12:03 <Kwiboo> CounterPillow: I have just pushed an updated branch at https://github.com/Kwiboo/linux-rockchip/commits/next-20250428-rk35xx-scmi-gpu-clk/ with some new rk3576 related scmi gpu clk related changes

12:04 <Kwiboo> main change is PCLK_GPU_ROOT is marked as critical, as it seem to be required to be enabled when pvtpll is used, and also panfrost changes to only set scmi clk rate when gpu is not pm suspended

12:06 <Kwiboo> did not really see any performance diff when using plain clk_gpu vs scmi_clk_gpu at 900mhz, only marginal gain, so not sure it is worth the trouble to change to pvtpll on rk3576

12:09 <CounterPillow> Kwiboo: thanks! I'll check on my side tomorrow. This may highly depend on silicon quality, so having n>1 for testers would be great

12:45 <Kwiboo> CounterPillow: thanks and yes, more testing is always good :-), for rk3588 there was a big gain due to the normal pll rates that was being used, for rk3576 the main diff would probably only be stability related to silicon quality

12:47 <Kwiboo> also during my testing on rk3576 there was not much gain from 700mhz to 800mhz to 900mhz, testing with glmark2-es2-gbm (latest master commit), not sure what that is about

12:49 <Kwiboo> on rk3588 the glmark2 result was more linear related to the rate, and on rk3576 it felt more linear for 200-700, and after that it seemed to flatten out

12:52 <CounterPillow> possibly memory bandwidth bound, the rk3576 has a narrower memory bus

13:52 <diederik> Kwiboo: Do you have a good example to look at for switching from snps,reset-* to reset properties in the Ethernet PHY node?

14:20 <diederik> nvm ... you already described it further down the email :-D