Index: lib/Transforms/Scalar/NewGVN.cpp
===================================================================
--- lib/Transforms/Scalar/NewGVN.cpp
+++ lib/Transforms/Scalar/NewGVN.cpp
@@ -4094,10 +4094,13 @@
           // It's about to be alive again.
           if (LeaderUseCount == 0 && isa<Instruction>(DominatingLeader))
             ProbablyDead.erase(cast<Instruction>(DominatingLeader));
-          // Copy instructions, however, are still dead because we use their
-          // operand as the leader.
-          if (LeaderUseCount == 0 && isSSACopy)
-            ProbablyDead.insert(II);
+          // For copy instructions, we use their operand as a leader,
+          // which means we remove a user of the copy and it may become dead.
+          if (isSSACopy) {
+            unsigned &IIUseCount = UseCounts[II];
+            if (--IIUseCount == 0)
+              ProbablyDead.insert(II);
+          }
           ++LeaderUseCount;
           AnythingReplaced = true;
         }
Index: test/Transforms/NewGVN/eliminate-ssacopy.ll
===================================================================
--- /dev/null
+++ test/Transforms/NewGVN/eliminate-ssacopy.ll
@@ -0,0 +1,50 @@
+; RUN: opt -newgvn -S < %s | FileCheck %s
+
+; Make sure the created ssa copies are cleaned up.
+
+; CHECK-NOT: ssa_copy
+
+@b = external dso_local local_unnamed_addr global i32, align 4
+@a = external dso_local local_unnamed_addr global i8, align 1
+@f = external dso_local local_unnamed_addr global i16, align 2
+
+define void @g() {
+entry:
+  %tobool = icmp eq i32 undef, 0
+  br i1 %tobool, label %for.cond1thread-pre-split, label %for.cond.preheader
+
+for.cond.preheader:                               ; preds = %entry
+  unreachable
+
+for.cond1thread-pre-split:                        ; preds = %entry
+  br label %for.end4.split
+
+for.end4.split:                                   ; preds = %for.cond1thread-pre-split
+  br i1 %tobool, label %for.cond6.preheader, label %if.end11
+
+for.cond6.preheader:                              ; preds = %for.end4.split
+  br i1 undef, label %for.cond6.preheader3, label %if.end11.loopexit
+
+for.cond6.preheader3:                             ; preds = %for.cond6.preheader
+  br label %if.end11.loopexit
+
+if.end11.loopexit:                                ; preds = %for.cond6.preheader3, %for.cond6.preheader
+  %storemerge.lcssa = phi i32 [ 0, %for.cond6.preheader ], [ 1, %for.cond6.preheader3 ]
+  store i32 %storemerge.lcssa, i32* @b, align 4
+  br label %if.end11
+
+if.end11:                                         ; preds = %if.end11.loopexit, %for.end4.split
+  %0 = load i32, i32* @b, align 4
+  %1 = load i8, i8* @a, align 1
+  %conv = sext i8 %1 to i32
+  %cmp12 = icmp eq i32 %0, %conv
+  br i1 %cmp12, label %if.then14, label %if.end16
+
+if.then14:                                        ; preds = %if.end11
+  %conv15 = trunc i32 %0 to i16
+  store i16 %conv15, i16* @f, align 2
+  unreachable
+
+if.end16:                                         ; preds = %if.end11
+  ret void
+}