Re: simple string comparison for efficiency

[etj]

I'd use PDL, with some overly-clever finagling to maximise locality:

• Get each string as columns 0 and 1 of a pdl(byte, 2, 1e8). With PDL's row-major storage, that would put each pair of characters next to each other in memory.
• Set its "badvalue" to be ord('N'), because the ne operator will ignore that pair if either value is "bad".
• Run PDL::any(PDL::ne($p->using(0,1)))
• Sadly this will currently not (yet) loop-fuse the ne and the any, but locality could be maintained by doing a splitdim into smaller chunks, and doing the any on each of those as you go.

Yes, I enjoy that in the third bit, "any" sounds like the normal pronunciation of "ne". I'm only human.

If you want to make a custom transformation that does loop-fusion manually, see the SYNOPSIS of Inline::Pdlpp for an also-bio-inspired example.

Edit: if one chooses to not loop-fuse, the ne produces a mask that is true for any non-matches; that can then be fed into which to give the positions of those.

Edit 2: the above should take a matter of minutes to write, which would be much quicker than writing it in C/C++, and should run pretty quickly. But the naive, 2,1e8 solution has the benefit of broadcasting, which would mean it would also benefit from auto-pthreading, which in C/C++ would be a lot of hard work.