Error flatten logarithm approximation for graphics processing unit

This paper propose a high performance algorithm for logarithm approximation. A piecewise linear interpolation with non-uniform partition is introduced to produce a flatten error distribution. Base on error flatten algorithm our method can produce greatly more accurate results than state-of-art logarithm approximation. Our proposed logarithm conversion is further optimized for hardware implementation. Analysis of simulation results in Matlab and hardware resources in FPGA illustrate that our method are superior or comparable to state-of-art logarithm conversion for graphics hardware.