docker中cuda9.2+cudnn7+caffe

万事开头难, 所以尽量找别人开好的头, 现在nvidia官方给的都cuda9.2了, 开源的gpu-caffe基本还是cuda8.0, 经过综合考虑, 还是在docker上面重新搭建, 这样不影响本地主机的环境又方便以后的移植部署.

     1. Pull docker image. (docker的安装请自行百度下.)

# sudo docker pull nvidia/cuda:9.2-cudnn7-devel-ubuntu16.04 

     2. Git clone caffe source. (Git的安装我是用 apt安装的)

# cd workspace && git clone https://github.com/BVLC/caffe.git

    3. 启动docker 镜像, 记得一定要用nvidia-docker命令,这个不是docker自带的, 安装请百度下.

# sudo nvidia-docker run -it -v #HOME/workspace:/var/workspace --name cuda-caffe nvidia/cuda:9.2-cudnn7-devel-ubuntu16.04 /bin/bash

Ps:启动过后, 以后进入docker的bash需要执行

# sudo nvidia-docker exec -it cuda-caffe /bin/bash

或者

# sudo docker exec -it cuda-caffe /bin/bash

  4. 准备caffe环境, 参考的caffe官网 installation, 其实这里说得很清楚了, 我这里也是做一个搬运工.

         a. Cuda, 官方说需要cuda7以上版本, 以前的版本也是可以的, 但是不保证没得小问题. 当然我们docker image直接是cuda9.2,所以不考虑这个问题.

         b. Blas, 好像是一个数学的计算库, 包括矩阵计算啥的, 有MKL, ATLAS, OpenBLAS等都可以, 默认支持MKL, 这个选择不同配置的makefile-config. 我看OpenBLAS亲切, 于是就上的OpenBLAS

# apt install libopenblas-dev

         c. Boost, 这个是c++的一个封装库, 很强大, 有很多模块. 如果不考虑pycaffe啥的, 可以直接用apt安装(# apt install libboost-all-dev), 但是我这里用的anaconda安装的python3.6, 考虑到apt安装的默认python3.5版本不兼容, 所以下载的boost重新编译的.

Boost没有立即git最新版本, 是以前下载的(1.66.0), 但和最新版本应该不超过三个月, 相差不大, 所以最新版本(1.68.0)应该也没有很大的问题, 灵活处理, 欢迎交流.

下载,编译与安装:

# cd workspace && wget https://dl.bintray.com/boostorg/release/1.66.0/source/boost_1_66_0.tar.gz

# tar xf boost_1_66_0.tar.gz

# cd boost_1_66_0

# ./bootstrap.sh --with-libraries=python --with-toolset=gcc

# ./b2 cflags='-fPIC' cxxflags='-fPIC' --with-python include="/root/anaconda3/include/python3.6m/" # 这里的include根据自己的python环境进行修改即可

# ./b2 install

         d. Protobuf, gflags, hdf5, glog. 前面三个在Ubuntu上可以用apt进行安装, 最后一个要git源码安装.

# apt install libprotobuf-dev libgflags-dev libhdf5-dev

 

# cd $HOME/workspace && git clone https://github.com/google/glog.git

# cd glog

# ./autogen.sh && ./configure && make && make install #如果没有安装autogen工具, 百度安装下即可

         e. Opencv 是现在霸屏似得图像处理工具, caffe官方推荐版本要大于2.4, 当然3.0以后也是没得问题的. 当然不知道怎么的, opencv3.3以后的版本没有c-api的, 需要注意下. 我这里不纠结直接用apt进行安装.

# apt install libopencv-dev

Ps: 当然这样安装是有坏处的, 首先版本不能控制, 这个完全看官方维护, 我安装完后查看了一下是 2.4.9.1 . 满足官方的要求, 我就没有纠结, 当然我们公司项目用的是opencv 3.2.0 . 那都是后话.

         f. Io 库, 官方需要 lmdb 和leveldb. 直接apt安装

# apt install liblmdb-dev libleveldb-dev

         g. Cudnn, 官方建议v6以上, 我们docker image自带v7. 所以这里不做修改.

 

5. 修改makefile.config

改的细节挺多的, 我这里直接把改好的config分享出来, 修改的地方用红色标准.

## Refer to http://caffe.berkeleyvision.org/installation.html

# Contributions simplifying and improving our build system are welcome!

 

# cuDNN acceleration switch (uncomment to build with cuDNN).

USE_CUDNN := 1

 

# CPU-only switch (uncomment to build without GPU support).

# CPU_ONLY := 1

 

# uncomment to disable IO dependencies and corresponding data layers

# USE_OPENCV := 0

# USE_LEVELDB := 0

# USE_LMDB := 0

# This code is taken from https://github.com/sh1r0/caffe-android-lib

# USE_HDF5 := 0

 

# uncomment to allow MDB_NOLOCK when reading LMDB files (only if necessary)

# You should not set this flag if you will be reading LMDBs with any

# possibility of simultaneous read and write

# ALLOW_LMDB_NOLOCK := 1

 

# Uncomment if you're using OpenCV 3

# OPENCV_VERSION := 3

 

# To customize your choice of compiler, uncomment and set the following.

# N.B. the default for Linux is g++ and the default for OSX is clang++

# CUSTOM_CXX := g++

 

# CUDA directory contains bin/ and lib/ directories that we need.

CUDA_DIR := /usr/local/cuda

# On Ubuntu 14.04, if cuda tools are installed via

# "sudo apt-get install nvidia-cuda-toolkit" then use this instead:

# CUDA_DIR := /usr

 

# CUDA architecture setting: going with all of them.

# For CUDA < 6.0, comment the *_50 through *_61 lines for compatibility.

# For CUDA < 8.0, comment the *_60 and *_61 lines for compatibility.

# For CUDA >= 9.0, comment the *_20 and *_21 lines for compatibility.

# CUDA_ARCH := -gencode arch=compute_20,code=sm_20 \

# -gencode arch=compute_20,code=sm_21 \

# -gencode arch=compute_30,code=sm_30 \

# -gencode arch=compute_35,code=sm_35

CUDA_ARCH := -gencode arch=compute_50,code=sm_50 \

-gencode arch=compute_52,code=sm_52 \

-gencode arch=compute_60,code=sm_60 \

-gencode arch=compute_61,code=sm_61 \

-gencode arch=compute_61,code=compute_61

 

# BLAS choice:

# atlas for ATLAS (default)

# mkl for MKL

# open for OpenBlas

#BLAS := atlas

BLAS := open

# Custom (MKL/ATLAS/OpenBLAS) include and lib directories.

# Leave commented to accept the defaults for your choice of BLAS

# (which should work)!

# BLAS_INCLUDE := /path/to/your/blas

# BLAS_LIB := /path/to/your/blas

 

# Homebrew puts openblas in a directory that is not on the standard search path

# BLAS_INCLUDE := $(shell brew --prefix openblas)/include

# BLAS_LIB := $(shell brew --prefix openblas)/lib

 

# This is required only if you will compile the matlab interface.

# MATLAB directory should contain the mex binary in /bin.

# MATLAB_DIR := /usr/local

# MATLAB_DIR := /Applications/MATLAB_R2012b.app

 

# NOTE: this is required only if you will compile the python interface.

# We need to be able to find Python.h and numpy/arrayobject.h.

#PYTHON_INCLUDE := /usr/include/python2.7 \

# /usr/lib/python2.7/dist-packages/numpy/core/include

# Anaconda Python distribution is quite popular. Include path:

# Verify anaconda location, sometimes it's in root.

# ANACONDA_HOME := $(HOME)/anaconda

# PYTHON_INCLUDE := $(ANACONDA_HOME)/include \

# $(ANACONDA_HOME)/include/python2.7 \

# $(ANACONDA_HOME)/lib/python2.7/site-packages/numpy/core/include

 

PYTHON_LIBRARIES := boost_python3 python3.6m

ANACONDA_HOME := $(HOME)/anaconda3

PYTHON_INCLUDE := $(ANACONDA_HOME)/include \

 $(ANACONDA_HOME)/include/python3.6m \

 $(ANACONDA_HOME)/lib/python3.6/site-packages/numpy/core/include

 

# Uncomment to use Python 3 (default is Python 2)

# PYTHON_LIBRARIES := boost_python3 python3.5m

# PYTHON_INCLUDE := /usr/include/python3.5m \

#                 /usr/lib/python3.5/dist-packages/numpy/core/include

 

# We need to be able to find libpythonX.X.so or .dylib.

# PYTHON_LIB := /usr/lib

 PYTHON_LIB := $(ANACONDA_HOME)/lib

 

# Homebrew installs numpy in a non standard path (keg only)

# PYTHON_INCLUDE += $(dir $(shell python -c 'import numpy.core; print(numpy.core.__file__)'))/include

# PYTHON_LIB += $(shell brew --prefix numpy)/lib

 

# Uncomment to support layers written in Python (will link against Python libs)

# WITH_PYTHON_LAYER := 1

 

# Whatever else you find you need goes here.

INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial/

LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/x86_64-linux-gnu/hdf5/serial/

 

# If Homebrew is installed at a non standard location (for example your home directory) and you use it for general dependencies

# INCLUDE_DIRS += $(shell brew --prefix)/include

# LIBRARY_DIRS += $(shell brew --prefix)/lib

 

# NCCL acceleration switch (uncomment to build with NCCL)

# https://github.com/NVIDIA/nccl (last tested version: v1.2.3-1+cuda8.0)

# USE_NCCL := 1

 

# Uncomment to use `pkg-config` to specify OpenCV library paths.

# (Usually not necessary -- OpenCV libraries are normally installed in one of the above $LIBRARY_DIRS.)

# USE_PKG_CONFIG := 1

 

# N.B. both build and distribute dirs are cleared on `make clean`

BUILD_DIR := build

DISTRIBUTE_DIR := distribute

 

# Uncomment for debugging. Does not work on OSX due to https://github.com/BVLC/caffe/issues/171

# DEBUG := 1

 

# The ID of the GPU that 'make runtest' will use to run unit tests.

TEST_GPUID := 0

 

# enable pretty build (comment to see full commands)

Q ?= @

 

   

6. 编译caffe与测试

# cd $HOME/caffe

# make all -j8

# make pycaffe

# make test

7. Python 环境添加 caffe

a. 在.bashrc中添加一下内容

export PYTHONPATH="/var/source/caffe/python:$PYTHONPATH"

export LD_LIBRARY_PATH="/usr/lib/x86_64-linux-gnu:/usr/lib/x86_64-linux-gnu/hdf5/serial:/usr/local/lib:$LD_LIBRARY_PATH"

   

b. 更新环境

# source $HOME/.bashrc

8. 总结

总体来看, 没有修改caffe代码或者库, 唯一麻烦的就是配置下Makefile.config文件. 哈哈, 大功告成, 到这里基本搞定了, 但博客总有遗漏的地方, 欢迎大家积极交流.

阅读更多
换一批

没有更多推荐了,返回首页