The Design And Implementation Of Floating Point Unit Based On ARMv7 Floating Point Instruction Set

This article studies the FPU architecture and its implementation,designed and implemented a FPU support both single and double precision floating-point arithmetic based on the ARMv7 floating-point instruction set.The FPU implements a total of 28 floating-point instructions,including floating-point add and subtract instructions,floating-point multiply instructions,floating-point multiply accumulate instructions, floating-point fused multiply add instructions, floating point divide instructions, floating-point square root instructions and floating-point data convert instructions and load/store instructions.It implements in five stages pipeline.Arthmetic excute unit can share one round unit.Designs five execution units, including data transfer unit,fused floating-point multiply accumulate unit,data storage unit,divide unit.Data transfer unit, floating-point fused multiply add unit,data storage unit are fully pipelined units,they can execute instructions in parallel.Floating-point division unit implement improve in SRT Algorithm, division and square root instrcution can execute in the same unit.This paper studies quotient lookup table and quotient dynamic conversion unit deeply.It is synthesized by Design Complier using SMIC 0.13μm technology library. Optimizing critical path, reduce the critical path delay in 5ns.Compare with the conventional SRT divider, it uses lesser cycles when execute division or square root instruction..Fused floating-point multiply add unit based on dual path design, which proposed by T.Lang and J.D.Bruguera.And improve it. It can use the same round unit with other arithmetic unit.It is synthesized by Design Compiler using SMIC 0.13μm technology library.And optimze the critical path, reduce the caritical path delay in 5ns.And also propose a floating point data transfer unit.It can supporte four rounding modes, specified in IEEE 754 floating-point rule.