chrisf has quit [Remote host closed the connection]
chrisf has joined #panfrost
popolon has quit [Quit: WeeChat 3.0]
mixfix41 has quit [Ping timeout: 246 seconds]
<alyssa>
robclark: here's hoping :p
<alyssa>
it's still named "sgx" in the dt so idk
<macc24>
alyssa: iirc some bifrost in android is named "mali-midgard"
<alyssa>
macc24: yeah, bifrost is midgard with the ISA swapped out :p
archetech has joined #panfrost
<macc24>
alyssa: so bifrost is basically midgard, but with most of midgard swapped out for bifrost parts?
<HdkR>
Yes, the ISA :P
archetech has quit [Ping timeout: 260 seconds]
stikonas has quit [Remote host closed the connection]
atler is now known as Guest46768
Guest46768 has quit [Killed (tepper.freenode.net (Nickname regained by services))]
atler has joined #panfrost
vstehle has quit [Ping timeout: 256 seconds]
kaspter has joined #panfrost
kaspter has quit [Excess Flood]
kaspter has joined #panfrost
ndufresne8 has joined #panfrost
italove has quit [Read error: Connection reset by peer]
anarsoul has quit [Remote host closed the connection]
Depau has joined #panfrost
bbrezillon has quit [Ping timeout: 246 seconds]
anarsoul has joined #panfrost
Depau_ has quit [Read error: Connection reset by peer]
rtp has quit [Ping timeout: 264 seconds]
ndufresne has quit [Ping timeout: 264 seconds]
bbrezillon has joined #panfrost
ndufresne8 is now known as ndufresne
rtp has joined #panfrost
<icecream95>
macc24: I already have tracing set up on CrOS, including for dEQP
* alyssa
doesn't have tracing anywhere since she broke her setup whoops
warpme_ has quit [Quit: Connection closed for inactivity]
camus has joined #panfrost
kaspter has quit [Ping timeout: 264 seconds]
camus is now known as kaspter
robmur01 has joined #panfrost
robmur01 has quit [Ping timeout: 246 seconds]
davidlt has joined #panfrost
camus has joined #panfrost
kaspter has quit [Ping timeout: 246 seconds]
camus is now known as kaspter
Lyude has quit [Quit: WeeChat 2.8]
vstehle has joined #panfrost
chewitt has quit [Quit: Zzz..]
chewitt has joined #panfrost
Lyude has joined #panfrost
chewitt has quit [Quit: Zzz..]
chewitt has joined #panfrost
camus has joined #panfrost
kaspter has quit [Ping timeout: 260 seconds]
camus is now known as kaspter
chewitt has quit [Read error: Connection reset by peer]
chewitt_ has joined #panfrost
chewitt_ has quit [Quit: Adios!]
_whitelogger has joined #panfrost
<icecream95>
alyssa: You Won't Believe These Two Simple Tricks For Increasing A Piglit drawoverhead Subtest's Draws/Second By 17%
kaspter has quit [Ping timeout: 260 seconds]
kaspter has joined #panfrost
warpme_ has joined #panfrost
<macc24>
icecream95: ok, so i won't rush getting cros kernel to compile and boot on duet haha
davidlt has quit [Remote host closed the connection]
davidlt has joined #panfrost
stikonas has joined #panfrost
thecycoone1 has quit [Ping timeout: 272 seconds]
karolherbst has joined #panfrost
raster has joined #panfrost
<kinkinkijkin>
i too love making new gpus by simply changing the native isa of an existing gpu and saying the gpu is different now
<kinkinkijkin>
technically correct, the best correct
<kinkinkijkin>
i have an x86_64 laptop now but it lacks a charger and is likely slower than my duet
<kinkinkijkin>
but It'll Run Video Games Natively so enh
<kinkinkijkin>
intel graphics 4400
camus has joined #panfrost
kaspter has quit [Ping timeout: 246 seconds]
camus is now known as kaspter
italove2 has joined #panfrost
orkid has joined #panfrost
<orkid>
and by the way, thanks for all the work on the drivers. :)
robmur01 has joined #panfrost
robmur01 has quit [Read error: Connection reset by peer]
robmur01 has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
raster has joined #panfrost
<alyssa>
icecream95: Graphics Driver Developers Hate Them!
leper` has quit [Quit: .]
leper` has joined #panfrost
stikonas has quit [*.net *.split]
anarsoul has quit [*.net *.split]
jernej has quit [*.net *.split]
dstzd has quit [*.net *.split]
endrift has quit [*.net *.split]
robertfoss_ has quit [*.net *.split]
xperia64 has quit [*.net *.split]
ente has joined #panfrost
enty has quit [Read error: Connection reset by peer]
stikonas has joined #panfrost
jernej has joined #panfrost
endrift has joined #panfrost
anarsoul has joined #panfrost
dstzd has joined #panfrost
robertfoss_ has joined #panfrost
xperia64 has joined #panfrost
exit70[m] has quit [Ping timeout: 258 seconds]
icecream95 has quit [Ping timeout: 244 seconds]
wiizzard has quit [Ping timeout: 244 seconds]
Ke has quit [Ping timeout: 244 seconds]
l-as has quit [Ping timeout: 258 seconds]
raster has quit [Quit: Gettin' stinky!]
raster has joined #panfrost
clementp[m] has quit [Ping timeout: 240 seconds]
l-as has joined #panfrost
exit70[m] has joined #panfrost
urjaman has quit [Read error: Connection reset by peer]
urjaman has joined #panfrost
l-as has quit [Ping timeout: 258 seconds]
exit70[m] has quit [Ping timeout: 260 seconds]
<macc24>
kinkinkijkin: what are specs of it?
<kinkinkijkin>
an ultra-low voltage i3 and aforementioned intel gpu
<kinkinkijkin>
4gb ddr3
<kinkinkijkin>
and a spinning rust disk
wiizzard has joined #panfrost
<macc24>
oh god
<macc24>
i played video games on slower computers
<kinkinkijkin>
i currently play video games on a slower computer
<kinkinkijkin>
my mother's desktop
<kinkinkijkin>
only thing that'll run trackmania
<kinkinkijkin>
pentium dual-core E5200 and an NVIDIA GT 220 GAMING GRAPHICS GPU UNIT CARD
<kinkinkijkin>
and 3gb ddr2
<kinkinkijkin>
well correction, the duet runs trackmania wii at half speed fine
<kinkinkijkin>
that's the catch though, half speed
<kinkinkijkin>
and i dont remember dolphin still having frameskip
<kinkinkijkin>
in fact i pretty explicitly remember it being removed like 7 years ago
icecream95 has joined #panfrost
<alyssa>
icecream95: Could you send your magic 2x compression function?
<alyssa>
Thanks
<kinkinkijkin>
oh by the way, i thought id drop this in here, with PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT, i was able to get dolphin running out-of-box with a git build on my duet, and it can get full-speed during cpu video decode as long as there's nothing going through the gpu
clementp[m] has joined #panfrost
<kinkinkijkin>
as soon as you render something 3d and game-like it drops speed fast
Ke has joined #panfrost
m][sko has joined #panfrost
<alyssa>
COHERENT should be ok for Mali
<alyssa>
and PERSISTENT should also be ok
<alyssa>
but the interaction of the two is bad news for us
cowsay_ has quit [Quit: No Ping reply in 180 seconds.]
<alyssa>
coherency only applies at job buffer boundaries
<macc24>
kinkinkijkin: i bet gt220 is faster than ironlake graphics haha
<alyssa>
Doing PERSISTENT+COHERENT on mali might have subtle deviance from the spec if you update a resource CPU side while the GPU is using it
<macc24>
minecraft should run comfortably on that box at lowest settings
<alyssa>
but I'm not sure how this is specced exactly
<macc24>
terraria too
cowsay has joined #panfrost
l-as has joined #panfrost
exit70[m] has joined #panfrost
<kinkinkijkin>
the gpu in the desktop being slightly faster doesn't really matter much when the cpu is so slow it can't even saturate it
<macc24>
my t400 had p8600. it was my gaming rig for a while.
<kinkinkijkin>
alyssa idk i havent had any unique bugs since enabling it, and have used dolphin extensively (which uses ARB_buffer_storage, which requires this)
<kinkinkijkin>
macc24 again the desktop has a pentium e5200
<kinkinkijkin>
2008 bottom-end cpu
<macc24>
kinkinkijkin: pentium e5200 is almost the same in performance as p8600
<kinkinkijkin>
oh dang
<macc24>
p8600 is a dualcore 2008 laptop chip
<kinkinkijkin>
same struggle, same struggle
<kinkinkijkin>
though this desktop is running windows 10 because my mother uses windows
paulk-leonov has quit [Remote host closed the connection]
<macc24>
oh god
<kinkinkijkin>
the laptop i can put linux on though
raster has quit [Quit: Gettin' stinky!]
<macc24>
there's ##panfrost-offtopic btw
<kinkinkijkin>
fun thing in trackmania is when your cpu is so slow that in order to speed up the framerate you need to switch to nice shaders so that drawcall count is cut down
<kinkinkijkin>
there is?
<macc24>
yes
raster has joined #panfrost
raster has quit [Client Quit]
raster has joined #panfrost
kaspter has quit [Quit: kaspter]
paulk-leonov has joined #panfrost
paulk-leonov has quit [Remote host closed the connection]
paulk-leonov has joined #panfrost
paulk-leonov has quit [Read error: Connection reset by peer]
paulk-leonov has joined #panfrost
<paulk-leonov>
uit
<paulk-leonov>
oops
paulk-leonov has quit [Client Quit]
paulk-leonov has joined #panfrost
<bbrezillon>
alyssa: hm, looks like nir_lower_ssbo() is generating 64b ALUs (load_ssbo_address + ssbo_offset)
<bbrezillon>
I wonder why it didn't show up with the SSBO tests
<alyssa>
paulk-leonov: hullo there :)
<alyssa>
bbrezillon: That's expected
<paulk-leonov>
alyssa: o/
<alyssa>
bbrezillon: But then lower_int64 and friends lower it away so it's fine.
<alyssa>
Possibly that patch hasn't merged yet for Bifrost
<bbrezillon>
oh, looks like ssbo is GLES 3.1
m][sko has quit [Quit: Connection closed]
<alyssa>
bbrezillon: Yeah, icecream95's opencl branch has all of that working though
<bbrezillon>
tried with all int64 lowering enabled, but the assert() in bi_alu_src_index() fails
<bbrezillon>
alyssa: ack, I'll try it out
<alyssa>
there's one patch blocked but that's blocked on clover bug
<bbrezillon>
italove: feel free to send a separate MR for that one
karolherbst has quit [Quit: duh 🐧]
karolherbst has joined #panfrost
urjaman has quit [Ping timeout: 256 seconds]
Lyude has quit [Ping timeout: 256 seconds]
vstehle has quit [Ping timeout: 256 seconds]
urjaman has joined #panfrost
vstehle has joined #panfrost
Lyude has joined #panfrost
karolherbst has quit [Quit: duh 🐧]
karolherbst has joined #panfrost
shadeslayer has joined #panfrost
<dschuermann>
alyssa: are 16bit vec2 shifts also broken for you? how do you deal with them having 32bit shift operands?
<alyssa>
dschuermann: I haven't fully debugged int16 yet since I was busy rushing some hw bringup through the branch point (and even as-is, vec2 is one of my 6 side projects for when I get frustrated with scheduling/RA stuff :p)
<alyssa>
But -
<alyssa>
Bifrost takes an 8-bit shift regardless of the other operand size.
<alyssa>
The hardware semantic is to ignore all but the bottom log2(other_operand_size) bits, effectively doing a bitwise AND.
<alyssa>
I believe that matches the NIR semantic.
<alyssa>
The upshot is we can totally ignore the NIR bit size when emitting a shift and always do an implicit downcast to 8-bit.
<alyssa>
At least, that should be conformant. It does unfortunately have two bad cases for perf:
<alyssa>
- if the app does a 16-bit shift at the GLSL level, NIR will insert a u2u32 which our backend isn't smart enough to see through (maybe a peephole opt pass?)
<alyssa>
- if the app does a shift by a multiple 8 on the shift operand, we could do that for free by picking the appropriate 8b lane (as opposed to always taking the bottom 8bits)... this seems like a _seriously_ obscure pattern
<alyssa>
also, depending on how aggressive nir_opt_vectorize is, the calculation of the shift might be needlessly scalarized
<alyssa>
Given that all of this is broken right now, I haven't been too perf worried :p
<alyssa>
TL;DR kinda annoying that NIR forces 32-bit but doesn't break anything
<alyssa>
I don't remember off-hand what the semantic is on Midgard
<alyssa>
ok, this is getting ridiculous, how hard can it be to decouple the panwrap build from the rest of mesa
<alyssa>
HdkR: ~~if only to get you to valhall~~
<HdkR>
wah?
<HdkR>
gimme it
<alyssa>
HdkR: IIRC you were blocked on panwrap being totally broken build system
<alyssa>
that's also screwing me up for bifrost so... yeah, probably a good idea to shave that yak
<alyssa>
panwrap source does need to live in mesa but it should be totally self-contained
<HdkR>
ah
<HdkR>
I forgot about that
<macc24>
what is panwrap
<alyssa>
er, *pandecode
<alyssa>
macc24: pandecode is our home grown debugger, ingesting GPU memory and pretty-printing everything
<alyssa>
powers PAN_MESA_DEBUG=trace
<alyssa>
panwrap is the glue code to use pandecode on the blob
<alyssa>
pandecode was sort of obsoleted by GenXML since we can autogenerate most of it nowadays
<alyssa>
HdkR: Ideally just the XML definitions, a Python file, and some supporting C code should be it, no external deps
<alyssa>
Particularly if you don't want to link the disassemblers
<HdkR>
Guess that might work. I believe I was just caught up on the android build system problems though
<alyssa>
Getting close, I uh hope
<dschuermann>
alyssa: what I meant was that you also get 32bit vec2 operands. do you repack that in the backend?
<alyssa>
dschuermann: Oh.. haven't come across those yet..
<alyssa>
er-
<alyssa>
shoot
<alyssa>
I think I specifically scalarized shifts to deal with that
<alyssa>
yeah, this is totally broken for us too :)
m][sko has joined #panfrost
<alyssa>
Update: I have a PoC decoupling pandecode + bifrost disasm from the rest of mesa
<alyssa>
so I can now build that on HdkR's bifrost board
<alyssa>
==> I have access to blob traces again
<HdkR>
:)
<alyssa>
HdkR: That also means you can start hacking on Valhall without fighting android build system 😇
<alyssa>
Since there's no more dependencies on any mesa or anything
<alyssa>
It's just good old-fashioned C99 and the stdlib
<HdkR>
Very nice
<alyssa>
I have to run now but can dump the list o' links in a bit :)
<icecream95>
alyssa: I removed the global binding commit from my MR, so it can be merged now