[julia-users] huge performance hit if for-loops in separate function.

[Jürgen Bohnert]

Hi everyone,

I'm new to using julia and I would be grateful if anyone could shed some 
light on the following performance-issue that I cannot explain.
For some reason wrapping 2 for-loops into a separate function and then 
calling that function is 2 times slower than executing the for-loops 
directly.

*My goal:*
Perform timeconsuming but easily parallelizable operation '*operator*' over 
moderate/large parameter-space on a SINGLE Machine with multiple cores. -> 
Put loops into function and call '*pmap*' on that.

function operator(x::Float64, y::Float64, extraparameters::Float64 ...)
        # ... Does something
        # returns Float64 scalar
end

The function '*operator*' works like a charm and fast.

*What I did:*
Execution over a parameter-space:  *x::Array{Float64, 1}, y::Array{Float64, 
1}*
via the function '*loops_simple*' (below)

function loops_simple(different_arguments ...)

    # generate x::Array{Float64, 1} and y::Array{Float64, 1} and 
extraparameters
    # preallocate output container
    output = Array(Float64, (length(x), length(y)))

    t = time()

    for i=1:length(y)
        for j=1:length(x)
            output[j,i] = operator(x[j], y[i], extraparameters ...)
        end
    end

    t = time() - t
    println("time taken: $t")

    return output
end

This runs really fast as expected. Trouble comes as soon as I stuff the 
loops into a separate function '*wrapper*' (for easier parallel execution 
but I didn't get that far.).

function wrapper(x::Array{Float64, 1}, y::Array{Float64, 1}, extraparameters
::Float64 ...)
    output = Array(Float64, (length(x), length(y)))
    for i=1:length(y)
        for j=1:length(x)
            output[j,i] = operator (x[j], y[i], extraparameters ...)
        end
    end
end


Now calling:

function loops_wrapped(different_arguments ...)
    # generate x::Array{Float64, 1} and y::Array{Float64, 1} and 
extraparameters
    t = time()
    output = wrapper (x, y, extraparameters ...)
    t = time() - t
    println("time taken: $t")

    return output
end

the process ends up taking twice the time of the '*loops_simple*' case. Can 
anyone explain what I'm doing wrong?

Thanks in advance.

I'm running julia Version 0.4.0-dev+584; Commit 114abf4; x86_64-linux-gnu