Scalar static vector multiplication triggers `unsupported call to the Julia runtime`

[mohamed82008]

The following code triggers unsupported call to the Julia runtime (call to jl_invoke).

using CUDAnative, CUDAdrv, CuArrays, StaticArrays

k = 3
n = 10

Ks = [SMatrix{k, k, Float64}(rand(k,k)) for i in 1:n] |> CuArrays.CuArray
fs = [SVector{k, Float64}(rand(k)) for i in 1:n] |> CuArrays.CuArray
newfs = similar(fs)

dev = CUDAdrv.device()
ctx = CUDAdrv.CuContext(dev)
MAX_THREADS_PER_BLOCK = CUDAdrv.attribute(dev, CUDAdrv.MAX_THREADS_PER_BLOCK)
threads = min(n, MAX_THREADS_PER_BLOCK)
blocks = ceil.(Int, n / threads)

function kernel(newfs, Ks, fs::AbstractVector{TV}, n) where {N, T, TV<:SVector{N,T}}
    i = (blockIdx().x-1) * blockDim().x + threadIdx().x
    m = 2.0
    if i <= n
	    newfs[i] = m * (Ks[i] * fs[i])
	    #newfs[i] = SVector{1,T}((m,)) .* (Ks[i] * fs[i])
    end
    return
end

@cuda blocks=blocks threads=threads kernel(newfs,  Ks, fs, n)

Replacing the multiplication line with the commented one works fine.

[maleadt]

Works here. Julia 1.0.2, StaticArrays v0.9.2, GPU packages from master.

[mohamed82008]

I can confirm that it works with Julia 1.0.1, but it fails in Julia master from 19/11.

julia> versioninfo()
Julia Version 1.1.0-DEV.681
Commit 8a4f20b887 (2018-11-19 01:50 UTC)
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: Intel(R) Core(TM) i7-6700 CPU @ 3.40GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-6.0.1 (ORCJIT, skylake)

[maleadt]

Bisected to JuliaLang/julia@9e98386 (cc @vtjnash). Care to reduce this and file an upstream issue?

[maleadt]

Fixed upstream.