Dana Jacobsen

This is very concerning. Unfortunately the GCC compiler farm no longer has HP/UX or Itanium machines. I ran on MIPS, ARM, and AIX-Power8 and didn't see it. It is possible...

This looks very similar to the range_construct_almost_prime function in almost_primes.c. It's been languishing there 3 years now. Which version (sieve or construct) is better depends on a lot of things...

So this does turn out to be significantly faster for many examples. Sadly they are not in order so if we're using the values we still need a sort. The...

In theory this is resolved in ad82672b471441051a3c09caea34c506cff77193 today. foralmostprimes: ``` time perl -Iblib/lib -Iblib/arch /tmp/f68-2.pl Count = 10427 Took: 0.001043 ``` almost primes then count the array: ``` time perl...

Should be fixed in PP foralmostprimes {} now as well.

It looks like that was fixed in February in github, with the same behavior. Those changes also tried to consistently use "non-negative" rather than "positive" unless zero is not allowed.

It's definitely useful, and I *greatly* appreciate the comment! Yes, the phi_walk was interesting for its time, but the memory use makes it fairly limited. Your caching phi is extremely...

It is a nice improvement, thank you! I also saw your main loop in src/phi.cpp stops recursing when we know all further answers come from the cache. That's a *big*...

I don't know if it is practical in the caching situation, but for the walk, I used a reduction for both 2, 3, and 5: ``` // Valid for a...

Nice! I just tested collapsing the values themselves, but used the identical 2x reduction (ceil_div(x,2)) for the actual cache storage. So the memory use would be the same, but the...