alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - - Logs - <daniels> avoiding X is a huge feature
NeuroScr has quit [Read error: Connection reset by peer]
NeuroScr has joined #panfrost
<hanetzer> if I could get petitboot working as a coreboot payload it would be rather interesting if I could cook a minimal uImage which supports all the gru based chromebooks
<alyssa> hanetzer: #linux-rockchip please ..
<hanetzer> fair. though, does anyone here have any non-kevin gru chromebooks who can test stuff for it?
<tomeu> milloni, TheCycoONE: I think robertfoss has ran panfrost on a rockpi4
<tomeu> alyssa: do you have an opinion on what should be the initial subset of tests to run?
<tomeu> otherwise I'm going to try running all of them at first, then start pruning the ones that cause hangs
fysa has joined #panfrost
hrw has left #panfrost [#panfrost]
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
vstehle has joined #panfrost
afaerber has joined #panfrost
bbrezillon has quit [Ping timeout: 250 seconds]
mateo` has quit [Ping timeout: 246 seconds]
mateo` has joined #panfrost
stikonas_ has joined #panfrost
stikonas has quit [Ping timeout: 246 seconds]
afaerber has quit [Quit: Leaving]
<alyssa> tomeu: No opinions here (I know -- a first for me) :)
<alyssa> tomeu: I am a little concerned about results being meaningful, given we're not nearly conformant right now
<alyssa> I.e. it'll look like all commits fail, which doesn't tell us anything about regressions (or lack thereof)
yann has quit [Ping timeout: 245 seconds]
<alyssa> 17,433 functional tests in total, huh
stikonas_ has quit [Remote host closed the connection]
stikonas_ has joined #panfrost
stikonas_ has quit [Remote host closed the connection]
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
* alyssa wonders how register spilling works
<alyssa> Got a trace of register spilling. Now the fun begins :p
<alyssa> Semantics are still not obvious, I'm guessing we're missing some detail for whatever reason. Keep poking..
embed-3d has joined #panfrost
<alyssa> Ah, I have a hunch what might be going on. Need to modify the disassembler to confirm it.
<alyssa> Also pandecode. Seriously, does all of our infrastructure have bugs?
<alyssa> ("Didn't you write both of the relevant pieces of code?" "...Mmyes..")
_whitelogger has joined #panfrost
<alyssa> A thought -- the only field I see changing in the cmdstream (there may be others we're not dumping, of course) is not in the shader descriptor at all, but in the framebuffer descriptor
<alyssa> Given the layout of both descriptors, it actually... seems more likely the spill memory is in the framebuffer descriptor for whatever reason
<alyssa> There are a lot of random chunks of memory tossed around in the FBD -- might one of them be for spilled memory?
<alyssa> If that's the case, though, I'm not sure why we need to allocate them for shaders that don't spill.
stikonas has quit [Remote host closed the connection]
fysa has quit [Remote host closed the connection]
fysa has joined #panfrost
TheCycoONE has quit [Quit: ZNC 1.7.2 -]
TheCycoONE has joined #panfrost
<alyssa> tomeu: What's the status with devfreq on Panfrost?
<alyssa> I'm on kind of an old kernel version, but rn I'm ending up at low frequencies a lot of the time and it's a little annoying.. If I try to force performance governor, the kernel crashes..
<alyssa> Sweet! common_functions is now up to 400/400 passed.
<alyssa> Running the clipping tests for the first time since there are a bunch.. looking good though.. :)
<alyssa> 602/602 there. Sweet!
<alyssa> Really should fix those crashing/hanging/etc tests so we can do a full run and figure out where we are there :p
<alyssa> Next up, indirect loads/stores
<alyssa> This will be... interesting
<alyssa> I think more broadly giving attention to load/store ops in shaders is a good place to be at right now
<alyssa> (Ties into register spilling stuff as well)
<alyssa> It looks like indirect varying loads are accomplished by normal varying ops
<alyssa> The index is in r27.w
<alyssa> The magic paramemter being 0x79E
<alyssa> Let's get that working, shall we?
<alyssa> ("Why do you use IRC like Twitter?" "Why not?")
cwabbott has quit [Ping timeout: 245 seconds]