Quick fix for threaded assembly example on Julia 1.8

[kimauth]

This guarantees the threaded assembly example to work correctly even on Julia 1.8. I am not really sure though what's the best practice pattern to use now - a channel with scratch values I guess?

[fredrikekre]

I am not really sure though what's the best practice pattern to use now - a channel with scratch values I guess?

Probably something like

Ferrite.jl/docs/src/literate/threaded_assembly_spawn.jl

Lines 108 to 121 in 0e3f4d2

	@sync begin
	workqueue = Channel{Int}(Inf)
	foreach(x -> put!(workqueue, x), color)
	close(workqueue)
	for _ in 1:Threads.nthreads()
	Threads.@spawn begin
	local scratch = create_scratchbuffer(K, f, dh)
	for ci in workqueue
	# println("Now I am doing color $(ci)")
	assemble_cell!(scratch, ci, grid, dh, C, b)
	end
	end
	end
	end

, but :static should be fine too.

[KristofferC]

Have someone benchmarked the Channel approach?

[fredrikekre]

IIRC it is more or less equivalent to the :static approach for this example.

[lijas]

This is what I get:

Channels:

function doassemble(K::SparseMatrixCSC, f, colors, grid::Grid, dh::DofHandler, C::SymmetricTensor{4, dim}) where {dim}
    for color in colors
        @sync begin
            workqueue = Channel{Int}(Inf)
            foreach(x -> put!(workqueue, x), color)
            close(workqueue)
            for _ in 1:Threads.nthreads()
                Threads.@spawn begin
                    local scratch = create_scratchbuffer(K, f, dh)
                    for ci in workqueue
                        # println("Now I am doing color $(ci)")
                        assemble_cell!(scratch, ci, grid, dh, C)
                    end
                end
            end
        end
    end
end

$ julia --project -t 1 docs/src/literate/threaded_assembly.jl 
  0.775692 seconds (250.56 k allocations: 11.765 MiB, 6.53% gc time)
$ julia --project -t 4 docs/src/literate/threaded_assembly.jl 
  0.240576 seconds (1.00 M allocations: 44.171 MiB, 25.31% gc time)

@ threads:

function doassemble(K::SparseMatrixCSC, f, colors, grid::Grid, dh::DofHandler, C::SymmetricTensor{4, dim}) where {dim}
scratches = create_scratchvalues(K, f, dh)
for color in colors
        ## Each color is safe to assemble threaded
        Threads.@threads :static for i in 1:length(color)
            assemble_cell!(scratches[Threads.threadid()], color[i], grid, dh, C)
        end
    end
end

$ julia --project -t 1 docs/src/literate/threaded_assembly.jl 
  0.535967 seconds (14.03 k allocations: 632.516 KiB)
$ julia --project -t 4 docs/src/literate/threaded_assembly.jl 
  0.170042 seconds (55.96 k allocations: 2.461 MiB)

But here create_scratchvalues(..) is also included in the time

Edit: Is it possible to pre-allocate the local sratch_buffers and workqueue in some way (when using channels in the first approach)? For the scratch_buffers, one could use some sort of "pool" which the threads can get the sratch-buffers from...