primes

A submission for the Software Drag Race

• primes_faithful_st.ml is around 8-10x faster than the current submission.
  • primes_faithful_im.ml is the same implementation but with imperative constructs, showing that OCaml can optimize tailrec functions transparently.
• primes_faithful_mt.ml is around 30-50% faster than its single-threaded counterpart with nproc=4
• primes_faithful_mc.ml uses the Multicore OCaml compiler, the perf differences between it and conventional threaded OCaml code are marginal*
• primes_faithful_fk.ml uses fork() and pipe(), hitting upwards of 2x speedup for nproc=4 compared to single-threaded. Amazing what the classics can achieve.
• primes_faithful_1b.ml uses a compact 1-bit repr. Right now it's quite slow but the plan is to batch-modify 8-bits at a time and see if this gives any nice speedup.
• primes_faithful_hb.ml is built on the premise that keeping all data limited to minor-heap capacity could produce faster code. The speedup is balanced by the cost of extra indirection in practice.
• primes_faithful_b{b,s,m}.ml is a direct translation of Rust submission's bitstripes implementation to see if it brings any good speedup. It does. Around 40%.

* NOTE that, when I compare the single-threaded Trunk OCaml compiled binaries with the Multicore OCaml ones. The performance then appears marginal. However, when comparing single-threaded Multicore-compiled binaries with parallel Multicore-compiled ones, I do get that relative 2x speedup as well. Unfortunately this exercise aims to hit the highest raw performance out of all OCaml compilers, not compare running times within the context of the Multicore compiler only.