CuBLonD - GPU Version of BLonD

Requirments

1. Cuda
2. PyCuda
3. Scikit-Cuda

Installation Guide

Cuda

Download and install cuda from the following link https://developer.nvidia.com/cuda-downloads.

PyCuda

To install pycuda open a terminal and type

Scikit-Cuda

To install scikit-cuda open a terminal and type $ pip install scikit-cuda

Verification

To verify your installation, in a python terminal type the following:

import pycuda.autoinit
from pycuda import gpuarray
import numpy as np
a = gpuarray.to_gpu(np.zeros(1000,np.float64))

GPU-BLonD

Finally after cloning the repo and adding it to your pythonpath, install it like the default BLonD with

$ python blond/compile.py install and the flags of your choice

How to use your GPU

To use the GPU version of BLonD you need to follow these 2 steps

1. You need to do that import import blond.utils.bmath as bm
2. Right before your main loop you need to add the following line of code:

bm.use_gpu()

and call the use_gpu() method of all your basic components. For example if you have a profile, a tracker and a TotalInducedVoltage object you need to write these lines before your main loop

bm.use_gpu()
my_tracker.use_gpu()
my_profile.use_gpu()
my_totalinducedvoltage().use_gpu()

More information

• You do not need to call the use_gpu method for components you pass to your tracker as arguments.
• You can enable an optimization with bm.enable_gpucache()
• If you have multiple GPUs and you want to use a specific one you can choose which one by giving its id as an argument to the bm.use_gpu() like that bm.use_gpu(1). To view your GPUs you can type nvidia-smi in your terminal.

For lxplus Users

You need to add these lines to your ~/.bashrc

export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
export PYTHONPATH=~/work/GPU-BLonD:$PYTHONPATH