Optimising Pseudoknot in ΓCMC

Benchmarking implementations is fundamental to allow analysing performance amongst different platforms. The choice of a benchmark that makes possible a reliable and fair comparison of a particular aspect is a difficult task, however. The Pseudoknot benchmark is a floating-point intensive application taken from molecular biology which was used to compare the compile-time and execution-time performance of over 25 different implementations of functional languages. Amongst those implementations was ΓCMC, an abstract machine for efficient implementation of lazy functional languages. ΓCMC pioneered the transference of the control of the execution flow to C, as much as possible, to take advantage of the extremely low cost of procedure calls in modern RISC architectures. ΓCMC was amongst the machines that presented good Pseudoknot figures, although it did not use some of the sophisticated optimisations of most of the other implementations. The experience of implementing Pseudoknot in ΓCMC was most valuable in providing insights for new ways in optimising it. This paper describes several optimisations introduced to ΓCMC which bring a better Pseudoknot performance