We now have almost complete results for the latest hgcd2.c tweaks. We have 5 div1 variants (as chosen by HGCD2_DIV1_METHOD). There are results and brief documentation here:
https://gmplib.org/devel/thres/HGCD2_DIV1_METHOD.html About half of our test machines prefer '/', i.e., whatever word division the compiler comes up with. About one quarter prefers method 3, which handles n / d < 8 with plain and branch-free code, but '/' for greater divisors. About one quarter prefers table-driven inverse-and-multiply, which is in practice branch-free. The even-numbered methods are clearly unpopular. -- Torbjörn Please encrypt, key id 0xC8601622 _______________________________________________ gmp-devel mailing list gmp-devel@gmplib.org https://gmplib.org/mailman/listinfo/gmp-devel