TridiagSolver (local): embed row permutation in rank1 solver

[albestro]

Distributed and local implementations have slightly diverged in the last months.

This PR addresses the main (and probably the only ones) differences. With #831 distributed implementation got row permutations and setUnitDiag embedded in rank1 solution, removing the need of explicitly doing so just before doing the gemm. Local implementation was still doing the permutation separately.

With this PR both setUnitDiag and row permutation happen during rank1 solver also for the local implementation.

[albestro]

Note:

Permutation is done in a slightly different way due to different optimization requirements between local and distributed. In the latter one, in order to avoid additional (and unnecessary) communications, the permutation is done in the first steps of the rank1-solver, and all other steps have to take this into account. On the other hand, in the local implementation the permutation is done in the last step, since there is no reason to do it immediately, keeping rank1-solver "clean" from permutations till the very end.

This also introduces a difference for k == 2, for which the permutations should happen even if it is handled differently from other cases. Since in local implementation is done at the very end, the aforementioned case has to handled differently by performing an ad-hoc permutation just before the early exit of the special case. It must be highlighted that these special cases are not tested extensively for the moment, and I'm going to open an issue to keep track of this.

[rasolca]

cscs-ci run

[codecov-commenter]

Codecov Report

Merging #936 (e728046) into master (3d86f2b) will decrease coverage by 0.01%.
The diff coverage is 96.42%.

❗ Your organization is not using the GitHub App Integration. As a result you may experience degraded service beginning May 15th. Please install the Github App Integration for your organization. Read more.

@@            Coverage Diff             @@
##           master     #936      +/-   ##
==========================================
- Coverage   93.20%   93.19%   -0.01%     
==========================================
  Files         136      136              
  Lines        8327     8318       -9     
  Branches     1081     1083       +2     
==========================================
- Hits         7761     7752       -9     
  Misses        391      391              
  Partials      175      175              

Impacted Files	Coverage Δ
include/dlaf/eigensolver/tridiag_solver/kernels.h	100.00% <ø> (ø)
src/eigensolver/tridiag_solver/kernels.cpp	100.00% <ø> (ø)
src/eigensolver/tridiag_solver/kernels.cu	41.48% <ø> (-6.66%)	⬇️
include/dlaf/eigensolver/tridiag_solver/merge.h	99.81% <96.42%> (-0.19%)	⬇️

[albestro]

cscs-ci run

[albestro]

cscs-ci run