This repository contains several example of ROC GPU programming using Numba. It is a fork and update of examples written for HSA https://github.com/ContinuumIO/numba-hsa-examples.
WARNING: The examples here and the underlying compilation chain are both under development!
For the purposes of this README, the $
symbol indicates the command prompt.
-
First, install the ROC stack from AMD as per the installation instructions provided here .
-
Create and activate a conda environment (named e.g.
amd_roc
) as follows:
$ conda create -n amd_roc -c numba numba roctools jupyter bokeh statsmodels \
python=3 h5py -y
$ source activate amd_roc
- Note that users should be a member of the
video
group to have access to the GPU.
There are two examples working at present, both revolving around kernel density
estimation. The first is a Jupyter notebook, multi_variate_kde_example.ipynb
which can be launched with the following:
$ jupyter notebook numba_roc_examples/kerneldensity/
You can see the rendered notebook here. The AMD GPU implementation used in notebook can be found here.
The second is a bokeh
application that can be launched following the
instructions in the README.md
of the numba_roc_examples/kde_bokeh
directory.
Run the full test suite, it is expected that this will fail (still under development):
$ ./runtests.sh
Selectively run tests from subdirectories:
$ ./runtests.sh <subdirectory>
Example:
$ ./runtests.sh numba_roc_examples/kerneldensity