Generalize from Array to DenseArray where possible in stridelayout

[brenhinkeller]

This would help enable LoopVectorization of custom array types that subtype <: DenseArray <: AbstractArray. Since DenseArray is defined to be column-major

DenseArray is an abstract subtype of AbstractArray intended to include all arrays where elements are stored contiguously in column-major order

this should be kosher AFAIU.

[ChrisRackauckas]

https://github.com/JuliaArrays/ArrayInterface.jl/runs/5182799084?check_suite_focus=true#step:6:213
https://github.com/JuliaArrays/ArrayInterface.jl/runs/5182799146?check_suite_focus=true#step:6:172

@Tokazama We should revert on Friday if this all doesn't get fixed. Downstream shouldn't be failing for more than a week or two without a downstream PR.

[chriselrod]

I think this makes sense, but as this is expressly for LoopVectorization support, I'd like to see
JuliaSIMD/LoopVectorization.jl#383
merged first so we can run downstream tests on it.

But LV could also depend on this and explicitly test and document a simple API for adding LV support to custom array types.

[codecov]

Codecov Report

Merging #243 (e1cf4dc) into master (b2304bd) will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #243   +/-   ##
=======================================
  Coverage   89.14%   89.14%           
=======================================
  Files          11       11           
  Lines        1751     1751           
=======================================
  Hits         1561     1561           
  Misses        190      190           

Impacted Files	Coverage Δ
src/stridelayout.jl	87.18% <100.00%> (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 b2304bd...e1cf4dc. Read the comment docs.

[brenhinkeller]

No rush on my end! This'll ultimately hopefully support LV of MallocArrays from StaticTools.jl but I've got lots of other things to worry about in the meanwhile, so can wait and see how JuliaSIMD/LoopVectorization.jl#383 / the downstream fixes play out first

[brenhinkeller]

Any progress on the downstream stuff lately @Tokazama? I'd try to help, but a lot of the LV internals are still beyond me

[Tokazama]

Sorry I've been recovering from COVID the last week and haven't been very active on GitHub. LoopVectorization has some odd failures that I don't understand why they occur with the change in question.

[brenhinkeller]

Is this potentially back on the table now? Looks like downstream tests are passing

[brenhinkeller]

Oh hmm, are these expected? It was all green before the rerun.. Seems like something is broken on Nightly but that might be unrelated? I can't see the logs though due to some permissions setting.

[Tokazama]

This is an inference issue we've ran into on nightly that I'm pretty sure is unrelated

[Tokazama]

@brenhinkeller, if you think this should be released in a new update then there needs to be an update to Project.toml too.

[brenhinkeller]

Version updated!