GPU memory movement optimisation between Gauge actions derive -> Staple / Cshift

[qiUip]

There are lot of not understood memory movements around the Cshifts called from Staple (CovShift) which we need to investigate (see profile).

Concerns are around:

1. Small Host to Device memory transfer in between loops over the dimensions (small green lines between the 4 large red / green blocks).
2. The large transfers at the start / end of each dim loop.

• Investigate the origins of the small memory transfers from Staple to Cshift #21
• Investigate using autoView directly to pass the GaugeField to Staple. #20

See notes from 17th of Jan and 31st of Jan.