Algebraic Transformation of Descriptive Vector Byte-code Sequences

Abstract

Both high-productivity and high-performance are two often sought after aspects of scientific programming. Python gives the programmer high-productivity, but even with NumPy it is often not high-performant because of the GIL1, which makes it inherently single threaded. Bohrium intercepts NumPy calls and generates an intermediate language, Bohrium byte-code, before being compiled to OpenCL kernels. It thus grants Python/NumPy the ability to be easily run on multicore systems or GPUs, without changing the source code. The Bohrium byte-code can be optimized, by transforming byte-code sequences into more performant ones. This way, the scientific programmer will not need to change her code to utilize special performant constructs.