The R programming language is being widely used by application programmers in statistics and bioinformatics where large data sets are processed. R is an
interpreted language that currently does not make use of the multiple processors available in modern computers. Hence, processing large data sets is slow, and applications such as string matching or genom sequence analysis often result in computations taking several hours.
ALCHEMY is an experimentation platform for the automatic parallelization analysis of R programs and the execution of parallelized R programs. The platform and several parallelization analysis modules (PAM) have already been developed; yet, ALCHEMY currently lacks a backend for fast parallel execution.
The goal of this thesis is to provide an ALCHEMY backend for fast parallel execution of parallelized R programs. This requires mapping parallel code structures that result from the parallelization analysis of the PAMs to primitives of the target platform. As the target platform, we use Intels C++ based Array Building Blocks (ArBB). ArBB provides means to handle parallelization at a high level of abstraction and transparently supports various multicore processors.