Matt Borland

@jzmaddock The return of `boost::math::tools::bisect` is `std::pair` but it in all test cases, and the example from @L0ric0 `.first` is immediately called from the return. Is `.second` ever a useful...

@NAThompson That sounds like [wheel factorization](https://en.wikipedia.org/wiki/Wheel_factorization). That method is added to the sieve of Eratosthenes, but would not be applicable here.

@NAThompson It would be fairly easy to create light wrappers to the current `prime_sieve` implementation like in the uni-variate statistics library. I will have to do some digging to see...

@NAThompson I am seeing a similar result using a segmented sieve. Currently faster than kimwalish, and `joy_of_factoring_bm` from your post until 524288. After that kimwalish pulls ahead.

@NAThompson I have found some performance improvements using C-style arrays instead of `std::vector::reserve()` which makes sense given that data.

@NAThompson This latest iteration's benchmark delivered much better results: prime_sieve/real_time_BigO 2.87 N 0.22 N prime_sieve/real_time_RMS 11 % 27 %

@NAThompson This should fit the bill. Performance ticked down a bit to 3.07N/0.23N because I realized I was relying on the prime table to be statically linked. If you dynamically...

@NAThompson That makes sense. I can add the notice.

@NAThompson What are your thoughts on defining a `seq_prime_sieve` function instead of wrapping the parallel `prime_sieve` in a block that requires an execution policy. My concern is that execution policies...

@NAThompson Works for me. I saw the notice that C++03 is just now being deprecated so I figured I would ask. I believe that GCC 11 is going to have...