alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - Logs https://freenode.irclog.whitequark.org/panfrost - <daniels> avoiding X is a huge feature
narmstrong has quit [Read error: Connection reset by peer]
narmstrong has joined #panfrost
steev has joined #panfrost
ezequielg has joined #panfrost
lvrp16 has quit [Ping timeout: 246 seconds]
ezequielg has quit [Read error: Connection reset by peer]
austriancoder has quit [Ping timeout: 250 seconds]
steev has quit [Read error: Connection reset by peer]
robher has quit [Ping timeout: 245 seconds]
daniels has quit [Ping timeout: 260 seconds]
robclark has quit [Ping timeout: 264 seconds]
narmstrong has quit [Ping timeout: 245 seconds]
jstultz has quit [Ping timeout: 260 seconds]
ric96 has quit [Ping timeout: 248 seconds]
Green has quit [Ping timeout: 248 seconds]
megi has quit [Ping timeout: 240 seconds]
nerdboy has quit [Ping timeout: 265 seconds]
nerdboy has joined #panfrost
ezequielg has joined #panfrost
nerdboy has quit [Ping timeout: 260 seconds]
stikonas has quit [Ping timeout: 272 seconds]
ezequielg has quit [Ping timeout: 272 seconds]
ezequielg has joined #panfrost
ezequielg has quit [Read error: Connection reset by peer]
ezequielg has joined #panfrost
<alyssa> anarsoul: Okay, compromise is to do it like we used to but with an OFFSET(..) macro
<alyssa> it's not as elegant as having extra srcx/srcy parameters but it keeps us from mucking with the inner loops
<anarsoul> can't you just calculate your offset inside the actual routine when you have src_x and src_y?
<anarsoul> exposing optimizations in interface is usually not a good idea
<anarsoul> I don't have strong preference though as long as you keep maintaining this code :)
ezequielg has quit [Ping timeout: 272 seconds]
bbrezillon has quit [Ping timeout: 268 seconds]
bbrezillon has joined #panfrost
<alyssa> anarsoul: :P
<alyssa> It's not in interface
<alyssa> The interface to these routines is the load/store functions
<alyssa> That *does* change in this series -- from using a pipe_box to x/y/w/h directly since I don't want to depend on Gallium (not an opt)
<alyssa> All of the offset goofiness is contained to static routines in the file
<alyssa> aaaand deqp regressed weeeeeee
<alyssa> 2 steps forward..
NeuroScr has quit [Quit: NeuroScr]
<alyssa> fixed
<alyssa> It's pretty wild to me that panfrost *feels* tangibly faster today (with this series) than yesterday
<alyssa> it's like woah, we're fast of all a sudden?!
<alyssa> So a lot of the yak shaving is paying off :)
daniels has joined #panfrost
daniels has quit [Excess Flood]
daniels has joined #panfrost
<anarsoul> alyssa: I believe it's mostly -O3 thing :)
daniels has quit [Ping timeout: 245 seconds]
anarsoul has quit [Ping timeout: 268 seconds]
anarsoul has joined #panfrost
<icecream95> Darkplaces now works with lighting on full. :)
<icecream95> I won't say it's perfect, but it's definitely playable, at least with vblank_mode=3
<icecream95> You'll also need to set MESA_GL_VERSION_OVERRIDE=3.3 to get it to work
<icecream95> OpenMW crashes with ../src/panfrost/midgard/midgard_compile.c:1651:emit_intrinsic: Assertion `0' failed.
<HdkR> Unsupported intrinsic looks like
davidlt has joined #panfrost
<icecream95> The intrinsic is nir_intrinsic_load_barycentric_centroid
<HdkR> `centroid varying vec4 lighting;`
<icecream95> I added an empty case statement for it and now OpenMW with shadows enabled "works"...
<icecream95> Only about 1 fps, and a pile of GL_INVALID_ENUM in glEnable(GL_DEPTH_CLAMP) errors
<HdkR> How surprising that they use centroid. Guess it is so if someone enables AA then things don't break?
<HdkR> supposedly centroid is supposed to require GL 4.0 or gpu_shader5. Looks like a game bug :P
<HdkR> oh, sample keyword requires that
<HdkR> centroid was available in glsl 120, dang :D
<HdkR> Time to implement centroid interpolation
NeuroScr has joined #panfrost
<icecream95> The sample keyword is also in ES 3.2
<HdkR> That it is
<tomeu> anarsoul: O3 makes a big difference when compared to O2?
<anarsoul> tomeu: I doubt we use -O2 with debug builds
<tomeu> anarsoul: debugoptimized uses O2
<anarsoul> oh
<anarsoul> then yeah
<tomeu> ah, so you were using debugoptimized and see a big difference when using O3?
<anarsoul> 3m20s vs 3m35s for functional.texture.* run
<anarsoul> I ran it several times and result is consistent
<icecream95> Is that still with assertions enabled?
<anarsoul> I believe you have a lima board (we all got one on XDC), you can try it yourself :)
<anarsoul> icecream95: yes
<icecream95> CFLAGS=-flto LDFLAGS=-flto=4 also helps performance by a few percent, but makes linking take around 5 minutes.
<tomeu> cool, will rebuild and profile STK again with O3
<anarsoul> tomeu: I was talking about tiling/untiling routines
<tomeu> yes, but I see some functions in panfrost itself that appear high in the profile without good reason
<anarsoul> oh, OK
<tomeu> don't want to waste time looking at those if it's a matter of the compiler not doing its job
<icecream95> Shadows in OpenMW: https://gitlab.freedesktop.org/snippets/833
austriancoder has joined #panfrost
vstehle has joined #panfrost
Green has joined #panfrost
robclark has joined #panfrost
steev has joined #panfrost
ric96 has joined #panfrost
narmstrong has joined #panfrost
lvrp16 has joined #panfrost
TheKit has quit [Ping timeout: 268 seconds]
ezequielg has joined #panfrost
daniels has joined #panfrost
jstultz has joined #panfrost
cowsay_ has joined #panfrost
cowsay has quit [Ping timeout: 260 seconds]
guillaume_g has joined #panfrost
robher has joined #panfrost
anarsoul|c has joined #panfrost
yann has quit [Ping timeout: 268 seconds]
<tomeu> bbrezillon robmur01 robher: is this one known? http://paste.debian.net/1126206/
<bbrezillon> tomeu: might be related to the problem fixed by https://patchwork.freedesktop.org/patch/348937/
<bbrezillon> tomeu: is this patch applied?
<bbrezillon> tomeu: I mean, applied on the tree you're testing
<bbrezillon> I know it's not applied to drm-fixes yet
<tomeu> bbrezillon: this was deqp with surfaceless, so I'm not sure there's any importing going on at all
<bbrezillon> tomeu: importing?
<bbrezillon> the fix is not about imported BOs
<bbrezillon> well, not only
<tomeu> oh, I see now that it also adds refcounting
<bbrezillon> alyssa, tomeu: I think I'll go back to an ad-hoc solution for wallpapering -> reloading ZS buffers makes things even more complicated than they already are
yann|work has joined #panfrost
marex-cloud has joined #panfrost
pH5 has joined #panfrost
stikonas has joined #panfrost
icecream95 has quit [Ping timeout: 268 seconds]
davidlt has quit [Remote host closed the connection]
raster has joined #panfrost
<tomeu> bbrezillon: drop using the blitter for wallpapering?
<tomeu> robmur01: I saw you are using a nanopc-t4 for testing, I undusted mine today and it's always hanging at an indeterminate point after starting to retry deferred probes
<bbrezillon> tomeu: yes
<tomeu> robmur01: it looks a bit like if the hw was shutting down due to thermal
<tomeu> bbrezillon: cool, curious about any performance impact
<bbrezillon> it's not optimal and forces us to add hacks all over the place to deal with tile reloading specificities
<tomeu> robmur01: if you aren't seeing this issue, what heatsink do you have? mine looks really slim
<bbrezillon> tomeu: had to add even more hacks from ZS reloading => https://gitlab.freedesktop.org/bbrezillon/mesa/commit/8b3da85bbd80513fee24b26272fa302a113f81df
<bbrezillon> *for
stikonas has quit [Ping timeout: 272 seconds]
davidlt has joined #panfrost
megi has joined #panfrost
anarsoul|c has quit [Quit: Connection closed for inactivity]
NeuroScr has quit [Quit: NeuroScr]
<bbrezillon> alyssa: while working on the ZS reloading stuff, I faced this issue => http://code.bulix.org/tze5sd-1091837
<bbrezillon> not sure my fix is correct though
warpme_ has joined #panfrost
kaspter has quit [Quit: kaspter]
NeuroScr has joined #panfrost
megi has quit [Ping timeout: 268 seconds]
alpernebbi has joined #panfrost
<alyssa> icecream95: GL_DEPTH_CLAMP shouldn't be too hard to implement, I think
<alyssa> icecream95: I have long suspected that clip_minz/clip_maxz in mali_viewport do depth clipping (is that different from depth clamping?)
<alyssa> clamping is probably lowerable anyway
<alyssa> Okay, next up is cleaning up my tiling stuff and merging it in with the compressed textures work so we can land both together without conflicts
megi has joined #panfrost
<robmur01> tomeu: I have the very original heatsink, that's even smaller than the one currently offered
<robmur01> I do have a 40mm fan on the PWM, but even without that I've never seen it get anywhere near thermal throttling just booting
<robmur01> I'd be more suspicious of firmware wonkiness - there's been a lot of that with RK3399, but upstream ATF + U-Boot is more or less OK as of quite recently
<robmur01> also: nice to see a use-after-free finally caught in the act, but DELAYED_BUS_FAULT is certainly a symptom I've never seen before :)
Xalius has joined #panfrost
<Xalius> has anyone tried Cube2?
<Xalius> 3D menus work fine, but with GL2.0 or 2.1 mode set it looks like the second picture, third is with llvmpipe
<Xalius> I get a lot of
<Xalius> [19701.717863] panfrost ff9a0000.gpu: js fault, js=0, status=INSTR_INVALID_ENC, head=0xcc00380, tail=0xcc00380
<Xalius> [19701.718804] panfrost ff9a0000.gpu: gpu sched timeout, js=0, config=0x3300, status=0x51, head=0xcc00380, tail=0xcc00380, sched_job=00000000e7dc7b53
<Xalius> in the log
davidlt has quit [Read error: Connection reset by peer]
davidlt has joined #panfrost
pak0st has joined #panfrost
pak0st has quit [Quit: Konversation terminated!]
anarsoul|c has joined #panfrost
tgall_foo has joined #panfrost
leinax has quit [Remote host closed the connection]
leinax has joined #panfrost
<tomeu> robmur01: it started working just fine after I mounted the board between the plastic sheets
<tomeu> maybe some bad contact that improved with the tension?
pH5 has quit [Quit: bye]
yann|work has quit [Ping timeout: 272 seconds]
<robmur01> tomeu: ha, a little alarming, but good that it works :)
<robmur01> it took many months until I got round to screwing the 'case' onto mine, and the board had definitely developed a slight curve from the heatsink mounting tension
guillaume_g has quit [Quit: Konversation terminated!]
TheKit has joined #panfrost
nerdboy has joined #panfrost
yann|work has joined #panfrost
<alyssa> Weee
Xalius has quit [Quit: Leaving]
<alyssa> Okay, I cherry-picked icecream95's awesome compressed textures stuff and added in support for tiling like in that series in the optimized tiling routine series
<alyssa> So that series is going to land with compressed texture support as well (since that requires mucking with tiling and better to go through the review cycle once at this point than have a rewrite shortly after)
<alyssa> Tests are passing which is fun
<alyssa> also implied adding bpp16 tiling support (for ASTC I guess?) which also means RGBA32 tiles again
<anarsoul> alyssa: you should probably squash some commits
<alyssa> anarsoul: absolutely
<alyssa> I did squash already and I plan to squash again
<alyssa> still WIP :)
stikonas has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
raster has joined #panfrost
<Lyude> alyssa: \o/
icecream95 has joined #panfrost
davidlt has quit [Ping timeout: 260 seconds]
raster has quit [Quit: Gettin' stinky!]
alpernebbi has quit [Quit: alpernebbi]
Stary has quit [Ping timeout: 268 seconds]
<icecream95> alyssa: Any ideas why MALI_MFBD_FORMAT_SRGB doesn't seem to have any affect?
nerdboy has quit [Ping timeout: 265 seconds]
Stary has joined #panfrost
Depau has joined #panfrost
Depau_ has joined #panfrost
Depau has quit [Ping timeout: 268 seconds]