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
cyrozap has joined #panfrost
stikonas has quit [Remote host closed the connection]
vstehle has quit [Ping timeout: 264 seconds]
kaspter has joined #panfrost
kaspter has quit [Excess Flood]
kaspter has joined #panfrost
AreaScout_ has quit [Ping timeout: 260 seconds]
davidlt has joined #panfrost
_whitelogger has joined #panfrost
davidlt has quit [Ping timeout: 260 seconds]
kaspter has quit [Ping timeout: 260 seconds]
kaspter has joined #panfrost
buzzmarshall has quit [Remote host closed the connection]
vstehle has joined #panfrost
nerdboy has quit [Ping timeout: 264 seconds]
davidlt has joined #panfrost
icecream95 has joined #panfrost
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 265 seconds]
camus1 is now known as kaspter
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 256 seconds]
camus1 is now known as kaspter
nerdboy has joined #panfrost
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 256 seconds]
camus1 is now known as kaspter
<tomeu> robmur01: echo 0x1ff > /sys/module/drm/parameters/debug will probably make the cause be printed on the console
raster has joined #panfrost
TheMojoMan has joined #panfrost
TheMojoMan has quit [Quit: Connection closed]
davidlt has quit [Remote host closed the connection]
davidlt has joined #panfrost
<icecream95> robmur01: kmscube wasn't working for me on 5.8, but after 348d374d013 it works. Are you sure you are using the right branch/remote for Mesa?
<robmur01> tomeu: derp, I should have thought of that - the one thing that stands out is "Invalid pixel format XR24 little-endian (0x34325258), modifier 0xffffffffffffffff"
<icecream95> robmur01: I get the same message for Mesa < 348d374d013
<tomeu> hrm, why would xrgb8888 be invalid...
<robmur01> icecream95: sigh, apparently the point at which I pulled yesterday was literally two commits before !6159 landed
<hanetzer> so, question. at any point during panfrost's development, was static analysis used at all?
<hanetzer> eg, objdump, ida, radare, ghidra
<robmur01> seems OK now
<HdkR> hanetzer: Nothing that contributed
<hanetzer> nifty.
<HdkR> It's a legal concern depending on region, so it shouldn't be done
<hanetzer> truthy statement, as far as anything regarding laws can be truthy
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 256 seconds]
camus1 is now known as kaspter
<robmur01> BTW, I did some more rvgl testing, and Profile=0,Shaders=1 with PAN_MESA_DEBUG=gl3 does indeed render correctly, but is the *only* combination that does so
<robmur01> without PAN_MESA_DEBUG it won't event start
<robmur01> Shaders=1,Profile=2 has a different visual bug
<robmur01> and anything with Shaders=0 still has the original glitches
<bbrezillon> tomeu: it's the modifier that's invalid
<tomeu> oh, ti has become invalid after the afbc work, right?
<bbrezillon> well, IIRC, 0 == linear, and 0xfffff... is pick the one you prefer
<robmur01> so whatever the fix was it doesn't seem to be applying to the fixed-function GL/ES corner of gallium
<tomeu> robmur01: fixed-function will hit some unrelated bug?
<icecream95> With Shaders=0 I'm hitting the assertion !zsa->alpha.enabled in panfrost_create_depth_stencil_state
<HdkR> womp womp
_whitelogger has joined #panfrost
kaspter has quit [Quit: kaspter]
<robmur01> FWIW that would also have been with yesterday's master at 371f6f42ad
<icecream95> alyssa: AFBC regresses performance in some situations for me - glmark2-es2 -b pulsar has gone from almost 1500 fps to 1150 fps
<urjaman> the difference in frame time with such big fps numbers isnt a lot tho
<icecream95> alyssa: AFBC also caused DarkPlaces to be unplayable - frame rates are in the single digits now
<icecream95> pan_js is using about 20% CPU, which seems a lot...
<HdkR> minmax calculation for indexed draws? :P
<icecream95> HdkR: These are kernel threads, so shouldn't be going anywhere near GPU job descriptors, but minmax calculation hasn't been a bottleneck for a while anyway
<tomeu> pan_js is the kernel thread that submits the jobs to the hw
<HdkR> ah
<tomeu> icecream95: do you have that in perf? if so, can you annotate and see what within it takes so long?
stikonas has joined #panfrost
<icecream95> tomeu: pan_js spends a lot of time in queue_delayed_work_on and cancel_delayed_work, and spinlock functions. drm_gem_object_put_unlocked is the top drm function that perf shows
<tomeu> ok, if there's spinlocks, then that's the problem
<tomeu> we shouldn't be spending any sensible amount of time with that
<tomeu> is it with a very specific workload?
<icecream95> tomeu: I've only seen it with DarkPlaces so far
<tomeu> oh, I guess what happens is that the GPU jobs take longer than the timeout
<tomeu> then we get into a cancel storm
<tomeu> we could increase the timeout to see if the storm doesn't happen, but I guess there's some problem with the GPU taking so long
<icecream95> tomeu: dmesg doesn't show anything about job timeouts
<tomeu> ok, then I guess we are using spinlocks to wait for BOs to be released by previous jobs?
<tomeu> sounds quite awful
robmur01 has quit [Read error: Connection reset by peer]
robmur01 has joined #panfrost
icecream95 has quit [Quit: leaving]
buzzmarshall has joined #panfrost
tomboy65 has quit [Ping timeout: 240 seconds]
<alyssa> "!zsa->alpha.enabled in panfrost_create_depth_stencil_state"
<alyssa> This definitely smells like a Gallium bug, we're not setting CAP_ALPHA_TEST
<alyssa> Hm. Not sure why AFBC would make things *slower* in practice ...
tomboy65 has joined #panfrost
<tomeu> alyssa: maybe there's some extraneous blits to linear?
<alyssa> tomeu: yeah, that's possible
indy has quit [Quit: ZNC - http://znc.sourceforge.net]
<alyssa> icecream95: I'm not able to reproduce a slowdown in -bpulsar with AFBC
<alyssa> indeed, under xwayland gnome-shell at least (buffer shared as tiled in both cases), seeing a 10% boost in fps
nerdboy has quit [Ping timeout: 265 seconds]
jolan has quit [Quit: leaving]
jolan has joined #panfrost
nerdboy has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
Elpaulo has joined #panfrost
<macc24> is ABFC on by default?
<alyssa> macc24: kinda
<alyssa> landed this week to master
<alyssa> on t760/t820/t860, it's used by default for textures and FBOs
<alyssa> but *not* for anything shared across processes (i.e. scanout, window back buffers, etc)
<HdkR> But it can be forced with PAN_MESA_DEBUG=afbc right? :)
<alyssa> correct
<alyssa> ymmv
<macc24> nice
<macc24> would be nicer if sway worked on my setup xD
<alyssa> oh, this is annoying
<alyssa> Midgard has attributes/images all indexed in the same table (see MALI_ATTR_IMAGE for how that works)
<alyssa> Gallium has it separated out
<alyssa> So someone has to budge
raster has joined #panfrost
<alyssa> I think the trick will be that we know at compile-time how many attributes a shader needs
<alyssa> So the descriptors can be assembled separately like that
<alyssa> what we *don't* know is if instancing is used
<alyssa> meh. probably no choice but draw-time, at least in part.
nerdboy has quit [Ping timeout: 265 seconds]
unoccupied has joined #panfrost
* alyssa trying to tease apart vertex code
davidlt has quit [Ping timeout: 265 seconds]
robmur01 has quit [Ping timeout: 256 seconds]
nerdboy has joined #panfrost
warpme_ has joined #panfrost
robmur01 has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
tlwoerner has quit [Quit: Leaving]
gcl_ has joined #panfrost