CompatHelper: bump compat for "SIMDPirates" to "0.6"

[github-actions]

This pull request changes the compat entry for the SIMDPirates package from 0.1,0.2,0.3,0.5 to 0.1,0.2,0.3,0.5, 0.6.

This keeps the compat entries for earlier versions.

Note: I have not tested your package with this new compat entry. It is your responsibility to make sure that your package tests pass before you merge this pull request.

[JeffreySarnoff]

@ffevotte I'm leaving this alone because it did not build. There is newer version of SIMDPirates 0.62, maybe that resolves it? @chriselrod

[chriselrod]

Thanks for pinging me, this is definitely related.
The bump from 0.5 to 0.6 was because of yet another API change. I hope things will be more stable now.
Comparison functions like vgreater used to return unsigned integers which were used as bit masks. I changed this, because it posed two problems:

1. It is hard to make them act like a Vec of Bools without committing type piracy. I needed to awkwardly work around operations like mask & b::Bool to make it act like you'd expect (i.e., I'd want mask == mask & true.
2. Unsigned integers don't carry their length. A UInt8 has to represent all masks of length <= 8.

This changed caused the test failure, because the masks are no longer acting like scalars in a broadcast. I think it's reasonable to make them behave in that way, so I'll do so and issue a new release -- unless you or someone else chimes in and suggests how/why making them broadcastable makes sense.
Normal scalar operations already work (elementwise) on Vecs, which act like discrete units. Individual operations, whether extracting elements, or gathering/scattering them to/from multiple memory locations are all slow. Your application is an example that favors this treatment.

After defining broadcastable(v::AbstractStructVec) = Ref(v):

Test Summary:      | Pass  Total
AccurateArithmetic |  180    180

size 10_000
  sum_oro  1.219 μs (0 allocations: 0 bytes)
  sum_kbn  1.451 μs (0 allocations: 0 bytes)

size 1_000_000
  sum_oro  350.304 μs (0 allocations: 0 bytes)
  sum_kbn  348.766 μs (0 allocations: 0 bytes)

size 100_000_000
  sum_oro  57.978 ms (0 allocations: 0 bytes)
  sum_kbn  54.237 ms (0 allocations: 0 bytes)
  sum      100.290 ms (0 allocations: 0 bytes)

  dot_oro  122.701 ms (0 allocations: 0 bytes)
  dot_kbn  107.203 ms (0 allocations: 0 bytes)
  dot      19.043 ms (0 allocations: 0 bytes)
    Testing AccurateArithmetic tests passed

So tests should pass with SIMDPirates v0.6.3+, which was just released.

Is there any convenient infrastructure for running the test suites of packages that depend on your own before issuing a release?

[ffevotte]

@chriselrod Yes it looks like 0.6.3 passes the tests, thanks!

I'm not aware of any infrastructure for testing reverse dependencies. However, I think that is more or less what Julia itself does: test all (registered) packages before every release. Maybe the project has some infrastructure that could at least in part be reused ? But I wouldn't know who to ask.