add multiplication with vectors method

[dkarrasch]

This seems to fix an oversight. Previously, there was no specialized *(::BlockDiagonal,::AbstractVector) method, such that this combination was handled by some generic LinearAlgebra method. Not surprisingly, this yields a giant performance gain.

I have kept the changes simple and minimal. There is quite some code duplication with the corresponding *(::BlockDiagonal,::AbstractMatrix) method, which can be reduced by means of a helper function and using selectdim(y, 1, st:ed) if desired.

[codecov]

Codecov Report

Merging #32 into master will increase coverage by 0.13%.
The diff coverage is 100%.

@@            Coverage Diff            @@
##           master     #32      +/-   ##
=========================================
+ Coverage   94.77%   94.9%   +0.13%     
=========================================
  Files           4       4              
  Lines         153     157       +4     
=========================================
+ Hits          145     149       +4     
  Misses          8       8

Impacted Files	Coverage Δ
src/base_maths.jl	100% <100%> (ø)	⬆️

---

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 891ae0d...ddcc05f. Read the comment docs.

[nickrobinson251]

🎉 Thanks a lot! LGTM

There is quite some code duplication with the corresponding *(::BlockDiagonal,::AbstractMatrix) method, which can be reduced by means of a helper function and using selectdim(y, 1, st:ed) if desired.

That'd be a great follow-up PR, if you have time for that?

For this PR, would you mind posting a simple benchmark giving a quick illustration of the speed-up (it'd be helpful when one day making proper benchmarks for the package #19)

And please bump the package version -- will tag a new release straight away!

[dkarrasch]

Here is a quick benchmark, slightly adopted from the test suite:

using BlockDiagonals, BenchmarkTools
N1, N2, N3 = 30, 40, 50
N = N1 + N2 + N3
b1 = BlockDiagonal([rand(N1, N1), rand(N2, N2), rand(N3, N3)])
a = rand(N)
@benchmark $b1*$a

Master:

BenchmarkTools.Trial: 
  memory estimate:  4.41 MiB
  allocs estimate:  43921
  --------------
  minimum time:     1.702 ms (0.00% GC)
  median time:      1.852 ms (0.00% GC)
  mean time:        2.195 ms (11.26% GC)
  maximum time:     7.055 ms (44.17% GC)
  --------------
  samples:          2270
  evals/sample:     1

This PR:

BenchmarkTools.Trial: 
  memory estimate:  2.63 KiB
  allocs estimate:  9
  --------------
  minimum time:     1.121 μs (0.00% GC)
  median time:      1.227 μs (0.00% GC)
  mean time:        1.633 μs (4.68% GC)
  maximum time:     180.318 μs (97.38% GC)
  --------------
  samples:          10000
  evals/sample:     10

[nickrobinson251]

This is great! Thanks a lot. One tiny style comment, then good to go.