//===- AlphaISelPattern.cpp - A pattern matching inst selector for Alpha --===// // // The LLVM Compiler Infrastructure // // This file was developed by the LLVM research group 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 Alpha. // //===----------------------------------------------------------------------===// #include "Alpha.h" #include "AlphaRegisterInfo.h" #include "llvm/Constants.h" // FIXME: REMOVE #include "llvm/Function.h" #include "llvm/Module.h" #include "llvm/CodeGen/MachineInstrBuilder.h" #include "llvm/CodeGen/MachineConstantPool.h" // FIXME: REMOVE #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/SelectionDAG.h" #include "llvm/CodeGen/SelectionDAGISel.h" #include "llvm/CodeGen/SSARegMap.h" #include "llvm/Target/TargetData.h" #include "llvm/Target/TargetLowering.h" #include "llvm/Support/MathExtras.h" #include "llvm/ADT/Statistic.h" #include "llvm/Support/Debug.h" #include "llvm/Support/CommandLine.h" #include #include using namespace llvm; namespace llvm { cl::opt EnableAlphaIDIV("enable-alpha-intfpdiv", cl::desc("Use the FP div instruction for integer div when possible"), cl::Hidden); cl::opt EnableAlphaFTOI("enable-alpha-FTOI", cl::desc("Enable use of ftoi* and itof* instructions (ev6 and higher)"), cl::Hidden); cl::opt EnableAlphaCT("enable-alpha-CT", cl::desc("Enable use of the ctpop, ctlz, and cttz instructions"), cl::Hidden); cl::opt EnableAlphaCount("enable-alpha-count", cl::desc("Print estimates on live ins and outs"), cl::Hidden); cl::opt EnableAlphaLSMark("enable-alpha-lsmark", cl::desc("Emit symbols to correlate Mem ops to LLVM Values"), cl::Hidden); } namespace { // Alpha Specific DAG Nodes namespace AlphaISD { enum NodeType { // Start the numbering where the builtin ops leave off. FIRST_NUMBER = ISD::BUILTIN_OP_END, //Convert an int bit pattern in an FP reg to a Double or Float //Has a dest type and a source CVTQ, //Move an Ireg to a FPreg ITOF, //Move a FPreg to an Ireg FTOI, }; } } //===----------------------------------------------------------------------===// // AlphaTargetLowering - Alpha Implementation of the TargetLowering interface namespace { class AlphaTargetLowering : public TargetLowering { int VarArgsOffset; // What is the offset to the first vaarg int VarArgsBase; // What is the base FrameIndex unsigned GP; //GOT vreg unsigned RA; //Return Address public: AlphaTargetLowering(TargetMachine &TM) : TargetLowering(TM) { // Set up the TargetLowering object. //I am having problems with shr n ubyte 1 setShiftAmountType(MVT::i64); setSetCCResultType(MVT::i64); setSetCCResultContents(ZeroOrOneSetCCResult); addRegisterClass(MVT::i64, Alpha::GPRCRegisterClass); addRegisterClass(MVT::f64, Alpha::FPRCRegisterClass); addRegisterClass(MVT::f32, Alpha::FPRCRegisterClass); setOperationAction(ISD::BRCONDTWOWAY, MVT::Other, Expand); setOperationAction(ISD::BRTWOWAY_CC, MVT::Other, Expand); setOperationAction(ISD::EXTLOAD, MVT::i1, Promote); setOperationAction(ISD::EXTLOAD, MVT::f32, Expand); setOperationAction(ISD::ZEXTLOAD, MVT::i1, Promote); setOperationAction(ISD::ZEXTLOAD, MVT::i32, Expand); setOperationAction(ISD::SEXTLOAD, MVT::i1, Promote); setOperationAction(ISD::SEXTLOAD, MVT::i8, Expand); setOperationAction(ISD::SEXTLOAD, MVT::i16, Expand); setOperationAction(ISD::SREM, MVT::f32, Expand); setOperationAction(ISD::SREM, MVT::f64, Expand); setOperationAction(ISD::UINT_TO_FP, MVT::i64, Expand); if (!EnableAlphaCT) { setOperationAction(ISD::CTPOP , MVT::i64 , Expand); setOperationAction(ISD::CTTZ , MVT::i64 , Expand); setOperationAction(ISD::CTLZ , MVT::i64 , Expand); } //If this didn't legalize into a div.... // setOperationAction(ISD::SREM , MVT::i64, Expand); // setOperationAction(ISD::UREM , MVT::i64, Expand); setOperationAction(ISD::MEMMOVE , MVT::Other, Expand); setOperationAction(ISD::MEMSET , MVT::Other, Expand); setOperationAction(ISD::MEMCPY , MVT::Other, Expand); // We don't support sin/cos/sqrt setOperationAction(ISD::FSIN , MVT::f64, Expand); setOperationAction(ISD::FCOS , MVT::f64, Expand); setOperationAction(ISD::FSQRT, MVT::f64, Expand); setOperationAction(ISD::FSIN , MVT::f32, Expand); setOperationAction(ISD::FCOS , MVT::f32, Expand); setOperationAction(ISD::FSQRT, MVT::f32, Expand); //Doesn't work yet setOperationAction(ISD::SETCC, MVT::f32, Promote); //Try a couple things with a custom expander //setOperationAction(ISD::SINT_TO_FP , MVT::i64 , Custom); computeRegisterProperties(); addLegalFPImmediate(+0.0); //F31 addLegalFPImmediate(-0.0); //-F31 } /// LowerOperation - Provide custom lowering hooks for some operations. /// virtual SDOperand LowerOperation(SDOperand Op, SelectionDAG &DAG); /// LowerArguments - This hook must be implemented to indicate how we should /// lower the arguments for the specified function, into the specified DAG. virtual std::vector LowerArguments(Function &F, SelectionDAG &DAG); /// LowerCallTo - This hook lowers an abstract call to a function into an /// actual call. virtual std::pair LowerCallTo(SDOperand Chain, const Type *RetTy, bool isVarArg, unsigned CC, bool isTailCall, SDOperand Callee, ArgListTy &Args, SelectionDAG &DAG); virtual SDOperand LowerVAStart(SDOperand Chain, SDOperand VAListP, Value *VAListV, SelectionDAG &DAG); virtual SDOperand LowerVACopy(SDOperand Chain, SDOperand SrcP, Value *SrcV, SDOperand DestP, Value *DestV, SelectionDAG &DAG); virtual std::pair LowerVAArg(SDOperand Chain, SDOperand VAListP, Value *VAListV, const Type *ArgTy, SelectionDAG &DAG); void restoreGP(MachineBasicBlock* BB) { BuildMI(BB, Alpha::BIS, 2, Alpha::R29).addReg(GP).addReg(GP); } void restoreRA(MachineBasicBlock* BB) { BuildMI(BB, Alpha::BIS, 2, Alpha::R26).addReg(RA).addReg(RA); } unsigned getRA() { return RA; } }; } /// LowerOperation - Provide custom lowering hooks for some operations. /// SDOperand AlphaTargetLowering::LowerOperation(SDOperand Op, SelectionDAG &DAG) { MachineFunction &MF = DAG.getMachineFunction(); switch (Op.getOpcode()) { default: assert(0 && "Should not custom lower this!"); #if 0 case ISD::SINT_TO_FP: { assert (Op.getOperand(0).getValueType() == MVT::i64 && "only quads can be loaded from"); SDOperand SRC; if (EnableAlphaFTOI) { std::vector RTs; RTs.push_back(Op.getValueType()); std::vector Ops; Ops.push_back(Op.getOperand(0)); SRC = DAG.getNode(AlphaISD::ITOF, RTs, Ops); } else { int SSFI = MF.getFrameInfo()->CreateStackObject(8, 8); SDOperand StackSlot = DAG.getFrameIndex(SSFI, getPointerTy()); SDOperand Store = DAG.getNode(ISD::STORE, MVT::Other, DAG.getEntryNode(), Op.getOperand(0), StackSlot, DAG.getSrcValue(NULL)); SRC = DAG.getLoad(Op.getValueType(), Store.getValue(0), StackSlot, DAG.getSrcValue(NULL)); } std::vector RTs; RTs.push_back(Op.getValueType()); std::vector Ops; Ops.push_back(SRC); return DAG.getNode(AlphaISD::CVTQ, RTs, Ops); } #endif } return SDOperand(); } /// AddLiveIn - This helper function adds the specified physical register to the /// MachineFunction as a live in value. It also creates a corresponding virtual /// register for it. static unsigned AddLiveIn(MachineFunction &MF, unsigned PReg, TargetRegisterClass *RC) { assert(RC->contains(PReg) && "Not the correct regclass!"); unsigned VReg = MF.getSSARegMap()->createVirtualRegister(RC); MF.addLiveIn(PReg, VReg); return VReg; } //http://www.cs.arizona.edu/computer.help/policy/DIGITAL_unix/AA-PY8AC-TET1_html/callCH3.html#BLOCK21 //For now, just use variable size stack frame format //In a standard call, the first six items are passed in registers $16 //- $21 and/or registers $f16 - $f21. (See Section 4.1.2 for details //of argument-to-register correspondence.) The remaining items are //collected in a memory argument list that is a naturally aligned //array of quadwords. In a standard call, this list, if present, must //be passed at 0(SP). //7 ... n 0(SP) ... (n-7)*8(SP) // //#define FP $15 // //#define RA $26 // //#define PV $27 // //#define GP $29 // //#define SP $30 std::vector AlphaTargetLowering::LowerArguments(Function &F, SelectionDAG &DAG) { std::vector ArgValues; MachineFunction &MF = DAG.getMachineFunction(); MachineFrameInfo*MFI = MF.getFrameInfo(); MachineBasicBlock& BB = MF.front(); unsigned args_int[] = {Alpha::R16, Alpha::R17, Alpha::R18, Alpha::R19, Alpha::R20, Alpha::R21}; unsigned args_float[] = {Alpha::F16, Alpha::F17, Alpha::F18, Alpha::F19, Alpha::F20, Alpha::F21}; int count = 0; GP = AddLiveIn(MF, Alpha::R29, getRegClassFor(MVT::i64)); RA = AddLiveIn(MF, Alpha::R26, getRegClassFor(MVT::i64)); for (Function::arg_iterator I = F.arg_begin(), E = F.arg_end(); I != E; ++I) { SDOperand argt; if (count < 6) { unsigned Vreg; MVT::ValueType VT = getValueType(I->getType()); switch (VT) { default: std::cerr << "Unknown Type " << VT << "\n"; abort(); case MVT::f64: case MVT::f32: args_float[count] = AddLiveIn(MF,args_float[count], getRegClassFor(VT)); argt = DAG.getCopyFromReg(DAG.getRoot(), args_float[count], VT); DAG.setRoot(argt.getValue(1)); break; case MVT::i1: case MVT::i8: case MVT::i16: case MVT::i32: case MVT::i64: args_int[count] = AddLiveIn(MF, args_int[count], getRegClassFor(MVT::i64)); argt = DAG.getCopyFromReg(DAG.getRoot(), args_int[count], MVT::i64); DAG.setRoot(argt.getValue(1)); if (VT != MVT::i64) argt = DAG.getNode(ISD::TRUNCATE, VT, argt); break; } } else { //more args // Create the frame index object for this incoming parameter... int FI = MFI->CreateFixedObject(8, 8 * (count - 6)); // Create the SelectionDAG nodes corresponding to a load //from this parameter SDOperand FIN = DAG.getFrameIndex(FI, MVT::i64); argt = DAG.getLoad(getValueType(I->getType()), DAG.getEntryNode(), FIN, DAG.getSrcValue(NULL)); } ++count; ArgValues.push_back(argt); } // If the functions takes variable number of arguments, copy all regs to stack if (F.isVarArg()) { VarArgsOffset = count * 8; std::vector LS; for (int i = 0; i < 6; ++i) { if (args_int[i] < 1024) args_int[i] = AddLiveIn(MF,args_int[i], getRegClassFor(MVT::i64)); SDOperand argt = DAG.getCopyFromReg(DAG.getRoot(), args_int[i], MVT::i64); int FI = MFI->CreateFixedObject(8, -8 * (6 - i)); if (i == 0) VarArgsBase = FI; SDOperand SDFI = DAG.getFrameIndex(FI, MVT::i64); LS.push_back(DAG.getNode(ISD::STORE, MVT::Other, DAG.getRoot(), argt, SDFI, DAG.getSrcValue(NULL))); if (args_float[i] < 1024) args_float[i] = AddLiveIn(MF,args_float[i], getRegClassFor(MVT::f64)); argt = DAG.getCopyFromReg(DAG.getRoot(), args_float[i], MVT::f64); FI = MFI->CreateFixedObject(8, - 8 * (12 - i)); SDFI = DAG.getFrameIndex(FI, MVT::i64); LS.push_back(DAG.getNode(ISD::STORE, MVT::Other, DAG.getRoot(), argt, SDFI, DAG.getSrcValue(NULL))); } //Set up a token factor with all the stack traffic DAG.setRoot(DAG.getNode(ISD::TokenFactor, MVT::Other, LS)); } // Finally, inform the code generator which regs we return values in. switch (getValueType(F.getReturnType())) { default: assert(0 && "Unknown type!"); case MVT::isVoid: break; case MVT::i1: case MVT::i8: case MVT::i16: case MVT::i32: case MVT::i64: MF.addLiveOut(Alpha::R0); break; case MVT::f32: case MVT::f64: MF.addLiveOut(Alpha::F0); break; } //return the arguments return ArgValues; } std::pair AlphaTargetLowering::LowerCallTo(SDOperand Chain, const Type *RetTy, bool isVarArg, unsigned CallingConv, bool isTailCall, SDOperand Callee, ArgListTy &Args, SelectionDAG &DAG) { int NumBytes = 0; if (Args.size() > 6) NumBytes = (Args.size() - 6) * 8; Chain = DAG.getNode(ISD::CALLSEQ_START, MVT::Other, Chain, DAG.getConstant(NumBytes, getPointerTy())); std::vector args_to_use; for (unsigned i = 0, e = Args.size(); i != e; ++i) { switch (getValueType(Args[i].second)) { default: assert(0 && "Unexpected ValueType for argument!"); case MVT::i1: case MVT::i8: case MVT::i16: case MVT::i32: // Promote the integer to 64 bits. If the input type is signed use a // sign extend, otherwise use a zero extend. if (Args[i].second->isSigned()) Args[i].first = DAG.getNode(ISD::SIGN_EXTEND, MVT::i64, Args[i].first); else Args[i].first = DAG.getNode(ISD::ZERO_EXTEND, MVT::i64, Args[i].first); break; case MVT::i64: case MVT::f64: case MVT::f32: break; } args_to_use.push_back(Args[i].first); } std::vector RetVals; MVT::ValueType RetTyVT = getValueType(RetTy); if (RetTyVT != MVT::isVoid) RetVals.push_back(RetTyVT); RetVals.push_back(MVT::Other); SDOperand TheCall = SDOperand(DAG.getCall(RetVals, Chain, Callee, args_to_use), 0); Chain = TheCall.getValue(RetTyVT != MVT::isVoid); Chain = DAG.getNode(ISD::CALLSEQ_END, MVT::Other, Chain, DAG.getConstant(NumBytes, getPointerTy())); return std::make_pair(TheCall, Chain); } SDOperand AlphaTargetLowering::LowerVAStart(SDOperand Chain, SDOperand VAListP, Value *VAListV, SelectionDAG &DAG) { // vastart stores the address of the VarArgsBase and VarArgsOffset SDOperand FR = DAG.getFrameIndex(VarArgsBase, MVT::i64); SDOperand S1 = DAG.getNode(ISD::STORE, MVT::Other, Chain, FR, VAListP, DAG.getSrcValue(VAListV)); SDOperand SA2 = DAG.getNode(ISD::ADD, MVT::i64, VAListP, DAG.getConstant(8, MVT::i64)); return DAG.getNode(ISD::TRUNCSTORE, MVT::Other, S1, DAG.getConstant(VarArgsOffset, MVT::i64), SA2, DAG.getSrcValue(VAListV, 8), DAG.getValueType(MVT::i32)); } std::pair AlphaTargetLowering:: LowerVAArg(SDOperand Chain, SDOperand VAListP, Value *VAListV, const Type *ArgTy, SelectionDAG &DAG) { SDOperand Base = DAG.getLoad(MVT::i64, Chain, VAListP, DAG.getSrcValue(VAListV)); SDOperand Tmp = DAG.getNode(ISD::ADD, MVT::i64, VAListP, DAG.getConstant(8, MVT::i64)); SDOperand Offset = DAG.getExtLoad(ISD::SEXTLOAD, MVT::i64, Base.getValue(1), Tmp, DAG.getSrcValue(VAListV, 8), MVT::i32); SDOperand DataPtr = DAG.getNode(ISD::ADD, MVT::i64, Base, Offset); if (ArgTy->isFloatingPoint()) { //if fp && Offset < 6*8, then subtract 6*8 from DataPtr SDOperand FPDataPtr = DAG.getNode(ISD::SUB, MVT::i64, DataPtr, DAG.getConstant(8*6, MVT::i64)); SDOperand CC = DAG.getSetCC(MVT::i64, Offset, DAG.getConstant(8*6, MVT::i64), ISD::SETLT); DataPtr = DAG.getNode(ISD::SELECT, MVT::i64, CC, FPDataPtr, DataPtr); } SDOperand Result; if (ArgTy == Type::IntTy) Result = DAG.getExtLoad(ISD::SEXTLOAD, MVT::i64, Offset.getValue(1), DataPtr, DAG.getSrcValue(NULL), MVT::i32); else if (ArgTy == Type::UIntTy) Result = DAG.getExtLoad(ISD::ZEXTLOAD, MVT::i64, Offset.getValue(1), DataPtr, DAG.getSrcValue(NULL), MVT::i32); else Result = DAG.getLoad(getValueType(ArgTy), Offset.getValue(1), DataPtr, DAG.getSrcValue(NULL)); SDOperand NewOffset = DAG.getNode(ISD::ADD, MVT::i64, Offset, DAG.getConstant(8, MVT::i64)); SDOperand Update = DAG.getNode(ISD::TRUNCSTORE, MVT::Other, Result.getValue(1), NewOffset, Tmp, DAG.getSrcValue(VAListV, 8), DAG.getValueType(MVT::i32)); Result = DAG.getNode(ISD::TRUNCATE, getValueType(ArgTy), Result); return std::make_pair(Result, Update); } SDOperand AlphaTargetLowering:: LowerVACopy(SDOperand Chain, SDOperand SrcP, Value *SrcV, SDOperand DestP, Value *DestV, SelectionDAG &DAG) { SDOperand Val = DAG.getLoad(getPointerTy(), Chain, SrcP, DAG.getSrcValue(SrcV)); SDOperand Result = DAG.getNode(ISD::STORE, MVT::Other, Val.getValue(1), Val, DestP, DAG.getSrcValue(DestV)); SDOperand NP = DAG.getNode(ISD::ADD, MVT::i64, SrcP, DAG.getConstant(8, MVT::i64)); Val = DAG.getExtLoad(ISD::SEXTLOAD, MVT::i64, Result, NP, DAG.getSrcValue(SrcV, 8), MVT::i32); SDOperand NPD = DAG.getNode(ISD::ADD, MVT::i64, DestP, DAG.getConstant(8, MVT::i64)); return DAG.getNode(ISD::TRUNCSTORE, MVT::Other, Val.getValue(1), Val, NPD, DAG.getSrcValue(DestV, 8), DAG.getValueType(MVT::i32)); } namespace { //===--------------------------------------------------------------------===// /// ISel - Alpha specific code to select Alpha machine instructions for /// SelectionDAG operations. //===--------------------------------------------------------------------===// class AlphaISel : public SelectionDAGISel { /// AlphaLowering - This object fully describes how to lower LLVM code to an /// Alpha-specific SelectionDAG. AlphaTargetLowering AlphaLowering; SelectionDAG *ISelDAG; // Hack to support us having a dag->dag transform // for sdiv and udiv until it is put into the future // dag combiner. /// ExprMap - As shared expressions are codegen'd, we keep track of which /// vreg the value is produced in, so we only emit one copy of each compiled /// tree. static const unsigned notIn = (unsigned)(-1); std::map ExprMap; //CCInvMap sometimes (SetNE) we have the inverse CC code for free std::map CCInvMap; int count_ins; int count_outs; bool has_sym; int max_depth; public: AlphaISel(TargetMachine &TM) : SelectionDAGISel(AlphaLowering), AlphaLowering(TM) {} /// 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()); count_ins = 0; count_outs = 0; max_depth = 0; has_sym = false; // Codegen the basic block. ISelDAG = &DAG; max_depth = DAG.getRoot().getNodeDepth(); Select(DAG.getRoot()); if(has_sym) ++count_ins; if(EnableAlphaCount) std::cerr << "COUNT: " << BB->getParent()->getFunction ()->getName() << " " << BB->getNumber() << " " << max_depth << " " << count_ins << " " << count_outs << "\n"; // Clear state used for selection. ExprMap.clear(); CCInvMap.clear(); } virtual void EmitFunctionEntryCode(Function &Fn, MachineFunction &MF); unsigned SelectExpr(SDOperand N); void Select(SDOperand N); void SelectAddr(SDOperand N, unsigned& Reg, long& offset); void SelectBranchCC(SDOperand N); void MoveFP2Int(unsigned src, unsigned dst, bool isDouble); void MoveInt2FP(unsigned src, unsigned dst, bool isDouble); //returns whether the sense of the comparison was inverted bool SelectFPSetCC(SDOperand N, unsigned dst); // dag -> dag expanders for integer divide by constant SDOperand BuildSDIVSequence(SDOperand N); SDOperand BuildUDIVSequence(SDOperand N); }; } void AlphaISel::EmitFunctionEntryCode(Function &Fn, MachineFunction &MF) { // If this function has live-in values, emit the copies from pregs to vregs at // the top of the function, before anything else. MachineBasicBlock *BB = MF.begin(); if (MF.livein_begin() != MF.livein_end()) { SSARegMap *RegMap = MF.getSSARegMap(); for (MachineFunction::livein_iterator LI = MF.livein_begin(), E = MF.livein_end(); LI != E; ++LI) { const TargetRegisterClass *RC = RegMap->getRegClass(LI->second); if (RC == Alpha::GPRCRegisterClass) { BuildMI(BB, Alpha::BIS, 2, LI->second).addReg(LI->first) .addReg(LI->first); } else if (RC == Alpha::FPRCRegisterClass) { BuildMI(BB, Alpha::CPYS, 2, LI->second).addReg(LI->first) .addReg(LI->first); } else { assert(0 && "Unknown regclass!"); } } } } static bool isSIntImmediate(SDOperand N, int64_t& Imm) { // test for constant if (ConstantSDNode *CN = dyn_cast(N)) { // retrieve value Imm = CN->getSignExtended(); // passes muster return true; } // not a constant return false; } // isSIntImmediateBounded - This method tests to see if a constant operand // bounded s.t. low <= Imm <= high // If so Imm will receive the 64 bit value. static bool isSIntImmediateBounded(SDOperand N, int64_t& Imm, int64_t low, int64_t high) { if (isSIntImmediate(N, Imm) && Imm <= high && Imm >= low) return true; return false; } static bool isUIntImmediate(SDOperand N, uint64_t& Imm) { // test for constant if (ConstantSDNode *CN = dyn_cast(N)) { // retrieve value Imm = (uint64_t)CN->getValue(); // passes muster return true; } // not a constant return false; } static bool isUIntImmediateBounded(SDOperand N, uint64_t& Imm, uint64_t low, uint64_t high) { if (isUIntImmediate(N, Imm) && Imm <= high && Imm >= low) return true; return false; } static void getValueInfo(const Value* v, int& type, int& fun, int& offset) { fun = type = offset = 0; if (v == NULL) { type = 0; } else if (const GlobalValue* GV = dyn_cast(v)) { type = 1; const Module* M = GV->getParent(); for(Module::const_global_iterator ii = M->global_begin(); &*ii != GV; ++ii) ++offset; } else if (const Argument* Arg = dyn_cast(v)) { type = 2; const Function* F = Arg->getParent(); const Module* M = F->getParent(); for(Module::const_iterator ii = M->begin(); &*ii != F; ++ii) ++fun; for(Function::const_arg_iterator ii = F->arg_begin(); &*ii != Arg; ++ii) ++offset; } else if (const Instruction* I = dyn_cast(v)) { assert(dyn_cast(I->getType())); type = 3; const BasicBlock* bb = I->getParent(); const Function* F = bb->getParent(); const Module* M = F->getParent(); for(Module::const_iterator ii = M->begin(); &*ii != F; ++ii) ++fun; for(Function::const_iterator ii = F->begin(); &*ii != bb; ++ii) offset += ii->size(); for(BasicBlock::const_iterator ii = bb->begin(); &*ii != I; ++ii) ++offset; } else if (const Constant* C = dyn_cast(v)) { //Don't know how to look these up yet type = 0; } else { assert(0 && "Error in value marking"); } //type = 4: register spilling //type = 5: global address loading or constant loading } static int getUID() { static int id = 0; return ++id; } //Factorize a number using the list of constants static bool factorize(int v[], int res[], int size, uint64_t c) { bool cont = true; while (c != 1 && cont) { cont = false; for(int i = 0; i < size; ++i) { if (c % v[i] == 0) { c /= v[i]; ++res[i]; cont=true; } } } return c == 1; } //Shamelessly adapted from PPC32 // Structure used to return the necessary information to codegen an SDIV as // a multiply. struct ms { int64_t m; // magic number int64_t s; // shift amount }; struct mu { uint64_t m; // magic number int64_t a; // add indicator int64_t s; // shift amount }; /// magic - calculate the magic numbers required to codegen an integer sdiv as /// a sequence of multiply and shifts. Requires that the divisor not be 0, 1, /// or -1. static struct ms magic(int64_t d) { int64_t p; uint64_t ad, anc, delta, q1, r1, q2, r2, t; const uint64_t two63 = 9223372036854775808ULL; // 2^63 struct ms mag; ad = llabs(d); t = two63 + ((uint64_t)d >> 63); anc = t - 1 - t%ad; // absolute value of nc p = 63; // initialize p q1 = two63/anc; // initialize q1 = 2p/abs(nc) r1 = two63 - q1*anc; // initialize r1 = rem(2p,abs(nc)) q2 = two63/ad; // initialize q2 = 2p/abs(d) r2 = two63 - q2*ad; // initialize r2 = rem(2p,abs(d)) do { p = p + 1; q1 = 2*q1; // update q1 = 2p/abs(nc) r1 = 2*r1; // update r1 = rem(2p/abs(nc)) if (r1 >= anc) { // must be unsigned comparison q1 = q1 + 1; r1 = r1 - anc; } q2 = 2*q2; // update q2 = 2p/abs(d) r2 = 2*r2; // update r2 = rem(2p/abs(d)) if (r2 >= ad) { // must be unsigned comparison q2 = q2 + 1; r2 = r2 - ad; } delta = ad - r2; } while (q1 < delta || (q1 == delta && r1 == 0)); mag.m = q2 + 1; if (d < 0) mag.m = -mag.m; // resulting magic number mag.s = p - 64; // resulting shift return mag; } /// magicu - calculate the magic numbers required to codegen an integer udiv as /// a sequence of multiply, add and shifts. Requires that the divisor not be 0. static struct mu magicu(uint64_t d) { int64_t p; uint64_t nc, delta, q1, r1, q2, r2; struct mu magu; magu.a = 0; // initialize "add" indicator nc = - 1 - (-d)%d; p = 63; // initialize p q1 = 0x8000000000000000ull/nc; // initialize q1 = 2p/nc r1 = 0x8000000000000000ull - q1*nc; // initialize r1 = rem(2p,nc) q2 = 0x7FFFFFFFFFFFFFFFull/d; // initialize q2 = (2p-1)/d r2 = 0x7FFFFFFFFFFFFFFFull - q2*d; // initialize r2 = rem((2p-1),d) do { p = p + 1; if (r1 >= nc - r1 ) { q1 = 2*q1 + 1; // update q1 r1 = 2*r1 - nc; // update r1 } else { q1 = 2*q1; // update q1 r1 = 2*r1; // update r1 } if (r2 + 1 >= d - r2) { if (q2 >= 0x7FFFFFFFFFFFFFFFull) magu.a = 1; q2 = 2*q2 + 1; // update q2 r2 = 2*r2 + 1 - d; // update r2 } else { if (q2 >= 0x8000000000000000ull) magu.a = 1; q2 = 2*q2; // update q2 r2 = 2*r2 + 1; // update r2 } delta = d - 1 - r2; } while (p < 64 && (q1 < delta || (q1 == delta && r1 == 0))); magu.m = q2 + 1; // resulting magic number magu.s = p - 64; // resulting shift return magu; } /// BuildSDIVSequence - Given an ISD::SDIV node expressing a divide by constant, /// return a DAG expression to select that will generate the same value by /// multiplying by a magic number. See: /// SDOperand AlphaISel::BuildSDIVSequence(SDOperand N) { int64_t d = (int64_t)cast(N.getOperand(1))->getSignExtended(); ms magics = magic(d); // Multiply the numerator (operand 0) by the magic value SDOperand Q = ISelDAG->getNode(ISD::MULHS, MVT::i64, N.getOperand(0), ISelDAG->getConstant(magics.m, MVT::i64)); // If d > 0 and m < 0, add the numerator if (d > 0 && magics.m < 0) Q = ISelDAG->getNode(ISD::ADD, MVT::i64, Q, N.getOperand(0)); // If d < 0 and m > 0, subtract the numerator. if (d < 0 && magics.m > 0) Q = ISelDAG->getNode(ISD::SUB, MVT::i64, Q, N.getOperand(0)); // Shift right algebraic if shift value is nonzero if (magics.s > 0) Q = ISelDAG->getNode(ISD::SRA, MVT::i64, Q, ISelDAG->getConstant(magics.s, MVT::i64)); // Extract the sign bit and add it to the quotient SDOperand T = ISelDAG->getNode(ISD::SRL, MVT::i64, Q, ISelDAG->getConstant(63, MVT::i64)); return ISelDAG->getNode(ISD::ADD, MVT::i64, Q, T); } /// BuildUDIVSequence - Given an ISD::UDIV node expressing a divide by constant, /// return a DAG expression to select that will generate the same value by /// multiplying by a magic number. See: /// SDOperand AlphaISel::BuildUDIVSequence(SDOperand N) { unsigned d = (unsigned)cast(N.getOperand(1))->getSignExtended(); mu magics = magicu(d); // Multiply the numerator (operand 0) by the magic value SDOperand Q = ISelDAG->getNode(ISD::MULHU, MVT::i64, N.getOperand(0), ISelDAG->getConstant(magics.m, MVT::i64)); if (magics.a == 0) { Q = ISelDAG->getNode(ISD::SRL, MVT::i64, Q, ISelDAG->getConstant(magics.s, MVT::i64)); } else { SDOperand NPQ = ISelDAG->getNode(ISD::SUB, MVT::i64, N.getOperand(0), Q); NPQ = ISelDAG->getNode(ISD::SRL, MVT::i64, NPQ, ISelDAG->getConstant(1, MVT::i64)); NPQ = ISelDAG->getNode(ISD::ADD, MVT::i64, NPQ, Q); Q = ISelDAG->getNode(ISD::SRL, MVT::i64, NPQ, ISelDAG->getConstant(magics.s-1, MVT::i64)); } return Q; } //These describe LDAx static const int IMM_LOW = -32768; static const int IMM_HIGH = 32767; static const int IMM_MULT = 65536; static long getUpper16(long l) { long y = l / IMM_MULT; if (l % IMM_MULT > IMM_HIGH) ++y; return y; } static long getLower16(long l) { long h = getUpper16(l); return l - h * IMM_MULT; } static unsigned GetRelVersion(unsigned opcode) { switch (opcode) { default: assert(0 && "unknown load or store"); return 0; case Alpha::LDQ: return Alpha::LDQr; case Alpha::LDS: return Alpha::LDSr; case Alpha::LDT: return Alpha::LDTr; case Alpha::LDL: return Alpha::LDLr; case Alpha::LDBU: return Alpha::LDBUr; case Alpha::LDWU: return Alpha::LDWUr; case Alpha::STB: return Alpha::STBr; case Alpha::STW: return Alpha::STWr; case Alpha::STL: return Alpha::STLr; case Alpha::STQ: return Alpha::STQr; case Alpha::STS: return Alpha::STSr; case Alpha::STT: return Alpha::STTr; } } void AlphaISel::MoveFP2Int(unsigned src, unsigned dst, bool isDouble) { unsigned Opc; if (EnableAlphaFTOI) { Opc = isDouble ? Alpha::FTOIT : Alpha::FTOIS; BuildMI(BB, Opc, 1, dst).addReg(src).addReg(Alpha::F31); } else { //The hard way: // Spill the integer to memory and reload it from there. unsigned Size = MVT::getSizeInBits(MVT::f64)/8; MachineFunction *F = BB->getParent(); int FrameIdx = F->getFrameInfo()->CreateStackObject(Size, 8); if (EnableAlphaLSMark) BuildMI(BB, Alpha::MEMLABEL, 4).addImm(4).addImm(0).addImm(0) .addImm(getUID()); Opc = isDouble ? Alpha::STT : Alpha::STS; BuildMI(BB, Opc, 3).addReg(src).addFrameIndex(FrameIdx).addReg(Alpha::F31); if (EnableAlphaLSMark) BuildMI(BB, Alpha::MEMLABEL, 4).addImm(4).addImm(0).addImm(0) .addImm(getUID()); Opc = isDouble ? Alpha::LDQ : Alpha::LDL; BuildMI(BB, Alpha::LDQ, 2, dst).addFrameIndex(FrameIdx).addReg(Alpha::F31); } } void AlphaISel::MoveInt2FP(unsigned src, unsigned dst, bool isDouble) { unsigned Opc; if (EnableAlphaFTOI) { Opc = isDouble?Alpha::ITOFT:Alpha::ITOFS; BuildMI(BB, Opc, 1, dst).addReg(src).addReg(Alpha::R31); } else { //The hard way: // Spill the integer to memory and reload it from there. unsigned Size = MVT::getSizeInBits(MVT::f64)/8; MachineFunction *F = BB->getParent(); int FrameIdx = F->getFrameInfo()->CreateStackObject(Size, 8); if (EnableAlphaLSMark) BuildMI(BB, Alpha::MEMLABEL, 4).addImm(4).addImm(0).addImm(0) .addImm(getUID()); Opc = isDouble ? Alpha::STQ : Alpha::STL; BuildMI(BB, Opc, 3).addReg(src).addFrameIndex(FrameIdx).addReg(Alpha::F31); if (EnableAlphaLSMark) BuildMI(BB, Alpha::MEMLABEL, 4).addImm(4).addImm(0).addImm(0) .addImm(getUID()); Opc = isDouble ? Alpha::LDT : Alpha::LDS; BuildMI(BB, Opc, 2, dst).addFrameIndex(FrameIdx).addReg(Alpha::F31); } } bool AlphaISel::SelectFPSetCC(SDOperand N, unsigned dst) { SDNode *SetCC = N.Val; unsigned Opc, Tmp1, Tmp2, Tmp3; ISD::CondCode CC = cast(SetCC->getOperand(2))->get(); bool rev = false; bool inv = false; switch (CC) { default: SetCC->dump(); assert(0 && "Unknown FP comparison!"); case ISD::SETEQ: Opc = Alpha::CMPTEQ; break; case ISD::SETLT: Opc = Alpha::CMPTLT; break; case ISD::SETLE: Opc = Alpha::CMPTLE; break; case ISD::SETGT: Opc = Alpha::CMPTLT; rev = true; break; case ISD::SETGE: Opc = Alpha::CMPTLE; rev = true; break; case ISD::SETNE: Opc = Alpha::CMPTEQ; inv = true; break; } ConstantFPSDNode *CN; if ((CN = dyn_cast(SetCC->getOperand(0))) && (CN->isExactlyValue(+0.0) || CN->isExactlyValue(-0.0))) Tmp1 = Alpha::F31; else Tmp1 = SelectExpr(N.getOperand(0)); if ((CN = dyn_cast(SetCC->getOperand(1))) && (CN->isExactlyValue(+0.0) || CN->isExactlyValue(-0.0))) Tmp2 = Alpha::F31; else Tmp2 = SelectExpr(N.getOperand(1)); //Can only compare doubles, and dag won't promote for me if (SetCC->getOperand(0).getValueType() == MVT::f32) { //assert(0 && "Setcc On float?\n"); std::cerr << "Setcc on float!\n"; Tmp3 = MakeReg(MVT::f64); BuildMI(BB, Alpha::CVTST, 1, Tmp3).addReg(Alpha::F31).addReg(Tmp1); Tmp1 = Tmp3; } if (SetCC->getOperand(1).getValueType() == MVT::f32) { //assert (0 && "Setcc On float?\n"); std::cerr << "Setcc on float!\n"; Tmp3 = MakeReg(MVT::f64); BuildMI(BB, Alpha::CVTST, 1, Tmp3).addReg(Alpha::F31).addReg(Tmp2); Tmp2 = Tmp3; } if (rev) std::swap(Tmp1, Tmp2); //do the comparison BuildMI(BB, Opc, 2, dst).addReg(Tmp1).addReg(Tmp2); return inv; } //Check to see if the load is a constant offset from a base register void AlphaISel::SelectAddr(SDOperand N, unsigned& Reg, long& offset) { unsigned opcode = N.getOpcode(); if (opcode == ISD::ADD && N.getOperand(1).getOpcode() == ISD::Constant && cast(N.getOperand(1))->getValue() <= 32767) { //Normal imm add Reg = SelectExpr(N.getOperand(0)); offset = cast(N.getOperand(1))->getValue(); return; } Reg = SelectExpr(N); offset = 0; return; } void AlphaISel::SelectBranchCC(SDOperand N) { assert(N.getOpcode() == ISD::BRCOND && "Not a BranchCC???"); MachineBasicBlock *Dest = cast(N.getOperand(2))->getBasicBlock(); unsigned Opc = Alpha::WTF; Select(N.getOperand(0)); //chain SDOperand CC = N.getOperand(1); if (CC.getOpcode() == ISD::SETCC) { ISD::CondCode cCode= cast(CC.getOperand(2))->get(); if (MVT::isInteger(CC.getOperand(0).getValueType())) { //Dropping the CC is only useful if we are comparing to 0 bool RightZero = CC.getOperand(1).getOpcode() == ISD::Constant && cast(CC.getOperand(1))->getValue() == 0; bool isNE = false; //Fix up CC if(cCode == ISD::SETNE) isNE = true; if (RightZero) { switch (cCode) { default: CC.Val->dump(); assert(0 && "Unknown integer comparison!"); case ISD::SETEQ: Opc = Alpha::BEQ; break; case ISD::SETLT: Opc = Alpha::BLT; break; case ISD::SETLE: Opc = Alpha::BLE; break; case ISD::SETGT: Opc = Alpha::BGT; break; case ISD::SETGE: Opc = Alpha::BGE; break; case ISD::SETULT: assert(0 && "x (unsigned) < 0 is never true"); break; case ISD::SETUGT: Opc = Alpha::BNE; break; //Technically you could have this CC case ISD::SETULE: Opc = Alpha::BEQ; break; case ISD::SETUGE: assert(0 && "x (unsgined >= 0 is always true"); break; case ISD::SETNE: Opc = Alpha::BNE; break; } unsigned Tmp1 = SelectExpr(CC.getOperand(0)); //Cond BuildMI(BB, Opc, 2).addReg(Tmp1).addMBB(Dest); return; } else { unsigned Tmp1 = SelectExpr(CC); if (isNE) BuildMI(BB, Alpha::BEQ, 2).addReg(CCInvMap[CC]).addMBB(Dest); else BuildMI(BB, Alpha::BNE, 2).addReg(Tmp1).addMBB(Dest); return; } } else { //FP //Any comparison between 2 values should be codegened as an folded //branch, as moving CC to the integer register is very expensive //for a cmp b: c = a - b; //a = b: c = 0 //a < b: c < 0 //a > b: c > 0 bool invTest = false; unsigned Tmp3; ConstantFPSDNode *CN; if ((CN = dyn_cast(CC.getOperand(1))) && (CN->isExactlyValue(+0.0) || CN->isExactlyValue(-0.0))) Tmp3 = SelectExpr(CC.getOperand(0)); else if ((CN = dyn_cast(CC.getOperand(0))) && (CN->isExactlyValue(+0.0) || CN->isExactlyValue(-0.0))) { Tmp3 = SelectExpr(CC.getOperand(1)); invTest = true; } else { unsigned Tmp1 = SelectExpr(CC.getOperand(0)); unsigned Tmp2 = SelectExpr(CC.getOperand(1)); bool isD = CC.getOperand(0).getValueType() == MVT::f64; Tmp3 = MakeReg(isD ? MVT::f64 : MVT::f32); BuildMI(BB, isD ? Alpha::SUBT : Alpha::SUBS, 2, Tmp3) .addReg(Tmp1).addReg(Tmp2); } switch (cCode) { default: CC.Val->dump(); assert(0 && "Unknown FP comparison!"); case ISD::SETEQ: Opc = invTest ? Alpha::FBNE : Alpha::FBEQ; break; case ISD::SETLT: Opc = invTest ? Alpha::FBGT : Alpha::FBLT; break; case ISD::SETLE: Opc = invTest ? Alpha::FBGE : Alpha::FBLE; break; case ISD::SETGT: Opc = invTest ? Alpha::FBLT : Alpha::FBGT; break; case ISD::SETGE: Opc = invTest ? Alpha::FBLE : Alpha::FBGE; break; case ISD::SETNE: Opc = invTest ? Alpha::FBEQ : Alpha::FBNE; break; } BuildMI(BB, Opc, 2).addReg(Tmp3).addMBB(Dest); return; } abort(); //Should never be reached } else { //Giveup and do the stupid thing unsigned Tmp1 = SelectExpr(CC); BuildMI(BB, Alpha::BNE, 2).addReg(Tmp1).addMBB(Dest); return; } abort(); //Should never be reached } unsigned AlphaISel::SelectExpr(SDOperand N) { unsigned Result; unsigned Tmp1, Tmp2 = 0, Tmp3; unsigned Opc = 0; unsigned opcode = N.getOpcode(); int64_t SImm; uint64_t UImm; SDNode *Node = N.Val; MVT::ValueType DestType = N.getValueType(); bool isFP = DestType == MVT::f64 || DestType == MVT::f32; unsigned &Reg = ExprMap[N]; if (Reg) return Reg; if (N.getOpcode() != ISD::CALL && N.getOpcode() != ISD::TAILCALL) Reg = Result = (N.getValueType() != MVT::Other) ? MakeReg(N.getValueType()) : notIn; else { // If this is a call instruction, make sure to prepare ALL of the result // values as well as the chain. if (Node->getNumValues() == 1) Reg = Result = notIn; // Void call, just a chain. else { Result = MakeReg(Node->getValueType(0)); ExprMap[N.getValue(0)] = Result; for (unsigned i = 1, e = N.Val->getNumValues()-1; i != e; ++i) ExprMap[N.getValue(i)] = MakeReg(Node->getValueType(i)); ExprMap[SDOperand(Node, Node->getNumValues()-1)] = notIn; } } switch (opcode) { default: Node->dump(); assert(0 && "Node not handled!\n"); case ISD::CTPOP: case ISD::CTTZ: case ISD::CTLZ: Opc = opcode == ISD::CTPOP ? Alpha::CTPOP : (opcode == ISD::CTTZ ? Alpha::CTTZ : Alpha::CTLZ); Tmp1 = SelectExpr(N.getOperand(0)); BuildMI(BB, Opc, 1, Result).addReg(Alpha::R31).addReg(Tmp1); return Result; case ISD::MULHU: Tmp1 = SelectExpr(N.getOperand(0)); Tmp2 = SelectExpr(N.getOperand(1)); BuildMI(BB, Alpha::UMULH, 2, Result).addReg(Tmp1).addReg(Tmp2); return Result; case ISD::MULHS: { //MULHU - Ra<63>*Rb - Rb<63>*Ra Tmp1 = SelectExpr(N.getOperand(0)); Tmp2 = SelectExpr(N.getOperand(1)); Tmp3 = MakeReg(MVT::i64); BuildMI(BB, Alpha::UMULH, 2, Tmp3).addReg(Tmp1).addReg(Tmp2); unsigned V1 = MakeReg(MVT::i64); unsigned V2 = MakeReg(MVT::i64); BuildMI(BB, Alpha::CMOVGE, 3, V1).addReg(Tmp2).addReg(Alpha::R31) .addReg(Tmp1); BuildMI(BB, Alpha::CMOVGE, 3, V2).addReg(Tmp1).addReg(Alpha::R31) .addReg(Tmp2); unsigned IRes = MakeReg(MVT::i64); BuildMI(BB, Alpha::SUBQ, 2, IRes).addReg(Tmp3).addReg(V1); BuildMI(BB, Alpha::SUBQ, 2, Result).addReg(IRes).addReg(V2); return Result; } case ISD::UNDEF: { BuildMI(BB, Alpha::IDEF, 0, Result); return Result; } case ISD::DYNAMIC_STACKALLOC: // Generate both result values. if (Result != notIn) ExprMap[N.getValue(1)] = notIn; // Generate the token else Result = ExprMap[N.getValue(0)] = MakeReg(N.getValue(0).getValueType()); // FIXME: We are currently ignoring the requested alignment for handling // greater than the stack alignment. This will need to be revisited at some // point. Align = N.getOperand(2); if (!isa(N.getOperand(2)) || cast(N.getOperand(2))->getValue() != 0) { std::cerr << "Cannot allocate stack object with greater alignment than" << " the stack alignment yet!"; abort(); } Select(N.getOperand(0)); if (isSIntImmediateBounded(N.getOperand(1), SImm, 0, 32767)) BuildMI(BB, Alpha::LDA, 2, Alpha::R30).addImm(-SImm).addReg(Alpha::R30); else { Tmp1 = SelectExpr(N.getOperand(1)); // Subtract size from stack pointer, thereby allocating some space. BuildMI(BB, Alpha::SUBQ, 2, Alpha::R30).addReg(Alpha::R30).addReg(Tmp1); } // Put a pointer to the space into the result register, by copying the stack // pointer. BuildMI(BB, Alpha::BIS, 2, Result).addReg(Alpha::R30).addReg(Alpha::R30); return Result; case ISD::ConstantPool: Tmp1 = cast(N)->getIndex(); AlphaLowering.restoreGP(BB); Tmp2 = MakeReg(MVT::i64); BuildMI(BB, Alpha::LDAHr, 2, Tmp2).addConstantPoolIndex(Tmp1) .addReg(Alpha::R29); BuildMI(BB, Alpha::LDAr, 2, Result).addConstantPoolIndex(Tmp1) .addReg(Tmp2); return Result; case ISD::FrameIndex: BuildMI(BB, Alpha::LDA, 2, Result) .addFrameIndex(cast(N)->getIndex()) .addReg(Alpha::F31); return Result; case ISD::EXTLOAD: case ISD::ZEXTLOAD: case ISD::SEXTLOAD: case ISD::LOAD: { // Make sure we generate both values. if (Result != notIn) ExprMap[N.getValue(1)] = notIn; // Generate the token else Result = ExprMap[N.getValue(0)] = MakeReg(N.getValue(0).getValueType()); SDOperand Chain = N.getOperand(0); SDOperand Address = N.getOperand(1); Select(Chain); bool fpext = true; if (opcode == ISD::LOAD) switch (Node->getValueType(0)) { default: Node->dump(); assert(0 && "Bad load!"); case MVT::i64: Opc = Alpha::LDQ; break; case MVT::f64: Opc = Alpha::LDT; break; case MVT::f32: Opc = Alpha::LDS; break; } else switch (cast(Node->getOperand(3))->getVT()) { default: Node->dump(); assert(0 && "Bad sign extend!"); case MVT::i32: Opc = Alpha::LDL; assert(opcode != ISD::ZEXTLOAD && "Not sext"); break; case MVT::i16: Opc = Alpha::LDWU; assert(opcode != ISD::SEXTLOAD && "Not zext"); break; case MVT::i1: //FIXME: Treat i1 as i8 since there are problems otherwise case MVT::i8: Opc = Alpha::LDBU; assert(opcode != ISD::SEXTLOAD && "Not zext"); break; } int i, j, k; if (EnableAlphaLSMark) getValueInfo(dyn_cast(N.getOperand(2))->getValue(), i, j, k); GlobalAddressSDNode *GASD = dyn_cast(Address); if (GASD && !GASD->getGlobal()->isExternal()) { Tmp1 = MakeReg(MVT::i64); AlphaLowering.restoreGP(BB); BuildMI(BB, Alpha::LDAHr, 2, Tmp1) .addGlobalAddress(GASD->getGlobal()).addReg(Alpha::R29); if (EnableAlphaLSMark) BuildMI(BB, Alpha::MEMLABEL, 4).addImm(i).addImm(j).addImm(k) .addImm(getUID()); BuildMI(BB, GetRelVersion(Opc), 2, Result) .addGlobalAddress(GASD->getGlobal()).addReg(Tmp1); } else if (ConstantPoolSDNode *CP = dyn_cast(Address)) { AlphaLowering.restoreGP(BB); has_sym = true; Tmp1 = MakeReg(MVT::i64); BuildMI(BB, Alpha::LDAHr, 2, Tmp1).addConstantPoolIndex(CP->getIndex()) .addReg(Alpha::R29); if (EnableAlphaLSMark) BuildMI(BB, Alpha::MEMLABEL, 4).addImm(i).addImm(j).addImm(k) .addImm(getUID()); BuildMI(BB, GetRelVersion(Opc), 2, Result) .addConstantPoolIndex(CP->getIndex()).addReg(Tmp1); } else if(Address.getOpcode() == ISD::FrameIndex) { if (EnableAlphaLSMark) BuildMI(BB, Alpha::MEMLABEL, 4).addImm(i).addImm(j).addImm(k) .addImm(getUID()); BuildMI(BB, Opc, 2, Result) .addFrameIndex(cast(Address)->getIndex()) .addReg(Alpha::F31); } else { long offset; SelectAddr(Address, Tmp1, offset); if (EnableAlphaLSMark) BuildMI(BB, Alpha::MEMLABEL, 4).addImm(i).addImm(j).addImm(k) .addImm(getUID()); BuildMI(BB, Opc, 2, Result).addImm(offset).addReg(Tmp1); } return Result; } case ISD::GlobalAddress: AlphaLowering.restoreGP(BB); has_sym = true; Reg = Result = MakeReg(MVT::i64); if (EnableAlphaLSMark) BuildMI(BB, Alpha::MEMLABEL, 4).addImm(5).addImm(0).addImm(0) .addImm(getUID()); BuildMI(BB, Alpha::LDQl, 2, Result) .addGlobalAddress(cast(N)->getGlobal()) .addReg(Alpha::R29); return Result; case ISD::ExternalSymbol: AlphaLowering.restoreGP(BB); has_sym = true; Reg = Result = MakeReg(MVT::i64); if (EnableAlphaLSMark) BuildMI(BB, Alpha::MEMLABEL, 4).addImm(5).addImm(0).addImm(0) .addImm(getUID()); BuildMI(BB, Alpha::LDQl, 2, Result) .addExternalSymbol(cast(N)->getSymbol()) .addReg(Alpha::R29); return Result; case ISD::TAILCALL: case ISD::CALL: { Select(N.getOperand(0)); // The chain for this call is now lowered. ExprMap[N.getValue(Node->getNumValues()-1)] = notIn; //grab the arguments std::vector argvregs; //assert(Node->getNumOperands() < 8 && "Only 6 args supported"); for(int i = 2, e = Node->getNumOperands(); i < e; ++i) argvregs.push_back(SelectExpr(N.getOperand(i))); //in reg args for(int i = 0, e = std::min(6, (int)argvregs.size()); i < e; ++i) { unsigned args_int[] = {Alpha::R16, Alpha::R17, Alpha::R18, Alpha::R19, Alpha::R20, Alpha::R21}; unsigned args_float[] = {Alpha::F16, Alpha::F17, Alpha::F18, Alpha::F19, Alpha::F20, Alpha::F21}; switch(N.getOperand(i+2).getValueType()) { default: Node->dump(); N.getOperand(i).Val->dump(); std::cerr << "Type for " << i << " is: " << N.getOperand(i+2).getValueType() << "\n"; assert(0 && "Unknown value type for call"); case MVT::i1: case MVT::i8: case MVT::i16: case MVT::i32: case MVT::i64: BuildMI(BB, Alpha::BIS, 2, args_int[i]).addReg(argvregs[i]) .addReg(argvregs[i]); break; case MVT::f32: case MVT::f64: BuildMI(BB, Alpha::CPYS, 2, args_float[i]).addReg(argvregs[i]) .addReg(argvregs[i]); break; } } //in mem args for (int i = 6, e = argvregs.size(); i < e; ++i) { switch(N.getOperand(i+2).getValueType()) { default: Node->dump(); N.getOperand(i).Val->dump(); std::cerr << "Type for " << i << " is: " << N.getOperand(i+2).getValueType() << "\n"; assert(0 && "Unknown value type for call"); case MVT::i1: case MVT::i8: case MVT::i16: case MVT::i32: case MVT::i64: BuildMI(BB, Alpha::STQ, 3).addReg(argvregs[i]).addImm((i - 6) * 8) .addReg(Alpha::R30); break; case MVT::f32: BuildMI(BB, Alpha::STS, 3).addReg(argvregs[i]).addImm((i - 6) * 8) .addReg(Alpha::R30); break; case MVT::f64: BuildMI(BB, Alpha::STT, 3).addReg(argvregs[i]).addImm((i - 6) * 8) .addReg(Alpha::R30); break; } } //build the right kind of call GlobalAddressSDNode *GASD = dyn_cast(N.getOperand(1)); if (GASD && !GASD->getGlobal()->isExternal()) { //use PC relative branch call AlphaLowering.restoreGP(BB); BuildMI(BB, Alpha::BSR, 1, Alpha::R26) .addGlobalAddress(GASD->getGlobal(),true); } else { //no need to restore GP as we are doing an indirect call Tmp1 = SelectExpr(N.getOperand(1)); BuildMI(BB, Alpha::BIS, 2, Alpha::R27).addReg(Tmp1).addReg(Tmp1); BuildMI(BB, Alpha::JSR, 2, Alpha::R26).addReg(Alpha::R27).addImm(0); } //push the result into a virtual register switch (Node->getValueType(0)) { default: Node->dump(); assert(0 && "Unknown value type for call result!"); case MVT::Other: return notIn; case MVT::i1: case MVT::i8: case MVT::i16: case MVT::i32: case MVT::i64: BuildMI(BB, Alpha::BIS, 2, Result).addReg(Alpha::R0).addReg(Alpha::R0); break; case MVT::f32: case MVT::f64: BuildMI(BB, Alpha::CPYS, 2, Result).addReg(Alpha::F0).addReg(Alpha::F0); break; } return Result+N.ResNo; } case ISD::SIGN_EXTEND_INREG: { //do SDIV opt for all levels of ints if not dividing by a constant if (EnableAlphaIDIV && N.getOperand(0).getOpcode() == ISD::SDIV && N.getOperand(0).getOperand(1).getOpcode() != ISD::Constant) { unsigned Tmp4 = MakeReg(MVT::f64); unsigned Tmp5 = MakeReg(MVT::f64); unsigned Tmp6 = MakeReg(MVT::f64); unsigned Tmp7 = MakeReg(MVT::f64); unsigned Tmp8 = MakeReg(MVT::f64); unsigned Tmp9 = MakeReg(MVT::f64); Tmp1 = SelectExpr(N.getOperand(0).getOperand(0)); Tmp2 = SelectExpr(N.getOperand(0).getOperand(1)); MoveInt2FP(Tmp1, Tmp4, true); MoveInt2FP(Tmp2, Tmp5, true); BuildMI(BB, Alpha::CVTQT, 1, Tmp6).addReg(Alpha::F31).addReg(Tmp4); BuildMI(BB, Alpha::CVTQT, 1, Tmp7).addReg(Alpha::F31).addReg(Tmp5); BuildMI(BB, Alpha::DIVT, 2, Tmp8).addReg(Tmp6).addReg(Tmp7); BuildMI(BB, Alpha::CVTTQ, 1, Tmp9).addReg(Alpha::F31).addReg(Tmp8); MoveFP2Int(Tmp9, Result, true); return Result; } //Alpha has instructions for a bunch of signed 32 bit stuff if(cast(Node->getOperand(1))->getVT() == MVT::i32) { switch (N.getOperand(0).getOpcode()) { case ISD::ADD: case ISD::SUB: case ISD::MUL: { bool isAdd = N.getOperand(0).getOpcode() == ISD::ADD; bool isMul = N.getOperand(0).getOpcode() == ISD::MUL; //FIXME: first check for Scaled Adds and Subs! if(!isMul && N.getOperand(0).getOperand(0).getOpcode() == ISD::SHL && isSIntImmediateBounded(N.getOperand(0).getOperand(0).getOperand(1), SImm, 2, 3)) { bool use4 = SImm == 2; Tmp1 = SelectExpr(N.getOperand(0).getOperand(0).getOperand(0)); Tmp2 = SelectExpr(N.getOperand(0).getOperand(1)); BuildMI(BB, isAdd?(use4?Alpha::S4ADDL:Alpha::S8ADDL):(use4?Alpha::S4SUBL:Alpha::S8SUBL), 2,Result).addReg(Tmp1).addReg(Tmp2); } else if(isAdd && N.getOperand(0).getOperand(1).getOpcode() == ISD::SHL && isSIntImmediateBounded(N.getOperand(0).getOperand(1).getOperand(1), SImm, 2, 3)) { bool use4 = SImm == 2; Tmp1 = SelectExpr(N.getOperand(0).getOperand(1).getOperand(0)); Tmp2 = SelectExpr(N.getOperand(0).getOperand(0)); BuildMI(BB, use4?Alpha::S4ADDL:Alpha::S8ADDL, 2,Result).addReg(Tmp1).addReg(Tmp2); } else if(isSIntImmediateBounded(N.getOperand(0).getOperand(1), SImm, 0, 255)) { //Normal imm add/sub Opc = isAdd ? Alpha::ADDLi : (isMul ? Alpha::MULLi : Alpha::SUBLi); Tmp1 = SelectExpr(N.getOperand(0).getOperand(0)); BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addImm(SImm); } else if(!isMul && isSIntImmediate(N.getOperand(0).getOperand(1), SImm) && (((SImm << 32) >> 32) >= -255) && (((SImm << 32) >> 32) <= 0)) { //handle canonicalization Opc = isAdd ? Alpha::SUBLi : Alpha::ADDLi; Tmp1 = SelectExpr(N.getOperand(0).getOperand(0)); SImm = 0 - ((SImm << 32) >> 32); assert(SImm >= 0 && SImm <= 255); BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addImm(SImm); } else { //Normal add/sub Opc = isAdd ? Alpha::ADDL : (isMul ? Alpha::MULL : Alpha::SUBL); Tmp1 = SelectExpr(N.getOperand(0).getOperand(0)); Tmp2 = SelectExpr(N.getOperand(0).getOperand(1)); BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addReg(Tmp2); } return Result; } default: break; //Fall Though; } } //Every thing else fall though too, including unhandled opcodes above Tmp1 = SelectExpr(N.getOperand(0)); //std::cerr << "SrcT: " << MVN->getExtraValueType() << "\n"; switch(cast(Node->getOperand(1))->getVT()) { default: Node->dump(); assert(0 && "Sign Extend InReg not there yet"); break; case MVT::i32: { BuildMI(BB, Alpha::ADDLi, 2, Result).addReg(Tmp1).addImm(0); break; } case MVT::i16: BuildMI(BB, Alpha::SEXTW, 1, Result).addReg(Alpha::R31).addReg(Tmp1); break; case MVT::i8: BuildMI(BB, Alpha::SEXTB, 1, Result).addReg(Alpha::R31).addReg(Tmp1); break; case MVT::i1: Tmp2 = MakeReg(MVT::i64); BuildMI(BB, Alpha::ANDi, 2, Tmp2).addReg(Tmp1).addImm(1); BuildMI(BB, Alpha::SUBQ, 2, Result).addReg(Alpha::R31).addReg(Tmp2); break; } return Result; } case ISD::SETCC: { ISD::CondCode CC = cast(N.getOperand(2))->get(); if (MVT::isInteger(N.getOperand(0).getValueType())) { bool isConst = false; int dir; //Tmp1 = SelectExpr(N.getOperand(0)); if(isSIntImmediate(N.getOperand(1), SImm) && SImm <= 255 && SImm >= 0) isConst = true; switch (CC) { default: Node->dump(); assert(0 && "Unknown integer comparison!"); case ISD::SETEQ: Opc = isConst ? Alpha::CMPEQi : Alpha::CMPEQ; dir=1; break; case ISD::SETLT: Opc = isConst ? Alpha::CMPLTi : Alpha::CMPLT; dir = 1; break; case ISD::SETLE: Opc = isConst ? Alpha::CMPLEi : Alpha::CMPLE; dir = 1; break; case ISD::SETGT: Opc = Alpha::CMPLT; dir = 2; break; case ISD::SETGE: Opc = Alpha::CMPLE; dir = 2; break; case ISD::SETULT: Opc = isConst ? Alpha::CMPULTi : Alpha::CMPULT; dir = 1; break; case ISD::SETUGT: Opc = Alpha::CMPULT; dir = 2; break; case ISD::SETULE: Opc = isConst ? Alpha::CMPULEi : Alpha::CMPULE; dir = 1; break; case ISD::SETUGE: Opc = Alpha::CMPULE; dir = 2; break; case ISD::SETNE: {//Handle this one special //std::cerr << "Alpha does not have a setne.\n"; //abort(); Tmp1 = SelectExpr(N.getOperand(0)); Tmp2 = SelectExpr(N.getOperand(1)); Tmp3 = MakeReg(MVT::i64); BuildMI(BB, Alpha::CMPEQ, 2, Tmp3).addReg(Tmp1).addReg(Tmp2); //Remeber we have the Inv for this CC CCInvMap[N] = Tmp3; //and invert BuildMI(BB, Alpha::CMPEQ, 2, Result).addReg(Alpha::R31).addReg(Tmp3); return Result; } } if (dir == 1) { Tmp1 = SelectExpr(N.getOperand(0)); if (isConst) { BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addImm(SImm); } else { Tmp2 = SelectExpr(N.getOperand(1)); BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addReg(Tmp2); } } else { //if (dir == 2) { Tmp1 = SelectExpr(N.getOperand(1)); Tmp2 = SelectExpr(N.getOperand(0)); BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addReg(Tmp2); } } else { //do the comparison Tmp1 = MakeReg(MVT::f64); bool inv = SelectFPSetCC(N, Tmp1); //now arrange for Result (int) to have a 1 or 0 Tmp2 = MakeReg(MVT::i64); BuildMI(BB, Alpha::ADDQi, 2, Tmp2).addReg(Alpha::R31).addImm(1); Opc = inv?Alpha::CMOVNEi_FP:Alpha::CMOVEQi_FP; BuildMI(BB, Opc, 3, Result).addReg(Tmp2).addImm(0).addReg(Tmp1); } return Result; } case ISD::CopyFromReg: { ++count_ins; // Make sure we generate both values. if (Result != notIn) ExprMap[N.getValue(1)] = notIn; // Generate the token else Result = ExprMap[N.getValue(0)] = MakeReg(N.getValue(0).getValueType()); SDOperand Chain = N.getOperand(0); Select(Chain); unsigned r = cast(Node->getOperand(1))->getReg(); //std::cerr << "CopyFromReg " << Result << " = " << r << "\n"; if (MVT::isFloatingPoint(N.getValue(0).getValueType())) BuildMI(BB, Alpha::CPYS, 2, Result).addReg(r).addReg(r); else BuildMI(BB, Alpha::BIS, 2, Result).addReg(r).addReg(r); return Result; } //Most of the plain arithmetic and logic share the same form, and the same //constant immediate test case ISD::XOR: //Match Not if (isSIntImmediate(N.getOperand(1), SImm) && SImm == -1) { Tmp1 = SelectExpr(N.getOperand(0)); BuildMI(BB, Alpha::ORNOT, 2, Result).addReg(Alpha::R31).addReg(Tmp1); return Result; } //Fall through case ISD::AND: //handle zap if (opcode == ISD::AND && isUIntImmediate(N.getOperand(1), UImm)) { unsigned int build = 0; for(int i = 0; i < 8; ++i) { if ((UImm & 0x00FF) == 0x00FF) build |= 1 << i; else if ((UImm & 0x00FF) != 0) { build = 0; break; } UImm >>= 8; } if (build) { Tmp1 = SelectExpr(N.getOperand(0)); BuildMI(BB, Alpha::ZAPNOTi, 2, Result).addReg(Tmp1).addImm(build); return Result; } } case ISD::OR: //Check operand(0) == Not if (N.getOperand(0).getOpcode() == ISD::XOR && isSIntImmediate(N.getOperand(0).getOperand(1), SImm) && SImm == -1) { switch(opcode) { case ISD::AND: Opc = Alpha::BIC; break; case ISD::OR: Opc = Alpha::ORNOT; break; case ISD::XOR: Opc = Alpha::EQV; break; } Tmp1 = SelectExpr(N.getOperand(1)); Tmp2 = SelectExpr(N.getOperand(0).getOperand(0)); BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addReg(Tmp2); return Result; } //Check operand(1) == Not if (N.getOperand(1).getOpcode() == ISD::XOR && isSIntImmediate(N.getOperand(1).getOperand(1), SImm) && SImm == -1) { switch(opcode) { case ISD::AND: Opc = Alpha::BIC; break; case ISD::OR: Opc = Alpha::ORNOT; break; case ISD::XOR: Opc = Alpha::EQV; break; } Tmp1 = SelectExpr(N.getOperand(0)); Tmp2 = SelectExpr(N.getOperand(1).getOperand(0)); BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addReg(Tmp2); return Result; } //Fall through case ISD::SHL: case ISD::SRL: case ISD::SRA: case ISD::MUL: if(isSIntImmediateBounded(N.getOperand(1), SImm, 0, 255)) { switch(opcode) { case ISD::AND: Opc = Alpha::ANDi; break; case ISD::OR: Opc = Alpha::BISi; break; case ISD::XOR: Opc = Alpha::XORi; break; case ISD::SHL: Opc = Alpha::SLi; break; case ISD::SRL: Opc = Alpha::SRLi; break; case ISD::SRA: Opc = Alpha::SRAi; break; case ISD::MUL: Opc = Alpha::MULQi; break; }; Tmp1 = SelectExpr(N.getOperand(0)); BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addImm(SImm); } else { switch(opcode) { case ISD::AND: Opc = Alpha::AND; break; case ISD::OR: Opc = Alpha::BIS; break; case ISD::XOR: Opc = Alpha::XOR; break; case ISD::SHL: Opc = Alpha::SL; break; case ISD::SRL: Opc = Alpha::SRL; break; case ISD::SRA: Opc = Alpha::SRA; break; case ISD::MUL: Opc = isFP ? (DestType == MVT::f64 ? Alpha::MULT : Alpha::MULS) : Alpha::MULQ; break; }; Tmp1 = SelectExpr(N.getOperand(0)); Tmp2 = SelectExpr(N.getOperand(1)); BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addReg(Tmp2); } return Result; case ISD::ADD: case ISD::SUB: if (isFP) { ConstantFPSDNode *CN; if (opcode == ISD::ADD) Opc = DestType == MVT::f64 ? Alpha::ADDT : Alpha::ADDS; else Opc = DestType == MVT::f64 ? Alpha::SUBT : Alpha::SUBS; if (opcode == ISD::SUB && (CN = dyn_cast(N.getOperand(0))) && (CN->isExactlyValue(+0.0) || CN->isExactlyValue(-0.0))) { Tmp2 = SelectExpr(N.getOperand(1)); BuildMI(BB, Alpha::CPYSN, 2, Result).addReg(Tmp2).addReg(Tmp2); } else { Tmp1 = SelectExpr(N.getOperand(0)); Tmp2 = SelectExpr(N.getOperand(1)); BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addReg(Tmp2); } return Result; } else { bool isAdd = opcode == ISD::ADD; //first check for Scaled Adds and Subs! //Valid for add and sub if(N.getOperand(0).getOpcode() == ISD::SHL && isSIntImmediate(N.getOperand(0).getOperand(1), SImm) && (SImm == 2 || SImm == 3)) { bool use4 = SImm == 2; Tmp2 = SelectExpr(N.getOperand(0).getOperand(0)); if (isSIntImmediateBounded(N.getOperand(1), SImm, 0, 255)) BuildMI(BB, isAdd?(use4?Alpha::S4ADDQi:Alpha::S8ADDQi):(use4?Alpha::S4SUBQi:Alpha::S8SUBQi), 2, Result).addReg(Tmp2).addImm(SImm); else { Tmp1 = SelectExpr(N.getOperand(1)); BuildMI(BB, isAdd?(use4?Alpha::S4ADDQi:Alpha::S8ADDQi):(use4?Alpha::S4SUBQi:Alpha::S8SUBQi), 2, Result).addReg(Tmp2).addReg(Tmp1); } } //Position prevents subs else if(N.getOperand(1).getOpcode() == ISD::SHL && isAdd && isSIntImmediate(N.getOperand(1).getOperand(1), SImm) && (SImm == 2 || SImm == 3)) { bool use4 = SImm == 2; Tmp2 = SelectExpr(N.getOperand(1).getOperand(0)); if (isSIntImmediateBounded(N.getOperand(0), SImm, 0, 255)) BuildMI(BB, use4?Alpha::S4ADDQi:Alpha::S8ADDQi, 2, Result).addReg(Tmp2).addImm(SImm); else { Tmp1 = SelectExpr(N.getOperand(0)); BuildMI(BB, use4?Alpha::S4ADDQ:Alpha::S8ADDQ, 2, Result).addReg(Tmp2).addReg(Tmp1); } } //small addi else if(isSIntImmediateBounded(N.getOperand(1), SImm, 0, 255)) { //Normal imm add/sub Opc = isAdd ? Alpha::ADDQi : Alpha::SUBQi; Tmp1 = SelectExpr(N.getOperand(0)); BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addImm(SImm); } else if(isSIntImmediateBounded(N.getOperand(1), SImm, -255, 0)) { //inverted imm add/sub Opc = isAdd ? Alpha::SUBQi : Alpha::ADDQi; Tmp1 = SelectExpr(N.getOperand(0)); BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addImm(-SImm); } //larger addi else if(isSIntImmediateBounded(N.getOperand(1), SImm, -32767, 32767)) { //LDA Tmp1 = SelectExpr(N.getOperand(0)); if (!isAdd) SImm = -SImm; BuildMI(BB, Alpha::LDA, 2, Result).addImm(SImm).addReg(Tmp1); } //give up and do the operation else { //Normal add/sub Opc = isAdd ? Alpha::ADDQ : Alpha::SUBQ; Tmp1 = SelectExpr(N.getOperand(0)); Tmp2 = SelectExpr(N.getOperand(1)); BuildMI(BB, Opc, 2, Result).addReg(Tmp1).addReg(Tmp2); } return Result; } case ISD::SDIV: if (isFP) { Tmp1 = SelectExpr(N.getOperand(0)); Tmp2 = SelectExpr(N.getOperand(1)); BuildMI(BB, DestType == MVT::f64 ? Alpha::DIVT : Alpha::DIVS, 2, Result) .addReg(Tmp1).addReg(Tmp2); return Result; } else { //check if we can convert into a shift! if (isSIntImmediate(N.getOperand(1), SImm) && SImm != 0 && isPowerOf2_64(llabs(SImm))) { unsigned k = Log2_64(llabs(SImm)); Tmp1 = SelectExpr(N.getOperand(0)); if (k == 1) Tmp2 = Tmp1; else { Tmp2 = MakeReg(MVT::i64); BuildMI(BB, Alpha::SRAi, 2, Tmp2).addReg(Tmp1).addImm(k - 1); } Tmp3 = MakeReg(MVT::i64); BuildMI(BB, Alpha::SRLi, 2, Tmp3).addReg(Tmp2).addImm(64-k); unsigned Tmp4 = MakeReg(MVT::i64); BuildMI(BB, Alpha::ADDQ, 2, Tmp4).addReg(Tmp3).addReg(Tmp1); if (SImm > 0) BuildMI(BB, Alpha::SRAi, 2, Result).addReg(Tmp4).addImm(k); else { unsigned Tmp5 = MakeReg(MVT::i64); BuildMI(BB, Alpha::SRAi, 2, Tmp5).addReg(Tmp4).addImm(k); BuildMI(BB, Alpha::SUBQ, 2, Result).addReg(Alpha::R31).addReg(Tmp5); } return Result; } } //Else fall through case ISD::UDIV: { if (isSIntImmediate(N.getOperand(1), SImm) && (SImm >= 2 || SImm <= -2)) { // If this is a divide by constant, we can emit code using some magic // constants to implement it as a multiply instead. ExprMap.erase(N); if (opcode == ISD::SDIV) return SelectExpr(BuildSDIVSequence(N)); else return SelectExpr(BuildUDIVSequence(N)); } } //else fall though case ISD::UREM: case ISD::SREM: { const char* opstr = 0; switch(opcode) { case ISD::UREM: opstr = "__remqu"; break; case ISD::SREM: opstr = "__remq"; break; case ISD::UDIV: opstr = "__divqu"; break; case ISD::SDIV: opstr = "__divq"; break; } Tmp1 = SelectExpr(N.getOperand(0)); Tmp2 = SelectExpr(N.getOperand(1)); SDOperand Addr = ISelDAG->getExternalSymbol(opstr, AlphaLowering.getPointerTy()); Tmp3 = SelectExpr(Addr); //set up regs explicitly (helps Reg alloc) BuildMI(BB, Alpha::BIS, 2, Alpha::R24).addReg(Tmp1).addReg(Tmp1); BuildMI(BB, Alpha::BIS, 2, Alpha::R25).addReg(Tmp2).addReg(Tmp2); BuildMI(BB, Alpha::BIS, 2, Alpha::R27).addReg(Tmp3).addReg(Tmp3); BuildMI(BB, Alpha::JSRs, 2, Alpha::R23).addReg(Alpha::R27).addImm(0); BuildMI(BB, Alpha::BIS, 2, Result).addReg(Alpha::R27).addReg(Alpha::R27); return Result; } case ISD::FP_TO_UINT: case ISD::FP_TO_SINT: { assert (DestType == MVT::i64 && "only quads can be loaded to"); MVT::ValueType SrcType = N.getOperand(0).getValueType(); assert (SrcType == MVT::f32 || SrcType == MVT::f64); Tmp1 = SelectExpr(N.getOperand(0)); // Get the operand register if (SrcType == MVT::f32) { Tmp2 = MakeReg(MVT::f64); BuildMI(BB, Alpha::CVTST, 1, Tmp2).addReg(Alpha::F31).addReg(Tmp1); Tmp1 = Tmp2; } Tmp2 = MakeReg(MVT::f64); BuildMI(BB, Alpha::CVTTQ, 1, Tmp2).addReg(Alpha::F31).addReg(Tmp1); MoveFP2Int(Tmp2, Result, true); return Result; } case ISD::SELECT: if (isFP) { //Tmp1 = SelectExpr(N.getOperand(0)); //Cond unsigned TV = SelectExpr(N.getOperand(1)); //Use if TRUE unsigned FV = SelectExpr(N.getOperand(2)); //Use if FALSE SDOperand CC = N.getOperand(0); if (CC.getOpcode() == ISD::SETCC && !MVT::isInteger(CC.getOperand(0).getValueType())) { //FP Setcc -> Select yay! //for a cmp b: c = a - b; //a = b: c = 0 //a < b: c < 0 //a > b: c > 0 bool invTest = false; unsigned Tmp3; ConstantFPSDNode *CN; if ((CN = dyn_cast(CC.getOperand(1))) && (CN->isExactlyValue(+0.0) || CN->isExactlyValue(-0.0))) Tmp3 = SelectExpr(CC.getOperand(0)); else if ((CN = dyn_cast(CC.getOperand(0))) && (CN->isExactlyValue(+0.0) || CN->isExactlyValue(-0.0))) { Tmp3 = SelectExpr(CC.getOperand(1)); invTest = true; } else { unsigned Tmp1 = SelectExpr(CC.getOperand(0)); unsigned Tmp2 = SelectExpr(CC.getOperand(1)); bool isD = CC.getOperand(0).getValueType() == MVT::f64; Tmp3 = MakeReg(isD ? MVT::f64 : MVT::f32); BuildMI(BB, isD ? Alpha::SUBT : Alpha::SUBS, 2, Tmp3) .addReg(Tmp1).addReg(Tmp2); } switch (cast(CC.getOperand(2))->get()) { default: CC.Val->dump(); assert(0 && "Unknown FP comparison!"); case ISD::SETEQ: Opc = invTest ? Alpha::FCMOVNE : Alpha::FCMOVEQ; break; case ISD::SETLT: Opc = invTest ? Alpha::FCMOVGT : Alpha::FCMOVLT; break; case ISD::SETLE: Opc = invTest ? Alpha::FCMOVGE : Alpha::FCMOVLE; break; case ISD::SETGT: Opc = invTest ? Alpha::FCMOVLT : Alpha::FCMOVGT; break; case ISD::SETGE: Opc = invTest ? Alpha::FCMOVLE : Alpha::FCMOVGE; break; case ISD::SETNE: Opc = invTest ? Alpha::FCMOVEQ : Alpha::FCMOVNE; break; } BuildMI(BB, Opc, 3, Result).addReg(FV).addReg(TV).addReg(Tmp3); return Result; } else { Tmp1 = SelectExpr(N.getOperand(0)); //Cond BuildMI(BB, Alpha::FCMOVEQ_INT, 3, Result).addReg(TV).addReg(FV) .addReg(Tmp1); // // Spill the cond to memory and reload it from there. // unsigned Tmp4 = MakeReg(MVT::f64); // MoveIntFP(Tmp1, Tmp4, true); // //now ideally, we don't have to do anything to the flag... // // Get the condition into the zero flag. // BuildMI(BB, Alpha::FCMOVEQ, 3, Result).addReg(TV).addReg(FV).addReg(Tmp4); return Result; } } else { //FIXME: look at parent to decide if intCC can be folded, or if setCC(FP) //and can save stack use //Tmp1 = SelectExpr(N.getOperand(0)); //Cond //Tmp2 = SelectExpr(N.getOperand(1)); //Use if TRUE //Tmp3 = SelectExpr(N.getOperand(2)); //Use if FALSE // Get the condition into the zero flag. //BuildMI(BB, Alpha::CMOVEQ, 2, Result).addReg(Tmp2).addReg(Tmp3).addReg(Tmp1); SDOperand CC = N.getOperand(0); if (CC.getOpcode() == ISD::SETCC && !MVT::isInteger(CC.getOperand(0).getValueType())) { //FP Setcc -> Int Select Tmp1 = MakeReg(MVT::f64); Tmp2 = SelectExpr(N.getOperand(1)); //Use if TRUE Tmp3 = SelectExpr(N.getOperand(2)); //Use if FALSE bool inv = SelectFPSetCC(CC, Tmp1); BuildMI(BB, inv?Alpha::CMOVNE_FP:Alpha::CMOVEQ_FP, 2, Result) .addReg(Tmp2).addReg(Tmp3).addReg(Tmp1); return Result; } if (CC.getOpcode() == ISD::SETCC) { //Int SetCC -> Select //Dropping the CC is only useful if we are comparing to 0 if(isSIntImmediateBounded(CC.getOperand(1), SImm, 0, 0)) { //figure out a few things bool useImm = isSIntImmediateBounded(N.getOperand(2), SImm, 0, 255); //Fix up CC ISD::CondCode cCode= cast(CC.getOperand(2))->get(); if (useImm) //Invert sense to get Imm field right cCode = ISD::getSetCCInverse(cCode, true); //Choose the CMOV switch (cCode) { default: CC.Val->dump(); assert(0 && "Unknown integer comparison!"); case ISD::SETEQ: Opc = useImm?Alpha::CMOVEQi:Alpha::CMOVEQ; break; case ISD::SETLT: Opc = useImm?Alpha::CMOVLTi:Alpha::CMOVLT; break; case ISD::SETLE: Opc = useImm?Alpha::CMOVLEi:Alpha::CMOVLE; break; case ISD::SETGT: Opc = useImm?Alpha::CMOVGTi:Alpha::CMOVGT; break; case ISD::SETGE: Opc = useImm?Alpha::CMOVGEi:Alpha::CMOVGE; break; case ISD::SETULT: assert(0 && "unsigned < 0 is never true"); break; case ISD::SETUGT: Opc = useImm?Alpha::CMOVNEi:Alpha::CMOVNE; break; //Technically you could have this CC case ISD::SETULE: Opc = useImm?Alpha::CMOVEQi:Alpha::CMOVEQ; break; case ISD::SETUGE: assert(0 && "unsgined >= 0 is always true"); break; case ISD::SETNE: Opc = useImm?Alpha::CMOVNEi:Alpha::CMOVNE; break; } Tmp1 = SelectExpr(CC.getOperand(0)); //Cond if (useImm) { Tmp3 = SelectExpr(N.getOperand(1)); //Use if FALSE BuildMI(BB, Opc, 2, Result).addReg(Tmp3).addImm(SImm).addReg(Tmp1); } else { Tmp2 = SelectExpr(N.getOperand(1)); //Use if TRUE Tmp3 = SelectExpr(N.getOperand(2)); //Use if FALSE BuildMI(BB, Opc, 2, Result).addReg(Tmp3).addReg(Tmp2).addReg(Tmp1); } return Result; } //Otherwise, fall though } Tmp1 = SelectExpr(N.getOperand(0)); //Cond Tmp2 = SelectExpr(N.getOperand(1)); //Use if TRUE Tmp3 = SelectExpr(N.getOperand(2)); //Use if FALSE BuildMI(BB, Alpha::CMOVEQ, 2, Result).addReg(Tmp2).addReg(Tmp3) .addReg(Tmp1); return Result; } case ISD::Constant: { int64_t val = (int64_t)cast(N)->getValue(); int zero_extend_top = 0; if (val > 0 && (val & 0xFFFFFFFF00000000ULL) == 0 && ((int32_t)val < 0)) { //try a small load and zero extend val = (int32_t)val; zero_extend_top = 15; } if (val <= IMM_HIGH && val >= IMM_LOW) { if(!zero_extend_top) BuildMI(BB, Alpha::LDA, 2, Result).addImm(val).addReg(Alpha::R31); else { Tmp1 = MakeReg(MVT::i64); BuildMI(BB, Alpha::LDA, 2, Tmp1).addImm(val).addReg(Alpha::R31); BuildMI(BB, Alpha::ZAPNOT, 2, Result).addReg(Tmp1).addImm(zero_extend_top); } } else if (val <= (int64_t)IMM_HIGH +(int64_t)IMM_HIGH* (int64_t)IMM_MULT && val >= (int64_t)IMM_LOW + (int64_t)IMM_LOW * (int64_t)IMM_MULT) { Tmp1 = MakeReg(MVT::i64); BuildMI(BB, Alpha::LDAH, 2, Tmp1).addImm(getUpper16(val)) .addReg(Alpha::R31); if (!zero_extend_top) BuildMI(BB, Alpha::LDA, 2, Result).addImm(getLower16(val)).addReg(Tmp1); else { Tmp3 = MakeReg(MVT::i64); BuildMI(BB, Alpha::LDA, 2, Tmp3).addImm(getLower16(val)).addReg(Tmp1); BuildMI(BB, Alpha::ZAPNOT, 2, Result).addReg(Tmp3).addImm(zero_extend_top); } } else { //re-get the val since we are going to mem anyway val = (int64_t)cast(N)->getValue(); MachineConstantPool *CP = BB->getParent()->getConstantPool(); ConstantUInt *C = ConstantUInt::get(Type::getPrimitiveType(Type::ULongTyID) , val); unsigned CPI = CP->getConstantPoolIndex(C); AlphaLowering.restoreGP(BB); has_sym = true; Tmp1 = MakeReg(MVT::i64); BuildMI(BB, Alpha::LDAHr, 2, Tmp1).addConstantPoolIndex(CPI) .addReg(Alpha::R29); if (EnableAlphaLSMark) BuildMI(BB, Alpha::MEMLABEL, 4).addImm(5).addImm(0).addImm(0) .addImm(getUID()); BuildMI(BB, Alpha::LDQr, 2, Result).addConstantPoolIndex(CPI) .addReg(Tmp1); } return Result; } case ISD::FNEG: if(ISD::FABS == N.getOperand(0).getOpcode()) { Tmp1 = SelectExpr(N.getOperand(0).getOperand(0)); BuildMI(BB, Alpha::CPYSN, 2, Result).addReg(Alpha::F31).addReg(Tmp1); } else { Tmp1 = SelectExpr(N.getOperand(0)); BuildMI(BB, Alpha::CPYSN, 2, Result).addReg(Tmp1).addReg(Tmp1); } return Result; case ISD::FABS: Tmp1 = SelectExpr(N.getOperand(0)); BuildMI(BB, Alpha::CPYS, 2, Result).addReg(Alpha::F31).addReg(Tmp1); return Result; case ISD::FP_ROUND: assert (DestType == MVT::f32 && N.getOperand(0).getValueType() == MVT::f64 && "only f64 to f32 conversion supported here"); Tmp1 = SelectExpr(N.getOperand(0)); BuildMI(BB, Alpha::CVTTS, 1, Result).addReg(Alpha::F31).addReg(Tmp1); return Result; case ISD::FP_EXTEND: assert (DestType == MVT::f64 && N.getOperand(0).getValueType() == MVT::f32 && "only f32 to f64 conversion supported here"); Tmp1 = SelectExpr(N.getOperand(0)); BuildMI(BB, Alpha::CVTST, 1, Result).addReg(Alpha::F31).addReg(Tmp1); return Result; case ISD::ConstantFP: if (ConstantFPSDNode *CN = dyn_cast(N)) { if (CN->isExactlyValue(+0.0)) { BuildMI(BB, Alpha::CPYS, 2, Result).addReg(Alpha::F31) .addReg(Alpha::F31); } else if ( CN->isExactlyValue(-0.0)) { BuildMI(BB, Alpha::CPYSN, 2, Result).addReg(Alpha::F31) .addReg(Alpha::F31); } else { abort(); } } return Result; case ISD::SINT_TO_FP: { assert (N.getOperand(0).getValueType() == MVT::i64 && "only quads can be loaded from"); Tmp1 = SelectExpr(N.getOperand(0)); // Get the operand register Tmp2 = MakeReg(MVT::f64); MoveInt2FP(Tmp1, Tmp2, true); Opc = DestType == MVT::f64 ? Alpha::CVTQT : Alpha::CVTQS; BuildMI(BB, Opc, 1, Result).addReg(Alpha::F31).addReg(Tmp2); return Result; } } return 0; } void AlphaISel::Select(SDOperand N) { unsigned Tmp1, Tmp2, Opc; unsigned opcode = N.getOpcode(); if (!ExprMap.insert(std::make_pair(N, notIn)).second) return; // Already selected. SDNode *Node = N.Val; switch (opcode) { default: Node->dump(); std::cerr << "\n"; assert(0 && "Node not handled yet!"); case ISD::BRCOND: { SelectBranchCC(N); return; } case ISD::BR: { MachineBasicBlock *Dest = cast(N.getOperand(1))->getBasicBlock(); Select(N.getOperand(0)); BuildMI(BB, Alpha::BR, 1, Alpha::R31).addMBB(Dest); return; } case ISD::ImplicitDef: ++count_ins; Select(N.getOperand(0)); BuildMI(BB, Alpha::IDEF, 0, cast(N.getOperand(1))->getReg()); return; case ISD::EntryToken: return; // Noop case ISD::TokenFactor: for (unsigned i = 0, e = Node->getNumOperands(); i != e; ++i) Select(Node->getOperand(i)); //N.Val->dump(); std::cerr << "\n"; //assert(0 && "Node not handled yet!"); return; case ISD::CopyToReg: ++count_outs; Select(N.getOperand(0)); Tmp1 = SelectExpr(N.getOperand(2)); Tmp2 = cast(N.getOperand(1))->getReg(); if (Tmp1 != Tmp2) { if (N.getOperand(2).getValueType() == MVT::f64 || N.getOperand(2).getValueType() == MVT::f32) BuildMI(BB, Alpha::CPYS, 2, Tmp2).addReg(Tmp1).addReg(Tmp1); else BuildMI(BB, Alpha::BIS, 2, Tmp2).addReg(Tmp1).addReg(Tmp1); } return; case ISD::RET: ++count_outs; switch (N.getNumOperands()) { default: std::cerr << N.getNumOperands() << "\n"; for (unsigned i = 0; i < N.getNumOperands(); ++i) std::cerr << N.getOperand(i).getValueType() << "\n"; Node->dump(); assert(0 && "Unknown return instruction!"); case 2: Select(N.getOperand(0)); Tmp1 = SelectExpr(N.getOperand(1)); switch (N.getOperand(1).getValueType()) { default: Node->dump(); assert(0 && "All other types should have been promoted!!"); case MVT::f64: case MVT::f32: BuildMI(BB, Alpha::CPYS, 2, Alpha::F0).addReg(Tmp1).addReg(Tmp1); break; case MVT::i32: case MVT::i64: BuildMI(BB, Alpha::BIS, 2, Alpha::R0).addReg(Tmp1).addReg(Tmp1); break; } break; case 1: Select(N.getOperand(0)); break; } // Just emit a 'ret' instruction AlphaLowering.restoreRA(BB); BuildMI(BB, Alpha::RET, 2, Alpha::R31).addReg(Alpha::R26).addImm(1); return; case ISD::TRUNCSTORE: case ISD::STORE: { SDOperand Chain = N.getOperand(0); SDOperand Value = N.getOperand(1); SDOperand Address = N.getOperand(2); Select(Chain); Tmp1 = SelectExpr(Value); //value if (opcode == ISD::STORE) { switch(Value.getValueType()) { default: assert(0 && "unknown Type in store"); case MVT::i64: Opc = Alpha::STQ; break; case MVT::f64: Opc = Alpha::STT; break; case MVT::f32: Opc = Alpha::STS; break; } } else { //ISD::TRUNCSTORE switch(cast(Node->getOperand(4))->getVT()) { default: assert(0 && "unknown Type in store"); case MVT::i1: //FIXME: DAG does not promote this load case MVT::i8: Opc = Alpha::STB; break; case MVT::i16: Opc = Alpha::STW; break; case MVT::i32: Opc = Alpha::STL; break; } } int i, j, k; if (EnableAlphaLSMark) getValueInfo(cast(N.getOperand(3))->getValue(), i, j, k); GlobalAddressSDNode *GASD = dyn_cast(Address); if (GASD && !GASD->getGlobal()->isExternal()) { Tmp2 = MakeReg(MVT::i64); AlphaLowering.restoreGP(BB); BuildMI(BB, Alpha::LDAHr, 2, Tmp2) .addGlobalAddress(GASD->getGlobal()).addReg(Alpha::R29); if (EnableAlphaLSMark) BuildMI(BB, Alpha::MEMLABEL, 4).addImm(i).addImm(j).addImm(k) .addImm(getUID()); BuildMI(BB, GetRelVersion(Opc), 3).addReg(Tmp1) .addGlobalAddress(GASD->getGlobal()).addReg(Tmp2); } else if(Address.getOpcode() == ISD::FrameIndex) { if (EnableAlphaLSMark) BuildMI(BB, Alpha::MEMLABEL, 4).addImm(i).addImm(j).addImm(k) .addImm(getUID()); BuildMI(BB, Opc, 3).addReg(Tmp1) .addFrameIndex(cast(Address)->getIndex()) .addReg(Alpha::F31); } else { long offset; SelectAddr(Address, Tmp2, offset); if (EnableAlphaLSMark) BuildMI(BB, Alpha::MEMLABEL, 4).addImm(i).addImm(j).addImm(k) .addImm(getUID()); BuildMI(BB, Opc, 3).addReg(Tmp1).addImm(offset).addReg(Tmp2); } return; } case ISD::EXTLOAD: case ISD::SEXTLOAD: case ISD::ZEXTLOAD: case ISD::LOAD: case ISD::CopyFromReg: case ISD::TAILCALL: case ISD::CALL: case ISD::DYNAMIC_STACKALLOC: ExprMap.erase(N); SelectExpr(N); return; case ISD::CALLSEQ_START: case ISD::CALLSEQ_END: Select(N.getOperand(0)); Tmp1 = cast(N.getOperand(1))->getValue(); Opc = N.getOpcode() == ISD::CALLSEQ_START ? Alpha::ADJUSTSTACKDOWN : Alpha::ADJUSTSTACKUP; BuildMI(BB, Opc, 1).addImm(Tmp1); return; case ISD::PCMARKER: Select(N.getOperand(0)); //Chain BuildMI(BB, Alpha::PCLABEL, 2) .addImm( cast(N.getOperand(1))->getValue()); return; } assert(0 && "Should not be reached!"); } /// createAlphaPatternInstructionSelector - This pass converts an LLVM function /// into a machine code representation using pattern matching and a machine /// description file. /// FunctionPass *llvm::createAlphaPatternInstructionSelector(TargetMachine &TM) { return new AlphaISel(TM); }