geofflangdale

@Kynx @ChuckMcManis @lcamtuf I don't get the big deal about Chesterton's Fence and think it's probably safe to remove it from our lexicon tbh

@regehr@mastodon.social the most haunting element of this is that in your average group of programmers or grad students you will never, never be allowed to forget this. One of our devs had a commit message involving "transistions" rather than "transitions" and no-one could resist using "transistions" in his presence for the next 5 years whenever NFA/DFA transitions came up. Which was *often*, since we were a automata-theory-based startup.

@regehr@mastodon.social Instructions I Won't Work With: PCMPESTRM

@regehr@mastodon.social unfortunately I think a chair-moistening CS nerd complaining about misbegotten instructions making his programs slightly slower lacks the built-in humor of TIWWW. It would be impossible to replicate his hilariously understatement in response to life-threatening explosions, toxic fumes and horrible stenches. Also, from a personal perspective, I would find my chosen take on some instructions (BEXTR and VPINTERSECT), ahem, career-limiting.

@synlogic4242@social.vivaldi.net @regehr@mastodon.social at the time they were delivered. Due to the fact that these instructions are quite unloved, they were not promoted to 256b in AVX2, much less 512b, so they have become even less competitive, but they were *never* competitive against the PSHUFB-table approach for most of the 'we're looking for a known-in-advance table of characters' use cases. Never. I guess I *did* just write the TIWWW article in your mentions, sorry.

@synlogic4242@social.vivaldi.net @regehr@mastodon.social To be honest, googling anything in SSE4.2 but CRC32 (why is this SSE4.2? who knows?) and PCMPGTQ (a 64b lanewise greater-than comparison, peculiarly absent previously) is mostly a waste of time. There are still some uses of the PCMP* string matching instructions - particularly on 16b quantities and in the context where you don't know either side of the comparison well in advance - but they have the dubious distinction of being obsolete for much of their intended use ...

@pinskia@hachyderm.io @penguin42@mastodon.org.uk @regehr@mastodon.social That's interesting. I never heard of the OzLabs thing until now. Companies are often arm-twisted by the Australian government to have some R&D here and abandon it the minute the incentives end (Microsoft, iirc, had its first "Labs" in Australia too).

@penguin42@mastodon.org.uk @pinskia@hachyderm.io @regehr@mastodon.social I've long wished that there was a sane and stable CS-focused broad-based research lab somewhere in Australia. NICTA turned into a "bonus system for existing players" (some quite good, but no-one had to quit their day job), CSIRO was a bullying factory with low pay, CiSRA was too weird (Silverbrook!), and most of the small corporate labs were "satisfy ausgov for X years".