[JuliaLowering] Tidy up node creation functions #60191
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
We currently use keyword arguments as arbitrary key-value pairs in a number of
places, which I've realized isn't really how they're intended to be used.
This change fixes this and reduces the JuliaLowering package image size by
about 20% on my machine from `26.910 MiB -> 21.701 MiB`, or when compiled
into sysbase.dylib, `+55MB -> +45MB`. Unfortunately no change to
precompile time as I had hoped, but I figured this is worth doing anyway.
I've just used `Vector{Pair}` or explicit `setattr` calls, but let me know if
there's a better way of doing this.
Also let me know if you have a reason for the difference in size when compiled
into Base---I don't think I'm accidentally producing two copies, but you
never know.
mlechu
added
the
compiler:lowering
topolarity
reviewed
Nov 25, 2025
| function set_scope_layer(ctx, ex, layer_id, force) | ||
| k = kind(ex) | ||
| scope_layer = force ? layer_id : get(ex, :scope_layer, layer_id) | ||
| attrs = [:scope_layer=>(force ? layer_id : get(ex, :scope_layer, layer_id))] |
Member
There was a problem hiding this comment.
Suggested change
| attrs = [:scope_layer=>(force ? layer_id : get(ex, :scope_layer, layer_id))] | |
| attrs = Pair{Symbol,LayerId}[ | |
| :scope_layer=>(force ? layer_id : get(ex, :scope_layer, layer_id)) | |
| ] |
if you're really trying to be strict about specializations, it can be a good idea to explicitly type most arrays, since:
Pairover-specializes just likeNamedTuple(so the strict array type will eitherconvertthis away or push you to fix the type instability), and- Whether this vector ends up well-typed depends on inference /
Base._return_type(esp. bad for recursive functions)
| elseif k == K"." | ||
| makenode(ctx, ex, ex, set_scope_layer(ctx, ex[1], layer_id, force), ex[2], | ||
| scope_layer=scope_layer) | ||
| children = [set_scope_layer(ctx, ex[1], layer_id, force), ex[2]] |
Member
There was a problem hiding this comment.
Suggested change
| children = [set_scope_layer(ctx, ex[1], layer_id, force), ex[2]] | |
| children = NodeId[set_scope_layer(ctx, ex[1], layer_id, force), ex[2]] |
see above
| elseif !is_leaf(ex) | ||
| mapchildren(e->set_scope_layer(ctx, e, layer_id, force), ctx, ex; | ||
| scope_layer=scope_layer) | ||
| mapchildren(e->set_scope_layer(ctx, e, layer_id, force), ctx, ex, attrs) |
Member
There was a problem hiding this comment.
It looks like this method doesn't exist, unless I misread something?
| function mapchildren(f::Function, ctx, ex::SyntaxTree, mapped_children::AbstractVector{<:Integer}; | ||
| extra_attrs...) | ||
| function mapchildren(f::Function, ctx, ex::SyntaxTree, | ||
| mapped_children::AbstractVector{<:Integer}) |
Member
There was a problem hiding this comment.
narrowing this to Vector{NodeId} would be a good idea, if it's possible
|
|
||
| function makenode(ctx, srcref, proto, children...; attrs...) | ||
| _makenode(ctx, srcref, proto, _node_ids(syntax_graph(ctx), children...); attrs...) | ||
| function makeleaf(ctx, srcref, proto, @nospecialize(attrs::AbstractVector)) |
Member
There was a problem hiding this comment.
Another normalization trick you might consider trying out at some point is:
@noinline function _makeleaf(..., attrs::Vector{Any})
# ... (complicated function)
end
@inline function makeleaf(..., @nospecialize(attrs::AbstractVector))
return _makeleaf(..., convert(Vector{Any}, attrs))
endThis preserves "good" codegen, while trying to limit specialized code to the caller side. Ofc, it's even better if we can avoid the Vector type explosion to begin with.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
We currently use keyword arguments as arbitrary key-value pairs in a number of
places, which I've realized isn't really how they're intended to be used.
This change fixes this and reduces the JuliaLowering package image size by
about 20% on my machine from
26.910 MiB -> 21.701 MiB, or when compiledinto sysbase.dylib,
+55MB -> +45MB. Unfortunately no change toprecompile time as I had hoped, but I figured this is worth doing anyway.
Lowering performance looks unchanged.
I've just used
Vector{Pair}or explicitsetattrcalls, but let me know ifthere's a better way of doing this.
Also let me know if you have a reason for the difference in size when compiled
into Base---I don't think I'm accidentally producing two copies, but you
never know.