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

test=develop

Author: jacquesqiao

.gitignore

@@ -28,3 +28,4 @@ third_party/
 build_*
 # clion workspace.
 cmake-build-*
+model_test

CMakeLists.txt

@@ -62,13 +62,12 @@ option(WITH_DISTRIBUTE  "Compile with distributed support"              OFF)
 option(USE_EIGEN_FOR_BLAS   "Use matrix multiplication in Eigen"        OFF)
 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_INFERENCE    "Compile fluid inference library"              ON)
+option(ON_INFER         "Turn on inference optimization."               OFF)
 option(WITH_INFERENCE_API_TEST   "Test fluid inference high-level api interface"  OFF)
 option(WITH_SYSTEM_BLAS   "Use system blas library"           OFF)
 option(PY_VERSION       "Compile PaddlePaddle with python3 support"     ${PY_VERSION})
@@ -127,6 +126,9 @@ set(THIRD_PARTY_PATH "${CMAKE_BINARY_DIR}/third_party" CACHE STRING
 set(FLUID_INSTALL_DIR "${CMAKE_BINARY_DIR}/fluid_install_dir" CACHE STRING
   "A path setting fluid shared and static libraries")
 
+set(FLUID_INFERENCE_INSTALL_DIR "${CMAKE_BINARY_DIR}/fluid_inference_install_dir" CACHE STRING
+  "A path setting fluid inference shared and static libraries")
+
 if (WITH_C_API AND WITH_PYTHON)
   message(WARNING "It is suggest not embedded a python interpreter in Paddle "
     "when using C-API. It will give an unpredictable behavior when using a "
@@ -176,6 +178,7 @@ include(external/eigen)     # download eigen3
 include(external/pybind11)  # download pybind11
 include(external/cares)
 include(external/cub)
+include(external/xxhash)    # download xxhash
 
 if (NOT WIN32)
 # there is no official support of snappystream, warpctc, nccl, cupti in windows
@@ -298,3 +301,11 @@ if(WITH_DOC)
     find_python_module(recommonmark REQUIRED)
     add_subdirectory(doc)
 endif()
+
+if (ON_INFER)
+    message(STATUS "On inference mode, will take place some specific optimization.")
+    add_definitions(-DPADDLE_ON_INFERENCE)
+else()
+    #TODO(luotao), combine this warning with `make inference_lib_dist` command.
+    message(WARNING "On inference mode, will take place some specific optimization. Turn on the ON_INFER flag when building inference_lib only.")
+endif()

Dockerfile

@@ -75,14 +75,14 @@ RUN pip3 install -U wheel && \
     pip3 install -U docopt PyYAML sphinx==1.5.6 && \
     pip3 install sphinx-rtd-theme==0.1.9 recommonmark && \
     easy_install -U pip && \
-    pip install -U wheel && \
+    pip install -U pip setuptools wheel && \
     pip install -U docopt PyYAML sphinx==1.5.6 && \
     pip install sphinx-rtd-theme==0.1.9 recommonmark
 
-RUN pip3 install pre-commit 'ipython==5.3.0' && \
+RUN pip3 install 'pre-commit==1.10.4' 'ipython==5.3.0' && \
     pip3 install 'ipykernel==4.6.0' 'jupyter==1.0.0' && \
     pip3 install opencv-python && \
-    pip install pre-commit 'ipython==5.3.0' && \
+    pip install 'pre-commit==1.10.4' 'ipython==5.3.0' && \
     pip install 'ipykernel==4.6.0' 'jupyter==1.0.0' && \
     pip install opencv-python
 

README.md

@@ -2,8 +2,8 @@
 
 
 [![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)
+[![Documentation Status](https://img.shields.io/badge/docs-latest-brightgreen.svg?style=flat)](http://paddlepaddle.org/documentation/docs/en/1.1/getstarted/index_en.html)
+[![Documentation Status](https://img.shields.io/badge/中文文档-最新-brightgreen.svg)](http://paddlepaddle.org/documentation/docs/zh/1.1/beginners_guide/index.html)
 [![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)
 
@@ -19,17 +19,17 @@ 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.
 
 
-### Latest PaddlePaddle Release: [Fluid 0.15.0](https://github.com/PaddlePaddle/Paddle/tree/v0.15.0)
+### Latest PaddlePaddle Release: [Fluid 1.1.0](https://github.com/PaddlePaddle/Paddle/tree/release/1.1)
 ### Install Latest Stable Release:
 ```
 # Linux CPU
 pip install paddlepaddle
 # Linux GPU cuda9cudnn7
 pip install paddlepaddle-gpu
 # Linux GPU cuda8cudnn7
-pip install paddlepaddle-gpu==0.15.0.post87
+pip install paddlepaddle-gpu==1.1.0.post87
 # Linux GPU cuda8cudnn5
-pip install paddlepaddle-gpu==0.15.0.post85
+pip install paddlepaddle-gpu==1.1.0.post85
 
 # For installation on other platform, refer to http://paddlepaddle.org/
 ```
@@ -76,26 +76,26 @@ pip install paddlepaddle-gpu==0.15.0.post85
 
 ## Installation
 
-It is recommended to read [this doc](http://paddlepaddle.org/documentation/docs/zh/0.15.0/new_docs/beginners_guide/install/install_doc.html) on our website.
+It is recommended to read [this doc](http://paddlepaddle.org/documentation/docs/zh/1.1/beginners_guide/index.html) on our website.
 
 ## Documentation
 
-We provide [English](http://paddlepaddle.org/documentation/docs/en/0.15.0/getstarted/index_en.html) and
-[Chinese](http://paddlepaddle.org/documentation/docs/zh/0.15.0/new_docs/beginners_guide/index.html) documentation.
+We provide [English](http://paddlepaddle.org/documentation/docs/en/1.1/getstarted/index_en.html) and
+[Chinese](http://paddlepaddle.org/documentation/docs/zh/1.1/beginners_guide/index.html) documentation.
 
 - [Deep Learning 101](https://github.com/PaddlePaddle/book)
 
   You might want to start from this online interactive book that can run in a Jupyter Notebook.
 
-- [Distributed Training](http://paddlepaddle.org/documentation/docs/zh/0.15.0/new_docs/user_guides/howto/training/cluster_howto.html)
+- [Distributed Training](http://paddlepaddle.org/documentation/docs/zh/1.1/user_guides/howto/training/cluster_howto.html)
 
   You can run distributed training jobs on MPI clusters.
 
-- [Python API](http://paddlepaddle.org/documentation/api/zh/0.15.0/fluid.html)
+- [Python API](http://paddlepaddle.org/documentation/api/zh/1.1/fluid.html)
 
    Our new API enables much shorter programs.
 
-- [How to Contribute](http://paddlepaddle.org/documentation/docs/zh/0.15.0/new_docs/advanced_usage/development/contribute_to_paddle.html)
+- [How to Contribute](http://paddlepaddle.org/documentation/docs/zh/1.1/advanced_usage/development/contribute_to_paddle.html)
 
    We appreciate your contributions!
 

benchmark/fluid/args.py

@@ -142,5 +142,10 @@ def parse_args():
         choices=['reduce', 'all_reduce'],
         default='all_reduce',
         help='Specify the reduce strategy, can be reduce, all_reduce')
+    parser.add_argument(
+        '--fuse_broadcast_op',
+        action='store_true',
+        help='If set, would fuse multiple broadcast operators into one fused_broadcast operator.'
+    )
     args = parser.parse_args()
     return args

benchmark/fluid/fluid_benchmark.py

@@ -177,6 +177,7 @@ def train_parallel(train_args, test_args, args, train_prog, test_prog,
     else:
         build_strategy.reduce_strategy = fluid.BuildStrategy(
         ).ReduceStrategy.AllReduce
+    build_strategy.fuse_broadcast_op = args.fuse_broadcast_op
 
     avg_loss = train_args[0]
 
@@ -240,7 +241,6 @@ def train_parallel(train_args, test_args, args, train_prog, test_prog,
 
             if args.use_fake_data or args.use_reader_op:
                 try:
-
                     fetch_ret = exe.run(fetch_list)
                 except fluid.core.EOFException as eof:
                     break

benchmark/fluid/run.sh

@@ -50,11 +50,7 @@ if(NOT WITH_PROFILER)
 endif(NOT WITH_PROFILER)
 
 if(NOT CMAKE_CROSSCOMPILING)
-    if(WITH_AVX AND AVX512F_FOUND)
-        set(SIMD_FLAG ${AVX512F_FLAG})
-    elseif(WITH_AVX AND AVX2_FOUND)
-        set(SIMD_FLAG ${AVX2_FLAG})
-    elseif(WITH_AVX AND AVX_FOUND)
+    if(WITH_AVX AND AVX_FOUND)
         set(SIMD_FLAG ${AVX_FLAG})
     elseif(SSE3_FOUND)
         set(SIMD_FLAG ${SSE3_FLAG})

cmake/configure.cmake

@@ -0,0 +1,50 @@
+INCLUDE(ExternalProject)
+
+set(XXHASH_SOURCE_DIR ${THIRD_PARTY_PATH}/xxhash)
+set(XXHASH_INSTALL_DIR ${THIRD_PARTY_PATH}/install/xxhash)
+set(XXHASH_INCLUDE_DIR "${XXHASH_INSTALL_DIR}/include")
+
+IF(WITH_STATIC_LIB)
+  SET(BUILD_CMD make lib)
+ELSE()
+  IF(APPLE)
+    SET(BUILD_CMD sed -i \"\" "s/-Wstrict-prototypes -Wundef/-Wstrict-prototypes -Wundef -fPIC/g" ${XXHASH_SOURCE_DIR}/src/extern_xxhash/Makefile && make lib)
+  ELSE(APPLE)
+    SET(BUILD_CMD sed -i "s/-Wstrict-prototypes -Wundef/-Wstrict-prototypes -Wundef -fPIC/g" ${XXHASH_SOURCE_DIR}/src/extern_xxhash/Makefile && make lib)
+  ENDIF(APPLE)
+ENDIF()
+
+ExternalProject_Add(
+    extern_xxhash
+    ${EXTERNAL_PROJECT_LOG_ARGS}
+    GIT_REPOSITORY  "https://github.com/Cyan4973/xxHash"
+    GIT_TAG         "v0.6.5"
+    PREFIX          ${XXHASH_SOURCE_DIR}
+    DOWNLOAD_NAME   "xxhash"
+    UPDATE_COMMAND  ""
+    CONFIGURE_COMMAND ""
+    BUILD_IN_SOURCE 1
+    PATCH_COMMAND
+    BUILD_COMMAND     ${BUILD_CMD}
+    INSTALL_COMMAND   export PREFIX=${XXHASH_INSTALL_DIR}/ && make install
+    TEST_COMMAND      ""
+)
+
+set(XXHASH_LIBRARIES "${XXHASH_INSTALL_DIR}/lib/libxxhash.a")
+INCLUDE_DIRECTORIES(${XXHASH_INCLUDE_DIR})
+
+add_library(xxhash STATIC IMPORTED GLOBAL)
+set_property(TARGET xxhash PROPERTY IMPORTED_LOCATION ${XXHASH_LIBRARIES})
+include_directories(${XXHASH_INCLUDE_DIR})
+add_dependencies(xxhash extern_xxhash)
+
+LIST(APPEND external_project_dependencies xxhash)
+
+IF(WITH_C_API)
+  INSTALL(DIRECTORY ${XXHASH_INCLUDE_DIR} DESTINATION third_party/xxhash)
+  IF(ANDROID)
+    INSTALL(FILES ${XXHASH_LIBRARIES} DESTINATION third_party/xxhash/lib/${ANDROID_ABI})
+  ELSE()
+    INSTALL(FILES ${XXHASH_LIBRARIES} DESTINATION third_party/xxhash/lib)
+  ENDIF()
+ENDIF()

cmake/external/xxhash.cmake

@@ -261,6 +261,13 @@ function(cc_library TARGET_NAME)
         add_dependencies(${TARGET_NAME} mklml)
         target_link_libraries(${TARGET_NAME} "-L${MKLML_LIB_DIR} -liomp5 -Wl,--as-needed")
       endif()
+      # remove link to python, see notes at:
+      # https://github.com/pybind/pybind11/blob/master/docs/compiling.rst#building-manually
+      if("${cc_library_DEPS};" MATCHES "python;")
+        list(REMOVE_ITEM cc_library_DEPS python)
+        add_dependencies(${TARGET_NAME} python)
+        target_link_libraries(${TARGET_NAME} "-Wl,-undefined,dynamic_lookup")
+      endif()
       target_link_libraries(${TARGET_NAME} ${cc_library_DEPS})
       add_dependencies(${TARGET_NAME} ${cc_library_DEPS})
     endif()
@@ -311,6 +318,8 @@ function(cc_test TARGET_NAME)
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_cpu_deterministic=true)
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_init_allocated_mem=true)
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_cudnn_deterministic=true)
+    # No unit test should exceed 10 minutes.
+    set_tests_properties(${TARGET_NAME} PROPERTIES TIMEOUT 600)
   endif()
 endfunction(cc_test)
 
@@ -629,6 +638,8 @@ function(py_test TARGET_NAME)
              PYTHONPATH=${PADDLE_BINARY_DIR}/python ${py_test_ENVS}
              ${PYTHON_EXECUTABLE} -u ${py_test_SRCS} ${py_test_ARGS}
              WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})
+    # No unit test should exceed 10 minutes.
+    set_tests_properties(${TARGET_NAME} PROPERTIES TIMEOUT 600)
   endif()
 endfunction()