stikonas_ has quit [Remote host closed the connection]
chrisf_ is now known as chrisf
<
HdkR>
gtucker: Oh wow, comfortable enough to enable by default now?
vstehle has quit [Ping timeout: 244 seconds]
sravn has quit [Ping timeout: 258 seconds]
sravn has joined #panfrost
buzzmarshall has quit [Remote host closed the connection]
sravn has quit [Ping timeout: 248 seconds]
sravn has joined #panfrost
davidlt has joined #panfrost
vstehle has joined #panfrost
davidlt_ has joined #panfrost
davidlt has quit [Ping timeout: 252 seconds]
pH5 has joined #panfrost
yann has quit [Ping timeout: 248 seconds]
davidlt_ is now known as davidlt
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
guillaume_g has joined #panfrost
<
HdkR>
Nice, another Bifrost board :)
yann has joined #panfrost
<
narmstrong>
gtucker: hmm, but olf already applied tomeu's and my patches, should already be enabled in linux-next and Ci
griffinp has joined #panfrost
<
gtucker>
narmstrong: yes I've found it in olf's soc tree
<
gtucker>
but it's not in -next yet, because I think the branch need to be pulled into arm-soc or another tree first
<
gtucker>
so in the meantime we have these configs enabled on staging.kernelci.org to ramp things up
<
shadeslayer[m]>
tomeu: Do I enable --debug then so that we hit the asserts in the CI so we can fix them?
afaerber has quit [Quit: Leaving]
afaerber has joined #panfrost
<
tomeu>
shadeslayer: no, we need for the CI to stay always green
<
tomeu>
you can run locally with the debug build type and fix those, no need to break Ci for that :)
<
shadeslayer[m]>
<tomeu "you can run locally with the deb"> Fair enough now that I have a board :D
<
shadeslayer[m]>
that patch series with 20 patches q.q
BenG83 has joined #panfrost
herbmilleriw has quit [Remote host closed the connection]
herbmilleriw has joined #panfrost
<
alyssa>
Yeah, sorry, I'm working on it
<
alyssa>
CI is upset with me
yann has quit [Ping timeout: 248 seconds]
_whitelogger has joined #panfrost
stikonas has joined #panfrost
stikonas has quit [Read error: Connection reset by peer]
stikonas_ has joined #panfrost
pH5 has quit [Quit: bye]
<
alyssa>
anarsoul: I reworked the tiling code
<
anarsoul>
alyssa: did you make it better?
<
alyssa>
Got rid of the "space filler" magic numbers entirely -- they're wholly unnecessary once you understand how the tiling actually works
<
alyssa>
I might've broken loading; waiting for CI :p
<
anarsoul>
alyssa: make sure you write down your findings
<
alyssa>
anarsoul: Hang on
<
anarsoul>
alyssa: btw, have you send your proposal for xdc?
<
anarsoul>
s/send/sent
<
anarsoul>
cool, I think I'm coming this year
<
anarsoul>
alyssa: Nice write up. Btw, I have strong suspicion that LUT for x will be faster than calculating it (16 bytes only)
<
alyssa>
anarsoul: Even though it's only 2 ALU ops?
<
alyssa>
Actually I think you're still right,lemme test
<
anarsoul>
alyssa: I'm talkin about space_4()
<
alyssa>
anarsoul: Hit refresh; space_4 isn't in the critical path
<
alyssa>
It's called at most once and not always
<
anarsoul>
yet I think it makes sense to use LUT for x for all the cases
<
alyssa>
yeah, I'm trying that now
<
alyssa>
anarsoul: Waow, that's way faster :D
<
alyssa>
..ONly because I did it wrong
<
anarsoul>
it shouldn't be
*way* faster or slower
<
alyssa>
anarsoul: Performance is suspiciously indentical
<
anarsoul>
alyssa: then don't use smart tricks
<
alyssa>
No, the clever thing is that that's tile invariant so I can pull this out and.. aa I bet this will be good
<
alyssa>
Perf is still suspiciously close
<
anarsoul>
please keep it simple :)
<
alyssa>
NEON doesn't do scatter writes, does it..
somy has joined #panfrost
<
alyssa>
anarsoul: Just threw the code onto the list, if you could maybe review?
<
anarsoul>
I'm not on the list
<
anarsoul>
did you CC me?
<
anarsoul>
OK, got it
yann has joined #panfrost
<
anarsoul>
alyssa: what's the point in splitting tiling/untiling into aligned and unaligned parts?
stikonas_ has quit [Read error: Connection reset by peer]
stikonas has joined #panfrost
<
alyssa>
anarsoul: The aligned part is able to be optimized a lot more
<
alyssa>
Since we can work at the tile level, rather than the pixel level (and therefore move a lot of calculation out of the inner loop)
<
alyssa>
Maybe we should just use the generic/slow path though
<
alyssa>
Yeah, let's do that
davidlt has quit [Ping timeout: 245 seconds]
BenG83 has quit [Quit: Leaving]
robert_ancell has joined #panfrost
afaerber has quit [Quit: Leaving]
Elpaulo has quit [Ping timeout: 258 seconds]
Elpaulo has joined #panfrost
Kwiboo has quit [Ping timeout: 244 seconds]
BenG83 has joined #panfrost
Kwiboo has joined #panfrost
BenG83 has quit [Quit: Leaving]
<
alyssa>
Mipmapping series (30 patches!) pushed :)
_whitelogger has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]