Handle unoptimized branches in CodeFolding (#7111)

CodeFolding previously did not consider br_on_* instructions at all, so it would happily merge tails even if there were br_on_* branches to the same label with non-matching tails. Fix the bug by making any label targeted by any instruction not explicitly handled by CodeFolding unoptimizable. This will gracefully handle other branching instructions like `resume` and `resume_throw` as well. Folding these branches properly is left as future work. Also rename the test file from code-folding_enable-threads.wast to just code-folding.wast and enable all features instead of just threads. The old name was left over from when the test was originally ported to lit, and the new feature is necessary because the new test uses GC instructions.
WebAssembly · Nov 25, 2024 · a2a8d2a · a2a8d2a
1 parent 3d39401
commit a2a8d2a
Show file tree

Hide file tree

Showing 2 changed files with 65 additions and 20 deletions.
diff --git a/src/passes/CodeFolding.cpp b/src/passes/CodeFolding.cpp
@@ -84,7 +84,9 @@ struct ExpressionMarker
   void visitExpression(Expression* expr) { marked.insert(expr); }
 };
 
-struct CodeFolding : public WalkerPass<ControlFlowWalker<CodeFolding>> {
+struct CodeFolding
+  : public WalkerPass<
+      ControlFlowWalker<CodeFolding, UnifiedExpressionVisitor<CodeFolding>>> {
   bool isFunctionParallel() override { return true; }
 
   std::unique_ptr<Pass> create() override {
@@ -138,6 +140,17 @@ struct CodeFolding : public WalkerPass<ControlFlowWalker<CodeFolding>> {
 
   // walking
 
+  void visitExpression(Expression* curr) {
+    // For any branching instruction not explicitly handled by this pass, mark
+    // the labels it branches to unoptimizable.
+    // TODO: Handle folding br_on* instructions. br_on_null could be folded with
+    // other kinds of branches and br_on_non_null, br_on_cast, and
+    // br_on_cast_fail instructions could be folded with other copies of
+    // themselves.
+    BranchUtils::operateOnScopeNameUses(
+      curr, [&](Name label) { unoptimizables.insert(label); });
+  }
+
   void visitBreak(Break* curr) {
     if (curr->condition || curr->value) {
       unoptimizables.insert(curr->name);
@@ -155,13 +168,6 @@ struct CodeFolding : public WalkerPass<ControlFlowWalker<CodeFolding>> {
     }
   }
 
-  void visitSwitch(Switch* curr) {
-    for (auto target : curr->targets) {
-      unoptimizables.insert(target);
-    }
-    unoptimizables.insert(curr->default_);
-  }
-
   void visitUnreachable(Unreachable* curr) {
     // we can only optimize if we are at the end of the parent block
     if (!controlFlowStack.empty()) {

diff --git a/...t/passes/code-folding_enable-threads.wast → test/lit/passes/code-folding.wast b/...t/passes/code-folding_enable-threads.wast → test/lit/passes/code-folding.wast
@@ -1,7 +1,7 @@
 ;; NOTE: Assertions have been generated by update_lit_checks.py --all-items and should not be edited.
 ;; NOTE: This test was ported using port_passes_tests_to_lit.py and could be cleaned up.
 
-;; RUN: foreach %s %t wasm-opt --code-folding --enable-threads -S -o - | filecheck %s
+;; RUN: foreach %s %t wasm-opt -all --code-folding -S -o - | filecheck %s
 
 (module
  ;; CHECK:      (type $0 (func))
@@ -15,13 +15,13 @@
  (memory $0 1 1)
  ;; CHECK:      (table $0 282 282 funcref)
 
- ;; CHECK:      (func $0
+ ;; CHECK:      (func $0 (type $0)
  ;; CHECK-NEXT:  (block $label$1
  ;; CHECK-NEXT:   (if
  ;; CHECK-NEXT:    (i32.const 1)
  ;; CHECK-NEXT:    (then
  ;; CHECK-NEXT:     (block $label$3
- ;; CHECK-NEXT:      (call_indirect (type $13)
+ ;; CHECK-NEXT:      (call_indirect $0 (type $13)
  ;; CHECK-NEXT:       (block $label$4
  ;; CHECK-NEXT:        (br $label$3)
  ;; CHECK-NEXT:       )
@@ -52,7 +52,7 @@
    )
   )
  )
- ;; CHECK:      (func $negative-zero (result f32)
+ ;; CHECK:      (func $negative-zero (type $1) (result f32)
  ;; CHECK-NEXT:  (if (result f32)
  ;; CHECK-NEXT:   (i32.const 0)
  ;; CHECK-NEXT:   (then
@@ -82,7 +82,7 @@
    )
   )
  )
- ;; CHECK:      (func $negative-zero-b (result f32)
+ ;; CHECK:      (func $negative-zero-b (type $1) (result f32)
  ;; CHECK-NEXT:  (drop
  ;; CHECK-NEXT:   (i32.const 0)
  ;; CHECK-NEXT:  )
@@ -105,7 +105,7 @@
    )
   )
  )
- ;; CHECK:      (func $negative-zero-c (result f32)
+ ;; CHECK:      (func $negative-zero-c (type $1) (result f32)
  ;; CHECK-NEXT:  (drop
  ;; CHECK-NEXT:   (i32.const 0)
  ;; CHECK-NEXT:  )
@@ -128,7 +128,7 @@
    )
   )
  )
- ;; CHECK:      (func $break-target-outside-of-return-merged-code
+ ;; CHECK:      (func $break-target-outside-of-return-merged-code (type $0)
  ;; CHECK-NEXT:  (block $label$A
  ;; CHECK-NEXT:   (if
  ;; CHECK-NEXT:    (unreachable)
@@ -202,7 +202,7 @@
    )
   )
  )
- ;; CHECK:      (func $break-target-inside-all-good
+ ;; CHECK:      (func $break-target-inside-all-good (type $0)
  ;; CHECK-NEXT:  (block $folding-inner0
  ;; CHECK-NEXT:   (block $label$A
  ;; CHECK-NEXT:    (if
@@ -269,7 +269,7 @@
    )
   )
  )
- ;; CHECK:      (func $leave-inner-block-type
+ ;; CHECK:      (func $leave-inner-block-type (type $0)
  ;; CHECK-NEXT:  (block $label$1
  ;; CHECK-NEXT:   (drop
  ;; CHECK-NEXT:    (block $label$2
@@ -312,7 +312,7 @@
  (memory $0 1 1 shared)
  ;; CHECK:      (export "func_2224" (func $0))
  (export "func_2224" (func $0))
- ;; CHECK:      (func $0 (result i32)
+ ;; CHECK:      (func $0 (type $0) (result i32)
  ;; CHECK-NEXT:  (local $var$0 i32)
  ;; CHECK-NEXT:  (if (result i32)
  ;; CHECK-NEXT:   (i32.const 0)
@@ -352,7 +352,7 @@
 
  ;; CHECK:      (global $global$0 (mut i32) (i32.const 10))
  (global $global$0 (mut i32) (i32.const 10))
- ;; CHECK:      (func $determinism
+ ;; CHECK:      (func $determinism (type $0)
  ;; CHECK-NEXT:  (block $folding-inner0
  ;; CHECK-NEXT:   (block
  ;; CHECK-NEXT:    (block $label$1
@@ -439,7 +439,7 @@
   )
   (unreachable)
  )
- ;; CHECK:      (func $careful-of-the-switch (param $0 i32)
+ ;; CHECK:      (func $careful-of-the-switch (type $1) (param $0 i32)
  ;; CHECK-NEXT:  (block $label$1
  ;; CHECK-NEXT:   (block $label$3
  ;; CHECK-NEXT:    (block $label$5
@@ -482,3 +482,42 @@
   )
  )
 )
+
+(module
+ ;; CHECK:      (type $0 (func))
+
+ ;; CHECK:      (func $br-on-null (type $0)
+ ;; CHECK-NEXT:  (block $block
+ ;; CHECK-NEXT:   (drop
+ ;; CHECK-NEXT:    (br_on_null $block
+ ;; CHECK-NEXT:     (ref.null none)
+ ;; CHECK-NEXT:    )
+ ;; CHECK-NEXT:   )
+ ;; CHECK-NEXT:   (drop
+ ;; CHECK-NEXT:    (block (result i32)
+ ;; CHECK-NEXT:     (call $br-on-null)
+ ;; CHECK-NEXT:     (br $block)
+ ;; CHECK-NEXT:    )
+ ;; CHECK-NEXT:   )
+ ;; CHECK-NEXT:   (call $br-on-null)
+ ;; CHECK-NEXT:  )
+ ;; CHECK-NEXT: )
+ (func $br-on-null
+  (block $block
+   (drop
+    ;; The other two tails are the same, but this br_on_null should inhibit code
+    ;; folding.
+    (br_on_null $block
+     (ref.null none)
+    )
+   )
+   (drop
+    (block (result i32)
+     (call $br-on-null)
+     (br $block)
+    )
+   )
+   (call $br-on-null)
+  )
+ )
+)