Improved integration of multithreading into the STGM

Abstract

variant of the Spineless Tagless G-Machine (STGM) which contains explicit support for multithreading is introduced in [1]. The main design decisions are the separation of demand for evaluation from case selection and the introduction of an abstract notion of thread boundaries and thread synchronisation. This report proposes an alternative solution which does not separate demand from selection. Instead, case selections are extended by additional alternatives which handle the appearance of long latency operations. The overhead, necessary to control multithreading, is reduced and sequentially evaluated parts of a program are more efficient.