Post #2766117
2025-12-31 00:14 UTC
@thephd@pony.social For the record, macOS thread-locals were notoriously slow back when I worked at Apple, because they’re hidden behind function calls in order to not make the special thread-local data register part of the ABI. I don’t know if that’s still the case, but it may be interesting for you (or someone else) to re-run on Linux. Or it may not be, I can’t imagine it’s ever going to be cheaper than the static case anyway.
Replies (1)
-
@thephd@pony.social 2025-12-31 16:32
@jrose@social.belkadan.com could be! I got similar base numbers on x64 Windows 10 but did not do all 150 repetitions, just one. If someone reminds me in 6 hours I can probably do a test again with repetitions on and graph generation.