Title: Accelerated Array Computations in Haskell

Manuel Chakravarty, University of NSW

Abstract:

Modern GPUs are increasingly used for general-purpose computations,
which are supported by several low-level programming interfaces.
However, harnessing the extra-ordinary computational power of GPUs
from high-level languages, such as Haskell, remains a challenge. In
particular, the lack of support for function pointers and the very
small frame stack of GPUs makes it very hard to execute arbitrary
Haskell code on these devices. In this talk, I will outline a GPU
compilation framework for an embedded domain-specific language (DSL)
of collective arrays operations paramterised by scalar Haskell
computations. Computations of the embedded DSL are reflected into an
expression tree, dynamically compiled, and finally dispatched to the
GPU by targeting existing low-level programming interfaces. I will
discuss the main challenges of this approach and our efforts to
overcome these challenges.