GUFA: Refine casts (#5805)

If we see (ref.cast $A) but we have inferred that a more refined type will
be present there at runtime $B then we can refine the cast to (ref.cast $B).

We could do the same even when a cast is not present, but that would
increase code size. This optimization keeps code size constant.

2 files changed, 22 insertions, 1 deletions

diff --git a/src/passes/GUFA.cpp b/src/passes/GUFA.cpp
index 137ffda17..8f58eaf8a 100644
--- a/src/passes/GUFA.cpp
+++ b/src/passes/GUFA.cpp
@@ -255,6 +255,25 @@ struct GUFAOptimizer
     }
   }
 
+  void visitRefCast(RefCast* curr) {
+    auto currType = curr->type;
+    auto inferredType = getContents(curr).getType();
+    if (inferredType.isRef() && inferredType != currType &&
+        Type::isSubType(inferredType, currType)) {
+      // We have inferred that this will only contain something of a more
+      // refined type, so we might as well cast to that more refined type.
+      //
+      // Note that we could in principle apply this in all expressions by adding
+      // a cast. However, to be careful with code size, we only refine existing
+      // casts for now.
+      curr->type = inferredType;
+    }
+
+    // Apply the usual optimizations as well, such as potentially replacing this
+    // with a constant.
+    visitExpression(curr);
+  }
+
   // TODO: If an instruction would trap on null, like struct.get, we could
   //       remove it here if it has no possible contents and if we are in
   //       traps-never-happen mode (that is, we'd have proven it can only trap,
diff --git a/src/passes/Heap2Local.cpp b/src/passes/Heap2Local.cpp
index 91298106d..fccbc3321 100644
--- a/src/passes/Heap2Local.cpp
+++ b/src/passes/Heap2Local.cpp
@@ -191,7 +191,9 @@ struct Heap2LocalOptimizer {
     localGraph.computeSetInfluences();
 
     // All the allocations in the function.
-    // TODO: Arrays (of constant size) as well.
+    // TODO: Arrays (of constant size) as well, if all element accesses use
+    //       constant indexes. One option might be to first convert such
+    //       nonescaping arrays into structs.
     FindAll<StructNew> allocations(func->body);
 
     for (auto* allocation : allocations.list) {