seeeath has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
capta1nt0ad has joined #asahi-gpu
capta1nt0ad has quit [Remote host closed the connection]
capta1nt0ad has joined #asahi-gpu
kesslerd has quit [Remote host closed the connection]
kesslerd has joined #asahi-gpu
alyssa has joined #asahi-gpu
<
alyssa>
It was blocked on !20836 for a minute there but !20836 is now merged so that can go in
<
alyssa>
and !20836 is kinda churn so the rebasing around has been really annoying for conflicts
<
alyssa>
luckily it's pretty small, since it's been landed piece by piece already (as !20836 and others)
<
alyssa>
with !20836, we now finally run most of our NIR passes at CSO create time (shader compile time from the app's perspective)
<
alyssa>
in theory that means draw time shader variants are cheaper
<
alyssa>
since now they're not starting "from scratch", but instead are just applying key-based optimizations to an already "almost ready" NIR shader
<
alyssa>
(and then running that NIR through the backend)
<
alyssa>
(I have a few other MRs up if you're feeling reviewy, but !21104 is the one that's causing rebase problems)
capta1nt0ad has quit [Ping timeout: 480 seconds]
capta1nt0ad has joined #asahi-gpu
capta1nt0ad has quit [Ping timeout: 480 seconds]
possiblemeatball has joined #asahi-gpu
kesslerd has quit [Remote host closed the connection]
Emantor has joined #asahi-gpu
kesslerd has joined #asahi-gpu
kesslerd has quit [Remote host closed the connection]
kesslerd has joined #asahi-gpu
i509vcb has quit [Quit: Connection closed for inactivity]
c10l has quit [Read error: Connection reset by peer]
c10l has joined #asahi-gpu
djorz has quit [Ping timeout: 480 seconds]
possiblemeatball has quit [Quit: Quit]
<
lina>
alyssa: I'll probably look at it tomorrow, busy today getting Luna packed up for her flight back ^^
stipa has quit [Ping timeout: 480 seconds]
stipa has joined #asahi-gpu
Z750 has quit [Read error: Connection reset by peer]
Z750 has joined #asahi-gpu
ChaosPrincess has quit [Quit: WeeChat 3.8]
ChaosPrincess has joined #asahi-gpu
possiblemeatball has joined #asahi-gpu
md_ has joined #asahi-gpu
possiblemeatball has quit [Quit: Quit]
possiblemeatball has joined #asahi-gpu
cylm has quit [Ping timeout: 480 seconds]
nate8 has joined #asahi-gpu
amarioguy has joined #asahi-gpu
<
alyssa>
with alpha to cov enabled, enabling writemask causes:
<
alyssa>
Shader::unk1 to be false
<
alyssa>
st_tile to be masked, of course
<
alyssa>
Occlusion query 2::Unknown from 0x1ffff -> 0x2
seeeath has joined #asahi-gpu
<
alyssa>
with no alpha to cov, we still have unk1 false and st_tile masked
<
alyssa>
and still 0x2
<
alyssa>
notably an... Opaque pass type? which seems broken?
<
alyssa>
It seems like all the difference is from this oq2 unknown
<
alyssa>
unless, hmm wait I did a green mask right
<
alyssa>
right, yes, blue mask is 0x4
<
alyssa>
so it's like a mask of what's enabled
<
alyssa>
where 0x1ffff means "everything is enabled"?
amarioguy has quit [Ping timeout: 480 seconds]
<
alyssa>
with blue on 4 render targets, 0x4444
<
alyssa>
but I wonder how that works with 8 rt?
<
alyssa>
with 8 render targets instead (and switched down to rgba8888) it's switched itself to translucnt and disabled early-z and unk2
<
alyssa>
also fragment_properties::unk3 from 0xf to 0x0 in this case
<
alyssa>
uhh with 5 render targets (rgba8) and blue
<
alyssa>
it's still no earlyz/unk2/unk3 and translcuent
<
alyssa>
but now the unknown is 0x2492
<
alyssa>
which makes even less sense
<
alyssa>
oh, 4 render targets (rgba8) and blue has the first set of changes, just with unknown 0x4444
<
alyssa>
with 2 render tagets, 0x44
<
alyssa>
but with early-z/etc still enabled
<
alyssa>
ditto for 3 rts, with 0x444
<
alyssa>
only at 4 does early-z come off
<
alyssa>
oh but this was with MSAA 4x
<
alyssa>
no still no early-z with 4 rts and MSAA
djorz has joined #asahi-gpu
kesslerd has quit [Ping timeout: 480 seconds]
<
alyssa>
as for fragment shader side effects
<
alyssa>
"set when frag shader spills" is set
<
alyssa>
so that's actually a "frag shader side effects" bit
<
alyssa>
lina: ^^ that one is uapi
<
alyssa>
unk 2: 2 -> 3
<
alyssa>
interestingly, the frag shader gets a `gl_FragDepth = gl_Position.z` write after the store but before the st_tile
<
alyssa>
early-z is disabled
<
alyssa>
tri merging is disabled
<
alyssa>
punchtrough pass type + conservative depth any thanks to that depth write
<
alyssa>
with [[early_fragment_tests]]
<
alyssa>
still "spills"
<
alyssa>
still unk2->3
<
alyssa>
no zs_emit, but now we have at the beginning of the shader `gl_SampleMask = 1`
<
alyssa>
and still early-z testing is disabled in hardware?!
<
alyssa>
passtype of translucent punch through (?!?!?!)
<
alyssa>
that's enough hw for my brain rn
skoobasteeve has joined #asahi-gpu
skoobasteeve has quit []
skoobasteeve has joined #asahi-gpu
<
alyssa>
my guess is that the unknown mask is related to HSR
<
alyssa>
lina: also, I wonder if we have the same ISP_CTL register as powervr
<
alyssa>
since that has some line style bits which will be useful later
<
alyssa>
(for a vulkan extension)
amarioguy has joined #asahi-gpu
<
alyssa>
maybe it's "tile_config"
<
alyssa>
shrug not too important
<
alyssa>
"Another feature that reduces HSR efficiency is write masking."
<
alyssa>
ding ding ding
seeeath has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
seeeath has joined #asahi-gpu
<
alyssa>
ahaha how did I miss this
<
alyssa>
"opaque", "feedback", "translucent"
<
alyssa>
so, the passtype enum. nice
skoobasteeve has joined #asahi-gpu
<
alyssa>
well, there's another bug I didn't know I had
possiblemeatball has quit [Quit: Quit]
i509vcb has joined #asahi-gpu
possiblemeatball has joined #asahi-gpu
seeeath has joined #asahi-gpu
amarioguy has quit [Ping timeout: 480 seconds]
amarioguy has joined #asahi-gpu
possiblemeatball has quit [Quit: Quit]
possiblemeatball has joined #asahi-gpu
possiblemeatball has quit []
possiblemeatball has joined #asahi-gpu