-
Notifications
You must be signed in to change notification settings - Fork 741
/
cppbuild.sh
executable file
·217 lines (190 loc) · 9.04 KB
/
cppbuild.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
#!/bin/bash
# This file is meant to be included by the parent cppbuild.sh script
if [[ -z "$PLATFORM" ]]; then
pushd ..
bash cppbuild.sh "$@" mxnet
popd
exit
fi
export ADD_CFLAGS="-DMXNET_USE_LAPACK=1"
export ADD_LDFLAGS=
export USE_OPENMP=1
export CUDA_ARCH=-arch=sm_35
export USE_CUDA=0
export USE_CUDNN=0
export USE_NCCL=0
export USE_CUDA_PATH=
export NVCC=
export USE_MKLDNN=1
if [[ "$EXTENSION" == *gpu ]]; then
export ADD_CFLAGS="$ADD_CFLAGS -DMXNET_USE_CUDA=1"
export USE_CUDA=1
export USE_CUDNN=1
if [[ "$PLATFORM" == linux* ]]; then
export USE_NCCL=1
fi
export USE_CUDA_PATH="/usr/local/cuda"
export NVCC="$USE_CUDA_PATH/bin/nvcc"
export ACTUAL_GCC_HOST_COMPILER_PATH=$(which -a gcc | grep -v /ccache/ | head -1) # skip ccache
fi
MXNET_VERSION=1.9.1
download https://downloads.apache.org/mxnet/$MXNET_VERSION/apache-mxnet-src-$MXNET_VERSION-incubating.tar.gz apache-mxnet-src-$MXNET_VERSION-incubating.tar.gz
#download https://github.com/apache/mxnet/releases/download/$MXNET_VERSION/apache-mxnet-src-$MXNET_VERSION-incubating.tar.gz apache-mxnet-src-$MXNET_VERSION-incubating.tar.gz
mkdir -p "$PLATFORM$EXTENSION"
cd "$PLATFORM$EXTENSION"
INSTALL_PATH=`pwd`
if [[ -f "lib/libmxnet.so" ]]; then
echo libmxnet.so already exists!
return 0
elif [[ -f "bin/libmxnet.dll" ]]; then
echo libmxnet.dll already exists!
return 0
fi
OPENCV_PATH="$INSTALL_PATH/../../../opencv/cppbuild/$PLATFORM/"
OPENBLAS_PATH="$INSTALL_PATH/../../../openblas/cppbuild/$PLATFORM/"
if [[ -n "${BUILD_PATH:-}" ]]; then
PREVIFS="$IFS"
IFS="$BUILD_PATH_SEPARATOR"
for P in $BUILD_PATH; do
if [[ -d "$P/include/opencv2" ]]; then
OPENCV_PATH="$P"
elif [[ -f "$P/include/openblas_config.h" ]]; then
OPENBLAS_PATH="$P"
fi
done
IFS="$PREVIFS"
fi
echo "Decompressing archives... (ignore any errors)"
tar --totals -xzf ../apache-mxnet-src-$MXNET_VERSION-incubating.tar.gz || true
cd apache-mxnet-src-$MXNET_VERSION-incubating
# https://github.com/apache/incubator-mxnet/pull/20207
patch -Np1 < ../../../mxnet.patch
# upgrade MKL-DNN
#sedinplace 's/0.18.1/0.19/g' 3rdparty/mkldnn/CMakeLists.txt
#sedinplace 's/0.18/0.19/g' 3rdparty/mkldnn/scripts/prepare_mkl.bat 3rdparty/mkldnn/scripts/prepare_mkl.sh
#sedinplace 's/2019.0.3.20190220/2019.0.5.20190502/g' 3rdparty/mkldnn/scripts/prepare_mkl.bat 3rdparty/mkldnn/scripts/prepare_mkl.sh
# actually use LAPACK from OpenBLAS
sedinplace 's/USE_LAPACK = 0/USE_LAPACK = 1/g' Makefile
sedinplace '/LDFLAGS += -llapack/d' Makefile
# patch up compile errors
sedinplace "s/cmake/$CMAKE/g" mkldnn.mk
sedinplace 's/-Werror//g' 3rdparty/mkldnn/cmake/platform.cmake
sedinplace '/include_directories("3rdparty\/nvidia_cub")/d' CMakeLists.txt
sedinplace 's/kCPU/Context::kCPU/g' src/operator/tensor/elemwise_binary_scalar_op_basic.cc
sedinplace 's:../../src/operator/tensor/:./:g' src/operator/tensor/cast_storage-inl.h
sedinplace 's/-Xcompiler "$(CFLAGS)/"-Xcompiler=$(CFLAGS)/g' Makefile
sedinplace '/CHECK(mshadow::DataType<DType>::kFlag == type_flag_)/{N;N;N;d;}' include/mxnet/tensor_blob.h
sedinplace 's/std::pow/powf/g' src/operator/contrib/multi_lamb.cu
sedinplace 's/round(/roundf(/g' src/operator/*.cu src/operator/contrib/*.cu
sedinplace 's/floor(/floorf(/g' src/operator/*.cu src/operator/contrib/*.cu
sedinplace 's/ceil(/ceilf(/g' src/operator/*.cu src/operator/contrib/*.cu
sedinplace '/#include <opencv2\/opencv.hpp>/a\
#include <opencv2/imgproc/types_c.h>\
' src/io/image_augmenter.h src/io/image_io.cc tools/im2rec.cc
sedinplace 's/CV_LOAD_IMAGE_COLOR/cv::IMREAD_COLOR/g' tools/im2rec.cc
sedinplace 's/CV_IMWRITE_PNG_COMPRESSION/cv::IMWRITE_PNG_COMPRESSION/g' tools/im2rec.cc
sedinplace 's/CV_IMWRITE_JPEG_QUALITY/cv::IMWRITE_JPEG_QUALITY/g' tools/im2rec.cc
# note: MXNet needs full path to ccache wrappers to use them
case $PLATFORM in
linux-x86)
export CC="$(which gcc) -m32"
export CXX="$(which g++) -m32"
export BLAS="openblas"
export USE_INTGEMM=0
export USE_MKLDNN=0
;;
linux-x86_64)
export CC="$(which gcc) -m64"
export CXX="$(which g++) -m64"
export BLAS="openblas"
sedinplace "s/-std=c++11/-std=c++14/g" Makefile
# libmklml_intel.so does not have a SONAME, so libmkldnn.so.0 needs an RPATH to be able to load
# sedinplace "s/$CMAKE/$CMAKE -DCMAKE_CXX_FLAGS='-Wl,-rpath,\$\$ORIGIN\/'/g" mkldnn.mk
if [[ -f /usr/local/cuda/bin/nvcccache ]]; then
export NVCC=/usr/local/cuda/bin/nvcccache
sedinplace 's/-ccbin $(CXX)/-ccbin $(ACTUAL_GCC_HOST_COMPILER_PATH) -Xcompiler -fPIC -Xcompiler -O3/g' Makefile
fi
;;
macosx-*)
# remove harmful changes to rpath
sedinplace '/install_name_tool/d' Makefile
export CC="$(which clang)"
export CXX="$(which clang++)"
export BLAS="openblas"
export USE_OPENMP=0
# export USE_OPENMP=1
# export ADD_CFLAGS="$ADD_CFLAGS -I/usr/local/include"
# export ADD_CFLAGS="$ADD_CFLAGS -L/usr/local/lib -lomp"
;;
windows-x86_64)
export CC="cl.exe"
export CXX="cl.exe"
# copy include files
mkdir -p ../include
cp -RLf include/mxnet 3rdparty/dmlc-core/include/dmlc 3rdparty/mshadow/mshadow ../include
# configure the build
mkdir -p ../build
cd ../build
USE_X="-DMXNET_CUDA_ARCH=3.5+PTX -DCUDA_ARCH_LIST=3.5+PTX -DUSE_CUDA=$USE_CUDA -DUSE_CUDNN=$USE_CUDNN -DUSE_OPENCV=ON -DUSE_MKLDNN=$USE_MKLDNN"
OPENCV="-DOpenCV_DIR=$OPENCV_PATH/ -DOpenCV_CONFIG_PATH=$OPENCV_PATH/"
OPENBLAS="-DOpenBLAS_INCLUDE_DIR=$OPENBLAS_PATH/include/ -DOpenBLAS_LIB=$OPENBLAS_PATH/lib/openblas.lib"
"$CMAKE" -G "Ninja" -DCMAKE_BUILD_TYPE=Release $USE_X $OPENCV $OPENBLAS ../apache-mxnet-src-$MXNET_VERSION-incubating
# try to rebuild the project without compiler parallelism to avoid "out of heap space"
ninja -j $MAKEJ || ninja -j 1
# copy binary files
mkdir -p ../bin
cp -f *.dll ../bin
# copy library files
mkdir -p ../lib
cp -f libmxnet.lib ../lib/mxnet.lib
cp -f mxnet_??.lib ../lib/mxnet.lib || true
# finish
cd ../apache-mxnet-src-$MXNET_VERSION-incubating
;;
*)
echo "Error: Platform \"$PLATFORM\" is not supported"
return 0
;;
esac
if [[ ! "$PLATFORM" == windows* ]]; then
export C_INCLUDE_PATH="$OPENBLAS_PATH/include/:$OPENCV_PATH/include/"
export CPLUS_INCLUDE_PATH="$C_INCLUDE_PATH"
export LIBRARY_PATH="$OPENBLAS_PATH/:$OPENBLAS_PATH/lib/:$OPENCV_PATH/:$OPENCV_PATH/lib/"
sed -i="" 's/$(shell pkg-config --cflags opencv)//' Makefile
sed -i="" 's/$(shell pkg-config --libs opencv)/-lopencv_highgui -lopencv_imgcodecs -lopencv_imgproc -lopencv_core/' Makefile
sed -i="" 's/$(shell pkg-config --cflags $(OPENCV_LIB))//' Makefile
sed -i="" 's/$(shell pkg-config --libs-only-L $(OPENCV_LIB))//' Makefile
sed -i="" 's/$(filter -lopencv_imgcodecs -lopencv_highgui, $(shell pkg-config --libs-only-l $(OPENCV_LIB)))/-lopencv_highgui -lopencv_imgcodecs/' Makefile
make -s -j $MAKEJ CC="$CC" CXX="$CXX" USE_BLAS="$BLAS" USE_OPENMP="$USE_OPENMP" CUDA_ARCH="$CUDA_ARCH" USE_CUDA="$USE_CUDA" USE_CUDNN="$USE_CUDNN" USE_NCCL="$USE_NCCL" USE_CUDA_PATH="$USE_CUDA_PATH" NVCC="$NVCC" USE_MKLDNN="$USE_MKLDNN" USE_F16C=0 ADD_CFLAGS="$ADD_CFLAGS" ADD_LDFLAGS="$ADD_LDFLAGS" lib/libmxnet.a lib/libmxnet.so
cp -RLf include lib 3rdparty/dmlc-core/include ..
cp -RLf 3rdparty/mshadow/mshadow ../include
unset CC
unset CXX
fi
# add Scala files we cannot get with CMake for Windows
# cd scala-package
# mvn clean install -Dmaven.test.skip
# rm -Rf `find ./ -iname target`
patch -Np1 < ../../../mxnet-scala.patch || true
# use mxnet-scala-cuda.patch to get a couple of CUDA ops,
# but an actual GPU becomes necessary for the build
# copy official JNI functions and adjust include directives
cp -RLf 3rdparty/dlpack/include/dlpack 3rdparty/tvm/nnvm/include/* ../include
cp -f src/common/cuda_utils.h scala-package/init-native/src/main/native/* scala-package/native/src/main/native/* ../include
sedinplace 's:../src/common/::g' ../include/*.cc
# copy/adjust Scala source files and work around loader issue in Base.scala
mkdir -p ../scala/init ../scala/core
cp -RLf scala-package/init/src/main/scala/* ../scala/init
cp -RLf scala-package/macros/src/main/scala/* ../scala/init
cp -RLf scala-package/core/src/main/scala/* ../scala/core
cp -RLf scala-package/infer/src/main/scala/* ../scala/core
cp -RLf scala-package/spark/src/main/scala/* ../scala/core
sedinplace 's/ tryLoadInitLibrary()/ org.bytedeco.javacpp.Loader.load(classOf[org.bytedeco.mxnet.presets.mxnet])/g' ../scala/init/org/apache/mxnet/init/Base.scala
sedinplace 's/ tryLoadLibraryOS("mxnet-scala")/ org.bytedeco.javacpp.Loader.load(classOf[org.bytedeco.mxnet.presets.mxnet])/g' ../scala/core/org/apache/mxnet/Base.scala
# fix library with correct rpath on Mac
case $PLATFORM in
macosx-*)
install_name_tool -add_rpath @loader_path/. -id @rpath/libmxnet.so ../lib/libmxnet.so
;;
esac
cd ../..