paulk has quit [Read error: Connection reset by peer]
Net147 has quit [Quit: Quit]
Net147 has joined #panfrost
aliceryhl has joined #panfrost
aliceryhl has quit [Ping timeout: 480 seconds]
aliceryhl has joined #panfrost
aliceryhl has quit [Ping timeout: 480 seconds]
aliceryhl has joined #panfrost
<dliviu>
bbrezillon: what's the best strategy for pushing my patch? wait until drm-misc catches up with v6.12-rc1 or push it into drm-misc-next-fixes?
aliceryhl has quit [Ping timeout: 480 seconds]
aliceryhl has joined #panfrost
manu has joined #panfrost
manu has quit []
manu has joined #panfrost
manu has quit []
paperManu has joined #panfrost
<stepri01>
dliviu: It should end up in drm-misc-fixes, but neither branch has the problematic commit yet and I believe it will be a compile error until that lands.
<stepri01>
So either it needs to find a faster route to Linus or wait for -rc1 (and the back-merge to drm-misc-fixes).
ente` is now known as ente
<dliviu>
yeah, I know, but apparently drm-misc-next-fixes can be used in such strange situations
<stepri01>
it would still need a back-merge first. Ask the DRM maintainers I guess
<paperManu>
hello! I'm doing some computer graphics development, maintaining a videomapping software in particular, and I wanted a confirmation that the Panfrost driver does not support geometry shaders? Is it meant to be added at some point?
<CounterPillow>
It does not support GS because the hardware doesn't support it
<CounterPillow>
It could be added by emulating it through compute shaders etc. eventually, but ideally you don't use geometry shaders even on hardware that supports it
<CounterPillow>
GS are a bad idea for performance everywhere regardless of hardware support for them.
<CounterPillow>
I don't know what your software does but chances are whatever you think of doing with GS is better solved with just better use of VBO+VAO+VS (e.g. instead of generating a bunch of geometry from a point primitive in a GS, have the geometry already generated and modify it in a vertex shader based on vertex attribs and such)
<paperManu>
ah, I wondered whether it was supproted by the hardware. Thanks for the clarification.
<paperManu>
and yes I know about the issues with geometry shaders. They are used to generate some geometry once, returned through transform feedback, and then used through another rendering pipeline during runtime
<paperManu>
I know that at some point I should convert some of this to compute shaders
aliceryhl has quit [Ping timeout: 480 seconds]
<bbrezillon>
dliviu: check with the drm-misc maintainers
<bbrezillon>
I tend to wait the -rc1 and ask for a backmerge in drm-misc-fixes
<bbrezillon>
but if you want the fix in before -rc1 is released, you'll probably have to apply to drm-misc-next-fixes
warpme has quit []
aliceryhl has joined #panfrost
<dliviu>
I've been slapped over the wrist twice for committing to drm-misc-next-fixes, at least now I have a good reason</joking>
aliceryhl has quit [Ping timeout: 480 seconds]
<alyssa>
paperManu: yeah, compute shaders are the preferred approach for that
<alyssa>
and a hypothetical driver impl would turn the GS+XFB into compute
<alyssa>
(asahi does this)
<alyssa>
and that compute shader would be surely slower than what you'd've written yourself
aliceryhl has joined #panfrost
<paperManu>
well, that would depend on how well I do it ;) Anyway, getting rid of geometry shaders has been on my todo list for some time, this might just be the good reason I needed to do it