Optimization of VLIW compatibility systems employing dynamic rescheduling

Abstract

Lack of object code compatibility in VLIW architectures is a severe limit to their adoption as a general-purpose computing paradigm. Previous approaches include hardware and software techniques, both of which have drawbacks. Hardware techniques add to the complexity of the architecture, whereas software techniques require multiple executables. This paper presents a technique called Dynamic Rescheduling that applies software techniques dynamically, using intervention by the OS: at each first-time page fault, the page of code is rescheduled for the new generation, if required. Results are presented to demonstrate the viability of the technique using the Illinois IMPACT compiler and the TINKER architectural framework. For the machine models and the workloads used in this study, performance of the rescheduled code compares well with the native scheduled code for a machine. The behavior of a subset of programs in the workload is such that they face a large number of first-time page faults. Due to this, their rescheduling overhead is higher relative to their total execution time. Such programs are called high-overhead programs. Caching of translated pages across multiple invocations of the program to reduce the rescheduling overhead, using a persistent rescheduled-page cache (PRC) (1) discussed. It was found that for the workload used in this evaluation, a PRC of size between 512 to 1024 pages, and which uses an overhead-based page replacement policy would be effective in reducing the overhead.