ChanServ changed the topic of #lima to: Development channel for open source lima driver for ARM Mali4** GPUs - Kernel has landed in mainline, userspace driver is part of mesa - Logs at https://people.freedesktop.org/~cbrill/dri-log/index.php?channel=lima and https://freenode.irclog.whitequark.org/lima - Contact ARM for binary driver support!
warpme_ has quit [Quit: Connection closed for inactivity]
<anarsoul> rellla: can you share your expected failres and skips for deqp?
<anarsoul> I'd like to re-enable CI for lima
<anarsoul> nevermind, found it
megi has quit [Ping timeout: 245 seconds]
archetech has joined #lima
smaeul has quit [Ping timeout: 250 seconds]
_whitelogger has joined #lima
_whitelogger has joined #lima
<anarsoul> narmstrong: so what happens on jan 6?
<narmstrong> anarsoul: everything goes back to normal !
<anarsoul> OK, what's wrong with the lab till then? :)
<rellla> anarsoul: will try your MR later. we probably should sort some of the skips out and restest them. or did you do that? maybe that can be done until jan 6th :p
<rellla> my skips list sounds a bit hacky (the latter part...)
<anarsoul> rellla: well, I reworked it a bit
<anarsoul> see MR
megi has joined #lima
megi has quit [Quit: WeeChat 2.6]
<narmstrong> anarsoul: we are doing some arrangements in the office and there will be regular power cuts
dddddd has quit [Remote host closed the connection]
smaeul has joined #lima
archetech has quit [Quit: Leaving]
archetech has joined #lima
abordado has joined #lima
abordado has quit [Ping timeout: 245 seconds]
archetech has quit [Quit: Leaving]
archetech has joined #lima
<rellla> fck. i grilled my cubieboards hdmi connector ...
<plaes> :(
<rellla> good to have a second one and just switch the sdcard. from now on, this one is headless :(
<plaes> how did you fry it?
<rellla> i had to put an eeprom in between the cable because the eeprom on the display was programmed with wrong edid. this setup was on a breadboard and poorly cable-soldered.
<rellla> i either adapted it in the wrong direction or i simply shortened sth, because i found a single cable afterwards wich wasn't soldered anymore...
<rellla> and all because of the mali blobs...
sphalerite has quit [Ping timeout: 246 seconds]
sphalerite has joined #lima
<plaes> you know that it's possible to push edid from userspace?
<rellla> anarsoul, enunes: yeah, i finally have deqp running with blobs! time to setup mali-syscall-tracker ...
<rellla> or should i do a full deqp run first :p
warpme_ has joined #lima
<rellla> yeah, it really works!! so now we can do a 1-to-1 stream comparison with lima and mali using deqp.
<rellla> time to check write.use.* again, as they fail with anarsouls MR ...
<enunes> rellla: sounds great! thanks for doing that
<enunes> useful to finally decide about the precision tests and others that might not be an issue with lima
<rellla> enunes: i will pimp mali-syscall-tracker later and do some tests.
<rellla> plaes: about edid, i could not flash edid onto the eeprom, because the edid/eeprom is kind of emulated with software/firmware
<rellla> i wasn't able to flash it, so i soldered one in between :/
<enunes> rellla: I'm not sure what the problem with that platform was, but you know you can force some fake edid with the kernel right?
<rellla> at least 2 years ago, when i soldered, i didn't :)
<rellla> maybe i should try that then :p
abordado has joined #lima
abordado has quit [Client Quit]
dddddd has joined #lima
cp- has quit [Quit: Disappeared in a puff of smoke]
cp- has joined #lima
cp- has quit [Quit: Disappeared in a puff of smoke]
cp- has joined #lima
<anarsoul> rellla: great!
<anarsoul> rellla: could you share your image later?
<anarsoul> I guess it should be easy enough to replace kernel...
niceplace has quit [Ping timeout: 258 seconds]
niceplaces has joined #lima
<rellla> anarsoul: i will try to upload the rootfs later
<rellla> or do you prefer a dd'ed image?
<anarsoul> rootfs is better
robertfoss has quit [Ping timeout: 258 seconds]
robertfoss has joined #lima
anarsoul|c has quit [Quit: Connection closed for inactivity]
Marex has quit [*.net *.split]
rellla has quit [*.net *.split]
rellla has joined #lima
BenG83 has joined #lima
Marex has joined #lima
BenG83 has quit [Quit: Leaving]
archetech has quit [Quit: Leaving]
adjtm has joined #lima
adjtm_ has quit [Read error: Connection reset by peer]
archetech has joined #lima
archetech has quit [Read error: Connection reset by peer]
archetech has joined #lima
archetech has quit [Ping timeout: 258 seconds]
<anarsoul> enunes: hey
<anarsoul> I'm not sure whether unrolling loops for GP is a good idea
<enunes> anarsoul: I'm totally ok with not doing that if that's what ends up being better
<enunes> didn't cause any regressions or any shader-db example to abort or hit the limit though
<anarsoul> fair enough
<anarsoul> I'd lower iterations limit though
<enunes> everybody seems to have copied those values from each other with some 'todo' parameter so far, hah
<anarsoul> :)
<anarsoul> OK, let's keep it at 32 and lower it if it fails
<anarsoul> what about const lowering bug?
<anarsoul> do you have a test case to reproduce it?
<enunes> I think it's a regression or bug from the lod bias patch
<enunes> we now have 2 ppir_src in ppir_load_texture_node
<enunes> and ppir_node_get_src doesnt handle the second one
<enunes> or something like that, didn't debug it completely yet
<anarsoul> can you share a shader that reproduces it?
<enunes> let me find one that reproduces with master
<enunes> we should be hitting it with piglit though
<enunes> rellla: do you have a recent piglit result available?
<rellla> enunes: no
<rellla> enunes: http://imkreisrum.de/piglit/mali450/ these are the last ones ...
<enunes> well shaders@glsl-fs-texture2d-bias reproduces it
<enunes> hmm I think the last one there from november still didnt have it
<anarsoul> thanks
<enunes> I plan to look into it more though
<anarsoul> OK, I'll leave it up to you then :)
<rellla> anarsoul: i think i got the write.use.* issue fixed with a oneliner ...
<anarsoul> great!
<anarsoul> I love oneliner fixes :)
<anarsoul> it usually takes few days to figure out what line it should be though :)
<anarsoul> rellla: so what line is it?
<rellla> fixes it again upon your drop-allocator patches
<rellla> this is the place, where the blob locates it
<rellla> here are the dumps with working mali and failing lima without that patch. test case write.use.index_array.array
<rellla> anarsoul: not sure about the draw_arrays case though. i will verify and do some more investigations tomorrow.
<anarsoul> sounds good
<anarsoul> thanks a lot for looking into it!
<rellla> mali dumps for deqp sound useful :)
<anarsoul> definitely
<anarsoul> lima-drop-suballocator has merged
<rellla> nice
<rellla> anarsoul: does marge bot automatically add rbs?
<anarsoul> nope :(
<anarsoul> you have to collect all the tags
<rellla> then you have been fast :)
<anarsoul> I've got an email that you pushed new commits
<anarsoul> check them - tag is here, assigned marge bot to this MR
<anarsoul> that's it
<anarsoul> literally took ~20 seconds :)
<rellla> fine. bedtime here :p
<anarsoul> good night :)
<anarsoul> rellla: I have strong suspicion that your fix may also fix lots of other tests that currently fail in deqp