mapreduce/mapfoldl unrolls fully, failing to compile for large static sizes

[chengchingwen]

MWE:

julia> x = SizedVector{30000}(randn(30000));

julia> 0.5 in x
ERROR: syntax: invalid syntax (memory-error out of gc handles)
Stacktrace:
 [1] top-level scope
   @ REPL[4]:1
 [2] _mapreduce
   @ ~/.julia/packages/StaticArrays/58yy1/src/mapreduce.jl:113 [inlined]
 [3] in(x::Float64, a::SizedVector{30000, Float64, Vector{Float64}})
   @ StaticArrays ~/.julia/packages/StaticArrays/58yy1/src/mapreduce.jl:267
 [4] top-level scope
   @ REPL[4]:1
 [5] top-level scope

[thchr]

This happens because Base.in is forwarded to _mapreduce, which in turn is forwarded to _mapfoldl which is fully unrolled without any size cut-off. I.e., the current implementation does a full unroll of the loop so the code size is out of hand for S = 30000.

The broader problem is discussed in #439.

[chengchingwen]

I see. BTW, I turn into using findfirst as a replacement and it doesn't seem to be forwarded to _mapreduce. I'm curious about would these "find" function also benefit from unrolling?

[mateuszbaran]

When you have more than 1000 elements in an array, then I doubt there is any function in StaticArrays.jl that would be actually faster than the plain Array implementation. Are you sure you actually need SizedVector?

[chengchingwen]

I use SizedVector merely for having the length value as type parameter and avoiding push!/pop! operations at the surface level. Maybe there is a better choice for this need?

[mateuszbaran]

SizedArray would be fine if you accessed the data field, for example 0.5 in x.data instead of 0.5 in x. I don't really know any other types that would work better.