mesa-panfrost-git belongs to the first tests of jgmdev who setup archlinuxdroid repo. He wrote you should better use mesa-arm-git instead because this contains the most recent updates of mesa.
Hm, that is surprising low.
Here are results of jgmdev running glmark2-es2 before it crashed:
glmark2 2020.04
=======================================================
OpenGL Information
GL_VENDOR: Panfrost
GL_RENDERER: Mali G52 (Panfrost)
GL_VERSION: 2.1 Mesa 21.0.0-devel (git-f89b29f708)
=======================================================
[build] use-vbo=false: FPS: 312 FrameTime: 3.205 ms
[build] use-vbo=true: FPS: 347 FrameTime: 2.882 ms
[texture] texture-filter=nearest: FPS: 404 FrameTime: 2.475 ms
[texture] texture-filter=linear: FPS: 405 FrameTime: 2.469 ms
[1] 2153 segmentation fault (core dumped) glmark2
Does not work perfectly for me. I still have dropped frames (~9 fps) on X96Max+ with Mali G31. Maybe because libvdpau_meson and libcuda are missing. Did not find it in one of the packages.
[themojoman@manjaro-arm64 ~]$ mpv --hwdec=auto --hwdec-codecs=all https://youtube.com/watch?v=BHACKCNDMW8
(+) Video --vid=1 (*) (h264 1920x1080 30.000fps)
(+) Audio --aid=1 --alang=eng (*) (opus 2ch 48000Hz)
Cannot load libcuda.so.1
[vaapi] libva: vaGetDriverNameByIndex() failed with unknown libva error, driver_name = (null)
Failed to open VDPAU backend libvdpau_meson.so: cannot open shared object file: No such file or directory
AO: [pulse] 48000Hz stereo 2ch float
Using hardware decoding (v4l2m2m-copy).
VO: [gpu] 1920x1080 nv12
AV: 00:01:40 / 03:00:08 (0%) A-V: 0.000 Dropped: 852 Cache: 140s/54MB
[ffmpeg/demuxer] mov,mp4,m4a,3gp,3g2,mj2: Packet corrupt (stream = 0, dts = 3776000).
[ffmpeg] NULL: Invalid NAL unit size (234662 > 206658).
[ffmpeg] NULL: missing picture in access unit with size 206716
Exiting... (Quit)
Wow is like up to 6 times slow 52 fps vs 312 fps for the first test! May be some other settings might be involved to optimize the performance.
There was an update of mesa-arm-git (Mesa 21.0.0-devel (git-1200f6da0b) and just update it and ran glmark2-es2 (glmark2 20.20.04) and GT King Pro score is slightly lower 41.
Looking at your glmark2-es2 results with a supposedly lower GPU (and with a much earlier mesa-git-20.3.0) than in GTKPro with G52 and your score is significantly better and the Odroid N2 results by jgmdev are way better. This lead me to suspect that Manjaro image for GT King Pro have not been optimized, missing drivers or enviromental set up?
I know you guys are excited for Panfrost on the Bifrost architectures. But itâs still very early days.
Itâs normal for results to be all over the place, when the development moves as quickly as it does.
You donât see users testing Alpha drivers for Nvidia cards, do you?
The fact that itâs actually just around the corner is a amazing to be honest. So I donât get the âyour benchmark score is higher than mineâ thing. It means absolutely nothing this early in the development cycle.
I see there was a bunch of commits on the 17th relating to Bifrost, so the mesa-git package that hits Unstable branch on sunday will be exciting I think.