# L-22 MCS 572 Wed 16 Oct 2024 : gpusum32cuda.jl

# Summing 32 numbers with 32 threads,
# using loop enrolling, with a different order.

using CUDA

"""
    function gpusum32!(a)

sums the 32 numbers in the array a.
On return a[1] contains the sum.
"""
function gpusum32!(a)
    i = threadIdx().x
    a[i] += a[i+16]
    a[i] += a[i+8]
    a[i] += a[i+4]
    a[i] += a[i+2]
    a[i] += a[i+1]
    return nothing
end

a_h = [convert(Float32, k) for k=1:32]
z_h = [0.0f0 for k=1:32] # padding with zeros
x_h = vcat(a_h, z_h)
println("the numbers to sum : ", x_h)
x_d = CuArray(x_h) 

# run with 32 threads

@cuda threads=32 gpusum32!(x_d)

println("the summed numbers : ", Array(x_d))