Skip to content

Fix type instability with many array references on Julia 1.11+ #573

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ranocha merged 1 commit into from
Oct 14, 2026
Merged

Fix type instability with many array references on Julia 1.11+ #573

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 10 additions & 1 deletion src/codegen/lowering.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1065,7 +1065,16 @@ function determine_eltype(ls::LoopSet, ortypdefined::Bool)::Union{Symbol,Expr}
push!(pt.args, Expr(:call, lv(:eltype), array))
end
push_outer_reduct_types!(pt, ls, ortypdefined)
return pt
# Nest calls of the two-argument `promote_type` instead of calling it once with all
# types: on Julia >= 1.11, `promote_type` with many (>= 36) arguments is not inferred
# (`Base.afoldl` has `max_args = 34`), which makes the vector width and thus the whole
# generated loop type-unstable. `ls.includedactualarrays` has one entry per array
# reference, so loops with many references to only a few arrays are affected.
nested = pt.args[2]
for i in 3:length(pt.args)
nested = Expr(:call, lv(:promote_type), nested, pt.args[i])
end
return nested
end
@inline _eltype(x) = eltype(x)
@inline _eltype(::BitArray) = VectorizationBase.Bit
Expand Down
11 changes: 11 additions & 0 deletions test/manyarrayrefs.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,3 +21,14 @@ end

@test isapprox(sum_way_too_unrolled(A, Val(17), Val(16)), sum(A))
end

@testset "Many Array References: type stability" begin
# The element type of the loop is determined by `promote_type` over all array
# references. With many references, a single `promote_type` call with many
# arguments is not inferred on Julia >= 1.11, which made the generated code
# type-unstable (dynamic dispatch and allocations in every iteration).
A = rand(6, 7, 10)
sum_way_too_unrolled(A, Val(6), Val(7)) # compile
@test isapprox(sum_way_too_unrolled(A, Val(6), Val(7)), sum(A))
@test (@allocated sum_way_too_unrolled(A, Val(6), Val(7))) == 0
end
Loading