Skip to content

[Custom Descriptors] Consider desc effects in HeapStoreOptimization #7772

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
merged 2 commits into from
Jul 31, 2025
Merged

[Custom Descriptors] Consider desc effects in HeapStoreOptimization #7772

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3155d1e
[Custom Descriptors] Consider desc effects in HeapStoreOptimization
tlively Jul 31, 2025
f86d699
fixes
tlively Jul 31, 2025
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
10 changes: 10 additions & 0 deletions src/passes/HeapStoreOptimization.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -248,6 +248,16 @@ struct HeapStoreOptimization
}
}

// Similarly, we must move the set's value past the allocation's descriptor,
// if it exists.
if (new_->desc) {
auto descEffects = effects(new_->desc);
if (descEffects.invalidates(setValueEffects)) {
// TODO: we could use locals to reorder everything
return false;
}
}

// We also cannot reorder if the struct.new itself has effects (which it can
// in the case of a descriptor) that interact with X' (from the comment
// above), as e.g. a struct.new trap happens before effects in X', but after
Expand Down
59 changes: 50 additions & 9 deletions test/lit/passes/heap-store-optimization-desc.wast
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,15 +6,17 @@
(rec
;; CHECK: (rec
;; CHECK-NEXT: (type $struct (descriptor $desc (struct (field (mut i32)))))
(type $struct (sub final (descriptor $desc (struct (field (mut i32))))))
;; CHECK: (type $desc (sub (describes $struct (struct))))
(type $desc (sub (describes $struct (struct))))
(type $struct (descriptor $desc (struct (field (mut i32)))))
;; CHECK: (type $desc (describes $struct (descriptor $meta (struct))))
(type $desc (describes $struct (descriptor $meta (struct))))
;; CHECK: (type $meta (describes $desc (struct)))
(type $meta (describes $desc (struct)))
)

;; CHECK: (import "" "" (func $effect (type $2)))
;; CHECK: (import "" "" (func $effect (type $3)))
(import "" "" (func $effect))

;; CHECK: (func $no-reorder (type $2)
;; CHECK: (func $no-reorder (type $3)
;; CHECK-NEXT: (local $struct (ref $struct))
;; CHECK-NEXT: (local.set $struct
;; CHECK-NEXT: (struct.new_default $struct
Expand Down Expand Up @@ -47,25 +49,64 @@
)
)

;; CHECK: (func $yes-reorder (type $2)
;; CHECK: (func $no-reorder-nested (type $3)
;; CHECK-NEXT: (local $struct (ref $struct))
;; CHECK-NEXT: (local.set $struct
;; CHECK-NEXT: (struct.new $struct
;; CHECK-NEXT: (block $block (result i32)
;; CHECK-NEXT: (call $effect)
;; CHECK-NEXT: (i32.const 0)
;; CHECK-NEXT: )
;; CHECK-NEXT: (struct.new_default $desc)
;; CHECK-NEXT: (struct.new_default $desc
;; CHECK-NEXT: (ref.null none)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (nop)
Copy link
Member

@kripken kripken Jul 31, 2025

Choose a reason for hiding this comment

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

This looks optimized..?

Copy link
Member Author

@tlively tlively Jul 31, 2025

Choose a reason for hiding this comment

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

Indeed, the test was out of date. Fixed.

;; CHECK-NEXT: )
(func $no-reorder-nested
(local $struct (ref $struct))
;; As above, but now it is not the top-level allocation that traps, but
;; rather its descriptor operand. We still cannot optimized.
Copy link
Member

@kripken kripken Jul 31, 2025

Choose a reason for hiding this comment

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

Suggested change
;; rather its descriptor operand. We still cannot optimized.
;; rather its descriptor operand. We still cannot optimize.

(local.set $struct
(struct.new_default $struct
(struct.new $desc
(ref.null none)
)
)
)
(struct.set $struct 0
(local.get $struct)
(block $block (result i32)
(call $effect)
(i32.const 0)
)
)
)

;; CHECK: (func $yes-reorder (type $3)
;; CHECK-NEXT: (local $struct (ref $struct))
;; CHECK-NEXT: (local.set $struct
;; CHECK-NEXT: (struct.new $struct
;; CHECK-NEXT: (block $block (result i32)
;; CHECK-NEXT: (call $effect)
;; CHECK-NEXT: (i32.const 0)
;; CHECK-NEXT: )
;; CHECK-NEXT: (struct.new_default $desc
;; CHECK-NEXT: (struct.new_default $meta)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (nop)
;; CHECK-NEXT: )
(func $yes-reorder
(local $struct (ref $struct))
;; As above, but now the descriptor does not trap, so we optimize.
;; Nothing traps, so we can reorder.
(local.set $struct
(struct.new_default $struct
(struct.new $desc)
(struct.new $desc
(struct.new $meta)
)
)
)
(struct.set $struct 0
Expand Down
Loading