Skip to Main Content
VLIW processors are statically scheduled processors and their performance depends on the quality of schedules generated by the compiler's scheduler. We propose a multi-threaded architectural support for speculative trace scheduling in VLIW processors. In this multithreaded architecture the next most probable trace is speculatively executed, overlapping the stall cycles of the processor during cache misses and page faults. Switching between traces is achieved with the help of special hardware units viz. operation state buffers and trace buffers. We observe an 8.39% reduction in the overall misprediction penalty as compared to that incurred when the stall cycles due to cache misses alone are not overlapped.