Specialize lmul!/rmul! for strided triangular matrices

[jishnub]

This improves performance, as we only need to loop over one half of the matrix. On master, rmul! for an UpperTriangular forwards the multiplication to the parent, so the entire array is looped over. We therefore obtain identical performance for rmul!(::Matrix, ::Diagonal) and rmul!(::UpperTriangular{<:Any, <:Matrix}, ::Diagonal).

julia> using LinearAlgebra, Chairmarks

julia> D = Diagonal(rand(4000));

julia> A = rand(size(D)...);

julia> U = UpperTriangular(A);

julia> @b (A, D) rmul!(_[1], _[2])
10.309 ms

julia> @b (U, D) rmul!(_[1], _[2])
10.370 ms

On this PR, the latter is faster, as the loop is only over half the indices:

julia> @b (U, D) rmul!(_[1], _[2])
7.216 ms

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 91.98%. Comparing base (508e77f) to head (b5473ee).
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1228      +/-   ##
==========================================
+ Coverage   91.97%   91.98%   +0.01%     
==========================================
  Files          34       34              
  Lines       15400    15430      +30     
==========================================
+ Hits        14164    14194      +30     
  Misses       1236     1236              

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