<robmur01>
based on trying G71 fairly recently i'mma say: yes
<macc24>
please let me have hope
<robmur01>
IIRC we're missing significant chunks of the Bifrost v6 ISA still
<macc24>
g31 works and it's newer soooo
<alyssa>
robmur01: Define significant.
<macc24>
kmscube works
<macc24>
before crashing
<robmur01>
newer = better: Bifrost support is pretty much starting at the end and working backwards ;)
<alyssa>
robmur01: tbf midgard kicked off on t760 and the later t720 code was a backport
<alyssa>
(mostly. Actually I was looking at t628 blobs and using awful t760 compat hw but I digress)
<robmur01>
I never meant to imply that starting on mature(er) hardware from once most of the bugs and quirks have been ironed out isn't massively sensible ;)
<macc24>
robmur01: panfrost started with t760 probably because asus c201pa has t760
<robmur01>
G71 I think was falling over somewhere in the compiler due to some entire class of instruction not being implemented
<macc24>
ugh
<alyssa>
robmur01: G71 omits support for fp32 transcendentals.
<alyssa>
Whether fp32 transcendentals are supported is implementation-defined, not architecture-defined.
<alyssa>
In practice everything except G71 supports them, because not supporting them is really silly.
<alyssa>
Upshot is that G72 is just a debug problem now, while G71 needs actual compiler stuff
<macc24>
poor G71
<macc24>
"just" a debug problem?
<alyssa>
Yes?
<macc24>
h o w
<alyssa>
?
<macc24>
how to fix it
<alyssa>
1. Email sales@collabora.com
<alyssa>
;P
<alyssa>
/s
<alyssa>
It's on the todo list, just low priority.
<macc24>
ok, how to fix it since all low priority stuff never gets fixed?
<robmur01>
oh, neat, seems we don't actually fall over in the compiler any more, just give the GPU some instructions it's too stupid to understand and let it choke. That's... a bit better, at least
* HdkR
looks at personal low priority
<HdkR>
Ugh, it's all filled with garbage
<robmur01>
BTW, bump:bump-render=height is another glmark2 scene where fp16 seems to go a bit wonky, even without the inf/nan patch
* robmur01
should probably fire up something other than T620 to double-check...
<robmur01>
I should have gone to bed at least 2 hours ago, but no, let's reflash FPGAs right now!
<robmur01>
stupid brain...
<icecream95>
So ThreadSanitizer is 64-bit only? Looks like I'll have to debug u_threaded_context the hard way…
<HdkR>
clang tsan?
<HdkR>
Could use valgrind instead
<alyssa>
icecream95: ...Do you need an arm64 machine?
<HdkR>
icecream95: My offer still stands
camus1 has joined #panfrost
kaspter has quit [Remote host closed the connection]
camus1 is now known as kaspter
<icecream95>
panfrost_create_sampler_view can be called from both threads at once, with no way of telling which it came from, and uses thread-unsafe rzalloc
<icecream95>
There are also races in panfrost_minmax_cache_invalidate, when a buffer is mapped and unmapped in two threads at once
<alyssa>
~~rewrite it in rust~~
stikonas_ has quit [Remote host closed the connection]
<icecream95>
"Note that transfer_unmap is always enqueued" … "Possible data race by thread #1 [in] panfrost_ptr_unmap"
kaspter has quit [Ping timeout: 246 seconds]
kaspter has joined #panfrost
vstehle has quit [Ping timeout: 240 seconds]
<icecream95>
I managed to get through the first level of Quake without it crashing, so that's good enough testing to push it to the stable branch
<archetech>
youtube Sophie Wilson
<archetech>
original arm isa
<HdkR>
What's this about?
<alyssa>
archetech: what about her?
<archetech>
when we see an instruction like mov pop in assembly that was her
<HdkR>
Okay? Is this relevant to something?
<archetech>
she was co-inventor of arm
<archetech>
relevelent to some as hsitory
<archetech>
history
<HdkR>
I'm going to recommend you stay on topic to panfrost specific things rather than randomly dropping a topic in to the channel
italove has joined #panfrost
mixfix41 has joined #panfrost
camus1 has joined #panfrost
kaspter has quit [Read error: Connection reset by peer]
camus1 is now known as kaspter
italove has quit [Ping timeout: 256 seconds]
robmur01 has quit [Ping timeout: 272 seconds]
vstehle has joined #panfrost
davidlt has joined #panfrost
icecrea105 has joined #panfrost
icecream95 has quit [Ping timeout: 240 seconds]
mixfix41 has quit [Ping timeout: 260 seconds]
rando25892 has quit [Ping timeout: 256 seconds]
icecrea105 has quit [Quit: Reconnecting]
icecream95 has joined #panfrost
rando25892 has joined #panfrost
mixfix41 has joined #panfrost
Elpaulo has quit [Quit: Elpaulo]
jernej has quit [Ping timeout: 264 seconds]
dstzd has quit [Ping timeout: 264 seconds]
jernej has joined #panfrost
dstzd has joined #panfrost
kaspter has quit [Quit: kaspter]
<icecream95>
alyssa: Is there any reason that vs_vary and fs_vary are regenerated for each draw?
raster has joined #panfrost
icecream95 has quit [Ping timeout: 264 seconds]
archetech has quit [Quit: Konversation terminated!]
HdkR has quit [Ping timeout: 246 seconds]
_whitelogger has joined #panfrost
stikonas has joined #panfrost
robmur01 has joined #panfrost
kaspter has joined #panfrost
italove has joined #panfrost
<alyssa>
icecream95: Gallium doesn't give us linking information
<alyssa>
^ If those land, we can probably move generation to CSO create time. But that's blocked on upstream stuff.
davidlt has quit [Ping timeout: 256 seconds]
klaxa has joined #panfrost
kaspter has quit [Quit: kaspter]
alyssa has quit [Quit: leaving]
chewitt has joined #panfrost
afaerber has quit [Ping timeout: 260 seconds]
alpernebbi has joined #panfrost
afaerber has joined #panfrost
yann has quit [Read error: No route to host]
yann has joined #panfrost
unoccupied has joined #panfrost
yann has quit [Ping timeout: 240 seconds]
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
yann has joined #panfrost
davidlt has joined #panfrost
yann has quit [Ping timeout: 272 seconds]
yann has joined #panfrost
yann has quit [Ping timeout: 264 seconds]
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
yann has joined #panfrost
sigmaris has quit [Ping timeout: 260 seconds]
sigmaris has joined #panfrost
AreaScout has joined #panfrost
<AreaScout>
alyssa I have tried latest mesa on odroid C4 and I get a strange versions string 'GLSL Version : OpenGL ES GLSL ES 1.0.16' it's the first time that I have seen a x.16 ... also I did run a shader test and got this as a result https://www.areascout.at/screen.png