Merge branch 'develop' of https://github.com/PaddlePaddle/paddle into complete-hsigmoid

.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)^.*$

.travis.yml

@@ -31,7 +31,7 @@ script:
     if [[ "$JOB" != "doc" ]]; then exit 0; fi;
     # For document only
     if [[ "$TRAVIS_PULL_REQUEST" != "false" ]]; then exit 0; fi;
-    if [[ "$TRAVIS_BRANCH" != "develop"  && ! "$TRAVIS_BRANCH" =~ ^v[[:digit:]]+\.[[:digit:]]+(\.[[:digit:]]+)?(-\S*)?$ ]]; then exit 0; fi;
+    if [[ "$TRAVIS_BRANCH" != "develop"  && ! "$TRAVIS_BRANCH" =~ ^v|release/[[:digit:]]+\.[[:digit:]]+(\.[[:digit:]]+)?(-\S*)?$ ]]; then exit 0; fi;
     export DEPLOY_DOCS_SH=https://raw.githubusercontent.com/PaddlePaddle/PaddlePaddle.org/master/scripts/deploy/deploy_docs.sh
     export DOCS_DIR=`pwd`
     cd ..

CMakeLists.txt

@@ -61,8 +61,11 @@ 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)
+option(WITH_SYSTEM_BLAS   "Use system blas library"           OFF)
 
 # CMAKE_BUILD_TYPE
 if(NOT CMAKE_BUILD_TYPE)
@@ -131,6 +134,10 @@ 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
@@ -153,12 +160,24 @@ 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)
@@ -178,7 +197,7 @@ 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")
 
@@ -222,7 +241,7 @@ 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

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)

Dockerfile

@@ -23,7 +23,7 @@ ENV HOME /root
 COPY ./paddle/scripts/docker/root/ /root/
 
 RUN apt-get update && \
-    apt-get install -y --allow-downgrades \
+    apt-get install -y --allow-downgrades patchelf \
     git python-pip python-dev python-opencv openssh-server bison \
     libnccl2=2.1.2-1+cuda8.0 libnccl-dev=2.1.2-1+cuda8.0 \
     wget unzip unrar tar xz-utils bzip2 gzip coreutils ntp \
@@ -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

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)
 
@@ -19,6 +18,8 @@ learning to many products at Baidu.
 Our vision is to enable deep learning for everyone via PaddlePaddle.
 Please refer to our [release announcement](https://github.com/PaddlePaddle/Paddle/releases) to track the latest feature of PaddlePaddle.
 
+### Lastest PaddlePaddle Version: [Fluid](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/fluid)
+
 ## Features
 
 - **Flexibility**

benchmark/fluid/Dockerfile

 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/

benchmark/fluid/args.py

@@ -122,5 +122,13 @@ 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, use inference transpiler to optimize the program.')
+    parser.add_argument(
+        '--no_random',
+        action='store_true',
+        help='If set, keep the random seed and do not shuffle the data.')
     args = parser.parse_args()
     return args

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 +131,7 @@ 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 not args.use_reader_op:
         feed_var_list = [
             var for var in train_prog.global_block().vars.itervalues()
@@ -181,6 +182,10 @@ def train(avg_loss, infer_prog, optimizer, train_reader, test_reader, batch_acc,
         print("Pass: %d, Loss: %f" % (pass_id, np.mean(train_losses))),
         # evaluation
         if not args.no_test and batch_acc and not args.use_reader_op:
+            if args.use_inference_transpiler:
+                t = fluid.InferenceTranspiler()
+                t.transpile(infer_prog, place)
+
             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:
@@ -301,9 +304,20 @@ 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()
+    if args.no_random:
+        fluid.default_startup_program().random_seed = 1
 
     # the unique trainer id, starting from 0, needed by trainer
     # only

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":

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')

benchmark/fluid/models/resnet.py

@@ -197,12 +197,12 @@ def get_model(args):
     optimizer = fluid.optimizer.Momentum(learning_rate=0.01, momentum=0.9)
 
     batched_train_reader = paddle.batch(
-        paddle.reader.shuffle(
+        train_reader if args.no_random else paddle.reader.shuffle(
             train_reader, buf_size=5120),
         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)
+        test_reader, batch_size=args.batch_size, drop_last=True)
 
     return avg_cost, inference_program, optimizer, batched_train_reader,\
                    batched_test_reader, batch_acc

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

cmake/cblas.cmake

@@ -83,18 +83,20 @@ else()
   set(REFERENCE_CBLAS_LIB_SEARCH_PATHS ${REFERENCE_CBLAS_ROOT}/lib)
 endif()
 
-find_path(REFERENCE_CBLAS_INCLUDE_DIR NAMES cblas.h PATHS
+if(WITH_SYSTEM_BLAS)
+  find_path(REFERENCE_CBLAS_INCLUDE_DIR NAMES cblas.h PATHS
         ${REFERENCE_CBLAS_INCLUDE_SEARCH_PATHS})
-find_library(REFERENCE_CBLAS_LIBRARY NAMES cblas PATHS
+  find_library(REFERENCE_CBLAS_LIBRARY NAMES cblas PATHS
         ${REFERENCE_CBLAS_LIB_SEARCH_PATHS})
 
-if(REFERENCE_CBLAS_INCLUDE_DIR AND REFERENCE_CBLAS_LIBRARY)
-  set(CBLAS_FOUND ON)
-  set(CBLAS_PROVIDER REFERENCE)
-  set(CBLAS_INC_DIR ${REFERENCE_CBLAS_INCLUDE_DIR})
-  set(CBLAS_LIBRARIES ${REFERENCE_CBLAS_LIBRARY})
-  add_definitions(-DPADDLE_USE_REFERENCE_CBLAS)
-  message(STATUS "Found reference-cblas (include: ${CBLAS_INC_DIR}, library: ${CBLAS_LIBRARIES})")
+  if(REFERENCE_CBLAS_INCLUDE_DIR AND REFERENCE_CBLAS_LIBRARY)
+    set(CBLAS_FOUND ON)
+    set(CBLAS_PROVIDER REFERENCE)
+    set(CBLAS_INC_DIR ${REFERENCE_CBLAS_INCLUDE_DIR})
+    set(CBLAS_LIBRARIES ${REFERENCE_CBLAS_LIBRARY})
+    add_definitions(-DPADDLE_USE_REFERENCE_CBLAS)
+    message(STATUS "Found reference-cblas (include: ${CBLAS_INC_DIR}, library: ${CBLAS_LIBRARIES})")
+  endif()
 endif()
 
 if(IOS_USE_VECLIB_FOR_BLAS AND VECLIB_FOUND)

cmake/configure.cmake

@@ -174,3 +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)

cmake/external/anakin.cmake

@@ -7,7 +7,17 @@ set(ANAKIN_INSTALL_DIR "${THIRD_PARTY_PATH}/install/anakin" CACHE PATH
 set(ANAKIN_INCLUDE "${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_COMPILE_EXTRA_FLAGS 
+    -Wno-error=unused-variable -Wno-unused-variable 
+    -Wno-error=format-extra-args -Wno-format-extra-args
+    -Wno-error=comment -Wno-comment 
+    -Wno-error=format -Wno-format 
+    -Wno-error=switch -Wno-switch
+    -Wno-error=return-type -Wno-return-type 
+    -Wno-error=non-virtual-dtor -Wno-non-virtual-dtor
+    -Wno-sign-compare
+    -Wno-reorder 
+    -Wno-error=cpp)
 
 set(ANAKIN_LIBRARY_URL "https://github.com/pangge/Anakin/releases/download/3.0/anakin_release_simple.tar.gz")
 
@@ -26,13 +36,15 @@ function(fetch_include_recursively root_dir)
     endforeach()
 endfunction()
 
-# 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 ${ANAKIN_INSTALL_DIR}; tar xzf anakin_release_simple.tar.gz")
+if (NOT EXISTS "${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")

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)

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 ""

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}

cmake/external/openblas.cmake

@@ -114,7 +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})
-TARGET_LINK_LIBRARIES(cblas ${CBLAS_LIBRARIES})
+
+IF("${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)

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} ${cc_library_DEPS})
     endif()
@@ -234,13 +257,14 @@ function(cc_test TARGET_NAME)
     set(multiValueArgs SRCS DEPS ARGS)
     cmake_parse_arguments(cc_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
     add_executable(${TARGET_NAME} ${cc_test_SRCS})
-    target_link_libraries(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main memory gtest gflags glog)
-    add_dependencies(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main memory gtest gflags glog)
+    target_link_libraries(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main lod_tensor memory gtest gflags glog)
+    add_dependencies(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main lod_tensor memory gtest gflags glog)
     add_test(NAME ${TARGET_NAME}
              COMMAND ${TARGET_NAME} ${cc_test_ARGS}
              WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})
     if (${cc_test_SERIAL})
         set_property(TEST ${TARGET_NAME} PROPERTY SERIAL 1)
+    set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_init_allocated_mem=true)
     endif()
   endif()
 endfunction(cc_test)
@@ -300,11 +324,12 @@ function(nv_test TARGET_NAME)
     set(multiValueArgs SRCS DEPS)
     cmake_parse_arguments(nv_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
     cuda_add_executable(${TARGET_NAME} ${nv_test_SRCS})
-    target_link_libraries(${TARGET_NAME} ${nv_test_DEPS} paddle_gtest_main memory gtest gflags glog)
-    add_dependencies(${TARGET_NAME} ${nv_test_DEPS} paddle_gtest_main memory gtest gflags glog)
+    target_link_libraries(${TARGET_NAME} ${nv_test_DEPS} paddle_gtest_main lod_tensor memory gtest gflags glog)
+    add_dependencies(${TARGET_NAME} ${nv_test_DEPS} paddle_gtest_main lod_tensor memory gtest gflags glog)
     add_test(${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()

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 "")
@@ -149,21 +136,33 @@ copy(memory_lib
   DSTS ${dst_dir}/${module} ${dst_dir}/${module}/detail
 )
 
-set(module "inference")
-copy(inference_lib DEPS paddle_fluid_shared paddle_fluid
-  SRCS ${src_dir}/${module}/*.h ${PADDLE_BINARY_DIR}/paddle/fluid/inference/libpaddle_fluid.*
-  DSTS ${dst_dir}/${module} ${dst_dir}/${module}
-)
+set(inference_deps paddle_fluid_shared paddle_fluid)
 
 if(WITH_CONTRIB)
-   set(contrib_dst_dir "${FLUID_INSTALL_DIR}/contrib/inference")
-   copy(contrib_inference_lib DEPS paddle_inference_api
+    message(STATUS "installing contrib")
+    set(contrib_dst_dir "${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}
-   )
+        ${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 ${inference_deps}
+  SRCS ${src_dir}/${module}/*.h ${PADDLE_BINARY_DIR}/paddle/fluid/inference/libpaddle_fluid.*
+  DSTS ${dst_dir}/${module} ${dst_dir}/${module}
+)
+
 set(module "platform")
 copy(platform_lib DEPS profiler_py_proto
   SRCS ${src_dir}/${module}/*.h ${src_dir}/${module}/dynload/*.h ${src_dir}/${module}/details/*.h

cmake/version.cmake

 # Get the latest git tag.
 set(PADDLE_VERSION $ENV{PADDLE_VERSION})
 set(tmp_version "HEAD")
+set(TAG_VERSION_REGEX "[0-9]+\\.[0-9]+\\.[0-9]+(\\.(a|b|rc)\\.[0-9]+)?")
+set(COMMIT_VERSION_REGEX "[0-9a-f]+[0-9a-f]+[0-9a-f]+[0-9a-f]+[0-9a-f]+")
 while ("${PADDLE_VERSION}" STREQUAL "")
   execute_process(
-    COMMAND ${GIT_EXECUTABLE} describe --tags --abbrev=0 ${tmp_version}
+    COMMAND ${GIT_EXECUTABLE} describe --tags --abbrev=0 --always ${tmp_version}
     WORKING_DIRECTORY ${PADDLE_SOURCE_DIR}
     OUTPUT_VARIABLE GIT_TAG_NAME
     RESULT_VARIABLE GIT_RESULT
     ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
   if (NOT ${GIT_RESULT})
     # Check the tag is a correct version
-    if (${GIT_TAG_NAME} MATCHES "v[0-9]+\\.[0-9]+\\.[0-9]+(\\.(a|b|rc)\\.[0-9]+)?")
+    if (${GIT_TAG_NAME} MATCHES "${COMMIT_VERSION_REGEX}")
+      # if no tag was found, set PADDLE_VERSION to latest
+      set(PADDLE_VERSION "latest")
+    elseif (${GIT_TAG_NAME} MATCHES "v${TAG_VERSION_REGEX}")
       string(REPLACE "v" "" PADDLE_VERSION ${GIT_TAG_NAME})
     else()  # otherwise, get the previous git tag name.
       set(tmp_version "${GIT_TAG_NAME}~1")

doc/about/about_us.rst

+=========
+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

doc/fluid/api/average.rst

+..  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:
+

doc/fluid/api/backward.rst

+..  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:
+

doc/fluid/api/clip.rst

 ..  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:
-

doc/fluid/api/data.rst

-==================================
-Data Reader Interface and DataSets
-==================================
-
-..  toctree::
-    :maxdepth: 1
-
-    data/data_reader.rst
-    data/image.rst
-    data/dataset.rst

doc/fluid/api/data_feeder.rst

 ..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
     !DO NOT EDIT THIS FILE MANUALLY!
 
-===========
-data_feeder
-===========
+=================
+fluid.data_feeder
+=================
+
+.. _api_fluid_data_feeder_DataFeeder:
 
 DataFeeder
 ----------

doc/fluid/api/evaluator.rst

-..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
-    !DO NOT EDIT THIS FILE MANUALLY!
-
-=========
-evaluator
-=========
-

doc/fluid/api/executor.rst

 ..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
     !DO NOT EDIT THIS FILE MANUALLY!
 
-========
-executor
-========
+==============
+fluid.executor
+==============
+
+.. _api_fluid_executor_Executor:
 
 Executor
 --------
@@ -12,24 +14,32 @@ Executor
     :members:
     :noindex:
 
+.. _api_fluid_executor_global_scope:
+
 global_scope
 ------------
 
 ..  autofunction:: paddle.fluid.executor.global_scope
     :noindex:
 
+.. _api_fluid_executor_scope_guard:
+
 scope_guard
 -----------
 
 ..  autofunction:: paddle.fluid.executor.scope_guard
     :noindex:
 
-switch_scope
-------------
+.. _api_fluid_executor__switch_scope:
+
+_switch_scope
+-------------
 
-..  autofunction:: paddle.fluid.executor.switch_scope
+..  autofunction:: paddle.fluid.executor._switch_scope
     :noindex:
 
+.. _api_fluid_executor_fetch_var:
+
 fetch_var
 ---------
 

doc/fluid/api/fluid.rst

+..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
+    !DO NOT EDIT THIS FILE MANUALLY!
+
+=====
+fluid
+=====
+
+.. _api_fluid_Block:
+
+Block
+-----
+
+..  autoclass:: paddle.fluid.Block
+    :members:
+    :noindex:
+
+.. _api_fluid_Variable:
+
+Variable
+--------
+
+..  autoclass:: paddle.fluid.Variable
+    :members:
+    :noindex:
+
+.. _api_fluid_Program:
+
+Program
+-------
+
+..  autoclass:: paddle.fluid.Program
+    :members:
+    :noindex:
+
+.. _api_fluid_Operator:
+
+Operator
+--------
+
+..  autoclass:: paddle.fluid.Operator
+    :members:
+    :noindex:
+
+.. _api_fluid_default_startup_program:
+
+default_startup_program
+-----------------------
+
+..  autofunction:: paddle.fluid.default_startup_program
+    :noindex:
+
+.. _api_fluid_default_main_program:
+
+default_main_program
+--------------------
+
+..  autofunction:: paddle.fluid.default_main_program
+    :noindex:
+
+.. _api_fluid_program_guard:
+
+program_guard
+-------------
+
+..  autofunction:: paddle.fluid.program_guard
+    :noindex:
+
+.. _api_fluid_get_var:
+
+get_var
+-------
+
+..  autofunction:: paddle.fluid.get_var
+    :noindex:
+
+.. _api_fluid_Executor:
+
+Executor
+--------
+
+..  autoclass:: paddle.fluid.Executor
+    :members:
+    :noindex:
+
+.. _api_fluid_global_scope:
+
+global_scope
+------------
+
+..  autofunction:: paddle.fluid.global_scope
+    :noindex:
+
+.. _api_fluid_scope_guard:
+
+scope_guard
+-----------
+
+..  autofunction:: paddle.fluid.scope_guard
+    :noindex:
+
+.. _api_fluid__switch_scope:
+
+_switch_scope
+-------------
+
+..  autofunction:: paddle.fluid._switch_scope
+    :noindex:
+
+.. _api_fluid_fetch_var:
+
+fetch_var
+---------
+
+..  autofunction:: paddle.fluid.fetch_var
+    :noindex:
+
+.. _api_fluid_Go:
+
+Go
+--
+
+..  autoclass:: paddle.fluid.Go
+    :members:
+    :noindex:
+
+.. _api_fluid_make_channel:
+
+make_channel
+------------
+
+..  autofunction:: paddle.fluid.make_channel
+    :noindex:
+
+.. _api_fluid_channel_send:
+
+channel_send
+------------
+
+..  autofunction:: paddle.fluid.channel_send
+    :noindex:
+
+.. _api_fluid_channel_recv:
+
+channel_recv
+------------
+
+..  autofunction:: paddle.fluid.channel_recv
+    :noindex:
+
+.. _api_fluid_channel_close:
+
+channel_close
+-------------
+
+..  autofunction:: paddle.fluid.channel_close
+    :noindex:
+
+.. _api_fluid_Select:
+
+Select
+------
+
+..  autoclass:: paddle.fluid.Select
+    :members:
+    :noindex:
+
+.. _api_fluid_Trainer:
+
+Trainer
+-------
+
+..  autoclass:: paddle.fluid.Trainer
+    :members:
+    :noindex:
+
+.. _api_fluid_BeginEpochEvent:
+
+BeginEpochEvent
+---------------
+
+..  autoclass:: paddle.fluid.BeginEpochEvent
+    :members:
+    :noindex:
+
+.. _api_fluid_EndEpochEvent:
+
+EndEpochEvent
+-------------
+
+..  autoclass:: paddle.fluid.EndEpochEvent
+    :members:
+    :noindex:
+
+.. _api_fluid_BeginStepEvent:
+
+BeginStepEvent
+--------------
+
+..  autoclass:: paddle.fluid.BeginStepEvent
+    :members:
+    :noindex:
+
+.. _api_fluid_EndStepEvent:
+
+EndStepEvent
+------------
+
+..  autoclass:: paddle.fluid.EndStepEvent
+    :members:
+    :noindex:
+
+.. _api_fluid_CheckpointConfig:
+
+CheckpointConfig
+----------------
+
+..  autoclass:: paddle.fluid.CheckpointConfig
+    :members:
+    :noindex:
+
+.. _api_fluid_Inferencer:
+
+Inferencer
+----------
+
+..  autoclass:: paddle.fluid.Inferencer
+    :members:
+    :noindex:
+
+.. _api_fluid_DistributeTranspiler:
+
+DistributeTranspiler
+--------------------
+
+..  autoclass:: paddle.fluid.DistributeTranspiler
+    :members:
+    :noindex:
+
+.. _api_fluid_memory_optimize:
+
+memory_optimize
+---------------
+
+..  autofunction:: paddle.fluid.memory_optimize
+    :noindex:
+
+.. _api_fluid_release_memory:
+
+release_memory
+--------------
+
+..  autofunction:: paddle.fluid.release_memory
+    :noindex:
+
+.. _api_fluid_ParallelExecutor:
+
+ParallelExecutor
+----------------
+
+..  autoclass:: paddle.fluid.ParallelExecutor
+    :members:
+    :noindex:
+
+.. _api_fluid_ExecutionStrategy:
+
+ExecutionStrategy
+-----------------
+
+..  autoclass:: paddle.fluid.ExecutionStrategy
+    :members:
+    :noindex:
+
+.. _api_fluid_BuildStrategy:
+
+BuildStrategy
+-------------
+
+..  autoclass:: paddle.fluid.BuildStrategy
+    :members:
+    :noindex:
+
+.. _api_fluid_create_lod_tensor:
+
+create_lod_tensor
+-----------------
+
+..  autofunction:: paddle.fluid.create_lod_tensor
+    :noindex:
+
+.. _api_fluid_create_random_int_lodtensor:
+
+create_random_int_lodtensor
+---------------------------
+
+..  autofunction:: paddle.fluid.create_random_int_lodtensor
+    :noindex:
+
+.. _api_fluid_LoDTensor:
+
+LoDTensor
+---------
+
+..  autoclass:: paddle.fluid.LoDTensor
+    :members:
+    :noindex:
+
+.. _api_fluid_CPUPlace:
+
+CPUPlace
+--------
+
+..  autoclass:: paddle.fluid.CPUPlace
+    :members:
+    :noindex:
+
+.. _api_fluid_CUDAPlace:
+
+CUDAPlace
+---------
+
+..  autoclass:: paddle.fluid.CUDAPlace
+    :members:
+    :noindex:
+
+.. _api_fluid_CUDAPinnedPlace:
+
+CUDAPinnedPlace
+---------------
+
+..  autoclass:: paddle.fluid.CUDAPinnedPlace
+    :members:
+    :noindex:
+
+.. _api_fluid_Tensor:
+
+Tensor
+------
+
+..  autoclass:: paddle.fluid.Tensor
+    :members:
+    :noindex:
+
+.. _api_fluid_ParamAttr:
+
+ParamAttr
+---------
+
+..  autoclass:: paddle.fluid.ParamAttr
+    :members:
+    :noindex:
+
+.. _api_fluid_WeightNormParamAttr:
+
+WeightNormParamAttr
+-------------------
+
+..  autoclass:: paddle.fluid.WeightNormParamAttr
+    :members:
+    :noindex:
+
+.. _api_fluid_DataFeeder:
+
+DataFeeder
+----------
+
+..  autoclass:: paddle.fluid.DataFeeder
+    :members:
+    :noindex:
+
+.. _api_fluid_Scope:
+
+Scope
+-----
+
+..  autoclass:: paddle.fluid.Scope
+    :members:
+    :noindex:
+

doc/fluid/api/gen_doc.py

@@ -29,19 +29,27 @@ def parse_arg():
 
 
 class DocGenerator(object):
-    def __init__(self, module_name, stream=sys.stdout):
+    def __init__(self, module_name=None, stream=sys.stdout):
+        if module_name == "":
+            module_name = None
         self.stream = stream
-        self.module_name = module_name
-        if not hasattr(fluid, module_name):
-            raise ValueError("Cannot find fluid.{0}".format(module_name))
+        if module_name is None:
+            self.module_name = "fluid"
         else:
-            self.module = getattr(fluid, module_name)
+            self.module_name = "fluid." + module_name
+        if module_name is None:
+            self.module = fluid
+        else:
+            if not hasattr(fluid, module_name):
+                raise ValueError("Cannot find fluid.{0}".format(module_name))
+            else:
+                self.module = getattr(fluid, module_name)
         self.stream.write('''..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
     !DO NOT EDIT THIS FILE MANUALLY!
 
 ''')
 
-        self._print_header_(module_name, dot='=', is_title=True)
+        self._print_header_(self.module_name, dot='=', is_title=True)
 
     def print_submodule(self, submodule_name):
         submodule = getattr(self.module, submodule_name)
@@ -60,25 +68,29 @@ class DocGenerator(object):
         self._print_header_(name, dot='=', is_title=False)
 
     def print_item(self, name):
-        item = getattr(self.module, name)
+        item = getattr(self.module, name, None)
+        if item is None:
+            return
         if isinstance(item, types.TypeType):
             self.print_class(name)
         elif isinstance(item, types.FunctionType):
             self.print_method(name)
         else:
-            raise RuntimeError("Unsupported item {0}".format(name))
+            pass
 
     def print_class(self, name):
+        self._print_ref_(name)
         self._print_header_(name, dot='-', is_title=False)
-        self.stream.write('''..  autoclass:: paddle.fluid.{0}.{1}
+        self.stream.write('''..  autoclass:: paddle.{0}.{1}
     :members:
     :noindex:
 
 '''.format(self.module_name, name))
 
     def print_method(self, name):
+        self._print_ref_(name)
         self._print_header_(name, dot='-', is_title=False)
-        self.stream.write('''..  autofunction:: paddle.fluid.{0}.{1}
+        self.stream.write('''..  autofunction:: paddle.{0}.{1}
     :noindex:
 
 '''.format(self.module_name, name))
@@ -94,6 +106,10 @@ class DocGenerator(object):
         self.stream.write('\n')
         self.stream.write('\n')
 
+    def _print_ref_(self, name):
+        self.stream.write(".. _api_{0}_{1}:\n\n".format("_".join(
+            self.module_name.split(".")), name))
+
 
 def main():
     args = parse_arg()

doc/fluid/api/gen_doc.sh

 #!/bin/bash
-python gen_doc.py layers --submodules control_flow device io nn ops tensor detection learning_rate_scheduler > layers.rst
+python gen_doc.py layers --submodules control_flow device io nn ops tensor learning_rate_scheduler detection metric_op tensor > layers.rst
 
-for module in data_feeder clip metrics executor initializer io nets optimizer param_attr profiler regularizer
+for module in data_feeder clip metrics executor initializer io nets optimizer param_attr profiler regularizer transpiler recordio_writer backward average profiler
 do
   python gen_doc.py ${module} > ${module}.rst
 done
+
+python gen_doc.py "" > fluid.rst

doc/fluid/api/index_en.rst

-======================
-Fluid
-======================
+=============
+API Reference
+=============
 
 ..  toctree::
     :maxdepth: 1
 
+    fluid.rst
     layers.rst
     data_feeder.rst
     executor.rst
@@ -18,3 +19,8 @@ Fluid
     regularizer.rst
     io.rst
     data.rst
+    transpiler.rst
+    recordio_writer.rst
+    backward.rst
+    average.rst
+    profiler.rst

doc/fluid/api/initializer.rst

 ..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
     !DO NOT EDIT THIS FILE MANUALLY!
 
-===========
-initializer
-===========
+=================
+fluid.initializer
+=================
+
+.. _api_fluid_initializer_Constant:
 
 Constant
 --------
@@ -12,6 +14,8 @@ Constant
     :members:
     :noindex:
 
+.. _api_fluid_initializer_Uniform:
+
 Uniform
 -------
 
@@ -19,6 +23,8 @@ Uniform
     :members:
     :noindex:
 
+.. _api_fluid_initializer_Normal:
+
 Normal
 ------
 
@@ -26,6 +32,8 @@ Normal
     :members:
     :noindex:
 
+.. _api_fluid_initializer_Xavier:
+
 Xavier
 ------
 
@@ -33,18 +41,42 @@ Xavier
     :members:
     :noindex:
 
+.. _api_fluid_initializer_Bilinear:
+
+Bilinear
+--------
+
+..  autoclass:: paddle.fluid.initializer.Bilinear
+    :members:
+    :noindex:
+
+.. _api_fluid_initializer_MSRA:
+
+MSRA
+----
+
+..  autoclass:: paddle.fluid.initializer.MSRA
+    :members:
+    :noindex:
+
+.. _api_fluid_initializer_force_init_on_cpu:
+
 force_init_on_cpu
 -----------------
 
 ..  autofunction:: paddle.fluid.initializer.force_init_on_cpu
     :noindex:
 
+.. _api_fluid_initializer_init_on_cpu:
+
 init_on_cpu
 -----------
 
 ..  autofunction:: paddle.fluid.initializer.init_on_cpu
     :noindex:
 
+.. _api_fluid_initializer_ConstantInitializer:
+
 ConstantInitializer
 -------------------
 
@@ -52,6 +84,8 @@ ConstantInitializer
     :members:
     :noindex:
 
+.. _api_fluid_initializer_UniformInitializer:
+
 UniformInitializer
 ------------------
 
@@ -59,6 +93,8 @@ UniformInitializer
     :members:
     :noindex:
 
+.. _api_fluid_initializer_NormalInitializer:
+
 NormalInitializer
 -----------------
 
@@ -66,6 +102,8 @@ NormalInitializer
     :members:
     :noindex:
 
+.. _api_fluid_initializer_XavierInitializer:
+
 XavierInitializer
 -----------------
 
@@ -73,3 +111,21 @@ XavierInitializer
     :members:
     :noindex:
 
+.. _api_fluid_initializer_BilinearInitializer:
+
+BilinearInitializer
+-------------------
+
+..  autoclass:: paddle.fluid.initializer.BilinearInitializer
+    :members:
+    :noindex:
+
+.. _api_fluid_initializer_MSRAInitializer:
+
+MSRAInitializer
+---------------
+
+..  autoclass:: paddle.fluid.initializer.MSRAInitializer
+    :members:
+    :noindex:
+

doc/fluid/api/io.rst

 ..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
     !DO NOT EDIT THIS FILE MANUALLY!
 
-==
-io
-==
+========
+fluid.io
+========
+
+.. _api_fluid_io_save_vars:
 
 save_vars
 ---------
@@ -11,51 +13,115 @@ save_vars
 ..  autofunction:: paddle.fluid.io.save_vars
     :noindex:
 
+.. _api_fluid_io_save_params:
+
 save_params
 -----------
 
 ..  autofunction:: paddle.fluid.io.save_params
     :noindex:
 
+.. _api_fluid_io_save_persistables:
+
 save_persistables
 -----------------
 
 ..  autofunction:: paddle.fluid.io.save_persistables
     :noindex:
 
+.. _api_fluid_io_load_vars:
+
 load_vars
 ---------
 
 ..  autofunction:: paddle.fluid.io.load_vars
     :noindex:
 
+.. _api_fluid_io_load_params:
+
 load_params
 -----------
 
 ..  autofunction:: paddle.fluid.io.load_params
     :noindex:
 
+.. _api_fluid_io_load_persistables:
+
 load_persistables
 -----------------
 
 ..  autofunction:: paddle.fluid.io.load_persistables
     :noindex:
 
+.. _api_fluid_io_save_inference_model:
+
 save_inference_model
 --------------------
 
 ..  autofunction:: paddle.fluid.io.save_inference_model
     :noindex:
 
+.. _api_fluid_io_load_inference_model:
+
 load_inference_model
 --------------------
 
 ..  autofunction:: paddle.fluid.io.load_inference_model
     :noindex:
 
+.. _api_fluid_io_get_inference_program:
+
 get_inference_program
 ---------------------
 
 ..  autofunction:: paddle.fluid.io.get_inference_program
     :noindex:
 
+.. _api_fluid_io_save_checkpoint:
+
+save_checkpoint
+---------------
+
+..  autofunction:: paddle.fluid.io.save_checkpoint
+    :noindex:
+
+.. _api_fluid_io_load_checkpoint:
+
+load_checkpoint
+---------------
+
+..  autofunction:: paddle.fluid.io.load_checkpoint
+    :noindex:
+
+.. _api_fluid_io_clean_checkpoint:
+
+clean_checkpoint
+----------------
+
+..  autofunction:: paddle.fluid.io.clean_checkpoint
+    :noindex:
+
+.. _api_fluid_io_load_persist_vars_without_grad:
+
+load_persist_vars_without_grad
+------------------------------
+
+..  autofunction:: paddle.fluid.io.load_persist_vars_without_grad
+    :noindex:
+
+.. _api_fluid_io_save_persist_vars_without_grad:
+
+save_persist_vars_without_grad
+------------------------------
+
+..  autofunction:: paddle.fluid.io.save_persist_vars_without_grad
+    :noindex:
+
+.. _api_fluid_io_get_latest_checkpoint_serial:
+
+get_latest_checkpoint_serial
+----------------------------
+
+..  autofunction:: paddle.fluid.io.get_latest_checkpoint_serial
+    :noindex:
+

doc/fluid/api/metrics.rst

 ..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
     !DO NOT EDIT THIS FILE MANUALLY!
 
-=======
-metrics
-=======
+=============
+fluid.metrics
+=============
+
+.. _api_fluid_metrics_MetricBase:
 
 MetricBase
 ----------
@@ -12,6 +14,8 @@ MetricBase
     :members:
     :noindex:
 
+.. _api_fluid_metrics_CompositeMetric:
+
 CompositeMetric
 ---------------
 
@@ -19,6 +23,26 @@ CompositeMetric
     :members:
     :noindex:
 
+.. _api_fluid_metrics_Precision:
+
+Precision
+---------
+
+..  autoclass:: paddle.fluid.metrics.Precision
+    :members:
+    :noindex:
+
+.. _api_fluid_metrics_Recall:
+
+Recall
+------
+
+..  autoclass:: paddle.fluid.metrics.Recall
+    :members:
+    :noindex:
+
+.. _api_fluid_metrics_Accuracy:
+
 Accuracy
 --------
 
@@ -26,6 +50,8 @@ Accuracy
     :members:
     :noindex:
 
+.. _api_fluid_metrics_ChunkEvaluator:
+
 ChunkEvaluator
 --------------
 
@@ -33,6 +59,8 @@ ChunkEvaluator
     :members:
     :noindex:
 
+.. _api_fluid_metrics_EditDistance:
+
 EditDistance
 ------------
 
@@ -40,6 +68,8 @@ EditDistance
     :members:
     :noindex:
 
+.. _api_fluid_metrics_DetectionMAP:
+
 DetectionMAP
 ------------
 
@@ -47,6 +77,8 @@ DetectionMAP
     :members:
     :noindex:
 
+.. _api_fluid_metrics_Auc:
+
 Auc
 ---
 

doc/fluid/api/nets.rst

 ..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
     !DO NOT EDIT THIS FILE MANUALLY!
 
-====
-nets
-====
+==========
+fluid.nets
+==========
+
+.. _api_fluid_nets_simple_img_conv_pool:
 
 simple_img_conv_pool
 --------------------
@@ -11,18 +13,24 @@ simple_img_conv_pool
 ..  autofunction:: paddle.fluid.nets.simple_img_conv_pool
     :noindex:
 
+.. _api_fluid_nets_sequence_conv_pool:
+
 sequence_conv_pool
 ------------------
 
 ..  autofunction:: paddle.fluid.nets.sequence_conv_pool
     :noindex:
 
+.. _api_fluid_nets_glu:
+
 glu
 ---
 
 ..  autofunction:: paddle.fluid.nets.glu
     :noindex:
 
+.. _api_fluid_nets_scaled_dot_product_attention:
+
 scaled_dot_product_attention
 ----------------------------
 

doc/fluid/api/optimizer.rst

 ..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
     !DO NOT EDIT THIS FILE MANUALLY!
 
-=========
-optimizer
-=========
+===============
+fluid.optimizer
+===============
+
+.. _api_fluid_optimizer_SGD:
 
 SGD
 ---
@@ -12,6 +14,8 @@ SGD
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_Momentum:
+
 Momentum
 --------
 
@@ -19,6 +23,8 @@ Momentum
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_Adagrad:
+
 Adagrad
 -------
 
@@ -26,6 +32,8 @@ Adagrad
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_Adam:
+
 Adam
 ----
 
@@ -33,6 +41,8 @@ Adam
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_Adamax:
+
 Adamax
 ------
 
@@ -40,6 +50,8 @@ Adamax
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_DecayedAdagrad:
+
 DecayedAdagrad
 --------------
 
@@ -47,6 +59,17 @@ DecayedAdagrad
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_Ftrl:
+
+Ftrl
+----
+
+..  autoclass:: paddle.fluid.optimizer.Ftrl
+    :members:
+    :noindex:
+
+.. _api_fluid_optimizer_SGDOptimizer:
+
 SGDOptimizer
 ------------
 
@@ -54,6 +77,8 @@ SGDOptimizer
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_MomentumOptimizer:
+
 MomentumOptimizer
 -----------------
 
@@ -61,6 +86,8 @@ MomentumOptimizer
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_AdagradOptimizer:
+
 AdagradOptimizer
 ----------------
 
@@ -68,6 +95,8 @@ AdagradOptimizer
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_AdamOptimizer:
+
 AdamOptimizer
 -------------
 
@@ -75,6 +104,8 @@ AdamOptimizer
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_AdamaxOptimizer:
+
 AdamaxOptimizer
 ---------------
 
@@ -82,6 +113,8 @@ AdamaxOptimizer
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_DecayedAdagradOptimizer:
+
 DecayedAdagradOptimizer
 -----------------------
 
@@ -89,6 +122,26 @@ DecayedAdagradOptimizer
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_RMSPropOptimizer:
+
+RMSPropOptimizer
+----------------
+
+..  autoclass:: paddle.fluid.optimizer.RMSPropOptimizer
+    :members:
+    :noindex:
+
+.. _api_fluid_optimizer_FtrlOptimizer:
+
+FtrlOptimizer
+-------------
+
+..  autoclass:: paddle.fluid.optimizer.FtrlOptimizer
+    :members:
+    :noindex:
+
+.. _api_fluid_optimizer_Adadelta:
+
 Adadelta
 --------
 
@@ -96,6 +149,8 @@ Adadelta
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_ModelAverage:
+
 ModelAverage
 ------------
 
@@ -103,6 +158,8 @@ ModelAverage
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_Optimizer:
+
 Optimizer
 ---------
 
@@ -110,3 +167,12 @@ Optimizer
     :members:
     :noindex:
 
+.. _api_fluid_optimizer_RMSPropOptimizer:
+
+RMSPropOptimizer
+----------------
+
+..  autoclass:: paddle.fluid.optimizer.RMSPropOptimizer
+    :members:
+    :noindex:
+

doc/fluid/api/param_attr.rst

 ..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
     !DO NOT EDIT THIS FILE MANUALLY!
 
-==========
-param_attr
-==========
+================
+fluid.param_attr
+================
+
+.. _api_fluid_param_attr_ParamAttr:
 
 ParamAttr
 ---------
@@ -12,6 +14,8 @@ ParamAttr
     :members:
     :noindex:
 
+.. _api_fluid_param_attr_WeightNormParamAttr:
+
 WeightNormParamAttr
 -------------------
 

doc/fluid/api/profiler.rst

 ..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
     !DO NOT EDIT THIS FILE MANUALLY!
 
-========
-profiler
-========
+==============
+fluid.profiler
+==============
+
+.. _api_fluid_profiler_cuda_profiler:
 
 cuda_profiler
 -------------
@@ -11,15 +13,35 @@ cuda_profiler
 ..  autofunction:: paddle.fluid.profiler.cuda_profiler
     :noindex:
 
+.. _api_fluid_profiler_reset_profiler:
+
 reset_profiler
 --------------
 
 ..  autofunction:: paddle.fluid.profiler.reset_profiler
     :noindex:
 
+.. _api_fluid_profiler_profiler:
+
 profiler
 --------
 
 ..  autofunction:: paddle.fluid.profiler.profiler
     :noindex:
 
+.. _api_fluid_profiler_start_profiler:
+
+start_profiler
+--------------
+
+..  autofunction:: paddle.fluid.profiler.start_profiler
+    :noindex:
+
+.. _api_fluid_profiler_stop_profiler:
+
+stop_profiler
+-------------
+
+..  autofunction:: paddle.fluid.profiler.stop_profiler
+    :noindex:
+

doc/fluid/api/recordio_writer.rst

+..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
+    !DO NOT EDIT THIS FILE MANUALLY!
+
+=====================
+fluid.recordio_writer
+=====================
+
+.. _api_fluid_recordio_writer_convert_reader_to_recordio_file:
+
+convert_reader_to_recordio_file
+-------------------------------
+
+..  autofunction:: paddle.fluid.recordio_writer.convert_reader_to_recordio_file
+    :noindex:
+
+.. _api_fluid_recordio_writer_convert_reader_to_recordio_files:
+
+convert_reader_to_recordio_files
+--------------------------------
+
+..  autofunction:: paddle.fluid.recordio_writer.convert_reader_to_recordio_files
+    :noindex:
+

doc/fluid/api/regularizer.rst

 ..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
     !DO NOT EDIT THIS FILE MANUALLY!
 
-===========
-regularizer
-===========
+=================
+fluid.regularizer
+=================
+
+.. _api_fluid_regularizer_append_regularization_ops:
 
 append_regularization_ops
 -------------------------
@@ -11,12 +13,7 @@ append_regularization_ops
 ..  autofunction:: paddle.fluid.regularizer.append_regularization_ops
     :noindex:
 
-WeightDecayRegularizer
-----------------------
-
-..  autoclass:: paddle.fluid.regularizer.WeightDecayRegularizer
-    :members:
-    :noindex:
+.. _api_fluid_regularizer_L1Decay:
 
 L1Decay
 -------
@@ -25,6 +22,8 @@ L1Decay
     :members:
     :noindex:
 
+.. _api_fluid_regularizer_L2Decay:
+
 L2Decay
 -------
 
@@ -32,6 +31,8 @@ L2Decay
     :members:
     :noindex:
 
+.. _api_fluid_regularizer_L1DecayRegularizer:
+
 L1DecayRegularizer
 ------------------
 
@@ -39,6 +40,8 @@ L1DecayRegularizer
     :members:
     :noindex:
 
+.. _api_fluid_regularizer_L2DecayRegularizer:
+
 L2DecayRegularizer
 ------------------
 

doc/fluid/api/transpiler.rst

+..  THIS FILE IS GENERATED BY `gen_doc.{py|sh}`
+    !DO NOT EDIT THIS FILE MANUALLY!
+
+================
+fluid.transpiler
+================
+
+.. _api_fluid_transpiler_DistributeTranspiler:
+
+DistributeTranspiler
+--------------------
+
+..  autoclass:: paddle.fluid.transpiler.DistributeTranspiler
+    :members:
+    :noindex:
+
+.. _api_fluid_transpiler_InferenceTranspiler:
+
+InferenceTranspiler
+-------------------
+
+..  autoclass:: paddle.fluid.transpiler.InferenceTranspiler
+    :members:
+    :noindex:
+
+.. _api_fluid_transpiler_memory_optimize:
+
+memory_optimize
+---------------
+
+..  autofunction:: paddle.fluid.transpiler.memory_optimize
+    :noindex:
+
+.. _api_fluid_transpiler_release_memory:
+
+release_memory
+--------------
+
+..  autofunction:: paddle.fluid.transpiler.release_memory
+    :noindex:
+
+.. _api_fluid_transpiler_HashName:
+
+HashName
+--------
+
+..  autoclass:: paddle.fluid.transpiler.HashName
+    :members:
+    :noindex:
+
+.. _api_fluid_transpiler_RoundRobin:
+
+RoundRobin
+----------
+
+..  autoclass:: paddle.fluid.transpiler.RoundRobin
+    :members:
+    :noindex:
+

doc/fluid/design/concepts/lod_tensor.md

@@ -173,6 +173,7 @@ are transformed into offsets of elements/words as follows:
 
 ## Slicing of LoD Tensors
 
+
 When we use the above 2-level LoD Tensor as the input to a nested-RNN, we need to retrieve certain sequences.  Here we define the sequence identified by branch <i,j,...> as the **<i,j,...>-slice**.
 
 For example, the <2>-slice of above example is
@@ -189,3 +190,22 @@ and the <2,0>-slice of above slice is
 10  12
   ||
 ```
+
+## Length Representation vs Offset Representation
+
+The offset representation is an implementation-oriented decision and it makes understanding the idea behind LoDTensor difficult.
+Hence, we encapsulate this implementation detail in C++ and expose the original length representation in our Python API. 
+Specifically, we call this length representation `recursive_sequence_lengths` and users can use the following code to set or get the `recursive_sequence_lengths` of a LoDTensor in Python:
+```Python
+# length representation of lod called recursive_sequence_lengths
+recursive_seq_lens = [[3, 1, 2], [2, 2, 1, 3, 1, 2]]
+# Create a LoDTensor that has the above recursive_sequence_lengths info.
+# This recursive_sequence_lengths will be converted to an offset representation of LoD in the C++ implementation under the hood.
+tensor = fluid.LoDTensor(lod)
+
+# Set/Change the recursive_sequence_lengths info of LoDTensor
+tensor.set_recursive_sequence_lengths([[3, 1, 2]])
+# Get the recursive_sequence_lengths info of a LoDTensor (the offset-based LoD representation stored in C++ will be converted 
+# back to length-based recursive_sequence_lengths), new_recursive_seq_lens = [[3, 1, 2]]
+new_recursive_seq_lens = tensor.recursive_sequence_lengths()
+```

doc/fluid/design/concepts/python_data_feeding.md

+# Python Data Feeding
+
+In the former implementation of Paddle Fluid, there are two ways to feed data:
+
+- Use `reader_op` in backend C++ side. This method only supports data feeding from recordio files and random data generators, but supports many kinds of `decorated_readers`. For examples, `double_buffer_reader` uses two threads to achieve better performance: one for time-consuming I/O operations, and the other for `Executor::Run()`. See [C++ Data Feeding](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/fluid/design/concepts/cpp_data_feeding.md) for details.
+
+- Feed data directly using `DataFeeder.feed()` in Python codes. It is more flexible than the first way. Many kinds of preprocessing steps can be performed before feeding using Python or any other languages, instead of adding many uncommon `operators` in C++ side. But this method is less efficient: the program cannot read the next mini-batch data before `Executor::Run()` ends. Moreover, `decorated_readers` such as `double_buffer_reader` cannot be used for better performance.
+
+In this document, we design a Python Data Feeding process combining the efficiency of the first way and the flexibility of the second way. A data queue `LoDTensorBlockingQueue` is designed to be shared by the Python and C++ side, while `LoDTensorArray` is pushed into the queue in Python side and `reader_op` in C++ side reads out the data from the queue.
+
+
+## Design of LoDTensorBlockingQueue
+`LoDTensorBlockingQueue` is a blocking queue with a fixed `capacity` and accepts `std::vector<framework::LoDTensor>` with shapes indicated by `dims`. Since `LoDTensorBlockingQueue` must be constructed using `capacity` and `dims`, it cannot be a `Variable` type. Therefore, a `LoDTensorBlockingQueueHolder` is designed to defer construction of `LoDTensorBlockingQueue`.
+
+```C++
+class LoDTensorBlockingQueueHolder;
+
+class LoDTensorBlockingQueue {
+  friend class LoDTensorBlockingQueueHolder;
+ private:
+  // `LoDTensorBlockingQueue` can only be constructed by 
+  // `LoDTensorBlockingQueueHolder::InitOnce()`
+  LoDTensorBlockingQueue(size_t capacity, const std::vector<framework::DDim>& dims);
+ 
+ public:
+  size_t Size() const { return queue_.Size(); } // Get the current size of the queue
+
+  size_t Cap() const { return queue_.Cap(); }// Get the capacity of the queue
+
+  void Close() { return queue_.Close(); }
+
+  bool IsClosed() const { return queue_.IsClosed(); }
+
+  // Block if Size() == Cap()
+  // Return false only when queue_.IsClosed() == true
+  bool Push(const std::vector<framework::LoDTensor> &lod_tensor_vec);
+  
+  // Block if Size() == 0.
+  // *Success == false when queue_.IsClosed() == true
+  std::vector<framework::LoDTensor> Pop(bool *success = nullptr);
+ 
+ private:
+  // Use reader::BlockingQueue as the inner data structure
+  BlockingQueue<std::vector<framework::LoDTensor>> queue_;
+  std::vector<framework::DDim> dims_;
+};
+
+class LoDTensorBlockingQueueHolder {
+ public:  
+  // Call the constructor of `LoDTensorBlockingQueue` to create queue_
+  // `InitOnce` can only called once, otherwise an exception would raise
+  void InitOnce(size_t capacity, const std::vector<framework::DDim>& dims) {
+    PADDLE_ENFORCE(queue_ == nullptr);
+    queue_.reset(new LoDTensorBlockingQueue(capacity, dims));
+  }
+
+  const std::shared_ptr<LoDTensorBlockingQueue>& GetQueue() const { return queue_; }
+
+ private:
+  std::shared_ptr<LoDTensorBlockingQueue> queue_;
+};
+```
+
+There are some major things that must be concerned:
+- `LoDTensorBlockingQueueHolder` should be a `Variable` in global scope, so that `reader_op` can find it when reading data.
+- A `Variable` of `LoDTensorBlockingQueueHolder` but not `VarDesc` must be created in Python code before `Executor::Run()` so that `Executor::Run()` can get the feeding data when it is called.
+- `Create_reader_op` should accept the name of the `LoDTensorBlockingQueueHolder` variable as an input.
+
+
+## Release of the GIL in pybind
+`Pybind11::gil_scoped_release` is used to release GIL (Global Interpreter Lock) when `LoDTensorBlockingQueue::Push()` or `Executor::Run()` method are invoked in Python side, making `LoDTensorBlockingQueue::Push()` and `Executor::Run()` run in parallel.
+
+
+## Design of PyReader
+`PyReader` is a reader which holds a `LoDTensorBlockingQueue` object.
+```C++
+class PyReader : public ReaderBase {
+ public:
+  explicit PyReader(const std::shared_ptr<LoDTensorBlockingQueue>& queue);
+  
+  void ReadNext(std::vector<framework::LoDTensor>* out) override {
+    bool success;
+    *out = queue_->Pop(&success);
+    if (!success) out->clear();
+  }
+  
+  void ReInit() override { return; }
+
+ private:
+  std::shared_ptr<LoDTensorBlockingQueue> queue_;
+};
+```
+
+
+## Design of CreatePyReaderOp
+`CreatePyReaderOp` is used to create the `PyReader` object. It requires an input `blocking_queue` which indicates the name of the `LoDTensorBlockingQueueHolder` variable.
+```C++
+class CreatePyReaderOp : public framework::OperatorBase {
+ public:
+  using framework::OperatorBase::OperatorBase;
+ private:
+  void RunImpl(const framework::Scope& scope,
+               const platform::Place& dev_place) const override {
+    auto* out = scope.FindVar(Output("Out"))
+                    ->template GetMutable<framework::ReaderHolder>();
+    if (out->Get() != nullptr) return;
+    
+    const std::string& queue_name = Input("blocking_queue");
+    auto* queue_holder_var = scope.FindVar(queue_name);
+    PADDLE_ENFORCE(queue_holder_var != nullptr);
+		auto* queue_holder = queue_holder_var
+                    ->template GetMutable<framework::LoDTensorBlockingQueueHolder>();
+    out->Reset(new PyReader(queue_holder->GetQueue()));
+  }
+};
+```
+
+## Design of Python codes
+The design of Python codes are as follows. First, we construct a variable of `LoDTensorBlockingQueueHolder` and init it with given parameters, returning the `LoDTensorBlockingQueue` object after initialization. After that, a layer of `CreatePyReaderOp` is constructed and accepts the name of the `LoDTensorBlockingQueueHolder` variable. The `LoDTensorBlockingQueue` object and result of the layer are both returned.
+```Python
+def py_reader(capacity, shapes):
+  queue_name = unique_name.generate("lod_tensor_blocking_queue")
+  var = global_scope().var(feeder_name) # create LoDTensorBlockingQueueHolder Variable
+  feed_queue = core.init_lod_tensor_blocking_queue(var, capacity, shapes) # init the queue
+  out = create_var()
+  create_py_reader_op_with_queue_name(
+      inputs={'blocking_queue': queue_name},
+      outputs={'Out':[out]})  
+  return out, feed_queue
+```

doc/fluid/design/dist_train/dist_train_nccl2.md

+# 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
+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 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.
+
+It have 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.
+
+In both two modes, this op can save the NCCL ID into current scope as a
+persistable variable, Then we can insert this op at the end of
+"startup program" of fluid, so that all workers can get the same ID to
+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
+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/design/dist_train/distributed_lookup_table_design.md

 # Design Doc: Distributed Lookup Table Operator
 
-A lookup table operator in PaddlePaddle where the table could be out
+A distribute lookup table operator in PaddlePaddle where the table could be out
 of the memory of a computer.
 
 ## Background
@@ -24,14 +24,14 @@ memory, so we'd need a distributed storage service, which supports the
 lookup of rows.
 
 The following figure illustrates the multiplication of x with two
-non-zero elements, or say, two symbols, and a lookup table W:
+non-zero elements, or say two symbols, and a lookup table W:
 
 ![lookup table](./src/lookup_table.png)
 
 ### The Backward Algorithm
 
 The backward algorithm computes W'(x) using W(x).  W'(x) has the same
-scale of size as W(x) and is much smaller than W.
+the scale of size as W(x) and is much smaller than W.
 
 To optimize W given W', we can do simple SGD update:
 
@@ -44,85 +44,46 @@ $$W = f(W, W')$$
 The following figure illustrates the backward pass of the lookup
 operator: ![lookup table training](./src/lookup_table_training.png)
 
-## Distributed Storage Service
-
-The forward algorithm requires a distributed storage service for W.
-The backward algorithm prefers that the storage system can apply the
-optimization algorithm on W.  The following two sections describe two
-solutions -- the former doesn't require that the storage service can
-do optimization, the latter does.
-
-### Storage Service Doesn't Optimize
-
-In this design, we use highly-optimized distributed storage, e.g.,
-memcached, as the storage service, and we run the optimization
-algorithm on parameter servers of PaddlePaddle.  The following figure
-illustrates the training process.
-
-<!--
-Note: please update the following URL when update this digraph.
-<img src='https://g.gravizo.com/svg?
-digraph G {
-  rankdir="LR";
-  subgraph cluster1 {
-  P1 [label="pserver 1"];
-  P2 [label="pserver 2"];
-  T1 [label="trainer 1"];
-  T2 [label="trainer 2"];
-  T3 [label="trainer 3"];
-  }
-  KV [label="memcached"];
-  T1 -> P1;
-  T1 -> P2;
-  T2 -> P1;
-  T2 -> P2;
-  T3 -> P1;
-  T3 -> P2;
-  P1 -> KV [color=gray, weight=0.1];
-  KV -> P1 [color=gray, weight=0.1];
-  P2 -> KV [color=gray, weight=0.1];
-  KV -> P2 [color=gray, weight=0.1];
-  KV -> T1 [color=gray, weight=0.1];
-  KV -> T2 [color=gray, weight=0.1];
-  KV -> T3 [color=gray, weight=0.1];
-}
-)
-'/>
--->
-
-<img src='https://g.gravizo.com/svg?%20digraph%20G%20{%20rankdir=%22LR%22;%20subgraph%20cluster1%20{%20P1%20[label=%22pserver%201%22];%20P2%20[label=%22pserver%202%22];%20T1%20[label=%22trainer%201%22];%20T2%20[label=%22trainer%202%22];%20T3%20[label=%22trainer%203%22];%20}%20KV%20[label=%22memcached%22];%20T1%20-%3E%20P1;%20T1%20-%3E%20P2;%20T2%20-%3E%20P1;%20T2%20-%3E%20P2;%20T3%20-%3E%20P1;%20T3%20-%3E%20P2;%20P1%20-%3E%20KV%20[color=gray,%20weight=0.1];%20KV%20-%3E%20P1%20[color=gray,%20weight=0.1];%20P2%20-%3E%20KV%20[color=gray,%20weight=0.1];%20KV%20-%3E%20P2%20[color=gray,%20weight=0.1];%20KV%20-%3E%20T1%20[color=gray,%20weight=0.1];%20KV%20-%3E%20T2%20[color=gray,%20weight=0.1];%20KV%20-%3E%20T3%20[color=gray,%20weight=0.1];%20}'/>
-
-Each trainer runs the forward and backward passes using their local
-data:
-
-1. In the forward pass, when a trainer runs the forward algorithm of a
-   lookup operator, it retrieves W(x) from the storage service.
-1. The trainer computes W'(x) in the backward pass using W(x).
-
-During the global update process:
-
-1. Each trainer uploads its W'(x) to parameter servers.
-1. The parameter server runs the optimization algorithm, e.g., the
-   Adam optimization algorithm, which requires that
-   1. The parameter server retrieves W(x) from memcached, and
-   1. The parameter server pushes $\Delta W(x)=f(W(x), lambda \sum_j
-      W'(x))$ to memcached, where $f$ denotes the optimization
-      algorithm.
-
-### Storage Service Does Optimize
-
-This design is very similar to the above one, except that the
-optimization algorithm $f$ runs on the storage service.
-
-- Pro: parameter servers do not retrieve W(x) from the storage
-  service, thus saves half network communication.
-- Con: the storage service needs to be able to run the optimization
-  algorithm.
-
-## Conclusion
-
-Let us do the "storage service does not optimize" solution first, as a
-baseline at least, because it is easier to use a well-optimized
-distributed storage service like memcached.  We can do the "storage
-service does optimize" solution later or at the same time, which, if
-implemented carefully, should have better performance than the former.
+## Distributed Lookup Table
+### Problem 1: The lookup table may be very large.
+
+ In the condition like the search engine and recommendation system, the number of feature Id may be very large, say 100,000,000,000, then for a float value lookup table of size 8, the total size of the table is:
+
+ ```
+ 100,000,000,000 * 8 * 4(Bytes) = 2980.23 GB
+ ```
+
+### Solution: Distributed storage
+
+1. Paddle use [SelectedRows](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/fluid/design/modules/selected_rows.md) as the storage format for the lookup table, the lookup table parameter will be split to multi-machine according to the hash of the feature ID, and data will also be split and send to the same machine to prefetch the parameter.
+
+1. For common parameters, the trainer will get the whole parameter for training, but for the big lookup table, the trainer can not store the whole parameter. Because the input data feature is very sparse, every time we only need a few parameters for training, so we use `prefetch_op` to only prefetch the parameter needed to trainer.
+
+### Problem 2. The Id in the lookup table is not sure before training.
+
+ The feature Id is calculated by the hash function because the feature data source is so large, we can not get all the Id before training. So we can not initialize the table before training.
+
+### Solution: Id auto growth
+
+At the beginning of training, paddle only malloc the memory for the lookup table at parameter server side, the Id and it's value will not be initialized. During training, when a parameter server received an Id, if it is already in the lookup table, it will return the existing parameter, if the Id does not exist, paddle will add it into the lookup table and initialize the value for it.
+
+### Problem 3: parameter load and save
+
+For common parameters, paddle use trainer to save and load them. But for distributed lookup table, trainer cannot do this because it's large size.
+
+### Solution: Parameter server side save and load
+
+Paddle support parameter server side save and load for distribute lookup table. Each machine of parameter servers will only save and load part of the whole table.
+
+## Architecture
+The whole architecture of the distribute lookup table is as below:
+
+### Training steps:
+1. Read a batch of data, the data is feature ids.
+1. The input ids will be split by `split_ids_op` with the same hash function of the lookup table.
+1. The `prefetch_op` use the split result to prefetch parameters back from the lookup table.
+1. Run forward-backward to get the gradient of the lookup table.
+1. `split_ids_op` split the gradient and then use `send_op` to the parameter server.
+1. parameter server update the table with the received gradient.
+
+![distribute lookup table](./src/distributed_lookup_table.jpeg)

doc/fluid/design/multi_devices/kernel_selection.md

@@ -74,10 +74,10 @@ void OperatorWithKernel::Run(
     auto kernel_type_for_var = this->GetKernelTypeForVar(...);
     if (kernel_type_for_var.place_ != expected_kernel_key.place_) {
       auto* trans_var = new_scope.Var(var_name);
-      auto* out = DataTransform(expected_kernel_key,
+      auto* out = TransformData(expected_kernel_key,
                                 kernel_type_for_var,
                                 *tensor_in);
-      CopyVariableWithTensor(...);
+      SetTensorToVariable(...);
     }
   }
 

doc/fluid/howto/cluster/fluid_cluster_train_cn.md

@@ -168,13 +168,13 @@ cd /paddle/python/paddle/fluid/tests/book
 
 第二步，启动Parameter Server：
 ```bash
-PADDLE_INIT_PORT=6174 PADDLE_INIT_PSERVERS=192.168.1.2 TRAINERS=2 POD_IP=192.168.1.2 PADDLE_INIT_TRAINER_ID=1 TRAINING_ROLE=PSERVER python test_fit_a_line.py
+PADDLE_PSERVER_PORT=6174 PADDLE_PSERVER_IPS=192.168.1.2 PADDLE_TRAINERS=2 PADDLE_CURRENT_IP=192.168.1.2 PADDLE_TRAINER_ID=1 PADDLE_TRAINING_ROLE=PSERVER python test_fit_a_line.py
 ```
 执行命令后请等待出现提示： ```Server listening on 192.168.1.2:6174 ```, 表示Paramter Server已经正常启动。
 
 第三步，启动Trainer：
 ```bash
-PADDLE_INIT_PORT=6174 PADDLE_INIT_PSERVERS=192.168.1.3 TRAINERS=2 POD_IP=192.168.1.3 PADDLE_INIT_TRAINER_ID=1 TRAINING_ROLE=TRAINER python test_fit_a_line.py
+PADDLE_PSERVER_PORT=6174 PADDLE_PSERVER_IPS=192.168.1.3 PADDLE_TRAINERS=2 PADDLE_CURRENT_IPP=192.168.1.3 PADDLE_TRAINER_ID=1 PADDLE_TRAINING_ROLE=TRAINER python test_fit_a_line.py
 ```
 由于我们定义的Trainer的数量是2个，因此需要在另外一个计算节点上再启动一个Trainer。
 

doc/fluid/howto/cluster/fluid_recordio.md

@@ -114,8 +114,8 @@ def gen_train_list(file_pattern, trainers, trainer_id):
            ret_list.append(f)
    return ret_list
 
-trainers = int(os.getenv("TRAINERS"))
-trainer_id = int(os.getenv("PADDLE_INIT_TRAINER_ID"))
+trainers = int(os.getenv("PADDLE_TRAINERS"))
+trainer_id = int(os.getenv("PADDLE_TRAINER_ID"))
 data_file = fluid.layers.io.open_files(
     filenames=gen_train_list("./mnist-[0-9]*.recordio", 2, 0),
     thread_num=1,

doc/fluid/howto/inference/build_and_install_lib_cn.rst

@@ -13,6 +13,7 @@ cpu_noavx_openblas       `fluid.tgz <https://guest:@paddleci.ngrok.io/repository
 cuda7.5_cudnn5_avx_mkl   `fluid.tgz <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda75cudnn5cp27cp27mu/.lastSuccessful/fluid.tgz>`_
 cuda8.0_cudnn5_avx_mkl   `fluid.tgz <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/fluid.tgz>`_
 cuda8.0_cudnn7_avx_mkl   `fluid.tgz <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/fluid.tgz>`_
+cuda9.0_cudnn7_avx_mkl   `fluid.tgz <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda90cudnn7avxMkl/.lastSuccessful/fluid.tgz>`_
 ======================   ========================================
 
 从源码编译

doc/fluid/howto/optimization/host_memory_profiling_cn.md

-## 堆内存分析和优化
+# 堆内存分析和优化
 
 计算机程序都可能有内存泄漏的风险。**内存泄漏**一般是由于程序在堆(heap)上分配了内存而没有释放，随着程序的运行占用的内存越来越大，一方面会影响程序的稳定性，可能让运行速度越来越慢，或者造成oom，甚至会影响运行程序的机器的稳定性，造成宕机。
 
@@ -20,11 +20,11 @@ Paddle也提供了基于gperftool的[CPU性能分析教程](https://github.com/P
 
 对于堆内存的分析，主要用到thread-caching malloc和heap-profiling using tcmalloc。
 
-## 使用流程
-#### 环境
+## 环境
+
 本教程基于paddle提供的Docker开发环境paddlepaddle/paddle:latest-dev，基于Ubuntu 16.04.4 LTS环境。
 
-#### 使用流程
+## 使用流程
 
 - 安装google-perftools
 

doc/fluid/howto/optimization/timeline_cn.md

+# 如何使用timeline工具做性能分析
+
+1. 在训练的主循环外加上`with profiler.profiler(...)`。运行之后，代码会在`/tmp/profile`目录下生成一个profile的记录文件。
+
+	**提示：**
+	请不要在timeline记录信息时运行太多次迭代，因为timeline中的记录数量和迭代次数是成正比的。
+
+	```python
+	with profiler.profiler('All', 'total', '/tmp/profile') as prof:
+	    for pass_id in range(pass_num):
+	        for batch_id, data in enumerate(train_reader()):
+	            exe.run(fluid.default_main_program(),
+	                    feed=feeder.feed(data),
+	                    fetch_list=[])
+	            ...
+	```
+
+1. 运行`python paddle/tools/timeline.py`来处理`/tmp/profile`，这个程序默认会生成一个`/tmp/timeline`文件，你也可以用命令行参数来修改这个路径，请参考[timeline.py](https://github.com/PaddlePaddle/Paddle/blob/develop/tools/timeline.py)。
+
+1. 打开chrome浏览器，访问<chrome://tracing/>，用`load`按钮来加载生成的`timeline`文件。
+
+	![chrome tracing](./tracing.jpeg)
+
+1. 结果如下图所示，可以放到来查看timetime的细节信息。
+
+	![chrome timeline](./timeline.jpeg)

doc/v2/design/cluster_train/large_model_dist_train.md

@@ -52,7 +52,7 @@ In `trainer_internal.cpp:L93 trainOneBatch`:
 
 When doing actual network forward and backward, at the beginning of each batch, the trainer will try to download one row of data from pserver.
 
-In `trainer/RemoteParameterUpdater.cpp`: `parameterUpdater_->getParametersRemote();`:
+In `legacy/trainer/RemoteParameterUpdater.cpp`: `parameterUpdater_->getParametersRemote();`:
 
 ```c++
 if (fullSize) {

doc/v2/design/interface/00.why_plain_c.md

@@ -65,7 +65,7 @@ paddle_error paddle_matrix_get_shape(paddle_matrix matrix,
 而在CPP里面实现这个C的接口，文件 `paddle_matrix.cpp`
 
 ```cpp
-#include "paddle/math/matrix.h"
+#include "paddle/legacy/math/matrix.h"
 extern "C"
 paddle_error paddle_matrix_shape(paddle_matrix matrix,
                                  uint64_t *width,

doc/v2/design/mkl/mkldnn.md

@@ -18,20 +18,20 @@ Figure 1. PaddlePaddle on IA
 具体的完成状态可以参见[这里](https://github.com/PaddlePaddle/Paddle/projects/21)。
 
 ## Contents
-
-- [Overview](#overview)
-- [Actions](#actions)
- 	- [CMake](#cmake)
- 	- [Matrix](#matrix)
-	- [Layers](#layers)
-	- [Activations](#activations)
-	- [Parameters](#parameters)
-	- [Gradients](#gradients)
-	- [Unit Tests](#unit-tests)
-	- [Python API](#python-api)
-	- [Benchmarking](#benchmarking)
-	- [Others](#others)
-- [Design Concerns](#design-concerns)
+
+- [Overview](#overview)
+- [Actions](#actions)
+ 	- [CMake](#cmake)
+ 	- [Matrix](#matrix)
+	- [Layers](#layers)
+	- [Activations](#activations)
+	- [Parameters](#parameters)
+	- [Gradients](#gradients)
+	- [Unit Tests](#unit-tests)
+	- [Python API](#python-api)
+	- [Benchmarking](#benchmarking)
+	- [Others](#others)
+- [Design Concerns](#design-concerns)
 
 ## Overview
 
@@ -218,20 +218,20 @@ if use_mkldnn
 我们总结出一些特别需要注意的点：
 
 1. 使用**deviceId_**。为了尽可能少的在父类Layer中添加变量或者函数，
-我们决定使用已有的`deviceId_`变量来区分layer的属性，定义`-2`为`MKLDNNLayer`特有的设备ID。
-2. 重写父类Layer的**init**函数，修改`deviceId_`为`-2`，代表这个layer是用于跑在MKL-DNN的环境下。
+我们决定使用已有的`deviceId_`变量来区分layer的属性，定义`-2`为`MKLDNNLayer`特有的设备ID。
+2. 重写父类Layer的**init**函数，修改`deviceId_`为`-2`，代表这个layer是用于跑在MKL-DNN的环境下。
 3. 创建`MKLDNNBase`，定义一些除了layer和memory相关的类和函数。
-包括MKL-DNN会用到`MKLDNNStream`和`CPUEngine`，和未来可能还会用到`FPGAEngine`等。
+包括MKL-DNN会用到`MKLDNNStream`和`CPUEngine`，和未来可能还会用到`FPGAEngine`等。
 4. 如果MKL-DNN layer的后面接有cpu device，那么就会使`output_.value`与`extOutVal_`共享内存，
 同时数据格式就是`NCHW`，这样下一个cpu device就能拿到正确的数据。
 在有普通的CPU layer时， `extOutVal_`和`extOutGrad_`的格式始终是`NCHW`或者`NC`。
 
 ## References
 1. [MKL small library](https://github.com/01org/mkl-dnn#linking-your-application)是[Intel MKL](https://software.intel.com/en-us/mkl)的一个子集。
-主要包括了深度学习相关的数学原语与操作，一般由MKL-DNN在发布[新版本](https://github.com/01org/mkl-dnn/releases)时一起更新。
+主要包括了深度学习相关的数学原语与操作，一般由MKL-DNN在发布[新版本](https://github.com/01org/mkl-dnn/releases)时一起更新。
 2. [MKL-DNN System Requirements](https://github.com/01org/mkl-dnn#system-requirements)。
 目前在PaddlePaddle中，仅会在支持AVX2指令集及以上的机器才使用MKL-DNN。
 3. [原来的方案](https://github.com/PaddlePaddle/Paddle/pull/3096)会引入**nextLayer**的信息。
-但是在PaddlePaddle中，无论是重构前的layer还是重构后的op，都不会想要知道next layer/op的信息。
+但是在PaddlePaddle中，无论是重构前的layer还是重构后的op，都不会想要知道next layer/op的信息。
 4. MKL-DNN的高性能格式与PaddlePaddle原有的`NCHW`不同(PaddlePaddle中的cuDNN部分使用的也是`NCHW`，所以不存在这个问题)。
-所以需要引入一个转换方法，并且只需要在必要的时候转换这种格式，才能更好的发挥MKL-DNN的性能。
+所以需要引入一个转换方法，并且只需要在必要的时候转换这种格式，才能更好的发挥MKL-DNN的性能。

doc/v2/dev/new_layer_cn.rst

@@ -58,7 +58,7 @@ PaddlePaddle的base layer类可以自动计算上面的导数。
 实现C++类
 ===================
 
-一个网络层的C++类需要实现初始化，前向和后向。全连接层的实现位于:code:`paddle/gserver/layers/FullyConnectedLayer.h`及:code:`paddle/gserver/layers/FullyConnectedLayer.cpp`。这里我们展示一份简化过的代码。
+一个网络层的C++类需要实现初始化，前向和后向。全连接层的实现位于:code:`paddle/legacy/gserver/layers/FullyConnectedLayer.h`及:code:`paddle/legacy/gserver/layers/FullyConnectedLayer.cpp`。这里我们展示一份简化过的代码。
 
 这个类需要继承 :code:`paddle::Layer` 这个基类，并且需要重写基类中的以下几个虚函数：
 
@@ -153,7 +153,7 @@ PaddlePaddle的base layer类可以自动计算上面的导数。
 
 - 每个层在其 :code:`forward` 函数的开头必须调用 :code:`Layer::forward(passType);` 。
 - 之后使用 :code:`reserveOutput(batchSize, size);` 为输出分配内存。由于我们支持训练数据有不同的批次大小，所以这一步是必要的。 :code:`reserveOutput`  会相应地改变输出的尺寸。为了保证效率，如果需要扩大矩阵，我们会重新分配内存；如果需要缩减矩阵，我们会继续使用现有的内存块。
-- 之后使用矩阵运算函数来计算 :math:`\sum_i W_i x + b`。:code:`getInput(i).value` 返回第i个输入矩阵。每个输入都是一个 :math:`batchSize \times dim` 的矩阵，每行表示一个批次中的单个输入。对于我们支持的全部矩阵操作，请参考 :code:`paddle/math/Matrix.h`和:code:`paddle/math/BaseMatrix.h` 。
+- 之后使用矩阵运算函数来计算 :math:`\sum_i W_i x + b`。:code:`getInput(i).value` 返回第i个输入矩阵。每个输入都是一个 :math:`batchSize \times dim` 的矩阵，每行表示一个批次中的单个输入。对于我们支持的全部矩阵操作，请参考 :code:`paddle/legacy/math/Matrix.h`和:code:`paddle/legacy/math/BaseMatrix.h` 。
 - 最终，使用 :code:`forwardActivation();` 进行激活操作。这会自动进行网络配置中声明的激活操作。
 
 
@@ -262,7 +262,7 @@ PaddlePaddle的base layer类可以自动计算上面的导数。
     REGISTER_LAYER(fc, FullyConnectedLayer);
     }
 
-若 :code:`cpp` 被放在 :code:`paddle/gserver/layers` 目录下，其会自动被加入编译列表。
+若 :code:`cpp` 被放在 :code:`paddle/legacy/gserver/layers` 目录下，其会自动被加入编译列表。
 
 
 写梯度检查单元测试
@@ -270,7 +270,7 @@ PaddlePaddle的base layer类可以自动计算上面的导数。
 
 写梯度检查单元测试是一个验证新实现的层是否正确的相对简单的办法。梯度检查单元测试通过有限差分法来验证一个层的梯度。首先对输入做一个小的扰动 :math:`\Delta x` ，然后观察到输出的变化为 :math:`\Delta y` ，那么，梯度就可以通过这个方程计算得到 :math:`\frac{\Delta y}{\Delta x }` 。之后，再用这个梯度去和 :code:`backward` 函数得到的梯度去对比，以保证梯度计算的正确性。需要注意的是梯度检查仅仅验证了梯度的计算，并不保证 :code:`forward` 和 :code:`backward` 函数的实现是正确的。你需要一些更复杂的单元测试来保证你实现的网络层是正确的。
 
-所有网络层的梯度检查单测都位于 :code:`paddle/gserver/tests/test_LayerGrad.cpp` 。我们建议你在写新网络层时把测试代码放入新的文件中。下面列出了全连接层的梯度检查单元测试。它包含以下几步：
+所有网络层的梯度检查单测都位于 :code:`paddle/legacy/gserver/tests/test_LayerGrad.cpp` 。我们建议你在写新网络层时把测试代码放入新的文件中。下面列出了全连接层的梯度检查单元测试。它包含以下几步：
 
 + 生成网络层配置。网络层配置包含以下几项：
    - 偏置参数的大小。（例子中是4096）
@@ -322,7 +322,7 @@ PaddlePaddle的base layer类可以自动计算上面的导数。
       }
     }
 
-如果你要为了测试而增加新的文件，例如 :code:`paddle/gserver/tests/testFCGrad.cpp` ，你需要把该文件加入 :code:`paddle/gserver/tests/CMakeLists.txt` 中。下面给出了一个例子。当你执行命令 :code:`make tests` 时，所有的单测都会被执行一次。注意，有些层可能需要高精度来保证梯度检查单测正确执行。你需要在配置cmake时将 :code:`WITH_DOUBLE` 设置为 `ON` 。
+如果你要为了测试而增加新的文件，例如 :code:`paddle/legacy/gserver/tests/testFCGrad.cpp` ，你需要把该文件加入 :code:`paddle/legacy/gserver/tests/CMakeLists.txt` 中。下面给出了一个例子。当你执行命令 :code:`make tests` 时，所有的单测都会被执行一次。注意，有些层可能需要高精度来保证梯度检查单测正确执行。你需要在配置cmake时将 :code:`WITH_DOUBLE` 设置为 `ON` 。
 
 .. code-block:: bash
 

doc/v2/dev/new_layer_en.rst

@@ -58,7 +58,7 @@ Finally we can use chain rule to calculate :math:`\frac{\partial z}{\partial x}`
 Implement C++ Class
 ===================
 
-The C++ class of the layer implements the initialization, forward, and backward part of the layer. The fully connected layer is at :code:`paddle/gserver/layers/FullyConnectedLayer.h` and :code:`paddle/gserver/layers/FullyConnectedLayer.cpp`. We list simplified version of the code below.
+The C++ class of the layer implements the initialization, forward, and backward part of the layer. The fully connected layer is at :code:`paddle/legacy/gserver/layers/FullyConnectedLayer.h` and :code:`paddle/legacy/gserver/layers/FullyConnectedLayer.cpp`. We list simplified version of the code below.
 
 It needs to derive the base class :code:`paddle::Layer`, and it needs to override the following functions:
 
@@ -154,7 +154,7 @@ The implementation of the forward part has the following steps.
 
 - Every layer must call :code:`Layer::forward(passType);` at the beginning of its :code:`forward` function.
 - Then it allocates memory for the output using :code:`reserveOutput(batchSize, size);`. This step is necessary because we support the batches to have different batch sizes. :code:`reserveOutput` will change the size of the output accordingly. For the sake of efficiency, we will allocate new memory if we want to expand the matrix, but we will reuse the existing memory block if we want to shrink the matrix.
-- Then it computes :math:`\sum_i W_i x + b` using Matrix operations. :code:`getInput(i).value` retrieve the matrix of the i-th input. Each input is a :math:`batchSize \times dim` matrix, where each row represents an single input in a batch. For a complete lists of supported matrix operations, please refer to :code:`paddle/math/Matrix.h` and :code:`paddle/math/BaseMatrix.h`.
+- Then it computes :math:`\sum_i W_i x + b` using Matrix operations. :code:`getInput(i).value` retrieve the matrix of the i-th input. Each input is a :math:`batchSize \times dim` matrix, where each row represents an single input in a batch. For a complete lists of supported matrix operations, please refer to :code:`paddle/legacy/math/Matrix.h` and :code:`paddle/legacy/math/BaseMatrix.h`.
 - Finally it applies the activation function using :code:`forwardActivation();`. It will automatically applies the corresponding activation function specifies in the network configuration.
 
 
@@ -263,7 +263,7 @@ Finally, you can use :code:`REGISTER_LAYER(fc, FullyConnectedLayer);` to registe
     REGISTER_LAYER(fc, FullyConnectedLayer);
     }
 
-If the :code:`cpp` file is put into :code:`paddle/gserver/layers`, it will be automatically added to the compilation list.
+If the :code:`cpp` file is put into :code:`paddle/legacy/gserver/layers`, it will be automatically added to the compilation list.
 
 
 Write Gradient Check Unit Test
@@ -271,7 +271,7 @@ Write Gradient Check Unit Test
 
 An easy way to verify the correctness of new layer's implementation is to write a gradient check unit test. Gradient check unit test utilizes finite difference method to verify the gradient of a layer. It modifies the input with a small perturbation :math:`\Delta x` and observes the changes of output :math:`\Delta y`, the gradient can be computed as :math:`\frac{\Delta y}{\Delta x }`. This gradient can be compared with the gradient computed by the :code:`backward` function of the layer to ensure the correctness of the gradient computation. Notice that the gradient check only tests the correctness of the gradient computation, it does not necessarily guarantee the correctness of the implementation of the :code:`forward` and :code:`backward` function. You need to write more sophisticated unit tests to make sure your layer is implemented correctly.
 
-All the gradient check unit tests are located in :code:`paddle/gserver/tests/test_LayerGrad.cpp`. You are recommended to put your test into a new test file if you are planning to write a new layer. The gradient test of the gradient check unit test of the fully connected layer is listed below. It has the following steps.
+All the gradient check unit tests are located in :code:`paddle/legacy/gserver/tests/test_LayerGrad.cpp`. You are recommended to put your test into a new test file if you are planning to write a new layer. The gradient test of the gradient check unit test of the fully connected layer is listed below. It has the following steps.
 
 + Create layer configuration. A layer configuration can include the following attributes:
    - size of the bias parameter. (4096 in our example)
@@ -323,7 +323,7 @@ All the gradient check unit tests are located in :code:`paddle/gserver/tests/tes
       }
     }
 
-If you are creating a new file for the test, such as :code:`paddle/gserver/tests/testFCGrad.cpp`, you need to add the file to :code:`paddle/gserver/tests/CMakeLists.txt`. An example is given below. All the unit tests will run when you execute the command :code:`make tests`. Notice that some layers might need high accuracy for the gradient check unit tests to work well. You need to configure :code:`WITH_DOUBLE` to `ON` when configuring cmake.
+If you are creating a new file for the test, such as :code:`paddle/legacy/gserver/tests/testFCGrad.cpp`, you need to add the file to :code:`paddle/legacy/gserver/tests/CMakeLists.txt`. An example is given below. All the unit tests will run when you execute the command :code:`make tests`. Notice that some layers might need high accuracy for the gradient check unit tests to work well. You need to configure :code:`WITH_DOUBLE` to `ON` when configuring cmake.
 
 .. code-block:: bash
 
@@ -339,7 +339,7 @@ If you are creating a new file for the test, such as :code:`paddle/gserver/tests
 Implement Python Wrapper
 ========================
 
-Implementing Python wrapper allows us to use the added layer in configuration files. All the Python wrappers are in file :code:`python/paddle/trainer/config_parser.py`. An example of the Python wrapper for fully connected layer is listed below. It has the following steps:
+Implementing Python wrapper allows us to use the added layer in configuration files. All the Python wrappers are in file :code:`python/paddle/legacy/trainer/config_parser.py`. An example of the Python wrapper for fully connected layer is listed below. It has the following steps:
 
 - Use :code:`@config_layer('fc')` at the decorator for all the Python wrapper class. :code:`fc` is the identifier of the layer.
 - Implements :code:`__init__` constructor function.

doc/v2/faq/build_and_install/index_cn.rst

@@ -213,3 +213,12 @@ virtualenv本身也是Python的一个包，可以用pip进行安装：
 保存并关闭文件。
 
 这样，每次打开终端时就会自动启动名为‘paddle’的Python环境了。
+
+10. 通过pip安装的PaddlePaddle在  :code:`import paddle.fluid` 报找不到 :code:`libmkldnn.so` 或 :code:`libmklml_intel.so`
+------------------------------------------------------------------------------------------
+出现这种问题的原因是在导入 :code:`paddle.fluid` 时需要加载 :code:`libmkldnn.so` 和 :code:`libmklml_intel.so`，
+但是系统没有找到该文件。一般通过pip安装PaddlePaddle时会将 :code:`libmkldnn.so` 和 :code:`libmklml_intel.so`
+拷贝到 :code:`/usr/local/lib` 路径下，所以解决办法是将该路径加到 :code:`LD_LIBRARY_PATH` 环境变量下，
+即： :code:`export LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH` 。
+
+**注意**：如果是在虚拟环境中安装PaddlePaddle， :code:`libmkldnn.so` 和 :code:`libmklml_intel.so` 可能不在 :code:`/usr/local/lib` 路径下。
\ No newline at end of file

doc/v2/faq/parameter/index_cn.rst

@@ -196,6 +196,6 @@ PaddlePaddle保存的模型参数文件内容由16字节头信息和网络参数
         obj="process",
         args={"src_dict_path": src_dict_path})
 
-完整源码可参考 `sequence_recurrent <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/gserver/tests/sequence_recurrent.py>`_ 示例。
+完整源码可参考 `sequence_recurrent <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/gserver/tests/sequence_recurrent.py>`_ 示例。
 
 

doc/v2/howto/capi/compile_paddle_lib_cn.md

@@ -18,7 +18,7 @@
 </tr>
 <tr>
 <td>cpu_avx_openblas</td>
-<td>暂无</td>
+<td><a href="https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxOpenblas/.lastSuccessful/paddle.tgz" rel="nofollow">paddle.tgz</a></td>
 </tr>
 <tr>
 <td>cpu_noavx_openblas</td>
@@ -35,7 +35,12 @@
 <tr>
 <td>cuda8.0_cudnn7_avx_mkl</td>
 <td><a href="https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddle.tgz" rel="nofollow">paddle.tgz</a></td>
-</tr></tbody></table>
+</tr>
+<tr>
+<td>cuda9.0_cudnn7_avx_mkl</td>
+<td><a href="https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda90cudnn7avxMkl/.lastSuccessful/paddle.tgz" rel="nofollow">paddle.tgz</a></td>
+</tr>
+</tbody></table>
 
 ### 从源码编译
 

doc/v2/howto/capi/compile_paddle_lib_en.md

@@ -17,7 +17,7 @@
 </tr>
 <tr>
 <td>cpu_avx_openblas</td>
-<td>-</td>
+<td><a href="https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxOpenblas/.lastSuccessful/paddle.tgz" rel="nofollow">paddle.tgz</a></td>
 </tr>
 <tr>
 <td>cpu_noavx_openblas</td>
@@ -34,7 +34,12 @@
 <tr>
 <td>cuda8.0_cudnn7_avx_mkl</td>
 <td><a href="https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddle.tgz" rel="nofollow">paddle.tgz</a></td>
-</tr></tbody></table>
+</tr>
+<tr>
+<td>cuda9.0_cudnn7_avx_mkl</td>
+<td><a href="https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda90cudnn7avxMkl/.lastSuccessful/paddle.tgz" rel="nofollow">paddle.tgz</a></td>
+</tr>
+</tbody></table>
 
 ### From source
 

doc/v2/howto/capi/workflow_of_capi_cn.md

@@ -28,9 +28,9 @@
 
 ### 准备预测模型
 
-准备预测模型部分，我们以手写数字识别任务为例进行介绍。手写数字识别任务定义了一个含有[两个隐层的简单全连接网络](https://github.com/PaddlePaddle/book/blob/develop/02.recognize_digits/README.cn.md#softmax回归softmax-regression)，网络接受一幅图片作为输入，将图片分类到 0 ~ 9 类别标签之一。完整代码可以查看[此目录](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/capi/examples/model_inference/dense) 中的相关脚本。
+准备预测模型部分，我们以手写数字识别任务为例进行介绍。手写数字识别任务定义了一个含有[两个隐层的简单全连接网络](https://github.com/PaddlePaddle/book/blob/develop/02.recognize_digits/README.cn.md#softmax回归softmax-regression)，网络接受一幅图片作为输入，将图片分类到 0 ~ 9 类别标签之一。完整代码可以查看[此目录](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/legacy/capi/examples/model_inference/dense) 中的相关脚本。
 
-调用C-API开发预测程序需要一个训练好的模型，运行[MNIST手写数字识别目录](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/capi/examples/model_inference/dense)下的[mnist_v2.py](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/capi/examples/model_inference/dense/mnist_v2.py)脚本，在终端执行`python mnist_v2.py`，会使用 PaddlePaddle 内置的 [MNIST 数据集](http://yann.lecun.com/exdb/mnist/)进行训练。训练好的模型默认保存在当前运行目录下的`models`目录中。
+调用C-API开发预测程序需要一个训练好的模型，运行[MNIST手写数字识别目录](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/legacy/capi/examples/model_inference/dense)下的[mnist_v2.py](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/capi/examples/model_inference/dense/mnist_v2.py)脚本，在终端执行`python mnist_v2.py`，会使用 PaddlePaddle 内置的 [MNIST 数据集](http://yann.lecun.com/exdb/mnist/)进行训练。训练好的模型默认保存在当前运行目录下的`models`目录中。
 
 下面，我们将训练结束后存储下来的模型转换成预测模型。
 
@@ -48,7 +48,7 @@
     dump_v2_config(predict, "trainer_config.bin", True)
     ```
 
-    对[手写数字识别](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/capi/examples/model_inference/dense)这个示例，[`mnist_v2.py`](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/capi/examples/model_inference/dense/mnist_v2.py)脚本集成了序列化神经网络结构的过程，可以直接运行 `python mnist_v2.py --task dump_config` 对神经网络结构进行序列化，结果会写入当前运行目录下的`trainer_config.bin`文件中。
+    对[手写数字识别](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/legacy/capi/examples/model_inference/dense)这个示例，[`mnist_v2.py`](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/legacy/capi/examples/model_inference/dense/mnist_v2.py)脚本集成了序列化神经网络结构的过程，可以直接运行 `python mnist_v2.py --task dump_config` 对神经网络结构进行序列化，结果会写入当前运行目录下的`trainer_config.bin`文件中。
 
     使用这种方式，需要**在运行时将神经网络的多个可学习参数放在同一个目录中**，C-API可以通过分别指定序列化后的网络结构文件和参数目录来加载训练好的模型。
 
@@ -68,7 +68,7 @@
     merge_v2_model(net, param_file, output_file)
     ```
 
-    对[手写数字识别](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/capi/examples/model_inference/dense)这个示例，可直接运行 `python` [merge_v2_model.py](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/capi/examples/model_inference/dense/merge_v2_model.py)。序列化结果会写入当前运行目录下的`output.paddle.model`文件中。使用这种方式，运行时C-API可以通过指定`output.paddle.model`文件的路径来加载预测模型。
+    对[手写数字识别](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/legacy/capi/examples/model_inference/dense)这个示例，可直接运行 `python` [merge_v2_model.py](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/legacy/capi/examples/model_inference/dense/merge_v2_model.py)。序列化结果会写入当前运行目录下的`output.paddle.model`文件中。使用这种方式，运行时C-API可以通过指定`output.paddle.model`文件的路径来加载预测模型。
 
 #### 注意事项
 1. 为使用C-API，在调用`dump_v2_config`序列化神经网络结构时，参数`binary`必须指定为`True`。
@@ -77,10 +77,10 @@
 
 ### 编写预测代码
 
-预测代码更多详细示例代码请参考[C-API使用示例](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/capi/examples/model_inference) 目录下的代码示例。这一节对图1中预测代码编写的5个步骤进行介绍和说明。
+预测代码更多详细示例代码请参考[C-API使用示例](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/legacy/capi/examples/model_inference) 目录下的代码示例。这一节对图1中预测代码编写的5个步骤进行介绍和说明。
 
 #### step 1. 初始化PaddlePaddle运行环境
-第一步需调用[`paddle_init`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/capi/main.h#L27) 初始化PaddlePaddle运行环境，该接口接受两个参数：参数的个数和参数列表。
+第一步需调用[`paddle_init`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/capi/main.h#L27) 初始化PaddlePaddle运行环境，该接口接受两个参数：参数的个数和参数列表。
 
 #### step2. 加载模型
 
@@ -88,8 +88,8 @@
 
 概念上，在 PaddlePaddle 内部，一个GradientMachine类的对象管理着一组计算层（PaddlePaddle Layers）来完成前向和反向计算，并处理与之相关的所有细节。在调用C-API预测时，只需进行前向计算而无需调用反向计算。这篇文档之后部分会使用`gradient machine`来特指调用PaddlePaddle C-API创建的GradientMachine类的对象。每一个 `gradient machine` 都会管理维护一份训练好的模型，下面是C-API提供的，两种常用的模型加载方式：
 
-1. 调用[`paddle_gradient_machine_load_parameter_from_disk`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/capi/gradient_machine.h#L61)接口，从磁盘加载预测模型。这时`gradient machine`会独立拥有一份训练好的模型；
-1. 调用[`paddle_gradient_machine_create_shared_param`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/capi/gradient_machine.h#L88)接口，与其它`gradient machine`的共享已经加载的预测模型。这种情况多出现在使用多线程预测时，通过多个线程共享同一个模型来减少内存开销。可参考[此示例](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/capi/examples/model_inference/multi_thread/main.c)。
+1. 调用[`paddle_gradient_machine_load_parameter_from_disk`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/capi/gradient_machine.h#L61)接口，从磁盘加载预测模型。这时`gradient machine`会独立拥有一份训练好的模型；
+1. 调用[`paddle_gradient_machine_create_shared_param`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/capi/gradient_machine.h#L88)接口，与其它`gradient machine`的共享已经加载的预测模型。这种情况多出现在使用多线程预测时，通过多个线程共享同一个模型来减少内存开销。可参考[此示例](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/capi/examples/model_inference/multi_thread/main.c)。
 
 - 注意事项
 
@@ -117,7 +117,7 @@ C-API支持的所有输入数据类型和他们的组织方式，请参考“输
 
 #### step 4. 前向计算
 
-完成上述准备之后，通过调用 [`paddle_gradient_machine_forward`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/capi/gradient_machine.h#L73) 接口完成神经网络的前向计算。
+完成上述准备之后，通过调用 [`paddle_gradient_machine_forward`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/capi/gradient_machine.h#L73) 接口完成神经网络的前向计算。
 
 #### step 5. 清理
 

doc/v2/howto/optimization/gpu_profiling_cn.rst

@@ -50,12 +50,12 @@ GPU则还需要高并行性，才能发挥其全部能力。这正是它们速
 **nvprof** 是Nvidia性能分析工具， **nvvp** 则是带GUI的Nvidia可视化性能分析工具。
 在这个教程中，我们主要会介绍nvprof和nvvp。
 
-:code:`test_GpuProfiler` from :code:`paddle/math/tests` directory will be used to evaluate
+:code:`test_GpuProfiler` from :code:`paddle/legacy/math/tests` directory will be used to evaluate
 above profilers.
 
-:code:`paddle/math/test` 目录中的 :code:`test_GpuProfiler` 就是用于展示上述分析工具的用法。
+:code:`paddle/legacy/math/test` 目录中的 :code:`test_GpuProfiler` 就是用于展示上述分析工具的用法。
 
-.. literalinclude:: ../../../../paddle/math/tests/test_GpuProfiler.cpp
+.. literalinclude:: ../../../../paddle/legacy/math/tests/test_GpuProfiler.cpp
    :language: c++
    :lines: 137-151
    :linenos:
@@ -83,7 +83,7 @@ program crashes when CPU version of PaddlePaddle invokes them.
 
 1. 加入 :code:`REGISTER_TIMER_INFO` 和 :code:`printAllStatus` 函数（如高亮部分）。
 
-    .. literalinclude:: ../../../../paddle/math/tests/test_GpuProfiler.cpp
+    .. literalinclude:: ../../../../paddle/legacy/math/tests/test_GpuProfiler.cpp
         :language: c++
         :lines: 137-151
         :emphasize-lines: 8-12,14
@@ -101,8 +101,8 @@ program crashes when CPU version of PaddlePaddle invokes them.
     .. code-block:: bash
         :emphasize-lines: 1,12-15
 
-        > ./paddle/math/tests/test_GpuProfiler
-        I1117 11:13:42.313065 2522362816 Util.cpp:155] commandline: ./paddle/math/tests/test_GpuProfiler
+        > ./paddle/legacy/math/tests/test_GpuProfiler
+        I1117 11:13:42.313065 2522362816 Util.cpp:155] commandline: ./paddle/legacy/math/tests/test_GpuProfiler
         I1117 11:13:42.845065 2522362816 Util.cpp:130] Calling runInitFunctions
         I1117 11:13:42.845208 2522362816 Util.cpp:143] Call runInitFunctions done.
         [==========] Running 1 test from 1 test case.
@@ -130,7 +130,7 @@ nvprof 工具
 
 1. 将 :code:`REGISTER_GPU_PROFILER` 函数加到代码中（参考强调部分）。
 
-    .. literalinclude:: ../../../../paddle/math/tests/test_GpuProfiler.cpp
+    .. literalinclude:: ../../../../paddle/legacy/math/tests/test_GpuProfiler.cpp
         :language: c++
         :lines: 137-151
         :emphasize-lines: 6-7
@@ -147,13 +147,13 @@ nvprof 工具
 
     .. code-block:: bash
 
-        nvprof  ./paddle/math/tests/test_GpuProfiler
+        nvprof  ./paddle/legacy/math/tests/test_GpuProfiler
 
 然后，您就能获得如下的分析结果：
 
 .. code-block:: bash
 
-    ==78544== Profiling application: ./paddle/math/tests/test_GpuProfiler
+    ==78544== Profiling application: ./paddle/legacy/math/tests/test_GpuProfiler
     ==78544== Profiling result:
     Time(%)     Time     Calls       Avg       Min       Max  Name
     27.60%  9.6305ms         5  1.9261ms  3.4560us  6.4035ms  [CUDA memcpy HtoD]

doc/v2/howto/optimization/gpu_profiling_en.rst

@@ -51,10 +51,10 @@ For general GPU profiling, a bunch of tools are provided from both NVIDIA and th
 **nvprof** is Nvidia profiler and **nvvp** is (GUI based) Nvidia visual profiler.
 In this tutorial, we will focus on nvprof and nvvp.
 
-:code:`test_GpuProfiler` from :code:`paddle/math/tests` directory will be used to evaluate
+:code:`test_GpuProfiler` from :code:`paddle/legacy/math/tests` directory will be used to evaluate
 above profilers.
 
-.. literalinclude:: ../../../../paddle/math/tests/test_GpuProfiler.cpp
+.. literalinclude:: ../../../../paddle/legacy/math/tests/test_GpuProfiler.cpp
    :language: c++
    :lines: 137-151
    :linenos:
@@ -80,7 +80,7 @@ As a simple example, consider the following:
 
 1. Add :code:`REGISTER_TIMER_INFO` and :code:`printAllStatus` functions (see the emphasize-lines).
 
-    .. literalinclude:: ../../../../paddle/math/tests/test_GpuProfiler.cpp
+    .. literalinclude:: ../../../../paddle/legacy/math/tests/test_GpuProfiler.cpp
         :language: c++
         :lines: 137-151
         :emphasize-lines: 8-12,14
@@ -98,8 +98,8 @@ As a simple example, consider the following:
     .. code-block:: bash
         :emphasize-lines: 1,12-15
 
-        > ./paddle/math/tests/test_GpuProfiler
-        I1117 11:13:42.313065 2522362816 Util.cpp:155] commandline: ./paddle/math/tests/test_GpuProfiler
+        > ./paddle/legacy/math/tests/test_GpuProfiler
+        I1117 11:13:42.313065 2522362816 Util.cpp:155] commandline: ./paddle/legacy/math/tests/test_GpuProfiler
         I1117 11:13:42.845065 2522362816 Util.cpp:130] Calling runInitFunctions
         I1117 11:13:42.845208 2522362816 Util.cpp:143] Call runInitFunctions done.
         [==========] Running 1 test from 1 test case.
@@ -127,7 +127,7 @@ To use this command line profiler **nvprof**, you can simply issue the following
 
 1. Add :code:`REGISTER_GPU_PROFILER` function (see the emphasize-lines).
 
-    .. literalinclude:: ../../../../paddle/math/tests/test_GpuProfiler.cpp
+    .. literalinclude:: ../../../../paddle/legacy/math/tests/test_GpuProfiler.cpp
         :language: c++
         :lines: 137-151
         :emphasize-lines: 6-7
@@ -144,13 +144,13 @@ To use this command line profiler **nvprof**, you can simply issue the following
 
     .. code-block:: bash
 
-        nvprof  ./paddle/math/tests/test_GpuProfiler
+        nvprof  ./paddle/legacy/math/tests/test_GpuProfiler
 
 Then, you can get the following profiling result:
 
 .. code-block:: bash
 
-    ==78544== Profiling application: ./paddle/math/tests/test_GpuProfiler
+    ==78544== Profiling application: ./paddle/legacy/math/tests/test_GpuProfiler
     ==78544== Profiling result:
     Time(%)     Time     Calls       Avg       Min       Max  Name
     27.60%  9.6305ms         5  1.9261ms  3.4560us  6.4035ms  [CUDA memcpy HtoD]

doc/v2/howto/rnn/hrnn_rnn_api_compare_cn.rst

@@ -4,7 +4,7 @@
 单双层RNN API对比介绍
 #####################
 
-本文以PaddlePaddle的双层RNN单元测试为示例，用多对效果完全相同的、分别使用单双层RNN作为网络配置的模型，来讲解如何使用双层RNN。本文中所有的例子，都只是介绍双层RNN的API接口，并不是使用双层RNN解决实际的问题。如果想要了解双层RNN在具体问题中的使用，请参考\ :ref:`algo_hrnn_demo`\ 。本文中示例所使用的单元测试文件是\ `test_RecurrentGradientMachine.cpp <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/gserver/tests/test_RecurrentGradientMachine.cpp>`_\ 。
+本文以PaddlePaddle的双层RNN单元测试为示例，用多对效果完全相同的、分别使用单双层RNN作为网络配置的模型，来讲解如何使用双层RNN。本文中所有的例子，都只是介绍双层RNN的API接口，并不是使用双层RNN解决实际的问题。如果想要了解双层RNN在具体问题中的使用，请参考\ :ref:`algo_hrnn_demo`\ 。本文中示例所使用的单元测试文件是\ `test_RecurrentGradientMachine.cpp <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/gserver/tests/test_RecurrentGradientMachine.cpp>`_\ 。
 
 示例1：双层RNN，子序列间无Memory
 ================================
@@ -13,8 +13,8 @@
 
 在本示例中，单层RNN和双层RNN的网络配置，都是将每一句分好词后的句子，使用LSTM作为encoder，压缩成一个向量。区别是RNN使用两层序列模型，将多句话看成一个整体同时使用encoder压缩。二者语意上完全一致。这组语义相同的示例配置如下：
 
-* 单层RNN\: `sequence_layer_group.conf <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/gserver/tests/sequence_layer_group.conf>`_
-* 双层RNN\: `sequence_nest_layer_group.conf <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/gserver/tests/sequence_nest_layer_group.conf>`_
+* 单层RNN\: `sequence_layer_group.conf <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/gserver/tests/sequence_layer_group.conf>`_
+* 双层RNN\: `sequence_nest_layer_group.conf <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/gserver/tests/sequence_nest_layer_group.conf>`_
 
 
 读取双层序列数据
@@ -24,18 +24,18 @@
 
 - 本例中的原始数据一共有10个样本。每个样本由两部分组成，一个label（此处都为2）和一个已经分词后的句子。这个数据也被单层RNN网络直接使用。
 
-..  literalinclude:: ../../../../paddle/gserver/tests/Sequence/tour_train_wdseg
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/Sequence/tour_train_wdseg
     :language: text
 
 
 - 双层序列数据一共有4个样本。 每个样本间用空行分开，整体数据和原始数据完全一样。但于双层序列的LSTM来说，第一个样本同时encode两条数据成两个向量。这四条数据同时处理的句子数量为\ :code:`[2, 3, 2, 3]`\ 。
 
-..  literalinclude:: ../../../../paddle/gserver/tests/Sequence/tour_train_wdseg.nest
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/Sequence/tour_train_wdseg.nest
     :language: text
 
-其次，对于两种不同的输入数据类型，不同DataProvider对比如下(`sequenceGen.py <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/gserver/tests/sequenceGen.py>`_)\：
+其次，对于两种不同的输入数据类型，不同DataProvider对比如下(`sequenceGen.py <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/gserver/tests/sequenceGen.py>`_)\：
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequenceGen.py
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequenceGen.py
     :language: python
     :lines: 21-39
     :linenos:
@@ -47,7 +47,7 @@
     - words是原始数据中的每一句话，所对应的词表index数组。它是integer_value_sequence类型的，即整数数组。words即为这个数据中的单层时间序列。
     - label是原始数据中对于每一句话的分类标签，它是integer_value类型的。
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequenceGen.py
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequenceGen.py
     :language: python
     :lines: 42-71
     :linenos:
@@ -64,7 +64,7 @@
 
 首先，我们看一下单层RNN的配置。代码中9-15行(高亮部分)即为单层RNN序列的使用代码。这里使用了PaddlePaddle预定义好的RNN处理函数。在这个函数中，RNN对于每一个时间步通过了一个LSTM网络。
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequence_layer_group.conf
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequence_layer_group.conf
     :language: python
     :lines: 38-63
     :linenos:
@@ -85,7 +85,7 @@
 
 * 至此，\ :code:`lstm_last`\ 便和单层RNN配置中的\ :code:`lstm_last`\ 具有相同的结果了。
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequence_nest_layer_group.conf
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequence_nest_layer_group.conf
     :language: python
     :lines: 38-64
     :linenos:
@@ -107,7 +107,7 @@
 
 - 单层RNN：过了一个很简单的recurrent_group。每一个时间步，当前的输入y和上一个时间步的输出rnn_state做了一个全链接。
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequence_rnn.conf
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequence_rnn.conf
     :language: python
     :lines: 36-48
 
@@ -116,7 +116,7 @@
   - 内层inner_step的recurrent_group和单层序列的几乎一样。除了boot_layer=outer_mem，表示将外层的outer_mem作为内层memory的初始状态。外层outer_step中，outer_mem是一个子句的最后一个向量，即整个双层group是将前一个子句的最后一个向量，作为下一个子句memory的初始状态。
   - 从输入数据上看，单双层序列的句子是一样的，只是双层序列将其又做了子序列划分。因此双层序列的配置中，必须将前一个子句的最后一个元素，作为boot_layer传给下一个子句的memory，才能保证和单层序列的配置中“每个时间步都用了上一个时间步的输出结果”一致。
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequence_nest_rnn.conf
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequence_nest_rnn.conf
     :language: python
     :lines: 39-66
 
@@ -134,7 +134,7 @@
 
 **输入不等长** 是指recurrent_group的多个输入序列，在每个时间步的子序列长度可以不相等。但序列输出时，需要指定与某一个输入的序列信息是一致的。使用\ :red:`targetInlink`\ 可以指定哪一个输入和输出序列信息一致，默认指定第一个输入。 
 
-示例3的配置分别为\ `单层不等长RNN <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/gserver/tests/sequence_rnn_multi_unequalength_inputs.py>`_\ 和\ `双层不等长RNN <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/gserver/tests/sequence_nest_rnn_multi_unequalength_inputs.py>`_\ 。
+示例3的配置分别为\ `单层不等长RNN <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/gserver/tests/sequence_rnn_multi_unequalength_inputs.py>`_\ 和\ `双层不等长RNN <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/gserver/tests/sequence_nest_rnn_multi_unequalength_inputs.py>`_\ 。
 
 示例3对于单层RNN和双层RNN数据完全相同。
 
@@ -152,14 +152,14 @@
 
 * 单层RNN\:
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequence_rnn_multi_unequalength_inputs.py
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequence_rnn_multi_unequalength_inputs.py
     :language: python
     :lines: 42-59
     :linenos:
 
 * 双层RNN\ \:
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequence_nest_rnn_multi_unequalength_inputs.py
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequence_nest_rnn_multi_unequalength_inputs.py
     :language: python
     :lines: 41-80
     :linenos:

doc/v2/howto/rnn/hrnn_rnn_api_compare_en.rst

@@ -4,7 +4,7 @@
 API comparision between RNN and hierarchical RNN
 #####################
 
-This article takes PaddlePaddle's hierarchical RNN unit test as an example. We will use several examples to illestrate the usage of single-layer and hierarchical RNNs. Each example has two model configurations, one for single-layer, and the other for hierarchical RNN. Although the implementations are different, both the two model configurations' effects are the same. All of the examples in this article only describe the API interface of the hierarchical RNN, while we do not use this hierarchical RNN to solve practical problems. If you want to understand the use of hierarchical RNN in specific issues, please refer to \ :ref:`algo_hrnn_demo`\ 。The unit test file used in this article's example is \ `test_RecurrentGradientMachine.cpp <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/gserver/tests/test_RecurrentGradientMachine.cpp>`_\ 。
+This article takes PaddlePaddle's hierarchical RNN unit test as an example. We will use several examples to illestrate the usage of single-layer and hierarchical RNNs. Each example has two model configurations, one for single-layer, and the other for hierarchical RNN. Although the implementations are different, both the two model configurations' effects are the same. All of the examples in this article only describe the API interface of the hierarchical RNN, while we do not use this hierarchical RNN to solve practical problems. If you want to understand the use of hierarchical RNN in specific issues, please refer to \ :ref:`algo_hrnn_demo`\ 。The unit test file used in this article's example is \ `test_RecurrentGradientMachine.cpp <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/gserver/tests/test_RecurrentGradientMachine.cpp>`_\ 。
 
 Example 1：Hierarchical RNN without Memory between subsequences
 ================================
@@ -13,8 +13,8 @@ The classical case in the hierarchical RNN is to perform sequence operations on
 
 In this example, the network configuration of single-layer RNNs and hierarchical RNNs are all to use LSTM as en encoder to compress a word-segmented sentence into a vector. The difference is that, RNN uses a hierarchical RNN model, treating multiple sentences as a whole to use encoder to compress simultaneously. They are completely consistent in their semantic meanings. This pair of semantically identical example configurations is as follows：
 
-* RNN\: `sequence_layer_group.conf <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/gserver/tests/sequence_layer_group.conf>`_
-* Hierarchical RNN\: `sequence_nest_layer_group.conf <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/gserver/tests/sequence_nest_layer_group.conf>`_
+* RNN\: `sequence_layer_group.conf <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/gserver/tests/sequence_layer_group.conf>`_
+* Hierarchical RNN\: `sequence_nest_layer_group.conf <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/gserver/tests/sequence_nest_layer_group.conf>`_
 
 
 Reading hierarchical sequence data
@@ -24,18 +24,18 @@ Firstly, the original data in this example is as follows \:
 
 - The original data in this example has 10 samples. Each of the sample includes two components: a lable(all 2 here), and a word-segmented sentence. This data is used by single RNN as well. 
 
-..  literalinclude:: ../../../../paddle/gserver/tests/Sequence/tour_train_wdseg
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/Sequence/tour_train_wdseg
     :language: text
 
 
 - The data for hierarchical RNN has 4 samples. Every sample is seperated by a blank line, while the content of the data is the same as the original data. But as for hierarchical LSTM, the first sample will encode two sentences into two vectors simultaneously. The sentence count dealed simultaneously by this 4 samples are \ :code:`[2, 3, 2, 3]`\ .
 
-..  literalinclude:: ../../../../paddle/gserver/tests/Sequence/tour_train_wdseg.nest
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/Sequence/tour_train_wdseg.nest
     :language: text
 
-Secondly, as for these two types of different input data formats, the contrast of different DataProviders are as follows (`sequenceGen.py <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/gserver/tests/sequenceGen.py>`_)\：
+Secondly, as for these two types of different input data formats, the contrast of different DataProviders are as follows (`sequenceGen.py <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/gserver/tests/sequenceGen.py>`_)\：
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequenceGen.py
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequenceGen.py
     :language: python
     :lines: 21-39
     :linenos:
@@ -47,7 +47,7 @@ Secondly, as for these two types of different input data formats, the contrast o
     - "words" is a list of word table indices corresponding to each word in the sentence in the original data. Its data type is integer_value_sequence, that is integer list. So, "words" is a singler-layer time series in the data. 
     - "label" is the categorical label of each sentence, whose data type is integer_value. 
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequenceGen.py
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequenceGen.py
     :language: python
     :lines: 42-71
     :linenos:
@@ -64,7 +64,7 @@ Model configuration
 
 Firstly, let's look at the configuration of single-layer RNN. The hightlighted part of line 9 to line 15 is the usage of single-layer RNN. Here we use the pre-defined RNN process function in PaddlePaddle. In this function, for each time step, RNN passes through an LSTM network. 
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequence_layer_group.conf
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequence_layer_group.conf
     :language: python
     :lines: 38-63
     :linenos:
@@ -85,7 +85,7 @@ Secondly, let's look at the model configuration of hierarchical RNN which has th
 
 * Till now, \ :code:`lstm_last`\ has the same result as \ :code:`lstm_last`\ in single-layer RNN configuration. 
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequence_nest_layer_group.conf
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequence_nest_layer_group.conf
     :language: python
     :lines: 38-64
     :linenos:
@@ -107,7 +107,7 @@ We select the different parts between single-layer RNN and hierarchical RNN conf
 
 - single-layer RNN：passes through a simple recurrent_group. For each time step, the current input y and the last time step's output rnn_state pass through a fully-connected layer. 
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequence_rnn.conf
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequence_rnn.conf
     :language: python
     :lines: 36-48
 
@@ -116,7 +116,7 @@ We select the different parts between single-layer RNN and hierarchical RNN conf
   - The recurrent_group of inner layer's inner_step is nearly the same as single-layer sequence, except for the case of boot_layer=outer_mem, which means using the outer layer's outer_mem as the initial state for the inner layer's memory. In the outer layer's out_step, outer_mem is the last vector of a subsequence, that is, the whole hierarchical group uses the last vector of the previous subsequence as the initial state for the next subsequence's memory. 
   - From the aspect of the input data, sentences from single-layer and hierarchical RNN are the same. The only difference is that, hierarchical RNN disassembes the sequence into subsequences. So in the hierarchical RNN configuration, we must use the last element of the previous subsequence as a boot_layer for the memory of the next subsequence, so that it makes no difference with "every time step uses the output of last time step" in the sigle-layer RNN configuration. 
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequence_nest_rnn.conf
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequence_nest_rnn.conf
     :language: python
     :lines: 39-66
 
@@ -134,7 +134,7 @@ Example 3：hierarchical RNN with unequal length inputs
 
 **unequal length inputs** means in the multiple input sequences of recurrent_group, the lengths of subsequences can be unequal. But the output of the sequence, needs to be consistent with one of the input sequences. Using \ :red:`targetInlink`\ can help you specify which of the input sequences and the output sequence can be consistent, by default is the first input. 
 
-The configurations of Example 3 are \ `sequence_rnn_multi_unequalength_inputs <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/gserver/tests/sequence_rnn_multi_unequalength_inputs.py>`_ \ and \ `sequence_nest_rnn_multi_unequalength_inputs <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/gserver/tests/sequence_nest_rnn_multi_unequalength_inputs.py>`_\ . 
+The configurations of Example 3 are \ `sequence_rnn_multi_unequalength_inputs <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/gserver/tests/sequence_rnn_multi_unequalength_inputs.py>`_ \ and \ `sequence_nest_rnn_multi_unequalength_inputs <https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/legacy/gserver/tests/sequence_nest_rnn_multi_unequalength_inputs.py>`_\ .
 
 The data for the configurations of Example 3's single-layer RNN and hierarchical RNN are exactly the same. 
 
@@ -152,14 +152,14 @@ Similar to Example 2's configuration, Example 3's configuration uses single-laye
 
 * single-layer RNN\:
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequence_rnn_multi_unequalength_inputs.py
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequence_rnn_multi_unequalength_inputs.py
     :language: python
     :lines: 42-59
     :linenos:
 
 * hierarchical RNN\ \:
 
-..  literalinclude:: ../../../../paddle/gserver/tests/sequence_nest_rnn_multi_unequalength_inputs.py
+..  literalinclude:: ../../../../paddle/legacy/gserver/tests/sequence_nest_rnn_multi_unequalength_inputs.py
     :language: python
     :lines: 41-80
     :linenos:

go/pserver/optimizer.go

@@ -16,7 +16,7 @@ package pserver
 
 // #cgo CFLAGS: -I ../../
 // #cgo LDFLAGS: ${SRCDIR}/client/c/libpaddle_go_optimizer.a -lstdc++ -lm
-// #include "paddle/optimizer/optimizer.h"
+// #include "paddle/legacy/optimizer/optimizer.h"
 // #include <stdlib.h>
 // #include <string.h>
 import "C"

paddle/CMakeLists.txt

 if(NOT WITH_FLUID_ONLY)
-  add_subdirectory(cuda)
-  add_subdirectory(function)
-  add_subdirectory(utils)
-  add_subdirectory(math)
-  add_subdirectory(gserver)
-  add_subdirectory(parameter)
+  add_subdirectory(legacy/cuda)
+  add_subdirectory(legacy/function)
+  add_subdirectory(legacy/utils)
+  add_subdirectory(legacy/math)
+  add_subdirectory(legacy/gserver)
+  add_subdirectory(legacy/parameter)
 
   if(MOBILE_INFERENCE)
-    add_subdirectory(capi)
+    add_subdirectory(legacy/capi)
   else()
-    add_subdirectory(pserver)
-    add_subdirectory(trainer)
+    add_subdirectory(legacy/pserver)
+    add_subdirectory(legacy/trainer)
     add_subdirectory(scripts)
 
     if(WITH_C_API)
-      add_subdirectory(capi)
+      add_subdirectory(legacy/capi)
     endif()
 
     if(WITH_SWIG_PY)
-      add_subdirectory(api)
+      add_subdirectory(legacy/api)
     endif()
   endif()
 endif()

paddle/contrib/CMakeLists.txt

@@ -14,4 +14,3 @@
 #
 
 add_subdirectory(inference)
-add_subdirectory(tape)