libera/#sbcl - IRC Chatlog

17:56:07 stassats flip214: avx has nothing to do with fasls

17:57:34 stassats the comments around bignum-truncate are quite wrong...

17:58:56 stassats so are they wrong and it's ok, or is the code wrong

18:28:58 flip214 stassats: so I can build sb-simd on a non-avx machine and use compiled libraries on an avx machine, with speedup? And the other way around as well?

18:29:15 stassats you could, except that you can't

18:29:27 stassats as sb-simd uses avx when building

18:30:00 stassats not sure if just during constant folding

18:32:14 |3b| presumably simd code would do a runtime check and not call the avx versions of functions if they won't work

18:32:58 stassats that happens with the avx2 routines that coincidentally are in this release, but not related to sb-simd

18:33:51 |3b| ACTION hopes the ability to do that stuff ended up in the final thing, since that was one of my requirements. requiring it to build is more or less OK, but need to be able to make binaries that run anywhere (even if that needs some coding to enable)

18:34:24 stassats don't know about sb-simd and multi-versioned functions

18:35:14 _death I think the sb-simd docs mentions this

18:35:33 stassats in normal sbcl: #'sb-vm::simd-nreverse8 => #<FUNCTION SB-VM::SIMD-NREVERSE8-AVX2>, gets redefined at startup

18:36:21 stassats sb-ext:*init-hooks* => (SB-SIMD-INTERNALS::UPDATE-IDISPATCH-INDICES)

18:36:26 stassats i suppose that's the thing?

18:40:33 |3b| yeah, looks like that's probably it

18:41:59 |3b| looks like users use instruction-set-case, which presumably uses that hook

20:12:05 Shinmera neato