Oct. 7, 2014, 12:01 p.m. - Oct. 7, 2014, 1:01 p.m.
McConnell Eng. room 320
Velociraptor project: Tools for compiling array-based languages to CPUs and GPUs”
Compiling array-based languages such as MATLAB and Python to CPUs and GPUs is a challenging problem. I have built a collection of tools that you can use to build such compilers. The tools include an optimizing compiler toolkit called Velociraptor that generates LLVM and OpenCL, a fast portable matrix operations library called RaijinCL for GPUs a type inference and compilation engine for a subset of Python+NumPy, an extension of McVM (a virtual machine for MATLAB built in our lab) to multicores and GPUs and more! Come learn about various tools we built that might be of interest to you. We will introduce the tools, how they interoperate, strengths and weaknesses and how and when to use them.
The talk will lean towards implementation details, rather than theory or experimental results.
The talk will be of interest to people interested in compilers. I will also discuss future projects of various complexity that might be of interest to any students looking for compiler-related projects