"Marco Bodrato" <bodr...@mail.dm.unipi.it> writes: May I suggest to write:
u0 = (u0 >> c) | (u1 << (GMP_LIMB_BITS - c)); instead of u0 = (u0 >> c) || (u1 << (GMP_LIMB_BITS - c)); You may. And it is indeed a great improvement! The generated code for Niels' C code is awful for x86. It consists of mainly register-to-register copying. We might want another C variant for machines where we don't provide assembly. I have written just arm32, arm64, x86-64, ppc64-power7, and ppc64-power9 code. I might try an x86-32 but then I'm done. I believe an amalgamation of Niels code which uses an implicit lsb and his present _22 code would be best. The present requirement for sub_mddmmss is a portability inconvenience, and for subtract-with-carry challenged processors a major hassle. -- Torbjörn Please encrypt, key id 0xC8601622 _______________________________________________ gmp-devel mailing list gmp-devel@gmplib.org https://gmplib.org/mailman/listinfo/gmp-devel
