# L-17 MCS 572 Fri 4 Oct 2024 : gpuadd1metal.jl

# The CUDA version is copied from the tutorial at
# https://cuda.juliagpu.org/stable/tutorials/introduction/
# and adjusted for use on an M1 MacBook Air with Metal

using Metal
using Test

function gpu_add1!(y, x)
    for i = 1:length(y)
        @inbounds y[i] += x[i]
    end
    return nothing
end

N = 32
x_d = Metal.fill(1.0f0, N)  # filled with Float32 1.0 on GPU
y_d = Metal.fill(2.0f0, N)  # filled with Float32 2.0

# run with N threads

@metal threads=N gpu_add1!(y_d, x_d)
result = (@test all(Array(y_d) .== 3.0f0))

println(result)