<alyssa>
We really need to get Panfrost merged upstream, oy
<alyssa>
Fixed... again... rebuilding...
<alyssa>
We're... missing.. files.. from.. git.. gah
<alyssa>
Well, it's been nearly 2 months since the last rebase, so most of mesa is recompiling
<alyssa>
Guess I'll let that stew until it gets to driver files and I find a dozen NIR APIs broke
* alyssa
blinks
<alyssa>
$ git diff upstream/master | wc -l
<alyssa>
26981
<alyssa>
That's.. not good
<alyssa>
Here there are, lovely, lovely
<alyssa>
Alrighty
<alyssa>
Latest panfrost/mesa master is rebased on latest upstream mesa master
<tomeu>
robher: hi there, have you pushed your 5.0 rebase somewhere?
<alyssa>
Eh, well, let me figure out how much work it is to get some code upstream
<tomeu>
alyssa: amazing :)
<alyssa>
tomeu: Not like.. real usable code
<alyssa>
But at least some stubs so I can stop going crazy with rebase
<alyssa>
Going through the diff of master/us and cherrypicking, essentially
<alyssa>
(Implicitly squishing history)
<alyssa>
Things I'm skipping: buffer age (should be its own thing, I didn't write it), blend lowering (should also be its own thing, I did write it tho :P), ..
<alyssa>
NIR vectorize (already submitted upstream, I didn't write it)
<alyssa>
- panwrap (it's a lot of code, most of it of questionable quality, and it's not needed upstream, at least not yet)
<alyssa>
What's left has no real blockers to merge, but is also not useful since kernel :p
<alyssa>
Well, there'll be the standalone Midgard shader toolchain, which is useful in and of itself
<tomeu>
looks good to me!
<alyssa>
But anyways, the point of this rn isn't getting it shipping, it's just having the code upstream to make rebasing breezy again
<tomeu>
alyssa: what's going to contain the first series to be submitted?
_whitelogger has joined #panfrost
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
pH5 has joined #panfrost
robertfoss_ has quit [Ping timeout: 244 seconds]
robertfoss_ has joined #panfrost
raster has joined #panfrost
<robher>
tomeu: pushed panfrost-5.0 to my fdo tree. There wasn't much to it. It all just applied and one function changed.
<tomeu>
robher: ok, are you planning to work on gem shmem helpers?
<robher>
tomeu: yes, or give up and just copy the etnaviv code...
<tomeu>
robher: I have started copying the etnaviv code :)
<robher>
tomeu: perhaps look at the MMU. My first question is freedreno uses the iommu api and etnaviv does not.
<tomeu>
robher: which iommu api are you referring to?
<robher>
tomeu: the kernel's...
<tomeu>
robher: but etnaviv does use the iommu subsystem
<tomeu>
oh wait, maybe I mixed things
<robher>
tomeu: NM, I was confused...
<robher>
tomeu: msm calls iommu_map_sg and uses iommu_domain for example. etnaviv doesn't.
<tomeu>
yeah, I was looking at both as well and somehow came to think that it was etnaviv that was using it... :/
<tomeu>
robher: I think I will keep banging on etnaviv's for now
<tomeu>
I really want to get to command submission as soon as possible, as I think at that point is easier for more people to work on this in parallel
yann has joined #panfrost
Elpaulo has quit [Quit: Elpaulo]
Elpaulo has joined #panfrost
BenG83 has joined #panfrost
<alyssa>
tomeu: Mali Midgard+ is unlike a lot of GPUs in that its "command stream" is composed of descriptors in (shared/mapped) memory... so you really do need MMU going before you can think about job submission..
<tomeu>
the idea was to implement enough of PANFROST_GEM_NEW to get panfrost_drm_allocate_slab working
<robher>
tomeu: hum, that's what I'm reworking on etnaviv...
<robher>
BTW, AFAICT none of the userptr stuff in etnaviv is used anywhere.
<tomeu>
robher: reworking?
<robher>
tomeu: to use shmem lib. I was taking the approach to rework in etnaviv and then copy rather than copy and then rework.
<tomeu>
ah, cool
<tomeu>
are you having to write a iommu driver as well?
cwabbott has quit [Quit: cwabbott]
<robher>
things like etnaviv_gem_ops can be replaced with the standard GEM funcs struct. Like any new driver, what's recommended practice has changed since any in tree driver was written...
<robher>
tomeu: As soon as we want h/w to touch any GEM buffer, we'll need it. I've only looked at it in the sense of what GEM obj data does the MMU code use.
<tomeu>
ok, sounds great!
<HdkR>
alyssa: Nice. Fix the typo in the commit message but I approve :)
yann has quit [Ping timeout: 250 seconds]
BenG83 has quit [Quit: Leaving]
stikonas has joined #panfrost
pH5 has quit [Quit: bye]
pH5 has joined #panfrost
raster has quit [Remote host closed the connection]
Elpaulo has quit [Remote host closed the connection]
Elpaulo has joined #panfrost
yann has joined #panfrost
Elpaulo has quit [Quit: Elpaulo]
cwabbott has joined #panfrost
NeuroScr has quit [Read error: Connection reset by peer]
NeuroScr has joined #panfrost
<stikonas>
hanetzer: you need to rename rockchip -> kmsro in your gentoo ebuild.
<stikonas>
it compiles with that change
<hanetzer>
stikonas: compilation wasn't a problem as far as I knew, just its not getting the mali regulator for me :<
* alyssa
updates instructions
<hanetzer>
stikonas: I assume you mean the use_enable bits?
<alyssa>
Instructions updated
<HdkR>
alyssa: btw, is that a patch for upstream that you posted?