NVML: Add support for clock queries.

[maleadt]

julia> NVML.clock_event_reasons(dev)
(idle = false, display_setting = false, hw_power_brake = false, sw_power_cap = false, application_setting = false, hw_slow = false, sync_boost = false, sw_thermal = false, hw_thermal = false)

julia> NVML.clock_info(dev)
(sm = 915, graphics = 915, memory = 5000, video = 1500)

Can be useful to ensure the GPU isn't throttled while benchmarking. Couldn't trigger that locally though.

[Zentrik]

I've often seen my gpu (gtx 1070) throttle during benchmarking on Windows 10. Not sure I ever figured out what the problem was (interestingly you could workaround this by playing a video at the same time) but I just switched to Linux.

[maleadt]

Ah, I couldn't notice anything because I had locked clocks (using nvidia-smi). After a reboot I can see the idle event:

julia> NVML.clock_event_reasons(dev)
(idle = true, display_setting = false, hw_power_brake = false, sw_power_cap = false, application_setting = false, hw_slow = false, sync_boost = false, sw_thermal = false, hw_thermal = false)

[codecov]

Codecov Report

Attention: 1 lines in your changes are missing coverage. Please review.

Comparison is base (a63e062) 72.72% compared to head (0a80269) 72.85%.
Report is 1 commits behind head on master.

Files	Patch %	Lines
lib/nvml/device.jl	98.79%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2194      +/-   ##
==========================================
+ Coverage   72.72%   72.85%   +0.12%     
==========================================
  Files         159      159              
  Lines       14556    14623      +67     
==========================================
+ Hits        10586    10653      +67     
  Misses       3970     3970              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.