Dynamic vectorization: a mechanism for exploiting far-flung ILP in ordinary programs
Sriram Vajapeyam, P.J. Joseph, Tulika Mitra · NOT FOUND REPOSITORY (Indian Institute of Science Bangalore) · 1999
Several ILP limit studies indicate the presence of considerable ILP across dynamically far-apart instruc-tions in program execution. This paper proposes a hardware mechanism, dynamic vectorization (DV), as n tool for quickly building up n large logical instruction window. Dynamic vectorization converts repetitive dynamic instruction sequences into vectorform, enabling the processing of imtructiom from beyond the corresponding program loop to be overlapped with the loop. This enables vector-like execution of programs with relatively complex static control fiw that my not be amenable to static, compile time vectorization. Experi-mental evaluation shows that a large fraction of the dynamic instructions of four of the six SPEClnt92 pro-grams can be captured in vectorform. Three of these pro-grams exhibit significant potential for ILP improvements from dynamic vectoriiation, with speedups of more than a factor of 2 in a scemrio of realistic branch prediction and perfect memory disambiguntion. Under perfect branch prediction conditions, a fourth program also shows well over a factor of 2 speedup from DV, The speedups are due to the overlap of post-loop processing with loop process-ing. 1.