Faster extrema on abstract arrays when there is no functions or selected dims

[longemen3000]

for reference, see this discourse thread:
https://discourse.julialang.org/t/perf-improvement-suggestion-for-simple-extrema/34776/2
the key is just writing a simple loop in the case of extrema(x::AbstractArray). the post uses an standard indexing, but i think this can be changed?

[longemen3000]

a prototype

function extrema(x::AbstractArray)
  a = b = first(x)
  @inbounds @simd for i in eachindex(x)
    if x[i] > b 
      b = x[i]
    elseif x[i] < a 
      a = x[i]
    end
  end
  a, b
end

[tkf]

Isn't it more or less a dup of #31442?

[longemen3000]

not technically, but it seems so

[johnnychen94]

I've checked against the loop version on both Intel's consumer CPU and server CPU, and the first direct conclusion is:
introducing extrema_loop as a specialization for extrema(A::AbstractArray) is a good change overall.

I'd be curious on why extrema_loop isn't faster than minmax for consumer CPU, but it doesn't make things worse 🤷‍♂

using BenchmarkTools

function extrema_mm(x)
    mn, mx = minimum(x), maximum(x)
    return (mn, mx)
end

function extrema_loop(x)
    mn, mx = x[1], x[1]
    @inbounds @simd for i in eachindex(x)
        v = x[i]
        mn = ifelse(v < mn, v, mn)
        mx = ifelse(v > mx, v, mx)
    end
    return (mn, mx)
end

A = rand(1_000_000)

@btime extrema(A)
@btime extrema_mm(A)
@btime extrema_loop(A)

CPU: Intel(R) Core(TM) i7-12700H

supported instruction set: Intel® SSE4.1, Intel® SSE4.2, Intel® AVX2

Julia Version 1.9.3
Commit bed2cd540a1 (2023-08-24 14:43 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 20 × 12th Gen Intel(R) Core(TM) i7-12700H
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.6 (ORCJIT, alderlake)
  Threads: 1 on 20 virtual cores

version	function	time (ms)
1.9.3	extrema	0.971
1.9.3	extrema_mm	1.185
1.9.3	extrema_loop	0.848
1.10.9	extrema	2.364
1.10.9	extrema_mm	1.161
1.10.9	extrema_loop	2.067
1.11.4	extrema	2.421
1.11.4	extrema_mm	1.391
1.11.4	extrema_loop	2.082
1.12.0-beta1	extrema	2.401
1.12.0-beta1	extrema_mm	0.981
1.12.0-beta1	extrema_loop	0.862

CPU: Intel(R) Xeon(R) Gold 5318Y

supported instruction set: Intel® SSE4.2, Intel® AVX, Intel® AVX2, Intel® AVX-512

Julia Version 1.9.3
Commit bed2cd540a1 (2023-08-24 14:43 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 96 × Intel(R) Xeon(R) Gold 5318Y CPU @ 2.10GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.6 (ORCJIT, icelake-server)
  Threads: 1 on 96 virtual cores

version	function	time (ms)
1.9.3	extrema	4.894
1.9.3	extrema_mm	1.562
1.9.3	extrema_loop	1.344
1.10.9	extrema	3.612
1.10.9	extrema_mm	1.861
1.10.9	extrema_loop	1.343
1.11.5	extrema	3.091
1.11.5	extrema_mm	1.477
1.11.5	extrema_loop	1.183
1.12.0-beta1	extrema	5.109
1.12.0-beta1	extrema_mm	1.660
1.12.0-beta1	extrema_loop	1.339

[barucden]

Note that the current extrema_loop is not equivalent to extrema for floats:

julia> function extrema_loop(x)
           mn, mx = x[1], x[1]
           @inbounds @simd for i in eachindex(x)
               v = x[i]
               mn = ifelse(v < mn, v, mn)
               mx = ifelse(v > mx, v, mx)
           end
           return (mn, mx)
       end
extrema_loop (generic function with 1 method)

julia> extrema([1., NaN])
(NaN, NaN)

julia> extrema_loop([1., NaN])
(1.0, 1.0)