<urjaman>
i thought you had ascended to an another level of consiousness
<alyssa>
Ш вщтэе лтщц
<alyssa>
црфе умут
<Lyude>
i think she's on another plane of existence now, urjaman...
<alyssa>
האלפ
<alyssa>
okay got back to english somehow
<alyssa>
Some on-screen keyboard popped up reading "Latin"
<alyssa>
then something happened and it said "full-width" and then the above madness
<alyssa>
and then trying to get it back I somehow got stuck in... Cyrillic?
<alyssa>
but somehow the hebrew mode still worked?
<alyssa>
(but not english or spanish)
<alyssa>
and now it's back and I'm scared to reboot
<Lyude>
i wanna know how you got the full width mode because that would beｖｅｒｙ ｕｓｅｆｕｌ ｆｏｒ ｍｅ
<alyssa>
I don't know
<urjaman>
"U+FF0F : FULLWIDTH SOLIDUS
<urjaman>
" ... okay
<Lyude>
you can't see the characters?
<urjaman>
i can see them, but i didnt know they existed, so i wanted to find in what the hell of a range they are :P
<Lyude>
ahhhh
<Lyude>
yay, time for me to go home
<urjaman>
about time :P (joke about it being 3 in the morning here...)
<Lyude>
yeah this is an hour earlier then I usually leave! ;)
<Lyude>
i am trying to get my work schedule back to something that makes sense
<Lyude>
hopefully I can get myself to leave a little earlier each day until then
<anarsoul|2>
alyssa: btw, cyrillic letters on kbd don't correspond phonetically to latin
<anarsoul|2>
so you wrote some garbage :)
<hanetzer>
urjaman: I'll check that when I'm done with homework and such :)
<alyssa>
anarsoul|2: awesome :p
<urjaman>
(and also, i do and dont see them ... that is, they happen to work on the CrOS hterm i happened to be using (because my linux C201 cant do audio now, and who could live without music so ffs), but on the linux one i have set a font that doesnt have them (Terminus) so just boxes...)
<alyssa>
woop
<alyssa>
urjaman: I can confirm that life without music is bad :(
<alyssa>
urjaman: yeah it was the commit that hanetzer asked for :V
<alyssa>
reverted in latest master
* urjaman
tests
<urjaman>
"Bus error (core dumped)" okay :P
<urjaman>
i didnt even have the kernel module loaded ...
<urjaman>
(that was an alignment thing ... of some sort, it seems, based on dmesg... i'm guessing i should go turn some bits around to make this thing try to be 32-bit? :P)
<alyssa>
urjaman: Did it, er, used to work?
<urjaman>
alyssa: i mean, last time i managed to make it fail on a mmap, but back then i actually did things (instead of just build and run :P)
<alyssa>
meep
<hanetzer>
alyssa: ah, so that doesn't work then.
<urjaman>
alyssa: ah yeah this is new (blend mode ...)
<urjaman>
i did comment out #define BIT64 and #define T8XX and set #define T6XX ... in two places (sigh, could we atleast have those in one place? :P)
<urjaman>
just to see that it was indeed new if i was doing the same things as before
<urjaman>
i'll go and look if i can make sense of this ...
<alyssa>
urjaman: there shouldn't be two places to #define that but I might've messed up :p
<alyssa>
...oops, sorry
<urjaman>
in mali-job.h and some context.h
<alyssa>
Yeah, I saw. will fix in a min
<hanetzer>
so T6XX for the t760 ?
<alyssa>
hanetzer: Yeah, for now
<alyssa>
urjaman: Fixed
<hanetzer>
so, if I eglinfo/es2_info, should I expect to get the 'libEGL warning: DRI2: failed to authenticate' message? or is that still showing my setup to be busted?
<urjaman>
alyssa: the #define?
<alyssa>
urjaman: Yeah, it's down to one
<alyssa>
hanetzer: probably yeah
<urjaman>
or should i stop chasing this misaligned pointer? :P
<urjaman>
ah ok
<hanetzer>
alyssa: and if I run glmark2 I should get multiple windows spawning?
<alyssa>
yes
<hanetzer>
should I build glmark with any particular flavors?
<urjaman>
I suddenly wish the bit field feature didnt exist
<urjaman>
in C.
<urjaman>
what is the size of anything anymore anyways
<alyssa>
hanetzer: no?
<alyssa>
urjaman: bit fields are convenient
<urjaman>
alyssa: anyways, remove the packed on struct mali_blend_equation and it manages to run fine :P
<alyssa>
urjaman: ....hm
<alyssa>
oic
<urjaman>
somehow that struct gets put at non-4 alignment in the struct it's embedded into because it's packed ...
<alyssa>
Forgot that's not legal on 32-bit
<hanetzer>
still no more than a single window
<urjaman>
then it tries to store that float into weirdly aligned pointer...
<alyssa>
urjaman: Would it fix it to have the packed part (without the float) as it's own thing,
<alyssa>
mali_blend_equation_internal
<alyssa>
and then the mali_blend_equation has that and the float together, unpacked?
<alyssa>
(Well, packed but it's all 32 aligned so)
<urjaman>
i dunno
<urjaman>
and i dont know the interaction of packed and bit fields :P
<urjaman>
it was very mindfuck to try to read wtf this ends up really doing ...
<urjaman>
anyways now it stops at the kernel module ABI being wrong, so theoretically I could start helping Lyude with that
<alyssa>
urjaman: Is my code that bad? :P
<urjaman>
alyssa: the code was simple enough, the data structures not so much ;P
<alyssa>
urjaman: I mean, that's the "magic"
<urjaman>
but really it's just me never using bitfields (because on the things where i care about such packing, the code generated by using them makes you barf)
<alyssa>
All the hardware s3cr3t sauce is in that file :P
<urjaman>
(atleast used to, got used to making very certain of the data type sizes and the bits contained within...)
<hanetzer>
urjaman: so, is there something I can do to get the current mali_kbase.ko to work? eg, not the r27 one or whatever.
<urjaman>
hanetzer: last i tried there was something odd with the old mali_kbase giving a >4GB "physical" (or was it GPU space? ... the amount of MMUs here makes my head spin :P) addresses and the kernel bailing when the driver tries to mmap them
<urjaman>
so nope i dont have anything on hand that i would expect to work on a 32bit system (except maybe the mali things that are on the 3.14 kernel? but i dont wanna touch that, and so i didnt test that either)
<alyssa>
urjaman: yeah there are two many MMUs
<urjaman>
lol
<alyssa>
mali_ptr is for GPU address space
<alyssa>
uintptr_t is for CPU address space
<alyssa>
former is always 64-bit, latter depends on the CPU arch
<urjaman>
anyways, i dont see a reason why that specific structure needs the packed flag, except that apparently it allows messing up the alignment on it...
<urjaman>
it defines 32 bits of bits and a float. they should align okay as is
<hanetzer>
well, can you not use the aligned flag instead of packed? I'm pretty sure that is a thing
<urjaman>
anyhows, again 7 in the morning and i'n not asleep
<urjaman>
nn
hanetzer has quit [Quit: WeeChat 2.3]
_whitelogger has joined #panfrost
hanetzer has joined #panfrost
tomeu has joined #panfrost
afaerber has joined #panfrost
BenG83 has quit [Quit: Leaving]
pH5 has joined #panfrost
cwabbott_ has joined #panfrost
cwabbott has quit [Ping timeout: 245 seconds]
cwabbott_ is now known as cwabbott
<cwabbott>
alyssa: i totally forgot, but there are actually builtin functions in glsl that do exactly what the blend shader is doing wrt packing/unpacking things for 8-bit
<cwabbott>
pack/unpack4x8Unorm() or something like that
<cwabbott>
which means there are also nir opcodes for it
<cwabbott>
as well as pack/unpack4x8Snorm() for the signed normalized formats
belgin has joined #panfrost
<cwabbott>
err, packUnorm4x8() and unpackUnorm4x8()
<cwabbott>
and apparently it's in GLES 3.1 at least
cwabbott_ has joined #panfrost
cwabbott has quit [Ping timeout: 252 seconds]
cwabbott_ is now known as cwabbott
afaerber has quit [Quit: Leaving]
afaerber has joined #panfrost
belgin has quit [Quit: Leaving]
robclark has quit [Read error: Connection reset by peer]
robclark has joined #panfrost
<alyssa>
cwabbott: oh, neato
BenG83 has joined #panfrost
pH5 has quit [Quit: bye]
pH5 has joined #panfrost
afaerber has quit [Quit: Leaving]
anarsoul|2 has joined #panfrost
afaerber has joined #panfrost
cwabbott has quit [Read error: Connection reset by peer]