massive slowdown for nested folding operations

[matthias314]

Nested folding operations may lead to massive slowdowns, apparently from 32 bit numbers on:

julia> N = 32; T = Int16; v = [rand(T, N) for _ in 1:1000]; w = map(SVector{N}, v);

julia> @b sum(sum, $v)
11.637 μs

julia> @b sum(sum, $w)
1.401 μs

julia> N = 16; T = Int32; v = [rand(T, N) for _ in 1:1000]; w = map(SVector{N}, v);

julia> @b sum(sum, $v)
10.120 μs

julia> @b sum(sum, $w)
28.138 μs

julia> N = 16; T = Float32; v = [rand(T, N) for _ in 1:1000]; w = map(SVector{N}, v);

julia> @b sum(sum, $v)
10.656 μs

julia> @b sum(sum, $w)
29.554 μs

I've tried out several processors (all x86-64), with the same result. Maybe it's an LLVM issue. Is it known?

Julia Version 1.11.2
Commit 5e9a32e7af2 (2024-12-01 20:02 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 4 × Intel(R) Core(TM) i3-10110U CPU @ 2.10GHz
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, skylake)

StaticArrays v1.9.13