Some more justification for my patch
I wrote some code that replaces mpz_millerrabin with a static lookup and
ran it for a number of values.
Existing Code (with real mpz_millerrabin)
100, gap: 1533, 0.00164 seconds
200, gap: 2547, 0.00675 seconds
300, gap: 3757, 0.01161 seconds
500, gap: 3765, 0.
I tried using a segmented sieve. this avoids multiplication or mod in the
inner loop.
It's an improvement and the code is easy to comprehend.
$ ./tune/speed -s 100-300,500,1000 -f 1.1 mpz_nextprime -p mpz_nextprime
$ hg import ~/Downloads/nextprime_segment.patch --no-commit
$ make -j4; make -C tun
Hi,
mini-gmp currently always passes old_size == 0 to the custom
reallocate function, causing custom allocators for GMP that use
old_size to determine the amount of data to copy data to newly
allocated space to malfunction with mini-gmp.
According to GMP documentation chapter 13 for reallocate_fu
