Presentation: "Making GPUs Accessible Through High Level Languages"

Time: Tuesday 12:05 - 13:05

Location: Robinson/Whitman

Abstract:
"Over the  past decade  GPU hardware  and programming  languages have become  powerful and  expressive  enough to  be  useful for  general computation. Still,  GPU programs are  written in a C  dialect, and, while  that  achieves great  performance,  it  can sometimes  hinder productivity.  To make  GPU  programming simple,  both research  and industry  provided  either  a  foreign function  interface  for  GPU programs, ability to overload  programming language operators to run on the  GPU, or provide  DSLs for the  domain that can  be translated into GPU code. In this talk we  will examine the state of the art of GPU computing  and research, how GPUs are  being used with high level languages today, and what the future of GPU programs will be.
"

Abdul Dakkak, Kernel Developer, Wolfram Research

 Abdul  Dakkak
Abdul Dakkak is a Kernel Developer at Wolfram Research and Lead Developer of Mathematica's GPU initiative - CUDALink and OpenCLLink.
He is also part of a team developing computational geometry, SymbolicC, CCompilerDriver, and the Mathematica Compiler. Currently, he is researching ways of making GPU programming simpler by exploiting the functional and expressive nature of the Mathematica language.