Index: llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.h
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.h
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.h
@@ -78,7 +78,6 @@
   bool shouldCombineMemoryType(EVT VT) const;
   SDValue performLoadCombine(SDNode *N, DAGCombinerInfo &DCI) const;
   SDValue performStoreCombine(SDNode *N, DAGCombinerInfo &DCI) const;
-  SDValue performClampCombine(SDNode *N, DAGCombinerInfo &DCI) const;
   SDValue performAssertSZExtCombine(SDNode *N, DAGCombinerInfo &DCI) const;
 
   SDValue splitBinaryBitConstantOpImpl(DAGCombinerInfo &DCI, const SDLoc &SL,
Index: llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -2898,28 +2898,6 @@
                       SN->getBasePtr(), SN->getMemOperand());
 }
 
-SDValue AMDGPUTargetLowering::performClampCombine(SDNode *N,
-                                                  DAGCombinerInfo &DCI) const {
-  ConstantFPSDNode *CSrc = dyn_cast<ConstantFPSDNode>(N->getOperand(0));
-  if (!CSrc)
-    return SDValue();
-
-  const APFloat &F = CSrc->getValueAPF();
-  APFloat Zero = APFloat::getZero(F.getSemantics());
-  APFloat::cmpResult Cmp0 = F.compare(Zero);
-  if (Cmp0 == APFloat::cmpLessThan ||
-      (Cmp0 == APFloat::cmpUnordered && Subtarget->enableDX10Clamp())) {
-    return DCI.DAG.getConstantFP(Zero, SDLoc(N), N->getValueType(0));
-  }
-
-  APFloat One(F.getSemantics(), "1.0");
-  APFloat::cmpResult Cmp1 = F.compare(One);
-  if (Cmp1 == APFloat::cmpGreaterThan)
-    return DCI.DAG.getConstantFP(One, SDLoc(N), N->getValueType(0));
-
-  return SDValue(CSrc, 0);
-}
-
 // FIXME: This should go in generic DAG combiner with an isTruncateFree check,
 // but isTruncateFree is inaccurate for i16 now because of SALU vs. VALU
 // issues.
@@ -3960,8 +3938,6 @@
     return performLoadCombine(N, DCI);
   case ISD::STORE:
     return performStoreCombine(N, DCI);
-  case AMDGPUISD::CLAMP:
-    return performClampCombine(N, DCI);
   case AMDGPUISD::RCP: {
     if (const auto *CFP = dyn_cast<ConstantFPSDNode>(N->getOperand(0))) {
       // XXX - Should this flush denormals?
Index: llvm/trunk/lib/Target/AMDGPU/AMDGPUInstrInfo.td
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUInstrInfo.td
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUInstrInfo.td
@@ -168,8 +168,6 @@
   [SDNPCommutative, SDNPAssociative]
 >;
 
-def AMDGPUclamp : SDNode<"AMDGPUISD::CLAMP", SDTFPUnaryOp>;
-
 // out = min(a, b) a and b are floats, where a nan comparison fails.
 def AMDGPUfmin_legacy : SDNode<"AMDGPUISD::FMIN_LEGACY", SDTFPBinOp,
   []
Index: llvm/trunk/lib/Target/AMDGPU/R600ISelLowering.cpp
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/R600ISelLowering.cpp
+++ llvm/trunk/lib/Target/AMDGPU/R600ISelLowering.cpp
@@ -287,13 +287,6 @@
       return AMDGPUTargetLowering::EmitInstrWithCustomInserter(MI, BB);
     }
     break;
-  case AMDGPU::CLAMP_R600: {
-    MachineInstr *NewMI = TII->buildDefaultInstruction(
-        *BB, I, AMDGPU::MOV, MI.getOperand(0).getReg(),
-        MI.getOperand(1).getReg());
-    TII->addFlag(*NewMI, 0, MO_FLAG_CLAMP);
-    break;
-  }
 
   case AMDGPU::FABS_R600: {
     MachineInstr *NewMI = TII->buildDefaultInstruction(
@@ -2180,20 +2173,6 @@
       if (FoldOperand(Node, i, Src, FakeOp, FakeOp, FakeOp, FakeOp, DAG))
         return DAG.getMachineNode(Opcode, SDLoc(Node), Node->getVTList(), Ops);
     }
-  } else if (Opcode == AMDGPU::CLAMP_R600) {
-    SDValue Src = Node->getOperand(0);
-    if (!Src.isMachineOpcode() ||
-        !TII->hasInstrModifiers(Src.getMachineOpcode()))
-      return Node;
-    int ClampIdx = TII->getOperandIdx(Src.getMachineOpcode(),
-        AMDGPU::OpName::clamp);
-    if (ClampIdx < 0)
-      return Node;
-    SDLoc DL(Node);
-    std::vector<SDValue> Ops(Src->op_begin(), Src->op_end());
-    Ops[ClampIdx - 1] = DAG.getTargetConstant(1, DL, MVT::i32);
-    return DAG.getMachineNode(Src.getMachineOpcode(), DL,
-                              Node->getVTList(), Ops);
   } else {
     if (!TII->hasInstrModifiers(Opcode))
       return Node;
Index: llvm/trunk/lib/Target/AMDGPU/R600Instructions.td
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/R600Instructions.td
+++ llvm/trunk/lib/Target/AMDGPU/R600Instructions.td
@@ -663,13 +663,6 @@
 
 let usesCustomInserter = 1  in {
 
-class CLAMP <RegisterClass rc> : AMDGPUShaderInst <
-  (outs rc:$dst),
-  (ins rc:$src0),
-  "CLAMP $dst, $src0",
-  [(set f32:$dst, (AMDGPUclamp f32:$src0))]
->;
-
 class FABS <RegisterClass rc> : AMDGPUShaderInst <
   (outs rc:$dst),
   (ins rc:$src0),
@@ -1194,7 +1187,6 @@
   let Itinerary = TransALU;
 }
 
-def CLAMP_R600 :  CLAMP <R600_Reg32>;
 def FABS_R600 : FABS<R600_Reg32>;
 def FNEG_R600 : FNEG<R600_Reg32>;
 
Index: llvm/trunk/lib/Target/AMDGPU/SIISelLowering.h
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIISelLowering.h
+++ llvm/trunk/lib/Target/AMDGPU/SIISelLowering.h
@@ -131,6 +131,7 @@
   SDValue performFSubCombine(SDNode *N, DAGCombinerInfo &DCI) const;
   SDValue performSetCCCombine(SDNode *N, DAGCombinerInfo &DCI) const;
   SDValue performCvtF32UByteNCombine(SDNode *N, DAGCombinerInfo &DCI) const;
+  SDValue performClampCombine(SDNode *N, DAGCombinerInfo &DCI) const;
 
   bool isLegalFlatAddressingMode(const AddrMode &AM) const;
   bool isLegalGlobalAddressingMode(const AddrMode &AM) const;
Index: llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp
+++ llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -7087,6 +7087,29 @@
   return SDValue();
 }
 
+SDValue SITargetLowering::performClampCombine(SDNode *N,
+                                              DAGCombinerInfo &DCI) const {
+  ConstantFPSDNode *CSrc = dyn_cast<ConstantFPSDNode>(N->getOperand(0));
+  if (!CSrc)
+    return SDValue();
+
+  const APFloat &F = CSrc->getValueAPF();
+  APFloat Zero = APFloat::getZero(F.getSemantics());
+  APFloat::cmpResult Cmp0 = F.compare(Zero);
+  if (Cmp0 == APFloat::cmpLessThan ||
+      (Cmp0 == APFloat::cmpUnordered && Subtarget->enableDX10Clamp())) {
+    return DCI.DAG.getConstantFP(Zero, SDLoc(N), N->getValueType(0));
+  }
+
+  APFloat One(F.getSemantics(), "1.0");
+  APFloat::cmpResult Cmp1 = F.compare(One);
+  if (Cmp1 == APFloat::cmpGreaterThan)
+    return DCI.DAG.getConstantFP(One, SDLoc(N), N->getValueType(0));
+
+  return SDValue(CSrc, 0);
+}
+
+
 SDValue SITargetLowering::PerformDAGCombine(SDNode *N,
                                             DAGCombinerInfo &DCI) const {
   switch (N->getOpcode()) {
@@ -7179,6 +7202,8 @@
     return performFMed3Combine(N, DCI);
   case AMDGPUISD::CVT_PKRTZ_F16_F32:
     return performCvtPkRTZCombine(N, DCI);
+  case AMDGPUISD::CLAMP:
+    return performClampCombine(N, DCI);
   case ISD::SCALAR_TO_VECTOR: {
     SelectionDAG &DAG = DCI.DAG;
     EVT VT = N->getValueType(0);
Index: llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.td
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.td
+++ llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.td
@@ -33,6 +33,8 @@
 // SI DAG Nodes
 //===----------------------------------------------------------------------===//
 
+def AMDGPUclamp : SDNode<"AMDGPUISD::CLAMP", SDTFPUnaryOp>;
+
 def SIload_constant : SDNode<"AMDGPUISD::LOAD_CONSTANT",
   SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisVT<1, v4i32>, SDTCisVT<2, i32>]>,
                       [SDNPMayLoad, SDNPMemOperand]