swaggie has joined #asahi-gpu
swaggie has quit [Ping timeout: 480 seconds]
cylm_ has quit [Ping timeout: 480 seconds]
swaggie has joined #asahi-gpu
swaggie has quit [Ping timeout: 480 seconds]
* alyssa
debugs texture buffers
<
alyssa>
ok, got the basic functionality working
<
alyssa>
Passed: 405/405 (100.0%)
<
alyssa>
that's nice, but no tests in that section for the actually hard part.
<
alyssa>
piglit has the tests
<
marcan>
alyssa: I haven't pushed last night's package out to most users, I can hold off if you think it regresses something
<
alyssa>
marcan: ack, please do
<
marcan>
please hold off?
<
alyssa>
I don't know if it regresses any actual workloads (i.e. anything that can be hit without overriding versions/extensions)
<
alyssa>
but I know it regresses gles3 stuff which means it's possible big gl 2 is regressed too, i haven't done enough testing to know
<
alyssa>
I worked out a solution that I'm working on
<
alyssa>
but it's a pretty big feature
<
marcan>
(that said, "push it and see if anyone complains"
*is* a valid strategy for mesa-asahi-edge, JFYI)
<
alyssa>
not multisampling big, but
swaggie has joined #asahi-gpu
<
alyssa>
buffer textures working, including with rgb32 support
<
alyssa>
honestly a little horrifying
<
alyssa>
Test case 'dEQP-GLES3.functional.pbo.native.clears'..
<
alyssa>
Unhandled intrinsic image_store
<
alyssa>
oh you've got to be kidding me.
pthariensflame has joined #asahi-gpu
pthariensflame has quit []
bcrumb has joined #asahi-gpu
bcrumb has quit [Quit: WeeChat 3.7.1]
bcrumb has joined #asahi-gpu
bcrumb has joined #asahi-gpu
bcrumb has joined #asahi-gpu
Dcow has joined #asahi-gpu
SSJ_GZ has joined #asahi-gpu
bcrumb has quit [Quit: WeeChat 3.7.1]
Dcow has quit [Ping timeout: 480 seconds]
bcrumb has joined #asahi-gpu
bcrumb has quit [Quit: WeeChat 3.7.1]
swaggie has quit [Remote host closed the connection]
swaggie has joined #asahi-gpu
Dcow has joined #asahi-gpu
Dcow has quit [Ping timeout: 480 seconds]
swaggie has quit [Ping timeout: 480 seconds]
swaggie has joined #asahi-gpu
Dcow has joined #asahi-gpu
swaggie has quit [Ping timeout: 480 seconds]
pjakobsson_ has joined #asahi-gpu
pjakobsson has quit [Ping timeout: 480 seconds]
bcrumb has joined #asahi-gpu
cylm_ has joined #asahi-gpu
<
lina>
alyssa: Nice! ^^
swaggie has joined #asahi-gpu
swaggie has quit [Ping timeout: 480 seconds]
<
alyssa>
if you squash that into asahi/main it should be good enough to ship
<
alyssa>
but not even compile tested
c10l has quit [Ping timeout: 480 seconds]
swaggie has joined #asahi-gpu
swaggie has quit [Ping timeout: 480 seconds]
c10l has joined #asahi-gpu
LinuxM1 has joined #asahi-gpu
swaggie has joined #asahi-gpu
bcrumb has joined #asahi-gpu
swaggie has quit [Ping timeout: 480 seconds]
swaggie has joined #asahi-gpu
<
marcan>
alyssa: I squashed it into the last commit, hope that's what you wanted :p
swaggie has quit [Ping timeout: 480 seconds]
<
marcan>
pushed to asahi-dev
<
alyssa>
sounds good
LinuxM1 has quit [Quit: Leaving]
<
alyssa>
1c: 0509400a00c81200 device_load.TODO 0, i32, x, r1, u0_u1, r2, unsigned, lsl 2, 0, 0, 4, 0
<
alyssa>
i wonder why metal is generating this alternate load in this case
<
alyssa>
constant address space, fragment shader, within uniform control flow
bcrumb has joined #asahi-gpu
<
alyssa>
also interestingly metal is not inserting jmp_exec_any branches despite memory access in here, hmhmhmhm
swaggie has joined #asahi-gpu
<
alyssa>
same in device space
<
alyssa>
this is for
<
alyssa>
if (uniform) { load } else { texture }
<
alyssa>
if I drop the texture, it uses a regular device_load, and it uses a jmp_exec_none
<
alyssa>
I wonder if there's some sort of predication interaction here
<
alyssa>
if I drop the load and leave only the texture, I do get a jmp_exec_none
<
alyssa>
if (uniform) { load } else { load, load }
<
alyssa>
the "if" uses a device_load.TODO with no jmp_exec_none
<
alyssa>
the "else" uses 2 device_loads with jmp_exec_none
<
alyssa>
the TODO version is just the version that unsets bit 26
bcrumb has joined #asahi-gpu
swaggie has quit [Ping timeout: 480 seconds]
<
alyssa>
also, no uniform branching. bleh
<
alyssa>
2c: 3185020a0062210020000000 texture_sample 0, 0b00, 0b01100, 0b0, 0b00000, x, 0b01, r1, 0, ts0, ss0, tex_2d, r1_r2, 2, 314
<
alyssa>
interesting way of doing lod(0) in a frag shader, yes
<
alyssa>
2c: 3185020a0062210020000000 texture_sample 0, 0b00, 0b01100, 0b0, 0b00000, x, 0b01, r1, 0, ts0, ss0, tex_2d, r1_r2, 2, 314
<
alyssa>
and lod 1, apparently
<
alyssa>
"Alyssa, that's the same instruction"
<
alyssa>
oh maybe it's getting the lod from a uniform
<
alyssa>
value is 10, maybe that's really u5l
<
alyssa>
yep, that's it.
<
alyssa>
level is provided as a half-float within uniform u5l
<
alyssa>
ok, easy enough to handle
<
alyssa>
ok, that's one mystery down
<
alyssa>
likewise for texture_load, the level is passed in a 16-bit int uniform.
<
alyssa>
which means our current txf implementation is broken. gg.
swaggie has joined #asahi-gpu
swaggie has quit [Ping timeout: 480 seconds]
bcrumb has quit [Quit: WeeChat 3.7.1]
WindowPain_ has joined #asahi-gpu
WindowPain has quit [Ping timeout: 480 seconds]
alyssa has quit [Quit: leaving]
bcrumb has joined #asahi-gpu
swaggie has joined #asahi-gpu
swaggie has quit [Ping timeout: 480 seconds]
bcrumb has quit [Quit: WeeChat 3.7.1]
swaggie has joined #asahi-gpu
chengsun has joined #asahi-gpu
chengsun_ has quit [Ping timeout: 480 seconds]
MajorBiscuit has joined #asahi-gpu
swaggie has quit [Ping timeout: 480 seconds]
WindowPain has joined #asahi-gpu
SSJ_GZ has quit [Ping timeout: 480 seconds]
swaggie has joined #asahi-gpu