Index: include/llvm/IR/Intrinsics.td
===================================================================
--- include/llvm/IR/Intrinsics.td
+++ include/llvm/IR/Intrinsics.td
@@ -619,7 +619,7 @@
 def int_coro_free : Intrinsic<[llvm_ptr_ty], [llvm_token_ty, llvm_ptr_ty], 
                               [IntrReadMem, IntrArgMemOnly, ReadOnly<1>, 
                                NoCapture<1>]>;
-def int_coro_end : Intrinsic<[], [llvm_ptr_ty, llvm_i1_ty], []>;
+def int_coro_end : Intrinsic<[llvm_i1_ty], [llvm_ptr_ty, llvm_i1_ty], []>;
 
 def int_coro_frame : Intrinsic<[llvm_ptr_ty], [], [IntrNoMem]>;
 def int_coro_size : Intrinsic<[llvm_anyint_ty], [], [IntrNoMem]>;
Index: lib/Transforms/Coroutines/CoroSplit.cpp
===================================================================
--- lib/Transforms/Coroutines/CoroSplit.cpp
+++ lib/Transforms/Coroutines/CoroSplit.cpp
@@ -145,6 +145,33 @@
   BB->getTerminator()->eraseFromParent();
 }
 
+// In Resumers, we replace unwind coro.end with True to force the immediate
+// unwind to caller.
+static void replaceUnwindCoroEnds(coro::Shape &Shape, ValueToValueMapTy &VMap) {
+  if (Shape.CoroEnds.empty())
+    return;
+
+  LLVMContext &Context = Shape.CoroEnds.front()->getContext();
+  auto *True = ConstantInt::getTrue(Context);
+  for (CoroEndInst *CE : Shape.CoroEnds) {
+    if (!CE->isUnwind())
+      continue;
+
+    auto *NewCE = cast<IntrinsicInst>(VMap[CE]);
+
+    // If coro.end has an associated bundle, add cleanupret instruction.
+    if (auto Bundle = NewCE->getOperandBundle(LLVMContext::OB_funclet)) {
+      Value *FromPad = Bundle->Inputs[0];
+      auto *CleanupRet = CleanupReturnInst::Create(FromPad, nullptr, NewCE);
+      NewCE->getParent()->splitBasicBlock(NewCE);
+      CleanupRet->getParent()->getTerminator()->eraseFromParent();
+    }
+
+    NewCE->replaceAllUsesWith(True);
+    NewCE->eraseFromParent();
+  }
+}
+
 // Rewrite final suspend point handling. We do not use suspend index to
 // represent the final suspend point. Instead we zero-out ResumeFnAddr in the
 // coroutine frame, since it is undefined behavior to resume a coroutine
@@ -271,8 +298,7 @@
 
   // Remove coro.end intrinsics.
   replaceFallthroughCoroEnd(Shape.CoroEnds.front(), VMap);
-  // FIXME: coming in upcoming patches:
-  // replaceUnwindCoroEnds(Shape.CoroEnds, VMap);
+  replaceUnwindCoroEnds(Shape, VMap);
 
   // Eliminate coro.free from the clones, replacing it with 'null' in cleanup,
   // to suppress deallocation code.
@@ -285,8 +311,16 @@
 }
 
 static void removeCoroEnds(coro::Shape &Shape) {
-  for (CoroEndInst *CE : Shape.CoroEnds)
+  if (Shape.CoroEnds.empty())
+    return;
+
+  LLVMContext &Context = Shape.CoroEnds.front()->getContext();
+  auto *False = ConstantInt::getFalse(Context);
+
+  for (CoroEndInst *CE : Shape.CoroEnds) {
+    CE->replaceAllUsesWith(False);
     CE->eraseFromParent();
+  }
 }
 
 static void replaceFrameSize(coro::Shape &Shape) {
@@ -510,16 +544,6 @@
     return;
   }
 
-  // If there are no suspend points, no split required, just remove
-  // the allocation and deallocation blocks, they are not needed.
-  if (Shape.CoroSuspends.empty()) {
-    handleNoSuspendCoroutine(Shape.CoroBegin, Shape.FrameTy);
-    removeCoroEnds(Shape);
-    postSplitCleanup(F);
-    coro::updateCallGraph(F, {}, CG, SCC);
-    return;
-  }
-
   auto *ResumeEntry = createResumeEntryBlock(F, Shape);
   auto ResumeClone = createClone(F, ".resume", Shape, ResumeEntry, 0);
   auto DestroyClone = createClone(F, ".destroy", Shape, ResumeEntry, 1);