This research is mainly looking at partial evaluation of MATLAB code. Much of the group use MATLAB as their main development tool and so the ability to generate efficient code quickly is very important. The aim is to allow researchers to develop general code and generate fast specialised codes for particular problems.

Partial evaluation of a function is the act of specialising the function for some specific inputs. This can lead to speed ups in the resultant code as expressions can then be fully evaluated, loops unrolled and conditional statements eliminated. This is useful in many scientific computations as often the same code will be run with only some of the parameters being varied. As the partial evaluation is a source code transformation, it will work in concert with MATLAB compilers, such as mcc, to produce even better results.

Partial evaluation also implies automatic optimisation and so a tool should be produced that can optimise code even if there are no fixed parameters. The specialisation is not limited to full specification of parameters but could also involve specification of the shape or size.

This work is part of a PhD by Dan Elphick with supervision from CED in the School of Engineering Sciences and DSSE in the Department of Electronics and Computer Science.