Specialize the sparse reducedim_init of maximum, minimum and extrema on f

[ViralBShah]

maximum, minimum and extrema along a dimension of a sparse array spent about 16 µs per call looking up an invoke at run time, however small the array.

Before (minimum time per call, 4×3 matrix and length-4 vector)

A = sparse([1,2,3,1], [1,1,2,3], [1.0,2,3,4], 4, 3)
maximum(A; dims=1)    # 16.7 µs, 1376 B   (dense 4×3: 0.17 µs)
extrema(A; dims=1)    # 17.7 µs, 1376 B
minimum(A'; dims=1)   # 17.0 µs, 1440 B
maximum(v; dims=1)    # 16.6 µs, 1072 B

After

maximum(A; dims=1)    # 0.41 µs, 512 B
extrema(A; dims=1)    # 0.14 µs, 464 B
minimum(A'; dims=1)   # 0.40 µs, 576 B
maximum(v; dims=1)    # 0.19 µs, 208 B

On a 10000×10000 matrix with density 1e-3, maximum(A; dims=1) goes from 347 µs to 320 µs and allocates 864 B less. sum along a dimension is unchanged.

_reducedim_init_dense and the sparse reducedim_init methods for max, min and _extrema_rf only pass f on, so Julia compiled them for f::Function. The invoke(Base.reducedim_init, Tuple{typeof(f), …}) that seeds the result from the first slice then became a dynamic invoke. Base's unspecialized method behind it dispatched isnan, isunordered, mapreduce and reducedim_initarray at run time. Writing f::F makes the invoke static. The set of methods and their behaviour are unchanged.

Left out: Base's _mapreduce_dim does not specialize on f and op either, so it dispatches reducedim_init and mapreducedim! dynamically. A dense Matrix behaves the same way.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.83%. Comparing base (a49c313) to head (7ee715a).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1104   +/-   ##
=======================================
  Coverage   94.83%   94.83%           
=======================================
  Files          16       16           
  Lines       10252    10252           
=======================================
  Hits         9722     9722           
  Misses        530      530           

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.