redundant gauge force?

[clarkedavida]

in gaugeActionDeriv.h: there is a claim symanzikGaugeActionDeriv is slower than gauge_force?. it seems like the only difference is one uses gsiteMu while the other uses both gsite and mu...

dennis suggests: you can write a short test, trace and profile and remove the slower one. If there is a perf difference, that might give us a hint towards our indexing issues. For example: on MI250x we are just using slightly too many registers to have perfect occupancy but enforcing the proper registers per thread then gives spilling. We looked at the device code and the register spills come from indexing. If that same issue is present in your case, that might give us a hint how to fix that