AMDPU: Cleanup CreateLiveInRegister

[arsenm]

This breaks an r600 test. Why does the SI version add the wrapper CopyFromReg and not r600?

[arsenm]

Add wrapper for r600 usage

[arsenm]

r305748

lib/Target/AMDGPU/AMDGPUISelLowering.h

public:
 
   unsigned ComputeNumSignBitsForTargetNode(SDValue Op, const APInt &DemandedElts,
                                            const SelectionDAG &DAG,
                                            unsigned Depth = 0) const override;
 
   /// \brief Helper function that adds Reg to the LiveIn list of the DAG's
   /// MachineFunction.
   ///
-  /// \returns a RegisterSDNode representing Reg.
-  virtual SDValue CreateLiveInRegister(SelectionDAG &DAG,
-                                       const TargetRegisterClass *RC,
-                                       unsigned Reg, EVT VT) const;
+  /// \returns a RegisterSDNode representing Reg if \p RawReg is true, otherwise
+  /// a copy from the register.
+  SDValue CreateLiveInRegister(SelectionDAG &DAG,
+                               unsigned Reg, EVT VT,
+                               const SDLoc &SL,
+                               bool RawReg = false) const;
+  SDValue CreateLiveInRegister(SelectionDAG &DAG,
+                               const TargetRegisterClass *RC,
+                               unsigned Reg, EVT VT) const {
+    return CreateLiveInRegister(DAG, RC, Reg, VT, SDLoc(DAG.getEntryNode()));
+  }
+
+  // Returns the raw live in register rather than a copy from it.
+  SDValue CreateLiveInRegisterRaw(SelectionDAG &DAG,
+                                  const TargetRegisterClass *RC,
+                                  unsigned Reg, EVT VT) const {
+    return CreateLiveInRegister(DAG, RC, Reg, VT, SDLoc(DAG.getEntryNode()), true);
+  }
 
   enum ImplicitParameter {
     FIRST_IMPLICIT,
     GRID_DIM = FIRST_IMPLICIT,
     GRID_OFFSET,
   };
 
   /// \brief Helper function that returns the byte offset of the given

lib/Target/AMDGPU/AMDGPUISelLowering.cpp

SDValue AMDGPUTargetLowering::PerformDAGCombine(SDNode *N,
   return SDValue();
 }
 
 //===----------------------------------------------------------------------===//
 // Helper functions
 //===----------------------------------------------------------------------===//
 
 SDValue AMDGPUTargetLowering::CreateLiveInRegister(SelectionDAG &DAG,
                                                    const TargetRegisterClass *RC,
-                                                   unsigned Reg, EVT VT) const {
+                                                   unsigned Reg, EVT VT,
+                                                   const SDLoc &SL,
+                                                   bool RawReg) const {
   MachineFunction &MF = DAG.getMachineFunction();
   MachineRegisterInfo &MRI = MF.getRegInfo();
-  unsigned VirtualRegister;
+  unsigned VReg;
+
   if (!MRI.isLiveIn(Reg)) {
-    VirtualRegister = MRI.createVirtualRegister(RC);
-    MRI.addLiveIn(Reg, VirtualRegister);
+    VReg = MRI.createVirtualRegister(RC);
+    MRI.addLiveIn(Reg, VReg);
   } else {
-    VirtualRegister = MRI.getLiveInVirtReg(Reg);
+    VReg = MRI.getLiveInVirtReg(Reg);
   }
-  return DAG.getRegister(VirtualRegister, VT);
+
+  if (RawReg)
+    return DAG.getRegister(VReg, VT);
+
+  return DAG.getCopyFromReg(DAG.getEntryNode(), SL, VReg, VT);
 }
 
 uint32_t AMDGPUTargetLowering::getImplicitParameterOffset(
     const AMDGPUMachineFunction *MFI, const ImplicitParameter Param) const {
   unsigned Alignment = Subtarget->getAlignmentForImplicitArgPtr();
   uint64_t ArgOffset = alignTo(MFI->getABIArgOffset(), Alignment);
   switch (Param) {
   case GRID_DIM:

lib/Target/AMDGPU/R600ISelLowering.cpp

case ISD::INTRINSIC_WO_CHAIN: {
     case Intrinsic::r600_read_local_size_x:
       return LowerImplicitParameter(DAG, VT, DL, 6);
     case Intrinsic::r600_read_local_size_y:
       return LowerImplicitParameter(DAG, VT, DL, 7);
     case Intrinsic::r600_read_local_size_z:
       return LowerImplicitParameter(DAG, VT, DL, 8);
 
     case Intrinsic::r600_read_tgid_x:
-      return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+      return CreateLiveInRegisterRaw(DAG, &AMDGPU::R600_TReg32RegClass,
                                      AMDGPU::T1_X, VT);
     case Intrinsic::r600_read_tgid_y:
-      return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+      return CreateLiveInRegisterRaw(DAG, &AMDGPU::R600_TReg32RegClass,
                                      AMDGPU::T1_Y, VT);
     case Intrinsic::r600_read_tgid_z:
-      return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+      return CreateLiveInRegisterRaw(DAG, &AMDGPU::R600_TReg32RegClass,
                                      AMDGPU::T1_Z, VT);
     case Intrinsic::r600_read_tidig_x:
-      return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+      return CreateLiveInRegisterRaw(DAG, &AMDGPU::R600_TReg32RegClass,
                                      AMDGPU::T0_X, VT);
     case Intrinsic::r600_read_tidig_y:
-      return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+      return CreateLiveInRegisterRaw(DAG, &AMDGPU::R600_TReg32RegClass,
                                      AMDGPU::T0_Y, VT);
     case Intrinsic::r600_read_tidig_z:
-      return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+      return CreateLiveInRegisterRaw(DAG, &AMDGPU::R600_TReg32RegClass,
                                      AMDGPU::T0_Z, VT);
 
     case Intrinsic::r600_recipsqrt_ieee:
       return DAG.getNode(AMDGPUISD::RSQ, DL, VT, Op.getOperand(1));
 
     case Intrinsic::r600_recipsqrt_clamped:
       return DAG.getNode(AMDGPUISD::RSQ_CLAMP, DL, VT, Op.getOperand(1));
     default:
       return Op;

lib/Target/AMDGPU/SIISelLowering.h

public:
   void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue> &Results,
                           SelectionDAG &DAG) const override;
 
   SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const override;
   SDNode *PostISelFolding(MachineSDNode *N, SelectionDAG &DAG) const override;
   void AdjustInstrPostInstrSelection(MachineInstr &MI,
                                      SDNode *Node) const override;
 
-  SDValue CreateLiveInRegister(SelectionDAG &DAG, const TargetRegisterClass *RC,
-                               unsigned Reg, EVT VT) const override;
   SDNode *legalizeTargetIndependentNode(SDNode *Node, SelectionDAG &DAG) const;
 
   MachineSDNode *wrapAddr64Rsrc(SelectionDAG &DAG, const SDLoc &DL,
                                 SDValue Ptr) const;
   MachineSDNode *buildRSRC(SelectionDAG &DAG, const SDLoc &DL, SDValue Ptr,
                            uint32_t RsrcDword1, uint64_t RsrcDword2And3) const;
   std::pair<unsigned, const TargetRegisterClass *>
   getRegForInlineAsmConstraint(const TargetRegisterInfo *TRI,

lib/Target/AMDGPU/SIISelLowering.cpp

const SDValue Ops[] = {
     DAG.getTargetConstant(AMDGPU::sub2, DL, MVT::i32),
     DataHi,
     DAG.getTargetConstant(AMDGPU::sub3, DL, MVT::i32)
   };
 
   return DAG.getMachineNode(AMDGPU::REG_SEQUENCE, DL, MVT::v4i32, Ops);
 }
 
-SDValue SITargetLowering::CreateLiveInRegister(SelectionDAG &DAG,
-                                               const TargetRegisterClass *RC,
-                                               unsigned Reg, EVT VT) const {
-  SDValue VReg = AMDGPUTargetLowering::CreateLiveInRegister(DAG, RC, Reg, VT);
-
-  return DAG.getCopyFromReg(DAG.getEntryNode(), SDLoc(DAG.getEntryNode()),
-                            cast<RegisterSDNode>(VReg)->getReg(), VT);
-}
-
 //===----------------------------------------------------------------------===//
 //                         SI Inline Assembly Support
 //===----------------------------------------------------------------------===//
 
 std::pair<unsigned, const TargetRegisterClass *>
 SITargetLowering::getRegForInlineAsmConstraint(const TargetRegisterInfo *TRI,
                                                StringRef Constraint,
                                                MVT VT) const {