Out of order floating point coprocessor for RISC V ISA

Abstract

Floating point operations are important and essential part of many scientific and engineering applications. Floating point coprocessor (FPU) performs operations like addition, subtraction, division, square root, multiplication, fused multiply and accumulate and compare. Floating point operations are part of ARM, MIPS, and RISC-V etc. instruction sets. The FPU can be a part of hardware or be implemented in software. This paper details an architectural exploration for floating point coprocessor enabled with RISC-V [1] floating point instructions. The Floating unit that has been designed with RISC-V floating point instructions is fully compatible with IEEE 754-2008 standard as well. The floating point coprocessor is capable of handling both single and double precision floating point data operands in out-of-order for execution and in-order commit. The front end of the floating point processor accepts three data operands, rounding mode and associated Opcode fields for decoding. Each floating point operation is tagged with an instruction token for in-order completion and commit. The output of the floating point unit is tagged with either single or double precision results along with floating point exceptions, if any, based on the RISC-V instruction set. The coprocessor for floating point integrates with integer pipeline. The proposed architecture for floating point coprocessor with out-of-order execution, in-order commit and completion/retire has been synthesized, tested and verified on Xilinx Virtex 6 xc6vlx550t-2ff1759 FPGA. A system frequency of 240MHz for single precision floating point and 180 MHz double precision floating point operations has been observed on FPGA.