01:07 orbea: imirkin: if you recall this GLES3 blackscreen with glupen64 and GL_EXT_buffer_storage. https://github.com/GLupeN64/GLupeN64/issues/107 Someone else tested it with intel and found it also occurs there, I guess its not specific to nouveau.
01:07 imirkin: coolio
11:44 zophyx: is there a nouveau driver for debian jesse that will use the Maxwell chip ?
11:54 pmoreau: Depends on which kernel is being shipped by Debian Jesse
14:37 NanoSector: [23:45:04] <Lekensteyn> NanoSector: I've survived for years using a 80GB Intel SSD :)
14:37 NanoSector: I don't manage to survive with a 256 GB SSD :(
14:37 NanoSector: muh gamez
14:50 karolherbst: pmoreau: jessie is ancient already :p 3.16 kernel afaik
14:50 pmoreau: …
14:51 pmoreau: It probably doesn’t even have modesetting
15:10 NanoSector: has debian ever been up to date?
15:21 pmoreau: Well, it is up-to-date with some point in the past ;-)
15:24 NanoSector: heh fair enough
18:07 urmet: eh. you can get 4.6 from jessie-backports
18:08 urmet: 4.7*
18:09 urmet: defaults are super conservative yes
18:09 urmet: but remember, jessie was released almost 2 years ago
20:19 karolherbst: imirkin: you wanted to look into the postraloadpropagation thing again, any new eta on this?
20:19 imirkin_: none
20:19 imirkin_: get hakzsam to review it
20:20 hakzsam: I can do that, but tomorrow, sorry busy with my scheduler
20:20 imirkin_: while i'm busy with my schedule :)
20:20 hakzsam: :)
20:22 karolherbst: ....
20:22 karolherbst: hakzsam: thanks
20:22 imirkin_: wasn't my question "why do we need this"
20:22 imirkin_: iirc i never got an explanation
20:22 karolherbst: huh?
20:22 karolherbst: regarding what?
20:23 imirkin_: post-ra load propagation
20:23 imirkin_: or was this the regs-have-to-be-just-right 3-arg mad/etc?
20:23 karolherbst: the posr ra load propagation cuts instructions count by 0.3% ;)
20:24 imirkin_: at a very high cost iirc
20:24 imirkin_: (but maybe not anymore)
20:24 karolherbst: well there was the RA thing
20:24 imirkin_: anyways, tbh, i have no clue which patches these are
20:24 karolherbst: the fmad with the limm
20:24 imirkin_: just vague memories
20:25 imirkin_: ah, that's worthwhile
20:25 karolherbst: and you wanted to do something about this: https://github.com/karolherbst/mesa/commit/96001592997c465450ee12514809676a5129e341
20:25 karolherbst: where I should only enable that RA part if there is an actual immediate
20:25 imirkin_: right - i don't think you can do that
20:25 karolherbst: well it works
20:25 imirkin_: heh
20:25 imirkin_: it works SOMETIMES
20:25 karolherbst: you just sai you wanted to check somethign regarding this
20:25 imirkin_: doesn't mean it always works
20:26 imirkin_: e.g. let's say you have
20:26 karolherbst: well, the entire shader-db run was fine
20:26 imirkin_: float x;
20:26 imirkin_: if (a) { x = 1.0; } else { x = uniform; }
20:26 imirkin_: fma(a, b, x)
20:26 imirkin_: your logic might decide that x is an immediate.
20:26 imirkin_: so ... no crash, on shader-db
20:26 imirkin_: but just wrong code generated.
20:26 karolherbst: why should it?
20:26 imirkin_: rtfs
20:27 imirkin_: getImmediate ain't magic
20:27 imirkin_: it really only works during SSA
20:27 karolherbst: I see
20:27 imirkin_: at this point in allocateRegisters iirc the defs have been merged
20:27 imirkin_: but maybe not
20:27 karolherbst: well we can get the stuff in without the RA part and that I can deal with later
20:27 imirkin_: either way, it'll require careful analysis
20:28 karolherbst: sure