Design and Implementation of Floating-Point Transcendental Function Processor Based on RISC- V Architecture
Bochen Qin, Gang Cai, Zhihong Huang · 2024
Floating-point (FP) transcendental function operations are widely used in deep learning, remote sensing radar, and other engineering and application fields. Computing these FP functions through integer arithmetic logic units requires a large number of clock cycles. Coordinate rotation digital computer (CORDIC) is one of the commonly used algorithms for hardware computation of transcendental functions. The high radix adaptive CORDIC (HCORDIC) algorithm, which has the advantages of fewer iterations and faster convergence, is implemented using the IEEE-754 FP standard. Due to the advantages of flexibility, customization, and scalability of the RISC- V instruction set architecture (ISA), a processor based on the RISC- V FP transcendental function extension is proposed. The proposed processor has been implemented on Xilinx Zynq X7CZ020 FPGA. The result shows that the processor executes transcendental functions at least 100 times faster than without the accelerator. The calculation accuracy ranges from 10–6to 10–8and the maximum operating frequency is 201.13 MHz, making it highly valuable in engineering applications.