Benchmarks with and without compiler optimizing

Torbjörn Granlund tg at
Tue Oct 29 15:18:23 UTC 2019

Hans Åberg <haberg-1 at> writes:

  But that is not somebody has benchmarked? — Somebody tested human vs
  computer optimization, and found that computers could do better, but
  then there might be a tradeoff with reliability.

You seem to be talking about two separate issues.  One is what happens
with an asm enabled build with no -O for the C code, and the other is
whether our asm code is better or worse than compiler generated code
(presumably with compiler optimisation turned on).

The first issue is something I'd suggest that you benchmark for your
intended application (although the results of such benchmarking don't
seem all that interesting).

If you want to compare our asm code to plain C code, just pass
--disable-assembly to configure.

Please encrypt, key id 0xC8601622

More information about the gmp-discuss mailing list