00:29 <alyssa> HdkR: I'll find out on Monda!

00:29 <anarsoul> :)

00:29 <anarsoul> what about Mesa/Lima?

00:30 <HdkR> alyssa: Apparently in GL/ES land it doesn't matter, so yolo

09:20 <shadeslayer[m]> <alyssa "So each vertex jo-- yeah, that s"> So I think the code is fine

09:21 <shadeslayer[m]> I /think/

09:21 <shadeslayer[m]> The way to visualize it is : Set Value -> Vertex -> Vertex -> Tiler -> Tiler

09:21 <shadeslayer[m]> then you generate a new Vertex job ( next_job )

09:22 <shadeslayer[m]> It links the last Vertex job to next_job

09:22 <shadeslayer[m]> and links next_job to the first_tiler job

15:31 <nuttersymbios> alyssa: you had questions if regalloc stuff can be changed to allow better scheduling.

15:32 <nuttersymbios> alyssa: answer is yes, but exactly how that goes i have not checked yet

16:37 <nuttersymbios> but in short, i am trained to do this kind of stuff. If you manage to start to behave like humans , i make the code without royalities to the end-user.

17:39 <nuttersymbios> this is only very complex issue because the field of scheduling is very wide/large, variety of code would do, what instructions compiler lifts to, how the FUs are remapped to queues, how you determine the decode wfids etc.

17:41 <nuttersymbios> but i am used to carrying weights like that, programmers have to land on mutual agreement to try various ways during different release cycles..release 91 has that type of scheduling agreed to be programmed 0.2 with slight changes etc. finally the best is chosen in 9.1

17:42 <nuttersymbios> 91/0.1

17:45 <nuttersymbios> when chip is not a very intelligent one as postfabbed, you can always fall back to wfid column tracing in instruction queues, which relies on the fact that fetch and wavepool pipeline is always deeper than an alu which asks which wfid was that i got branched into

17:55 <nuttersymbios> and yes, programming can be very complex and even stressful like doctors work when someone needs results in performance, it is exactly the same in sports actually, you can play without much effort put into the game easily, but when someone needs you to win the game, you got to give a lot more to earn a victory

18:00 <nuttersymbios> similarly you can do half-baked programming almost with ease, but if someone were to set the requirements to higher level of expertise, things get more complex, as if someone reuiring to win every game and or tournament

18:00 <HdkR> nuttersymbios: Keep the discussion panfrost related or I'm going to have to knock you out.

18:02 <shadeslayer[m]> royalty free? what?

18:03 <HdkR> shadeslayer[m]: Just don't

18:03 <shadeslayer[m]> heh ok

18:05 <nuttersymbios> HdkR: it was mostly related, you may knock yourself out , and do that please fucking troll! The fetch/wavepool/decode and schedule pipeline is roughly implemented the same on any kind of graphics chip, does not matter if that is panfrost, or frozendeludedHdKR

18:06 <shadeslayer[m]> Well, I guess it's time to have that beer

18:15 <HdkR> It comes down to this. If someone comes in here with a poisonous personality and start talking, I'm going to destroy them. Doesn't matter who they are. I refuse to allow someone to come in and ruin the mood of the channel

18:17 <HdkR> I'm fine with consuming someone's spite if it means the rest of the channel is improved

18:17 <urjaman> thank you

20:26 <hanetzer> wtf. where did that even come from?

20:27 <HdkR> They are a person that is common to all mesa related channels.

20:28 <hanetzer> ah, so mayhaps his comments were in regard to something said in another channel, gotcha.