<HdkR>
MIPS even has variable length delay slots depending on architecture. So it needs to be pretty flexible
<HdkR>
The dreaded....load delay slot
davidlt has joined #panfrost
<cyrozap>
Ohhhh, I'm suddenly starting to understand some things
<cyrozap>
MediaTek's docs say that the MD32 supports "parallel execution". I didn't really understand what that meant until now--remember those "paired" 16-bit instructions? My current hypothesis is that they're executed in a single instruction cycle, and that this is the "parallel execution" the docs were referring to.
<cyrozap>
And the whole delay slot thing is probably what they meant by "static branch prediction".
kaspter has quit [Ping timeout: 256 seconds]
kaspter has joined #panfrost
anarsoul|2 is now known as anarsoul
camus has joined #panfrost
vstehle has joined #panfrost
kaspter has quit [Ping timeout: 256 seconds]
camus is now known as kaspter
kaspter has quit [Ping timeout: 240 seconds]
kaspter has joined #panfrost
tomboy64 has quit [Ping timeout: 240 seconds]
tomboy64 has joined #panfrost
davidlt has quit [Ping timeout: 240 seconds]
camus has joined #panfrost
kaspter has quit [Ping timeout: 240 seconds]
camus is now known as kaspter
davidlt has joined #panfrost
camus has joined #panfrost
kaspter has quit [Ping timeout: 256 seconds]
camus is now known as kaspter
ckeepax has quit [Quit: WeeChat 2.7.1]
ckeepax has joined #panfrost
raster has joined #panfrost
chewitt has quit [Read error: Connection reset by peer]
chewitt has joined #panfrost
stikonas has joined #panfrost
chewitt has quit [Ping timeout: 256 seconds]
camus has joined #panfrost
kaspter has quit [Remote host closed the connection]
camus is now known as kaspter
camus has joined #panfrost
kaspter has quit [Ping timeout: 258 seconds]
camus is now known as kaspter
kaspter has quit [Ping timeout: 260 seconds]
kaspter has joined #panfrost
<alyssa>
cyrozap: Paired being parallel is typical for VLIW.
<alyssa>
Even though this obviously isn't "real" VLIW.
<robmur01>
as I suspected, it's more like "dual-issue: IA-64 edition"
camus has joined #panfrost
kaspter has quit [Ping timeout: 258 seconds]
camus is now known as kaspter
chewitt has joined #panfrost
tomboy64 has quit [Remote host closed the connection]
tomboy64 has joined #panfrost
popolon has joined #panfrost
kaspter has quit [Quit: kaspter]
<macc24>
bbrezillon: so gpu was clocking wayyyy too slow to have smooth scrolling in fullscreen firefox
<macc24>
or maybe gpu reclocking wasn't working hmm
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
raster has joined #panfrost
<alyssa>
devfreq is its own beast
raster has quit [Quit: Gettin' stinky!]
raster has joined #panfrost
rando25892 has joined #panfrost
warpme_ has joined #panfrost
<macc24>
now minetest is less unplayable on G72 at 1920x1200
<macc24>
:D
archetech has joined #panfrost
tomboy64 has quit [Remote host closed the connection]
tomboy64 has joined #panfrost
wiizzard has joined #panfrost
davidlt has quit [Ping timeout: 256 seconds]
<kinkinkijkin>
macc24 you're lucky to be able to check these things out
<kinkinkijkin>
i envy your ownership of a simple 1 dollar adapter
<macc24>
kinkinkijkin: that gives me an idea of a "live os" for cadmium
<macc24>
you can't get usb-c <-> usb-a adapter locally?
<kinkinkijkin>
cant do things that require money if you have none
<kinkinkijkin>
unfortunate reality of things
<macc24>
yeah i can relate to that
_whitelogger has joined #panfrost
_whitelogger has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
archetech has quit [Quit: Konversation terminated!]
archetech has joined #panfrost
raster has joined #panfrost
raster has quit [Client Quit]
raster has joined #panfrost
<kinkinkijkin>
something i forgot to say when i was testing my xu4 with games, the flickering looks like exactly half of the faces are drawn at any time
<kinkinkijkin>
and the flickering is just somehow related to how the faces drawn are chosen
<kinkinkijkin>
i would test it more, get some video, but that is a real pain on the eyes
<kinkinkijkin>
possibly a bug related to the 32-bit host
<HdkR>
Bug in CW versus CCW winding order?
<kinkinkijkin>
possibly that
<HdkR>
Is it some app that uses non-default winding order?
<kinkinkijkin>
theres a few things i can think of that it could be
<kinkinkijkin>
it's everything, everything flickers in the same way
<kinkinkijkin>
es2gears does not flicker within the application itself
<kinkinkijkin>
the window flickers in and out though, as sway flickers
<kinkinkijkin>
same with glxgears
<kinkinkijkin>
could also be related to backface culling
<kinkinkijkin>
what i was thinking it *might* be is some bug related to the 32-bit host where only half the faces of any given draw call are actually read
<kinkinkijkin>
would make sense that backface culling would affect it since it doubles the number of faces
<kinkinkijkin>
and es2gears and glxgears sometimes have one or two faces flicker out, rarely though
<kinkinkijkin>
i dont remember if es2gears does backface culling though
<kinkinkijkin>
ill have to look again
<kinkinkijkin>
es2gears uses cull_face without setting which face to cull
<kinkinkijkin>
mightve just found another bug as well
<kinkinkijkin>
sway and wlroots both do not contain any instance of glEnable(GL_CULL_FACE)
<kinkinkijkin>
interesting
<kinkinkijkin>
they have SCISSOR_TEST though
<kinkinkijkin>
im actually going to look into this myself, i have a few ideas what it might be and a few hours free
<kinkinkijkin>
alyssa: you there? looking for a few files to look into and mesa is massive
megi has quit [Ping timeout: 272 seconds]
alyssa has quit [Remote host closed the connection]
alyssa has joined #panfrost
<alyssa>
kinkinkijkin: here physically, if not mentally
raster has quit [Quit: Gettin' stinky!]
<kinkinkijkin>
think you can tell me where i might find a few files relevant to data control of vertex element arrays?
<kinkinkijkin>
as it relates to panfrost specifically
alyssa has quit [Remote host closed the connection]
<kinkinkijkin>
oop
alyssa has joined #panfrost
<kinkinkijkin>
think youd be up to tell me where i might find a few source files relevant to data control of element arrays and buffers?
<kinkinkijkin>
vertex elements specifically, but more may be relevant
<alyssa>
uhmmm
<kinkinkijkin>
in panfrost's source specifically
<kinkinkijkin>
if you dont feel up to it mentally rn dont worry
<alyssa>
Though tbh... t6xx is architecturally the same as t720 where all this stuff works. so whatever's going wrong is a hardware bug, so the fix in software will be icky.
archetech has quit [Quit: Konversation terminated!]
<kinkinkijkin>
32-bit hosts as well?
<alyssa>
rk3288 is 32-bit and pretty close to conformant on gles2
<alyssa>
I received an email recently where someone pinned down the t6xx instability to cache behaviour I think? Kernel stuff is over my head
<alyssa>
bbrezillon: ^^
<kinkinkijkin>
what im thinking is possibly related to t6xx on 32-bit... element size as read by the gpu may be half compared to 64-bit, which would be a hardware issue, yet not a hardware bug
<alyssa>
actually we saw simlar bugs with bifrost, and it sounds like it might be a similar fix. at any rate that's a kernel issue, not mesa.
<alyssa>
wdym element size?
tomboy64 has quit [Remote host closed the connection]
<kinkinkijkin>
the size of the uints pointing the elements
tomboy64 has joined #panfrost
<alyssa>
uhm I'll check
<kinkinkijkin>
since this isnt a stability issue, this specific thing does not cause a crash
<alyssa>
Flickering could still be a stability thing
<alyssa>
If data is corrupted, or jobs are only half executing, or whatever
<kinkinkijkin>
true
<kinkinkijkin>
but this is a little too uniform and cut-down-the-middle to be likely to be stability
archetech has joined #panfrost
<alyssa>
Sure
<kinkinkijkin>
its worth checking, and the fix would be dead simple if this is the case
<kinkinkijkin>
and looking at this code, if this is the issue, this code is exactly on-the-nose for code that would bring this issue up
<kinkinkijkin>
if this is the issue i can write up a fix and test it, just need to unpack my xu4 and hook it back up
<alyssa>
Aside - write value jobs might not be available on some t6xx hardware, that's probably unrelated but worth checking (I think we have an igt test for this)