···4343# on Cortex-A53 (or by 4 cycles per round).4444# (***) Super-impressive coefficients over gcc-generated code are4545# indication of some compiler "pathology", most notably code4646-# generated with -mgeneral-regs-only is significanty faster4646+# generated with -mgeneral-regs-only is significantly faster4747# and the gap is only 40-90%.4848#4949# October 2016.