alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - https://gitlab.freedesktop.org/panfrost - Logs https://freenode.irclog.whitequark.org/panfrost - <daniels> avoiding X is a huge feature
NeuroScr has joined #panfrost
<alyssa> Core transform feedback is working now. Still have to implement GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN before I can start passing any deqp tests
<alyssa> I think that's something i just keep track of myself in the driver since it's statically calculable (we don't spport geom shaders... yet... )
megi has quit [Ping timeout: 245 seconds]
vstehle has quit [Ping timeout: 272 seconds]
Elpaulo has quit [Read error: Connection reset by peer]
Elpaulo has joined #panfrost
<HdkR> Whoa, transform feedback, how fun
NeuroScr has quit [Quit: NeuroScr]
_whitelogger has joined #panfrost
vstehle has joined #panfrost
davidlt has joined #panfrost
pH5 has joined #panfrost
davidlt has quit [Ping timeout: 246 seconds]
raster has joined #panfrost
davidlt has joined #panfrost
davidlt has quit [Remote host closed the connection]
pH5 has quit [Quit: bye]
Kwiboo has quit [Ping timeout: 272 seconds]
megi has joined #panfrost
Kwiboo has joined #panfrost
Kwiboo has quit [Ping timeout: 248 seconds]
Kwiboo has joined #panfrost
davidlt has joined #panfrost
davidlt has quit [Remote host closed the connection]
davidlt has joined #panfrost
pH5 has joined #panfrost
davidlt has quit [Remote host closed the connection]
davidlt has joined #panfrost
adjtm_ has quit [Ping timeout: 248 seconds]
davidlt has quit [Ping timeout: 268 seconds]
_whitelogger has joined #panfrost
davidlt has joined #panfrost
davidlt has quit [Ping timeout: 246 seconds]
adjtm_ has joined #panfrost
raster has quit [Remote host closed the connection]
davidlt has joined #panfrost
davidlt has quit [Remote host closed the connection]
davidlt has joined #panfrost
davidlt has quit [Ping timeout: 272 seconds]
raster has joined #panfrost
davidlt has joined #panfrost
adjtm_ has quit [Ping timeout: 246 seconds]
davidlt has quit [Ping timeout: 248 seconds]
raster has quit [Remote host closed the connection]
raster has joined #panfrost
davidlt has joined #panfrost
davidlt has quit [Ping timeout: 258 seconds]
adjtm_ has joined #panfrost
davidlt has joined #panfrost
davidlt has quit [Read error: Connection reset by peer]
davidlt has joined #panfrost
davidlt has quit [Read error: Connection reset by peer]
davidlt has joined #panfrost
<alyssa> So, got XFB sort of working with 1 draw
<alyssa> Multiple draws and it's no happy
<alyssa> Guess I'm missing an offset.. somewhere...
<alyssa> ACtually I guess _I_ need to provide the offset
<alyssa> Ok, yeah, meh ok
<alyssa> Just passed my first XFB deqp test
<alyssa> Got some issues with packing vec2
pH5 has quit [Quit: bye]
davidlt has quit [Ping timeout: 245 seconds]
davidlt has joined #panfrost
JaceAlvejetti has joined #panfrost
JaceAlvejetti has quit [Client Quit]
JaceAlvejetti has joined #panfrost
JaceAlvejetti has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
<alyssa> Fixed those but uh
<alyssa> 2 steps forward, 1 step back
<alyssa> (I just "fixed" a bug that regressed all of the once-working XFB tests... that's how it goes smoetimes)
<raster> XFB?
<raster> u making x work?
<alyssa> raster: transform feedback
<raster> oh
<raster> btw
<alyssa> There's no hw support for it in Midgard, and it's kind of a checkbox feature at this point but I mean
<alyssa> I like anything trans* \shrug/
<raster> thanks for the perf discussion - i found some royal stupidity in efl/e that was proabbly huritng us
<alyssa> Nice :)
<raster> we created 91 textures then deleted them... per frame we rendered
<alyssa> O_O
<raster> with just a glmark window going
<raster> yes...
<alyssa> I don't suppose you're seeing a perf boost with other drivers too then?
<raster> well just fixed it and teste don intel
<raster> got about 8% up on there
<raster> still behind weston
<raster> i'm not going ot see what it does on panfrost
<raster> my guess is it'll have a much bigger impact.
<alyssa> :)
<raster> we still have blend mode enabled tho thanks to other optimizations that kidn of assumed this use case was for small regions
<raster> so it tries not to switch blend mode on/off so it can merge triangle "meshes" into single drawarrays better
<raster> makes sense if you are drwoing 200 buttons where the edges are "alpha" like a bevel and shadow
<raster> but the middle bit is solid
<raster> makes less sense for big fat window surfaces where the middle bit is solid with a surrounding CSd frame
JaceAlvejetti has joined #panfrost
<raster> (it just so happens glmark's frame is nil/empty)
<alyssa> Well, glad I can help stuff run better on E.. i know it can be a hard transition...
<raster> that nil frame also triggered another "slow path" where we drew the wallpaper below as we didnt "cut out" that opaque region becuase the borders were all 0's
<raster> and thus we went "oh all 0's - turn this feature off"
<raster> :)
JaceAlvejetti has quit [Client Quit]
JaceAlvejetti has joined #panfrost
<raster> and well that didnt help at all...
<alyssa> 0 my
NeuroScr has joined #panfrost
<raster> ... panfrost slowed down too
<raster> so... wtf?
<raster> i was going much faster yesterday ... i havent changed kernel or mesa... ugh.
<raster> yeah
<raster> 322 today
<raster> got 458 yesterday...
<raster> oooh i think i see where the perf diff will be
<raster> westonis using efl syncs
<raster> we're not
stikonas has joined #panfrost
adjtm_ has quit [Ping timeout: 268 seconds]
raster has quit [Remote host closed the connection]
raster has joined #panfrost
<daniels> we only use them for logging, they're not adding any additional sync
<daniels> it's all just implicit sync still
<alyssa> My code quality is decreasing by the second
<alyssa> Why won't you work dangit
adjtm_ has joined #panfrost
* alyssa got it
<alyssa> Well, next up is XFB on anything besides points...
raster has quit [Remote host closed the connection]
NeuroScr has quit [Quit: NeuroScr]
NeuroScr has joined #panfrost
<alyssa> Passed: 378/378 (100.0%)
<alyssa> ^^ For transform_feedback.basic_types.* :D
<alyssa> It... took me a disturbingly long time but whatever, it's working now
<HdkR> Nice
<anarsoul> congrats :)
<alyssa> HdkR: Granted there are like 700 other xfb tests in GLES3 alone
<alyssa> 3.0
<alyssa> anarsoul: You gonna add xfb to Utgard? :^)
<alyssa> *lima
<anarsoul> isn't it 3.0 thing?
stikonas has quit [Remote host closed the connection]
<alyssa> anarsoul: Yes, and?
<anarsoul> Utgard is GLES2.0 GPU
<alyssa> Y'all could probabbly support ES3.0 if you wanted to
<alyssa> anarsoul: So is Midgard, tbqh
<alyssa> ;P
<anarsoul> alyssa: I wish I had enough time to get cf support wired into our ppir compiler
<alyssa> Yes, higher prio that
<alyssa> transform feedback: Midgard doesn't have hw support
<alyssa> UBOs: I would assume Utgard can do them since generic I/O, or..?
<alyssa> occlusion queries: there's an ES2 extension, maybe utg implements
<anarsoul> I don't think Utgard can do UBOs
<alyssa> instanced: easy enough to lower, Midgard has to on some models because errata
<HdkR> precise occlusion queries?
<alyssa> MRT: a royal pain in the butt to emulate, but Midgard before T760 didn't have support and they did it anyway
<HdkR> There was never a public ES extension for precise occlusion queries. Nvidia is the only one to implement it in ES :P
<alyssa> etc
<HdkR> (Nvidia has a spec page internally that was never ratified as a real spec)
<alyssa> array.* also passing 100%
<alyssa> array_element.* is a mix
<anarsoul> alyssa: oh, also vertex shader can't have more than 512 instructions :)
<alyssa> anarsoul: ...Okay, that'll be the real limiting factor here... :)
<anarsoul> and it's scalar (no vec4 ops)
<alyssa> So more like 128 instruction limit?
<alyssa> XFB of gl_Position is legal?
<alyssa> Bleh
* alyssa rewrites more code
<HdkR> Just use a point per pixel on the screen and do XFB on it
<alyssa> APDIUHYF
<alyssa> Just relaized the problem
<alyssa> We're capturing the transformed output..
raster has joined #panfrost
raster has quit [Remote host closed the connection]
raster has joined #panfrost