Type-stable KroneckerPower

[jishnub]

This PR removes the exponent from the type-domain of a KroneckerPower, relying on runtime checks instead. This improves performance significantly, even though getmatrices loses type-stability. Fixes #101.

On master

julia> A = collect(reshape(1:16, 4, 4));

julia> K1 = kronecker(A, 3);

julia> @btime $K1[1,1]
  1.730 μs (10 allocations: 320 bytes)
1

This PR

julia> @btime $K1[1,1]
  97.187 ns (1 allocation: 32 bytes)
1

[codecov-commenter]

Codecov Report

Merging #102 (a1cf1cb) into master (08c40e2) will decrease coverage by 0.86%. The diff coverage is 85.71%.

@@            Coverage Diff             @@
##           master     #102      +/-   ##
==========================================
- Coverage   90.95%   90.09%   -0.87%     
==========================================
  Files          10       10              
  Lines         619      646      +27     
==========================================
+ Hits          563      582      +19     
- Misses         56       64       +8     

Impacted Files	Coverage Δ
src/kroneckerpowers.jl	93.18% <85.71%> (+1.87%)	:arrow_up:
src/indexedkroncker.jl	77.27% <0.00%> (-7.54%)	:arrow_down:
src/base.jl	92.85% <0.00%> (-0.58%)	:arrow_down:
src/multiply.jl	100.00% <0.00%> (ø)
src/kroneckergraphs.jl	100.00% <0.00%> (ø)
src/kroneckersum.jl	84.09% <0.00%> (+0.18%)	:arrow_up:
src/vectrick.jl	94.69% <0.00%> (+0.20%)	:arrow_up:

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 08c40e2...a1cf1cb. Read the comment docs.

[MichielStock]

Nice improvement. When testing this branch, I noticed products between KroneckerPower and KroneckerProduct do not (always) work. I fixed this in #103. OK if I close this PR and merge your changes via the other branch?

[jishnub]

Yes sure!