03:25 pabs3: imirkin: re that vblank issue you mentioned, I noticed 0ad works despite glxgears/etc not working. any idea what is going on there?
03:26 gnarface: off the top of my head, if it uses sdl, it would be capable of software-blitting failover as a last resort. that's sorta what sdl is for.
03:27 gnarface: sdl isn't using it
03:28 gnarface: at least, my binary here isn't built against it..
03:28 gnarface: *i mean glxgears isn't using it
03:28 imirkin: pabs3: check glxgears again. for me it starts workign on and off
03:40 pabs3: glxgears still dead
19:11 Subv: huh, there are multiple sets of blend methods in Maxwell? 0x133c is BLEND_SEPARATE_ALPHA but there's also an IBLEND stripe at 0x1e00
19:13 imirkin_: independend blend vs single
19:14 imirkin_: iirc this controls it: <reg32 offset="0x12e4" name="BLEND_INDEPENDENT" type="boolean"/>
19:15 imirkin_: check what nouveau does to be sure.
19:15 imirkin_: (i.e. one's per-mrt, the other is global)
20:00 Subv: ah, thanks
22:13 imirkin_: pendingchaos_: Value *sign = bld.mkOp2v(foo, bld.getSSA(), ...)
22:13 imirkin_: pendingchaos_: but i'm not sure what you're doing there....
22:14 imirkin_: an important tidbit to remember though is that -1 / 2 == 0, but -1 >> 1 == -1
22:14 pendingchaos_: is that code from the patch? I don't see it or remember writing it
22:15 imirkin_: pendingchaos_: it's not. it's a simpler way of doing it though.
22:15 pendingchaos_: ah yeah I misunderstood it
22:16 imirkin_: also util_is_power_of_two_or_zero(abs(x)) will return true for 0x80000000, i think.
22:16 imirkin_: which i guess is ok? hm.
22:17 pendingchaos_: it should be doing (a>>31 ? a + (b - 1) : a) >> log2(b)
22:17 imirkin_: ok. that's not intuitively obviously correct
22:18 imirkin_: is there something you can point to which declares it as correct?
22:29 pendingchaos_: gcc seems to do it: https://godbolt.org/g/VMr1U6
22:36 imirkin_: i was thinking more like a webpage we can reference :)
22:36 imirkin_: i'm sure that's right. it's just not intuitively obvious.
22:36 imirkin_: if you can point to e.g. nir doing it
22:37 imirkin_: or a resource
22:37 imirkin_: that'd be great :)
22:37 pendingchaos_: I'm actually having a bit of trouble finding a better source that gcc
22:37 pendingchaos_: I'll check NIR or something in a sec
22:37 imirkin_: nir_algebraic.py is the thing to check, i think
22:37 imirkin_: although... maybe not. that doesn't handle division super-great
22:53 imirkin_: i bet mwk knows this off the top of his head