diff --git a/llvm/lib/Target/X86/X86FastISel.cpp b/llvm/lib/Target/X86/X86FastISel.cpp
index bb95ed3ccdc53a..7ef800a35d9a96 100644
--- a/llvm/lib/Target/X86/X86FastISel.cpp
+++ b/llvm/lib/Target/X86/X86FastISel.cpp
@@ -2588,6 +2588,17 @@ bool X86FastISel::TryEmitSmallMemcpy(X86AddressMode DestAM,
   return true;
 }
 
+// Add an annotation to an intrinsic instruction, specifying whether the
+// intrinsic has been inlined or not.
+void annotateIntrinsic(const IntrinsicInst *II, bool Inlined) {
+  IntrinsicInst *CI = const_cast<IntrinsicInst *>(II);
+  LLVMContext& C = CI->getContext();
+  ConstantInt *CInt;
+  CInt = ConstantInt::get(C, APInt(1, Inlined ? 1: 0));
+  MDNode* N = MDNode::get(C, ConstantAsMetadata::get(CInt));
+  CI->setMetadata("yk.intrinsic.inlined", N);
+}
+
 bool X86FastISel::fastLowerIntrinsicCall(const IntrinsicInst *II) {
   // FIXME: Handle more intrinsics.
   switch (II->getIntrinsicID()) {
@@ -2725,6 +2736,7 @@ bool X86FastISel::fastLowerIntrinsicCall(const IntrinsicInst *II) {
       // without a call if possible.
       uint64_t Len = cast<ConstantInt>(MCI->getLength())->getZExtValue();
       if (IsMemcpySmall(Len)) {
+        annotateIntrinsic(II, true);
         X86AddressMode DestAM, SrcAM;
         if (!X86SelectAddress(MCI->getRawDest(), DestAM) ||
             !X86SelectAddress(MCI->getRawSource(), SrcAM))
@@ -2741,6 +2753,7 @@ bool X86FastISel::fastLowerIntrinsicCall(const IntrinsicInst *II) {
     if (MCI->getSourceAddressSpace() > 255 || MCI->getDestAddressSpace() > 255)
       return false;
 
+    annotateIntrinsic(II, false);
     return lowerCallTo(II, "memcpy", II->getNumArgOperands() - 1);
   }
   case Intrinsic::memset: {