# L-18 MCS 572 Mon 7 Oct 2024 : gpuadd3.jl

# Copied from the tutorial at
# https://cuda.juliagpu.org/stable/tutorials/introduction/

using CUDA
using Test

function gpu_add3!(y, x)
    index = (blockIdx().x - 1) * blockDim().x + threadIdx().x
    stride = gridDim().x * blockDim().x
    for i = index:stride:length(y)
        @inbounds y[i] += x[i]
    end
    return
end

N = 2^20
x_d = CUDA.fill(1.0f0, N)  # filled with Float32 1.0 on GPU
y_d = CUDA.fill(2.0f0, N)  # filled with Float32 2.0

# run with 256 threads per block

numblocks = ceil(Int, N/256)
@cuda threads=256 blocks=numblocks gpu_add3!(y_d, x_d)
result = (@test all(Array(y_d) .== 3.0f0))

println(result)