apeforest commented on a change in pull request #15045: [Dependency Update] Dependency update doc URL: https://github.com/apache/incubator-mxnet/pull/15045#discussion_r306452399
########## File path: tools/dependencies/README.md ########## @@ -40,3 +40,349 @@ This issue appeared in the OSX build with XCode version 8.0 above (reproduced on ``` --without-libidn2 ``` + +*** + +# Dependency Update Runbook + +MXNet is built on top of many dependencies. Managing those dependencies could be a big headache. This goal of this document is to give a overview of those dependencies and how to upgrade when new version of those are rolled out. + +## Overview + +The dependencies could be categorized by several groups: BLAS libraries, CPU-based performance boost library, i.e. MKLDNN and GPU-based performance boosting library including CUDA, cuDNN, NCCL. and others including OpenCV, Numpy, S3-related, PS-lite dependencies. The list below shows all the dependencies and their version. Except for CUDA, cuDNN, NCCL which the user is required to install on their environments, we statically link those dependencies into libmxnet.so when we build PyPi package. By doing this, the user can take advantage of these dependencies without being worry about it. + +| Dependencies | MXNet Version | +| :------------: |:-------------:| +|OpenBLAS| 0.3.3 | +|MKLDNN| 0.19 | +|CUDA| 10.1 | +|cuDNN| 7.5.1 | +|NCCL| 2.4.2 | +|numpy| >1.16.0,<2.0.0 | +|request| >=2.20.0,< 3.0.0 | +|graphviz| <0.9.0,>=0.8.1 | +|OpenCV| 3.4.2 | +|zlib| 1.2.6 | +|libjpeg-turbo| 2.0.2 | +|libpng| 1.6.35 | +|libtiff| 4-0-10 | +|eigen| 3.3.4 | +|libcurl| 7.61.0 | +|libssl-dev| 1.1.1b | +|zmq| 4.2.2 | +|protobuf| 3.5.1 | +|lz4| r130 | +|cityhash| 1.1.1 | + +## How to update them? + +### MKL, MKLDNN + +@pengzhao-intel (https://github.com/apache/incubator-mxnet/commits?author=pengzhao-intel) and his team are tracking and updating these versions. Kudos to them! + +### CUDA, cuDNN, NCCL +#### 1. Environment Setup +``` +# Take Ubuntu 16.04 for example +sudo apt update +sudo apt-get install -y git \ + cmake \ + libcurl4-openssl-dev \ + unzip \ + gcc-4.8 \ + g++-4.8 \ + gfortran \ + gfortran-4.8 \ + binutils \ + nasm \ + libtool \ + curl \ + wget \ + sudo \ + gnupg \ + gnupg2 \ + gnupg-agent \ + pandoc \ + python3-pip \ + automake \ + pkg-config \ + openjdk-8-jdk + +# CUDA installation +# Take CUDA 10 for example +wget https://developer.nvidia.com/compute/cuda/10.0/Prod/local_installers/cuda_10.0.130_410.48_linux +chmod +x cuda_10.0.130_410.48_linux && sudo ./cuda_10.0.130_410.48_linux +# Installation except: +# Install NVIDIA Accelerated Graphics Driver for Linux-x86_64 410.48? +# (y)es/(n)o/(q)uit: y +# +# Do you want to install the OpenGL libraries? +# (y)es/(n)o/(q)uit [ default is yes ]: +# +# Do you want to run nvidia-xconfig? +# This will update the system X configuration file so that the NVIDIA X driver +# is used. The pre-existing X configuration file will be backed up. +# This option should not be used on systems that require a custom +# X configuration, such as systems with multiple GPU vendors. +# (y)es/(n)o/(q)uit [ default is no ]: +# +# Install the CUDA 10.0 Toolkit? +# (y)es/(n)o/(q)uit: y +# +# Enter Toolkit Location +# [ default is /usr/local/cuda-10.0 ]: +# +# Do you want to install a symbolic link at /usr/local/cuda? +# (y)es/(n)o/(q)uit: y +# +# Install the CUDA 10.0 Samples? +# (y)es/(n)o/(q)uit: n + +# Set LD_LIBRARY_PATH +export LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH} + +# Check installation +nvidia-smi + +# cuDNN Setup +# Take cuDNN 7.5.0 with CUDA 10 for example +# https://developer.nvidia.com/rdp/cudnn-download +# Register with NVIDIA and download cudnn-10.0-linux-x64-v7.5.0.56.tgz +# scp it to your instance +# https://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html +tar -xvzf cudnn-10.0-linux-x64-v7.5.0.56.tgz +sudo cp cuda/include/cudnn.h /usr/local/cuda/include +sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64 +sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn* +# Check cuDNN version +cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2 +# #define CUDNN_MAJOR 7 +# #define CUDNN_MINOR 5 +# #define CUDNN_PATCHLEVEL 0 +# -- +# #define CUDNN_VERSION (CUDNN_MAJOR * 1000 + CUDNN_MINOR * 100 + CUDNN_PATCHLEVEL) +# +# #include "driver_types.h" + +# install NCCL +# take NCCL 2.4.2 for example +# https://developer.nvidia.com/nccl/nccl2-download-survey +# Register with NVIDIA and download nccl-repo-ubuntu1604-2.4.2-ga-cuda10.0_1-1_amd64.deb +sudo dpkg -i nccl-repo-ubuntu1604-2.4.2-ga-cuda10.0_1-1_amd64.deb +sudo apt-key add /var/nccl-repo-2.4.2-ga-cuda10.0/7fa2af80.pub +sudo apt update +sudo apt install libnccl2 libnccl-dev +# we will check the NCCL version later +``` +#### 2. Build +``` +# Clone MXNet repo +git clone --recursive https://github.com/apache/incubator-mxnet.git +cd incubator-mxnet +# Make sure you pin to specific commit for all the performance sanity check to make fair comparison +# Make corresponding change on tools/setup_gpu_build_tools.sh +# to upgrade CUDA version, please refer to PR #14887. +# Make sure you add new makefile and right debs CUDA uses on the website +# http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/ + +# Build PyPi package +tools/staticbuild/build.sh cu100mkl pip + +# Wait for 10 - 30 mins, you will find libmxnet.so under the incubator-mxnet/lib + +# Install python frontend +cd python Review comment: This can be simplified by pip install -e python ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
