00:54imirkin: anholt: i'm fine with the general idea of the UBO thing, but someone like karol should have a look. i haven't involved myself in the nir stuff at all.
00:55imirkin: anholt: you can take this as an a-b for any change to the nouveau/ci dir :)
00:58karolherbst: I am sure I already reviewed that one
00:58imirkin: ok, then that's sufficient review
00:58karolherbst: maybe I didn't? let's see
00:58imirkin: i think had that comment about addr regs being 2, but if the tgsi fe does 4, then it's all fine
00:59karolherbst: yeah, tgsi is all 4
00:59imirkin: i guess it's normalized somewhere. or it's not normalized anywhere and it's just wrong :)
01:00karolherbst: maybe we are just lucky
01:00imirkin: addr regs are weird
01:00imirkin: they have that sticky-bit logic to them too
01:00imirkin: (which iirc is disable-able somehow)
01:00karolherbst: what is it doing again?
01:00imirkin: so like
01:00imirkin: it's 16-bit
01:01imirkin: add $a0 $a0 4
01:01imirkin: if that goes to 0x10000
01:01imirkin: then nothing you do will change the $a0 value
01:01imirkin: (except moving an explicit value into it)
01:01imirkin: this makes it so that if you overflow the address, you don't start back at the bottom
01:01karolherbst: I see
01:02karolherbst: sounds like annoying magic behaviour I am sure somebody got annoyed by so the ditched it :)
01:03imirkin: well, no more 16-bit addr regs on fermi :)
01:03imirkin: i think this corresponds to some sort of implicit DX9 behavior
01:03imirkin: or SOMETHING
01:04imirkin: they didn't just put that in because they had transistors sitting around doing nothing
01:04imirkin: iirc it's even controllable via some class method
01:04karolherbst: who knows
01:05imirkin: and most importantly, who cares ;)
01:06karolherbst: I guess nvidia ended up using the file for more stuff than we do?
01:07imirkin: i think it's just to avoid buggy shader code
01:07imirkin: which increments some loop var
01:07imirkin: and does indirect indexing off it
01:08karolherbst: ohh sure.. but I mean we essentially just assign to it once, or do we also manipuate the value?
01:09airlied: I think it's only legal to assign to
01:09imirkin: you can also add
01:09imirkin: you can't even mov, technically
01:09imirkin: you can shl, and you can add
01:09imirkin: so the add is the sticky-bit-setting/preserving thing
01:10imirkin: i'm weak on its details
13:37Sarayan: the vbios tables parsing code, where should I look for it?
13:37Sarayan: for power/heat management
13:39Sarayan: in the kernel or in mesa?
13:48Sarayan: ah, kernel drm
19:29karolherbst: anholt: btw, did you update the shader-db stats in the meantime? I wouldn't be surprised if things got better on nv92
19:29anholt: I had.
19:30karolherbst: ohh, in the commit