The function mapreduce(f, op, X; dims = dim) can be used, but it seems that the parallel analog pmapreduce from the package ParallelUtilities does not support the dims argument.
How would one use simple parallel functions to perform the same operation and still get a big speed up?
Especially when X is of dims (very very big, big, very big) (say (10^5, 10^2, 10^3)) and the reduction happens on the first (largest) dimension. f are about 400ns and op are typically +.
Example of what I am doing
N = 10^5
n = 100
M = 1000
X = fill(rand(), N, M, n)
x = mapreduce(r -> sin(r, a, b), +, X, dims = 1)
# * What I want * #
using ParallelUtilities
x = pmapreduce(r -> sin(r, a, b), +, X, dims = 1)