Index: llvm/lib/Transforms/Utils/CodeExtractor.cpp
===================================================================
--- llvm/lib/Transforms/Utils/CodeExtractor.cpp
+++ llvm/lib/Transforms/Utils/CodeExtractor.cpp
@@ -456,13 +456,16 @@
         if (Info.LifeStart)
           return {};
         Info.LifeStart = IntrInst;
+        continue;
       }
       if (IntrInst->getIntrinsicID() == Intrinsic::lifetime_end) {
         if (Info.LifeEnd)
           return {};
         Info.LifeEnd = IntrInst;
+        continue;
       }
-      continue;
+      if (isa<DbgInfoIntrinsic>(IntrInst))
+        continue;
     }
     // Find untracked uses of the address, bail.
     if (!definedInRegion(Blocks, U))
Index: llvm/test/Transforms/CodeExtractor/LoopExtractor_alloca.ll
===================================================================
--- /dev/null
+++ llvm/test/Transforms/CodeExtractor/LoopExtractor_alloca.ll
@@ -0,0 +1,56 @@
+; RUN: opt -loop-extract -S < %s | FileCheck %s
+
+; This tests 2 cases:
+; 1. loop1 should be extracted into a function, without extracting %v1 alloca.
+; 2. loop2 should be extracted into a function, with the %v2 alloca.
+;
+; This used to produce an invalid IR, where `memcpy` will have a reference to
+; the, now, external value (local to the extracted loop function).
+
+; CHECK-LABEL: @test() {
+; CHECK-NEXT: entry:
+; CHECK-NEXT:   %v1 = alloca i32
+; CHECK-NEXT:   %p1 = bitcast i32* %v1 to i8*
+; CHECK-NEXT:   call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 undef, i8* %p1, i64 4, i1 true)
+
+; CHECK-LABEL: @test.loop2() {
+; CHECK-NEXT: newFuncRoot:
+; CHECK-NEXT:   %v2 = alloca i32
+; CHECK-NEXT:   %p2 = bitcast i32* %v2 to i8*
+
+; CHECK-LABEL: @test.loop1(i8* %p1) {
+; CHECK-NEXT: newFuncRoot:
+; CHECK-NOT:    alloca
+
+define void @test() {
+entry:
+  %v1 = alloca i32, align 4
+  %v2 = alloca i32, align 4
+  %p1 = bitcast i32* %v1 to i8*
+  %p2 = bitcast i32* %v2 to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 undef, i8* %p1, i64 4, i1 true)
+  br label %loop1
+
+loop1:
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* %p1)
+  %r1 = call i32 @foo(i8* %p1)
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* %p1)
+  %cmp1 = icmp ne i32 %r1, 0
+  br i1 %cmp1, label %loop1, label %loop2
+
+loop2:
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* %p2)
+  %r2 = call i32 @foo(i8* %p2)
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* %p2)
+  %cmp2 = icmp ne i32 %r2, 0
+  br i1 %cmp2, label %loop2, label %exit
+
+exit:
+  ret void
+}
+
+declare i32 @foo(i8*)
+
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture)
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture)
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* noalias nocapture writeonly, i8* noalias nocapture readonly, i64, i1 immarg)