Broadcast inference failure results in scalar iteration

[dpsanders]

Describe the bug
Broadcast failure with a CuArray{Interval{Float32}}, even though it works fine with Interval{Float64}.

To Reproduce
MWE for this bug:

julia> using IntervalArithmetic, CuArrays

julia> a = Interval{Float32}[1, 2, 3];

julia> a2 = CuArray(a);

julia> f(x) = x^2 - 2
f (generic function with 1 method)

julia> f.(a2)
ERROR: scalar getindex is disallowed
Stacktrace:
 [1] error(::String) at ./error.jl:33
 [2] assertscalar at /home/dpsanders/.julia/packages/GPUArrays/t8tJB/src/indexing.jl:8 [inlined]
 [3] getindex(::CuArray{Interval{Float32},1}, ::Int64) at /home/dpsanders/.julia/packages/GPUArrays/t8tJB/src/indexing.jl:44
 [4] _broadcast_getindex at ./broadcast.jl:544 [inlined]
 [5] _getindex at ./broadcast.jl:575 [inlined]
 [6] _broadcast_getindex at ./broadcast.jl:550 [inlined]
 [7] getindex at ./broadcast.jl:511 [inlined]
 [8] copy at ./broadcast.jl:787 [inlined]
 [9] materialize(::Base.Broadcast.Broadcasted{Base.Broadcast.ArrayStyle{CuArray},Nothing,typeof(f),Tuple{CuArray{Interval{Float32},1}}}) at ./broadcast.jl:753
 [10] top-level scope at none:0

julia> a = Interval{Float64}[1, 2, 3];

julia> a2 = CuArray(a);

julia> f.(a2)
3-element CuArray{Interval{Float64},1}:
 [-1, -1]
   [2, 2]
   [7, 7]

Expected behavior
The same as for 64-bit intervals.

Build log

# post the output of Pkg.build()
# make sure the error still reproduces after that.

Environment details (please complete this section)
Details on Julia:

julia> versioninfo()
Julia Version 1.1.0
Commit 80516ca202 (2019-01-21 21:24 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Xeon(R) CPU E5-2603 v4 @ 1.70GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-6.0.1 (ORCJIT, broadwell)

Julia packages:

• CuArrays.jl: 0.10.0
• CUDAnative.jl: v1.0.1+
• CUDAdrv v1.0.1+ #master

CUDA: toolkit and driver version
(Don't know how to find this out)

[vchuravy]

This looks like a type-inference issue since we are hitting the generic fallback, this indicates that the return type for f is abstract.

[dpsanders]

Apologies, you're right:

julia> using IntervalArithmetic

julia> x = Interval{Float32}(3, 4)
[3, 4]

julia> @code_warntype x^2
Body::Union{Interval{Float32}, Interval{Float64}}
...

Is it possible to make the error message more explicit about this?