Skip to content

Cast optimization improvement: also handle local.tee #6507

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
Apr 18, 2024
Merged

Cast optimization improvement: also handle local.tee #6507

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6cc2f1b
Cast optimization improvement: also handle local.tee
vouillon Apr 17, 2024
dc67f1a
Updates
vouillon Apr 18, 2024
File filter

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
53 changes: 26 additions & 27 deletions src/passes/OptimizeCasts.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,21 +93,10 @@
// RefAs with ExternInternalize and ExternExternalize are not considered casts
// when obtaining fallthroughs, and so are ignored.
//
// TODO: 1. Look past individual basic blocks? This may be worth considering
// given the pattern of a cast appearing in an if condition that is
// then used in an if arm, for example, where simple dominance shows
// the cast can be reused.
// TODO: 2. Look at LocalSet as well and not just Get. That would add some
// overlap with the other passes mentioned above (SimplifyLocals and
// RedundantSetElimination also track sets and can switch a get to use
// a better set's index when that refines the type). But once we do the
// first two TODOs above then we'd be adding some novel things here,
// as we could optimize "backwards" as well (TODO 1) and past basic
// blocks (TODO 2, though RedundantSetElimination does that as well).
// However, we should consider whether improving those other passes
// might make more sense (as it would help more than casts, if we could
// make them operate "backwards" and/or past basic blocks).
Copy link
Member

Choose a reason for hiding this comment

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

The first TODO is still relevant, isn't it?
Copy link
Contributor Author

Choose a reason for hiding this comment

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

Oh, it is indeed only partially done (with connectAdjacentBlocks set to true).
//
// TODO: Look past individual basic blocks? This may be worth considering
// given the pattern of a cast appearing in an if condition that is
// then used in an if arm, for example, where simple dominance shows
// the cast can be reused.

#include "ir/effects.h"
#include "ir/linear-execution.h"
Expand Down Expand Up @@ -453,20 +442,30 @@ struct BestCastFinder : public LinearExecutionWalker<BestCastFinder> {
void visitRefCast(RefCast* curr) { handleRefinement(curr); }

void handleRefinement(Expression* curr) {
auto* fallthrough = Properties::getFallthrough(curr, options, *getModule());
auto* teeFallthrough = Properties::getFallthrough(
curr, options, *getModule(), Properties::FallthroughBehavior::NoTeeBrIf);
if (auto* tee = teeFallthrough->dynCast<LocalSet>()) {
updateBestCast(curr, tee->index);
}
auto* fallthrough =
Properties::getFallthrough(teeFallthrough, options, *getModule());
if (auto* get = fallthrough->dynCast<LocalGet>()) {
auto*& bestCast = mostCastedGets[get->index];
if (!bestCast) {
// This is the first.
bestCast = curr;
return;
}
updateBestCast(curr, get->index);
}
}

// See if we are better than the current best.
if (curr->type != bestCast->type &&
Type::isSubType(curr->type, bestCast->type)) {
bestCast = curr;
}
void updateBestCast(Expression* curr, Index index) {
auto*& bestCast = mostCastedGets[index];
if (!bestCast) {
// This is the first.
bestCast = curr;
return;
}

// See if we are better than the current best.
if (curr->type != bestCast->type &&
Type::isSubType(curr->type, bestCast->type)) {
bestCast = curr;
}
}
};
Expand Down
37 changes: 37 additions & 0 deletions test/lit/passes/optimize-casts.wast
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1352,6 +1352,43 @@
)
)

;; CHECK: (func $local-tee (type $2) (param $x (ref struct))
;; CHECK-NEXT: (local $y (ref struct))
;; CHECK-NEXT: (local $2 (ref $A))
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (local.tee $2
;; CHECK-NEXT: (ref.cast (ref $A)
;; CHECK-NEXT: (local.tee $y
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (local.get $2)
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (local.get $2)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
(func $local-tee (param $x (ref struct))
(local $y (ref struct))
;; We should use the cast value after it has been computed, in both gets.
(drop
(ref.cast (ref $A)
(local.tee $y
(local.get $x)
)
)
)
(drop
(local.get $x)
)
(drop
(local.get $y)
)
)

;; CHECK: (func $get (type $11) (result (ref struct))
;; CHECK-NEXT: (unreachable)
;; CHECK-NEXT: )
Expand Down