tree 1e53c589244c476715b848cb522f2328e5062f59
parent df71589e19fe6de33cdd03577f5ce807d0d129cd
author Ben Vanik <ben.vanik@gmail.com> 1685995574 -0700
committer GitHub <noreply@github.com> 1685995574 -0700
gpgsig -----BEGIN PGP SIGNATURE-----
 
 wsBcBAABCAAQBQJkfkA2CRBK7hj4Ov3rIwAANOYIAEIyOMWv3jlxpgdTgSSmBbWi
 CDEJ5YrzyCnagavRBrnvm7wge3qEEsuV+31Kj7vvekUWuWoH2AV+o7rXTLfRbKjG
 FQsrdife0h1FkLxgZKd6T7A9dbCFEv3kALrO0wEqmRriWuP88bCoKBHSlOm/Lr/Y
 Ds9lKtXgAbNBNPDrPgZ6YKVYIfLPI3ZT8R1SCLOg5cOUIw/zdtPrP2IO5WUPmCN3
 ZpSq76/WWlsVrE6jeVivMRhz+RzWv8SsnaDk6GCmYzBOY0Sh0vBryEVRZ3k4OYJp
 70mCyP7yz2/SR6rrbl6G93Pxwr21wwalC7KpiGju3VKAbqqxEsNj9wouzDJ5Ffs=
 =UNyq
 -----END PGP SIGNATURE-----
 

Adding task system utilization tracing. (#13941)

Adds one new plot per executor in a process showing a 0-100% utilization
value as tracked by the workers (vs the CPU usage sampled a much lower
rate).


![image](https://github.com/openxla/iree/assets/75337/7fc353d2-3fe8-4e33-8b60-056751231381)

Inspecting the area under/over the curve is an easy way to spot
dispatches that could have improved distribution:

![image](https://github.com/openxla/iree/assets/75337/1383ad5d-c802-43ce-9352-e584eb040921)
(here 2x of the wall time is only 1x the utilization, meaning with
perfect distribution we could go 2x faster by 2xing the utilization)