Skip to content

Latest commit

 

History

History
272 lines (217 loc) · 6.65 KB

INSTALL.md

File metadata and controls

272 lines (217 loc) · 6.65 KB

Installation notes

General requirements on Linux/Unix

Normal build procedure with CMake

To build:

mkdir builddir
cd builddir
cmake ..
cmake --build .

Without GPU acceleration:

By default dw will be compiled with OpenCL/GPU support enabled. If that does not work for you, turn that option off by passing -DENABLE_GPU=OFF to cmake, i.e. use:

mkdir builddir
cd builddir
cmake -DENABLE_GPU=OFF ..
cmake --build .

Machine specific optimizations

To enable optimizations for your particular machine, pass the option -DENABLE_NATIVE_OPTIMIZATION=ON. Doing that might create binaries that run faster on your machine. However it might not be possible to run the same binaries on another machine.

To install

A direct install (files will be copied to default paths), use:

sudo make install

To use the system package manager is probably the better way since it offers a cleaner way to uninstall:

mkdir builddir
cd builddir
cmake ..
cpack -G DEB # see cpack --help for other types
sudo apt-get install ./deconwolf-0.3.8-Linux.deb # on Ubuntu

GPU Acceleration via OpenCL

  • If you have more than one OpenCL device, please use the --cldevice n argument to tell deconwolf which device to use. The devices are ordered with 0 being the first. If available use the command clinfo to list the OpenCL devices available on your system.

  • Under WSL2 and with the current version of pocl it does not seem to work, see issue #56. Hopefully it can work in the future.

  • If cmake can't find OpenCL and you are on a cluster, check out: [#55 (comment)]

macOS Big Sur

For building you will need XCode from the App Store and brew.

Then set up XCode and install the required packages:

xcode-select --install
brew install libopenmpt # Not sure if this is needed
brew install libomp
brew install libtiff
brew install fftw
brew install gsl

macOS Sonoma, Apple Silicon

For building you will need XCode from the App Store and brew.

Then set up XCode and install the required packages:

xcode-select --install
brew install cmake
brew install gcc
brew install pkg-config
brew install libomp
brew install libtiff
brew install fftw
brew install gsl

To build, make sure you use the Homebrew-installed version of gcc, in my case gcc-14:

mkdir builddir
cd builddir
CC=gcc-14 CXX=g++-14 cmake ..
CC=gcc-14 CXX=g++-14 cmake --build .
make install

Windows 10/11

Deconwolf can be built several different ways under Windows:

  1. Using WSL, then follow the instructions for Ubuntu. Most likely there will be a performance penalty, and it will not be possible to enable GPU acceleration, see issue #56.
  2. msys2 or cygwin, however those options will be slower since OpenMP will be using an pthreads emulation on top of windows threads. It might be possible to get OpenCL working.
  3. As a native windows program. This is the preferred way since it should work with OpenCL without any problems..

To build native windows programs, at least the following software is required:

The dependencies can get retrieved by vcpkg:

git clone https://github.com/microsoft/vcpkg
.\vcpkg\bootstrap-vcpkg.bat
.\vcpkg\vcpkg.exe install fftw3[threads]
.\vcpkg\vcpkg.exe install tiff
.\vcpkg\vcpkg.exe install gsl
.\vcpkg\vcpkg.exe install opencl
.\vcpkg\vcpkg.exe install getopt
.\vcpkg\vcpkg integrate install

Please note the value of the CMAKE_TOOLCHAIN_FILE as it will be used later.

A visual studio project can be created by

cd deconwolf
mkdir build
cd build
cmake "-DCMAKE_TOOLCHAIN_FILE=C:/YOUR/OWN/PATH/vcpkg/scripts/buildsystems/vcpkg.cmake" -T ClangCL -A x64 ../

Important: Please use the correct path to vcpkg.cmake.

Open the visual studio solution and oo to Linker->Input->Additional dependencies and add:

libomp.lib

Change the build type from debug to release and compile.

If you are a windows developer and reading this, please help us out to make the build process smoother!

FreeBSD

  • Use gmake, not make.

Packages:

pkg install git
pkg install gmake
pkg install fftw3
pkg install tiff
pkg install gsl
pkg install sudo

CentOS

Tested on CentOS Linux Release 7.8.2009 (Core).

Dependencies:

sudo yum install gcc gsl-devel libtiff-devel fftw-devel

Ubuntu 16.04

sudo apt-get update
sudo apt-get install gcc
sudo apt-get install pkg-config
sudo apt-get install libfftw3-single3
sudo apt-get install libfftw3-dev
sudo apt-get install openmp
sudo apt-get install libtiff-dev # only difference to 20.04
sudo apt-get install libgsl-dev
sudo apt-get install libomp-dev
sudo apt-get install libpng-dev

Ubuntu 22.04

First install the required packages:

sudo apt-get update
sudo apt-get install \
 cmake               \
 pkg-config          \
 gcc                 \
 libfftw3-single3    \
 libfftw3-dev        \
 libgsl-dev          \
 libomp-dev          \
 libpng-dev          \
 libtiff-dev

Ubuntu 23.04

Same as Ubuntu 22.04. Possibly also

apt-get install opencl-headers

Arch/ Manjaro

# remember to update system
sudo pacman -Suuyy
# install dependencies
sudo pacman -S fftw gsl openmp libtiff

Rapsberry PI (64-bit Debian bookworm)

sudo apt-get update
sudo apt-get upgrade
sudo apt-get install libfftw3-dev \
libtiff-dev \
libgsl-dev

MKL

FFTW3 is the default FFT backend for deconwolf but it is also possible to use Intel MKL. At some point it was possible to choose MKL via

sudo apt install intel-mkl
make MKL=1 -B

To set the number of threads, set the environmental variable MKL_NUM_THREADS, for example:

export MKL_NUM_THREADS=8
dw ...

If you are interested in using MKL please open a new issue.