14:12karolherbst: heh "(i->op == OP_SUB) != static_cast<bool>(i->src(1).mod.neg())"
14:49karolherbst: imirkin: mhh, actually, I think that CTSW_TIMEOUT I was hitting has something to do with increased workload on the GPU or something even more stupid :/ was playing some civ5 and the overall scene is getting more and more complex over time and those context switch errors are also becoming more and more frequent at the same time :/
14:50karolherbst: maybe we should just specify a max size for pushbuffers?
14:51karolherbst: skeggsb: any ideas?
15:04karolherbst: imirkin: https://github.com/karolherbst/mesa/commit/06a5b89e9dc88fc70d8f4354887a8fe16eec254e
19:40karolherbst: mhh, just 250 tests failing
20:01karolherbst: all fails are 64 bit related...
20:02HdkR: karolherbst: btw, you don't currently promote large immediates to uniform buffer?
20:35karolherbst: HdkR: nope, we don't
20:36karolherbst: I was thinking about using the driver constbuf for that as we should have enough space in there
20:36karolherbst: and we rebind it for every shader anyway
20:37HdkR: karolherbst: Currently you just materialize them using a move?
20:37HdkR: oof :P
20:37karolherbst: we have that RA opt for fmas though
20:38karolherbst: so it's not that bad
20:46karolherbst: .... I64NEG lowering depends on OP_SUB :/
20:48karolherbst: imirkin: do you think it would be fine emiting an add with the neg modifier directly when converting from tgsi?
20:49karolherbst: otherwise we would run into an infinite loop emiting it
20:49karolherbst: because I64NEG is sub(0, a) and we would have to emit add(0, neg(a)) without a sub :/
21:27karolherbst: mhhh, that doesn't work either :/