Last edited 5 months ago

How to monitor the GCNANO GPU load: Difference between revisions

Visual
deleted>Frq08942
mNo edit summary
 
Registered User
mNo edit summary
Tag: 2017 source edit
 
(11 intermediate revisions by 3 users not shown)
Line 1: Line 1:
<noinclude>
<noinclude>
{{ArticleMainWriter | PhilippeC }}
{{ApplicableFor
{{ReviewersList | Pierre-YvesM }}
|MPUs list=STM32MP15x, STM32MP23x, STM32MP25x
{{ArticleApprovedVersion | PhilippeC | Pierre-YvesM | No previous approved version | PhilipS - 15Jan'19 - 10256 | 15Jan'19}}
|MPUs checklist=STM32MP13x, STM32MP15x, STM32MP21x, STM32MP23x, STM32MP25x
 
}}
[[Category:GPU]]
[[Category:How to trace and debug]]
</noinclude>
</noinclude>
 
==Basic usage==
When a GPU animation is running on the display, the related GCNANO estimated GPU load can be monitored from the [[Vivante GCNANO GPU overview | GCNANO driver]] level, by using the following command:
When a GPU-based animation is running on the display, the related GCNANO estimated GPU load can be monitored from the GCNANO driver level, by using the following command:
  {{Board$}} (while true; do (tr -d '\n' < /sys/kernel/debug/gc/idle | awk -F" " '{printf("gpu load %.0f%%\n", $6*100/($4-$2));}'); sleep 4; done) &
  {{board$}}(while true; do \
gpu1=$(cat /sys/kernel/debug/gc/idle); \
sleep 4; \
gpu2=$(cat /sys/kernel/debug/gc/idle); \
echo $gpu1 $gpu2 | tr -d '\n' | tr -d ',' | tr -d 'ns' | awk -F" " '{printf("gpu load %.0f%%\n", ($10-$2)*100/($10+$12+$14+$16-$2-$4-$6-$8))}'; \
done) &


The GCNANO estimated GPU load is then periodically output in the user console as a percentage "%":
The GCNANO estimated GPU load is then periodically output in the user console as a percentage "%":
Line 18: Line 21:
</pre>
</pre>


Notes:
== Tips ==
* Stop monitoring the GPU load with the command "kill -9 `ps -o ppid= -C sleep`".
* Stop monitoring the GPU load with the command <code>kill -9 `ps -o ppid= -C sleep`</code>. This command may not work depending on your system configuration (busybox ps features, sh shell configuration...). Alternative method is to use the command <code>fg</code> followed by <code>CTRL+C</code> in your terminal.
* Adjust the GPU load update period by modifying the "sleep" value (4 seconds in the example).
* {{Highlight|Debugfs<ref>[[Debugfs]]</ref> configuration needs to be enabled}} in order to be able to monitor the GPU load from the debug directory "/sys/kernel/debug/gc".
* Use the command "dmesg -n8" to mix both user and kernel console outputs.
* Adjust the GPU load update period by modifying the "sleep" value (4 seconds in the above examples).
* [[Debugfs]] configuration needs to be enabled.
* Use the dmesg<ref>[[Dmesg and Linux kernel log]]</ref> command <code>dmesg -n8</code> to mix both user and kernel console outputs.
* The detailed calculation is: GPU load = On/(End-Start) with "On, End & Start" coming from the command:
* The detailed calculation is: <code>GPU load = (On-previous_On) / (Total-previous_Total)</code> with <code>Total=On+Off+Idle+Suspend</code>, all variables coming from the command:
  {{Board$}} cat /sys/kernel/debug/gc/idle
  {{board$}}cat /sys/kernel/debug/gc/idle
  Start:   3488941301162 ns
  On:               2,071,009,284,477 ns
  End:     3549591148649 ns
  Off:             11,480,071,864,263 ns
  On:     26672154276 ns
  Idle:                             0 ns
  Off:     0 ns
  Suspend:           1,242,043,838,898 ns
  ...
  ...
==References==
<references/>
<noinclude>
{{PublicationRequestId | 10256 | 2019-01-15 | PhilipS}}
[[Category:GPU]]
[[Category:How to trace and debug]]
</noinclude>

Latest revision as of 09:59, 8 October 2024↑

Applicable for STM32MP15x lines, STM32MP23x lines, STM32MP25x lines

1. Basic usage[edit | edit source]↑

When a GPU-based animation is running on the display, the related GCNANO estimated GPU load can be monitored from the GCNANO driver level, by using the following command: 

Template:Board$(while true; do \
gpu1=$(cat /sys/kernel/debug/gc/idle); \
sleep 4; \
gpu2=$(cat /sys/kernel/debug/gc/idle); \
echo $gpu1 $gpu2 | tr -d '\n' | tr -d ',' | tr -d 'ns' | awk -F" " '{printf("gpu load %.0f%%\n", ($10-$2)*100/($10+$12+$14+$16-$2-$4-$6-$8))}'; \
done) &

The GCNANO estimated GPU load is then periodically output in the user console as a percentage "%": 

gpu load 75%
gpu load 75%
gpu load 75%

2. Tips[edit | edit source]↑

  • Stop monitoring the GPU load with the command kill -9 `ps -o ppid= -C sleep`. This command may not work depending on your system configuration (busybox ps features, sh shell configuration...). Alternative method is to use the command fg followed by CTRL+C in your terminal.
  • Debugfs[1] configuration needs to be enabled in order to be able to monitor the GPU load from the debug directory "/sys/kernel/debug/gc".
  • Adjust the GPU load update period by modifying the "sleep" value (4 seconds in the above examples).
  • Use the dmesg[2] command dmesg -n8 to mix both user and kernel console outputs.
  • The detailed calculation is: GPU load = (On-previous_On) / (Total-previous_Total) with Total=On+Off+Idle+Suspend, all variables coming from the command:
Template:Board$cat /sys/kernel/debug/gc/idle
On:                2,071,009,284,477 ns
Off:              11,480,071,864,263 ns
Idle:                              0 ns
Suspend:           1,242,043,838,898 ns
...

3. References[edit | edit source]↑

  1. ↑ Debugfs
  2. ↑ Dmesg and Linux kernel log



Graphics processing units

Retrieved from "https://wiki.st.com/stm32mpu/index.php?title=How_to_monitor_the_GCNANO_GPU_load&oldid=108295"