Build support status
Host build
- Windows build (cpu and gpu)
- Linux build (cpu and gpu)
- MacOS build (cpu only)
- Android build (cpu only) at termux env
Cross build
- Windows cross build ARM-Android (ok)
- Windows cross build ARM-Linux (ok)
- Linux cross build ARM-Android (ok)
- Linux cross build ARM-Linux (ok)
- Linux cross build RISCV(support rvv)-Linux (ok)
- MacOS cross build ARM-Android (ok)
- MacOS cross build ARM-Linux (ok but experimental)
- MacOS cross build IOS (ok)
Build env prepare
Prerequisites
Most of the dependencies of MegBrain(MegEngine) are located in third_party directory, which can be prepared by executing:
./third_party/prepare.sh
./third_party/install-mkl.sh
Windows shell env(bash from windows-git), infact if you can use git command on Windows, which means you always install bash.exe at the same dir of git.exe, find it, then you can prepare third-party code by
bash.exe ./third_party/prepare.sh
bash.exe ./third_party/install-mkl.sh
if you are use github MegEngine and build for Windows XP, please
1: donwload mkl for xp from: http://registrationcenter-download.intel.com/akdlm/irc_nas/4617/w_mkl_11.1.4.237.exe
2: install exe, then from install dir:
2a: cp include file to third_party/mkl/x86_32/include/
2b: cp lib file to third_party/mkl/x86_32/lib/
About third_party/prepare.sh
, also support to be managed by CMake
, just config EXTRA_CMAKE_ARGS="-DMGE_SYNC_THIRD_PARTY=ON"
before run scripts/cmake-build/*.sh
But some dependencies need to be installed manually:
- CUDA(>=10.1), cuDNN(>=7.6) are required when building MegBrain with CUDA support.
- TensorRT(>=5.1.5) is required when building with TensorRT support.
- LLVM/Clang(>=6.0) is required when building with Halide JIT support.
- Python(>=3.6) and numpy are required to build Python modules.
Package install
Windows host build
0: about all windows config
* please check scripts/whl/windows/config.sh
1: install git (Windows GUI)
* download git-install.exe from https://github.com/git-for-windows/git/releases
* NOTICE: We find some version, which bigger than v2.35 may report: Bad address issue when run cmd.exe /C "dir", so we suggest install v2.34
* only need choose git-lfs component
* install to default dir: /c/Program\ Files/Git
2: install cuda components (Windows GUI) if you want to build with CUDA
* download cuda/cudnn/trt from NVIDIA or by scripts/whl/windows/cuda_cudnn_install.py
* export CUDA_ROOT_DIR/CUDNN_ROOT_DIR/TRT_ROOT_DIR to real cuda/cudnn/trt location
3: install all env except cuda env
* just run scripts/whl/windows/env_prepare.sh
4: edit system env variables (Windows GUI)
* append "Path" env value
C:\Program Files\Git\cmd
C:\Users\build\megengine_dev_tools\pyenv-win\pyenv-win\versions\3.10.1
change `build` to your real user name
Linux host build
0: we provide Dockerfile if you do not prepare local env
* check about scripts/whl/manylinux2014/build_image.sh
1: install Cmake, which version >= 3.15.2, ninja-build
2: install gcc/g++, which version >= 6, (gcc/g++ >= 7, if need build training mode)
3: install build-essential git git-lfs gfortran libgfortran-6-dev autoconf gnupg flex bison gperf curl zlib1g-dev gcc-multilib g++-multilib lib32ncurses5-dev libxml2-utils xsltproc unzip libtool librdmacm-dev rdmacm-utils python3-dev python3-numpy texinfo
4: CUDA env(if build with CUDA), please export CUDA/CUDNN/TRT env, for example:
export CUDA_ROOT_DIR=/path/to/cuda
export CUDNN_ROOT_DIR=/path/to/cudnn
export TRT_ROOT_DIR=/path/to/tensorrt
MacOS host build
1: install Cmake, which version >= 3.15.2
2: install brew: /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install.sh)"
3: brew install python python3 coreutils ninja
4: install at least xcode command line tool: https://developer.apple.com/xcode/
5: about cuda: we do not support CUDA on MacOS
6: python3 -m pip install numpy (if you want to build with training mode)
Cross build for ARM-Android
Now we support Windows/Linux/MacOS cross build to ARM-Android
2: download NDK from https://developer.android.google.cn/ndk/downloads/ for diff OS platform package, suggested NDK20 or NDK21
3: export NDK_ROOT=NDK_DIR at bash-like env
Cross build for ARM-Linux
Now we support ARM-Linux on Linux and Windows fully, also experimental on MacOS
1: download toolchains from http://releases.linaro.org/components/toolchain/binaries/ or https://developer.arm.com/tools-and-software/open-source-software/developer-tools/gnu-toolchain/gnu-a/downloads if use Windows or Linux
2: download toolchains from https://github.com/thinkski/osx-arm-linux-toolchains if use MacOS
Cross build for RISCV-Linux
Now we support RISCV-Linux
1: download toolchains from https://github.com/riscv-collab/riscv-gnu-toolchain
Cross build for IOS
Now we only support cross build to IOS from MACOS
1: install full xcode: https://developer.apple.com/xcode/
How to build
With bash env(Linux/MacOS/Windows-git-bash)
-
host build just use scripts:scripts/cmake-build/host_build.sh
builds MegBrain(MegEngine) that runs on the same host machine (i.e., no cross compiling)
The following command displays the usage:
scripts/cmake-build/host_build.sh -h
more example:
1a: build for Windows for XP (sp3): (dbg) EXTRA_CMAKE_ARGS="-DMGE_DEPLOY_INFERENCE_ON_WINDOWS_XP=ON" ./scripts/cmake-build/host_build.sh -m -d
(opt) EXTRA_CMAKE_ARGS="-DMGE_DEPLOY_INFERENCE_ON_WINDOWS_XP=ON" ./scripts/cmake-build/host_build.sh -m
2a: build for Windows for XP (sp2): (dbg) EXTRA_CMAKE_ARGS="-DMGE_DEPLOY_INFERENCE_ON_WINDOWS_XP_SP2=ON" ./scripts/cmake-build/host_build.sh -m -d
(opt) EXTRA_CMAKE_ARGS="-DMGE_DEPLOY_INFERENCE_ON_WINDOWS_XP_SP2=ON" ./scripts/cmake-build/host_build.sh -m
-
cross build to ARM-Android: scripts/cmake-build/cross_build_android_arm_inference.sh
builds MegBrain(MegEngine) for inference on Android-ARM platforms.
The following command displays the usage:
scripts/cmake-build/cross_build_android_arm_inference.sh -h
-
cross build to ARM-Linux: scripts/cmake-build/cross_build_linux_arm_inference.sh
builds MegBrain(MegEngine) for inference on Linux-ARM platforms.
The following command displays the usage:
scripts/cmake-build/cross_build_linux_arm_inference.sh -h
-
cross build to RISCV-Linux: scripts/cmake-build/cross_build_linux_riscv_inference.sh
builds MegBrain(MegEngine) for inference on Linux-RISCV platforms.
The following command displays the usage:
scripts/cmake-build/cross_build_linux_riscv_inference.sh -h
- if board support RVV(at least 0.7), for example: nezha D1 , use -a rv64gcv0p7
- if board do not support RVV, use -a rv64norvv
-
cross build to IOS: scripts/cmake-build/cross_build_ios_arm_inference.sh
builds MegBrain(MegEngine) for inference on iOS (iPhone/iPad) platforms.
The following command displays the usage:
scripts/cmake-build/cross_build_ios_arm_inference.sh -h
Visual Studio GUI(only for Windows host)
1: import megengine src to Visual Studio as a project
2: right click CMakeLists.txt, choose config 'cmake config' choose clang_cl_x86 or clang_cl_x64
3: config other CMAKE config, eg, CUDA ON OR OFF
Other ARM-Linux-Like board support
It's easy to support other customized arm-linux-like board, example:
-
1: HISI 3516/3519, infact u can just use toolchains from arm developer or linaro
then call scripts/cmake-build/cross_build_linux_arm_inference.sh to build a ELF
binary, or if you get HISI official toolschain, you just need modify CMAKE_CXX_COMPILER
and CMAKE_C_COMPILER in toolchains/arm-linux-gnueabi* to a real name
-
2: about Raspberry, just use scripts/cmake-build/cross_build_linux_arm_inference.sh
About build args
All scripts/cmake-build/*.sh
support EXTRA_CMAKE_ARGS
to config more options
- get support options by
-l
, for example: scripts/cmake-build/cross_build_android_arm_inference.sh -l
- CMake support
Release
, Debug
, RelWithDebInfo
build type, all scripts/cmake-build/*.sh
default build type is Release
, can build Debug
type with -d
, if you want to build with RelWithDebInfo
, you can config with EXTRA_CMAKE_ARGS
, for example: EXTRA_CMAKE_ARGS="-DCMAKE_BUILD_TYPE=RelWithDebInfo" ./scripts/cmake-build/host_build.sh
, Notice: when build with Release
, we will disable some build components: RTTI
, MGB_ASSERT_LOC
, and MGB_ENABLE_DEBUG_UTIL
- CMake build all targets by default, if you just want build a specified target, you can build with
-e xxxx
, for example, only build with lite_shared
: ./scripts/cmake-build/cross_build_android_arm_inference.sh -e lite_shared
, Notice: when with -e
, will do not strip target, always for debug or need strip target manually
- About others build flag, please run with flag
-h