Currently the spectral_to_rgb function can't be auto vectorized by gcc due to "complicated access pattern".
This PR solves that by using a temporary array.
This changes the IR required per call according to callgrind from from 206 to 122 with sse2 and 99 with avx.
Currently the
spectral_to_rgb
function can't be auto vectorized by gcc due to "complicated access pattern". This PR solves that by using a temporary array.This changes the IR required per call according to callgrind from from 206 to 122 with sse2 and 99 with avx.