Skip to content

Add @nospecialize to mapreduce dispatch chain to cut compile time #66

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
lkdvos merged 2 commits into from
Jun 19, 2026
Merged

Add @nospecialize to mapreduce dispatch chain to cut compile time #66

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
94f2005
Add @nospecialize to mapreduce dispatch chain to cut compile time
lkdvos Jun 18, 2026
25b35b4
Apply suggestion from @lkdvos
lkdvos Jun 18, 2026
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
26 changes: 15 additions & 11 deletions src/mapreduce.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,17 +33,17 @@ function Base._mapreduce_dim(f, op, ::NamedTuple{()}, A::StridedView, dims)
end

function Base.map(
f::F, a1::StridedView{<:Any, N},
@nospecialize(f), a1::StridedView{<:Any, N},
A::Vararg{StridedView{<:Any, N}}
) where {F, N}
) where {N}
T = Base.promote_eltype(a1, A...)
return map!(f, similar(a1, T), a1, A...)
end

function Base.map!(
f::F, b::StridedView{<:Any, N}, a1::StridedView{<:Any, N},
@nospecialize(f), b::StridedView{<:Any, N}, a1::StridedView{<:Any, N},
A::Vararg{StridedView{<:Any, N}}
) where {F, N}
) where {N}
dims = size(b)

# Check dimesions
Expand All @@ -59,7 +59,7 @@ function Base.map!(
return b
end

function _mapreduce(f, op, A::StridedView{T}, nt = nothing) where {T}
function _mapreduce(@nospecialize(f), @nospecialize(op), A::StridedView{T}, nt = nothing) where {T}
if isempty(A)
b = Base.mapreduce_empty(f, op, T)
return nt === nothing ? b : op(b, nt.init)
Expand All @@ -79,7 +79,7 @@ function _mapreduce(f, op, A::StridedView{T}, nt = nothing) where {T}
end

function Base.mapreducedim!(
f, op, b::StridedView{<:Any, N},
@nospecialize(f), @nospecialize(op), b::StridedView{<:Any, N},
a1::StridedView{<:Any, N},
A::Vararg{StridedView{<:Any, N}}
) where {N}
Expand All @@ -93,7 +93,7 @@ function Base.mapreducedim!(
end

function _mapreducedim!(
(f), (op), (initop),
@nospecialize(f), @nospecialize(op), @nospecialize(initop),
dims::Dims, arrays::Tuple{Vararg{StridedView}}
)
if any(isequal(0), dims)
Expand All @@ -107,7 +107,7 @@ function _mapreducedim!(
end

function _mapreduce_fuse!(
(f), (op), (initop),
@nospecialize(f), @nospecialize(op), @nospecialize(initop),
dims::Dims, arrays::Tuple{Vararg{StridedView}}
)
# Fuse dimensions if possible: assume that at least one array, e.g. the output array in
Expand All @@ -130,7 +130,7 @@ function _mapreduce_fuse!(
end

function _mapreduce_order!(
(f), (op), (initop),
@nospecialize(f), @nospecialize(op), @nospecialize(initop),
dims, strides, arrays
)
M = length(arrays)
Expand All @@ -155,7 +155,7 @@ end

const MINTHREADLENGTH = 1 << 15 # minimal length before any kind of threading is applied
function _mapreduce_block!(
(f), (op), (initop),
@nospecialize(f), @nospecialize(op), @nospecialize(initop),
dims, strides, offsets, costs, arrays
)
bytestrides = map((s, stride) -> s .* stride, sizeof.(eltype.(arrays)), strides)
Expand Down Expand Up @@ -214,7 +214,7 @@ end
# nthreads: number of threads spacing: extra addition to offset of array 1, to account for
# reduction
function _mapreduce_threaded!(
(f), (op), (initop),
@nospecialize(f), @nospecialize(op), @nospecialize(initop),
dims, blocks, strides, offsets, costs, arrays, nthreads,
spacing, taskindex
)
Expand Down Expand Up @@ -261,6 +261,10 @@ end
strides::NTuple{M, NTuple{N, Int}},
offsets::NTuple{M, Int}
) where {N, M}
return _mapreduce_kernel_expr(f, op, initop, N, M)
end

function _mapreduce_kernel_expr(f, op, initop, N::Int, M::Int)

@Jutho Jutho Jun 19, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a benefit to this approach? Maybe this is actually where some of the compilation time gain comes from? That the code that generates the code no longer needs to be recompiled for different types?

@lkdvos lkdvos Jun 19, 2026

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I measured this at some point, including the actual runtime of this function, and this seemed somewhat negligible. The main reason I did this is specifically for that, where I was benchmarking the expression generation :). I'm happy to revert this, but it is also just convenient to actually inspect the generated code without having to work around the generated function.


# many variables
blockloopvars = Array{Symbol}(undef, N)
Expand Down
Loading