diff --git a/lib/Target/R600/SIAnnotateControlFlow.cpp b/lib/Target/R600/SIAnnotateControlFlow.cpp
index 91eb60beb6b..79f6532f1d7 100644
--- a/lib/Target/R600/SIAnnotateControlFlow.cpp
+++ b/lib/Target/R600/SIAnnotateControlFlow.cpp
@@ -14,6 +14,7 @@
 
 #include "AMDGPU.h"
 #include "llvm/ADT/DepthFirstIterator.h"
+#include "llvm/Analysis/LoopInfo.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/Dominators.h"
 #include "llvm/IR/Instructions.h"
@@ -66,6 +67,8 @@ class SIAnnotateControlFlow : public FunctionPass {
   DominatorTree *DT;
   StackVector Stack;
 
+  LoopInfo *LI;
+
   bool isTopOfStack(BasicBlock *BB);
 
   Value *popSaved();
@@ -99,6 +102,7 @@ public:
   }
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.addRequired<LoopInfoWrapperPass>();
     AU.addRequired<DominatorTreeWrapperPass>();
     AU.addPreserved<DominatorTreeWrapperPass>();
     FunctionPass::getAnalysisUsage(AU);
@@ -277,10 +281,26 @@ void SIAnnotateControlFlow::handleLoop(BranchInst *Term) {
 
   Term->setCondition(CallInst::Create(Loop, Arg, "", Term));
   push(Term->getSuccessor(0), Arg);
-}
-
-/// \brief Close the last opened control flow
+}/// \brief Close the last opened control flow
 void SIAnnotateControlFlow::closeControlFlow(BasicBlock *BB) {
+  llvm::Loop *L = LI->getLoopFor(BB);
+
+  if (L && L->getHeader() == BB) {
+    // We can't insert an EndCF call into a loop header, because it will
+    // get executed on every iteration of the loop, when it should be
+    // executed only once before the loop.
+    SmallVector <BasicBlock*, 8> Latches;
+    L->getLoopLatches(Latches);
+
+    std::vector<BasicBlock*> Preds;
+    for (pred_iterator PI = pred_begin(BB), PE = pred_end(BB); PI != PE; ++PI) {
+      if (std::find(Latches.begin(), Latches.end(), *PI) == Latches.end())
+        Preds.push_back(*PI);
+    }
+    BB = llvm::SplitBlockPredecessors(BB, Preds, "endcf.split", nullptr, DT,
+                                      LI, false);
+  }
+
   CallInst::Create(EndCf, popSaved(), "", BB->getFirstInsertionPt());
 }
 
@@ -288,6 +308,7 @@ void SIAnnotateControlFlow::closeControlFlow(BasicBlock *BB) {
 /// recognize if/then/else and loops.
 bool SIAnnotateControlFlow::runOnFunction(Function &F) {
   DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
+  LI = &getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
 
   for (df_iterator<BasicBlock *> I = df_begin(&F.getEntryBlock()),
        E = df_end(&F.getEntryBlock()); I != E; ++I) {
diff --git a/test/CodeGen/R600/endcf-loop-header.ll b/test/CodeGen/R600/endcf-loop-header.ll
new file mode 100644
index 00000000000..e3c5b3c1c36
--- /dev/null
+++ b/test/CodeGen/R600/endcf-loop-header.ll
@@ -0,0 +1,34 @@
+; RUN: llc < %s -march=amdgcn -mcpu=SI -verify-machineinstrs | FileCheck %s
+
+; This tests that the llvm.SI.end.cf intrinsic is not inserted into the
+; loop block.  This intrinsic will be lowered to s_or_b64 by the code
+; generator.
+
+; CHECK-LABEL: {{^}}test:
+
+; This is was lowered from the llvm.SI.end.cf intrinsic:
+; CHECK: s_or_b64 exec, exec
+
+; CHECK: [[LOOP_LABEL:[0-9A-Za-z_]+]]: ; %loop{{$}}
+; CHECK-NOT: s_or_b64 exec, exec
+; CHECK: s_cbranch_execnz [[LOOP_LABEL]]
+define void @test(i32 addrspace(1)* %out, i32 %cond) {
+entry:
+  %tmp0 = icmp eq i32 %cond, 0
+  br i1 %tmp0, label %if, label %loop
+
+if:
+  store i32 0, i32 addrspace(1)* %out
+  br label %loop
+
+loop:
+  %tmp1 = phi i32 [0, %entry], [0, %if], [%inc, %loop]
+  %inc = add i32 %tmp1, %cond
+  %tmp2 = icmp ugt i32 %inc, 10
+  br i1 %tmp2, label %done, label %loop
+
+done:
+  %tmp3 = getelementptr i32 addrspace(1)* %out, i64 1
+  store i32 %inc, i32 addrspace(1)* %tmp3
+  ret void
+}