Merge branch 'develop' into softmax_with_cross_entropy_op

.gitignore

@@ -27,3 +27,4 @@ CMakeFiles
 cmake_install.cmake
 paddle/.timestamp
 python/paddlepaddle.egg-info/
+paddle/pybind/pybind.h

.travis.yml

@@ -36,10 +36,6 @@ before_install:
   # protobuf version.
   - sudo pip install -r $TRAVIS_BUILD_DIR/python/requirements.txt
   - sudo pip install wheel sphinx==1.5.6 recommonmark sphinx-rtd-theme==0.1.9 virtualenv pre-commit LinkChecker
-  - curl https://glide.sh/get | bash
-  - eval "$(GIMME_GO_VERSION=1.8.3 gimme)"
-  - go get -u github.com/alecthomas/gometalinter
-  - gometalinter --install
   - |
     function timeout() { perl -e 'alarm shift; exec @ARGV' "$@"; }
 script:

CMakeLists.txt

@@ -27,7 +27,7 @@ if(NOT CMAKE_CROSSCOMPILING)
 endif(NOT CMAKE_CROSSCOMPILING)
 find_package(Git REQUIRED)
 find_package(Threads REQUIRED)
-if(NOT ANDROID)
+if(NOT ANDROID AND NOT IOS)
     find_package(Boost QUIET)
 endif()
 
@@ -64,27 +64,29 @@ if(NOT CMAKE_BUILD_TYPE)
       FORCE)
 endif()
 
-if(ANDROID)
-    if(${CMAKE_SYSTEM_VERSION} VERSION_LESS "16")
-        message(FATAL_ERROR "Unsupport standalone toolchains with Android API level lower than 16")
-    elseif(${CMAKE_SYSTEM_VERSION} VERSION_LESS "21")
-        # TODO: support glog for Android api 16 ~ 19 in the future
-        message(WARNING "Using the unofficial git repository <https://github.com/Xreki/glog.git> instead")
+if(ANDROID OR IOS)
+    if(ANDROID)
+        if(AND ${CMAKE_SYSTEM_VERSION} VERSION_LESS "16")
+            message(FATAL_ERROR "Unsupport standalone toolchains with Android API level lower than 16")
+        elseif(${CMAKE_SYSTEM_VERSION} VERSION_LESS "21")
+            # TODO: support glog for Android api 16 ~ 19 in the future
+            message(WARNING "Using the unofficial git repository <https://github.com/Xreki/glog.git> instead")
+        endif()
     endif()
 
     set(WITH_GPU OFF CACHE STRING
-        "Disable GPU when cross-compiling for Android" FORCE)
+        "Disable GPU when cross-compiling for Android and iOS" FORCE)
     set(WITH_AVX OFF CACHE STRING
-        "Disable AVX when cross-compiling for Android" FORCE)
+        "Disable AVX when cross-compiling for Android and iOS" FORCE)
     set(WITH_PYTHON OFF CACHE STRING
-        "Disable PYTHON when cross-compiling for Android" FORCE)
+        "Disable PYTHON when cross-compiling for Android and iOS" FORCE)
     set(WITH_RDMA OFF CACHE STRING
-        "Disable RDMA when cross-compiling for Android" FORCE)
+        "Disable RDMA when cross-compiling for Android and iOS" FORCE)
     set(WITH_MKLDNN OFF CACHE STRING
-        "Disable MKLDNN when cross-compiling for Android" FORCE)
+        "Disable MKLDNN when cross-compiling for Android and iOS" FORCE)
     set(WITH_MKLML OFF CACHE STRING
-        "Disable MKLML package when cross-compiling for Android" FORCE)
-endif(ANDROID)
+        "Disable MKLML package when cross-compiling for Android and iOS" FORCE)
+endif()
 
 set(THIRD_PARTY_PATH "${CMAKE_BINARY_DIR}/third_party" CACHE STRING
   "A path setting third party libraries download & build directories.")

cmake/cblas.cmake

@@ -171,3 +171,10 @@ if (REFERENCE_CBLAS_INCLUDE_DIR AND REFERENCE_CBLAS_LIBRARY)
   add_definitions(-DPADDLE_USE_REFERENCE_CBLAS)
   message(STATUS "Found reference-cblas (include: ${CBLAS_INC_DIR}, library: ${CBLAS_LIBRARIES})")
 endif()
+
+if(IOS_USE_VECLIB_FOR_BLAS AND VECLIB_FOUND)
+  set(CBLAS_FOUND ON)
+  set(CBLAS_PROVIDER vecLib)
+  set(CBLAS_INC_DIR ${VECLIB_INC_DIR})
+  add_definitions(-DPADDLE_USE_VECLIB)
+endif()

cmake/external/gflags.cmake

@@ -39,13 +39,14 @@ ExternalProject_Add(
     PREFIX          ${GFLAGS_SOURCES_DIR}
     UPDATE_COMMAND  ""
     CMAKE_ARGS      -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
-    CMAKE_ARGS      -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
-    CMAKE_ARGS      -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
-    CMAKE_ARGS      -DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}
-    CMAKE_ARGS      -DCMAKE_INSTALL_PREFIX=${GFLAGS_INSTALL_DIR}
-    CMAKE_ARGS      -DCMAKE_POSITION_INDEPENDENT_CODE=ON
-    CMAKE_ARGS      -DBUILD_TESTING=OFF
-    CMAKE_ARGS      -DCMAKE_BUILD_TYPE=Release
+                    -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
+                    -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
+                    -DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}
+                    -DCMAKE_INSTALL_PREFIX=${GFLAGS_INSTALL_DIR}
+                    -DCMAKE_POSITION_INDEPENDENT_CODE=ON
+                    -DBUILD_TESTING=OFF
+                    -DCMAKE_BUILD_TYPE=Release
+                    ${EXTERNAL_OPTIONAL_ARGS}
     CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${GFLAGS_INSTALL_DIR}
                      -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON
                      -DCMAKE_BUILD_TYPE:STRING=Release

cmake/external/glog.cmake

@@ -34,16 +34,17 @@ ExternalProject_Add(
     PREFIX          ${GLOG_SOURCES_DIR}
     UPDATE_COMMAND  ""
     CMAKE_ARGS      -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
-    CMAKE_ARGS      -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
-    CMAKE_ARGS      -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
-    CMAKE_ARGS      -DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}
-    CMAKE_ARGS      -DCMAKE_INSTALL_PREFIX=${GLOG_INSTALL_DIR}
-    CMAKE_ARGS      -DCMAKE_INSTALL_LIBDIR=${GLOG_INSTALL_DIR}/lib
-    CMAKE_ARGS      -DCMAKE_POSITION_INDEPENDENT_CODE=ON
-    CMAKE_ARGS      -DWITH_GFLAGS=ON
-    CMAKE_ARGS      -Dgflags_DIR=${GFLAGS_INSTALL_DIR}/lib/cmake/gflags
-    CMAKE_ARGS      -DBUILD_TESTING=OFF
-    CMAKE_ARGS      -DCMAKE_BUILD_TYPE=Release
+                    -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
+                    -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
+                    -DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}
+                    -DCMAKE_INSTALL_PREFIX=${GLOG_INSTALL_DIR}
+                    -DCMAKE_INSTALL_LIBDIR=${GLOG_INSTALL_DIR}/lib
+                    -DCMAKE_POSITION_INDEPENDENT_CODE=ON
+                    -DWITH_GFLAGS=ON
+                    -Dgflags_DIR=${GFLAGS_INSTALL_DIR}/lib/cmake/gflags
+                    -DBUILD_TESTING=OFF
+                    -DCMAKE_BUILD_TYPE=Release
+                    ${EXTERNAL_OPTIONAL_ARGS}
     CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${GLOG_INSTALL_DIR}
                      -DCMAKE_INSTALL_LIBDIR:PATH=${GLOG_INSTALL_DIR}/lib
                      -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON

cmake/external/gtest.cmake

@@ -48,15 +48,16 @@ IF(WITH_TESTING)
         PREFIX          ${GTEST_SOURCES_DIR}
         UPDATE_COMMAND  ""
         CMAKE_ARGS      -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
-        CMAKE_ARGS      -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
-        CMAKE_ARGS      -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
-        CMAKE_ARGS      -DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}
-        CMAKE_ARGS      -DCMAKE_INSTALL_PREFIX=${GTEST_INSTALL_DIR}
-        CMAKE_ARGS      -DCMAKE_POSITION_INDEPENDENT_CODE=ON
-        CMAKE_ARGS      -DBUILD_GMOCK=ON
-        CMAKE_ARGS      -Dgtest_disable_pthreads=ON
-        CMAKE_ARGS      -Dgtest_force_shared_crt=ON
-        CMAKE_ARGS      -DCMAKE_BUILD_TYPE=Release
+                        -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
+                        -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
+                        -DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}
+                        -DCMAKE_INSTALL_PREFIX=${GTEST_INSTALL_DIR}
+                        -DCMAKE_POSITION_INDEPENDENT_CODE=ON
+                        -DBUILD_GMOCK=ON
+                        -Dgtest_disable_pthreads=ON
+                        -Dgtest_force_shared_crt=ON
+                        -DCMAKE_BUILD_TYPE=Release
+                        ${EXTERNAL_OPTIONAL_ARGS}
         CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${GTEST_INSTALL_DIR}
                          -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON
                          -DCMAKE_BUILD_TYPE:STRING=Release

cmake/external/openblas.cmake

@@ -29,30 +29,41 @@ IF(NOT ${CBLAS_FOUND})
         "${CBLAS_INSTALL_DIR}/lib/${CMAKE_STATIC_LIBRARY_PREFIX}openblas${CMAKE_STATIC_LIBRARY_SUFFIX}"
         CACHE FILEPATH "openblas library." FORCE)
 
-    IF(APPLE)
-        SET(OPENBLAS_CC "${CMAKE_C_COMPILER} -isysroot ${CMAKE_OSX_SYSROOT}")
-        SET(COMMON_ARGS CC=${OPENBLAS_CC} NO_SHARED=1 NO_LAPACK=1 libs)
-    ELSE()
-        SET(COMMON_ARGS CC=${CMAKE_C_COMPILER} NO_SHARED=1 NO_LAPACK=1 libs)
-    ENDIF()
+    SET(OPENBLAS_CC "${CMAKE_C_COMPILER}")
 
     IF(CMAKE_CROSSCOMPILING)
+        SET(OPTIONAL_ARGS HOSTCC=${HOST_C_COMPILER})
+        GET_FILENAME_COMPONENT(CROSS_SUFFIX ${CMAKE_C_COMPILER} DIRECTORY)
+        SET(CROSS_SUFFIX ${CROSS_SUFFIX}/)
         IF(ANDROID)
             # arm_soft_fp_abi branch of OpenBLAS to support softfp
             #   https://github.com/xianyi/OpenBLAS/tree/arm_soft_fp_abi
             SET(OPENBLAS_COMMIT "b5c96fcfcdc82945502a2303116a64d89985daf5")
             IF(ANDROID_ABI MATCHES "^armeabi(-v7a)?$")
-                SET(TARGET "ARMV7")
+                SET(OPTIONAL_ARGS ${OPTIONAL_ARGS} TARGET=ARMV7 ARM_SOFTFP_ABI=1 USE_THREAD=0)
             ELSEIF(ANDROID_ABI STREQUAL "arm64-v8a")
-                SET(TARGET "ARMV8")
+                SET(OPTIONAL_ARGS ${OPTIONAL_ARGS} TARGET=ARMV8 BINARY=64 USE_THREAD=0)
+            ENDIF()
+        ELSEIF(IOS)
+            # FIXME(liuyiqun): support multiple architectures
+            SET(OPENBLAS_COMMIT "b5c96fcfcdc82945502a2303116a64d89985daf5")
+            SET(OPENBLAS_CC "${OPENBLAS_CC} ${CMAKE_C_FLAGS} -isysroot ${CMAKE_OSX_SYSROOT}")
+            IF(CMAKE_OSX_ARCHITECTURES MATCHES "armv7")
+                SET(OPENBLAS_CC "${OPENBLAS_CC} -arch armv7")
+                SET(OPTIONAL_ARGS ${OPTIONAL_ARGS} TARGET=ARMV7 ARM_SOFTFP_ABI=1 USE_THREAD=0)
+            ELSEIF(CMAKE_OSX_ARCHITECTURES MATCHES "arm64")
+                SET(OPENBLAS_CC "${OPENBLAS_CC} -arch arm64")
+                SET(OPTIONAL_ARGS ${OPTIONAL_ARGS} TARGET=ARMV8 BINARY=64 USE_THREAD=0 CROSS_SUFFIX=${CROSS_SUFFIX})
             ENDIF()
-            SET(OPTIONAL_ARGS HOSTCC=${HOST_C_COMPILER} TARGET=${TARGET} ARM_SOFTFP_ABI=1 USE_THREAD=0)
         ELSEIF(RPI)
             # use hardfp
             SET(OPENBLAS_COMMIT "v0.2.20")
-            SET(OPTIONAL_ARGS HOSTCC=${HOST_C_COMPILER} TARGET=ARMV7 USE_THREAD=0)
+            SET(OPTIONAL_ARGS ${OPTIONAL_ARGS} TARGET=ARMV7 USE_THREAD=0)
         ENDIF()
     ELSE()
+        IF(APPLE)
+            SET(OPENBLAS_CC "${CMAKE_C_COMPILER} -isysroot ${CMAKE_OSX_SYSROOT}")
+        ENDIF()
         SET(OPENBLAS_COMMIT "v0.2.20")
         SET(OPTIONAL_ARGS "")
         IF(CMAKE_SYSTEM_PROCESSOR MATCHES "^x86(_64)?$")
@@ -60,6 +71,8 @@ IF(NOT ${CBLAS_FOUND})
         ENDIF()
     ENDIF()
 
+    SET(COMMON_ARGS CC=${OPENBLAS_CC} NO_SHARED=1 NO_LAPACK=1 libs)
+
     ExternalProject_Add(
         extern_openblas
         ${EXTERNAL_PROJECT_LOG_ARGS}

cmake/external/protobuf.cmake

@@ -173,7 +173,8 @@ FUNCTION(build_protobuf TARGET_NAME BUILD_FOR_HOST)
             "-DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}"
             "-DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}"
             "-Dprotobuf_WITH_ZLIB=ON"
-            "-DZLIB_ROOT:FILEPATH=${ZLIB_ROOT}")
+            "-DZLIB_ROOT:FILEPATH=${ZLIB_ROOT}"
+            ${EXTERNAL_OPTIONAL_ARGS})
         SET(OPTIONAL_CACHE_ARGS "-DZLIB_ROOT:STRING=${ZLIB_ROOT}")
     ENDIF()
 

cmake/external/python.cmake

@@ -12,16 +12,17 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-INCLUDE(ExternalProject)
+IF(NOT WITH_PYTHON)
+    return()
+ENDIF()
+
 INCLUDE(python_module)
 
 FIND_PACKAGE(PythonInterp 2.7)
-IF(WITH_PYTHON)
-    FIND_PACKAGE(PythonLibs 2.7)
-    # Fixme: Maybe find a static library. Get SHARED/STATIC by FIND_PACKAGE.
-    ADD_LIBRARY(python SHARED IMPORTED GLOBAL)
-    SET_PROPERTY(TARGET python PROPERTY IMPORTED_LOCATION ${PYTHON_LIBRARIES})
-ENDIF(WITH_PYTHON)
+FIND_PACKAGE(PythonLibs 2.7)
+# Fixme: Maybe find a static library. Get SHARED/STATIC by FIND_PACKAGE.
+ADD_LIBRARY(python SHARED IMPORTED GLOBAL)
+SET_PROPERTY(TARGET python PROPERTY IMPORTED_LOCATION ${PYTHON_LIBRARIES})
 
 SET(py_env "")
 IF(PYTHONINTERP_FOUND)
@@ -36,9 +37,5 @@ IF(PYTHONINTERP_FOUND)
     ENDIF()
 ENDIF(PYTHONINTERP_FOUND)
 
-IF(WITH_PYTHON)
-    INCLUDE_DIRECTORIES(${PYTHON_INCLUDE_DIR})
-    INCLUDE_DIRECTORIES(${PYTHON_NUMPY_INCLUDE_DIR})
-ELSE()
-    SET(PYTHON_LIBRARIES "")
-ENDIF()
+INCLUDE_DIRECTORIES(${PYTHON_INCLUDE_DIR})
+INCLUDE_DIRECTORIES(${PYTHON_NUMPY_INCLUDE_DIR})

cmake/external/swig.cmake

@@ -12,6 +12,10 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+IF(NOT WITH_SWIG_PY)
+    return()
+ENDIF()
+
 FIND_PACKAGE(SWIG)
 
 IF(NOT SWIG_FOUND)

cmake/external/warpctc.cmake

@@ -16,25 +16,14 @@ INCLUDE(ExternalProject)
 
 SET(WARPCTC_SOURCES_DIR ${THIRD_PARTY_PATH}/warpctc)
 SET(WARPCTC_INSTALL_DIR ${THIRD_PARTY_PATH}/install/warpctc)
-SET(WARPCTC_INCLUDE_DIR "${WARPCTC_INSTALL_DIR}/include" CACHE PATH "Warp-ctc Directory" FORCE)
 
-INCLUDE_DIRECTORIES(${WARPCTC_INCLUDE_DIR})
-
-SET(WARPCTC_LIB_DIR "${WARPCTC_INSTALL_DIR}/lib" CACHE PATH "Warp-ctc Library Directory" FORCE)
-
-IF(WIN32)
-    SET(WARPCTC_LIBRARIES
-        "${WARPCTC_INSTALL_DIR}/lib/warpctc.dll" CACHE FILEPATH "Warp-ctc Library" FORCE)
-ELSE(WIN32)
-    IF(APPLE)
-        SET(_warpctc_SHARED_SUFFIX dylib)
-    ELSE(APPLE)
-        SET(_warpctc_SHARED_SUFFIX so)
-    ENDIF(APPLE)
-
-    SET(WARPCTC_LIBRARIES
-        "${WARPCTC_INSTALL_DIR}/lib/libwarpctc.${_warpctc_SHARED_SUFFIX}" CACHE FILEPATH "Warp-ctc Library" FORCE)
-ENDIF(WIN32)
+SET(WARPCTC_INCLUDE_DIR "${WARPCTC_INSTALL_DIR}/include"
+    CACHE PATH "Warp-ctc Directory" FORCE)
+# Used in unit test test_WarpCTCLayer
+SET(WARPCTC_LIB_DIR "${WARPCTC_INSTALL_DIR}/lib"
+    CACHE PATH "Warp-ctc Library Directory" FORCE)
+SET(WARPCTC_LIBRARIES "${WARPCTC_INSTALL_DIR}/lib/libwarpctc${CMAKE_SHARED_LIBRARY_SUFFIX}"
+    CACHE FILEPATH "Warp-ctc Library" FORCE)
 
 IF(CMAKE_CXX_COMPILER_ID STREQUAL "Clang" OR CMAKE_CXX_COMPILER_ID STREQUAL "AppleClang" )
     SET(USE_OMP OFF)
@@ -49,22 +38,26 @@ ExternalProject_Add(
     PREFIX          ${WARPCTC_SOURCES_DIR}
     UPDATE_COMMAND  ""
     CMAKE_ARGS      -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
-    CMAKE_ARGS      -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
-    CMAKE_ARGS      -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
-    CMAKE_ARGS      -DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}
-    CMAKE_ARGS      -DCMAKE_INSTALL_PREFIX=${WARPCTC_INSTALL_DIR}
-    CMAKE_ARGS      -DWITH_GPU=${WITH_GPU}
-    CMAKE_ARGS      -DWITH_OMP=${USE_OMP}
-    CMAKE_ARGS      -DWITH_TORCH=OFF
-    CMAKE_ARGS      -DCMAKE_DISABLE_FIND_PACKAGE_Torch=ON
-    CMAKE_ARGS      -DBUILD_SHARED=ON
-    CMAKE_ARGS      -DCMAKE_POSITION_INDEPENDENT_CODE=ON
-    CMAKE_ARGS      -DCMAKE_BUILD_TYPE=Release
+                    -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
+                    -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
+                    -DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}
+                    -DCMAKE_INSTALL_PREFIX=${WARPCTC_INSTALL_DIR}
+                    -DWITH_GPU=${WITH_GPU}
+                    -DWITH_OMP=${USE_OMP}
+                    -DWITH_TORCH=OFF
+                    -DCMAKE_DISABLE_FIND_PACKAGE_Torch=ON
+                    -DBUILD_SHARED=ON
+                    -DCMAKE_POSITION_INDEPENDENT_CODE=ON
+                    -DCMAKE_BUILD_TYPE=Release
+                    ${EXTERNAL_OPTIONAL_ARGS}
     CMAKE_CACHE_ARGS -DCMAKE_BUILD_TYPE:STRING=Release
                      -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON
                      -DCMAKE_INSTALL_PREFIX:PATH=${WARPCTC_INSTALL_DIR}
 )
 
+MESSAGE(STATUS "warp-ctc library: ${WARPCTC_LIBRARIES}")
+INCLUDE_DIRECTORIES(${WARPCTC_INCLUDE_DIR})
+
 ADD_LIBRARY(warpctc STATIC IMPORTED GLOBAL)
 SET_PROPERTY(TARGET warpctc PROPERTY IMPORTED_LOCATION ${WARPCTC_LIBRARIES})
 ADD_DEPENDENCIES(warpctc extern_warpctc)

cmake/external/zlib.cmake

@@ -34,15 +34,16 @@ ExternalProject_Add(
     GIT_TAG         "v1.2.8"
     PREFIX          ${ZLIB_SOURCES_DIR}
     UPDATE_COMMAND  ""
-    CMAKE_ARGS      -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
     CMAKE_ARGS      -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
-    CMAKE_ARGS      -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
-    CMAKE_ARGS      -DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}
-    CMAKE_ARGS      -DCMAKE_INSTALL_PREFIX=${ZLIB_INSTALL_DIR}
-    CMAKE_ARGS      -DBUILD_SHARED_LIBS=OFF
-    CMAKE_ARGS      -DCMAKE_POSITION_INDEPENDENT_CODE=ON
-    CMAKE_ARGS      -DCMAKE_MACOSX_RPATH=ON
-    CMAKE_ARGS      -DCMAKE_BUILD_TYPE=Release
+                    -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
+                    -DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}
+                    -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
+                    -DCMAKE_INSTALL_PREFIX=${ZLIB_INSTALL_DIR}
+                    -DBUILD_SHARED_LIBS=OFF
+                    -DCMAKE_POSITION_INDEPENDENT_CODE=ON
+                    -DCMAKE_MACOSX_RPATH=ON
+                    -DCMAKE_BUILD_TYPE=Release
+                    ${EXTERNAL_OPTIONAL_ARGS}
     CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${ZLIB_INSTALL_DIR}
                      -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON
                      -DCMAKE_BUILD_TYPE:STRING=Release

cmake/flags.cmake

@@ -128,8 +128,10 @@ set(GPU_COMMON_FLAGS
 )
 
 if (APPLE)
-    # On Mac OS X build fat binaries with x86_64 architectures by default.
-    set (CMAKE_OSX_ARCHITECTURES "x86_64" CACHE STRING "Build architectures for OSX" FORCE)
+    if(NOT CMAKE_CROSSCOMPILING)
+        # On Mac OS X build fat binaries with x86_64 architectures by default.
+        set (CMAKE_OSX_ARCHITECTURES "x86_64" CACHE STRING "Build architectures for OSX" FORCE)
+    endif()
 else()
     set(GPU_COMMON_FLAGS
         -Wall

cmake/system.cmake

@@ -24,11 +24,10 @@ IF(WIN32)
     SET(HOST_SYSTEM "win32")
 ELSE(WIN32)
     IF(APPLE)
-        EXEC_PROGRAM (sw_vers ARGS -productVersion OUTPUT_VARIABLE MACOSX_VERSION)
-        STRING(REGEX MATCH "[0-9]+.[0-9]+" VERSION "${MACOSX_VERSION}")
-        SET(MACOS_VERSION ${VERSION})
         SET(HOST_SYSTEM "macosx")
-        IF(NOT DEFINED ENV{MACOSX_DEPLOYMENT_TARGET})
+        EXEC_PROGRAM(sw_vers ARGS -productVersion OUTPUT_VARIABLE HOST_SYSTEM_VERSION)
+        STRING(REGEX MATCH "[0-9]+.[0-9]+" MACOS_VERSION "${HOST_SYSTEM_VERSION}")
+        IF(NOT DEFINED $ENV{MACOSX_DEPLOYMENT_TARGET})
             # Set cache variable - end user may change this during ccmake or cmake-gui configure.
             SET(CMAKE_OSX_DEPLOYMENT_TARGET ${MACOS_VERSION} CACHE STRING
                 "Minimum OS X version to target for deployment (at runtime); newer APIs weak linked. Set to empty string for default value.")
@@ -49,6 +48,8 @@ ELSE(WIN32)
             ELSEIF(LINUX_ISSUE MATCHES "Fedora")
                 SET(HOST_SYSTEM "fedora")
             ENDIF()
+
+            STRING(REGEX MATCH "(([0-9]+)\\.)+([0-9]+)" HOST_SYSTEM_VERSION "${LINUX_ISSUE}")
         ENDIF(EXISTS "/etc/issue")
 
         IF(EXISTS "/etc/redhat-release")
@@ -70,7 +71,7 @@ CMAKE_HOST_SYSTEM_INFORMATION(RESULT CPU_CORES QUERY NUMBER_OF_LOGICAL_CORES)
 
 MARK_AS_ADVANCED(HOST_SYSTEM CPU_CORES)
 
-MESSAGE(STATUS "Found Paddle host system: ${HOST_SYSTEM}")
+MESSAGE(STATUS "Found Paddle host system: ${HOST_SYSTEM}, version: ${HOST_SYSTEM_VERSION}")
 MESSAGE(STATUS "Found Paddle host system's CPU: ${CPU_CORES} cores")
 
 # configuration for cross-compiling
@@ -82,6 +83,9 @@ IF(DEFINED CMAKE_SYSTEM_NAME)
     ELSEIF(${CMAKE_SYSTEM_NAME} STREQUAL "RPi")
         SET(RPI TRUE)
         INCLUDE(cross_compiling/raspberry_pi)
+    ELSEIF(${CMAKE_SYSTEM_NAME} STREQUAL "iOS")
+        SET(IOS TRUE)
+        INCLUDE(cross_compiling/ios)
     ENDIF()
 ENDIF()
 

cmake/util.cmake

@@ -25,7 +25,9 @@ function(target_circle_link_libraries TARGET_NAME)
             endif()
         endforeach()
         if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang" OR "${CMAKE_CXX_COMPILER_ID}" STREQUAL "AppleClang")
-            list(APPEND LIBS "-undefined dynamic_lookup")
+            if(IOS AND NOT IOS_ENABLE_BITCODE)
+                list(APPEND LIBS "-undefined dynamic_lookup")
+            endif()
         endif()
         list(REVERSE libsInArgn)
         target_link_libraries(${TARGET_NAME}

doc/faq/index_cn.rst

@@ -321,3 +321,55 @@ pip uninstall py_paddle paddle
 然后安装paddle的python环境, 在build目录下执行
 
 pip install python/dist/paddle*.whl && pip install ../paddle/dist/py_paddle*.whl
+
+16. PaddlePaddle存储的参数格式是什么，如何和明文进行相互转化
+---------------------------------------------------------
+
+PaddlePaddle保存的模型参数文件内容由16字节头信息和网络参数两部分组成。头信息中，1~4字节表示PaddlePaddle版本信息，请直接填充0；5~8字节表示每个参数占用的字节数，当保存的网络参数为float类型时为4，double类型时为8；9~16字节表示保存的参数总个数。
+
+将PaddlePaddle保存的模型参数还原回明文时，可以使用相应数据类型的 :code:`numpy.array` 加载具体网络参数，此时可以跳过PaddlePaddle模型参数文件的头信息。若在PaddlePaddle编译时，未指定按照double精度编译，默认情况下按照float精度计算，保存的参数也是float类型。这时在使用 :code:`numpy.array` 时，一般设置 :code:`dtype=float32` 。示例如下：
+
+..  code-block:: python
+
+    def read_parameter(fname, width):
+        s = open(fname).read()
+        # skip header
+        vec = np.fromstring(s[16:], dtype=np.float32)
+        # width is the size of the corresponding layer
+        np.savetxt(fname + ".csv", vec.reshape(width, -1),
+                fmt="%.6f", delimiter=",")
+
+
+将明文参数转化为PaddlePaddle可加载的模型参数时，首先构造头信息，再写入网络参数。下面的代码将随机生成的矩阵转化为可以被PaddlePaddle加载的模型参数。
+
+..  code-block:: python
+
+    def gen_rand_param(param_file, width, height, need_trans):
+        np.random.seed()
+        header = struct.pack("iil", 0, 4, height * width)
+        param = np.float32(np.random.rand(height, width))
+        with open(param_file, "w") as fparam:
+            fparam.write(header + param.tostring())
+
+17. 如何加载预训练参数
+------------------------------
+
+* 对加载预训练参数的层，设置其参数属性 :code:`is_static=True`，使该层的参数在训练过程中保持不变。以embedding层为例，代码如下：
+
+..  code-block:: python
+
+    emb_para = paddle.attr.Param(name='emb', is_static=True)
+    paddle.layer.embedding(size=word_dim, input=x, param_attr=emb_para)
+
+
+* 从模型文件将预训练参数载入 :code:`numpy.array`，在创建parameters后，使用 :code:`parameters.set()` 加载预训练参数。PaddlePaddle保存的模型参数文件前16字节为头信息，用户将参数载入 :code:`numpy.array` 时须从第17字节开始。以embedding层为例，代码如下：
+
+..  code-block:: python
+
+    def load_parameter(file_name, h, w):
+        with open(file_name, 'rb') as f:
+            f.read(16)  # skip header.
+            return np.fromfile(f, dtype=np.float32).reshape(h, w)
+
+    parameters = paddle.parameters.create(my_cost)
+    parameters.set('emb', load_parameter(emb_param_file, 30000, 256))

paddle/CMakeLists.txt

@@ -19,7 +19,7 @@ if(Boost_FOUND)
 endif()
 
 if(WITH_C_API)
-    add_subdirectory(capi)
+  add_subdirectory(capi)
 endif()
 
 if(WITH_SWIG_PY)

paddle/capi/CMakeLists.txt

@@ -28,42 +28,38 @@ add_style_check_target(paddle_capi ${CAPI_SOURCES} ${CAPI_HEADER}
 
 add_dependencies(paddle_capi paddle_proto)
 
-
 # combine all paddle static libraries together, into libpaddle_capi_whole.a
 # user should use PaddleCAPI as -lpaddle_capi_whole
-set(capi_whole_library libpaddle_capi_whole.a)
-add_custom_target(paddle_capi_whole ALL
-        COMMAND mkdir -p o_files/capi && cd o_files/capi/ && ar -x $<TARGET_FILE:paddle_capi>
-        COMMAND mkdir -p o_files/utils && cd o_files/utils/ && ar -x $<TARGET_FILE:paddle_utils>
-        COMMAND mkdir -p o_files/parameter && cd o_files/parameter/ && ar -x $<TARGET_FILE:paddle_parameter>
-        COMMAND mkdir -p o_files/math && cd o_files/math/  && ar -x $<TARGET_FILE:paddle_math>
-        COMMAND mkdir -p o_files/cuda && cd o_files/cuda/ && ar -x $<TARGET_FILE:paddle_cuda>
-        COMMAND mkdir -p o_files/function && cd o_files/function/ && ar -x $<TARGET_FILE:paddle_function>
-        COMMAND mkdir -p o_files/gserver && cd o_files/gserver/ && ar -x $<TARGET_FILE:paddle_gserver>
-        COMMAND mkdir -p o_files/proto && cd o_files/proto/ && ar -x $<TARGET_FILE:paddle_proto>
-        COMMAND mkdir -p o_files/network && cd o_files/network/ && ar -x $<TARGET_FILE:paddle_network>
-        COMMAND mkdir -p o_files/pserver && cd o_files/pserver/ && ar -x $<TARGET_FILE:paddle_pserver>
-        COMMAND ar crs ${capi_whole_library} `find ./o_files -name '*.o'`
-        COMMAND rm -rf o_files
-        WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
-        DEPENDS paddle_capi paddle_utils paddle_parameter paddle_math
-                paddle_cuda paddle_function paddle_gserver
-                paddle_proto paddle_pserver paddle_network
-        )
-set_target_properties(paddle_capi_whole
-  PROPERTIES IMPORTED_LOCATION ${CMAKE_CURRENT_BINARY_DIR}/${capi_whole_library})
+set(PADDLE_CAPI_INFER_LIBS
+    paddle_utils
+    paddle_parameter
+    paddle_math
+    paddle_cuda
+    paddle_function
+    paddle_gserver
+    paddle_proto
+    paddle_pserver
+    paddle_network)
+
+cc_library(paddle_capi_whole DEPS paddle_capi ${PADDLE_CAPI_INFER_LIBS})
 
-set(LINK_FLAGS " -Wl,--retain-symbols-file ${CMAKE_CURRENT_SOURCE_DIR}/export.sym -Wl,--version-script ${CMAKE_CURRENT_SOURCE_DIR}/export.map")
-# TODO: merge mkl into paddle_capi_shared
-add_library(paddle_capi_shared SHARED ${CAPI_SOURCES})
-set_target_properties(paddle_capi_shared	PROPERTIES LINK_FLAGS "${LINK_FLAGS}")
-target_include_directories(paddle_capi_shared PUBLIC ${CMAKE_CURRENT_BINARY_DIR})
-link_paddle_exe(paddle_capi_shared)
+# No shared library for iOS
+if(NOT IOS)
+  set(LINK_FLAGS " -Wl,--retain-symbols-file ${CMAKE_CURRENT_SOURCE_DIR}/export.sym -Wl,--version-script ${CMAKE_CURRENT_SOURCE_DIR}/export.map")
+  # TODO: merge mkl into paddle_capi_shared
+  add_library(paddle_capi_shared SHARED ${CAPI_SOURCES})
+  set_target_properties(paddle_capi_shared	PROPERTIES LINK_FLAGS "${LINK_FLAGS}")
+  target_include_directories(paddle_capi_shared PUBLIC ${CMAKE_CURRENT_BINARY_DIR})
+  link_paddle_exe(paddle_capi_shared)
+endif()
 
 # install library & headers.
 install(FILES ${CAPI_HEADERS} DESTINATION include/paddle)
 install(FILES ${CMAKE_CURRENT_BINARY_DIR}/config.h DESTINATION include/paddle)
 if(ANDROID)
+  install(TARGETS paddle_capi_whole paddle_capi_shared
+          ARCHIVE DESTINATION lib/${ANDROID_ABI}
+          LIBRARY DESTINATION lib/${ANDROID_ABI})
   execute_process(
     COMMAND ${GIT_EXECUTABLE} log --pretty=oneline -1
     OUTPUT_VARIABLE GIT_COMMITS_LIST
@@ -72,9 +68,6 @@ if(ANDROID)
   if(${GIT_COMMITS_LIST_RESULT})
     set(GIT_COMMITS_LIST "No commits.")
   endif()
-  install(FILES ${CMAKE_CURRENT_BINARY_DIR}/${capi_whole_library}
-          DESTINATION lib/${ANDROID_ABI})
-  install(TARGETS paddle_capi_shared DESTINATION lib/${ANDROID_ABI})
   install(CODE "FILE(WRITE ${CMAKE_INSTALL_PREFIX}/lib/${ANDROID_ABI}/BUILD.txt
           \"Compiler:\n\"
           \"\\t${CMAKE_C_COMPILER}\\n\"
@@ -88,8 +81,11 @@ if(ANDROID)
       )"
   )
 else(ANDROID)
-  install(FILES ${CMAKE_CURRENT_BINARY_DIR}/${capi_whole_library} DESTINATION lib)
-  install(TARGETS paddle_capi_shared DESTINATION lib)
+  install(TARGETS paddle_capi_whole
+          ARCHIVE DESTINATION lib)
+  if(NOT IOS)
+    install(TARGETS paddle_capi_shared DESTINATION lib)
+  endif()
 endif(ANDROID)
 
 # this variable used for unittest

paddle/cuda/include/hl_cuda_cudnn.h

@@ -22,10 +22,10 @@ limitations under the License. */
  */
 typedef enum {
   HL_POOLING_MAX = 0,
-  // average includes padded values
-  HL_POOLING_AVERAGE = 1,
   // average does not include padded values
-  HL_POOLING_AVERAGE_EXCLUDE_PADDING = 2,
+  HL_POOLING_AVERAGE = 1,
+  // average includes padded values
+  HL_POOLING_AVERAGE_INCLUDE_PADDING = 2,
   HL_POOLING_END
 } hl_pooling_mode_t;
 

paddle/cuda/include/hl_tensor_ops.h

@@ -461,7 +461,7 @@ class add<float32x4_t> {
 public:
   INLINE float32x4_t operator()(const float32x4_t a,
                                 const float32x4_t b) const {
-    return vmulq_f32(a, b);
+    return vaddq_f32(a, b);
   }
 };
 

paddle/cuda/src/hl_cuda_cnn.cu

@@ -211,13 +211,11 @@ __global__ void KeAvgPoolForward(const int nthreads,
 
     int hstart = ph * strideH - padH;
     int wstart = pw * strideW - padW;
-    int hend = min(hstart + sizeY, height + padH);
-    int wend = min(wstart + sizeX, width + padW);
-    int pool_size = (hend - hstart) * (wend - wstart);
+    int hend = min(hstart + sizeY, height);
+    int wend = min(wstart + sizeX, width);
     hstart = max(hstart, 0);
     wstart = max(wstart, 0);
-    hend = min(hend, height);
-    wend = min(wend, width);
+    int pool_size = (hend - hstart) * (wend - wstart);
 
     real aveval = 0;
     inputData += (frameNum * channels + c) * height * width;
@@ -299,12 +297,14 @@ __global__ void KeAvgPoolBackward(const int nthreads,
     outGrad += (frameNum * outStride + offsetC * pooledH * pooledW);
 
     for (int ph = phstart; ph < phend; ++ph) {
+      int hstart = ph * strideH - padH;
+      int hend = min(hstart + sizeY, height);
+      hstart = max(hstart, 0);
       for (int pw = pwstart; pw < pwend; ++pw) {
         // figure out the pooling size
-        int hstart = ph * strideH - padH;
         int wstart = pw * strideW - padW;
-        int hend = min(hstart + sizeY, height + padH);
-        int wend = min(wstart + sizeX, width + padW);
+        int wend = min(wstart + sizeX, width);
+        wstart = max(wstart, 0);
         int poolsize = (hend - hstart) * (wend - wstart);
         gradient += outGrad[ph * pooledW + pw] / poolsize;
       }
@@ -600,16 +600,13 @@ __global__ void KeAvgPool3DForward(const int nthreads,
     int dstart = pd * strideD - padD;
     int hstart = ph * strideH - padH;
     int wstart = pw * strideW - padW;
-    int dend = min(dstart + sizeZ, depth + padD);
-    int hend = min(hstart + sizeY, height + padH);
-    int wend = min(wstart + sizeX, width + padW);
-    int pool_size = (dend - dstart) * (hend - hstart) * (wend - wstart);
+    int dend = min(dstart + sizeZ, depth);
+    int hend = min(hstart + sizeY, height);
+    int wend = min(wstart + sizeX, width);
     dstart = max(dstart, 0);
     hstart = max(hstart, 0);
     wstart = max(wstart, 0);
-    dend = min(dend, depth);
-    hend = min(hend, height);
-    wend = min(wend, width);
+    int pool_size = (dend - dstart) * (hend - hstart) * (wend - wstart);
 
     real aveval = 0;
     inputData += (frameNum * channels + c) * depth * height * width;
@@ -712,15 +709,18 @@ __global__ void KeAvgPool3DBackward(const int nthreads,
     outGrad += (frameNum * channels + offsetC) * pooledD * pooledH * pooledW;
 
     for (int pd = pdstart; pd < pdend; ++pd) {
+      int dstart = pd * strideD - padD;
+      int dend = min(dstart + sizeZ, depth);
+      dstart = max(dstart, 0);
       for (int ph = phstart; ph < phend; ++ph) {
+        int hstart = ph * strideH - padH;
+        int hend = min(hstart + sizeY, height);
+        hstart = max(hstart, 0);
         for (int pw = pwstart; pw < pwend; ++pw) {
           // figure out the pooling size
-          int dstart = pd * strideD - padD;
-          int hstart = ph * strideH - padH;
           int wstart = pw * strideW - padW;
-          int dend = min(dstart + sizeZ, depth + padD);
-          int hend = min(hstart + sizeY, height + padH);
-          int wend = min(wstart + sizeX, width + padW);
+          int wend = min(wstart + sizeX, width);
+          wstart = max(wstart, 0);
           int poolsize = (dend - dstart) * (hend - hstart) * (wend - wstart);
           gradient += outGrad[(pd * pooledH + ph) * pooledW + pw] / poolsize;
         }