X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTarget%2FR600%2FAMDGPUISelLowering.h;h=5001116f5a06d1721cd88472de9a5b10bcc6e2f3;hb=00e08fcaa02286dd7da9cf9a8d158545532ab832;hp=c2a79ea99959fe97301a52403c1ca9def372763e;hpb=17ea10cb792832c99677afa13b9b866098bc4679;p=oota-llvm.git diff --git a/lib/Target/R600/AMDGPUISelLowering.h b/lib/Target/R600/AMDGPUISelLowering.h index c2a79ea9995..5001116f5a0 100644 --- a/lib/Target/R600/AMDGPUISelLowering.h +++ b/lib/Target/R600/AMDGPUISelLowering.h @@ -13,95 +13,158 @@ // //===----------------------------------------------------------------------===// -#ifndef AMDGPUISELLOWERING_H -#define AMDGPUISELLOWERING_H +#ifndef LLVM_LIB_TARGET_R600_AMDGPUISELLOWERING_H +#define LLVM_LIB_TARGET_R600_AMDGPUISELLOWERING_H #include "llvm/Target/TargetLowering.h" namespace llvm { +class AMDGPUMachineFunction; +class AMDGPUSubtarget; class MachineRegisterInfo; class AMDGPUTargetLowering : public TargetLowering { +protected: + const AMDGPUSubtarget *Subtarget; + private: + SDValue LowerConstantInitializer(const Constant* Init, const GlobalValue *GV, + const SDValue &InitPtr, + SDValue Chain, + SelectionDAG &DAG) const; + SDValue LowerFrameIndex(SDValue Op, SelectionDAG &DAG) const; + SDValue LowerEXTRACT_SUBVECTOR(SDValue Op, SelectionDAG &DAG) const; + SDValue LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) const; SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const; + /// \brief Lower vector stores by merging the vector elements into an integer + /// of the same bitwidth. + SDValue MergeVectorStore(const SDValue &Op, SelectionDAG &DAG) const; + /// \brief Split a vector store into multiple scalar stores. + /// \returns The resulting chain. + SDValue LowerUDIVREM(SDValue Op, SelectionDAG &DAG) const; + SDValue LowerFCEIL(SDValue Op, SelectionDAG &DAG) const; + SDValue LowerFTRUNC(SDValue Op, SelectionDAG &DAG) const; + SDValue LowerFRINT(SDValue Op, SelectionDAG &DAG) const; + SDValue LowerFNEARBYINT(SDValue Op, SelectionDAG &DAG) const; + SDValue LowerFFLOOR(SDValue Op, SelectionDAG &DAG) const; + + SDValue LowerUINT_TO_FP(SDValue Op, SelectionDAG &DAG) const; + + SDValue ExpandSIGN_EXTEND_INREG(SDValue Op, + unsigned BitsDiff, + SelectionDAG &DAG) const; + SDValue LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const; + + SDValue performStoreCombine(SDNode *N, DAGCombinerInfo &DCI) const; + SDValue performMulCombine(SDNode *N, DAGCombinerInfo &DCI) const; protected: + static EVT getEquivalentMemType(LLVMContext &Context, EVT VT); + static EVT getEquivalentLoadRegType(LLVMContext &Context, EVT VT); - /// \brief Helper function that adds Reg to the LiveIn list of the DAG's - /// MachineFunction. - /// - /// \returns a RegisterSDNode representing Reg. - SDValue CreateLiveInRegister(SelectionDAG &DAG, const TargetRegisterClass *RC, - unsigned Reg, EVT VT) const; + virtual SDValue LowerGlobalAddress(AMDGPUMachineFunction *MFI, SDValue Op, + SelectionDAG &DAG) const; + + /// \brief Split a vector load into a scalar load of each component. + SDValue ScalarizeVectorLoad(SDValue Op, SelectionDAG &DAG) const; + + /// \brief Split a vector load into 2 loads of half the vector. + SDValue SplitVectorLoad(SDValue Op, SelectionDAG &DAG) const; + /// \brief Split a vector store into a scalar store of each component. + SDValue ScalarizeVectorStore(SDValue Op, SelectionDAG &DAG) const; + + /// \brief Split a vector store into 2 stores of half the vector. + SDValue SplitVectorStore(SDValue Op, SelectionDAG &DAG) const; + + SDValue LowerLOAD(SDValue Op, SelectionDAG &DAG) const; + SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG) const; + SDValue LowerSDIVREM(SDValue Op, SelectionDAG &DAG) const; + SDValue LowerDIVREM24(SDValue Op, SelectionDAG &DAG, bool sign) const; bool isHWTrueValue(SDValue Op) const; bool isHWFalseValue(SDValue Op) const; + /// The SelectionDAGBuilder will automatically promote function arguments + /// with illegal types. However, this does not work for the AMDGPU targets + /// since the function arguments are stored in memory as these illegal types. + /// In order to handle this properly we need to get the origianl types sizes + /// from the LLVM IR Function and fixup the ISD:InputArg values before + /// passing them to AnalyzeFormalArguments() + void getOriginalFunctionArgs(SelectionDAG &DAG, + const Function *F, + const SmallVectorImpl &Ins, + SmallVectorImpl &OrigIns) const; void AnalyzeFormalArguments(CCState &State, const SmallVectorImpl &Ins) const; public: AMDGPUTargetLowering(TargetMachine &TM); - virtual SDValue LowerReturn(SDValue Chain, CallingConv::ID CallConv, - bool isVarArg, - const SmallVectorImpl &Outs, - const SmallVectorImpl &OutVals, - DebugLoc DL, SelectionDAG &DAG) const; - virtual SDValue LowerCall(CallLoweringInfo &CLI, - SmallVectorImpl &InVals) const { - CLI.Callee.dump(); - llvm_unreachable("Undefined function"); - } + bool isFAbsFree(EVT VT) const override; + bool isFNegFree(EVT VT) const override; + bool isTruncateFree(EVT Src, EVT Dest) const override; + bool isTruncateFree(Type *Src, Type *Dest) const override; + + bool isZExtFree(Type *Src, Type *Dest) const override; + bool isZExtFree(EVT Src, EVT Dest) const override; + bool isZExtFree(SDValue Val, EVT VT2) const override; + + bool isNarrowingProfitable(EVT VT1, EVT VT2) const override; + + MVT getVectorIdxTy() const override; + bool isSelectSupported(SelectSupportKind) const override; + + bool isFPImmLegal(const APFloat &Imm, EVT VT) const override; + bool ShouldShrinkFPConstant(EVT VT) const override; + + bool isLoadBitCastBeneficial(EVT, EVT) const override; + SDValue LowerReturn(SDValue Chain, CallingConv::ID CallConv, + bool isVarArg, + const SmallVectorImpl &Outs, + const SmallVectorImpl &OutVals, + SDLoc DL, SelectionDAG &DAG) const override; + SDValue LowerCall(CallLoweringInfo &CLI, + SmallVectorImpl &InVals) const override; + + SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const override; + SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const override; + void ReplaceNodeResults(SDNode * N, + SmallVectorImpl &Results, + SelectionDAG &DAG) const override; - virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const; SDValue LowerIntrinsicIABS(SDValue Op, SelectionDAG &DAG) const; SDValue LowerIntrinsicLRP(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerMinMax(SDValue Op, SelectionDAG &DAG) const; - virtual const char* getTargetNodeName(unsigned Opcode) const; + SDValue CombineMinMax(SDNode *N, SelectionDAG &DAG) const; + const char* getTargetNodeName(unsigned Opcode) const override; - virtual SDNode *PostISelFolding(MachineSDNode *N, SelectionDAG &DAG) const { + virtual SDNode *PostISelFolding(MachineSDNode *N, + SelectionDAG &DAG) const { return N; } -// Functions defined in AMDILISelLowering.cpp -public: - /// \brief Determine which of the bits specified in \p Mask are known to be /// either zero or one and return them in the \p KnownZero and \p KnownOne /// bitsets. - virtual void computeMaskedBitsForTargetNode(const SDValue Op, - APInt &KnownZero, - APInt &KnownOne, - const SelectionDAG &DAG, - unsigned Depth = 0) const; - - virtual bool getTgtMemIntrinsic(IntrinsicInfo &Info, - const CallInst &I, unsigned Intrinsic) const; + void computeKnownBitsForTargetNode(const SDValue Op, + APInt &KnownZero, + APInt &KnownOne, + const SelectionDAG &DAG, + unsigned Depth = 0) const override; - /// We want to mark f32/f64 floating point values as legal. - bool isFPImmLegal(const APFloat &Imm, EVT VT) const; + virtual unsigned ComputeNumSignBitsForTargetNode( + SDValue Op, + const SelectionDAG &DAG, + unsigned Depth = 0) const override; - /// We don't want to shrink f64/f32 constants. - bool ShouldShrinkFPConstant(EVT VT) const; - -private: - void InitAMDILLowering(); - SDValue LowerSREM(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerSREM8(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerSREM16(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerSREM32(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerSREM64(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerSDIV(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerSDIV24(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerSDIV32(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerSDIV64(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const; - EVT genIntType(uint32_t size = 32, uint32_t numEle = 1) const; - SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const; + /// \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; }; namespace AMDGPUISD { @@ -111,14 +174,17 @@ enum { FIRST_NUMBER = ISD::BUILTIN_OP_END, CALL, // Function call based on a single integer UMUL, // 32bit unsigned multiplication - DIV_INF, // Divide with infinity returned on zero divisor RET_FLAG, BRANCH_COND, // End AMDIL ISD Opcodes - BITALIGN, - BUFFER_STORE, DWORDADDR, FRACT, + CLAMP, + + // SIN_HW, COS_HW - f32 for SI, 1 ULP max error, valid from -100 pi to 100 pi. + // Denormals handled on some parts. + COS_HW, + SIN_HW, FMAX, SMAX, UMAX, @@ -126,10 +192,58 @@ enum { SMIN, UMIN, URECIP, + DIV_SCALE, + DIV_FMAS, + DIV_FIXUP, + TRIG_PREOP, // 1 ULP max error for f64 + + // RCP, RSQ - For f32, 1 ULP max error, no denormal handling. + // For f64, max error 2^29 ULP, handles denormals. + RCP, + RSQ, + RSQ_LEGACY, + RSQ_CLAMPED, + DOT4, + BFE_U32, // Extract range of bits with zero extension to 32-bits. + BFE_I32, // Extract range of bits with sign extension to 32-bits. + BFI, // (src0 & src1) | (~src0 & src2) + BFM, // Insert a range of bits into a 32-bit word. + BREV, // Reverse bits. + MUL_U24, + MUL_I24, + MAD_U24, + MAD_I24, + TEXTURE_FETCH, EXPORT, CONST_ADDRESS, REGISTER_LOAD, REGISTER_STORE, + LOAD_INPUT, + SAMPLE, + SAMPLEB, + SAMPLED, + SAMPLEL, + + // These cvt_f32_ubyte* nodes need to remain consecutive and in order. + CVT_F32_UBYTE0, + CVT_F32_UBYTE1, + CVT_F32_UBYTE2, + CVT_F32_UBYTE3, + /// This node is for VLIW targets and it is used to represent a vector + /// that is stored in consecutive registers with the same channel. + /// For example: + /// |X |Y|Z|W| + /// T0|v.x| | | | + /// T1|v.y| | | | + /// T2|v.z| | | | + /// T3|v.w| | | | + BUILD_VERTICAL_VECTOR, + /// Pointer to the start of the shader's constant data. + CONST_DATA_PTR, + FIRST_MEM_OPCODE_NUMBER = ISD::FIRST_TARGET_MEMORY_OPCODE, + STORE_MSKOR, + LOAD_CONSTANT, + TBUFFER_STORE_FORMAT, LAST_AMDGPU_ISD_NUMBER }; @@ -138,4 +252,4 @@ enum { } // End namespace llvm -#endif // AMDGPUISELLOWERING_H +#endif