ONE iteration Newton-Raphson reciprocal

for better performance,
i want to revised the recirpocal function in reference source code(Vincent)
from 2 ietration to only one iteration without loss precision ,but i don’t know how to do this. does anyone know that?


i have solve it, but i have another question, for x86 CPU, are there exist a count leading zero instruction?

Yap, take a look at the documentation for the BSR (bit-scan-reverse) assembler instruction. You’ll need to use your compiler’s inline assembler to generate the opcode. Alternatively, check if your compiler supports this as an intrinsic (GCC for example does).


So how were you able to get the same precisicion with less iterations?

  • HM

This topic was automatically closed 183 days after the last reply. New replies are no longer allowed.