Closed
Description
openedon Sep 28, 2014
I've found a case where compiling a function that calls a stagedfunction fails, due to a failure in type inference. If one checks out #8501 (and preferably, inserts a println(P) here), here's a simple way to reproduce:
function as_sub(x::AbstractMatrix)
y = similar(x, eltype(x), tuple(([size(x)...]*2)...))
@show typeof(y)
y = sub(y, 2:2:size(y,1), 2:2:size(y,2))
for j=1:size(x,2)
for i=1:size(x,1)
y[i,j] = x[i,j]
end
end
y
end
as_sub(eye(2))This gives the following output:
julia> as_sub(eye(2))
Array{Float64,N}
Array{Float64,N}
Array{Float64,N}
ERROR: `ndims` has no method matching ndims(::Type{Any})
in ndims at ./abstractarray.jl:22 (repeats 3 times)
in setindex! at ./subarray2.jl:35One can see that it fails at compile-time, not runtime: that @show typeof(y) never gets called.
A modified version of the function (which declares the type of y):
function as_sub1(x::AbstractMatrix)
y = similar(x, eltype(x), tuple(([size(x)...]*2)...))::typeof(x)
@show typeof(y)
y = sub(y, 2:2:size(y,1), 2:2:size(y,2))
for j=1:size(x,2)
for i=1:size(x,1)
y[i,j] = x[i,j]
end
end
y
endyields this:
julia> as_sub1(eye(2))
Array{Float64,2}
Array{Float64,2}
typeof(y) = Array{Float64,2}
2x2 SubArray{Float64,2,Array{Float64,2},(StepRange{Int64,Int64},StepRange{Int64,Int64})}:
1.0 0.0
0.0 1.0Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment