Get rid of union!

[gdalle]

Modify promote_order and distributive_merge to use reduce

[adrhill]

Can you run a before-and-after benchmark on the Brusselator to make sure we don't lose performance on BitSet?

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 72.10%. Comparing base (83cb0b0) to head (15b0cba).

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #42      +/-   ##
==========================================
- Coverage   72.30%   72.10%   -0.21%     
==========================================
  Files           8        8              
  Lines         278      276       -2     
==========================================
- Hits          201      199       -2     
  Misses         77       77              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[gdalle]

Setup code:

using BenchmarkTools
using SparseConnectivityTracer

include("brusselator.jl")

function benchmark_brusselator(N::Integer)
    dims = (N, N, 2)
    A = 1.0
    B = 1.0
    alpha = 1.0
    xyd = fill(1.0, N)
    dx = 1.0
    p = (A, B, alpha, xyd, dx, N)

    u = rand(dims...)
    du = similar(u)
    f!(du, u) = brusselator_2d_loop(du, u, p, nothing)
    return @benchmark jacobian_pattern($f!, $du, $u)
end

## Run Brusselator benchmarks
for N in (6, 24)
    @info "Benchmarking Brusselator of size $N"
    b = benchmark_brusselator(N)
    display(b)
end

On main:

[ Info: Benchmarking Brusselator of size 6
BenchmarkTools.Trial: 10000 samples with 1 evaluation.
 Range (min … max):  51.700 μs …   4.742 ms  ┊ GC (min … max): 0.00% … 95.55%
 Time  (median):     55.865 μs               ┊ GC (median):    0.00%
 Time  (mean ± σ):   63.044 μs ± 109.052 μs  ┊ GC (mean ± σ):  7.22% ±  4.22%

   ▁▆██▇▃                                                       
  ▂███████▆▄▄▃▂▃▂▂▂▂▂▂▂▂▂▂▂▁▂▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▂
  51.7 μs         Histogram: frequency by time         97.4 μs <

 Memory estimate: 108.00 KiB, allocs estimate: 1252.
[ Info: Benchmarking Brusselator of size 24
BenchmarkTools.Trial: 3286 samples with 1 evaluation.
 Range (min … max):  1.086 ms …  10.365 ms  ┊ GC (min … max): 0.00% … 66.99%
 Time  (median):     1.248 ms               ┊ GC (median):    0.00%
 Time  (mean ± σ):   1.515 ms ± 712.712 μs  ┊ GC (mean ± σ):  7.19% ± 12.84%

  ▃██▆▅▅▄▃▃▁▁▁ ▁                                              ▁
  ██████████████▇▇▇█▇▇▆▇████▆▇▆▆▆▆▆▅▅▅▆▅▅▆▅▃▄▅▆██▇▆▆▅▅▅▅▃▅▅▄▃ █
  1.09 ms      Histogram: log(frequency) by time      4.21 ms <

 Memory estimate: 2.15 MiB, allocs estimate: 23416.

On no_union!:

[ Info: Benchmarking Brusselator of size 6
BenchmarkTools.Trial: 10000 samples with 1 evaluation.
 Range (min … max):  54.363 μs …  3.114 ms  ┊ GC (min … max): 0.00% … 94.77%
 Time  (median):     62.468 μs              ┊ GC (median):    0.00%
 Time  (mean ± σ):   67.276 μs ± 82.905 μs  ┊ GC (mean ± σ):  5.55% ±  4.46%

     ▁▁▂▃▃▃▄▄▄▅▆▇█▇▆▆▅▄▃▂▂▁▁              ▁                   ▂
  ▅▄██████████████████████████▇▇▇▇▇███████████▇▇▇▇▇▆▆▆▅▆▆▆▆▅▅ █
  54.4 μs      Histogram: log(frequency) by time      84.6 μs <

 Memory estimate: 108.00 KiB, allocs estimate: 1252.
[ Info: Benchmarking Brusselator of size 24
BenchmarkTools.Trial: 3496 samples with 1 evaluation.
 Range (min … max):  1.060 ms …  14.582 ms  ┊ GC (min … max): 0.00% … 81.11%
 Time  (median):     1.203 ms               ┊ GC (median):    0.00%
 Time  (mean ± σ):   1.424 ms ± 588.615 μs  ┊ GC (mean ± σ):  6.73% ± 12.66%

   ▅█▇▅▄▃▃▃▃▂▂  ▁                                             ▁
  ▆██████████████████▇▇▇▇▆▇▆▆▆▆▇▇▇▆██▇▇▅▅▆█████▆▇▆▄▅▆▇▅▅▁▁▄▄▄ █
  1.06 ms      Histogram: log(frequency) by time      3.58 ms <

 Memory estimate: 2.15 MiB, allocs estimate: 23416.

[gdalle]

No obvious changes

[gdalle]

Wait don't merge yet

[adrhill]

Reopened as #44.