Index: lib/Target/AMDGPU/R600EmitClauseMarkers.cpp
===================================================================
--- lib/Target/AMDGPU/R600EmitClauseMarkers.cpp
+++ lib/Target/AMDGPU/R600EmitClauseMarkers.cpp
@@ -196,6 +196,7 @@
                         MachineBasicBlock::iterator Def,
                         MachineBasicBlock::iterator BBEnd) {
     const R600RegisterInfo &TRI = TII->getRegisterInfo();
+    //TODO: change this to defs?
     for (MachineInstr::const_mop_iterator
            MOI = Def->operands_begin(),
            MOE = Def->operands_end(); MOI != MOE; ++MOI) {
@@ -218,15 +219,18 @@
         if (AluInstCount >= TII->getMaxAlusPerClause())
           return false;
 
+        // TODO: Is this true? kill flag appears to work OK below
         // Register kill flags have been cleared by the time we get to this
         // pass, but it is safe to assume that all uses of this register
         // occur in the same basic block as its definition, because
         // it is illegal for the scheduler to schedule them in
         // different blocks.
-        if (UseI->findRegisterUseOperandIdx(MOI->getReg()))
+        if (UseI->findRegisterUseOperandIdx(MOI->getReg()) != -1) {
           LastUseCount = AluInstCount;
+        }
 
-        if (UseI != Def && UseI->findRegisterDefOperandIdx(MOI->getReg()) != -1)
+        // Exit early if the current use kills the register
+        if (UseI != Def && UseI->findRegisterUseOperandIdx(MOI->getReg(), true) != -1)
           break;
       }
       if (LastUseCount)
Index: test/CodeGen/AMDGPU/r600.alu-limits.ll
===================================================================
--- test/CodeGen/AMDGPU/r600.alu-limits.ll
+++ test/CodeGen/AMDGPU/r600.alu-limits.ll
@@ -1,8 +1,8 @@
 ; RUN: opt -loop-unroll -unroll-threshold=2000 -S < %s | llc -march=r600 -mcpu=cypress | FileCheck %s
-; XFAIL: *
 ; REQUIRES: asserts
 
-; CHECK: {{^}}@alu_limits:
+; CHECK: {{^}}alu_limits:
+; CHECK: CF_END
 
 %struct.foo = type {i32, i32, i32}