21:22:56alandipertpushed a JACL project update that suggests possible future overlap with SICL via ECL https://gist.github.com/alandipert/a79a345f1b4c822c4d8ddaf992a5fc30
0:01:13no-defun-allowedBike: Do you know which instructions I should emit for an unconditional atomic store? G++ emits MOV then MFENCE for a std::atomic<int>.
0:01:50Bikelike do you want it to be sequentially consistent? i think you do need a fence for that
0:02:52Bikebut i've mostly worked through LLVm so i'm not edealing with the actual instructions much
0:03:33no-defun-allowedGood question. /me checks a list of memory orders.
0:03:54Bikeyou probabgly want sequentially consistent, just because anything less gets hard to understand very quickly
0:04:11no-defun-allowedI'd do it just after a (successful) CAS, updating the metadata table, so there'd probably no point in trying to find a weaker model.
0:05:07no-defun-allowedAnd it'd replace something like that CAS loop you mentioned some time ago.
0:05:32Bikeright. the default is sequential conssitency, that's what the CAS is doing