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!
kaspter has quit [Ping timeout: 245 seconds]
kaspter has joined #lima
zombah_ has joined #lima
Net147_ has joined #lima
enunes has quit [*.net *.split]
megi has quit [*.net *.split]
Net147 has quit [*.net *.split]
bshah has quit [*.net *.split]
smaeul has quit [*.net *.split]
champagneg has quit [*.net *.split]
zombah has quit [*.net *.split]
bshah has joined #lima
smaeul has joined #lima
megi has joined #lima
Barada has joined #lima
Barada has quit [Client Quit]
chewitt has quit [Quit: Zzz..]
chewitt has joined #lima
monstr has joined #lima
kaspter has quit [Ping timeout: 260 seconds]
kaspter has joined #lima
Barada has joined #lima
kaspter has quit [Ping timeout: 264 seconds]
camus has joined #lima
camus is now known as kaspter
monstr has quit [Ping timeout: 265 seconds]
monstr has joined #lima
Barada has quit [Quit: Barada]
megi has quit [Quit: WeeChat 3.0.1]
megi has joined #lima
kaspter has quit [Ping timeout: 260 seconds]
camus has joined #lima
camus is now known as kaspter
kaspter has quit [Ping timeout: 245 seconds]
kaspter has joined #lima
chewitt has quit [Quit: Zzz..]
megi has quit [Quit: WeeChat 3.0.1]
megi has joined #lima
maciejjo has quit [Remote host closed the connection]
maciejjo has joined #lima
dev1990 has joined #lima
monstr has quit [Remote host closed the connection]
chewitt has joined #lima
<linkmauve> I’m still looking at how the filesystem cache could possibly be plugged in, and the blob apparently returns a MBS1 file on glGetProgramBinary(), is there still some information missing from there to build the final shader binary?
<linkmauve> Does the blob emulates the same formats as Lima?
<linkmauve> Or maybe, is this “emulation” only needed with desktop GL, and in the GLES case there is no additional swizzle needed? Is this about GL_EXT_bgra?
<anarsoul> linkmauve: blob doesn't emulate R and RG formats
<linkmauve> Ah, that’d be bad for e.g. Weston’s emulation of various YUV formats.
chewitt has quit [Read error: Connection reset by peer]
chewitt_ has joined #lima
<anarsoul> linkmauve: for FS cache you need to store sampler swizzles with the binary, you can use the binary only if all the swizzles match
<anarsoul> look at etnaviv disk cache for reference?
<linkmauve> Do you have a test program doing various swizzles so that I could compare stuff?
<linkmauve> Also, would it make sense to do binary patching instead of recompiling?
<anarsoul> recompile it
<anarsoul> see 77af1ca690f4f49e305c095d97b9af798bfea307 for the reference on how disk cache was implemented for etnaviv
<daniels> lack of R/RG support is absolutely the least painful thing about trying to use the blob
chewitt_ has quit [Read error: Connection reset by peer]
chewitt has joined #lima
<anarsoul> daniels: definitely, it's much easier to get mesa working
<linkmauve> I don’t even have access to the blob atm, IIRC only Linux 3.4 was available for my boards. ^^'
<linkmauve> My distribution definitely won’t support a kernel that ancient.
<anarsoul> there's sunxi-mali for allwinner boards
<linkmauve> Is there anything left to discover from this driver?
<anarsoul> who knows? :)
<linkmauve> ^^
chewitt has quit [Quit: Zzz..]
yann has quit [Ping timeout: 265 seconds]
yann has joined #lima
<linkmauve> Ok, I’m debugging the crash from yesterday (https://linkmauve.fr/files/gtk-lima.txt) and obviously there are not 818 slots in a ppir_op_undef (there are actually none).
<linkmauve> What does an undef op mean?
<anarsoul> can you try running it with NIR_PRINT=1 to see what we get from NIR?
<linkmauve> I’m adding a PPIR_INSTR_SLOT_END there, hopefully it’ll at least prevent the segfault.
<linkmauve> Sure.
<linkmauve> Wow, it spams a lot.
<linkmauve> With ssh compression it goes better.
<anarsoul> linkmauve: you're not supposed to get undef there if I'm not mistaken, so it's a bug in compiler
<linkmauve> Ok.
<linkmauve> Would it make sense to at least assert it’s not?
<anarsoul> can you share NIR_PRINT=1 output and the shader?
<linkmauve> The only two values I see in the tmux buffer (I can redirect to a file if you want it entirely) are:
<linkmauve> /* preds: */
<linkmauve> vec3 32 ssa_1 = undefined
<linkmauve> vec3 32 ssa_2 = undefined
<anarsoul> share complete output (just use pastebin)
<anarsoul> and fragment shader
<linkmauve> It doesn’t seem to happen every time, when it works it generates a 121 MiB file.
<anarsoul> :\
<anarsoul> with the same shader?
<linkmauve> With the same everything.
<linkmauve> Which might be why yesterday shader-db was running fine.
<anarsoul> well, it's important to pinpoint the shader that triggers it
<linkmauve> Yesterday it was blend.glsl.
<linkmauve> But I also noticed it was the very first shader being compiled.
<linkmauve> Oh, coredumpctl!
<linkmauve> From link_program() I can access shProg->Shaders[1]->Source, but the output is truncated. :(
<linkmauve> And it won’t give me the NIR_PRINT=1 anyway, nvm.
<anarsoul> :(
<linkmauve> Also, when it works, it still takes ~18s (on master) instead of ~6s (on 20.3.4) to launch gtk4-demo, so something clearly isn’t right.
<anarsoul> you can set MESA_SHADER_DUMP_PATH to some path to dump the shaders and run your program till it crashes. Make sure you clean the directory before each run
<linkmauve> I’ve tried a loop of shader-run, doesn’t seem to trigger.
<linkmauve> anarsoul, got one! https://linkmauve.fr/files/nir_print.txt
<linkmauve> Oh, that’s not transferred fully yet.
<linkmauve> There, done.
<anarsoul> wow, that's a huge shader
<linkmauve> It’s probably still blend.glsl, I’m trying to make sure of that.
<anarsoul> I don't see ssa1 or ssa2 being used anywhere, so it should be optimized out
<linkmauve> Yes, it is blend.glsl, in NO_CLIP mode (that is, the less expensive version).
adema- has joined #lima
kaspter has quit [Ping timeout: 276 seconds]
kaspter has joined #lima