Index: lib/Target/WebAssembly/WebAssemblyRegStackify.cpp
===================================================================
--- lib/Target/WebAssembly/WebAssemblyRegStackify.cpp
+++ lib/Target/WebAssembly/WebAssemblyRegStackify.cpp
@@ -81,6 +81,7 @@
 // more precise.
 static bool IsSafeToMove(const MachineInstr *Def, const MachineInstr *Insert,
                          AliasAnalysis &AA) {
+  assert(Def->getParent() == Insert->getParent());
   bool SawStore = false, SawSideEffects = false;
   MachineBasicBlock::const_iterator D(Def), I(Insert);
   for (--I; I != D; --I)
@@ -155,17 +156,15 @@
             Def->getOpcode() == WebAssembly::ARGUMENT_F64)
           continue;
 
-        // Single-use expression trees require defs that have one use, or that
-        // they be trivially clonable.
+        // Single-use expression trees require defs that have one use.
         // TODO: Eventually we'll relax this, to take advantage of set_local
         // returning its result.
         if (!MRI.hasOneUse(Reg))
           continue;
 
-        // For now, be conservative and don't look across block boundaries,
-        // unless we have something trivially clonable.
+        // For now, be conservative and don't look across block boundaries.
         // TODO: Be more aggressive.
-        if (Def->getParent() != &MBB && !Def->isMoveImmediate())
+        if (Def->getParent() != &MBB)
           continue;
 
         // Don't move instructions that have side effects or memory dependencies
Index: lib/Target/WebAssembly/WebAssemblyStoreResults.cpp
===================================================================
--- lib/Target/WebAssembly/WebAssemblyStoreResults.cpp
+++ lib/Target/WebAssembly/WebAssemblyStoreResults.cpp
@@ -89,12 +89,18 @@
         for (auto I = MRI.use_begin(FromReg), E = MRI.use_end(); I != E;) {
           MachineOperand &O = *I++;
           MachineInstr *Where = O.getParent();
-          if (Where->getOpcode() == TargetOpcode::PHI)
-            Where = Where->getOperand(&O - &Where->getOperand(0) + 1)
-                        .getMBB()
-                        ->getFirstTerminator();
-          if (&MI == Where || !MDT.dominates(&MI, Where))
-            continue;
+          if (Where->getOpcode() == TargetOpcode::PHI) {
+            // For a PHI, check that MI dominates the edge on which the operand
+            // is used.
+            MachineBasicBlock *Pred = Where->getOperand(&O - &Where->getOperand(0) + 1).getMBB();
+            if (!MDT.dominates(&MBB, Pred))
+              continue;
+          } else {
+            // For a non-PHI, check that MI dominates the instruction in the
+            // normal way.
+            if (&MI == Where || !MDT.dominates(&MI, Where))
+              continue;
+          }
           DEBUG(dbgs() << "Setting operand " << O << " in " << *Where <<
                 " from " << MI <<"\n");
           O.setReg(ToReg);
Index: test/CodeGen/WebAssembly/store-results.ll
===================================================================
--- test/CodeGen/WebAssembly/store-results.ll
+++ test/CodeGen/WebAssembly/store-results.ll
@@ -16,3 +16,44 @@
   store i32 0, i32* %p
   ret i32 0
 }
+
+; Test interesting corner cases for wasm-store-results.
+
+%class.Vec3 = type { float, float, float }
+
+@pos = global %class.Vec3 zeroinitializer, align 4
+
+; CHECK-LABEL: foo:
+; CHECK: i32.store $discard=, $pop0, $0
+define void @foo() {
+for.body.i:
+  br label %for.body5.i
+
+for.body5.i:
+  %i.0168.i = phi i32 [ 0, %for.body.i ], [ %inc.i, %for.body5.i ]
+  %conv6.i = sitofp i32 %i.0168.i to float
+  store volatile float 0.0, float* getelementptr inbounds (%class.Vec3, %class.Vec3* @pos, i32 0, i32 0)
+  %inc.i = add nuw nsw i32 %i.0168.i, 1
+  %exitcond.i = icmp eq i32 %inc.i, 256
+  br i1 %exitcond.i, label %for.cond.cleanup4.i, label %for.body5.i
+
+for.cond.cleanup4.i:
+  ret void
+}
+
+; CHECK-LABEL: bar:
+; CHECK: i32.store $discard=, $0, $pop0
+define void @bar() {
+for.body.i:
+  br label %for.body5.i
+
+for.body5.i:
+  %i.0168.i = phi float [ 0.0, %for.body.i ], [ %inc.i, %for.body5.i ]
+  store volatile float 0.0, float* getelementptr inbounds (%class.Vec3, %class.Vec3* @pos, i32 0, i32 0)
+  %inc.i = fadd float %i.0168.i, 1.0
+  %exitcond.i = fcmp oeq float %inc.i, 256.0
+  br i1 %exitcond.i, label %for.cond.cleanup4.i, label %for.body5.i
+
+for.cond.cleanup4.i:
+  ret void
+}