Our next meeting will be held on Wednesday 15th of February, at 5.00pm, in LG.11, David Hume Tower (on the lower ground floor - map below). As usual, the meeting will be followed by drinks and chat in the Potting Shed. Meetings are open for all to attend and newcomers/beginners are very welcome.
Our speaker is Juan Fumero from the School of Informatics at the University of Edinburgh, who will talk about:
Automatic Acceleration of R programs with GPUs
Graphics Processor Units (GPUs) are interesting hardware to parallelize and speed-up applications. However, the programmability of these devices is quite complex for most of the users and the existing standards for GPU programming are mostly available for low-level languages such as C.
Academia and industry have proposed and implemented higher-level abstractions on top of the most common dynamic programming languages to program GPUs much more easily. For example in R, GPU programming is possible through external libraries (for a certain number of operations) or via wrappers in which the GPU code is normally written in C. But, would it not be better if the R interpreter could offload the potentially parallel R code to the GPU automatically?
In this talk we present an R interpreter that automatically offloads parts of the R program into the GPU without any changes in the source code. We will show some preliminary results and a demo with a few examples. We will also show an overview about how the translation process (R to the GPU with OpenCL) is performed.
To download the paper associated with this talk, click here.
For any newcomers, here’s a map of where we’ll be.