ReductionToBand, BTReductionToBand, TFactor: Taus on GPU

[albestro]

The target of this PR is moving TAUS from being stored on CPU to be stored on GPU memory.
It's more a cleanup than an optimisation.

Changelog:

• (trivially) some changes in the API (and test accordingly too)
• reduction to band still computes taus on CPU, but then it copies them immediately to GPU. This required allocation of an additional support matrix on CPU (i.e. ComputePanelHelper<GPU>::mat_taus_cpu).
• in TFactor a couple of copies have been skipped: one becuase it was a leftover, the other because now having taus on GPU they can be used in-place without copying them in the diagonal of the TFactor tile.
• in TFactor Since CPU and GPU differs a bit in the implementation, it is now explicit, by making them private, that loop_gemv and loop_trmv are internal functions. These clarifies that these functions might have different signatures between the two different backends. Actually, in GPU version of loop_gemv, taus parameter was superfluous so it has been dropped now.

TODO

• Run benchmarks

[albestro]

cscs-ci run

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 95.07%. Comparing base (7ce9103) to head (4f2b707).

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1304   +/-   ##
=======================================
  Coverage   95.07%   95.07%           
=======================================
  Files         141      141           
  Lines        8654     8656    +2     
  Branches     1110     1110           
=======================================
+ Hits         8228     8230    +2     
  Misses        239      239           
  Partials      187      187           

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[msimberg]

Any idea about performance changes? Should have no effect, but have you tested if that's the case?

[albestro]

Any idea about performance changes? Should have no effect, but have you tested if that's the case?

It's a good point. I still have it in my plan, but before doing that I'm double checking HMM Heterogenous Memory Management 😉 Let me add a todo/reminder in the PR 😉