Use propagate_inbounds in AbstractArray iterator

[haampie]

I'm sure this was removed accidentally at some point.

julia> function simple_sum(V)
         s = 0.0
         @fastmath @inbounds for x in V
           s += x
         end
         s
       end

# Current master
julia> @btime simple_sum(v) setup = (v = rand(500))
  51.872 ns (0 allocations: 0 bytes)
251.22882057537214

# Current master
julia> @btime simple_sum(v) setup = (v = view(rand(500, 20), :, 1))
  553.941 ns (0 allocations: 0 bytes)
244.35868876917397

# This pr
julia> @btime simple_sum(v) setup = (v = view(rand(500, 20), :, 1))
  52.354 ns (0 allocations: 0 bytes)
254.29673869300711

[Keno]

This seems fine, but is a bit odd, since iterate is not getindex. Perhaps we're just missing something along the lines of #27386 for views?

[haampie]

I agree, but #27384 (comment) argues that something similar might be impossible. Also #25743 shows that a specialized iterator for views will probably not be safe.

[ararslan]

@nanosoldier runbenchmarks(ALL, vs=":master")

[haampie]

Seems like nanosoldier is hanging?

[ararslan]

Yeah it's definitely taking longer than normal, but the log suggests that it's actually still going. I'll give it a bit and kill it if doesn't respond in the next couple hours.

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan

[haampie]

Can't really explain this report 🔮

[KristofferC]

@nanosoldier runbenchmarks(ALL, vs=":master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan