2. Getting started guide

If you have already installed Bifrost, look below to Creating your first pipeline

2.1. Installation

Bifrost requires several dependencies, depending on how you want to use it. If you don’t know what you are doing, assume that you want all the dependencies - we will walk you through this process.

You will need a relatively new gcc and CUDA - we have used Bifrost with gcc 4.8 and CUDA 8, but higher ones should also work.

2.1.1. Python dependencies

Bifrost is compatible with Python >3.6.

2.1.1.1. pip

pip is a package manager for other Python dependencies. Once you have pip, installing additional python dependencies should be straightforward. pip comes with setuptools, which is required for installing Bifrost. The detailed instructions for pip can be found here, but the basics are as follows:

  1. Download `get-pip.py <https://bootstrap.pypa.io/get-pip.py>`__

  2. Navigate to the download directory, and run python get-pip.py --user, which will install a local copy of pip.

  3. Check pip is working with pip list, which will give the versions of pip and setuptools.

2.1.1.2. numpy, matplotlib, contextlib2, simplejson, pint, graphviz, ctypesgen

If you have already installed pip, this step should be as simple as pip install --user numpy matplotlib contextlib2 simplejson pint graphviz ctypesgen==1.0.2.

2.1.2. C++ dependencies

2.1.2.1. CUDA

CUDA allows you to program your GPU from C and C++. You will need an NVIDIA GPU to do this. If this is your first time trying out Bifrost, and you don’t have CUDA yet, we recommend that you skip this step, and try out a CPU-only version of Bifrost. Then, once you have that first experience, you can come back here for a speedup.

If you are ready to work with a GPU, you will want to get the newest CUDA toolkit. Follow the operating system-specific instructions to install. Be sure to install a kernel driver that works with the version of toolkit that you are using since version mismatches can lead to runtime errors.

The table below indicates which CUDA toolkit and kernel driver versions Bifrost has been tested against.

OS

Linux Kernel

Driver Version

GPU

Toolkit

Status

Ubuntu 20.04

5.4.0-177-generic

520.61.05

RTX 2080Ti

11.0.3

Working

Ubuntu 20.04

5.4.0-177-generic

520.61.05

RTX 2080Ti

11.1.1

Working

Ubuntu 20.04

5.4.0-177-generic

520.61.05

RTX 2080Ti

11.2.2

Working

Ubuntu 20.04

5.4.0-177-generic

520.61.05

RTX 2080Ti

11.3.1

Working

Ubuntu 20.04

5.4.0-186-generic

470.239.06

Quadro K2200

11.4.4

Working

Ubuntu 20.04

5.4.0-186-generic

495.29.05

Quadro K2200

11.5.2

Working

Ubuntu 18.04

4.15.0-88-generic

510.85.02

A4000

11.6.124

Working

Ubuntu 18.04

4.15.0-88-generic

510.85.02

RTX 2080Ti

11.6.124

Working

Ubuntu 20.04

4.4.0-174-generic

525.147.05

Titan RTX

11.6.55

Working

Ubuntu 20.04

5.4.0-186-generic

520.61.05

Quadro K2200

11.8.0

Known FIR and FFT Problems

Debian 12

6.1.0-21-amd64

525.147.05

Quadro K2200

12.0.0

Working

Ubuntu 20.04

5.4.0-147-generic

525.125.06

A5000

12.0.140

Working

Ubuntu 20.04

5.4.0-144-generic

525.125.06

A5000

12.0.140

Working

Ubuntu 20.04

5.4.0-144-generic

525.147.05

A4000

12.0.140

Working

Debian 12

6.1.0-21-amd64

530.30.02

Quadro K2200

12.1.1

Working

Debian 12

6.1.0-21-amd64

535.104.05

Quadro K2200

12.2.2

FFT bug: zeroed out data

Debian 12

6.1.0-21-amd64

545.23.08

Quadro K2200

12.3.2

Working

Debian 12

6.1.0-21-amd64

550.54.15

Quadro K2200

12.4.1

Working

Ubuntu 22.04

5.15.0-106-generic

555.42.06

GTX 980

12.5

Working

2.1.2.2. Other Dependencies

  • universal-ctags

  • Basic build tools (make, gcc, etc.)

On Ubuntu, the following command should grab everything you need:

sudo apt-get install build-essential software-properties-common universal-ctags

2.1.3. Bifrost install

Now you are ready to install Bifrost. Clone the GitHub master branch with

git clone https://github.com/lwa-project/bifrost.

You will want to run configure to tailor Bifrost to you system. At the end of configure you will get a summary of how Bifrost will be built:

config.status: creating src/bifrost/config.h config.status: executing libtool commands

configure: cuda: yes - 50 52 configure: numa: yes configure: hwloc: yes configure: libvma: no configure: python bindings: yes configure: memory alignment: 4096 configure: logging directory: /dev/shm/bifrost configure: options: native

Bifrost is now ready to be compiled. Please run ‘make’ ```

Now you can call make, and make install to install Bifrost.

Trying to call import bifrost inside of a Python program will tell you if your install was successful or not.