ChanServ changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard + Bifrost + Valhall - Logs https://oftc.irclog.whitequark.org/panfrost - I don't know anything about WSI. That's my story and I'm sticking to it.
jdavidberger has joined #panfrost
pentallelogram has quit [Remote host closed the connection]
<alyssa>
gfxstrand: don't worry about it ;p
<alyssa>
I think it was stated on the corresponding agx MR
<hays>
panfrost drivers are humming along on this Odroid N2L -- got it playing MK64 very nicely
<hays>
just some positive feedback for y'all
davidlt_ has joined #panfrost
davidlt has quit [Ping timeout: 480 seconds]
davidlt_ has quit [Ping timeout: 480 seconds]
jdavidberger has joined #panfrost
<jdavidberger>
I can't find much information on the memory arch for bifrost. Trying to speed up memory access in compute shaders and was wondering if there is any chance of using host invisible super fast gpu memory for SSBO sized(~5mb) things?
<HdkR>
Itty bitty tiny caches
<jdavidberger>
so just the 64k uniform slots essentially?
<jdavidberger>
or you mean its just at the mercy of a tiny L2 cache?
<HdkR>
At the mercy of the tiny cache
<HdkR>
Even UBOs may not necessarily be amazingly quick :)
<jdavidberger>
if i didn't know better i'd think they didn't mean for this thing to be used heavily for compute
<robclark>
for heavy compute you want a big power hungry video card w/ real vram.. tilers make 3d workloads go fast by cleverness (aka tiling) reducing memory bandwidth requirements... but that trick doesn't apply to compute workloads where you need more brute force