I am trying to benchmark an OpenCL code (github.com/.../mcxcl) over the Mali-G610 GPU on a recently purchased orange pi 5 max.
in my first attempt, when running the kernel for more than 5 second, it is killed with the following error
[80084.794346] mali fb000000.gpu: [5243519185] Iterator PROGRESS_TIMER timeout notification received for group 0 of ctx 578518_53 on slot 0[80084.795416] mali fb000000.gpu: Notify the event notification thread, forward progress timeout (2621440000 cycles)
[80084.794346] mali fb000000.gpu: [5243519185] Iterator PROGRESS_TIMER timeout notification received for group 0 of ctx 578518_53 on slot 0
[80084.795416] mali fb000000.gpu: Notify the event notification thread, forward progress timeout (2621440000 cycles)
after using echo 99999999999 > /sys/class/misc/mali0/device/progress_timeout command I found at this link, I was able to let my kernel run for over 20 seconds, however, when it runs more than 20 seconds, I am seeing a new error
[83338.111056] mali fb000000.gpu: Ctx 605122_59 Group 0 CSG 0 CSI: 0CS_FAULT.EXCEPTION_TYPE: 0x69 (RESOURCE_EVICTION_TIMEOUT)CS_FAULT.EXCEPTION_DATA: 0x0CS_FAULT_INFO.EXCEPTION_DATA: 0x1
[83338.111056] mali fb000000.gpu: Ctx 605122_59 Group 0 CSG 0 CSI: 0
CS_FAULT.EXCEPTION_TYPE: 0x69 (RESOURCE_EVICTION_TIMEOUT)
CS_FAULT.EXCEPTION_DATA: 0x0
CS_FAULT_INFO.EXCEPTION_DATA: 0x1
I could not find a fix for this, unless I kill the graphics using sudo init 3.
I am wondering what is the setting to allow an OpenCL to run longer period of time, I want to keep the graphics/desktop-environment, even the graphics is frozen.
thanks