Problems with Optimising Function

General Usage Performance
1

I’m having a tough time optimising and parallelising this function that executes a simplex extrusion (solid modeling). Anything to make it faster would be great, even a couple macros.

function extrudeSimplicial(model::Lar.LAR, pattern)
    V = [model[1][:,k] for k=1:size(model[1],2)]
    FV = model[2]
    d, m = length(FV[1]), length(pattern)
    coords = collect(cumsum(append!([0], abs.(pattern))))
    offset, outcells, rangelimit, i = length(V), [], d*m, 0
    for cell in FV
        i += 1
        tube = [v+k*offset for k in range(0, length=m+1) for v in cell]
        cellTube = [tube[k:k+d] for k in range(1, length=rangelimit)]
        if i==1 outcells = reshape(cellTube, d, m)
        else outcells = vcat(outcells, reshape(cellTube, d, m)) end
    end
    cellGroups = []
    for i in 1:size(outcells, 2)
        if pattern[i]>0
            cellGroups = vcat(cellGroups, outcells[:, i])
        end
    end
    outVertices = [vcat(v, [z]) for z in coords for v in V]
    cellGroups = convert(Array{Array{Int, 1}, 1}, cellGroups)
    outModel = outVertices, cellGroups
    hcat(outVertices...), cellGroups
end
function extrudeSimplicial(model::Union{Any,Lar.Cells}, pattern)
    V,FV = model
    d, m = length(FV[1]), length(pattern)
    coords = collect(cumsum(append!([0], abs.(pattern))))
    offset, outcells, rangelimit, i = length(V), [], d*m, 0
    for cell in FV
        i += 1
        tube = [v+k*offset for k in range(0, length=m+1) for v in cell]
        cellTube = [tube[k:k+d] for k in range(1, length=rangelimit)]
        if i==1 outcells = reshape(cellTube, d, m)
        else outcells = vcat(outcells, reshape(cellTube, d, m)) end
    end
    cellGroups = []
    for i in 1:size(outcells, 2)
        if pattern[i]>0
            cellGroups = vcat(cellGroups, outcells[:, i])
        end
    end
    outVertices = [vcat(v, [z]) for z in coords for v in V]
    cellGroups = convert(Array{Array{Int, 1}, 1}, cellGroups)
    outModel = outVertices, cellGroups
    hcat(outVertices...), cellGroups
end
2

Have you read the performance tips?

https://docs.julialang.org/en/v1/manual/performance-tips/#man-performance-tips

Some simple ones:

  • [:, k] type indexing copies, you should probably use @views
  • [] creates an untyped array, which is terrible for performance
  • hcat/vcat can be slow when splatting large collections, you should probably use reduce(vcat, ...) instead
2 Likes
3

Yeah, I believe the most important section should be Performance Tips · The Julia Language w.r.t. profiling. @mauri5566: did you consider to give us some setup demonstrating calling your performance critical functions?