This benchmark, lut and divide&conquer lut, seems to be really slow compared to paul's new itoa. His biggest trick is to preallocate the target buffer size, and to avoid all this heavy branching.