Detection of miscalculation caused by precision degradation

点击翻译按钮获取中文摘要

To achieve the efficient accuracy degradation detection on floating point number calculation, we developed a vector co-processor that has two or more floating-point arithmetic units with accuracy degradation detection function. And we investigatedthat it worked correctly. Moreover, we compared the number of execution cycles of the workload, which actually causes accuracy degradation, with a conventional processor. Because our vector processor can execute by about 1/5 number of execution cycles, we can practically use this system as a prototype system.