Handle atomics in GTO (#7160)

GTO removes fields that are never read and also removes sets to those
fields. Update the pass to add a seqcst fence when removing a seqcst set
to preserve its effect on the global order of seqcst operations.

2 files changed, 80 insertions, 7 deletions

diff --git a/src/passes/GlobalTypeOptimization.cpp b/src/passes/GlobalTypeOptimization.cpp
index 0baecdf43..b2ba23b0f 100644
--- a/src/passes/GlobalTypeOptimization.cpp
+++ b/src/passes/GlobalTypeOptimization.cpp
@@ -514,13 +514,23 @@ struct GlobalTypeOptimization : public Pass {
           // operations here: the trap on a null ref happens after the value,
           // which might have side effects.
           Builder builder(*getModule());
-          auto flipped = getResultOfFirst(curr->ref,
-                                          builder.makeDrop(curr->value),
-                                          getFunction(),
-                                          getModule(),
-                                          getPassOptions());
-          replaceCurrent(
-            builder.makeDrop(builder.makeRefAs(RefAsNonNull, flipped)));
+          auto* flipped = getResultOfFirst(curr->ref,
+                                           builder.makeDrop(curr->value),
+                                           getFunction(),
+                                           getModule(),
+                                           getPassOptions());
+          Expression* replacement =
+            builder.makeDrop(builder.makeRefAs(RefAsNonNull, flipped));
+          if (curr->order == MemoryOrder::SeqCst) {
+            // If the removed set is sequentially consistent, we must insert a
+            // seqcst fence to preserve the effect on the global order of seqcst
+            // operations. No fence is necessary for release sets because there
+            // are no reads for them to synchronize with given that we are
+            // removing the field.
+            replacement =
+              builder.makeSequence(replacement, builder.makeAtomicFence());
+          }
+          replaceCurrent(replacement);
         }
       }
 
diff --git a/test/lit/passes/gto-removals.wast b/test/lit/passes/gto-removals.wast
index a6fd9c228..6ab126611 100644
--- a/test/lit/passes/gto-removals.wast
+++ b/test/lit/passes/gto-removals.wast
@@ -1564,3 +1564,66 @@
   (export "globalB" (global $globalB))
 )
 
+;; Removed atomic sets needs special handling.
+(module
+  ;; CHECK:      (rec
+  ;; CHECK-NEXT:  (type $A (shared (struct)))
+  (type $A (shared (struct (mut i32))))
+
+  ;; CHECK:       (type $1 (func (param (ref $A))))
+
+  ;; CHECK:      (func $sets (type $1) (param $0 (ref $A))
+  ;; CHECK-NEXT:  (drop
+  ;; CHECK-NEXT:   (ref.as_non_null
+  ;; CHECK-NEXT:    (block (result (ref $A))
+  ;; CHECK-NEXT:     (drop
+  ;; CHECK-NEXT:      (i32.const 1)
+  ;; CHECK-NEXT:     )
+  ;; CHECK-NEXT:     (local.get $0)
+  ;; CHECK-NEXT:    )
+  ;; CHECK-NEXT:   )
+  ;; CHECK-NEXT:  )
+  ;; CHECK-NEXT:  (drop
+  ;; CHECK-NEXT:   (ref.as_non_null
+  ;; CHECK-NEXT:    (block (result (ref $A))
+  ;; CHECK-NEXT:     (drop
+  ;; CHECK-NEXT:      (i32.const 1)
+  ;; CHECK-NEXT:     )
+  ;; CHECK-NEXT:     (local.get $0)
+  ;; CHECK-NEXT:    )
+  ;; CHECK-NEXT:   )
+  ;; CHECK-NEXT:  )
+  ;; CHECK-NEXT:  (block
+  ;; CHECK-NEXT:   (drop
+  ;; CHECK-NEXT:    (ref.as_non_null
+  ;; CHECK-NEXT:     (block (result (ref $A))
+  ;; CHECK-NEXT:      (drop
+  ;; CHECK-NEXT:       (i32.const 1)
+  ;; CHECK-NEXT:      )
+  ;; CHECK-NEXT:      (local.get $0)
+  ;; CHECK-NEXT:     )
+  ;; CHECK-NEXT:    )
+  ;; CHECK-NEXT:   )
+  ;; CHECK-NEXT:   (atomic.fence)
+  ;; CHECK-NEXT:  )
+  ;; CHECK-NEXT: )
+  (func $sets (param (ref $A))
+    ;; Normal set is optimizable.
+    (struct.set $A 0
+      (local.get 0)
+      (i32.const 1)
+    )
+    ;; Release set is optimizable without a fence because there is no get to
+    ;; synchronize with.
+    (struct.atomic.set acqrel $A 0
+      (local.get 0)
+      (i32.const 1)
+    )
+    ;; This requires a fence to keep the effect on the global order of seqcst
+    ;; operations.
+    (struct.atomic.set $A 0
+      (local.get 0)
+      (i32.const 1)
+    )
+  )
+)