inline OffsetArray constructors

[johnnychen94]

To reduce the overhead

julia> a = zeros(5,5,5,5);

julia> @btime OffsetArray($a, -2:2, -2:2, -2:2, -2:2);
  3.088 ns (0 allocations: 0 bytes) # this PR; 1.5.2
  2.463 ns (0 allocations: 0 bytes) # this PR; 1.6-dev
  16.180 ns (0 allocations: 0 bytes) # master; 1.5.2
  15.693 ns (0 allocations: 0 bytes) # master; 1.6-dev

[codecov]

Codecov Report

Merging #164 into master will decrease coverage by 0.26%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #164      +/-   ##
==========================================
- Coverage   99.21%   98.95%   -0.27%     
==========================================
  Files           4        8       +4     
  Lines         256      477     +221     
==========================================
+ Hits          254      472     +218     
- Misses          2        5       +3     

Impacted Files	Coverage Δ
src/OffsetArrays.jl	99.44% <100.00%> (-0.01%)	⬇️
D:/a/OffsetArrays.jl/OffsetArrays.jl/src/utils.jl	100.00% <0.00%> (ø)
D:/a/OffsetArrays.jl/OffsetArrays.jl/src/origin.jl	100.00% <0.00%> (ø)
D:/a/OffsetArrays.jl/OffsetArrays.jl/src/axes.jl	97.56% <0.00%> (ø)
...ffsetArrays.jl/OffsetArrays.jl/src/OffsetArrays.jl	98.70% <0.00%> (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 08f969b...ea29ae7. Read the comment docs.

[timholy]

It doesn't get inlined now due to length(::UnitRange{Int}) having two checked-arithmetic operations. (See Base.print_statement_costs on 1.6.) This seems reasonable, and gives Julia a chance to elide the wrapper in the caller.

[fredrikekre]

It seems like this would allocate an array, is that completely elided?

[timholy]

Good thought. Really this should be a foreach call.

[johnnychen94]

Hmmm, turns out map is faster here.

• map: 3.004ns
• foreach: 5.261 ns

Is there a way to check if a method is properly inlined?

[timholy]

foreach doesn't force-specialize on the function. It could probably use some love...

I would guess, though, that it's elided. It should be creating a tuple (not an array) and that should get elided.

Is there a way to check if a method is properly inlined?

@code_typed or Cthulhu

[fredrikekre]

I would guess, though, that it's elided. It should be creating a tuple

Yea looks from the benchmarks that it is elided, it's just not immediatly obvious from the code (I didnt't think about the inputs beeing tuples). map is probably fine then (with a comment?) although something like foreach would be more typical when you don't care about the result I think.

[johnnychen94]

map on tuple can be optimized by the compiler and thus not using foreach here

This is what I added. Unsure what's the terminology here, "constant folding"?

[timholy]

You could just say "optimized away." Another option would be DCE.