Int64 not supported on AMD GPUs? about metal.jl HOT 5 CLOSED

Int64 and Float64 are not supported and lead to those crashes. Try to convert the data to single-precision

from metal.jl.

b = convert(Array{Int32}, a)
1-element Vector{Int32}:
 1

julia> b .+ 1
1-element Vector{Int64}:
 2

from metal.jl.

b = convert(Array{Int32}, a)
1-element Vector{Int32}:
1

julia> b .+ 1
1-element Vector{Int64}:
2

... and? Of course CPU vectors support int64 addition, it's the GPU that's more limited here.

@PhilipVinc Int64 seems to work fine on my M1 though, and I can't imagine it not working since every literal in Julia is a 64-bit integer.

from metal.jl.

julia> using Metal

julia> a = MtlArray([1])
1-element MtlArray{Int64, 1}:
 1

julia> b = convert(Array{Float32}, a)
1-element Vector{Float32}:
 1.0

julia> b .+ 1
1-element Vector{Float32}:
 2.0

from metal.jl.

Again, what are you trying to prove here? You're still broadcasting over a CPU vector. Please take the time to add some explanation, and not just dump code snippets; this is not helpful.

from metal.jl.