//===----------------------------------------------------------------------===//
-#ifndef SIINSTRINFO_H
-#define SIINSTRINFO_H
+#ifndef LLVM_LIB_TARGET_R600_SIINSTRINFO_H
+#define LLVM_LIB_TARGET_R600_SIINSTRINFO_H
#include "AMDGPUInstrInfo.h"
+#include "SIDefines.h"
#include "SIRegisterInfo.h"
namespace llvm {
const TargetRegisterClass *RC,
const MachineOperand &Op) const;
- void splitScalar64BitOp(SmallVectorImpl<MachineInstr *> & Worklist,
- MachineInstr *Inst, unsigned Opcode) const;
+ void swapOperands(MachineBasicBlock::iterator Inst) const;
+ void splitScalar64BitUnaryOp(SmallVectorImpl<MachineInstr *> &Worklist,
+ MachineInstr *Inst, unsigned Opcode) const;
+
+ void splitScalar64BitBinaryOp(SmallVectorImpl<MachineInstr *> &Worklist,
+ MachineInstr *Inst, unsigned Opcode) const;
+
+ void splitScalar64BitBCNT(SmallVectorImpl<MachineInstr *> &Worklist,
+ MachineInstr *Inst) const;
+ void splitScalar64BitBFE(SmallVectorImpl<MachineInstr *> &Worklist,
+ MachineInstr *Inst) const;
+
+ void addDescImplicitUseDef(const MCInstrDesc &Desc, MachineInstr *MI) const;
+
+ bool checkInstOffsetsDoNotOverlap(MachineInstr *MIa,
+ MachineInstr *MIb) const;
+
+ unsigned findUsedSGPR(const MachineInstr *MI, int OpIndices[3]) const;
public:
- explicit SIInstrInfo(AMDGPUTargetMachine &tm);
+ explicit SIInstrInfo(const AMDGPUSubtarget &st);
- const SIRegisterInfo &getRegisterInfo() const {
+ const SIRegisterInfo &getRegisterInfo() const override {
return RI;
}
- virtual void copyPhysReg(MachineBasicBlock &MBB,
- MachineBasicBlock::iterator MI, DebugLoc DL,
- unsigned DestReg, unsigned SrcReg,
- bool KillSrc) const;
+ bool areLoadsFromSameBasePtr(SDNode *Load1, SDNode *Load2,
+ int64_t &Offset1,
+ int64_t &Offset2) const override;
+
+ bool getLdStBaseRegImmOfs(MachineInstr *LdSt,
+ unsigned &BaseReg, unsigned &Offset,
+ const TargetRegisterInfo *TRI) const final;
+
+ bool shouldClusterLoads(MachineInstr *FirstLdSt,
+ MachineInstr *SecondLdSt,
+ unsigned NumLoads) const final;
+
+ void copyPhysReg(MachineBasicBlock &MBB,
+ MachineBasicBlock::iterator MI, DebugLoc DL,
+ unsigned DestReg, unsigned SrcReg,
+ bool KillSrc) const override;
+
+ unsigned calculateLDSSpillAddress(MachineBasicBlock &MBB,
+ MachineBasicBlock::iterator MI,
+ RegScavenger *RS,
+ unsigned TmpReg,
+ unsigned Offset,
+ unsigned Size) const;
void storeRegToStackSlot(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
unsigned SrcReg, bool isKill, int FrameIndex,
const TargetRegisterClass *RC,
- const TargetRegisterInfo *TRI) const;
+ const TargetRegisterInfo *TRI) const override;
void loadRegFromStackSlot(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
unsigned DestReg, int FrameIndex,
const TargetRegisterClass *RC,
- const TargetRegisterInfo *TRI) const;
+ const TargetRegisterInfo *TRI) const override;
+ bool expandPostRAPseudo(MachineBasicBlock::iterator MI) const override;
+
+ // \brief Returns an opcode that can be used to move a value to a \p DstRC
+ // register. If there is no hardware instruction that can store to \p
+ // DstRC, then AMDGPU::COPY is returned.
+ unsigned getMovOpcode(const TargetRegisterClass *DstRC) const;
unsigned commuteOpcode(unsigned Opcode) const;
- virtual MachineInstr *commuteInstruction(MachineInstr *MI,
- bool NewMI=false) const;
+ MachineInstr *commuteInstruction(MachineInstr *MI,
+ bool NewMI = false) const override;
+ bool findCommutedOpIndices(MachineInstr *MI,
+ unsigned &SrcOpIdx1,
+ unsigned &SrcOpIdx2) const override;
- virtual unsigned getIEQOpcode() const {
- llvm_unreachable("Unimplemented");
- }
+ bool isTriviallyReMaterializable(const MachineInstr *MI,
+ AliasAnalysis *AA = nullptr) const;
+
+ bool areMemAccessesTriviallyDisjoint(
+ MachineInstr *MIa, MachineInstr *MIb,
+ AliasAnalysis *AA = nullptr) const override;
MachineInstr *buildMovInstr(MachineBasicBlock *MBB,
MachineBasicBlock::iterator I,
- unsigned DstReg, unsigned SrcReg) const;
- virtual bool isMov(unsigned Opcode) const;
-
- virtual bool isSafeToMoveRegClassDefs(const TargetRegisterClass *RC) const;
- bool isDS(uint16_t Opcode) const;
- int isMIMG(uint16_t Opcode) const;
- int isSMRD(uint16_t Opcode) const;
- bool isVOP1(uint16_t Opcode) const;
- bool isVOP2(uint16_t Opcode) const;
- bool isVOP3(uint16_t Opcode) const;
- bool isVOPC(uint16_t Opcode) const;
+ unsigned DstReg, unsigned SrcReg) const override;
+ bool isMov(unsigned Opcode) const override;
+
+ bool isSafeToMoveRegClassDefs(const TargetRegisterClass *RC) const override;
+
+ bool isSALU(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::SALU;
+ }
+
+ bool isVALU(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::VALU;
+ }
+
+ bool isSOP1(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::SOP1;
+ }
+
+ bool isSOP2(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::SOP2;
+ }
+
+ bool isSOPC(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::SOPC;
+ }
+
+ bool isSOPK(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::SOPK;
+ }
+
+ bool isSOPP(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::SOPP;
+ }
+
+ bool isVOP1(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::VOP1;
+ }
+
+ bool isVOP2(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::VOP2;
+ }
+
+ bool isVOP3(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::VOP3;
+ }
+
+ bool isVOPC(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::VOPC;
+ }
+
+ bool isMUBUF(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::MUBUF;
+ }
+
+ bool isMTBUF(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::MTBUF;
+ }
+
+ bool isSMRD(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::SMRD;
+ }
+
+ bool isDS(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::DS;
+ }
+
+ bool isMIMG(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::MIMG;
+ }
+
+ bool isFLAT(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::FLAT;
+ }
+
+ bool isWQM(uint16_t Opcode) const {
+ return get(Opcode).TSFlags & SIInstrFlags::WQM;
+ }
+
+ bool isInlineConstant(const APInt &Imm) const;
bool isInlineConstant(const MachineOperand &MO) const;
bool isLiteralConstant(const MachineOperand &MO) const;
- virtual bool verifyInstruction(const MachineInstr *MI,
- StringRef &ErrInfo) const;
+ bool isImmOperandLegal(const MachineInstr *MI, unsigned OpNo,
+ const MachineOperand &MO) const;
+
+ /// \brief Return true if the given offset Size in bytes can be folded into
+ /// the immediate offsets of a memory instruction for the given address space.
+ bool canFoldOffset(unsigned OffsetSize, unsigned AS) const;
+
+ /// \brief Return true if this 64-bit VALU instruction has a 32-bit encoding.
+ /// This function will return false if you pass it a 32-bit instruction.
+ bool hasVALU32BitEncoding(unsigned Opcode) const;
+
+ /// \brief Returns true if this operand uses the constant bus.
+ bool usesConstantBus(const MachineRegisterInfo &MRI,
+ const MachineOperand &MO) const;
+
+ /// \brief Return true if this instruction has any modifiers.
+ /// e.g. src[012]_mod, omod, clamp.
+ bool hasModifiers(unsigned Opcode) const;
+
+ bool hasModifiersSet(const MachineInstr &MI,
+ unsigned OpName) const;
+
+ bool verifyInstruction(const MachineInstr *MI,
+ StringRef &ErrInfo) const override;
- bool isSALUInstr(const MachineInstr &MI) const;
static unsigned getVALUOp(const MachineInstr &MI);
bool isSALUOpSupportedOnVALU(const MachineInstr &MI) const;
/// instead of MOV.
void legalizeOpWithMove(MachineInstr *MI, unsigned OpIdx) const;
+ /// \brief Check if \p MO is a legal operand if it was the \p OpIdx Operand
+ /// for \p MI.
+ bool isOperandLegal(const MachineInstr *MI, unsigned OpIdx,
+ const MachineOperand *MO = nullptr) const;
+
/// \brief Legalize all operands in this instruction. This function may
/// create new instruction and insert them before \p MI.
void legalizeOperands(MachineInstr *MI) const;
+ /// \brief Split an SMRD instruction into two smaller loads of half the
+ // size storing the results in \p Lo and \p Hi.
+ void splitSMRD(MachineInstr *MI, const TargetRegisterClass *HalfRC,
+ unsigned HalfImmOp, unsigned HalfSGPROp,
+ MachineInstr *&Lo, MachineInstr *&Hi) const;
+
+ void moveSMRDToVALU(MachineInstr *MI, MachineRegisterInfo &MRI) const;
+
/// \brief Replace this instruction's opcode with the equivalent VALU
/// opcode. This function will also move the users of \p MI to the
/// VALU if necessary.
void moveToVALU(MachineInstr &MI) const;
- virtual unsigned calculateIndirectAddress(unsigned RegIndex,
- unsigned Channel) const;
+ unsigned calculateIndirectAddress(unsigned RegIndex,
+ unsigned Channel) const override;
- virtual const TargetRegisterClass *getIndirectAddrRegClass() const;
+ const TargetRegisterClass *getIndirectAddrRegClass() const override;
- virtual MachineInstrBuilder buildIndirectWrite(MachineBasicBlock *MBB,
- MachineBasicBlock::iterator I,
- unsigned ValueReg,
- unsigned Address,
- unsigned OffsetReg) const;
+ MachineInstrBuilder buildIndirectWrite(MachineBasicBlock *MBB,
+ MachineBasicBlock::iterator I,
+ unsigned ValueReg,
+ unsigned Address,
+ unsigned OffsetReg) const override;
- virtual MachineInstrBuilder buildIndirectRead(MachineBasicBlock *MBB,
- MachineBasicBlock::iterator I,
- unsigned ValueReg,
- unsigned Address,
- unsigned OffsetReg) const;
+ MachineInstrBuilder buildIndirectRead(MachineBasicBlock *MBB,
+ MachineBasicBlock::iterator I,
+ unsigned ValueReg,
+ unsigned Address,
+ unsigned OffsetReg) const override;
void reserveIndirectRegisters(BitVector &Reserved,
const MachineFunction &MF) const;
void LoadM0(MachineInstr *MoveRel, MachineBasicBlock::iterator I,
unsigned SavReg, unsigned IndexReg) const;
+
+ void insertNOPs(MachineBasicBlock::iterator MI, int Count) const;
+
+ /// \brief Returns the operand named \p Op. If \p MI does not have an
+ /// operand named \c Op, this function returns nullptr.
+ MachineOperand *getNamedOperand(MachineInstr &MI, unsigned OperandName) const;
+
+ const MachineOperand *getNamedOperand(const MachineInstr &MI,
+ unsigned OpName) const {
+ return getNamedOperand(const_cast<MachineInstr &>(MI), OpName);
+ }
+
+ uint64_t getDefaultRsrcDataFormat() const;
+
};
namespace AMDGPU {
int getVOPe64(uint16_t Opcode);
+ int getVOPe32(uint16_t Opcode);
int getCommuteRev(uint16_t Opcode);
int getCommuteOrig(uint16_t Opcode);
+ int getAddr64Inst(uint16_t Opcode);
+ int getAtomicRetOp(uint16_t Opcode);
+ int getAtomicNoRetOp(uint16_t Opcode);
const uint64_t RSRC_DATA_FORMAT = 0xf00000000000LL;
-
+ const uint64_t RSRC_TID_ENABLE = 1LL << 55;
} // End namespace AMDGPU
-} // End namespace llvm
+namespace SI {
+namespace KernelInputOffsets {
+
+/// Offsets in bytes from the start of the input buffer
+enum Offsets {
+ NGROUPS_X = 0,
+ NGROUPS_Y = 4,
+ NGROUPS_Z = 8,
+ GLOBAL_SIZE_X = 12,
+ GLOBAL_SIZE_Y = 16,
+ GLOBAL_SIZE_Z = 20,
+ LOCAL_SIZE_X = 24,
+ LOCAL_SIZE_Y = 28,
+ LOCAL_SIZE_Z = 32
+};
-namespace SIInstrFlags {
- enum Flags {
- // First 4 bits are the instruction encoding
- VM_CNT = 1 << 0,
- EXP_CNT = 1 << 1,
- LGKM_CNT = 1 << 2
- };
-}
+} // End namespace KernelInputOffsets
+} // End namespace SI
+
+} // End namespace llvm
-#endif //SIINSTRINFO_H
+#endif