Compilation error with type-converting round functions, e.g., ceil(Int32, 1.2f0)

Created by: cdsousa

Some time ago, CUDA.jl won't work with calls to rounding functions that perform conversion to integer types. Currently, it already works (I don't know why since those versions are still not defined https://github.com/JuliaGPU/CUDA.jl/blob/4c9a03d6202afe816707ccd12ed22acfa65a3df5/src/device/intrinsics/math.jl#L243)

However, KernelAbstractions.jl still doesn't work, but instead it gives a seemingly unrelated error:

using CUDA
using KernelAbstractions

x_d = CUDA.cu(rand(Float32, 1))
y_d = CUDA.zeros(Int32, 1)

f(x::Float32) = ceil(Int32, x)

## CUDA

function cu!(y, x)
    y[1] = f(x[1])
    return nothing
end

y_d .= 0
@cuda cu!(y_d, x_d)
y_d

## KA

function main(y_d)
    @kernel function ka!(y, x)
        y[1] = f(x[1]) 
    end
    event = ka!(CUDAKernels.CUDADevice(), 1)(y_d, x_d, ndrange=(1,))
    wait(event)
end

y_d .= 0
main(y_d)
y_d