<lina>
In the VS, messing with fetch_clip_area causes it to work or not, seemingly depending on the order/polarity of the condition there
<lina>
There whole thing seems to spill into the FS, my guess is NaN propagation? That's the only thing that makes sense since the FS breaks even when all code paths must return good values.
<lina>
From which I deduce that cmpsel stuff propagates NaNs? If not, something else is even more wrong...
<lina>
The FS side manifests in do_clip() and happens even if all paths there return good near-1.0 alpha values (but not if all return 1.0 because that causes everything to be optimized out)
<alyssa>
how do I use this apitrace?
<alyssa>
i'm not seeing anything
<lina>
Umm try --sb
<lina>
and --loop=300 to see the last frame more
<alyssa>
yeah ok thanks
<lina>
The square is supposed to be a map thing, not glitched out
<lina>
Do we know how NaN propagation works in general on AGX?
<alyssa>
`--sb --wait` was my friend
<alyssa>
and, IDK, correctly?
<lina>
I mean what about cmpsel type stuff?
<alyssa>
in general I'd expect the hardware to be IEEE compliant
<alyssa>
but GL isn't expected to be IEEE compliant and we're allowed to optimize assuming that there are no NaNs and such
<alyssa>
so if a GL shader is depending on IEEE compliant NaN behaviour, that's usually its own fault
<alyssa>
(If this were VK and it used KHR_float_controls, we'd have a different conversation)
<lina>
So one thing that fixes it is changing the condition in fetch_clip_area() to `if (index >= CLIP_TASK_EMPTY && index != 0xdead) {` (which should be a no-op as long as the index isn't that weird value)
<lina>
But one side effect of that is the order of the basic blocks, and the sense of the condition variable, get flipped in the asm
<lina>
So I think it has to do something with that
<alyssa>
alright
<alyssa>
general compiler snafu not related to floating point crimes sounds likely then
<alyssa>
lina: It's fixed on my next/ branch
<lina>
Yeah, I didn't see anything obvious but I didn't trace every single line...
<lina>
Oh.
<alyssa>
should I bisect?
<lina>
Please...
<lina>
That one's been broken in firefox for ages...
<alyssa>
Weird, usually I only fix fake bugs..
<lina>
It could also be randomly "fixed" though, given that it's a heisenbug..
<alyssa>
lina: 424ad888aeb ("agx: Don't reuse ssa_to_reg across blocks")
<alyssa>
register allocation bug fix
* alyssa
didn't realize that real apps were hitting that
i509vcb has quit [Quit: Connection closed for inactivity]
bisko has joined #asahi-gpu
bisko has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
possiblemeatball has joined #asahi-gpu
bisko has joined #asahi-gpu
bisko has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
bisko has joined #asahi-gpu
MajorBiscuit has quit [Quit: WeeChat 3.6]
bisko has quit []
<alyssa>
Mary: i'm rebasing agxv now fwiw
bisko has joined #asahi-gpu
<alyssa>
and rebased
<alyssa>
there is now NVK in the AGXV tree for easier copying :3
bisko has quit []
Bey0ndB1nary has joined #asahi-gpu
Armada has left #asahi-gpu [#asahi-gpu]
jeisom has quit [Quit: Leaving]
jeisom has joined #asahi-gpu
jeisom has quit [Quit: Leaving]
jeisom has joined #asahi-gpu
gi0 has joined #asahi-gpu
jeisom has quit [Remote host closed the connection]
jeisom has joined #asahi-gpu
Bey0ndB1nary has quit []
Guest8521 has quit [Quit: Bridge terminating on SIGTERM]
rhysmdnz has quit [Quit: Bridge terminating on SIGTERM]
rhysmdnz has joined #asahi-gpu
jeisom has quit [Quit: Leaving]
Guest8625 has joined #asahi-gpu
Bey0ndB1nary has joined #asahi-gpu
ju has quit [Remote host closed the connection]
ju has joined #asahi-gpu
billak has joined #asahi-gpu
jeisom has joined #asahi-gpu
jeisom has quit [Remote host closed the connection]
jeisom has joined #asahi-gpu
nela has quit [Quit: bye!]
nela has joined #asahi-gpu
jeisom has quit [Ping timeout: 480 seconds]
Bey0ndB1nary has quit []
jeisom has joined #asahi-gpu
gi0 has quit []
ATiltedTree_ has joined #asahi-gpu
maximbaz has quit [Quit: bye]
ATiltedTree has quit [Ping timeout: 480 seconds]
maximbaz has joined #asahi-gpu
ATiltedTree_ is now known as ATiltedTree
billak has quit [Quit: Konversation terminated!]
Z750 has quit [Quit: Ping timeout (120 seconds)]
Z750 has joined #asahi-gpu
jeisom has quit [Remote host closed the connection]
jeisom has joined #asahi-gpu
jeisom has quit [Ping timeout: 480 seconds]
<Mary>
yay
linuxgemini has quit [Read error: Connection reset by peer]
linuxgemini has joined #asahi-gpu
possiblemeatball has quit [Quit: Quit]
darkapex has quit [Remote host closed the connection]