This is a straighforward port of Dave's algorithm to Forth (tested with GForth).
The state of the sieve is stored in dictionary-allocated memory, and allocated, initialized and deallocated on every iteration.
There are two versions:
prime-bitarray.fs, storing the state in a bit fieldprime-bytearray.fs, storing every prime/not prime flag in a byte
prime-bitarray is closest to the original, prime-bytearray.fs is faster.
./run.sh
gforth 0.7.3:
tjol-1bit;173 ;5.018379 ;1;algorithm=base,faithful=no,bits=1
tjol-8bit;265 ;5.001464 ;1;algorithm=base,faithful=no,bits=8
gforth 0.7.9_20210701:
tjol-1bit;465 ;5.002018 ;1;algorithm=base,faithful=no,bits=1
tjol-8bit;475 ;5.000922 ;1;algorithm=base,faithful=no,bits=8
gforth 0.7.9_20190627 in the Docker container
tjol-1bit;256 ;5.001076 ;1;algorithm=base,faithful=no,bits=1
tjol-8bit;418 ;5.0083 ;1;algorithm=base,faithful=no,bits=8
This is on a Ryzen 5 2600 running OpenSUSE Tumbleweed.