Bechmark of slicedim for BItArray

benchsd.jl

using BenchmarkTools

function dobench(n, nd)
    x = falses(ntuple(x -> n, nd)...)
    @benchmark slicedim($x, 1, 3)
end

function dobench()
    println(dobench(5, 1))
    println(dobench(10000, 1))
    println(dobench(100, 3))
    println(dobench(10, 8))
end

println("Without specialization")
dobench()

function Base.slicedim(A::BitArray, d::Integer, i::Integer)
    d_in = size(A)
    leading = d_in[1:(d-1)]
    d_out = tuple(leading..., d_in[(d+1):end]...)

    M = prod(leading)
    N = length(A)
    stride = M * d_in[d]

    B = BitArray(d_out)
    index_offset = 1 + (i-1)*M

    l = 1

    if M == 1
        for j = 0:stride:(N-stride)
            B[l] = A[j + index_offset]
            l += 1
        end
    else
        for j = 0:stride:(N-stride)
            offs = j + index_offset
            for k = 0:(M-1)
                B[l] = A[offs + k]
                l += 1
            end
        end
    end
    return B
end

println("With specialization")
dobench()

results.txt

Without specialization
BenchmarkTools.Trial: 
  samples:          10000
  evals/sample:     996
  time tolerance:   5.00%
  memory tolerance: 1.00%
  memory estimate:  16.00 bytes
  allocs estimate:  1
  minimum time:     24.00 ns (0.00% GC)
  median time:      24.00 ns (0.00% GC)
  mean time:        24.77 ns (1.45% GC)
  maximum time:     1.01 μs (96.53% GC)
BenchmarkTools.Trial: 
  samples:          10000
  evals/sample:     996
  time tolerance:   5.00%
  memory tolerance: 1.00%
  memory estimate:  16.00 bytes
  allocs estimate:  1
  minimum time:     22.00 ns (0.00% GC)
  median time:      23.00 ns (0.00% GC)
  mean time:        24.34 ns (1.48% GC)
  maximum time:     989.00 ns (96.36% GC)
BenchmarkTools.Trial: 
  samples:          10000
  evals/sample:     1
  time tolerance:   5.00%
  memory tolerance: 1.00%
  memory estimate:  1.58 kb
  allocs estimate:  12
  minimum time:     25.21 μs (0.00% GC)
  median time:      25.49 μs (0.00% GC)
  mean time:        25.83 μs (0.00% GC)
  maximum time:     65.44 μs (0.00% GC)
BenchmarkTools.Trial: 
  samples:          208
  evals/sample:     1
  time tolerance:   5.00%
  memory tolerance: 1.00%
  memory estimate:  1.19 mb
  allocs estimate:  58
  minimum time:     24.00 ms (0.00% GC)
  median time:      24.04 ms (0.00% GC)
  mean time:        24.11 ms (0.06% GC)
  maximum time:     25.18 ms (2.68% GC)
With specialization
BenchmarkTools.Trial: 
  samples:          10000
  evals/sample:     6
  time tolerance:   5.00%
  memory tolerance: 1.00%
  memory estimate:  352.00 bytes
  allocs estimate:  6
  minimum time:     5.30 μs (0.00% GC)
  median time:      5.44 μs (0.00% GC)
  mean time:        5.48 μs (0.00% GC)
  maximum time:     13.01 μs (0.00% GC)
BenchmarkTools.Trial: 
  samples:          10000
  evals/sample:     6
  time tolerance:   5.00%
  memory tolerance: 1.00%
  memory estimate:  416.00 bytes
  allocs estimate:  10
  minimum time:     5.29 μs (0.00% GC)
  median time:      5.44 μs (0.00% GC)
  mean time:        5.52 μs (0.67% GC)
  maximum time:     386.14 μs (96.07% GC)
BenchmarkTools.Trial: 
  samples:          2378
  evals/sample:     1
  time tolerance:   5.00%
  memory tolerance: 1.00%
  memory estimate:  930.95 kb
  allocs estimate:  49482
  minimum time:     2.00 ms (0.00% GC)
  median time:      2.06 ms (0.00% GC)
  mean time:        2.10 ms (1.96% GC)
  maximum time:     3.55 ms (41.02% GC)
BenchmarkTools.Trial: 
  samples:          2
  evals/sample:     1
  time tolerance:   5.00%
  memory tolerance: 1.00%
  memory estimate:  1.19 gb
  allocs estimate:  69999346
  minimum time:     2.95 s (1.85% GC)
  median time:      2.95 s (1.87% GC)
  mean time:        2.95 s (1.87% GC)
  maximum time:     2.95 s (1.88% GC)