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
<icecream95> dual-source blending for logicOp would be useful for masking. glLogicOp(GL_OR_XOR) maybe?
<HdkR> Blend shaders being user programmable would be fun as well though :P
robink has quit [Ping timeout: 260 seconds]
robink has joined #panfrost
<icecream95> Someone needs to make an animated glxgears kart for SuperTuxKart
<nhp> SuperMesaKart
<nhp> including such characters as glxgears, vkcube and long glitched triangle
<alyssa> nhp: I feel called out.
<alyssa> :p
<alyssa> HdkR: The problem with programmable blend shaders on mali is that work registers are shared between frag and blend
<alyssa> For a single RT, that's a minor annoyance but doesn't matter; for MRT, that severely restricts what you can do.
<alyssa> OTOH extended blend as it is doesn't work with MRT IIRC, so maybe you could say blend shader sonly with 1 RT
<HdkR> Sounds fun
stikonas_ has quit [Remote host closed the connection]
yann has quit [Ping timeout: 265 seconds]
vstehle has quit [Ping timeout: 272 seconds]
robmur01 has joined #panfrost
robmur01 has quit [Ping timeout: 256 seconds]
nerdboy has quit [Ping timeout: 272 seconds]
davidlt has joined #panfrost
cwabbott_ has joined #panfrost
cwabbott has quit [Ping timeout: 260 seconds]
cwabbott_ is now known as cwabbott
<icecream95> alyssa: Half of the logicOp modes got broken by an earlier commit. I'm bisecting now
nerdboy has joined #panfrost
buzzmarshall has quit [Remote host closed the connection]
vstehle has joined #panfrost
nerdboy has quit [Ping timeout: 256 seconds]
<icecream95> alyssa: Piglit shader_test file for the bug: https://gitlab.freedesktop.org/snippets/1043
<HdkR> `1.-y` Whoa jeez, that's an accepted swizzle? I hate it
_mixfix41_ has joined #panfrost
nlhowell has joined #panfrost
Elpaulo has joined #panfrost
<daniels> icecream95: yeah I think we really definitely want replayable traces rather than trying to start real-life apps - far too many moving parts in the latter, and I would also say almost certainly too slow
yann has joined #panfrost
raster has joined #panfrost
unoccupied has quit [Quit: WeeChat 2.5]
icecream95 has quit [Quit: Lost terminal]
_mixfix41_ has quit [Ping timeout: 272 seconds]
stikonas has joined #panfrost
mixfix41 has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
raster has joined #panfrost
raster has quit [Client Quit]
raster has joined #panfrost
robmur01 has joined #panfrost
robmur01_ has joined #panfrost
robmur01 has quit [Ping timeout: 265 seconds]
robmur01_ is now known as robmur01
macc24 has joined #panfrost
HdkR has quit [Ping timeout: 258 seconds]
kinkinkijkin has quit [Ping timeout: 258 seconds]
guillaume_g has joined #panfrost
stikonas has quit [Ping timeout: 246 seconds]
stikonas has joined #panfrost
plaes has joined #panfrost
macc24 has quit [Ping timeout: 272 seconds]
macc24 has joined #panfrost
<alyssa> HdkR: Pretty sure that parses as `1.0 - y`
macc24 has quit [Quit: WeeChat 2.8]
<alyssa> and.. yeah, I see the bug, fixing
raster has quit [Quit: Gettin' stinky!]
raster has joined #panfrost
robmur01 has quit [Ping timeout: 265 seconds]
macc24 has joined #panfrost
guillaume_g has quit [Quit: Konversation terminated!]
<alyssa> nya
<alyssa> MR sent
nerdboy has joined #panfrost
macc24_ has joined #panfrost
macc24 has quit [Ping timeout: 265 seconds]
macc24_ has quit [Quit: WeeChat 2.8]
macc24 has joined #panfrost
kinkinkijkin has joined #panfrost
<alyssa> flags_lo 0x200 I'm really only seeing for compute jobs
<alyssa> or for fragment precompute which is a TILER job technically(?) but de facto compute
yann|work has joined #panfrost
<alyssa> by being clever! check the video!
<alyssa> I also see it with SSBOs!
<alyssa> which the TILER precompute does
<alyssa> so it's a "can write global memory?" deal, eh?
<alyssa> which explains why setting it needlessly on gles2 doesn't cause bugs but costs some performance
yann has quit [Ping timeout: 260 seconds]
<alyssa> nya!
<alyssa> 0x200 set (and EARLY_Z cleared) exactly when we write to global memory
<alyssa> reading is insufficient to trigger that
<alyssa> nya..
<alyssa> taking a shader with SSBO write and declaring layout(early_fragment_tests) in;
<alyssa> gets no changes in the cmdstream except EARLY_Z is now set again
<alyssa> so that confirms MALI_EARLY_Z works exactly as we thought :D
nerdboy has quit [Ping timeout: 256 seconds]
raster has quit [Quit: Gettin' stinky!]
raster has joined #panfrost
yann|work has quit [Read error: Connection reset by peer]
yann|work has joined #panfrost
<alyssa> alyssa@kevin:/dev/shm$ grep -r varying_ pandecode.dump.0008 | grep '\[' | grep = | sort | uniq | sed -e 's/.*\[//' | sed -e 's/\].*//' | paste -sd+ | bc
<alyssa> I feel sad that just came off my keyboard.
<alyssa> 74746720
<alyssa> So a single frame is chewing through 71MB of varyings (?)
<alyssa> so 71MB of write and 71MB of read
<alyssa> We could maybe get a real win from 16-bit varyings then.
<alyssa> if not for the bw, to reduce cache pressure, I hope
davidlt has quit [Ping timeout: 265 seconds]
nerdboy has joined #panfrost
macc24 has quit [Quit: WeeChat 2.8]
<alyssa> Likewise we're using a lot of attributes here, if I clamp that (render breaks but) fps jumps
<alyssa> OTOH we know from counters that the load/store pipe itself isn't doing much
<alyssa> so all of this points, I think, to thrashing the cache.
<alyssa> fp16 varyings are a minor win on geometry heavy stuff, I think
<alyssa> -bshading:model=bunny goes from 159fps->167fps
<alyssa> (Gouraud shading, not Phong. 200k vertices on the bunny)
stikonas has quit [Ping timeout: 272 seconds]
<plaes> heya! when trying to run KiCAD on machine with Mali T-860, it complains about not having enough OpenGL framebuffers (requiring 3)
<plaes> is this a hardware limitation, or just something that is not yet supported?
<alyssa> urjaman: reported this, still trying to figure out the proper fix
<alyssa> if you enable gles3 that error goes away but then something else in KiCAD breaks, so... :|
<urjaman> so yeah known thing
<plaes> cool, thanks
<urjaman> (if you need to look at a kicad thing rn, LIBGL_ALWAYS_SOFTWARE=1 kicad # will give you the llvmpipe software render you got before panfrost)
<urjaman> just incase you didnt know
<alyssa> unless you're like me and build w/o sw rasterizers
<alyssa> live on the edge? :P
<plaes> yeah, fortunately KiCAD is intelligent enough to fall back to software-rendering itself :)
macc24 has joined #panfrost
<urjaman> yeah for me (for whatever reason) the cairo softrender crashes
<alyssa> if $pandemic didn't make hw procurement complicated, I might be dogfooding bifrost. if it runs weston, it's useable, right?! :p
<urjaman> (and before when it still worked it was slower than llvmpipe, so i've just used llvmpipe for now)
<alyssa> urjaman: interesting... I sort of expect llvmpipe to be slower than a dedicated non-GL sw renderer since the GL overhead is nontrivial for sw
<alyssa> otoh llvmpipe's hot paths are probably much more efficient than cairo's sw
<alyssa> (but still slower than, say, a hand-optimized NEON routine vs llvmpipe's version from a GL shader? I dunno)
stikonas has joined #panfrost
<urjaman> yeah i had kinda assumed it was because 1. llvmpipe not badly optimized and 2. llvmpipe uses all of the cores (but i dunno if cairo can or cant do that)
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
<plaes> yeah, can confirm that PAN_MESA_DEBUG=gles3 has some issues with kicad..
<alyssa> plaes: :w
stikonas has quit [Read error: Connection reset by peer]
stikonas has joined #panfrost
stikonas has quit [Ping timeout: 246 seconds]
stikonas has joined #panfrost
stikonas has quit [Read error: Connection reset by peer]
stikonas has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
robmur01 has joined #panfrost
robmur01 has quit [Ping timeout: 240 seconds]
<anarsoul> alyssa: any plans to remove chromium from blacklist?
<alyssa> anarsoul: That will be the commit after the one that fixes chromium..
<anarsoul> it worked fine for me when I tested it this weekend, the only issue is BUGs in dmesg due to evicted pages
<anarsoul> but no rendering issues so far
<alyssa> anarsoul: with desktop GL or with GLES?
<alyssa> also, did you test WebGL?
<alyssa> I know at some point chromium gles+webgl probably worked, but I don't think desktop GL (default here) has.
<anarsoul> alyssa: I think it was with desktop GL, I didn't try webgl demos
<alyssa> hm ok
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
macc24 has quit [Ping timeout: 260 seconds]
macc24_ has joined #panfrost
macc24_ is now known as macc24
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
buzzmarshall has joined #panfrost
macc24 has quit [Quit: WeeChat 2.8]
robmur01 has joined #panfrost
macc24 has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
robmur01 has quit [Ping timeout: 258 seconds]
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
macc24 has quit [Quit: WeeChat 2.8]
macc24 has joined #panfrost
raster has joined #panfrost
NeuroScr has joined #panfrost
macc24 has quit [Ping timeout: 265 seconds]
NeuroScr has quit [Ping timeout: 258 seconds]
stikonas has quit [Ping timeout: 246 seconds]
NeuroScr has joined #panfrost
stikonas has joined #panfrost
stikonas has quit [Ping timeout: 246 seconds]
stikonas has joined #panfrost
stikonas has quit [Excess Flood]
jernej_ has joined #panfrost
stikonas has joined #panfrost
jernej has quit [Read error: Connection reset by peer]
jernej_ is now known as jernej
stikonas has quit [Ping timeout: 246 seconds]
stikonas has joined #panfrost