Merge branch 'develop' into updateWriteDocsCN

updateWriteDocsCN
Jeff Wang 7 years ago committed by GitHub
commit 048cc30e72
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

@ -25,4 +25,3 @@ AllowAllParametersOfDeclarationOnNextLine: true
BinPackParameters: false
BinPackArguments: false
...

@ -42,7 +42,7 @@ before_install:
script:
- |
timeout 2580 paddle/scripts/travis/${JOB}.sh # 43min timeout
RESULT=$?; if [ $RESULT -eq 0 ] || [ $RESULT -eq 142 ]; then true; else false; fi;
RESULT=$?; if [ $RESULT -eq 0 ] || [ $RESULT -eq 142 ]; then true ;else exit 1; fi;
- |
if [[ "$JOB" != "build_doc" ]]; then exit 0; fi;
if [[ "$TRAVIS_PULL_REQUEST" != "false" ]]; then exit 0; fi;

@ -36,8 +36,7 @@ include(simd)
################################ Configurations #######################################
option(WITH_GPU "Compile PaddlePaddle with NVIDIA GPU" ${CUDA_FOUND})
option(WITH_AVX "Compile PaddlePaddle with AVX intrinsics" ${AVX_FOUND})
option(WITH_MKLDNN "Compile PaddlePaddle with mkl-dnn support." ${AVX_FOUND})
option(WITH_MKLML "Compile PaddlePaddle with mklml package." ${AVX_FOUND})
option(WITH_MKL "Compile PaddlePaddle with MKL support." ${AVX_FOUND})
option(WITH_DSO "Compile PaddlePaddle with dynamic linked CUDA" ON)
option(WITH_TESTING "Compile PaddlePaddle with unit testing" ON)
option(WITH_SWIG_PY "Compile PaddlePaddle with inference api" ON)
@ -82,10 +81,8 @@ if(ANDROID OR IOS)
"Disable PYTHON when cross-compiling for Android and iOS" FORCE)
set(WITH_RDMA OFF CACHE STRING
"Disable RDMA when cross-compiling for Android and iOS" FORCE)
set(WITH_MKLDNN OFF CACHE STRING
"Disable MKLDNN when cross-compiling for Android and iOS" FORCE)
set(WITH_MKLML OFF CACHE STRING
"Disable MKLML package when cross-compiling for Android and iOS" FORCE)
set(WITH_MKL OFF CACHE STRING
"Disable MKL when cross-compiling for Android and iOS" FORCE)
# Compile PaddlePaddle mobile inference library
if (NOT WITH_C_API)
@ -111,6 +108,14 @@ else()
set(THIRD_PARTY_BUILD_TYPE Release)
endif()
set(WITH_MKLML ${WITH_MKL})
if (WITH_MKL AND AVX2_FOUND)
set(WITH_MKLDNN ON)
else()
message(STATUS "Do not have AVX2 intrinsics and disabled MKL-DNN")
set(WITH_MKLDNN OFF)
endif()
########################################################################################
include(external/mklml) # download mklml package
@ -128,6 +133,8 @@ include(external/any) # download libn::any
include(external/eigen) # download eigen3
include(external/pybind11) # download pybind11
include(external/nccl)
include(external/cares)
include(external/grpc)
include(cudnn) # set cudnn libraries, must before configure
include(configure) # add paddle env configuration
@ -158,14 +165,15 @@ set(EXTERNAL_LIBS
)
if(WITH_GPU)
list(APPEND EXTERNAL_LIBS ${CUDA_LIBRARIES} ${CUDA_rt_LIBRARY})
if(NOT WITH_DSO)
list(APPEND EXTERNAL_LIBS ${CUDNN_LIBRARY} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_curand_LIBRARY} ${NCCL_LIBRARY})
endif(NOT WITH_DSO)
include(cuda)
endif(WITH_GPU)
if(WITH_MKLML)
list(APPEND EXTERNAL_LIBS ${MKLML_IOMP_LIB})
endif()
if(WITH_MKLDNN)
list(APPEND EXTERNAL_LIBS ${MKLDNN_LIB} ${MKLDNN_IOMP_LIB})
list(APPEND EXTERNAL_LIBS ${MKLDNN_LIB})
endif()
if(USE_NNPACK)

@ -29,7 +29,7 @@ RUN apt-get update && \
automake locales clang-format swig doxygen cmake \
liblapack-dev liblapacke-dev libboost-dev \
clang-3.8 llvm-3.8 libclang-3.8-dev \
net-tools && \
net-tools libtool && \
apt-get clean -y
# Install Go and glide

@ -12,11 +12,11 @@ Machine:
System: CentOS release 6.3 (Final), Docker 1.12.1.
PaddlePaddle: paddlepaddle/paddle:latest (TODO: will rerun after 0.11.0)
- MKL-DNN tag v0.10
- MKLML 2018.0.20170720
PaddlePaddle: paddlepaddle/paddle:latest (for MKLML and MKL-DNN), paddlepaddle/paddle:latest-openblas (for OpenBLAS)
- MKL-DNN tag v0.11
- MKLML 2018.0.1.20171007
- OpenBLAS v0.2.20
(TODO: will rerun after 0.11.0)
On each machine, we will test and compare the performance of training on single node using MKL-DNN / MKLML / OpenBLAS respectively.
@ -31,17 +31,37 @@ Input image size - 3 * 224 * 224, Time: images/second
| BatchSize | 64 | 128 | 256 |
|--------------|-------| -----| --------|
| OpenBLAS | 7.82 | 8.62 | 10.34 |
| MKLML | 11.02 | 12.86 | 15.33 |
| MKL-DNN | 27.69 | 28.8 | 29.27 |
| OpenBLAS | 7.80 | 9.00 | 10.80 |
| MKLML | 12.12 | 13.70 | 16.18 |
| MKL-DNN | 28.46 | 29.83 | 30.44 |
chart on batch size 128
TBD
- ResNet-50
| BatchSize | 64 | 128 | 256 |
|--------------|-------| ------| -------|
| OpenBLAS | 25.22 | 25.68 | 27.12 |
| MKLML | 32.52 | 31.89 | 33.12 |
| MKL-DNN | 81.69 | 82.35 | 84.08 |
chart on batch size 128
TBD
- ResNet
- GoogLeNet
| BatchSize | 64 | 128 | 256 |
|--------------|-------| ------| -------|
| OpenBLAS | 89.52 | 96.97 | 108.25 |
| MKLML | 128.46| 137.89| 158.63 |
| MKL-DNN     | 250.46| 264.83| 269.50 |
chart on batch size 128
TBD
### Laptop
TBD
### Desktop

@ -5,6 +5,7 @@ height = 224
width = 224
num_class = 1000
batch_size = get_config_arg('batch_size', int, 128)
use_gpu = get_config_arg('use_gpu', bool, True)
args = {'height': height, 'width': width, 'color': True, 'num_class': num_class}
define_py_data_sources2(
@ -16,6 +17,8 @@ settings(
learning_method=MomentumOptimizer(0.9),
regularization=L2Regularization(0.0005 * batch_size))
conv_projection = conv_projection if use_gpu else img_conv_layer
def inception2(name, input, channels, \
filter1,
filter3R, filter3,
@ -138,7 +141,7 @@ def inception(name, input, channels, \
cat = concat_layer(
name=name,
input=[cov1, cov3, cov5, covprj],
bias_attr=True,
bias_attr=True if use_gpu else False,
act=ReluActivation())
return cat

@ -40,6 +40,7 @@ fi
for use_mkldnn in True False; do
for batchsize in 64 128 256; do
train vgg 19 $batchsize $use_mkldnn
train resnet 50 $batchsize $use_mkldnn
train resnet 50 $batchsize $use_mkldnn
train googlenet v1 $batchsize $use_mkldnn
done
done

@ -76,27 +76,14 @@ else()
include_directories(${CUDA_TOOLKIT_INCLUDE})
endif(NOT WITH_GPU)
if(WITH_MKLDNN)
add_definitions(-DPADDLE_USE_MKLDNN)
if (WITH_MKLML AND MKLDNN_IOMP_DIR)
message(STATUS "Enable Intel OpenMP at ${MKLDNN_IOMP_DIR}")
set(OPENMP_FLAGS "-fopenmp")
set(CMAKE_C_CREATE_SHARED_LIBRARY_FORBIDDEN_FLAGS ${OPENMP_FLAGS})
set(CMAKE_CXX_CREATE_SHARED_LIBRARY_FORBIDDEN_FLAGS ${OPENMP_FLAGS})
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OPENMP_FLAGS}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OPENMP_FLAGS}")
else()
find_package(OpenMP)
if(OPENMP_FOUND)
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OpenMP_C_FLAGS}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OpenMP_CXX_FLAGS}")
else()
message(WARNING "Can not find OpenMP."
"Some performance features in MKLDNN may not be available")
endif()
endif()
endif(WITH_MKLDNN)
if (WITH_MKLML AND MKLML_IOMP_LIB)
message(STATUS "Enable Intel OpenMP with ${MKLML_IOMP_LIB}")
set(OPENMP_FLAGS "-fopenmp")
set(CMAKE_C_CREATE_SHARED_LIBRARY_FORBIDDEN_FLAGS ${OPENMP_FLAGS})
set(CMAKE_CXX_CREATE_SHARED_LIBRARY_FORBIDDEN_FLAGS ${OPENMP_FLAGS})
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OPENMP_FLAGS}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OPENMP_FLAGS}")
endif()
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${SIMD_FLAG}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${SIMD_FLAG}")

@ -76,11 +76,9 @@ set(IOS_PLATFORM ${IOS_PLATFORM} CACHE STRING "Type of iOS Platform")
# Set the architecture for iOS
if(NOT DEFINED IOS_ARCH)
if(IOS_PLATFORM STREQUAL "OS")
# FIXME(liuyiqun): support "armv7;armv7s;arm64" future
set(IOS_ARCH "arm64")
set(IOS_ARCH "armv7;armv7s;arm64")
elseif(IOS_PLATFORM STREQUAL "SIMULATOR")
# FIXME(liuyiqun): support "i386;x86_64" future
set(IOS_ARCH "x86_64")
set(IOS_ARCH "i386;x86_64")
endif()
endif()
set(CMAKE_OSX_ARCHITECTURES ${IOS_ARCH} CACHE string "Build architecture for iOS")
@ -248,7 +246,7 @@ set(IOS_COMPILER_FLAGS "${XCODE_IOS_PLATFORM_VERSION_FLAGS} ${XCODE_IOS_BITCODE_
# Hidden visibilty is required for cxx on iOS
set(CMAKE_C_FLAGS "${IOS_COMPILER_FLAGS} ${CMAKE_C_FLAGS}" CACHE STRING "C flags")
set(CMAKE_CXX_FLAGS "${IOS_COMPILER_FLAGS} -fvisibility-inlines-hidden ${CMAKE_CXX_FLAGS}" CACHE STRING "CXX flags")
set(CMAKE_CXX_FLAGS "${IOS_COMPILER_FLAGS} -fvisibility=hidden -fvisibility-inlines-hidden ${CMAKE_CXX_FLAGS}" CACHE STRING "CXX flags")
set(IOS_LINK_FLAGS "${XCODE_IOS_PLATFORM_VERSION_FLAGS} -Wl,-search_paths_first")

@ -0,0 +1,188 @@
if(NOT WITH_GPU)
return()
endif()
set(paddle_known_gpu_archs "30 35 50 52 60 61 70")
set(paddle_known_gpu_archs7 "30 35 50 52")
set(paddle_known_gpu_archs8 "30 35 50 52 60 61")
######################################################################################
# A function for automatic detection of GPUs installed (if autodetection is enabled)
# Usage:
# detect_installed_gpus(out_variable)
function(detect_installed_gpus out_variable)
if(NOT CUDA_gpu_detect_output)
set(cufile ${PROJECT_BINARY_DIR}/detect_cuda_archs.cu)
file(WRITE ${cufile} ""
"#include <cstdio>\n"
"int main() {\n"
" int count = 0;\n"
" if (cudaSuccess != cudaGetDeviceCount(&count)) return -1;\n"
" if (count == 0) return -1;\n"
" for (int device = 0; device < count; ++device) {\n"
" cudaDeviceProp prop;\n"
" if (cudaSuccess == cudaGetDeviceProperties(&prop, device))\n"
" std::printf(\"%d.%d \", prop.major, prop.minor);\n"
" }\n"
" return 0;\n"
"}\n")
execute_process(COMMAND "${CUDA_NVCC_EXECUTABLE}" "-ccbin=${CUDA_HOST_COMPILER}"
"--run" "${cufile}"
WORKING_DIRECTORY "${PROJECT_BINARY_DIR}/CMakeFiles/"
RESULT_VARIABLE nvcc_res OUTPUT_VARIABLE nvcc_out
ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
if(nvcc_res EQUAL 0)
# only keep the last line of nvcc_out
STRING(REGEX REPLACE ";" "\\\\;" nvcc_out "${nvcc_out}")
STRING(REGEX REPLACE "\n" ";" nvcc_out "${nvcc_out}")
list(GET nvcc_out -1 nvcc_out)
string(REPLACE "2.1" "2.1(2.0)" nvcc_out "${nvcc_out}")
set(CUDA_gpu_detect_output ${nvcc_out} CACHE INTERNAL "Returned GPU architetures from detect_installed_gpus tool" FORCE)
endif()
endif()
if(NOT CUDA_gpu_detect_output)
message(STATUS "Automatic GPU detection failed. Building for all known architectures.")
set(${out_variable} ${paddle_known_gpu_archs} PARENT_SCOPE)
else()
set(${out_variable} ${CUDA_gpu_detect_output} PARENT_SCOPE)
endif()
endfunction()
########################################################################
# Function for selecting GPU arch flags for nvcc based on CUDA_ARCH_NAME
# Usage:
# select_nvcc_arch_flags(out_variable)
function(select_nvcc_arch_flags out_variable)
# List of arch names
set(archs_names "Kepler" "Maxwell" "Pascal" "All" "Manual")
set(archs_name_default "All")
if(NOT CMAKE_CROSSCOMPILING)
list(APPEND archs_names "Auto")
endif()
# set CUDA_ARCH_NAME strings (so it will be seen as dropbox in CMake-Gui)
set(CUDA_ARCH_NAME ${archs_name_default} CACHE STRING "Select target NVIDIA GPU achitecture.")
set_property( CACHE CUDA_ARCH_NAME PROPERTY STRINGS "" ${archs_names} )
mark_as_advanced(CUDA_ARCH_NAME)
# verify CUDA_ARCH_NAME value
if(NOT ";${archs_names};" MATCHES ";${CUDA_ARCH_NAME};")
string(REPLACE ";" ", " archs_names "${archs_names}")
message(FATAL_ERROR "Only ${archs_names} architeture names are supported.")
endif()
if(${CUDA_ARCH_NAME} STREQUAL "Manual")
set(CUDA_ARCH_BIN ${paddle_known_gpu_archs} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
set(CUDA_ARCH_PTX "50" CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
mark_as_advanced(CUDA_ARCH_BIN CUDA_ARCH_PTX)
else()
unset(CUDA_ARCH_BIN CACHE)
unset(CUDA_ARCH_PTX CACHE)
endif()
if(${CUDA_ARCH_NAME} STREQUAL "Kepler")
set(cuda_arch_bin "30 35")
elseif(${CUDA_ARCH_NAME} STREQUAL "Maxwell")
set(cuda_arch_bin "50")
elseif(${CUDA_ARCH_NAME} STREQUAL "Pascal")
set(cuda_arch_bin "60 61")
elseif(${CUDA_ARCH_NAME} STREQUAL "Volta")
set(cuda_arch_bin "70")
elseif(${CUDA_ARCH_NAME} STREQUAL "All")
set(cuda_arch_bin ${paddle_known_gpu_archs})
elseif(${CUDA_ARCH_NAME} STREQUAL "Auto")
detect_installed_gpus(cuda_arch_bin)
else() # (${CUDA_ARCH_NAME} STREQUAL "Manual")
set(cuda_arch_bin ${CUDA_ARCH_BIN})
endif()
# remove dots and convert to lists
string(REGEX REPLACE "\\." "" cuda_arch_bin "${cuda_arch_bin}")
string(REGEX REPLACE "\\." "" cuda_arch_ptx "${CUDA_ARCH_PTX}")
string(REGEX MATCHALL "[0-9()]+" cuda_arch_bin "${cuda_arch_bin}")
string(REGEX MATCHALL "[0-9]+" cuda_arch_ptx "${cuda_arch_ptx}")
list(REMOVE_DUPLICATES cuda_arch_bin)
list(REMOVE_DUPLICATES cuda_arch_ptx)
set(nvcc_flags "")
set(nvcc_archs_readable "")
# Tell NVCC to add binaries for the specified GPUs
foreach(arch ${cuda_arch_bin})
if(arch MATCHES "([0-9]+)\\(([0-9]+)\\)")
# User explicitly specified PTX for the concrete BIN
list(APPEND nvcc_flags -gencode arch=compute_${CMAKE_MATCH_2},code=sm_${CMAKE_MATCH_1})
list(APPEND nvcc_archs_readable sm_${CMAKE_MATCH_1})
else()
# User didn't explicitly specify PTX for the concrete BIN, we assume PTX=BIN
list(APPEND nvcc_flags -gencode arch=compute_${arch},code=sm_${arch})
list(APPEND nvcc_archs_readable sm_${arch})
endif()
endforeach()
# Tell NVCC to add PTX intermediate code for the specified architectures
foreach(arch ${cuda_arch_ptx})
list(APPEND nvcc_flags -gencode arch=compute_${arch},code=compute_${arch})
list(APPEND nvcc_archs_readable compute_${arch})
endforeach()
string(REPLACE ";" " " nvcc_archs_readable "${nvcc_archs_readable}")
set(${out_variable} ${nvcc_flags} PARENT_SCOPE)
set(${out_variable}_readable ${nvcc_archs_readable} PARENT_SCOPE)
endfunction()
message(STATUS "CUDA detected: " ${CUDA_VERSION})
if (${CUDA_VERSION} LESS 7.0)
set(paddle_known_gpu_archs ${paddle_known_gpu_archs})
elseif (${CUDA_VERSION} LESS 8.0) # CUDA 7.x
set(paddle_known_gpu_archs ${paddle_known_gpu_archs7})
list(APPEND CUDA_NVCC_FLAGS "-D_MWAITXINTRIN_H_INCLUDED")
list(APPEND CUDA_NVCC_FLAGS "-D__STRICT_ANSI__")
elseif (${CUDA_VERSION} LESS 9.0) # CUDA 8.x
set(paddle_known_gpu_archs ${paddle_known_gpu_archs8})
list(APPEND CUDA_NVCC_FLAGS "-D_MWAITXINTRIN_H_INCLUDED")
list(APPEND CUDA_NVCC_FLAGS "-D__STRICT_ANSI__")
# CUDA 8 may complain that sm_20 is no longer supported. Suppress the
# warning for now.
list(APPEND CUDA_NVCC_FLAGS "-Wno-deprecated-gpu-targets")
endif()
include_directories(${CUDA_INCLUDE_DIRS})
list(APPEND EXTERNAL_LIBS ${CUDA_LIBRARIES} ${CUDA_rt_LIBRARY})
if(NOT WITH_DSO)
list(APPEND EXTERNAL_LIBS ${CUDNN_LIBRARY} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_curand_LIBRARY} ${NCCL_LIBRARY})
endif(NOT WITH_DSO)
# setting nvcc arch flags
select_nvcc_arch_flags(NVCC_FLAGS_EXTRA)
list(APPEND CUDA_NVCC_FLAGS ${NVCC_FLAGS_EXTRA})
message(STATUS "Added CUDA NVCC flags for: ${NVCC_FLAGS_EXTRA_readable}")
# Set C++11 support
set(CUDA_PROPAGATE_HOST_FLAGS OFF)
# Release/Debug flags set by cmake. Such as -O3 -g -DNDEBUG etc.
# So, don't set these flags here.
list(APPEND CUDA_NVCC_FLAGS "-std=c++11")
list(APPEND CUDA_NVCC_FLAGS "--use_fast_math")
list(APPEND CUDA_NVCC_FLAGS "-Xcompiler -fPIC")
# Set :expt-relaxed-constexpr to suppress Eigen warnings
list(APPEND CUDA_NVCC_FLAGS "--expt-relaxed-constexpr")
if(CMAKE_BUILD_TYPE STREQUAL "Debug")
list(APPEND CUDA_NVCC_FLAGS ${CMAKE_CXX_FLAGS_DEBUG})
elseif(CMAKE_BUILD_TYPE STREQUAL "Release")
list(APPEND CUDA_NVCC_FLAGS ${CMAKE_CXX_FLAGS_RELEASE})
elseif(CMAKE_BUILD_TYPE STREQUAL "RelWithDebInfo")
list(APPEND CUDA_NVCC_FLAGS ${CMAKE_CXX_FLAGS_RELWITHDEBINFO})
elseif(CMAKE_BUILD_TYPE STREQUAL "MinSizeRel")
list(APPEND CUDA_NVCC_FLAGS ${CMAKE_CXX_FLAGS_MINSIZEREL})
endif()
mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD)
mark_as_advanced(CUDA_SDK_ROOT_DIR CUDA_SEPARABLE_COMPILATION)

@ -0,0 +1,45 @@
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
IF(MOBILE_INFERENCE)
return()
ENDIF()
include (ExternalProject)
# NOTE: c-ares is needed when linking with grpc.
SET(CARES_SOURCES_DIR ${THIRD_PARTY_PATH}/cares)
SET(CARES_INSTALL_DIR ${THIRD_PARTY_PATH}/install/cares)
SET(CARES_INCLUDE_DIR "${CARES_INSTALL_DIR}/include/" CACHE PATH "cares include directory." FORCE)
ExternalProject_Add(
extern_cares
GIT_REPOSITORY "https://github.com/c-ares/c-ares.git"
GIT_TAG "cares-1_13_0"
PREFIX ${CARES_SOURCES_DIR}
UPDATE_COMMAND ""
CONFIGURE_COMMAND ./buildconf && ./configure --disable-shared --prefix=${CARES_INSTALL_DIR}
BUILD_IN_SOURCE 1
BUILD_COMMAND make
INSTALL_COMMAND make install
)
ADD_LIBRARY(cares STATIC IMPORTED GLOBAL)
SET_PROPERTY(TARGET cares PROPERTY IMPORTED_LOCATION
"${CARES_INSTALL_DIR}/lib/libcares.a")
include_directories(${CARES_INCLUDE_DIR})
ADD_DEPENDENCIES(cares extern_cares)

@ -28,15 +28,8 @@ INCLUDE_DIRECTORIES(${GFLAGS_INCLUDE_DIR})
ExternalProject_Add(
extern_gflags
${EXTERNAL_PROJECT_LOG_ARGS}
# TODO(yiwang): The annoying warnings mentioned in
# https://github.com/PaddlePaddle/Paddle/issues/3277 are caused by
# gflags. I fired a PR https://github.com/gflags/gflags/pull/230
# to fix it. Before it gets accepted by the gflags team, we use
# my personal fork, which contains above fix, temporarily. Let's
# change this back to the official Github repo once my PR is
# merged.
GIT_REPOSITORY "https://github.com/wangkuiyi/gflags.git"
GIT_TAG 986964c07427ecb9cdb5bd73f73ebbd40e54dadb
GIT_REPOSITORY "https://github.com/gflags/gflags.git"
GIT_TAG 77592648e3f3be87d6c7123eb81cbad75f9aef5a
PREFIX ${GFLAGS_SOURCES_DIR}
UPDATE_COMMAND ""
CMAKE_ARGS -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}

@ -0,0 +1,66 @@
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
IF(MOBILE_INFERENCE)
return()
ENDIF()
include (ExternalProject)
SET(GRPC_SOURCES_DIR ${THIRD_PARTY_PATH}/grpc)
SET(GRPC_INSTALL_DIR ${THIRD_PARTY_PATH}/install/grpc)
SET(GRPC_INCLUDE_DIR "${GRPC_INSTALL_DIR}/include/" CACHE PATH "grpc include directory." FORCE)
SET(GRPC_CPP_PLUGIN "${GRPC_INSTALL_DIR}/bin/grpc_cpp_plugin" CACHE FILEPATH "GRPC_CPP_PLUGIN" FORCE)
IF(APPLE)
SET(BUILD_CMD make -n | sed "s/-Werror//g" | sh)
ELSE()
SET(BUILD_CMD make)
ENDIF()
ExternalProject_Add(
extern_grpc
DEPENDS protobuf zlib
GIT_REPOSITORY "https://github.com/grpc/grpc.git"
GIT_TAG "v1.7.x"
PREFIX ${GRPC_SOURCES_DIR}
UPDATE_COMMAND ""
CONFIGURE_COMMAND ""
BUILD_IN_SOURCE 1
# NOTE(yuyang18):
# Disable -Werror, otherwise the compile will fail in MacOS.
# It seems that we cannot configure that by make command.
# Just dry run make command and remove `-Werror`, then use a shell to run make commands
BUILD_COMMAND ${BUILD_CMD} HAS_SYSTEM_PROTOBUF=false -s -j8 static grpc_cpp_plugin
INSTALL_COMMAND make prefix=${GRPC_INSTALL_DIR} install
)
# FIXME(typhoonzero): hack to get static lib path, try a better way like merge them.
ADD_LIBRARY(grpc++_unsecure STATIC IMPORTED GLOBAL)
SET_PROPERTY(TARGET grpc++_unsecure PROPERTY IMPORTED_LOCATION
"${GRPC_INSTALL_DIR}/lib/libgrpc++_unsecure.a")
ADD_LIBRARY(grpc++ STATIC IMPORTED GLOBAL)
SET_PROPERTY(TARGET grpc++ PROPERTY IMPORTED_LOCATION
"${GRPC_INSTALL_DIR}/lib/libgrpc++.a")
ADD_LIBRARY(gpr STATIC IMPORTED GLOBAL)
SET_PROPERTY(TARGET gpr PROPERTY IMPORTED_LOCATION
"${GRPC_INSTALL_DIR}/lib/libgpr.a")
ADD_LIBRARY(grpc_unsecure STATIC IMPORTED GLOBAL)
SET_PROPERTY(TARGET grpc_unsecure PROPERTY IMPORTED_LOCATION
"${GRPC_INSTALL_DIR}/lib/libgrpc_unsecure.a")
include_directories(${GRPC_INCLUDE_DIR})
ADD_DEPENDENCIES(grpc++_unsecure extern_grpc)

@ -40,10 +40,9 @@ INCLUDE_DIRECTORIES(${MKLDNN_INC_DIR})
IF(${CBLAS_PROVIDER} STREQUAL "MKLML")
SET(MKLDNN_DEPENDS ${MKLML_PROJECT})
SET(MKLDNN_MKLROOT ${MKLML_ROOT})
SET(MKLDNN_IOMP_LIB ${MKLML_IOMP_LIB})
SET(MKLDNN_IOMP_DIR ${MKLML_LIB_DIR})
MESSAGE(STATUS "Build MKLDNN with ${MKLDNN_MKLROOT}")
MESSAGE(STATUS "Build MKLDNN with MKLML ${MKLML_ROOT}")
ELSE()
MESSAGE(FATAL_ERROR "Should enable MKLML when build MKLDNN")
ENDIF()
SET(MKLDNN_CFLAG "${CMAKE_C_FLAGS} -Wno-error=strict-overflow")
@ -57,15 +56,16 @@ ExternalProject_Add(
PREFIX ${MKLDNN_SOURCES_DIR}
UPDATE_COMMAND ""
CMAKE_ARGS -DCMAKE_INSTALL_PREFIX=${MKLDNN_INSTALL_DIR}
CMAKE_ARGS -DMKLROOT=${MKLDNN_MKLROOT}
CMAKE_ARGS -DMKLROOT=${MKLML_ROOT}
CMAKE_ARGS -DCMAKE_C_FLAGS=${MKLDNN_CFLAG}
CMAKE_ARGS -DCMAKE_CXX_FLAGS=${MKLDNN_CXXFLAG}
CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${MKLDNN_INSTALL_DIR}
-DMKLROOT:PATH=${MKLDNN_MKLROOT}
-DMKLROOT:PATH=${MKLML_ROOT}
)
ADD_LIBRARY(mkldnn SHARED IMPORTED GLOBAL)
SET_PROPERTY(TARGET mkldnn PROPERTY IMPORTED_LOCATION ${MKLDNN_LIB})
ADD_DEPENDENCIES(mkldnn ${MKLDNN_PROJECT})
MESSAGE(STATUS "Mkldnn library: ${MKLDNN_LIB}")
MESSAGE(STATUS "MKLDNN library: ${MKLDNN_LIB}")
add_definitions(-DPADDLE_USE_MKLDNN)
LIST(APPEND external_project_dependencies mkldnn)

@ -45,15 +45,14 @@ IF(NOT ${CBLAS_FOUND})
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 ${OPTIONAL_ARGS} TARGET=ARMV7 ARM_SOFTFP_ABI=1 USE_THREAD=0)
ELSEIF(CMAKE_OSX_ARCHITECTURES MATCHES "arm64")
IF(CMAKE_OSX_ARCHITECTURES MATCHES "arm64")
SET(OPENBLAS_COMMIT "b5c96fcfcdc82945502a2303116a64d89985daf5")
SET(OPENBLAS_CC "${OPENBLAS_CC} ${CMAKE_C_FLAGS} -isysroot ${CMAKE_OSX_SYSROOT}")
SET(OPENBLAS_CC "${OPENBLAS_CC} -arch arm64")
SET(OPTIONAL_ARGS ${OPTIONAL_ARGS} TARGET=ARMV8 BINARY=64 USE_THREAD=0 CROSS_SUFFIX=${CROSS_SUFFIX})
ELSE()
MESSAGE(FATAL_ERROR "OpenBLAS only support arm64 architectures on iOS. "
"You can set IOS_USE_VECLIB_FOR_BLAS=ON or USE_EIGEN_FOR_BLAS=ON to use other blas library instead.")
ENDIF()
ELSEIF(RPI)
# use hardfp

@ -15,7 +15,18 @@
INCLUDE(ExternalProject)
# Always invoke `FIND_PACKAGE(Protobuf)` for importing function protobuf_generate_cpp
FIND_PACKAGE(Protobuf QUIET)
SET(PROTOBUF_FOUND "OFF")
macro(UNSET_VAR VAR_NAME)
UNSET(${VAR_NAME} CACHE)
UNSET(${VAR_NAME})
endmacro()
UNSET_VAR(PROTOBUF_INCLUDE_DIR)
UNSET_VAR(PROTOBUF_FOUND)
UNSET_VAR(PROTOBUF_PROTOC_EXECUTABLE)
UNSET_VAR(PROTOBUF_PROTOC_LIBRARY)
UNSET_VAR(PROTOBUF_LITE_LIBRARY)
UNSET_VAR(PROTOBUF_LIBRARY)
UNSET_VAR(PROTOBUF_INCLUDE_DIR)
UNSET_VAR(Protobuf_PROTOC_EXECUTABLE)
if(NOT COMMAND protobuf_generate_python) # before cmake 3.4, protobuf_genrerate_python is not defined.
function(protobuf_generate_python SRCS)
@ -110,7 +121,6 @@ macro(PROMPT_PROTOBUF_LIB)
# FIND_Protobuf.cmake uses `Protobuf_PROTOC_EXECUTABLE`.
# make `protobuf_generate_cpp` happy.
SET(Protobuf_PROTOC_EXECUTABLE ${PROTOBUF_PROTOC_EXECUTABLE})
FOREACH(dep ${protobuf_DEPS})
ADD_DEPENDENCIES(protobuf ${dep})
ADD_DEPENDENCIES(protobuf_lite ${dep})
@ -128,11 +138,11 @@ endmacro()
set(PROTOBUF_ROOT "" CACHE PATH "Folder contains protobuf")
if (NOT "${PROTOBUF_ROOT}" STREQUAL "")
find_path(PROTOBUF_INCLUDE_DIR google/protobuf/message.h PATHS ${PROTOBUF_ROOT}/include)
find_library(PROTOBUF_LIBRARY protobuf PATHS ${PROTOBUF_ROOT}/lib)
find_library(PROTOBUF_LITE_LIBRARY protobuf-lite PATHS ${PROTOBUF_ROOT}/lib)
find_library(PROTOBUF_PROTOC_LIBRARY protoc PATHS ${PROTOBUF_ROOT}/lib)
find_program(PROTOBUF_PROTOC_EXECUTABLE protoc PATHS ${PROTOBUF_ROOT}/bin)
find_path(PROTOBUF_INCLUDE_DIR google/protobuf/message.h PATHS ${PROTOBUF_ROOT}/include NO_DEFAULT_PATH)
find_library(PROTOBUF_LIBRARY protobuf PATHS ${PROTOBUF_ROOT}/lib NO_DEFAULT_PATH)
find_library(PROTOBUF_LITE_LIBRARY protobuf-lite PATHS ${PROTOBUF_ROOT}/lib NO_DEFAULT_PATH)
find_library(PROTOBUF_PROTOC_LIBRARY protoc PATHS ${PROTOBUF_ROOT}/lib NO_DEFAULT_PATH)
find_program(PROTOBUF_PROTOC_EXECUTABLE protoc PATHS ${PROTOBUF_ROOT}/bin NO_DEFAULT_PATH)
if (PROTOBUF_INCLUDE_DIR AND PROTOBUF_LIBRARY AND PROTOBUF_LITE_LIBRARY AND PROTOBUF_PROTOC_LIBRARY AND PROTOBUF_PROTOC_EXECUTABLE)
message(STATUS "Using custom protobuf library in ${PROTOBUF_ROOT}.")
SET_PROTOBUF_VERSION()

@ -12,6 +12,10 @@
# See the License for the specific language governing permissions and
# limitations under the License.
IF(MOBILE_INFERENCE)
return()
ENDIF()
INCLUDE(ExternalProject)
SET(WARPCTC_SOURCES_DIR ${THIRD_PARTY_PATH}/warpctc)

@ -50,6 +50,8 @@ ExternalProject_Add(
)
LIST(APPEND external_project_dependencies zlib)
ADD_LIBRARY(zlib_target STATIC IMPORTED GLOBAL)
SET_PROPERTY(TARGET zlib_target PROPERTY IMPORTED_LOCATION ${ZLIB_LIBRARIES})
IF(WITH_C_API)
INSTALL(DIRECTORY ${ZLIB_INCLUDE_DIR} DESTINATION third_party/zlib)

@ -149,58 +149,3 @@ endforeach()
foreach(flag ${GPU_COMMON_FLAGS})
safe_set_nvflag(${flag})
endforeach()
set(CUDA_PROPAGATE_HOST_FLAGS OFF)
# Release/Debug flags set by cmake. Such as -O3 -g -DNDEBUG etc.
# So, don't set these flags here.
LIST(APPEND CUDA_NVCC_FLAGS -std=c++11)
LIST(APPEND CUDA_NVCC_FLAGS --use_fast_math)
if(CMAKE_BUILD_TYPE STREQUAL "Debug")
LIST(APPEND CUDA_NVCC_FLAGS ${CMAKE_CXX_FLAGS_DEBUG})
elseif(CMAKE_BUILD_TYPE STREQUAL "Release")
LIST(APPEND CUDA_NVCC_FLAGS ${CMAKE_CXX_FLAGS_RELEASE})
elseif(CMAKE_BUILD_TYPE STREQUAL "RelWithDebInfo")
LIST(APPEND CUDA_NVCC_FLAGS ${CMAKE_CXX_FLAGS_RELWITHDEBINFO})
elseif(CMAKE_BUILD_TYPE STREQUAL "MinSizeRel")
LIST(APPEND CUDA_NVCC_FLAGS ${CMAKE_CXX_FLAGS_MINSIZEREL})
endif()
function(specify_cuda_arch cuda_version cuda_arch)
if(${cuda_version} VERSION_GREATER "8.0")
foreach(capability 61 62)
if(${cuda_arch} STREQUAL ${capability})
list(APPEND __arch_flags " -gencode arch=compute_${cuda_arch},code=sm_${cuda_arch}")
endif()
endforeach()
elseif(${cuda_version} VERSION_GREATER "7.0" and ${cuda_arch} STREQUAL "53")
list(APPEND __arch_flags " -gencode arch=compute_${cuda_arch},code=sm_${cuda_arch}")
endif()
endfunction()
# Common gpu architectures: Kepler, Maxwell
foreach(capability 30 35 50)
list(APPEND __arch_flags " -gencode arch=compute_${capability},code=sm_${capability}")
endforeach()
if (CUDA_VERSION VERSION_GREATER "7.0" OR CUDA_VERSION VERSION_EQUAL "7.0")
list(APPEND __arch_flags " -gencode arch=compute_52,code=sm_52")
endif()
# Modern gpu architectures: Pascal
if (CUDA_VERSION VERSION_GREATER "8.0" OR CUDA_VERSION VERSION_EQUAL "8.0")
list(APPEND __arch_flags " -gencode arch=compute_60,code=sm_60")
list(APPEND CUDA_NVCC_FLAGS --expt-relaxed-constexpr)
endif()
# Custom gpu architecture
set(CUDA_ARCH)
if(CUDA_ARCH)
specify_cuda_arch(${CUDA_VERSION} ${CUDA_ARCH})
endif()
set(CUDA_NVCC_FLAGS ${__arch_flags} ${CUDA_NVCC_FLAGS})

@ -459,11 +459,58 @@ function(py_test TARGET_NAME)
if(WITH_TESTING)
set(options STATIC static SHARED shared)
set(oneValueArgs "")
set(multiValueArgs SRCS DEPS)
cmake_parse_arguments(py_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
set(multiValueArgs SRCS DEPS ARGS)
cmake_parse_arguments(py_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
add_test(NAME ${TARGET_NAME}
COMMAND env PYTHONPATH=${PADDLE_PYTHON_BUILD_DIR}/lib-python
python2 ${py_test_SRCS}
${PYTHON_EXECUTABLE} -u ${py_test_SRCS} ${py_test_ARGS}
WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
endif()
endfunction()
# grpc_library generate grpc code using grpc_cpp_plugin and protoc
# then build the generated protobuf code and grpc code with your
# implementation source codes together. Use SRCS argument for your
# implementation source files and PROTO argument for your .proto
# files.
#
# Usage: grpc_library(my_target SRCS my_client.cc PROTO my_target.proto DEPS my_dep)
function(grpc_library TARGET_NAME)
set(oneValueArgs PROTO)
set(multiValueArgs SRCS DEPS)
set(options "")
cmake_parse_arguments(grpc_library "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
message(STATUS "generating grpc ${grpc_library_PROTO}")
get_filename_component(ABS_PROTO ${grpc_library_PROTO} ABSOLUTE)
get_filename_component(PROTO_WE ${grpc_library_PROTO} NAME_WE)
get_filename_component(PROTO_PATH ${ABS_PROTO} PATH)
protobuf_generate_cpp(grpc_proto_srcs grpc_proto_hdrs "${ABS_PROTO}")
set(grpc_grpc_srcs "${CMAKE_CURRENT_BINARY_DIR}/${PROTO_WE}.grpc.pb.cc")
set(grpc_grpc_hdrs "${CMAKE_CURRENT_BINARY_DIR}/${PROTO_WE}.grpc.pb.h")
cc_library("${TARGET_NAME}_proto" SRCS "${grpc_proto_srcs}")
add_custom_command(
OUTPUT "${grpc_grpc_srcs}" "${grpc_grpc_hdrs}"
COMMAND ${PROTOBUF_PROTOC_EXECUTABLE}
ARGS --grpc_out "${CMAKE_CURRENT_BINARY_DIR}" -I "${PROTO_PATH}"
--plugin=protoc-gen-grpc="${GRPC_CPP_PLUGIN}" "${ABS_PROTO}"
DEPENDS "${ABS_PROTO}" ${PROTOBUF_PROTOC_EXECUTABLE} extern_grpc)
# FIXME(typhoonzero): grpc generated code do not generate virtual-dtor, mark it
# as compiler warnings instead of error. Should try remove the warnings also.
set_source_files_properties(
${grpc_grpc_srcs}
PROPERTIES
COMPILE_FLAGS "-Wno-error=non-virtual-dtor -Wno-error=delete-non-virtual-dtor")
cc_library("${TARGET_NAME}_grpc" SRCS "${grpc_grpc_srcs}")
set_source_files_properties(
${grpc_library_SRCS}
PROPERTIES
COMPILE_FLAGS "-Wno-error=non-virtual-dtor -Wno-error=delete-non-virtual-dtor")
cc_library("${TARGET_NAME}" SRCS "${grpc_library_SRCS}" DEPS "${TARGET_NAME}_grpc" "${TARGET_NAME}_proto" "${grpc_library_DEPS}")
endfunction()

@ -115,8 +115,8 @@ function(link_paddle_exe TARGET_NAME)
target_link_libraries(${TARGET_NAME} log)
endif(ANDROID)
if(WITH_MKLDNN AND WITH_MKLML AND MKLDNN_IOMP_DIR)
target_link_libraries(${TARGET_NAME} "-L${MKLDNN_IOMP_DIR} -liomp5 -Wl,--as-needed")
if(WITH_MKLML AND MKLML_LIB_DIR AND MKLML_IOMP_LIB)
target_link_libraries(${TARGET_NAME} "-L${MKLML_LIB_DIR} -liomp5 -Wl,--as-needed")
endif()
add_dependencies(${TARGET_NAME} ${external_project_dependencies})
@ -168,17 +168,3 @@ function(create_resources res_file output_file)
COMMAND python ARGS ${PADDLE_SOURCE_DIR}/cmake/make_resource.py ${res_file} ${output_file}
DEPENDS ${res_file} ${PADDLE_SOURCE_DIR}/cmake/make_resource.py)
endfunction()
# Create a python unittest using run_python_tests.sh,
# which takes care of making correct running environment
function(add_python_test TEST_NAME)
foreach(arg ${ARGN})
get_filename_component(py_fn ${arg} NAME_WE)
set(TRG_NAME ${TEST_NAME}_${py_fn})
add_test(NAME ${TRG_NAME}
COMMAND env PYTHONPATH=${PADDLE_PYTHON_PACKAGE_DIR}
python2 ${arg}
WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
endforeach()
endfunction()

@ -54,7 +54,7 @@ img_conv
.. _api_v2.layer_context_projection:
context_projection
context_projection
------------------
.. autoclass:: paddle.v2.layer.context_projection
:noindex:
@ -70,7 +70,7 @@ Image Pooling Layer
img_pool
--------
.. autoclass:: paddle.v2.layer.img_pool
:noindex:
:noindex:
spp
---
@ -104,7 +104,7 @@ sum_to_one_norm
---------------
.. autoclass:: paddle.v2.layer.sum_to_one_norm
:noindex:
cross_channel_norm
------------------
.. autoclass:: paddle.v2.layer.cross_channel_norm
@ -114,7 +114,7 @@ row_l2_norm
-----------
.. autoclass:: paddle.v2.layer.row_l2_norm
:noindex:
Recurrent Layers
================
@ -335,6 +335,16 @@ bilinear_interp
.. autoclass:: paddle.v2.layer.bilinear_interp
:noindex:
dot_prod
---------
.. autoclass:: paddle.v2.layer.dot_prod
:noindex:
out_prod
--------
.. autoclass:: paddle.v2.layer.out_prod
:noindex:
power
-----
.. autoclass:: paddle.v2.layer.power
@ -372,6 +382,11 @@ cos_sim
.. autoclass:: paddle.v2.layer.cos_sim
:noindex:
l2_distance
-----------
.. autoclass:: paddle.v2.layer.l2_distance
:noindex:
trans
-----
.. autoclass:: paddle.v2.layer.trans
@ -400,6 +415,13 @@ multiplex
.. autoclass:: paddle.v2.layer.multiplex
:noindex:
Factorization Machine Layer
============================
factorization_machine
---------------------
.. autoclass:: paddle.v2.layer.factorization_machine
:noindex:
Slicing and Joining Layers
==========================

@ -28,6 +28,51 @@ The goal of float16 is to serve as a key for the executor to find and run the co
- [Eigen](https://github.com/RLovelett/eigen) >= 3.3 supports float16 calculation on both GPU and CPU using the `Eigen::half` class. It is mostly useful for Nvidia GPUs because of the overloaded arithmetic operators using cuda intrinsics. It falls back to using software emulation on CPU for calculation and there is no special treatment to ARM processors.
- [ARM compute library](https://github.com/ARM-software/ComputeLibrary) >= 17.02.01 supports NEON FP16 kernels (requires ARMv8.2-A CPU).
### CUDA version issue
There are currently three versions of CUDA that supports `__half` data type, namely, CUDA 7.5, 8.0, and 9.0.
CUDA 7.5 and 8.0 define `__half` as a simple struct that has a `uint16_t` data (see [`cuda_fp16.h`](https://github.com/ptillet/isaac/blob/9212ab5a3ddbe48f30ef373f9c1fb546804c7a8c/include/isaac/external/CUDA/cuda_fp16.h)) as follows:
```
typedef struct __align__(2) {
unsigned short x;
} __half;
typedef __half half;
```
This struct does not define any overloaded arithmetic operators. So you have to directly use `__hadd` instead of `+` to correctly add two half types:
```
__global__ void Add() {
half a, b, c;
c = __hadd(a, b); // correct
c = a + b; // compiler error: no operator "+" matches these operands
}
```
CUDA 9.0 provides a major update to the half data type. The related code can be found in the updated [`cuda_fp16.h`](https://github.com/ptillet/isaac/blob/master/include/isaac/external/CUDA/cuda_fp16.h) and the newly added [`cuda_fp16.hpp`](https://github.com/ptillet/isaac/blob/master/include/isaac/external/CUDA/cuda_fp16.hpp).
Essentially, CUDA 9.0 renames the original `__half` type in 7.5 and 8.0 as `__half_raw`, and defines a new `__half` class type that has constructors, conversion operators, and also provides overloaded arithmetic operators such as follows:
```
typedef struct __CUDA_ALIGN__(2) {
unsigned short x;
} __half_raw;
struct __CUDA_ALIGN__(2) __half {
protected:
unsigned short __x;
public:
// constructors and conversion operators from/to
// __half_raw and other built-in data types
}
typedef __half half;
__device__ __forceinline__
__half operator+(const __half &lh, const __half &rh) {
return __hadd(lh, rh);
}
// Other overloaded operators
```
This new design makes `c = a + b` work correctly for CUDA half data type.
## Implementation
The float16 class holds a 16-bit `uint16_t` data internally.

@ -36,13 +36,13 @@ Figure 1. PaddlePaddle on IA.
我们把集成方案大致分为了如下几个方面。
### CMake
我们会在`CMakeLists.txt`中会添加`WITH_MKLDNN`的选项,当设置这个值为`ON`的时候会启用编译MKL-DNN功能。同时会自动开启OpenMP用于提高MKL-DNN的性能
我们会在`CMakeLists.txt`中会给用户添加一个`WITH_MKL`的开关,他是负责`WITH_MKLML`和`WITH_MKLDNN`的总开关
同时,我们会引入`WITH_MKLML`选项用于选择是否使用MKL-DNN自带的MKLML安装包。这个安装包可以独立于MKL-DNN使用但是建议在开启MKL-DNN的同时也打开MKLML的开关这样才能发挥最好的性能。
当打开`WITH_MKL`时会开启MKLML的功能作为PaddlePaddle的CBLAS和LAPACK库同时会开启Intel OpenMP用于提高MKLML的性能。 如果系统支持AVX2指令集及以上同时会开启MKL-DNN功能。
所以,我们会在`cmake/external`目录新建`mkldnn.cmake`和`mklml.cmake`文件它们会在编译PaddlePaddle的时候下载对应的软件包并放到PaddlePaddle的third party目录中
当关闭`WITH_MKL`时MKLML和MKL-DNN功能会同时关闭
**备注**:当`WITH_MKLML=ON`的时候会优先使用这个包作为PaddlePaddle的CBLAS和LAPACK库所以会稍微改动`cmake/cblas.cmake`中的逻辑
所以,我们会在`cmake/external`目录新建`mkldnn.cmake`和`mklml.cmake`文件它们会在编译PaddlePaddle的时候下载对应的软件包并放到PaddlePaddle的third party目录中
### Layers
所有MKL-DNN相关的C++ layers都会按照PaddlePaddle的目录结构存放在

@ -1,25 +1,25 @@
# Python Data Reader Design Doc
At training and testing time, PaddlePaddle programs need to read data. To ease the users' work to write data reading code, we define that
During the training and testing phases, PaddlePaddle programs need to read data. To help the users write code that performs reading input data, we define the following:
- A *reader* is a function that reads data (from file, network, random number generator, etc) and yields data items.
- A *reader creator* is a function that returns a reader function.
- A *reader decorator* is a function, which accepts one or more readers, and returns a reader.
- A *batch reader* is a function that reads data (from *reader*, file, network, random number generator, etc) and yields a batch of data items.
- A *reader*: A function that reads data (from file, network, random number generator, etc) and yields the data items.
- A *reader creator*: A function that returns a reader function.
- A *reader decorator*: A function, which takes in one or more readers, and returns a reader.
- A *batch reader*: A function that reads data (from *reader*, file, network, random number generator, etc) and yields a batch of data items.
and provide function which converts reader to batch reader, frequently used reader creators and reader decorators.
and also provide a function which can convert a reader to a batch reader, frequently used reader creators and reader decorators.
## Data Reader Interface
Indeed, *data reader* doesn't have to be a function that reads and yields data items. It can be any function with no parameter that creates a iterable (anything can be used in `for x in iterable`):
*Data reader* doesn't have to be a function that reads and yields data items. It can just be any function without any parameters that creates an iterable (anything can be used in `for x in iterable`) as follows:
```
iterable = data_reader()
```
Element produced from the iterable should be a **single** entry of data, **not** a mini batch. That entry of data could be a single item, or a tuple of items. Item should be of [supported type](http://www.paddlepaddle.org/doc/ui/data_provider/pydataprovider2.html?highlight=dense_vector#input-types) (e.g., numpy 1d array of float32, int, list of int)
The item produced from the iterable should be a **single** entry of data and **not** a mini batch. The entry of data could be a single item or a tuple of items. Item should be of one of the [supported types](http://www.paddlepaddle.org/doc/ui/data_provider/pydataprovider2.html?highlight=dense_vector#input-types) (e.g., numpy 1d array of float32, int, list of int etc.)
An example implementation for single item data reader creator:
An example implementation for single item data reader creator is as follows:
```python
def reader_creator_random_image(width, height):
@ -29,7 +29,7 @@ def reader_creator_random_image(width, height):
return reader
```
An example implementation for multiple item data reader creator:
An example implementation for multiple item data reader creator is as follows:
```python
def reader_creator_random_image_and_label(width, height, label):
def reader():
@ -40,9 +40,10 @@ def reader_creator_random_image_and_label(width, height, label):
## Batch Reader Interface
*batch reader* can be any function with no parameter that creates a iterable (anything can be used in `for x in iterable`). The output of the iterable should be a batch (list) of data items. Each item inside the list must be a tuple.
*Batch reader* can be any function without any parameters that creates an iterable (anything can be used in `for x in iterable`). The output of the iterable should be a batch (list) of data items. Each item inside the list should be a tuple.
Here are some valid outputs:
Here are valid outputs:
```python
# a mini batch of three data items. Each data item consist three columns of data, each of which is 1.
[(1, 1, 1),
@ -58,20 +59,22 @@ Here are valid outputs:
Please note that each item inside the list must be a tuple, below is an invalid output:
```python
# wrong, [1,1,1] needs to be inside a tuple: ([1,1,1],).
# Otherwise it's ambiguous whether [1,1,1] means a single column of data [1, 1, 1],
# or three column of datas, each of which is 1.
# Otherwise it is ambiguous whether [1,1,1] means a single column of data [1, 1, 1],
# or three columns of data, each of which is 1.
[[1,1,1],
[2,2,2],
[3,3,3]]
```
It's easy to convert from reader to batch reader:
It is easy to convert from a reader to a batch reader:
```python
mnist_train = paddle.dataset.mnist.train()
mnist_train_batch_reader = paddle.batch(mnist_train, 128)
```
Also easy to create custom batch reader:
It is also straight forward to create a custom batch reader:
```python
def custom_batch_reader():
while True:
@ -85,7 +88,8 @@ mnist_random_image_batch_reader = custom_batch_reader
## Usage
batch reader, mapping from item(s) read to data layer, batch size and number of total pass will be passed into `paddle.train`:
Following is how we can use the reader with PaddlePaddle:
The batch reader, a mapping from item(s) to data layer, the batch size and the number of total passes will be passed into `paddle.train` as follows:
```python
# two data layer is created:
@ -99,13 +103,13 @@ paddle.train(batch_reader, {"image":0, "label":1}, 128, 10, ...)
## Data Reader Decorator
*Data reader decorator* takes a single or multiple data reader, returns a new data reader. It is similar to a [python decorator](https://wiki.python.org/moin/PythonDecorators), but it does not use `@` syntax.
The *Data reader decorator* takes in a single reader or multiple data readers and returns a new data reader. It is similar to a [python decorator](https://wiki.python.org/moin/PythonDecorators), but it does not use `@` in the syntax.
Since we have a strict interface for data readers (no parameter, return a single data item). Data reader can be used flexiable via data reader decorators. Following are a few examples:
Since we have a strict interface for data readers (no parameters and return a single data item), a data reader can be used in a flexible way using data reader decorators. Following are a few examples:
### Prefetch Data
Since reading data may take time and training can not proceed without data. It is generally a good idea to prefetch data.
Since reading data may take some time and training can not proceed without data, it is generally a good idea to prefetch the data.
Use `paddle.reader.buffered` to prefetch data:
@ -117,9 +121,9 @@ buffered_reader = paddle.reader.buffered(paddle.dataset.mnist.train(), 100)
### Compose Multiple Data Readers
For example, we want to use a source of real images (reusing mnist dataset), and a source of random images as input for [Generative Adversarial Networks](https://arxiv.org/abs/1406.2661).
For example, if we want to use a source of real images (say reusing mnist dataset), and a source of random images as input for [Generative Adversarial Networks](https://arxiv.org/abs/1406.2661).
We can do:
We can do the following :
```python
def reader_creator_random_image(width, height):
@ -139,13 +143,13 @@ false_reader = reader_creator_bool(False)
reader = paddle.reader.compose(paddle.dataset.mnist.train(), data_reader_creator_random_image(20, 20), true_reader, false_reader)
# Skipped 1 because paddle.dataset.mnist.train() produces two items per data entry.
# And we don't care second item at this time.
# And we don't care about the second item at this time.
paddle.train(paddle.batch(reader, 128), {"true_image":0, "fake_image": 2, "true_label": 3, "false_label": 4}, ...)
```
### Shuffle
Given shuffle buffer size `n`, `paddle.reader.shuffle` will return a data reader that buffers `n` data entries and shuffle them before a data entry is read.
Given the shuffle buffer size `n`, `paddle.reader.shuffle` returns a data reader that buffers `n` data entries and shuffles them before a data entry is read.
Example:
```python
@ -154,21 +158,21 @@ reader = paddle.reader.shuffle(paddle.dataset.mnist.train(), 512)
## Q & A
### Why reader return only a single entry, but not a mini batch?
### Why does a reader return only a single entry, and not a mini batch?
Always returning a single entry make reusing existing data readers much easier (e.g., if existing reader return not a single entry but 3 entries, training code will be more complex because it need to handle cases like batch size 2).
Returning a single entry makes reusing existing data readers much easier (for example, if an existing reader returns 3 entries instead if a single entry, the training code will be more complicated because it need to handle cases like a batch size 2).
We provide function `paddle.batch` to turn (single entry) reader into batch reader.
We provide a function: `paddle.batch` to turn (a single entry) reader into a batch reader.
### Why do we need batch reader, isn't train take reader and batch_size as arguments sufficient?
### Why do we need a batch reader, isn't is sufficient to give the reader and batch_size as arguments during training ?
In most of the case, train taking reader and batch_size as arguments would be sufficent. However sometimes user want to customize order of data entries inside a mini batch. Or even change batch size dynamically.
In most of the cases, it would be sufficient to give the reader and batch_size as arguments to the train method. However sometimes the user wants to customize the order of data entries inside a mini batch, or even change the batch size dynamically. For these cases using a batch reader is very efficient and helpful.
### Why use a dictionary but not a list to provide mapping?
### Why use a dictionary instead of a list to provide mapping?
We decided to use dictionary (`{"image":0, "label":1}`) instead of list (`["image", "label"]`) is because that user can easily resue item (e.g., using `{"image_a":0, "image_b":0, "label":1}`) or skip item (e.g., using `{"image_a":0, "label":2}`).
Using a dictionary (`{"image":0, "label":1}`) instead of a list (`["image", "label"]`) gives the advantage that the user can easily reuse the items (e.g., using `{"image_a":0, "image_b":0, "label":1}`) or even skip an item (e.g., using `{"image_a":0, "label":2}`).
### How to create custom data reader creator
### How to create a custom data reader creator ?
```python
def image_reader_creator(image_path, label_path, n):
@ -192,7 +196,7 @@ paddle.train(paddle.batch(reader, 128), {"image":0, "label":1}, ...)
### How is `paddle.train` implemented
An example implementation of paddle.train could be:
An example implementation of paddle.train is:
```python
def train(batch_reader, mapping, batch_size, total_pass):

Some files were not shown because too many files have changed in this diff Show More

Loading…
Cancel
Save