Nicolai Hähnle

Math, compilers, GPUs, horrible puns, off-by-one errors, and Oxford commas.

Last Notes

2026-05-12 06:28:09 UTC

- reply

@nprofile…r9vg @nprofile…5wht That requires better CI. One of my fundamental beliefs is that we're all doing CI wrong.
If a PR causes a test to fail 5% of the time, chances are high that the PR will successfully land. And then cause problems down the line.
Passing all tests on every PR isn't the correct goal for a number of reasons, one of which is that tests can be flaky. The correct goal is zero regressions.

2025-11-17 23:32:37 UTC

- reply

@nprofile…xlvy I'm only aware due to my niblings who aren't that old yet, so I don't know.

2025-11-16 21:46:04 UTC

- reply

@nprofile…xlvy The mentality there seems to be similar to vaccination, basically. It's something that your pediatrician will tell you to do as part of the standard health plan for kids.

2025-11-16 21:45:03 UTC

- reply

@nprofile…xlvy Kids in the Netherlands get tiny doses of peanut almost from birth apparently.

2025-08-20 14:28:44 UTC

- reply

@nprofile…6cfc Very nice. With a slight tweak to the curve you could make a perfect loop.

2025-08-13 15:16:56 UTC

The trick for swapping two values without extra storage and without a dedicated swap instruction by three XORs is reasonably well known.
If you want to rotate N values, you can just do this as a sequence of swaps which under these constraints means 3(N-1) XORs.
Is there a cheaper sequence of XORs? If so, what's the asymptotic bound?

2025-07-29 12:44:26 UTC

- reply

@nprofile…gvnm You wouldn't download a bird!

2025-04-15 12:26:04 UTC

- reply

@nprofile…cfuw @nprofile…7pn8 Porque no los dos?

2025-02-27 19:36:48 UTC

- reply

@nprofile…qmju Link doesn't work for me.