ChanServ changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - Logs https://oftc.irclog.whitequark.org/panfrost - <macc24> i have been here before it was popular
vstehle has quit [Ping timeout: 480 seconds]
hanetzer1 has joined #panfrost
Daanct12 has joined #panfrost
hanetzer has quit [Ping timeout: 480 seconds]
Daanct12 has quit [Ping timeout: 480 seconds]
camus1 has joined #panfrost
atler is now known as Guest48
atler has joined #panfrost
camus has quit [Ping timeout: 480 seconds]
Guest48 has quit [Ping timeout: 480 seconds]
Daanct12 has joined #panfrost
Daanct12 has quit [Read error: Connection reset by peer]
Daanct12 has joined #panfrost
Daanct12 has quit [Remote host closed the connection]
Daanct12 has joined #panfrost
camus has joined #panfrost
camus1 has quit [Read error: Connection reset by peer]
Daanct12 has quit [Remote host closed the connection]
Daanct12 has joined #panfrost
Daanct12 has quit [Remote host closed the connection]
Daanct12 has joined #panfrost
Daanct12 has quit [Remote host closed the connection]
Daanct12 has joined #panfrost
Daaanct12 has joined #panfrost
Daaanct12 has quit [Remote host closed the connection]
Daanct12 has quit [Remote host closed the connection]
Daanct12 has joined #panfrost
Daanct12 has quit [Remote host closed the connection]
Daanct12 has joined #panfrost
davidlt has joined #panfrost
Daanct12 has quit [Remote host closed the connection]
Daanct12 has joined #panfrost
Daanct12 has quit [Remote host closed the connection]
Daanct12 has joined #panfrost
Daanct12 has quit [Remote host closed the connection]
Daanct12 has joined #panfrost
Daanct12 has quit [Remote host closed the connection]
davidlt has quit [Ping timeout: 480 seconds]
vstehle has joined #panfrost
davidlt has joined #panfrost
Daanct12 has joined #panfrost
MoeIcenowy has quit [Quit: ZNC 1.7.2+deb3 - https://znc.in]
Daanct12 has quit [Read error: Connection reset by peer]
Daanct12 has joined #panfrost
tchebb has joined #panfrost
Daanct12 has quit [Remote host closed the connection]
Daanct12 has joined #panfrost
warpme_ has joined #panfrost
camus1 has joined #panfrost
camus has quit [Ping timeout: 480 seconds]
mixfix41 has joined #panfrost
mixfix41_ has quit [Ping timeout: 480 seconds]
Daanct12 has quit [Remote host closed the connection]
Daanct12 has joined #panfrost
Daanct12 has quit [Remote host closed the connection]
Daanct12 has joined #panfrost
Daanct12 has quit [Read error: Connection reset by peer]
davidlt has joined #panfrost
rasterman has joined #panfrost
camus has joined #panfrost
camus1 has quit [Ping timeout: 480 seconds]
<icecream95>
Fun fact: Midgard can calculate sines for values 64x times the maximum for Bifrost (0x400000τ vs 0x10000τ)
robmur01 has quit [Quit: Leaving]
robmur01 has joined #panfrost
rasterman has quit [Quit: Gettin' stinky!]
rasterman has joined #panfrost
camus1 has joined #panfrost
camus has quit [Ping timeout: 480 seconds]
<macc24>
icecream95: why would that even be a thing
<macc24>
just sin(val % tau)
nlhowell has quit [Ping timeout: 480 seconds]
<robmur01>
but surely that's extra work, and possibly an extra intermediate register, and possibly extra rounding error? :P
<macc24>
...rounding error?
pjakobsson has quit [Remote host closed the connection]
<robmur01>
as in successive FP operations inevitably lose precision due to the inexactness of the format - sure it's highly unlikely to be *significant* in that case, but it's still technically a thing if the sin() operation can have higher precision internally ;)
<macc24>
oh
* macc24
wishes for degree trig functions
<robmur01>
better have some BCD arithmetic while you're at it :P
<HdkR>
I vote for 80 bit floats
<macc24>
robmur01: some hacker is going to make risc-v soc with this arithmetic just for the sake of it
<robmur01>
but why bother when the overgrown 1970s calculator chip in everyone's current desktop/laptop can do BCD just fine!
<macc24>
oh can it?
<alyssa>
Trivia: The 6502 in the Nintendo Entertainment System omits BCD support for IP reasons. It is otherwise a regular 6502.
mriesch has joined #panfrost
<macc24>
isn't 6502 in nes some sharp chip or am i remembering it wrong?
<robmur01>
to be fair you do have to get out of 64-bit mode first, but AAA/AAS and DAA/DAS are still there
<alyssa>
macc24: Ricoh 2A03
<macc24>
alyssa: huh, at least i knew that it wasn't mos 6502 :p
<HdkR>
robmur01: To be fair, I just wrote code today to kick an x86-64 process in to 32-bit code space and used AAD to test if it worked :<
<macc24>
just use x86 far jump and jit it to aarch64
<HdkR>
I need to jump in to Jazelle
<alyssa>
HdkR: i'm not even a little ok with that code ;p
<HdkR>
:D
<HdkR>
bwehehehe. goood...goooood...
* macc24
is guilty of not knowing what's wrong with that code
<alyssa>
macc24: bitbanging random x86 registers with binary code (*not* even assembly code, honest to goodness .byte in an asm directive) sprinkled in
<macc24>
oh god
<alyssa>
it makes Q_rsqrt look downright reasonable
<HdkR>
clang was being a pain and not accepting the mnemonic :<
<robmur01>
HdkR: AFAIK a far jump is called "BR x16" :P
<HdkR>
lol
<HdkR>
We need a new BXJ of course :>
<alyssa>
Yo I heard you like Java, so we put Java in your cpu
<macc24>
no.
<HdkR>
You got it. a full 80 ops that map directly to java
<HdkR>
Surely we could backport FJCVTZS and help Java as well :thonk:
<macc24>
hmm
<macc24>
if it's going to make minecraft faster go ahead :p
<robmur01>
HdkR: FWIW the Morello ISA does add BX, so there's that ;)
<HdkR>
...What is this
<alyssa>
HdkR: Morello also adds 128-bit pointers so there's that ;)
<alyssa>
HdkR: s e c u r i t y
<alyssa>
Arm ISA except rewritten in Rust
<suihkulokki>
Jazelle, now that's an word I haven't heard in a long time
<HdkR>
Yea, I've got the ISA pulled up. trying to figure out wtf it is
<alyssa>
HdkR: Capabilities baked into every pointer to make attacks harder against broken C code
<HdkR>
Wacky
camus1 has quit []
Bozzer has joined #panfrost
Bozzer has quit [Remote host closed the connection]
megi has quit [Quit: WeeChat 3.2]
megi has joined #panfrost
<urja>
.... sigh... T760 is still spewing translation faults left and right -.- (and a couple of DATA_INVALID_FAULTs here too it looks like)
<alyssa>
urja: ruh roh :|
<alyssa>
workload? mesa version? anything interesting in PAN_MESA_DEBUG=trace? does T860 not have the same bugs?
thecycoone has quit []
<urja>
i dunno, i dont have T860 :P... workload is Xorg, XFCE, firefox and a terminal (that has transparent background tbh, but ...)
<urja>
hmm will check mesa (i did a git pull and compiled it), kernel is 5.12.14 (yeah maybe i should 5.13 but I wanted to use something with patchlevels as a stepping stone :P)
<urja>
oh
<urja>
old mesa lol
thecycoone has joined #panfrost
<urja>
main vs master
<alyssa>
Ah :p
<urja>
also I ran out of disk space... on my microSD card *blink* (it used to always be the eMMC :P)
<macc24>
alyssa: i have t860 and can test things
SomeNick643 has joined #panfrost
* alyssa
looks for a needle in a pile of needles
SomeNick643 has quit []
davidlt has quit [Ping timeout: 480 seconds]
<macc24>
alyssa: around how much faster does OoO scheduling make g31?
<macc24>
or any other bifrost
<alyssa>
meh
<macc24>
worth patching my local mesa tree for perf benefits?
<alyssa>
meh
<urja>
oh yeah updated mesa and it's way better
<urja>
(a while back but forgot to report about it)
<urja>
this thing is still a thing, but rarely: WARNING: CPU: 1 PID: 10203 at drivers/iommu/io-pgtable-arm.c:281
<macc24>
urja: uhh, isn't this not fault of panfrost?
<urja>
the backtrace is that it comes from panfrost_gem_open -> panfrost_mmu_map etc
<macc24>
hmm
<urja>
and then to the generic system
<alyssa>
robmur01: i see iommu
<alyssa>
i ping you and pretend i didn't see anything
<macc24>
urja: what kernel version?
<urja>
i remember mentioning this before and to my memory the response was "huh, that shouldnt be possible" :P
<robmur01>
it mostly manifests as a barrage of translation faults on process exit; occasionally an attempt to actually handle one of those faults crashes the pagetable code
<alyssa>
robmur01: this is the "ctrl-c during manhattan freezes the GPU and needs a reboot" bug? :p
<robmur01>
even more occasionally, things line up just right such that a fault looks like it growing a heap BO, so it tries to map the page as normal. Into the wrong AS, because the right one's already gone. Hilarity ensues...
<alyssa>
Oof
<robmur01>
alyssa: don't think I've ever seen it be that terminal, but I guess it depends on what job is running when it starts falling fall over
<robmur01>
kmscube in hardly the most involved workload :)
<robmur01>
*is
<alyssa>
Indeed
<robmur01>
and I guess for people whose GPUs clock at more than 40MHz, the jobs that run long enough to hit the race probably are mostly the gnarliest ones
warpme_ has quit [Quit: Connection closed for inactivity]