dicecco1 / fpga_cpfp

HLS Custom-Precision Floating-Point Library
MIT License
13 stars 4 forks source link
floating-point fpga hls sdaccel

fpga_cpfp

HLS Custom-Precision Floating-Point Library

The figures below show the resource utilization of the multiplier and adder for different settings of exponent width, mantissa width, and rounding modes. The library results were gathered using Vivado HLS 2016.3, while the native floating-point results were gathered with Vivado HLS 2017.1. To test the adder core a simple function was synthesized consisting of a single addition between two cpfp numbers. Likewise to test the multiplier core a simple function was synthesized consisting of a single multiplication between two cpfp numbers.

An example use-case can be found at: https://github.com/dicecco1/fpga_caffe/blob/master/src/fpga_caffe/layers/crp_layer_hwcn_cpfp.cpp

adder_round

adder_round_off

mul_round

mul_round_off