Specifying a set of algorithms

General Usage
2

I’ve been thinking about the same thing recently.

If you know that each alg has an in-place implementation, you might want to preallocate your result before the loop.

using BenchmarkTools
using LinearAlgebra

# Original
A(x) = 2x
B(x) = 3x
C(x) = 4x

function f(algorithms, x)
    for alg in algorithms
        x = alg(x)
    end

    return x
end

println("Original:")
@btime f($(A, B, C, A, B), $[1.5, 3.0])


# In-place
A!(x) = lmul!(2,x)
B!(x) = lmul!(3,x)
C!(x) = lmul!(4,x)

function finplace(algorithms, x)
    # Allocates once
    xwork = copy(x)
    
    for alg in algorithms
        alg(xwork)
    end

    return xwork
end

println("In-place:")
@btime finplace($(A!, B!, C!, A!, B!), $[1.5, 3.0])

This prints out

Original:
373.892 ns (5 allocations: 480 bytes)
In-place:
193.564 ns (1 allocation: 96 bytes)

Perhaps you have a MWE where this assumption fails?

1 Like