11:48 RSpliet: karolherbst: yes that's a very useful tool. Thanks!
11:51 ccaione: r
13:02 karolherbst: RSpliet: so you toyed a bit around with that (or read the source) or just a comment about having something like this in nice and useable?
13:03 pmoreau: karolherbst: So, when can we expect a shader debugger for Nouveau? ;-)
13:03 karolherbst: :D later
13:03 karolherbst: first the falcon one
13:03 pmoreau: Eh eh :-)
13:04 karolherbst: getting the source code will be messy
13:04 karolherbst: and stuff
13:04 pmoreau: I'd like to help on the shader debugger, so feel free to ping me whenever you start looking into it
13:04 pmoreau: Sure
13:05 karolherbst: well, you can start by reing everything :p
13:05 pmoreau: I'll have a look at the code sometime, for curiosity reason
13:05 karolherbst: no idea if somebody already did that
13:06 pmoreau: Ah ah ah! I'll start with moving codegen to handling multiple entrypoints and fix the Niuveau image…
13:06 karolherbst: :)
13:06 karolherbst: we need a trap handler first aka a way to detect broken shaders
13:07 pmoreau: True
13:10 karolherbst: the next thing I want to implement in the debugger is falcon detection. I don't know if there will be any good source code parsing anyhow and I also don't really care. We have the "address: symbol" comments and parsing those should be enough already
13:10 karolherbst: and then we get intr+0x1f in the debugger
13:13 pmoreau: So, too have something similar to `x/i $pc` in gdb?
13:13 pmoreau: s/too/to
13:18 karolherbst: yes
13:18 karolherbst: allthough it is already in the script
13:19 karolherbst: but you don't get the offset to the symbol/marker
13:19 karolherbst: so you know what instructions is at the $pc, but you have no clue from where it comes
13:34 pmoreau: ok
13:35 karolherbst: and the .fuc*.h file path is hardcoded
13:35 karolherbst: I could pimp envyas to put some debugging information into the .h file though
13:35 karolherbst: and just read it out in the debugger
13:43 RSpliet: karolherbst: no didn't toy around, just encouraging you to keep up the good work
13:44 RSpliet: wish I had time to toy around... again :-C
13:45 karolherbst: well there is always next year
13:46 RSpliet: well, considering the rate at which I'm going bald and grey, I'm not so sure about that anymore...
13:51 karolherbst: well if there is no year anymore, the last one person who cares is yourself anyway
15:38 Tom^: RSpliet: oddly enough we all bald up and gray at the same rate. :P
15:39 Tom^: time is such silly constant
15:48 karolherbst: Tom^: no we don't
16:21 Tom^: i guess that came out wrong, i meant more of we all age one day each day that passes. :P
19:14 pmoreau: I find it super highly suspicious that hell didn't break loose yet… I must be doing something wrongly. :-/
19:14 karolherbst: or maybe nothing at all :p
19:14 pmoreau: :-D
19:15 pmoreau: Well, I did remove the main attribute from Program, and tested a program with two kernels, each having its own Graph
19:16 karolherbst: using the system libraries is a common mistake :p
19:18 pmoreau: :-p
19:56 Lyude: radeon is such a miss, every time I have to fix bugs in it I miss working on nouveau :(
19:57 Lyude: *mess
19:58 karolherbst: Lyude: then just work on nouveau :p
19:58 airlied: Lyude: it's abandonware pretty much
19:58 airlied: feel free to port it to atomic and rewrite :)
20:07 Lyude: karolherbst: hehe, unfortunately I accidentally fell into a very deep rabbit hole here
20:07 Lyude: i'm trying to confirm this on another machine to make sure nothing bizarre is happening here, but vblank handling in radeon appears to be kind of broken
20:08 Lyude: and it appears to have been broken for a lot longer then I would expect it to have been
20:08 karolherbst: I have two radeon cards right beside me :D
20:08 karolherbst: wanna have a powermac G4 card 9200 pro with ADC? :p
20:21 Lyude: karolherbst: nah, wouldn't have much use for it :P
20:21 Lyude: kinda try to avoid radeon.ko
20:22 Lyude: airlied: to be honest I would like to also see amdgpu rewritten for atomic commit, but i basically stay away from that too because i have no idea if anything I do on it is just going to get dumped for <whatever AMD ends up renaming DAL to next> at some point anyway...
20:23 karolherbst: Lyude: they have their own devs :p
20:24 Lyude: that as well
20:35 pmoreau: I shouldn't have complained that it works: it just became a huge mess to propagate the information further down, going through various structures here and there…
22:25 pmoreau: Shared mem variables are declared globally in SPIR-V… that is annoying
22:27 pmoreau: I'll have to manually track where they are used, and bumped the corresponding kernels' smem usage.