//===-- PPC32ISelDAGToDAG.cpp - PPC32 pattern matching inst selector ------===// // // The LLVM Compiler Infrastructure // // This file was developed by Chris Lattner and is distributed under // the University of Illinois Open Source License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // // This file defines a pattern matching instruction selector for 32 bit PowerPC, // converting from a legalized dag to a PPC dag. // //===----------------------------------------------------------------------===// #include "PowerPC.h" #include "PPC32TargetMachine.h" #include "PPC32ISelLowering.h" #include "llvm/CodeGen/SelectionDAG.h" #include "llvm/CodeGen/SelectionDAGISel.h" #include "llvm/Target/TargetOptions.h" #include "llvm/ADT/Statistic.h" #include "llvm/Support/Debug.h" #include "llvm/Support/MathExtras.h" using namespace llvm; namespace { Statistic<> Recorded("ppc-codegen", "Number of recording ops emitted"); Statistic<> FusedFP ("ppc-codegen", "Number of fused fp operations"); Statistic<> FrameOff("ppc-codegen", "Number of frame idx offsets collapsed"); //===--------------------------------------------------------------------===// /// PPC32DAGToDAGISel - PPC32 specific code to select PPC32 machine /// instructions for SelectionDAG operations. /// class PPC32DAGToDAGISel : public SelectionDAGISel { PPC32TargetLowering PPC32Lowering; unsigned GlobalBaseReg; bool GlobalBaseInitialized; public: PPC32DAGToDAGISel(TargetMachine &TM) : SelectionDAGISel(PPC32Lowering), PPC32Lowering(TM) {} /// runOnFunction - Override this function in order to reset our /// per-function variables. virtual bool runOnFunction(Function &Fn) { // Make sure we re-emit a set of the global base reg if necessary GlobalBaseInitialized = false; return SelectionDAGISel::runOnFunction(Fn); } /// getI32Imm - Return a target constant with the specified value, of type /// i32. inline SDOperand getI32Imm(unsigned Imm) { return CurDAG->getTargetConstant(Imm, MVT::i32); } // Select - Convert the specified operand from a target-independent to a // target-specific node if it hasn't already been changed. SDOperand Select(SDOperand Op); SDNode *SelectIntImmediateExpr(SDOperand LHS, SDOperand RHS, unsigned OCHi, unsigned OCLo, bool IsArithmetic = false, bool Negate = false); /// InstructionSelectBasicBlock - This callback is invoked by /// SelectionDAGISel when it has created a SelectionDAG for us to codegen. virtual void InstructionSelectBasicBlock(SelectionDAG &DAG) { DEBUG(BB->dump()); // Codegen the basic block. Select(DAG.getRoot()); DAG.RemoveDeadNodes(); DAG.viewGraph(); } virtual const char *getPassName() const { return "PowerPC DAG->DAG Pattern Instruction Selection"; } }; } // isIntImmediate - This method tests to see if a constant operand. // If so Imm will receive the 32 bit value. static bool isIntImmediate(SDNode *N, unsigned& Imm) { if (N->getOpcode() == ISD::Constant) { Imm = cast(N)->getValue(); return true; } return false; } // isOprShiftImm - Returns true if the specified operand is a shift opcode with // a immediate shift count less than 32. static bool isOprShiftImm(SDNode *N, unsigned& Opc, unsigned& SH) { Opc = N->getOpcode(); return (Opc == ISD::SHL || Opc == ISD::SRL || Opc == ISD::SRA) && isIntImmediate(N->getOperand(1).Val, SH) && SH < 32; } // isRunOfOnes - Returns true iff Val consists of one contiguous run of 1s with // any number of 0s on either side. The 1s are allowed to wrap from LSB to // MSB, so 0x000FFF0, 0x0000FFFF, and 0xFF0000FF are all runs. 0x0F0F0000 is // not, since all 1s are not contiguous. static bool isRunOfOnes(unsigned Val, unsigned &MB, unsigned &ME) { if (isShiftedMask_32(Val)) { // look for the first non-zero bit MB = CountLeadingZeros_32(Val); // look for the first zero bit after the run of ones ME = CountLeadingZeros_32((Val - 1) ^ Val); return true; } else if (isShiftedMask_32(Val = ~Val)) { // invert mask // effectively look for the first zero bit ME = CountLeadingZeros_32(Val) - 1; // effectively look for the first one bit after the run of zeros MB = CountLeadingZeros_32((Val - 1) ^ Val) + 1; return true; } // no run present return false; } // isRotateAndMask - Returns true if Mask and Shift can be folded in to a rotate // and mask opcode and mask operation. static bool isRotateAndMask(SDNode *N, unsigned Mask, bool IsShiftMask, unsigned &SH, unsigned &MB, unsigned &ME) { unsigned Shift = 32; unsigned Indeterminant = ~0; // bit mask marking indeterminant results unsigned Opcode = N->getOpcode(); if (!isIntImmediate(N->getOperand(1).Val, Shift) || (Shift > 31)) return false; if (Opcode == ISD::SHL) { // apply shift left to mask if it comes first if (IsShiftMask) Mask = Mask << Shift; // determine which bits are made indeterminant by shift Indeterminant = ~(0xFFFFFFFFu << Shift); } else if (Opcode == ISD::SRA || Opcode == ISD::SRL) { // apply shift right to mask if it comes first if (IsShiftMask) Mask = Mask >> Shift; // determine which bits are made indeterminant by shift Indeterminant = ~(0xFFFFFFFFu >> Shift); // adjust for the left rotate Shift = 32 - Shift; } else { return false; } // if the mask doesn't intersect any Indeterminant bits if (Mask && !(Mask & Indeterminant)) { SH = Shift; // make sure the mask is still a mask (wrap arounds may not be) return isRunOfOnes(Mask, MB, ME); } return false; } // isOpcWithIntImmediate - This method tests to see if the node is a specific // opcode and that it has a immediate integer right operand. // If so Imm will receive the 32 bit value. static bool isOpcWithIntImmediate(SDNode *N, unsigned Opc, unsigned& Imm) { return N->getOpcode() == Opc && isIntImmediate(N->getOperand(1).Val, Imm); } // isOprNot - Returns true if the specified operand is an xor with immediate -1. static bool isOprNot(SDNode *N) { unsigned Imm; return isOpcWithIntImmediate(N, ISD::XOR, Imm) && (signed)Imm == -1; } // Immediate constant composers. // Lo16 - grabs the lo 16 bits from a 32 bit constant. // Hi16 - grabs the hi 16 bits from a 32 bit constant. // HA16 - computes the hi bits required if the lo bits are add/subtracted in // arithmethically. static unsigned Lo16(unsigned x) { return x & 0x0000FFFF; } static unsigned Hi16(unsigned x) { return Lo16(x >> 16); } static unsigned HA16(unsigned x) { return Hi16((signed)x - (signed short)x); } // isIntImmediate - This method tests to see if a constant operand. // If so Imm will receive the 32 bit value. static bool isIntImmediate(SDOperand N, unsigned& Imm) { if (ConstantSDNode *CN = dyn_cast(N)) { Imm = (unsigned)CN->getSignExtended(); return true; } return false; } // SelectIntImmediateExpr - Choose code for integer operations with an immediate // operand. SDNode *PPC32DAGToDAGISel::SelectIntImmediateExpr(SDOperand LHS, SDOperand RHS, unsigned OCHi, unsigned OCLo, bool IsArithmetic, bool Negate) { // Check to make sure this is a constant. ConstantSDNode *CN = dyn_cast(RHS); // Exit if not a constant. if (!CN) return 0; // Extract immediate. unsigned C = (unsigned)CN->getValue(); // Negate if required (ISD::SUB). if (Negate) C = -C; // Get the hi and lo portions of constant. unsigned Hi = IsArithmetic ? HA16(C) : Hi16(C); unsigned Lo = Lo16(C); // If two instructions are needed and usage indicates it would be better to // load immediate into a register, bail out. if (Hi && Lo && CN->use_size() > 2) return false; // Select the first operand. SDOperand Opr0 = Select(LHS); if (Lo) // Add in the lo-part. Opr0 = CurDAG->getTargetNode(OCLo, MVT::i32, Opr0, getI32Imm(Lo)); if (Hi) // Add in the hi-part. Opr0 = CurDAG->getTargetNode(OCHi, MVT::i32, Opr0, getI32Imm(Hi)); return Opr0.Val; } // Select - Convert the specified operand from a target-independent to a // target-specific node if it hasn't already been changed. SDOperand PPC32DAGToDAGISel::Select(SDOperand Op) { SDNode *N = Op.Val; if (N->getOpcode() >= ISD::BUILTIN_OP_END) return Op; // Already selected. switch (N->getOpcode()) { default: std::cerr << "Cannot yet select: "; N->dump(); std::cerr << "\n"; abort(); case ISD::EntryToken: // These leaves remain the same. case ISD::UNDEF: return Op; case ISD::TokenFactor: { SDOperand New; if (N->getNumOperands() == 2) { SDOperand Op0 = Select(N->getOperand(0)); SDOperand Op1 = Select(N->getOperand(1)); New = CurDAG->getNode(ISD::TokenFactor, MVT::Other, Op0, Op1); } else { std::vector Ops; for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) Ops.push_back(Select(N->getOperand(0))); New = CurDAG->getNode(ISD::TokenFactor, MVT::Other, Ops); } if (New.Val != N) { CurDAG->ReplaceAllUsesWith(N, New.Val); N = New.Val; } break; } case ISD::CopyFromReg: { SDOperand Chain = Select(N->getOperand(0)); if (Chain == N->getOperand(0)) return Op; // No change SDOperand New = CurDAG->getCopyFromReg(Chain, cast(N->getOperand(1))->getReg(), N->getValueType(0)); return New.getValue(Op.ResNo); } case ISD::CopyToReg: { SDOperand Chain = Select(N->getOperand(0)); SDOperand Reg = N->getOperand(1); SDOperand Val = Select(N->getOperand(2)); if (Chain != N->getOperand(0) || Val != N->getOperand(2)) { SDOperand New = CurDAG->getNode(ISD::CopyToReg, MVT::Other, Chain, Reg, Val); CurDAG->ReplaceAllUsesWith(N, New.Val); N = New.Val; } break; } case ISD::Constant: { assert(N->getValueType(0) == MVT::i32); unsigned v = (unsigned)cast(N)->getValue(); if ((unsigned)(short)v == v) { CurDAG->SelectNodeTo(N, MVT::i32, PPC::LI, getI32Imm(v)); break; } else { SDOperand Top = CurDAG->getTargetNode(PPC::LIS, MVT::i32, getI32Imm(unsigned(v) >> 16)); CurDAG->SelectNodeTo(N, MVT::i32, PPC::ORI, Top, getI32Imm(v & 0xFFFF)); break; } } case ISD::SIGN_EXTEND_INREG: switch(cast(N->getOperand(1))->getVT()) { default: assert(0 && "Illegal type in SIGN_EXTEND_INREG"); break; case MVT::i16: CurDAG->SelectNodeTo(N, MVT::i32, PPC::EXTSH, Select(N->getOperand(0))); break; case MVT::i8: CurDAG->SelectNodeTo(N, MVT::i32, PPC::EXTSB, Select(N->getOperand(0))); break; } break; case ISD::CTLZ: assert(N->getValueType(0) == MVT::i32); CurDAG->SelectNodeTo(N, MVT::i32, PPC::CNTLZW, Select(N->getOperand(0))); break; case ISD::ADD: { MVT::ValueType Ty = N->getValueType(0); if (Ty == MVT::i32) { if (SDNode *I = SelectIntImmediateExpr(N->getOperand(0), N->getOperand(1), PPC::ADDIS, PPC::ADDI, true)) { CurDAG->ReplaceAllUsesWith(N, I); N = I; } else { CurDAG->SelectNodeTo(N, Ty, PPC::ADD, Select(N->getOperand(0)), Select(N->getOperand(1))); } break; } if (!NoExcessFPPrecision) { // Match FMA ops if (N->getOperand(0).getOpcode() == ISD::MUL && N->getOperand(0).Val->hasOneUse()) { ++FusedFP; // Statistic CurDAG->SelectNodeTo(N, Ty, Ty == MVT::f64 ? PPC::FMADD : PPC::FMADDS, Select(N->getOperand(0).getOperand(0)), Select(N->getOperand(0).getOperand(1)), Select(N->getOperand(1))); break; } else if (N->getOperand(1).getOpcode() == ISD::MUL && N->getOperand(1).hasOneUse()) { ++FusedFP; // Statistic CurDAG->SelectNodeTo(N, Ty, Ty == MVT::f64 ? PPC::FMADD : PPC::FMADDS, Select(N->getOperand(1).getOperand(0)), Select(N->getOperand(1).getOperand(1)), Select(N->getOperand(0))); break; } } CurDAG->SelectNodeTo(N, Ty, Ty == MVT::f64 ? PPC::FADD : PPC::FADDS, Select(N->getOperand(0)), Select(N->getOperand(1))); break; } case ISD::SUB: { MVT::ValueType Ty = N->getValueType(0); if (Ty == MVT::i32) { unsigned Imm; if (isIntImmediate(N->getOperand(0), Imm) && isInt16(Imm)) { CurDAG->SelectNodeTo(N, Ty, PPC::SUBFIC, Select(N->getOperand(1)), getI32Imm(Lo16(Imm))); break; } if (SDNode *I = SelectIntImmediateExpr(N->getOperand(0), N->getOperand(1), PPC::ADDIS, PPC::ADDI, true, true)) { CurDAG->ReplaceAllUsesWith(N, I); N = I; } else { CurDAG->SelectNodeTo(N, Ty, PPC::SUBF, Select(N->getOperand(1)), Select(N->getOperand(0))); } break; } if (!NoExcessFPPrecision) { // Match FMA ops if (N->getOperand(0).getOpcode() == ISD::MUL && N->getOperand(0).Val->hasOneUse()) { ++FusedFP; // Statistic CurDAG->SelectNodeTo(N, Ty, Ty == MVT::f64 ? PPC::FMSUB : PPC::FMSUBS, Select(N->getOperand(0).getOperand(0)), Select(N->getOperand(0).getOperand(1)), Select(N->getOperand(1))); break; } else if (N->getOperand(1).getOpcode() == ISD::MUL && N->getOperand(1).Val->hasOneUse()) { ++FusedFP; // Statistic CurDAG->SelectNodeTo(N, Ty, Ty == MVT::f64 ? PPC::FNMSUB : PPC::FNMSUBS, Select(N->getOperand(1).getOperand(0)), Select(N->getOperand(1).getOperand(1)), Select(N->getOperand(0))); break; } } CurDAG->SelectNodeTo(N, Ty, Ty == MVT::f64 ? PPC::FSUB : PPC::FSUBS, Select(N->getOperand(0)), Select(N->getOperand(1))); break; } case ISD::MUL: { unsigned Imm, Opc; if (isIntImmediate(N->getOperand(1), Imm) && isInt16(Imm)) { CurDAG->SelectNodeTo(N, N->getValueType(0), PPC::MULLI, Select(N->getOperand(0)), getI32Imm(Lo16(Imm))); break; } switch (N->getValueType(0)) { default: assert(0 && "Unhandled multiply type!"); case MVT::i32: Opc = PPC::MULLW; break; case MVT::f32: Opc = PPC::FMULS; break; case MVT::f64: Opc = PPC::FMUL; break; } CurDAG->SelectNodeTo(N, N->getValueType(0), Opc, Select(N->getOperand(0)), Select(N->getOperand(1))); break; } case ISD::MULHS: assert(N->getValueType(0) == MVT::i32); CurDAG->SelectNodeTo(N, MVT::i32, PPC::MULHW, Select(N->getOperand(0)), Select(N->getOperand(1))); break; case ISD::MULHU: assert(N->getValueType(0) == MVT::i32); CurDAG->SelectNodeTo(N, MVT::i32, PPC::MULHWU, Select(N->getOperand(0)), Select(N->getOperand(1))); break; case ISD::AND: { unsigned Imm, SH, MB, ME; // If this is an and of a value rotated between 0 and 31 bits and then and'd // with a mask, emit rlwinm if (isIntImmediate(N->getOperand(1), Imm) && (isShiftedMask_32(Imm) || isShiftedMask_32(~Imm))) { SDOperand Val; if (isRotateAndMask(N->getOperand(0).Val, Imm, false, SH, MB, ME)) { Val = Select(N->getOperand(0).getOperand(0)); } else { Val = Select(N->getOperand(0)); isRunOfOnes(Imm, MB, ME); SH = 0; } CurDAG->SelectNodeTo(N, MVT::i32, PPC::RLWINM, Val, getI32Imm(SH), getI32Imm(MB), getI32Imm(ME)); break; } // If this is an and with an immediate that isn't a mask, then codegen it as // high and low 16 bit immediate ands. if (SDNode *I = SelectIntImmediateExpr(N->getOperand(0), N->getOperand(1), PPC::ANDISo, PPC::ANDIo)) { CurDAG->ReplaceAllUsesWith(N, I); N = I; break; } // Finally, check for the case where we are being asked to select // and (not(a), b) or and (a, not(b)) which can be selected as andc. if (isOprNot(N->getOperand(0).Val)) CurDAG->SelectNodeTo(N, MVT::i32, PPC::ANDC, Select(N->getOperand(1)), Select(N->getOperand(0).getOperand(0))); else if (isOprNot(N->getOperand(1).Val)) CurDAG->SelectNodeTo(N, MVT::i32, PPC::ANDC, Select(N->getOperand(0)), Select(N->getOperand(1).getOperand(0))); else CurDAG->SelectNodeTo(N, MVT::i32, PPC::AND, Select(N->getOperand(0)), Select(N->getOperand(1))); break; } case ISD::XOR: // Check whether or not this node is a logical 'not'. This is represented // by llvm as a xor with the constant value -1 (all bits set). If this is a // 'not', then fold 'or' into 'nor', and so forth for the supported ops. if (isOprNot(N)) { unsigned Opc; SDOperand Val = Select(N->getOperand(0)); switch (Val.getTargetOpcode()) { default: Opc = 0; break; case PPC::OR: Opc = PPC::NOR; break; case PPC::AND: Opc = PPC::NAND; break; case PPC::XOR: Opc = PPC::EQV; break; } if (Opc) CurDAG->SelectNodeTo(N, MVT::i32, Opc, Val.getOperand(0), Val.getOperand(1)); else CurDAG->SelectNodeTo(N, MVT::i32, PPC::NOR, Val, Val); break; } // If this is a xor with an immediate other than -1, then codegen it as high // and low 16 bit immediate xors. if (SDNode *I = SelectIntImmediateExpr(N->getOperand(0), N->getOperand(1), PPC::XORIS, PPC::XORI)) { CurDAG->ReplaceAllUsesWith(N, I); N = I; break; } // Finally, check for the case where we are being asked to select // xor (not(a), b) which is equivalent to not(xor a, b), which is eqv if (isOprNot(N->getOperand(0).Val)) CurDAG->SelectNodeTo(N, MVT::i32, PPC::EQV, Select(N->getOperand(0).getOperand(0)), Select(N->getOperand(1))); else CurDAG->SelectNodeTo(N, MVT::i32, PPC::XOR, Select(N->getOperand(0)), Select(N->getOperand(1))); break; case ISD::FABS: CurDAG->SelectNodeTo(N, N->getValueType(0), PPC::FABS, Select(N->getOperand(0))); break; case ISD::FP_EXTEND: assert(MVT::f64 == N->getValueType(0) && MVT::f32 == N->getOperand(0).getValueType() && "Illegal FP_EXTEND"); CurDAG->SelectNodeTo(N, MVT::f64, PPC::FMR, Select(N->getOperand(0))); break; case ISD::FP_ROUND: assert(MVT::f32 == N->getValueType(0) && MVT::f64 == N->getOperand(0).getValueType() && "Illegal FP_ROUND"); CurDAG->SelectNodeTo(N, MVT::f32, PPC::FRSP, Select(N->getOperand(0))); break; case ISD::FNEG: { SDOperand Val = Select(N->getOperand(0)); MVT::ValueType Ty = N->getValueType(0); if (Val.Val->hasOneUse()) { unsigned Opc; switch (Val.getTargetOpcode()) { default: Opc = 0; break; case PPC::FABS: Opc = PPC::FNABS; break; case PPC::FMADD: Opc = PPC::FNMADD; break; case PPC::FMADDS: Opc = PPC::FNMADDS; break; case PPC::FMSUB: Opc = PPC::FNMSUB; break; case PPC::FMSUBS: Opc = PPC::FNMSUBS; break; } // If we inverted the opcode, then emit the new instruction with the // inverted opcode and the original instruction's operands. Otherwise, // fall through and generate a fneg instruction. if (Opc) { if (PPC::FNABS == Opc) CurDAG->SelectNodeTo(N, Ty, Opc, Val.getOperand(0)); else CurDAG->SelectNodeTo(N, Ty, Opc, Val.getOperand(0), Val.getOperand(1), Val.getOperand(2)); break; } } CurDAG->SelectNodeTo(N, Ty, PPC::FNEG, Val); break; } case ISD::FSQRT: { MVT::ValueType Ty = N->getValueType(0); CurDAG->SelectNodeTo(N, Ty, Ty == MVT::f64 ? PPC::FSQRT : PPC::FSQRTS, Select(N->getOperand(0))); break; } case ISD::RET: { SDOperand Chain = Select(N->getOperand(0)); // Token chain. if (N->getNumOperands() > 1) { SDOperand Val = Select(N->getOperand(1)); switch (N->getOperand(1).getValueType()) { default: assert(0 && "Unknown return type!"); case MVT::f64: case MVT::f32: Chain = CurDAG->getCopyToReg(Chain, PPC::F1, Val); break; case MVT::i32: Chain = CurDAG->getCopyToReg(Chain, PPC::R3, Val); break; } if (N->getNumOperands() > 2) { assert(N->getOperand(1).getValueType() == MVT::i32 && N->getOperand(2).getValueType() == MVT::i32 && N->getNumOperands() == 2 && "Unknown two-register ret value!"); Val = Select(N->getOperand(2)); Chain = CurDAG->getCopyToReg(Chain, PPC::R4, Val); } } // Finally, select this to a blr (return) instruction. CurDAG->SelectNodeTo(N, MVT::Other, PPC::BLR, Chain); break; } } return SDOperand(N, 0); } /// createPPC32ISelDag - This pass converts a legalized DAG into a /// PowerPC-specific DAG, ready for instruction scheduling. /// FunctionPass *llvm::createPPC32ISelDag(TargetMachine &TM) { return new PPC32DAGToDAGISel(TM); }