diff --git a/llvm/include/llvm/Analysis/DivergenceAnalysis.h b/llvm/include/llvm/Analysis/DivergenceAnalysis.h
--- a/llvm/include/llvm/Analysis/DivergenceAnalysis.h
+++ b/llvm/include/llvm/Analysis/DivergenceAnalysis.h
@@ -82,7 +82,7 @@
   void print(raw_ostream &OS, const Module *) const;
 
 private:
-  bool updateTerminator(const Instruction &Term) const;
+  bool updateTerminator(const Instruction &Term);
   bool updatePHINode(const PHINode &Phi) const;
 
   /// \brief Computes whether \p Inst is divergent based on the
@@ -171,6 +171,9 @@
   // Detected/marked divergent values.
   DenseSet<const Value *> DivergentValues;
 
+  // Terminators that were marked divergent outside the worklist.
+  DenseSet<const Instruction *> DeferredTerminators;
+
   // Internal worklist for divergence propagation.
   std::vector<const Instruction *> Worklist;
 };
diff --git a/llvm/lib/Analysis/DivergenceAnalysis.cpp b/llvm/lib/Analysis/DivergenceAnalysis.cpp
--- a/llvm/lib/Analysis/DivergenceAnalysis.cpp
+++ b/llvm/lib/Analysis/DivergenceAnalysis.cpp
@@ -107,7 +107,11 @@
   UniformOverrides.insert(&UniVal);
 }
 
-bool DivergenceAnalysis::updateTerminator(const Instruction &Term) const {
+bool DivergenceAnalysis::updateTerminator(const Instruction &Term) {
+  if (DeferredTerminators.count(&Term)) {
+    DeferredTerminators.erase(&Term);
+    return true;
+  }
   if (Term.getNumSuccessors() <= 1)
     return false;
   if (auto *BranchTerm = dyn_cast<BranchInst>(&Term)) {
@@ -233,8 +237,13 @@
         if (!OpInst)
           continue;
         if (DivLoop->contains(OpInst->getParent())) {
-          markDivergent(I);
-          pushUsers(I);
+          if (I.isTerminator()) {
+            DeferredTerminators.insert(&I);
+            Worklist.push_back(&I);
+          } else {
+            markDivergent(I);
+            pushUsers(I);
+          }
           break;
         }
       }
diff --git a/llvm/test/Analysis/DivergenceAnalysis/AMDGPU/propagate-loop-live-out.ll b/llvm/test/Analysis/DivergenceAnalysis/AMDGPU/propagate-loop-live-out.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/Analysis/DivergenceAnalysis/AMDGPU/propagate-loop-live-out.ll
@@ -0,0 +1,48 @@
+; RUN: opt -mtriple amdgcn-unknown-amdhsa -analyze -divergence -use-gpu-divergence-analysis %s | FileCheck %s
+
+; CHECK: bb6:
+; CHECK: DIVERGENT:       %.126.i355.i = phi i1 [ false, %bb5 ], [ true, %bb4 ]
+; CHECK: DIVERGENT:       br i1 %.126.i355.i, label %bb7, label %bb8
+
+; Function Attrs: nounwind readnone speculatable
+declare i32 @llvm.amdgcn.workitem.id.x() #0
+
+define protected amdgpu_kernel void @_Z23krnl_GPUITSFitterKerneli() {
+bb0:
+  %i4 = call i32 @llvm.amdgcn.workitem.id.x()
+  %i5 = icmp eq i32 %i4, -1
+  br label %bb1
+
+bb1:                                              ; preds = %bb3, %bb0
+  %lsr.iv = phi i32 [ %i1, %bb3 ], [ 7, %bb0 ]
+  br i1 %i5, label %bb2, label %bb3
+
+bb2:                                              ; preds = %bb1
+  %lsr.iv.next = add nsw i32 %lsr.iv, -1
+  %i14 = icmp eq i32 %lsr.iv.next, 0
+  br label %bb3
+
+bb3:                                              ; preds = %bb2, %bb1
+  %i1 = phi i32 [ %lsr.iv.next, %bb2 ], [ 0, %bb1 ]
+  %i2 = phi i1 [ false, %bb2 ], [ true, %bb1 ]
+  %i3 = phi i1 [ %i14, %bb2 ], [ true, %bb1 ]
+  br i1 %i3, label %bb4, label %bb1
+
+bb4:                                              ; preds = %bb3
+  br i1 %i2, label %bb5, label %bb6
+
+bb5:                                              ; preds = %bb4
+  br label %bb6
+
+bb6:                                              ; preds = %bb5, %bb4
+  %.126.i355.i = phi i1 [ false, %bb5 ], [ true, %bb4 ]
+  br i1 %.126.i355.i, label %bb7, label %bb8
+
+bb7:                                              ; preds = %bb6
+  br label %bb8
+
+bb8:                                              ; preds = %bb7, %bb6
+  ret void
+}
+
+attributes #0 = { nounwind readnone speculatable }