예제 #1
0
파일: ssa.go 프로젝트: achanda/go
func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
	s.SetLineno(v.Line)
	switch v.Op {
	case ssa.OpInitMem:
		// memory arg needs no code
	case ssa.OpArg:
		// input args need no code
	case ssa.OpSP, ssa.OpSB, ssa.OpGetG:
		// nothing to do
	case ssa.OpCopy, ssa.OpARMMOVWconvert, ssa.OpARMMOVWreg:
		if v.Type.IsMemory() {
			return
		}
		x := v.Args[0].Reg()
		y := v.Reg()
		if x == y {
			return
		}
		as := arm.AMOVW
		if v.Type.IsFloat() {
			switch v.Type.Size() {
			case 4:
				as = arm.AMOVF
			case 8:
				as = arm.AMOVD
			default:
				panic("bad float size")
			}
		}
		p := gc.Prog(as)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = x
		p.To.Type = obj.TYPE_REG
		p.To.Reg = y
	case ssa.OpARMMOVWnop:
		if v.Reg() != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		// nothing to do
	case ssa.OpLoadReg:
		if v.Type.IsFlags() {
			v.Fatalf("load flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(loadByType(v.Type))
		gc.AddrAuto(&p.From, v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpPhi:
		gc.CheckLoweredPhi(v)
	case ssa.OpStoreReg:
		if v.Type.IsFlags() {
			v.Fatalf("store flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(storeByType(v.Type))
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		gc.AddrAuto(&p.To, v)
	case ssa.OpARMUDIVrtcall:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = obj.Linklookup(gc.Ctxt, "udiv", 0)
	case ssa.OpARMADD,
		ssa.OpARMADC,
		ssa.OpARMSUB,
		ssa.OpARMSBC,
		ssa.OpARMRSB,
		ssa.OpARMAND,
		ssa.OpARMOR,
		ssa.OpARMXOR,
		ssa.OpARMBIC,
		ssa.OpARMMUL,
		ssa.OpARMADDF,
		ssa.OpARMADDD,
		ssa.OpARMSUBF,
		ssa.OpARMSUBD,
		ssa.OpARMMULF,
		ssa.OpARMMULD,
		ssa.OpARMDIVF,
		ssa.OpARMDIVD:
		r := v.Reg()
		r1 := v.Args[0].Reg()
		r2 := v.Args[1].Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r2
		p.Reg = r1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpARMADDS,
		ssa.OpARMSUBS:
		r := v.Reg0()
		r1 := v.Args[0].Reg()
		r2 := v.Args[1].Reg()
		p := gc.Prog(v.Op.Asm())
		p.Scond = arm.C_SBIT
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r2
		p.Reg = r1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpARMSLL,
		ssa.OpARMSRL,
		ssa.OpARMSRA:
		r := v.Reg()
		r1 := v.Args[0].Reg()
		r2 := v.Args[1].Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r2
		p.Reg = r1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpARMSRAcond:
		// ARM shift instructions uses only the low-order byte of the shift amount
		// generate conditional instructions to deal with large shifts
		// flag is already set
		// SRA.HS	$31, Rarg0, Rdst // shift 31 bits to get the sign bit
		// SRA.LO	Rarg1, Rarg0, Rdst
		r := v.Reg()
		r1 := v.Args[0].Reg()
		r2 := v.Args[1].Reg()
		p := gc.Prog(arm.ASRA)
		p.Scond = arm.C_SCOND_HS
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = 31
		p.Reg = r1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
		p = gc.Prog(arm.ASRA)
		p.Scond = arm.C_SCOND_LO
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r2
		p.Reg = r1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpARMADDconst,
		ssa.OpARMADCconst,
		ssa.OpARMSUBconst,
		ssa.OpARMSBCconst,
		ssa.OpARMRSBconst,
		ssa.OpARMRSCconst,
		ssa.OpARMANDconst,
		ssa.OpARMORconst,
		ssa.OpARMXORconst,
		ssa.OpARMBICconst,
		ssa.OpARMSLLconst,
		ssa.OpARMSRLconst,
		ssa.OpARMSRAconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMADDSconst,
		ssa.OpARMSUBSconst,
		ssa.OpARMRSBSconst:
		p := gc.Prog(v.Op.Asm())
		p.Scond = arm.C_SBIT
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg0()
	case ssa.OpARMSRRconst:
		genshift(arm.AMOVW, 0, v.Args[0].Reg(), v.Reg(), arm.SHIFT_RR, v.AuxInt)
	case ssa.OpARMADDshiftLL,
		ssa.OpARMADCshiftLL,
		ssa.OpARMSUBshiftLL,
		ssa.OpARMSBCshiftLL,
		ssa.OpARMRSBshiftLL,
		ssa.OpARMRSCshiftLL,
		ssa.OpARMANDshiftLL,
		ssa.OpARMORshiftLL,
		ssa.OpARMXORshiftLL,
		ssa.OpARMBICshiftLL:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm.SHIFT_LL, v.AuxInt)
	case ssa.OpARMADDSshiftLL,
		ssa.OpARMSUBSshiftLL,
		ssa.OpARMRSBSshiftLL:
		p := genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg0(), arm.SHIFT_LL, v.AuxInt)
		p.Scond = arm.C_SBIT
	case ssa.OpARMADDshiftRL,
		ssa.OpARMADCshiftRL,
		ssa.OpARMSUBshiftRL,
		ssa.OpARMSBCshiftRL,
		ssa.OpARMRSBshiftRL,
		ssa.OpARMRSCshiftRL,
		ssa.OpARMANDshiftRL,
		ssa.OpARMORshiftRL,
		ssa.OpARMXORshiftRL,
		ssa.OpARMBICshiftRL:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm.SHIFT_LR, v.AuxInt)
	case ssa.OpARMADDSshiftRL,
		ssa.OpARMSUBSshiftRL,
		ssa.OpARMRSBSshiftRL:
		p := genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg0(), arm.SHIFT_LR, v.AuxInt)
		p.Scond = arm.C_SBIT
	case ssa.OpARMADDshiftRA,
		ssa.OpARMADCshiftRA,
		ssa.OpARMSUBshiftRA,
		ssa.OpARMSBCshiftRA,
		ssa.OpARMRSBshiftRA,
		ssa.OpARMRSCshiftRA,
		ssa.OpARMANDshiftRA,
		ssa.OpARMORshiftRA,
		ssa.OpARMXORshiftRA,
		ssa.OpARMBICshiftRA:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm.SHIFT_AR, v.AuxInt)
	case ssa.OpARMADDSshiftRA,
		ssa.OpARMSUBSshiftRA,
		ssa.OpARMRSBSshiftRA:
		p := genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg0(), arm.SHIFT_AR, v.AuxInt)
		p.Scond = arm.C_SBIT
	case ssa.OpARMXORshiftRR:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm.SHIFT_RR, v.AuxInt)
	case ssa.OpARMMVNshiftLL:
		genshift(v.Op.Asm(), 0, v.Args[0].Reg(), v.Reg(), arm.SHIFT_LL, v.AuxInt)
	case ssa.OpARMMVNshiftRL:
		genshift(v.Op.Asm(), 0, v.Args[0].Reg(), v.Reg(), arm.SHIFT_LR, v.AuxInt)
	case ssa.OpARMMVNshiftRA:
		genshift(v.Op.Asm(), 0, v.Args[0].Reg(), v.Reg(), arm.SHIFT_AR, v.AuxInt)
	case ssa.OpARMMVNshiftLLreg:
		genregshift(v.Op.Asm(), 0, v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm.SHIFT_LL)
	case ssa.OpARMMVNshiftRLreg:
		genregshift(v.Op.Asm(), 0, v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm.SHIFT_LR)
	case ssa.OpARMMVNshiftRAreg:
		genregshift(v.Op.Asm(), 0, v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm.SHIFT_AR)
	case ssa.OpARMADDshiftLLreg,
		ssa.OpARMADCshiftLLreg,
		ssa.OpARMSUBshiftLLreg,
		ssa.OpARMSBCshiftLLreg,
		ssa.OpARMRSBshiftLLreg,
		ssa.OpARMRSCshiftLLreg,
		ssa.OpARMANDshiftLLreg,
		ssa.OpARMORshiftLLreg,
		ssa.OpARMXORshiftLLreg,
		ssa.OpARMBICshiftLLreg:
		genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), v.Reg(), arm.SHIFT_LL)
	case ssa.OpARMADDSshiftLLreg,
		ssa.OpARMSUBSshiftLLreg,
		ssa.OpARMRSBSshiftLLreg:
		p := genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), v.Reg0(), arm.SHIFT_LL)
		p.Scond = arm.C_SBIT
	case ssa.OpARMADDshiftRLreg,
		ssa.OpARMADCshiftRLreg,
		ssa.OpARMSUBshiftRLreg,
		ssa.OpARMSBCshiftRLreg,
		ssa.OpARMRSBshiftRLreg,
		ssa.OpARMRSCshiftRLreg,
		ssa.OpARMANDshiftRLreg,
		ssa.OpARMORshiftRLreg,
		ssa.OpARMXORshiftRLreg,
		ssa.OpARMBICshiftRLreg:
		genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), v.Reg(), arm.SHIFT_LR)
	case ssa.OpARMADDSshiftRLreg,
		ssa.OpARMSUBSshiftRLreg,
		ssa.OpARMRSBSshiftRLreg:
		p := genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), v.Reg0(), arm.SHIFT_LR)
		p.Scond = arm.C_SBIT
	case ssa.OpARMADDshiftRAreg,
		ssa.OpARMADCshiftRAreg,
		ssa.OpARMSUBshiftRAreg,
		ssa.OpARMSBCshiftRAreg,
		ssa.OpARMRSBshiftRAreg,
		ssa.OpARMRSCshiftRAreg,
		ssa.OpARMANDshiftRAreg,
		ssa.OpARMORshiftRAreg,
		ssa.OpARMXORshiftRAreg,
		ssa.OpARMBICshiftRAreg:
		genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), v.Reg(), arm.SHIFT_AR)
	case ssa.OpARMADDSshiftRAreg,
		ssa.OpARMSUBSshiftRAreg,
		ssa.OpARMRSBSshiftRAreg:
		p := genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), v.Reg0(), arm.SHIFT_AR)
		p.Scond = arm.C_SBIT
	case ssa.OpARMHMUL,
		ssa.OpARMHMULU:
		// 32-bit high multiplication
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REGREG
		p.To.Reg = v.Reg()
		p.To.Offset = arm.REGTMP // throw away low 32-bit into tmp register
	case ssa.OpARMMULLU:
		// 32-bit multiplication, results 64-bit, high 32-bit in out0, low 32-bit in out1
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REGREG
		p.To.Reg = v.Reg0()           // high 32-bit
		p.To.Offset = int64(v.Reg1()) // low 32-bit
	case ssa.OpARMMULA:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REGREG2
		p.To.Reg = v.Reg()                   // result
		p.To.Offset = int64(v.Args[2].Reg()) // addend
	case ssa.OpARMMOVWconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMMOVFconst,
		ssa.OpARMMOVDconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_FCONST
		p.From.Val = math.Float64frombits(uint64(v.AuxInt))
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMCMP,
		ssa.OpARMCMN,
		ssa.OpARMTST,
		ssa.OpARMTEQ,
		ssa.OpARMCMPF,
		ssa.OpARMCMPD:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		// Special layout in ARM assembly
		// Comparing to x86, the operands of ARM's CMP are reversed.
		p.From.Reg = v.Args[1].Reg()
		p.Reg = v.Args[0].Reg()
	case ssa.OpARMCMPconst,
		ssa.OpARMCMNconst,
		ssa.OpARMTSTconst,
		ssa.OpARMTEQconst:
		// Special layout in ARM assembly
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.Reg = v.Args[0].Reg()
	case ssa.OpARMCMPF0,
		ssa.OpARMCMPD0:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
	case ssa.OpARMCMPshiftLL:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), 0, arm.SHIFT_LL, v.AuxInt)
	case ssa.OpARMCMPshiftRL:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), 0, arm.SHIFT_LR, v.AuxInt)
	case ssa.OpARMCMPshiftRA:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), 0, arm.SHIFT_AR, v.AuxInt)
	case ssa.OpARMCMPshiftLLreg:
		genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), 0, arm.SHIFT_LL)
	case ssa.OpARMCMPshiftRLreg:
		genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), 0, arm.SHIFT_LR)
	case ssa.OpARMCMPshiftRAreg:
		genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), 0, arm.SHIFT_AR)
	case ssa.OpARMMOVWaddr:
		p := gc.Prog(arm.AMOVW)
		p.From.Type = obj.TYPE_ADDR
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()

		var wantreg string
		// MOVW $sym+off(base), R
		// the assembler expands it as the following:
		// - base is SP: add constant offset to SP (R13)
		//               when constant is large, tmp register (R11) may be used
		// - base is SB: load external address from constant pool (use relocation)
		switch v.Aux.(type) {
		default:
			v.Fatalf("aux is of unknown type %T", v.Aux)
		case *ssa.ExternSymbol:
			wantreg = "SB"
			gc.AddAux(&p.From, v)
		case *ssa.ArgSymbol, *ssa.AutoSymbol:
			wantreg = "SP"
			gc.AddAux(&p.From, v)
		case nil:
			// No sym, just MOVW $off(SP), R
			wantreg = "SP"
			p.From.Reg = arm.REGSP
			p.From.Offset = v.AuxInt
		}
		if reg := v.Args[0].RegName(); reg != wantreg {
			v.Fatalf("bad reg %s for symbol type %T, want %s", reg, v.Aux, wantreg)
		}

	case ssa.OpARMMOVBload,
		ssa.OpARMMOVBUload,
		ssa.OpARMMOVHload,
		ssa.OpARMMOVHUload,
		ssa.OpARMMOVWload,
		ssa.OpARMMOVFload,
		ssa.OpARMMOVDload:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMMOVBstore,
		ssa.OpARMMOVHstore,
		ssa.OpARMMOVWstore,
		ssa.OpARMMOVFstore,
		ssa.OpARMMOVDstore:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpARMMOVWloadidx:
		// this is just shift 0 bits
		fallthrough
	case ssa.OpARMMOVWloadshiftLL:
		p := genshift(v.Op.Asm(), 0, v.Args[1].Reg(), v.Reg(), arm.SHIFT_LL, v.AuxInt)
		p.From.Reg = v.Args[0].Reg()
	case ssa.OpARMMOVWloadshiftRL:
		p := genshift(v.Op.Asm(), 0, v.Args[1].Reg(), v.Reg(), arm.SHIFT_LR, v.AuxInt)
		p.From.Reg = v.Args[0].Reg()
	case ssa.OpARMMOVWloadshiftRA:
		p := genshift(v.Op.Asm(), 0, v.Args[1].Reg(), v.Reg(), arm.SHIFT_AR, v.AuxInt)
		p.From.Reg = v.Args[0].Reg()
	case ssa.OpARMMOVWstoreidx:
		// this is just shift 0 bits
		fallthrough
	case ssa.OpARMMOVWstoreshiftLL:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_SHIFT
		p.To.Reg = v.Args[0].Reg()
		p.To.Offset = int64(makeshift(v.Args[1].Reg(), arm.SHIFT_LL, v.AuxInt))
	case ssa.OpARMMOVWstoreshiftRL:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_SHIFT
		p.To.Reg = v.Args[0].Reg()
		p.To.Offset = int64(makeshift(v.Args[1].Reg(), arm.SHIFT_LR, v.AuxInt))
	case ssa.OpARMMOVWstoreshiftRA:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_SHIFT
		p.To.Reg = v.Args[0].Reg()
		p.To.Offset = int64(makeshift(v.Args[1].Reg(), arm.SHIFT_AR, v.AuxInt))
	case ssa.OpARMMOVBreg,
		ssa.OpARMMOVBUreg,
		ssa.OpARMMOVHreg,
		ssa.OpARMMOVHUreg:
		a := v.Args[0]
		for a.Op == ssa.OpCopy || a.Op == ssa.OpARMMOVWreg || a.Op == ssa.OpARMMOVWnop {
			a = a.Args[0]
		}
		if a.Op == ssa.OpLoadReg {
			t := a.Type
			switch {
			case v.Op == ssa.OpARMMOVBreg && t.Size() == 1 && t.IsSigned(),
				v.Op == ssa.OpARMMOVBUreg && t.Size() == 1 && !t.IsSigned(),
				v.Op == ssa.OpARMMOVHreg && t.Size() == 2 && t.IsSigned(),
				v.Op == ssa.OpARMMOVHUreg && t.Size() == 2 && !t.IsSigned():
				// arg is a proper-typed load, already zero/sign-extended, don't extend again
				if v.Reg() == v.Args[0].Reg() {
					return
				}
				p := gc.Prog(arm.AMOVW)
				p.From.Type = obj.TYPE_REG
				p.From.Reg = v.Args[0].Reg()
				p.To.Type = obj.TYPE_REG
				p.To.Reg = v.Reg()
				return
			default:
			}
		}
		fallthrough
	case ssa.OpARMMVN,
		ssa.OpARMCLZ,
		ssa.OpARMSQRTD,
		ssa.OpARMNEGF,
		ssa.OpARMNEGD,
		ssa.OpARMMOVWF,
		ssa.OpARMMOVWD,
		ssa.OpARMMOVFW,
		ssa.OpARMMOVDW,
		ssa.OpARMMOVFD,
		ssa.OpARMMOVDF:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMMOVWUF,
		ssa.OpARMMOVWUD,
		ssa.OpARMMOVFWU,
		ssa.OpARMMOVDWU:
		p := gc.Prog(v.Op.Asm())
		p.Scond = arm.C_UBIT
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMCMOVWHSconst:
		p := gc.Prog(arm.AMOVW)
		p.Scond = arm.C_SCOND_HS
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMCMOVWLSconst:
		p := gc.Prog(arm.AMOVW)
		p.Scond = arm.C_SCOND_LS
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMCALLstatic:
		if v.Aux.(*gc.Sym) == gc.Deferreturn.Sym {
			// Deferred calls will appear to be returning to
			// the CALL deferreturn(SB) that we are about to emit.
			// However, the stack trace code will show the line
			// of the instruction byte before the return PC.
			// To avoid that being an unrelated instruction,
			// insert an actual hardware NOP that will have the right line number.
			// This is different from obj.ANOP, which is a virtual no-op
			// that doesn't make it into the instruction stream.
			ginsnop()
		}
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(v.Aux.(*gc.Sym))
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARMCALLclosure:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Offset = 0
		p.To.Reg = v.Args[0].Reg()
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARMCALLdefer:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Deferproc.Sym)
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARMCALLgo:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Newproc.Sym)
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARMCALLinter:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Offset = 0
		p.To.Reg = v.Args[0].Reg()
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARMDUFFZERO:
		p := gc.Prog(obj.ADUFFZERO)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Pkglookup("duffzero", gc.Runtimepkg))
		p.To.Offset = v.AuxInt
	case ssa.OpARMDUFFCOPY:
		p := gc.Prog(obj.ADUFFCOPY)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Pkglookup("duffcopy", gc.Runtimepkg))
		p.To.Offset = v.AuxInt
	case ssa.OpARMLoweredNilCheck:
		// Issue a load which will fault if arg is nil.
		p := gc.Prog(arm.AMOVB)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = arm.REGTMP
		if gc.Debug_checknil != 0 && v.Line > 1 { // v.Line==1 in generated wrappers
			gc.Warnl(v.Line, "generated nil check")
		}
	case ssa.OpARMLoweredZero:
		// MOVW.P	Rarg2, 4(R1)
		// CMP	Rarg1, R1
		// BLE	-2(PC)
		// arg1 is the address of the last element to zero
		// arg2 is known to be zero
		// auxint is alignment
		var sz int64
		var mov obj.As
		switch {
		case v.AuxInt%4 == 0:
			sz = 4
			mov = arm.AMOVW
		case v.AuxInt%2 == 0:
			sz = 2
			mov = arm.AMOVH
		default:
			sz = 1
			mov = arm.AMOVB
		}
		p := gc.Prog(mov)
		p.Scond = arm.C_PBIT
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = arm.REG_R1
		p.To.Offset = sz
		p2 := gc.Prog(arm.ACMP)
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = v.Args[1].Reg()
		p2.Reg = arm.REG_R1
		p3 := gc.Prog(arm.ABLE)
		p3.To.Type = obj.TYPE_BRANCH
		gc.Patch(p3, p)
	case ssa.OpARMLoweredMove:
		// MOVW.P	4(R1), Rtmp
		// MOVW.P	Rtmp, 4(R2)
		// CMP	Rarg2, R1
		// BLE	-3(PC)
		// arg2 is the address of the last element of src
		// auxint is alignment
		var sz int64
		var mov obj.As
		switch {
		case v.AuxInt%4 == 0:
			sz = 4
			mov = arm.AMOVW
		case v.AuxInt%2 == 0:
			sz = 2
			mov = arm.AMOVH
		default:
			sz = 1
			mov = arm.AMOVB
		}
		p := gc.Prog(mov)
		p.Scond = arm.C_PBIT
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = arm.REG_R1
		p.From.Offset = sz
		p.To.Type = obj.TYPE_REG
		p.To.Reg = arm.REGTMP
		p2 := gc.Prog(mov)
		p2.Scond = arm.C_PBIT
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = arm.REGTMP
		p2.To.Type = obj.TYPE_MEM
		p2.To.Reg = arm.REG_R2
		p2.To.Offset = sz
		p3 := gc.Prog(arm.ACMP)
		p3.From.Type = obj.TYPE_REG
		p3.From.Reg = v.Args[2].Reg()
		p3.Reg = arm.REG_R1
		p4 := gc.Prog(arm.ABLE)
		p4.To.Type = obj.TYPE_BRANCH
		gc.Patch(p4, p)
	case ssa.OpVarDef:
		gc.Gvardef(v.Aux.(*gc.Node))
	case ssa.OpVarKill:
		gc.Gvarkill(v.Aux.(*gc.Node))
	case ssa.OpVarLive:
		gc.Gvarlive(v.Aux.(*gc.Node))
	case ssa.OpKeepAlive:
		gc.KeepAlive(v)
	case ssa.OpARMEqual,
		ssa.OpARMNotEqual,
		ssa.OpARMLessThan,
		ssa.OpARMLessEqual,
		ssa.OpARMGreaterThan,
		ssa.OpARMGreaterEqual,
		ssa.OpARMLessThanU,
		ssa.OpARMLessEqualU,
		ssa.OpARMGreaterThanU,
		ssa.OpARMGreaterEqualU:
		// generate boolean values
		// use conditional move
		p := gc.Prog(arm.AMOVW)
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = 0
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
		p = gc.Prog(arm.AMOVW)
		p.Scond = condBits[v.Op]
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = 1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpSelect0, ssa.OpSelect1:
		// nothing to do
	case ssa.OpARMLoweredGetClosurePtr:
		// Closure pointer is R7 (arm.REGCTXT).
		gc.CheckLoweredGetClosurePtr(v)
	case ssa.OpARMFlagEQ,
		ssa.OpARMFlagLT_ULT,
		ssa.OpARMFlagLT_UGT,
		ssa.OpARMFlagGT_ULT,
		ssa.OpARMFlagGT_UGT:
		v.Fatalf("Flag* ops should never make it to codegen %v", v.LongString())
	case ssa.OpARMInvertFlags:
		v.Fatalf("InvertFlags should never make it to codegen %v", v.LongString())
	default:
		v.Fatalf("genValue not implemented: %s", v.LongString())
	}
}
예제 #2
0
파일: ssa.go 프로젝트: Harvey-OS/go
func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
	s.SetLineno(v.Line)
	switch v.Op {
	case ssa.OpInitMem:
		// memory arg needs no code
	case ssa.OpArg:
		// input args need no code
	case ssa.OpSP, ssa.OpSB, ssa.OpGetG:
		// nothing to do
	case ssa.OpSelect0, ssa.OpSelect1:
		// nothing to do
	case ssa.OpCopy, ssa.OpMIPSMOVWconvert, ssa.OpMIPSMOVWreg:
		t := v.Type
		if t.IsMemory() {
			return
		}
		x := v.Args[0].Reg()
		y := v.Reg()
		if x == y {
			return
		}
		as := mips.AMOVW
		if isFPreg(x) && isFPreg(y) {
			as = mips.AMOVF
			if t.Size() == 8 {
				as = mips.AMOVD
			}
		}

		p := gc.Prog(as)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = x
		p.To.Type = obj.TYPE_REG
		p.To.Reg = y
		if isHILO(x) && isHILO(y) || isHILO(x) && isFPreg(y) || isFPreg(x) && isHILO(y) {
			// cannot move between special registers, use TMP as intermediate
			p.To.Reg = mips.REGTMP
			p = gc.Prog(mips.AMOVW)
			p.From.Type = obj.TYPE_REG
			p.From.Reg = mips.REGTMP
			p.To.Type = obj.TYPE_REG
			p.To.Reg = y
		}
	case ssa.OpMIPSMOVWnop:
		if v.Reg() != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		// nothing to do
	case ssa.OpLoadReg:
		if v.Type.IsFlags() {
			v.Fatalf("load flags not implemented: %v", v.LongString())
			return
		}
		r := v.Reg()
		p := gc.Prog(loadByType(v.Type, r))
		gc.AddrAuto(&p.From, v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
		if isHILO(r) {
			// cannot directly load, load to TMP and move
			p.To.Reg = mips.REGTMP
			p = gc.Prog(mips.AMOVW)
			p.From.Type = obj.TYPE_REG
			p.From.Reg = mips.REGTMP
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		}
	case ssa.OpStoreReg:
		if v.Type.IsFlags() {
			v.Fatalf("store flags not implemented: %v", v.LongString())
			return
		}
		r := v.Args[0].Reg()
		if isHILO(r) {
			// cannot directly store, move to TMP and store
			p := gc.Prog(mips.AMOVW)
			p.From.Type = obj.TYPE_REG
			p.From.Reg = r
			p.To.Type = obj.TYPE_REG
			p.To.Reg = mips.REGTMP
			r = mips.REGTMP
		}
		p := gc.Prog(storeByType(v.Type, r))
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r
		gc.AddrAuto(&p.To, v)
	case ssa.OpMIPSADD,
		ssa.OpMIPSSUB,
		ssa.OpMIPSAND,
		ssa.OpMIPSOR,
		ssa.OpMIPSXOR,
		ssa.OpMIPSNOR,
		ssa.OpMIPSSLL,
		ssa.OpMIPSSRL,
		ssa.OpMIPSSRA,
		ssa.OpMIPSADDF,
		ssa.OpMIPSADDD,
		ssa.OpMIPSSUBF,
		ssa.OpMIPSSUBD,
		ssa.OpMIPSMULF,
		ssa.OpMIPSMULD,
		ssa.OpMIPSDIVF,
		ssa.OpMIPSDIVD,
		ssa.OpMIPSMUL:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpMIPSSGT,
		ssa.OpMIPSSGTU:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpMIPSSGTzero,
		ssa.OpMIPSSGTUzero:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.Reg = mips.REGZERO
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpMIPSADDconst,
		ssa.OpMIPSSUBconst,
		ssa.OpMIPSANDconst,
		ssa.OpMIPSORconst,
		ssa.OpMIPSXORconst,
		ssa.OpMIPSNORconst,
		ssa.OpMIPSSLLconst,
		ssa.OpMIPSSRLconst,
		ssa.OpMIPSSRAconst,
		ssa.OpMIPSSGTconst,
		ssa.OpMIPSSGTUconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpMIPSMULT,
		ssa.OpMIPSMULTU,
		ssa.OpMIPSDIV,
		ssa.OpMIPSDIVU:
		// result in hi,lo
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.Reg = v.Args[0].Reg()
	case ssa.OpMIPSMOVWconst:
		r := v.Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
		if isFPreg(r) || isHILO(r) {
			// cannot move into FP or special registers, use TMP as intermediate
			p.To.Reg = mips.REGTMP
			p = gc.Prog(mips.AMOVW)
			p.From.Type = obj.TYPE_REG
			p.From.Reg = mips.REGTMP
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		}
	case ssa.OpMIPSMOVFconst,
		ssa.OpMIPSMOVDconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_FCONST
		p.From.Val = math.Float64frombits(uint64(v.AuxInt))
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpMIPSCMOVZ:
		if v.Reg() != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpMIPSCMOVZzero:
		if v.Reg() != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.Reg = mips.REGZERO
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpMIPSCMPEQF,
		ssa.OpMIPSCMPEQD,
		ssa.OpMIPSCMPGEF,
		ssa.OpMIPSCMPGED,
		ssa.OpMIPSCMPGTF,
		ssa.OpMIPSCMPGTD:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.Reg = v.Args[1].Reg()
	case ssa.OpMIPSMOVWaddr:
		p := gc.Prog(mips.AMOVW)
		p.From.Type = obj.TYPE_ADDR
		var wantreg string
		// MOVW $sym+off(base), R
		// the assembler expands it as the following:
		// - base is SP: add constant offset to SP (R29)
		//               when constant is large, tmp register (R23) may be used
		// - base is SB: load external address with relocation
		switch v.Aux.(type) {
		default:
			v.Fatalf("aux is of unknown type %T", v.Aux)
		case *ssa.ExternSymbol:
			wantreg = "SB"
			gc.AddAux(&p.From, v)
		case *ssa.ArgSymbol, *ssa.AutoSymbol:
			wantreg = "SP"
			gc.AddAux(&p.From, v)
		case nil:
			// No sym, just MOVW $off(SP), R
			wantreg = "SP"
			p.From.Reg = mips.REGSP
			p.From.Offset = v.AuxInt
		}
		if reg := v.Args[0].RegName(); reg != wantreg {
			v.Fatalf("bad reg %s for symbol type %T, want %s", reg, v.Aux, wantreg)
		}
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpMIPSMOVBload,
		ssa.OpMIPSMOVBUload,
		ssa.OpMIPSMOVHload,
		ssa.OpMIPSMOVHUload,
		ssa.OpMIPSMOVWload,
		ssa.OpMIPSMOVFload,
		ssa.OpMIPSMOVDload:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpMIPSMOVBstore,
		ssa.OpMIPSMOVHstore,
		ssa.OpMIPSMOVWstore,
		ssa.OpMIPSMOVFstore,
		ssa.OpMIPSMOVDstore:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpMIPSMOVBstorezero,
		ssa.OpMIPSMOVHstorezero,
		ssa.OpMIPSMOVWstorezero:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = mips.REGZERO
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpMIPSMOVBreg,
		ssa.OpMIPSMOVBUreg,
		ssa.OpMIPSMOVHreg,
		ssa.OpMIPSMOVHUreg:
		a := v.Args[0]
		for a.Op == ssa.OpCopy || a.Op == ssa.OpMIPSMOVWreg || a.Op == ssa.OpMIPSMOVWnop {
			a = a.Args[0]
		}
		if a.Op == ssa.OpLoadReg {
			t := a.Type
			switch {
			case v.Op == ssa.OpMIPSMOVBreg && t.Size() == 1 && t.IsSigned(),
				v.Op == ssa.OpMIPSMOVBUreg && t.Size() == 1 && !t.IsSigned(),
				v.Op == ssa.OpMIPSMOVHreg && t.Size() == 2 && t.IsSigned(),
				v.Op == ssa.OpMIPSMOVHUreg && t.Size() == 2 && !t.IsSigned():
				// arg is a proper-typed load, already zero/sign-extended, don't extend again
				if v.Reg() == v.Args[0].Reg() {
					return
				}
				p := gc.Prog(mips.AMOVW)
				p.From.Type = obj.TYPE_REG
				p.From.Reg = v.Args[0].Reg()
				p.To.Type = obj.TYPE_REG
				p.To.Reg = v.Reg()
				return
			default:
			}
		}
		fallthrough
	case ssa.OpMIPSMOVWF,
		ssa.OpMIPSMOVWD,
		ssa.OpMIPSTRUNCFW,
		ssa.OpMIPSTRUNCDW,
		ssa.OpMIPSMOVFD,
		ssa.OpMIPSMOVDF,
		ssa.OpMIPSNEGF,
		ssa.OpMIPSNEGD,
		ssa.OpMIPSSQRTD,
		ssa.OpMIPSCLZ:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpMIPSNEG:
		// SUB from REGZERO
		p := gc.Prog(mips.ASUBU)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.Reg = mips.REGZERO
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpMIPSLoweredZero:
		// SUBU	$4, R1
		// MOVW	R0, 4(R1)
		// ADDU	$4, R1
		// BNE	Rarg1, R1, -2(PC)
		// arg1 is the address of the last element to zero
		var sz int64
		var mov obj.As
		switch {
		case v.AuxInt%4 == 0:
			sz = 4
			mov = mips.AMOVW
		case v.AuxInt%2 == 0:
			sz = 2
			mov = mips.AMOVH
		default:
			sz = 1
			mov = mips.AMOVB
		}
		p := gc.Prog(mips.ASUBU)
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = sz
		p.To.Type = obj.TYPE_REG
		p.To.Reg = mips.REG_R1
		p2 := gc.Prog(mov)
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = mips.REGZERO
		p2.To.Type = obj.TYPE_MEM
		p2.To.Reg = mips.REG_R1
		p2.To.Offset = sz
		p3 := gc.Prog(mips.AADDU)
		p3.From.Type = obj.TYPE_CONST
		p3.From.Offset = sz
		p3.To.Type = obj.TYPE_REG
		p3.To.Reg = mips.REG_R1
		p4 := gc.Prog(mips.ABNE)
		p4.From.Type = obj.TYPE_REG
		p4.From.Reg = v.Args[1].Reg()
		p4.Reg = mips.REG_R1
		p4.To.Type = obj.TYPE_BRANCH
		gc.Patch(p4, p2)
	case ssa.OpMIPSLoweredMove:
		// SUBU	$4, R1
		// MOVW	4(R1), Rtmp
		// MOVW	Rtmp, (R2)
		// ADDU	$4, R1
		// ADDU	$4, R2
		// BNE	Rarg2, R1, -4(PC)
		// arg2 is the address of the last element of src
		var sz int64
		var mov obj.As
		switch {
		case v.AuxInt%4 == 0:
			sz = 4
			mov = mips.AMOVW
		case v.AuxInt%2 == 0:
			sz = 2
			mov = mips.AMOVH
		default:
			sz = 1
			mov = mips.AMOVB
		}
		p := gc.Prog(mips.ASUBU)
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = sz
		p.To.Type = obj.TYPE_REG
		p.To.Reg = mips.REG_R1
		p2 := gc.Prog(mov)
		p2.From.Type = obj.TYPE_MEM
		p2.From.Reg = mips.REG_R1
		p2.From.Offset = sz
		p2.To.Type = obj.TYPE_REG
		p2.To.Reg = mips.REGTMP
		p3 := gc.Prog(mov)
		p3.From.Type = obj.TYPE_REG
		p3.From.Reg = mips.REGTMP
		p3.To.Type = obj.TYPE_MEM
		p3.To.Reg = mips.REG_R2
		p4 := gc.Prog(mips.AADDU)
		p4.From.Type = obj.TYPE_CONST
		p4.From.Offset = sz
		p4.To.Type = obj.TYPE_REG
		p4.To.Reg = mips.REG_R1
		p5 := gc.Prog(mips.AADDU)
		p5.From.Type = obj.TYPE_CONST
		p5.From.Offset = sz
		p5.To.Type = obj.TYPE_REG
		p5.To.Reg = mips.REG_R2
		p6 := gc.Prog(mips.ABNE)
		p6.From.Type = obj.TYPE_REG
		p6.From.Reg = v.Args[2].Reg()
		p6.Reg = mips.REG_R1
		p6.To.Type = obj.TYPE_BRANCH
		gc.Patch(p6, p2)
	case ssa.OpMIPSCALLstatic:
		if v.Aux.(*gc.Sym) == gc.Deferreturn.Sym {
			// Deferred calls will appear to be returning to
			// the CALL deferreturn(SB) that we are about to emit.
			// However, the stack trace code will show the line
			// of the instruction byte before the return PC.
			// To avoid that being an unrelated instruction,
			// insert an actual hardware NOP that will have the right line number.
			// This is different from obj.ANOP, which is a virtual no-op
			// that doesn't make it into the instruction stream.
			ginsnop()
		}
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(v.Aux.(*gc.Sym))
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpMIPSCALLclosure:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Offset = 0
		p.To.Reg = v.Args[0].Reg()
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpMIPSCALLdefer:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Deferproc.Sym)
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpMIPSCALLgo:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Newproc.Sym)
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpMIPSCALLinter:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Offset = 0
		p.To.Reg = v.Args[0].Reg()
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpMIPSLoweredAtomicLoad:
		gc.Prog(mips.ASYNC)

		p := gc.Prog(mips.AMOVW)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg0()

		gc.Prog(mips.ASYNC)
	case ssa.OpMIPSLoweredAtomicStore:
		gc.Prog(mips.ASYNC)

		p := gc.Prog(mips.AMOVW)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()

		gc.Prog(mips.ASYNC)
	case ssa.OpMIPSLoweredAtomicStorezero:
		gc.Prog(mips.ASYNC)

		p := gc.Prog(mips.AMOVW)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = mips.REGZERO
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()

		gc.Prog(mips.ASYNC)
	case ssa.OpMIPSLoweredAtomicExchange:
		// SYNC
		// MOVW Rarg1, Rtmp
		// LL	(Rarg0), Rout
		// SC	Rtmp, (Rarg0)
		// BEQ	Rtmp, -3(PC)
		// SYNC
		gc.Prog(mips.ASYNC)

		p := gc.Prog(mips.AMOVW)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = mips.REGTMP

		p1 := gc.Prog(mips.ALL)
		p1.From.Type = obj.TYPE_MEM
		p1.From.Reg = v.Args[0].Reg()
		p1.To.Type = obj.TYPE_REG
		p1.To.Reg = v.Reg0()

		p2 := gc.Prog(mips.ASC)
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = mips.REGTMP
		p2.To.Type = obj.TYPE_MEM
		p2.To.Reg = v.Args[0].Reg()

		p3 := gc.Prog(mips.ABEQ)
		p3.From.Type = obj.TYPE_REG
		p3.From.Reg = mips.REGTMP
		p3.To.Type = obj.TYPE_BRANCH
		gc.Patch(p3, p)

		gc.Prog(mips.ASYNC)
	case ssa.OpMIPSLoweredAtomicAdd:
		// SYNC
		// LL	(Rarg0), Rout
		// ADDU Rarg1, Rout, Rtmp
		// SC	Rtmp, (Rarg0)
		// BEQ	Rtmp, -3(PC)
		// SYNC
		// ADDU Rarg1, Rout
		gc.Prog(mips.ASYNC)

		p := gc.Prog(mips.ALL)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg0()

		p1 := gc.Prog(mips.AADDU)
		p1.From.Type = obj.TYPE_REG
		p1.From.Reg = v.Args[1].Reg()
		p1.Reg = v.Reg0()
		p1.To.Type = obj.TYPE_REG
		p1.To.Reg = mips.REGTMP

		p2 := gc.Prog(mips.ASC)
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = mips.REGTMP
		p2.To.Type = obj.TYPE_MEM
		p2.To.Reg = v.Args[0].Reg()

		p3 := gc.Prog(mips.ABEQ)
		p3.From.Type = obj.TYPE_REG
		p3.From.Reg = mips.REGTMP
		p3.To.Type = obj.TYPE_BRANCH
		gc.Patch(p3, p)

		gc.Prog(mips.ASYNC)

		p4 := gc.Prog(mips.AADDU)
		p4.From.Type = obj.TYPE_REG
		p4.From.Reg = v.Args[1].Reg()
		p4.Reg = v.Reg0()
		p4.To.Type = obj.TYPE_REG
		p4.To.Reg = v.Reg0()

	case ssa.OpMIPSLoweredAtomicAddconst:
		// SYNC
		// LL	(Rarg0), Rout
		// ADDU $auxInt, Rout, Rtmp
		// SC	Rtmp, (Rarg0)
		// BEQ	Rtmp, -3(PC)
		// SYNC
		// ADDU $auxInt, Rout
		gc.Prog(mips.ASYNC)

		p := gc.Prog(mips.ALL)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg0()

		p1 := gc.Prog(mips.AADDU)
		p1.From.Type = obj.TYPE_CONST
		p1.From.Offset = v.AuxInt
		p1.Reg = v.Reg0()
		p1.To.Type = obj.TYPE_REG
		p1.To.Reg = mips.REGTMP

		p2 := gc.Prog(mips.ASC)
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = mips.REGTMP
		p2.To.Type = obj.TYPE_MEM
		p2.To.Reg = v.Args[0].Reg()

		p3 := gc.Prog(mips.ABEQ)
		p3.From.Type = obj.TYPE_REG
		p3.From.Reg = mips.REGTMP
		p3.To.Type = obj.TYPE_BRANCH
		gc.Patch(p3, p)

		gc.Prog(mips.ASYNC)

		p4 := gc.Prog(mips.AADDU)
		p4.From.Type = obj.TYPE_CONST
		p4.From.Offset = v.AuxInt
		p4.Reg = v.Reg0()
		p4.To.Type = obj.TYPE_REG
		p4.To.Reg = v.Reg0()

	case ssa.OpMIPSLoweredAtomicAnd,
		ssa.OpMIPSLoweredAtomicOr:
		// SYNC
		// LL	(Rarg0), Rtmp
		// AND/OR	Rarg1, Rtmp
		// SC	Rtmp, (Rarg0)
		// BEQ	Rtmp, -3(PC)
		// SYNC
		gc.Prog(mips.ASYNC)

		p := gc.Prog(mips.ALL)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = mips.REGTMP

		p1 := gc.Prog(v.Op.Asm())
		p1.From.Type = obj.TYPE_REG
		p1.From.Reg = v.Args[1].Reg()
		p1.Reg = mips.REGTMP
		p1.To.Type = obj.TYPE_REG
		p1.To.Reg = mips.REGTMP

		p2 := gc.Prog(mips.ASC)
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = mips.REGTMP
		p2.To.Type = obj.TYPE_MEM
		p2.To.Reg = v.Args[0].Reg()

		p3 := gc.Prog(mips.ABEQ)
		p3.From.Type = obj.TYPE_REG
		p3.From.Reg = mips.REGTMP
		p3.To.Type = obj.TYPE_BRANCH
		gc.Patch(p3, p)

		gc.Prog(mips.ASYNC)

	case ssa.OpMIPSLoweredAtomicCas:
		// MOVW $0, Rout
		// SYNC
		// LL	(Rarg0), Rtmp
		// BNE	Rtmp, Rarg1, 4(PC)
		// MOVW Rarg2, Rout
		// SC	Rout, (Rarg0)
		// BEQ	Rout, -4(PC)
		// SYNC
		p := gc.Prog(mips.AMOVW)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = mips.REGZERO
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg0()

		gc.Prog(mips.ASYNC)

		p1 := gc.Prog(mips.ALL)
		p1.From.Type = obj.TYPE_MEM
		p1.From.Reg = v.Args[0].Reg()
		p1.To.Type = obj.TYPE_REG
		p1.To.Reg = mips.REGTMP

		p2 := gc.Prog(mips.ABNE)
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = v.Args[1].Reg()
		p2.Reg = mips.REGTMP
		p2.To.Type = obj.TYPE_BRANCH

		p3 := gc.Prog(mips.AMOVW)
		p3.From.Type = obj.TYPE_REG
		p3.From.Reg = v.Args[2].Reg()
		p3.To.Type = obj.TYPE_REG
		p3.To.Reg = v.Reg0()

		p4 := gc.Prog(mips.ASC)
		p4.From.Type = obj.TYPE_REG
		p4.From.Reg = v.Reg0()
		p4.To.Type = obj.TYPE_MEM
		p4.To.Reg = v.Args[0].Reg()

		p5 := gc.Prog(mips.ABEQ)
		p5.From.Type = obj.TYPE_REG
		p5.From.Reg = v.Reg0()
		p5.To.Type = obj.TYPE_BRANCH
		gc.Patch(p5, p1)

		gc.Prog(mips.ASYNC)

		p6 := gc.Prog(obj.ANOP)
		gc.Patch(p2, p6)

	case ssa.OpVarDef:
		gc.Gvardef(v.Aux.(*gc.Node))
	case ssa.OpVarKill:
		gc.Gvarkill(v.Aux.(*gc.Node))
	case ssa.OpVarLive:
		gc.Gvarlive(v.Aux.(*gc.Node))
	case ssa.OpKeepAlive:
		gc.KeepAlive(v)
	case ssa.OpPhi:
		gc.CheckLoweredPhi(v)
	case ssa.OpMIPSLoweredNilCheck:
		// Issue a load which will fault if arg is nil.
		p := gc.Prog(mips.AMOVB)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = mips.REGTMP
		if gc.Debug_checknil != 0 && v.Line > 1 { // v.Line==1 in generated wrappers
			gc.Warnl(v.Line, "generated nil check")
		}
	case ssa.OpMIPSFPFlagTrue,
		ssa.OpMIPSFPFlagFalse:
		// MOVW		$1, r
		// CMOVF	R0, r

		cmov := mips.ACMOVF
		if v.Op == ssa.OpMIPSFPFlagFalse {
			cmov = mips.ACMOVT
		}
		p := gc.Prog(mips.AMOVW)
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = 1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
		p1 := gc.Prog(cmov)
		p1.From.Type = obj.TYPE_REG
		p1.From.Reg = mips.REGZERO
		p1.To.Type = obj.TYPE_REG
		p1.To.Reg = v.Reg()

	case ssa.OpMIPSLoweredGetClosurePtr:
		// Closure pointer is R22 (mips.REGCTXT).
		gc.CheckLoweredGetClosurePtr(v)
	default:
		v.Fatalf("genValue not implemented: %s", v.LongString())
	}
}
예제 #3
0
파일: ssa.go 프로젝트: achanda/go
func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
	s.SetLineno(v.Line)

	if gc.Thearch.Use387 {
		if ssaGenValue387(s, v) {
			return // v was handled by 387 generation.
		}
	}

	switch v.Op {
	case ssa.Op386ADDL:
		r := v.Reg()
		r1 := v.Args[0].Reg()
		r2 := v.Args[1].Reg()
		switch {
		case r == r1:
			p := gc.Prog(v.Op.Asm())
			p.From.Type = obj.TYPE_REG
			p.From.Reg = r2
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		case r == r2:
			p := gc.Prog(v.Op.Asm())
			p.From.Type = obj.TYPE_REG
			p.From.Reg = r1
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		default:
			p := gc.Prog(x86.ALEAL)
			p.From.Type = obj.TYPE_MEM
			p.From.Reg = r1
			p.From.Scale = 1
			p.From.Index = r2
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		}

	// 2-address opcode arithmetic
	case ssa.Op386SUBL,
		ssa.Op386MULL,
		ssa.Op386ANDL,
		ssa.Op386ORL,
		ssa.Op386XORL,
		ssa.Op386SHLL,
		ssa.Op386SHRL, ssa.Op386SHRW, ssa.Op386SHRB,
		ssa.Op386SARL, ssa.Op386SARW, ssa.Op386SARB,
		ssa.Op386ADDSS, ssa.Op386ADDSD, ssa.Op386SUBSS, ssa.Op386SUBSD,
		ssa.Op386MULSS, ssa.Op386MULSD, ssa.Op386DIVSS, ssa.Op386DIVSD,
		ssa.Op386PXOR,
		ssa.Op386ADCL,
		ssa.Op386SBBL:
		r := v.Reg()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		opregreg(v.Op.Asm(), r, v.Args[1].Reg())

	case ssa.Op386ADDLcarry, ssa.Op386SUBLcarry:
		// output 0 is carry/borrow, output 1 is the low 32 bits.
		r := v.Reg0()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output[0] not in same register %s", v.LongString())
		}
		opregreg(v.Op.Asm(), r, v.Args[1].Reg())

	case ssa.Op386ADDLconstcarry, ssa.Op386SUBLconstcarry:
		// output 0 is carry/borrow, output 1 is the low 32 bits.
		r := v.Reg0()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output[0] not in same register %s", v.LongString())
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r

	case ssa.Op386DIVL, ssa.Op386DIVW,
		ssa.Op386DIVLU, ssa.Op386DIVWU,
		ssa.Op386MODL, ssa.Op386MODW,
		ssa.Op386MODLU, ssa.Op386MODWU:

		// Arg[0] is already in AX as it's the only register we allow
		// and AX is the only output
		x := v.Args[1].Reg()

		// CPU faults upon signed overflow, which occurs when most
		// negative int is divided by -1.
		var j *obj.Prog
		if v.Op == ssa.Op386DIVL || v.Op == ssa.Op386DIVW ||
			v.Op == ssa.Op386MODL || v.Op == ssa.Op386MODW {

			var c *obj.Prog
			switch v.Op {
			case ssa.Op386DIVL, ssa.Op386MODL:
				c = gc.Prog(x86.ACMPL)
				j = gc.Prog(x86.AJEQ)
				gc.Prog(x86.ACDQ) //TODO: fix

			case ssa.Op386DIVW, ssa.Op386MODW:
				c = gc.Prog(x86.ACMPW)
				j = gc.Prog(x86.AJEQ)
				gc.Prog(x86.ACWD)
			}
			c.From.Type = obj.TYPE_REG
			c.From.Reg = x
			c.To.Type = obj.TYPE_CONST
			c.To.Offset = -1

			j.To.Type = obj.TYPE_BRANCH
		}

		// for unsigned ints, we sign extend by setting DX = 0
		// signed ints were sign extended above
		if v.Op == ssa.Op386DIVLU || v.Op == ssa.Op386MODLU ||
			v.Op == ssa.Op386DIVWU || v.Op == ssa.Op386MODWU {
			c := gc.Prog(x86.AXORL)
			c.From.Type = obj.TYPE_REG
			c.From.Reg = x86.REG_DX
			c.To.Type = obj.TYPE_REG
			c.To.Reg = x86.REG_DX
		}

		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = x

		// signed division, rest of the check for -1 case
		if j != nil {
			j2 := gc.Prog(obj.AJMP)
			j2.To.Type = obj.TYPE_BRANCH

			var n *obj.Prog
			if v.Op == ssa.Op386DIVL || v.Op == ssa.Op386DIVW {
				// n * -1 = -n
				n = gc.Prog(x86.ANEGL)
				n.To.Type = obj.TYPE_REG
				n.To.Reg = x86.REG_AX
			} else {
				// n % -1 == 0
				n = gc.Prog(x86.AXORL)
				n.From.Type = obj.TYPE_REG
				n.From.Reg = x86.REG_DX
				n.To.Type = obj.TYPE_REG
				n.To.Reg = x86.REG_DX
			}

			j.To.Val = n
			j2.To.Val = s.Pc()
		}

	case ssa.Op386HMULL, ssa.Op386HMULW, ssa.Op386HMULB,
		ssa.Op386HMULLU, ssa.Op386HMULWU, ssa.Op386HMULBU:
		// the frontend rewrites constant division by 8/16/32 bit integers into
		// HMUL by a constant
		// SSA rewrites generate the 64 bit versions

		// Arg[0] is already in AX as it's the only register we allow
		// and DX is the only output we care about (the high bits)
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()

		// IMULB puts the high portion in AH instead of DL,
		// so move it to DL for consistency
		if v.Type.Size() == 1 {
			m := gc.Prog(x86.AMOVB)
			m.From.Type = obj.TYPE_REG
			m.From.Reg = x86.REG_AH
			m.To.Type = obj.TYPE_REG
			m.To.Reg = x86.REG_DX
		}

	case ssa.Op386MULLQU:
		// AX * args[1], high 32 bits in DX (result[0]), low 32 bits in AX (result[1]).
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()

	case ssa.Op386ADDLconst:
		r := v.Reg()
		a := v.Args[0].Reg()
		if r == a {
			if v.AuxInt == 1 {
				p := gc.Prog(x86.AINCL)
				p.To.Type = obj.TYPE_REG
				p.To.Reg = r
				return
			}
			if v.AuxInt == -1 {
				p := gc.Prog(x86.ADECL)
				p.To.Type = obj.TYPE_REG
				p.To.Reg = r
				return
			}
			p := gc.Prog(v.Op.Asm())
			p.From.Type = obj.TYPE_CONST
			p.From.Offset = v.AuxInt
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
			return
		}
		p := gc.Prog(x86.ALEAL)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = a
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r

	case ssa.Op386MULLconst:
		r := v.Reg()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
		// TODO: Teach doasm to compile the three-address multiply imul $c, r1, r2
		// then we don't need to use resultInArg0 for these ops.
		//p.From3 = new(obj.Addr)
		//p.From3.Type = obj.TYPE_REG
		//p.From3.Reg = v.Args[0].Reg()

	case ssa.Op386SUBLconst,
		ssa.Op386ADCLconst,
		ssa.Op386SBBLconst,
		ssa.Op386ANDLconst,
		ssa.Op386ORLconst,
		ssa.Op386XORLconst,
		ssa.Op386SHLLconst,
		ssa.Op386SHRLconst, ssa.Op386SHRWconst, ssa.Op386SHRBconst,
		ssa.Op386SARLconst, ssa.Op386SARWconst, ssa.Op386SARBconst,
		ssa.Op386ROLLconst, ssa.Op386ROLWconst, ssa.Op386ROLBconst:
		r := v.Reg()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.Op386SBBLcarrymask:
		r := v.Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.Op386LEAL1, ssa.Op386LEAL2, ssa.Op386LEAL4, ssa.Op386LEAL8:
		r := v.Args[0].Reg()
		i := v.Args[1].Reg()
		p := gc.Prog(x86.ALEAL)
		switch v.Op {
		case ssa.Op386LEAL1:
			p.From.Scale = 1
			if i == x86.REG_SP {
				r, i = i, r
			}
		case ssa.Op386LEAL2:
			p.From.Scale = 2
		case ssa.Op386LEAL4:
			p.From.Scale = 4
		case ssa.Op386LEAL8:
			p.From.Scale = 8
		}
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = r
		p.From.Index = i
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.Op386LEAL:
		p := gc.Prog(x86.ALEAL)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.Op386CMPL, ssa.Op386CMPW, ssa.Op386CMPB,
		ssa.Op386TESTL, ssa.Op386TESTW, ssa.Op386TESTB:
		opregreg(v.Op.Asm(), v.Args[1].Reg(), v.Args[0].Reg())
	case ssa.Op386UCOMISS, ssa.Op386UCOMISD:
		// Go assembler has swapped operands for UCOMISx relative to CMP,
		// must account for that right here.
		opregreg(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg())
	case ssa.Op386CMPLconst, ssa.Op386CMPWconst, ssa.Op386CMPBconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_CONST
		p.To.Offset = v.AuxInt
	case ssa.Op386TESTLconst, ssa.Op386TESTWconst, ssa.Op386TESTBconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Args[0].Reg()
	case ssa.Op386MOVLconst:
		x := v.Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = x
		// If flags are live at this instruction, suppress the
		// MOV $0,AX -> XOR AX,AX optimization.
		if v.Aux != nil {
			p.Mark |= x86.PRESERVEFLAGS
		}
	case ssa.Op386MOVSSconst, ssa.Op386MOVSDconst:
		x := v.Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_FCONST
		p.From.Val = math.Float64frombits(uint64(v.AuxInt))
		p.To.Type = obj.TYPE_REG
		p.To.Reg = x
	case ssa.Op386MOVSSconst1, ssa.Op386MOVSDconst1:
		var literal string
		if v.Op == ssa.Op386MOVSDconst1 {
			literal = fmt.Sprintf("$f64.%016x", uint64(v.AuxInt))
		} else {
			literal = fmt.Sprintf("$f32.%08x", math.Float32bits(float32(math.Float64frombits(uint64(v.AuxInt)))))
		}
		p := gc.Prog(x86.ALEAL)
		p.From.Type = obj.TYPE_MEM
		p.From.Name = obj.NAME_EXTERN
		p.From.Sym = obj.Linklookup(gc.Ctxt, literal, 0)
		p.From.Sym.Set(obj.AttrLocal, true)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.Op386MOVSSconst2, ssa.Op386MOVSDconst2:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()

	case ssa.Op386MOVSSload, ssa.Op386MOVSDload, ssa.Op386MOVLload, ssa.Op386MOVWload, ssa.Op386MOVBload, ssa.Op386MOVBLSXload, ssa.Op386MOVWLSXload:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.Op386MOVSDloadidx8:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.From.Scale = 8
		p.From.Index = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.Op386MOVLloadidx4, ssa.Op386MOVSSloadidx4:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.From.Scale = 4
		p.From.Index = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.Op386MOVWloadidx2:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.From.Scale = 2
		p.From.Index = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.Op386MOVBloadidx1, ssa.Op386MOVWloadidx1, ssa.Op386MOVLloadidx1, ssa.Op386MOVSSloadidx1, ssa.Op386MOVSDloadidx1:
		r := v.Args[0].Reg()
		i := v.Args[1].Reg()
		if i == x86.REG_SP {
			r, i = i, r
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = r
		p.From.Scale = 1
		p.From.Index = i
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.Op386MOVSSstore, ssa.Op386MOVSDstore, ssa.Op386MOVLstore, ssa.Op386MOVWstore, ssa.Op386MOVBstore:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
	case ssa.Op386MOVSDstoreidx8:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		p.To.Scale = 8
		p.To.Index = v.Args[1].Reg()
		gc.AddAux(&p.To, v)
	case ssa.Op386MOVSSstoreidx4, ssa.Op386MOVLstoreidx4:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		p.To.Scale = 4
		p.To.Index = v.Args[1].Reg()
		gc.AddAux(&p.To, v)
	case ssa.Op386MOVWstoreidx2:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		p.To.Scale = 2
		p.To.Index = v.Args[1].Reg()
		gc.AddAux(&p.To, v)
	case ssa.Op386MOVBstoreidx1, ssa.Op386MOVWstoreidx1, ssa.Op386MOVLstoreidx1, ssa.Op386MOVSSstoreidx1, ssa.Op386MOVSDstoreidx1:
		r := v.Args[0].Reg()
		i := v.Args[1].Reg()
		if i == x86.REG_SP {
			r, i = i, r
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = r
		p.To.Scale = 1
		p.To.Index = i
		gc.AddAux(&p.To, v)
	case ssa.Op386MOVLstoreconst, ssa.Op386MOVWstoreconst, ssa.Op386MOVBstoreconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		sc := v.AuxValAndOff()
		p.From.Offset = sc.Val()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux2(&p.To, v, sc.Off())
	case ssa.Op386MOVLstoreconstidx1, ssa.Op386MOVLstoreconstidx4, ssa.Op386MOVWstoreconstidx1, ssa.Op386MOVWstoreconstidx2, ssa.Op386MOVBstoreconstidx1:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		sc := v.AuxValAndOff()
		p.From.Offset = sc.Val()
		r := v.Args[0].Reg()
		i := v.Args[1].Reg()
		switch v.Op {
		case ssa.Op386MOVBstoreconstidx1, ssa.Op386MOVWstoreconstidx1, ssa.Op386MOVLstoreconstidx1:
			p.To.Scale = 1
			if i == x86.REG_SP {
				r, i = i, r
			}
		case ssa.Op386MOVWstoreconstidx2:
			p.To.Scale = 2
		case ssa.Op386MOVLstoreconstidx4:
			p.To.Scale = 4
		}
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = r
		p.To.Index = i
		gc.AddAux2(&p.To, v, sc.Off())
	case ssa.Op386MOVWLSX, ssa.Op386MOVBLSX, ssa.Op386MOVWLZX, ssa.Op386MOVBLZX,
		ssa.Op386CVTSL2SS, ssa.Op386CVTSL2SD,
		ssa.Op386CVTTSS2SL, ssa.Op386CVTTSD2SL,
		ssa.Op386CVTSS2SD, ssa.Op386CVTSD2SS:
		opregreg(v.Op.Asm(), v.Reg(), v.Args[0].Reg())
	case ssa.Op386DUFFZERO:
		p := gc.Prog(obj.ADUFFZERO)
		p.To.Type = obj.TYPE_ADDR
		p.To.Sym = gc.Linksym(gc.Pkglookup("duffzero", gc.Runtimepkg))
		p.To.Offset = v.AuxInt
	case ssa.Op386DUFFCOPY:
		p := gc.Prog(obj.ADUFFCOPY)
		p.To.Type = obj.TYPE_ADDR
		p.To.Sym = gc.Linksym(gc.Pkglookup("duffcopy", gc.Runtimepkg))
		p.To.Offset = v.AuxInt

	case ssa.OpCopy, ssa.Op386MOVLconvert: // TODO: use MOVLreg for reg->reg copies instead of OpCopy?
		if v.Type.IsMemory() {
			return
		}
		x := v.Args[0].Reg()
		y := v.Reg()
		if x != y {
			opregreg(moveByType(v.Type), y, x)
		}
	case ssa.OpLoadReg:
		if v.Type.IsFlags() {
			v.Fatalf("load flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(loadByType(v.Type))
		gc.AddrAuto(&p.From, v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()

	case ssa.OpStoreReg:
		if v.Type.IsFlags() {
			v.Fatalf("store flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(storeByType(v.Type))
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		gc.AddrAuto(&p.To, v)
	case ssa.OpPhi:
		gc.CheckLoweredPhi(v)
	case ssa.OpInitMem:
		// memory arg needs no code
	case ssa.OpArg:
		// input args need no code
	case ssa.Op386LoweredGetClosurePtr:
		// Closure pointer is DX.
		gc.CheckLoweredGetClosurePtr(v)
	case ssa.Op386LoweredGetG:
		r := v.Reg()
		// See the comments in cmd/internal/obj/x86/obj6.go
		// near CanUse1InsnTLS for a detailed explanation of these instructions.
		if x86.CanUse1InsnTLS(gc.Ctxt) {
			// MOVL (TLS), r
			p := gc.Prog(x86.AMOVL)
			p.From.Type = obj.TYPE_MEM
			p.From.Reg = x86.REG_TLS
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		} else {
			// MOVL TLS, r
			// MOVL (r)(TLS*1), r
			p := gc.Prog(x86.AMOVL)
			p.From.Type = obj.TYPE_REG
			p.From.Reg = x86.REG_TLS
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
			q := gc.Prog(x86.AMOVL)
			q.From.Type = obj.TYPE_MEM
			q.From.Reg = r
			q.From.Index = x86.REG_TLS
			q.From.Scale = 1
			q.To.Type = obj.TYPE_REG
			q.To.Reg = r
		}
	case ssa.Op386CALLstatic:
		if v.Aux.(*gc.Sym) == gc.Deferreturn.Sym {
			// Deferred calls will appear to be returning to
			// the CALL deferreturn(SB) that we are about to emit.
			// However, the stack trace code will show the line
			// of the instruction byte before the return PC.
			// To avoid that being an unrelated instruction,
			// insert an actual hardware NOP that will have the right line number.
			// This is different from obj.ANOP, which is a virtual no-op
			// that doesn't make it into the instruction stream.
			ginsnop()
		}
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(v.Aux.(*gc.Sym))
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.Op386CALLclosure:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Args[0].Reg()
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.Op386CALLdefer:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Deferproc.Sym)
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.Op386CALLgo:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Newproc.Sym)
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.Op386CALLinter:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Args[0].Reg()
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.Op386NEGL,
		ssa.Op386BSWAPL,
		ssa.Op386NOTL:
		r := v.Reg()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.Op386BSFL, ssa.Op386BSFW,
		ssa.Op386BSRL, ssa.Op386BSRW,
		ssa.Op386SQRTSD:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpSP, ssa.OpSB, ssa.OpSelect0, ssa.OpSelect1:
		// nothing to do
	case ssa.Op386SETEQ, ssa.Op386SETNE,
		ssa.Op386SETL, ssa.Op386SETLE,
		ssa.Op386SETG, ssa.Op386SETGE,
		ssa.Op386SETGF, ssa.Op386SETGEF,
		ssa.Op386SETB, ssa.Op386SETBE,
		ssa.Op386SETORD, ssa.Op386SETNAN,
		ssa.Op386SETA, ssa.Op386SETAE:
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()

	case ssa.Op386SETNEF:
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
		q := gc.Prog(x86.ASETPS)
		q.To.Type = obj.TYPE_REG
		q.To.Reg = x86.REG_AX
		opregreg(x86.AORL, v.Reg(), x86.REG_AX)

	case ssa.Op386SETEQF:
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
		q := gc.Prog(x86.ASETPC)
		q.To.Type = obj.TYPE_REG
		q.To.Reg = x86.REG_AX
		opregreg(x86.AANDL, v.Reg(), x86.REG_AX)

	case ssa.Op386InvertFlags:
		v.Fatalf("InvertFlags should never make it to codegen %v", v.LongString())
	case ssa.Op386FlagEQ, ssa.Op386FlagLT_ULT, ssa.Op386FlagLT_UGT, ssa.Op386FlagGT_ULT, ssa.Op386FlagGT_UGT:
		v.Fatalf("Flag* ops should never make it to codegen %v", v.LongString())
	case ssa.Op386REPSTOSL:
		gc.Prog(x86.AREP)
		gc.Prog(x86.ASTOSL)
	case ssa.Op386REPMOVSL:
		gc.Prog(x86.AREP)
		gc.Prog(x86.AMOVSL)
	case ssa.OpVarDef:
		gc.Gvardef(v.Aux.(*gc.Node))
	case ssa.OpVarKill:
		gc.Gvarkill(v.Aux.(*gc.Node))
	case ssa.OpVarLive:
		gc.Gvarlive(v.Aux.(*gc.Node))
	case ssa.OpKeepAlive:
		gc.KeepAlive(v)
	case ssa.Op386LoweredNilCheck:
		// Issue a load which will fault if the input is nil.
		// TODO: We currently use the 2-byte instruction TESTB AX, (reg).
		// Should we use the 3-byte TESTB $0, (reg) instead?  It is larger
		// but it doesn't have false dependency on AX.
		// Or maybe allocate an output register and use MOVL (reg),reg2 ?
		// That trades clobbering flags for clobbering a register.
		p := gc.Prog(x86.ATESTB)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = x86.REG_AX
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
		if gc.Debug_checknil != 0 && v.Line > 1 { // v.Line==1 in generated wrappers
			gc.Warnl(v.Line, "generated nil check")
		}
	case ssa.Op386FCHS:
		v.Fatalf("FCHS in non-387 mode")
	default:
		v.Fatalf("genValue not implemented: %s", v.LongString())
	}
}
예제 #4
0
파일: ssa.go 프로젝트: achanda/go
func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
	s.SetLineno(v.Line)
	switch v.Op {
	case ssa.OpAMD64ADDQ, ssa.OpAMD64ADDL:
		r := v.Reg()
		r1 := v.Args[0].Reg()
		r2 := v.Args[1].Reg()
		switch {
		case r == r1:
			p := gc.Prog(v.Op.Asm())
			p.From.Type = obj.TYPE_REG
			p.From.Reg = r2
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		case r == r2:
			p := gc.Prog(v.Op.Asm())
			p.From.Type = obj.TYPE_REG
			p.From.Reg = r1
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		default:
			var asm obj.As
			if v.Op == ssa.OpAMD64ADDQ {
				asm = x86.ALEAQ
			} else {
				asm = x86.ALEAL
			}
			p := gc.Prog(asm)
			p.From.Type = obj.TYPE_MEM
			p.From.Reg = r1
			p.From.Scale = 1
			p.From.Index = r2
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		}
	// 2-address opcode arithmetic
	case ssa.OpAMD64SUBQ, ssa.OpAMD64SUBL,
		ssa.OpAMD64MULQ, ssa.OpAMD64MULL,
		ssa.OpAMD64ANDQ, ssa.OpAMD64ANDL,
		ssa.OpAMD64ORQ, ssa.OpAMD64ORL,
		ssa.OpAMD64XORQ, ssa.OpAMD64XORL,
		ssa.OpAMD64SHLQ, ssa.OpAMD64SHLL,
		ssa.OpAMD64SHRQ, ssa.OpAMD64SHRL, ssa.OpAMD64SHRW, ssa.OpAMD64SHRB,
		ssa.OpAMD64SARQ, ssa.OpAMD64SARL, ssa.OpAMD64SARW, ssa.OpAMD64SARB,
		ssa.OpAMD64ADDSS, ssa.OpAMD64ADDSD, ssa.OpAMD64SUBSS, ssa.OpAMD64SUBSD,
		ssa.OpAMD64MULSS, ssa.OpAMD64MULSD, ssa.OpAMD64DIVSS, ssa.OpAMD64DIVSD,
		ssa.OpAMD64PXOR:
		r := v.Reg()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		opregreg(v.Op.Asm(), r, v.Args[1].Reg())

	case ssa.OpAMD64DIVQU, ssa.OpAMD64DIVLU, ssa.OpAMD64DIVWU:
		// Arg[0] (the dividend) is in AX.
		// Arg[1] (the divisor) can be in any other register.
		// Result[0] (the quotient) is in AX.
		// Result[1] (the remainder) is in DX.
		r := v.Args[1].Reg()

		// Zero extend dividend.
		c := gc.Prog(x86.AXORL)
		c.From.Type = obj.TYPE_REG
		c.From.Reg = x86.REG_DX
		c.To.Type = obj.TYPE_REG
		c.To.Reg = x86.REG_DX

		// Issue divide.
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r

	case ssa.OpAMD64DIVQ, ssa.OpAMD64DIVL, ssa.OpAMD64DIVW:
		// Arg[0] (the dividend) is in AX.
		// Arg[1] (the divisor) can be in any other register.
		// Result[0] (the quotient) is in AX.
		// Result[1] (the remainder) is in DX.
		r := v.Args[1].Reg()

		// CPU faults upon signed overflow, which occurs when the most
		// negative int is divided by -1. Handle divide by -1 as a special case.
		var c *obj.Prog
		switch v.Op {
		case ssa.OpAMD64DIVQ:
			c = gc.Prog(x86.ACMPQ)
		case ssa.OpAMD64DIVL:
			c = gc.Prog(x86.ACMPL)
		case ssa.OpAMD64DIVW:
			c = gc.Prog(x86.ACMPW)
		}
		c.From.Type = obj.TYPE_REG
		c.From.Reg = r
		c.To.Type = obj.TYPE_CONST
		c.To.Offset = -1
		j1 := gc.Prog(x86.AJEQ)
		j1.To.Type = obj.TYPE_BRANCH

		// Sign extend dividend.
		switch v.Op {
		case ssa.OpAMD64DIVQ:
			gc.Prog(x86.ACQO)
		case ssa.OpAMD64DIVL:
			gc.Prog(x86.ACDQ)
		case ssa.OpAMD64DIVW:
			gc.Prog(x86.ACWD)
		}

		// Issue divide.
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r

		// Skip over -1 fixup code.
		j2 := gc.Prog(obj.AJMP)
		j2.To.Type = obj.TYPE_BRANCH

		// Issue -1 fixup code.
		// n / -1 = -n
		n1 := gc.Prog(x86.ANEGQ)
		n1.To.Type = obj.TYPE_REG
		n1.To.Reg = x86.REG_AX

		// n % -1 == 0
		n2 := gc.Prog(x86.AXORL)
		n2.From.Type = obj.TYPE_REG
		n2.From.Reg = x86.REG_DX
		n2.To.Type = obj.TYPE_REG
		n2.To.Reg = x86.REG_DX

		// TODO(khr): issue only the -1 fixup code we need.
		// For instance, if only the quotient is used, no point in zeroing the remainder.

		j1.To.Val = n1
		j2.To.Val = s.Pc()

	case ssa.OpAMD64HMULQ, ssa.OpAMD64HMULL, ssa.OpAMD64HMULW, ssa.OpAMD64HMULB,
		ssa.OpAMD64HMULQU, ssa.OpAMD64HMULLU, ssa.OpAMD64HMULWU, ssa.OpAMD64HMULBU:
		// the frontend rewrites constant division by 8/16/32 bit integers into
		// HMUL by a constant
		// SSA rewrites generate the 64 bit versions

		// Arg[0] is already in AX as it's the only register we allow
		// and DX is the only output we care about (the high bits)
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()

		// IMULB puts the high portion in AH instead of DL,
		// so move it to DL for consistency
		if v.Type.Size() == 1 {
			m := gc.Prog(x86.AMOVB)
			m.From.Type = obj.TYPE_REG
			m.From.Reg = x86.REG_AH
			m.To.Type = obj.TYPE_REG
			m.To.Reg = x86.REG_DX
		}

	case ssa.OpAMD64MULQU2:
		// Arg[0] is already in AX as it's the only register we allow
		// results hi in DX, lo in AX
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()

	case ssa.OpAMD64DIVQU2:
		// Arg[0], Arg[1] are already in Dx, AX, as they're the only registers we allow
		// results q in AX, r in DX
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()

	case ssa.OpAMD64AVGQU:
		// compute (x+y)/2 unsigned.
		// Do a 64-bit add, the overflow goes into the carry.
		// Shift right once and pull the carry back into the 63rd bit.
		r := v.Reg()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		p := gc.Prog(x86.AADDQ)
		p.From.Type = obj.TYPE_REG
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
		p.From.Reg = v.Args[1].Reg()
		p = gc.Prog(x86.ARCRQ)
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = 1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r

	case ssa.OpAMD64ADDQconst, ssa.OpAMD64ADDLconst:
		r := v.Reg()
		a := v.Args[0].Reg()
		if r == a {
			if v.AuxInt == 1 {
				var asm obj.As
				// Software optimization manual recommends add $1,reg.
				// But inc/dec is 1 byte smaller. ICC always uses inc
				// Clang/GCC choose depending on flags, but prefer add.
				// Experiments show that inc/dec is both a little faster
				// and make a binary a little smaller.
				if v.Op == ssa.OpAMD64ADDQconst {
					asm = x86.AINCQ
				} else {
					asm = x86.AINCL
				}
				p := gc.Prog(asm)
				p.To.Type = obj.TYPE_REG
				p.To.Reg = r
				return
			}
			if v.AuxInt == -1 {
				var asm obj.As
				if v.Op == ssa.OpAMD64ADDQconst {
					asm = x86.ADECQ
				} else {
					asm = x86.ADECL
				}
				p := gc.Prog(asm)
				p.To.Type = obj.TYPE_REG
				p.To.Reg = r
				return
			}
			p := gc.Prog(v.Op.Asm())
			p.From.Type = obj.TYPE_CONST
			p.From.Offset = v.AuxInt
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
			return
		}
		var asm obj.As
		if v.Op == ssa.OpAMD64ADDQconst {
			asm = x86.ALEAQ
		} else {
			asm = x86.ALEAL
		}
		p := gc.Prog(asm)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = a
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r

	case ssa.OpAMD64CMOVQEQ, ssa.OpAMD64CMOVLEQ:
		r := v.Reg()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r

	case ssa.OpAMD64MULQconst, ssa.OpAMD64MULLconst:
		r := v.Reg()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
		// TODO: Teach doasm to compile the three-address multiply imul $c, r1, r2
		// then we don't need to use resultInArg0 for these ops.
		//p.From3 = new(obj.Addr)
		//p.From3.Type = obj.TYPE_REG
		//p.From3.Reg = v.Args[0].Reg()

	case ssa.OpAMD64SUBQconst, ssa.OpAMD64SUBLconst,
		ssa.OpAMD64ANDQconst, ssa.OpAMD64ANDLconst,
		ssa.OpAMD64ORQconst, ssa.OpAMD64ORLconst,
		ssa.OpAMD64XORQconst, ssa.OpAMD64XORLconst,
		ssa.OpAMD64SHLQconst, ssa.OpAMD64SHLLconst,
		ssa.OpAMD64SHRQconst, ssa.OpAMD64SHRLconst, ssa.OpAMD64SHRWconst, ssa.OpAMD64SHRBconst,
		ssa.OpAMD64SARQconst, ssa.OpAMD64SARLconst, ssa.OpAMD64SARWconst, ssa.OpAMD64SARBconst,
		ssa.OpAMD64ROLQconst, ssa.OpAMD64ROLLconst, ssa.OpAMD64ROLWconst, ssa.OpAMD64ROLBconst:
		r := v.Reg()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpAMD64SBBQcarrymask, ssa.OpAMD64SBBLcarrymask:
		r := v.Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpAMD64LEAQ1, ssa.OpAMD64LEAQ2, ssa.OpAMD64LEAQ4, ssa.OpAMD64LEAQ8:
		r := v.Args[0].Reg()
		i := v.Args[1].Reg()
		p := gc.Prog(x86.ALEAQ)
		switch v.Op {
		case ssa.OpAMD64LEAQ1:
			p.From.Scale = 1
			if i == x86.REG_SP {
				r, i = i, r
			}
		case ssa.OpAMD64LEAQ2:
			p.From.Scale = 2
		case ssa.OpAMD64LEAQ4:
			p.From.Scale = 4
		case ssa.OpAMD64LEAQ8:
			p.From.Scale = 8
		}
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = r
		p.From.Index = i
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpAMD64LEAQ, ssa.OpAMD64LEAL:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpAMD64CMPQ, ssa.OpAMD64CMPL, ssa.OpAMD64CMPW, ssa.OpAMD64CMPB,
		ssa.OpAMD64TESTQ, ssa.OpAMD64TESTL, ssa.OpAMD64TESTW, ssa.OpAMD64TESTB:
		opregreg(v.Op.Asm(), v.Args[1].Reg(), v.Args[0].Reg())
	case ssa.OpAMD64UCOMISS, ssa.OpAMD64UCOMISD:
		// Go assembler has swapped operands for UCOMISx relative to CMP,
		// must account for that right here.
		opregreg(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg())
	case ssa.OpAMD64CMPQconst, ssa.OpAMD64CMPLconst, ssa.OpAMD64CMPWconst, ssa.OpAMD64CMPBconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_CONST
		p.To.Offset = v.AuxInt
	case ssa.OpAMD64TESTQconst, ssa.OpAMD64TESTLconst, ssa.OpAMD64TESTWconst, ssa.OpAMD64TESTBconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Args[0].Reg()
	case ssa.OpAMD64MOVLconst, ssa.OpAMD64MOVQconst:
		x := v.Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = x
		// If flags are live at this instruction, suppress the
		// MOV $0,AX -> XOR AX,AX optimization.
		if v.Aux != nil {
			p.Mark |= x86.PRESERVEFLAGS
		}
	case ssa.OpAMD64MOVSSconst, ssa.OpAMD64MOVSDconst:
		x := v.Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_FCONST
		p.From.Val = math.Float64frombits(uint64(v.AuxInt))
		p.To.Type = obj.TYPE_REG
		p.To.Reg = x
	case ssa.OpAMD64MOVQload, ssa.OpAMD64MOVSSload, ssa.OpAMD64MOVSDload, ssa.OpAMD64MOVLload, ssa.OpAMD64MOVWload, ssa.OpAMD64MOVBload, ssa.OpAMD64MOVBQSXload, ssa.OpAMD64MOVWQSXload, ssa.OpAMD64MOVLQSXload, ssa.OpAMD64MOVOload:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpAMD64MOVQloadidx8, ssa.OpAMD64MOVSDloadidx8:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.From.Scale = 8
		p.From.Index = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpAMD64MOVLloadidx4, ssa.OpAMD64MOVSSloadidx4:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.From.Scale = 4
		p.From.Index = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpAMD64MOVWloadidx2:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.From.Scale = 2
		p.From.Index = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpAMD64MOVBloadidx1, ssa.OpAMD64MOVWloadidx1, ssa.OpAMD64MOVLloadidx1, ssa.OpAMD64MOVQloadidx1, ssa.OpAMD64MOVSSloadidx1, ssa.OpAMD64MOVSDloadidx1:
		r := v.Args[0].Reg()
		i := v.Args[1].Reg()
		if i == x86.REG_SP {
			r, i = i, r
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = r
		p.From.Scale = 1
		p.From.Index = i
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpAMD64MOVQstore, ssa.OpAMD64MOVSSstore, ssa.OpAMD64MOVSDstore, ssa.OpAMD64MOVLstore, ssa.OpAMD64MOVWstore, ssa.OpAMD64MOVBstore, ssa.OpAMD64MOVOstore:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpAMD64MOVQstoreidx8, ssa.OpAMD64MOVSDstoreidx8:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		p.To.Scale = 8
		p.To.Index = v.Args[1].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpAMD64MOVSSstoreidx4, ssa.OpAMD64MOVLstoreidx4:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		p.To.Scale = 4
		p.To.Index = v.Args[1].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpAMD64MOVWstoreidx2:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		p.To.Scale = 2
		p.To.Index = v.Args[1].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpAMD64MOVBstoreidx1, ssa.OpAMD64MOVWstoreidx1, ssa.OpAMD64MOVLstoreidx1, ssa.OpAMD64MOVQstoreidx1, ssa.OpAMD64MOVSSstoreidx1, ssa.OpAMD64MOVSDstoreidx1:
		r := v.Args[0].Reg()
		i := v.Args[1].Reg()
		if i == x86.REG_SP {
			r, i = i, r
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = r
		p.To.Scale = 1
		p.To.Index = i
		gc.AddAux(&p.To, v)
	case ssa.OpAMD64MOVQstoreconst, ssa.OpAMD64MOVLstoreconst, ssa.OpAMD64MOVWstoreconst, ssa.OpAMD64MOVBstoreconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		sc := v.AuxValAndOff()
		p.From.Offset = sc.Val()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux2(&p.To, v, sc.Off())
	case ssa.OpAMD64MOVQstoreconstidx1, ssa.OpAMD64MOVQstoreconstidx8, ssa.OpAMD64MOVLstoreconstidx1, ssa.OpAMD64MOVLstoreconstidx4, ssa.OpAMD64MOVWstoreconstidx1, ssa.OpAMD64MOVWstoreconstidx2, ssa.OpAMD64MOVBstoreconstidx1:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		sc := v.AuxValAndOff()
		p.From.Offset = sc.Val()
		r := v.Args[0].Reg()
		i := v.Args[1].Reg()
		switch v.Op {
		case ssa.OpAMD64MOVBstoreconstidx1, ssa.OpAMD64MOVWstoreconstidx1, ssa.OpAMD64MOVLstoreconstidx1, ssa.OpAMD64MOVQstoreconstidx1:
			p.To.Scale = 1
			if i == x86.REG_SP {
				r, i = i, r
			}
		case ssa.OpAMD64MOVWstoreconstidx2:
			p.To.Scale = 2
		case ssa.OpAMD64MOVLstoreconstidx4:
			p.To.Scale = 4
		case ssa.OpAMD64MOVQstoreconstidx8:
			p.To.Scale = 8
		}
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = r
		p.To.Index = i
		gc.AddAux2(&p.To, v, sc.Off())
	case ssa.OpAMD64MOVLQSX, ssa.OpAMD64MOVWQSX, ssa.OpAMD64MOVBQSX, ssa.OpAMD64MOVLQZX, ssa.OpAMD64MOVWQZX, ssa.OpAMD64MOVBQZX,
		ssa.OpAMD64CVTTSS2SL, ssa.OpAMD64CVTTSD2SL, ssa.OpAMD64CVTTSS2SQ, ssa.OpAMD64CVTTSD2SQ,
		ssa.OpAMD64CVTSS2SD, ssa.OpAMD64CVTSD2SS:
		opregreg(v.Op.Asm(), v.Reg(), v.Args[0].Reg())
	case ssa.OpAMD64CVTSL2SD, ssa.OpAMD64CVTSQ2SD, ssa.OpAMD64CVTSQ2SS, ssa.OpAMD64CVTSL2SS:
		r := v.Reg()
		// Break false dependency on destination register.
		opregreg(x86.AXORPS, r, r)
		opregreg(v.Op.Asm(), r, v.Args[0].Reg())
	case ssa.OpAMD64DUFFZERO:
		off := duffStart(v.AuxInt)
		adj := duffAdj(v.AuxInt)
		var p *obj.Prog
		if adj != 0 {
			p = gc.Prog(x86.AADDQ)
			p.From.Type = obj.TYPE_CONST
			p.From.Offset = adj
			p.To.Type = obj.TYPE_REG
			p.To.Reg = x86.REG_DI
		}
		p = gc.Prog(obj.ADUFFZERO)
		p.To.Type = obj.TYPE_ADDR
		p.To.Sym = gc.Linksym(gc.Pkglookup("duffzero", gc.Runtimepkg))
		p.To.Offset = off
	case ssa.OpAMD64MOVOconst:
		if v.AuxInt != 0 {
			v.Fatalf("MOVOconst can only do constant=0")
		}
		r := v.Reg()
		opregreg(x86.AXORPS, r, r)
	case ssa.OpAMD64DUFFCOPY:
		p := gc.Prog(obj.ADUFFCOPY)
		p.To.Type = obj.TYPE_ADDR
		p.To.Sym = gc.Linksym(gc.Pkglookup("duffcopy", gc.Runtimepkg))
		p.To.Offset = v.AuxInt

	case ssa.OpCopy, ssa.OpAMD64MOVQconvert, ssa.OpAMD64MOVLconvert: // TODO: use MOVQreg for reg->reg copies instead of OpCopy?
		if v.Type.IsMemory() {
			return
		}
		x := v.Args[0].Reg()
		y := v.Reg()
		if x != y {
			opregreg(moveByType(v.Type), y, x)
		}
	case ssa.OpLoadReg:
		if v.Type.IsFlags() {
			v.Fatalf("load flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(loadByType(v.Type))
		gc.AddrAuto(&p.From, v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()

	case ssa.OpStoreReg:
		if v.Type.IsFlags() {
			v.Fatalf("store flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(storeByType(v.Type))
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		gc.AddrAuto(&p.To, v)
	case ssa.OpPhi:
		gc.CheckLoweredPhi(v)
	case ssa.OpInitMem:
		// memory arg needs no code
	case ssa.OpArg:
		// input args need no code
	case ssa.OpAMD64LoweredGetClosurePtr:
		// Closure pointer is DX.
		gc.CheckLoweredGetClosurePtr(v)
	case ssa.OpAMD64LoweredGetG:
		r := v.Reg()
		// See the comments in cmd/internal/obj/x86/obj6.go
		// near CanUse1InsnTLS for a detailed explanation of these instructions.
		if x86.CanUse1InsnTLS(gc.Ctxt) {
			// MOVQ (TLS), r
			p := gc.Prog(x86.AMOVQ)
			p.From.Type = obj.TYPE_MEM
			p.From.Reg = x86.REG_TLS
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		} else {
			// MOVQ TLS, r
			// MOVQ (r)(TLS*1), r
			p := gc.Prog(x86.AMOVQ)
			p.From.Type = obj.TYPE_REG
			p.From.Reg = x86.REG_TLS
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
			q := gc.Prog(x86.AMOVQ)
			q.From.Type = obj.TYPE_MEM
			q.From.Reg = r
			q.From.Index = x86.REG_TLS
			q.From.Scale = 1
			q.To.Type = obj.TYPE_REG
			q.To.Reg = r
		}
	case ssa.OpAMD64CALLstatic:
		if v.Aux.(*gc.Sym) == gc.Deferreturn.Sym {
			// Deferred calls will appear to be returning to
			// the CALL deferreturn(SB) that we are about to emit.
			// However, the stack trace code will show the line
			// of the instruction byte before the return PC.
			// To avoid that being an unrelated instruction,
			// insert an actual hardware NOP that will have the right line number.
			// This is different from obj.ANOP, which is a virtual no-op
			// that doesn't make it into the instruction stream.
			ginsnop()
		}
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(v.Aux.(*gc.Sym))
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpAMD64CALLclosure:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Args[0].Reg()
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpAMD64CALLdefer:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Deferproc.Sym)
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpAMD64CALLgo:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Newproc.Sym)
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpAMD64CALLinter:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Args[0].Reg()
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpAMD64NEGQ, ssa.OpAMD64NEGL,
		ssa.OpAMD64BSWAPQ, ssa.OpAMD64BSWAPL,
		ssa.OpAMD64NOTQ, ssa.OpAMD64NOTL:
		r := v.Reg()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpAMD64BSFQ, ssa.OpAMD64BSFL:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg0()
	case ssa.OpAMD64SQRTSD:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpSP, ssa.OpSB:
		// nothing to do
	case ssa.OpSelect0, ssa.OpSelect1:
		// nothing to do
	case ssa.OpAMD64SETEQ, ssa.OpAMD64SETNE,
		ssa.OpAMD64SETL, ssa.OpAMD64SETLE,
		ssa.OpAMD64SETG, ssa.OpAMD64SETGE,
		ssa.OpAMD64SETGF, ssa.OpAMD64SETGEF,
		ssa.OpAMD64SETB, ssa.OpAMD64SETBE,
		ssa.OpAMD64SETORD, ssa.OpAMD64SETNAN,
		ssa.OpAMD64SETA, ssa.OpAMD64SETAE:
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()

	case ssa.OpAMD64SETNEF:
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
		q := gc.Prog(x86.ASETPS)
		q.To.Type = obj.TYPE_REG
		q.To.Reg = x86.REG_AX
		// ORL avoids partial register write and is smaller than ORQ, used by old compiler
		opregreg(x86.AORL, v.Reg(), x86.REG_AX)

	case ssa.OpAMD64SETEQF:
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
		q := gc.Prog(x86.ASETPC)
		q.To.Type = obj.TYPE_REG
		q.To.Reg = x86.REG_AX
		// ANDL avoids partial register write and is smaller than ANDQ, used by old compiler
		opregreg(x86.AANDL, v.Reg(), x86.REG_AX)

	case ssa.OpAMD64InvertFlags:
		v.Fatalf("InvertFlags should never make it to codegen %v", v.LongString())
	case ssa.OpAMD64FlagEQ, ssa.OpAMD64FlagLT_ULT, ssa.OpAMD64FlagLT_UGT, ssa.OpAMD64FlagGT_ULT, ssa.OpAMD64FlagGT_UGT:
		v.Fatalf("Flag* ops should never make it to codegen %v", v.LongString())
	case ssa.OpAMD64AddTupleFirst32, ssa.OpAMD64AddTupleFirst64:
		v.Fatalf("AddTupleFirst* should never make it to codegen %v", v.LongString())
	case ssa.OpAMD64REPSTOSQ:
		gc.Prog(x86.AREP)
		gc.Prog(x86.ASTOSQ)
	case ssa.OpAMD64REPMOVSQ:
		gc.Prog(x86.AREP)
		gc.Prog(x86.AMOVSQ)
	case ssa.OpVarDef:
		gc.Gvardef(v.Aux.(*gc.Node))
	case ssa.OpVarKill:
		gc.Gvarkill(v.Aux.(*gc.Node))
	case ssa.OpVarLive:
		gc.Gvarlive(v.Aux.(*gc.Node))
	case ssa.OpKeepAlive:
		gc.KeepAlive(v)
	case ssa.OpAMD64LoweredNilCheck:
		// Issue a load which will fault if the input is nil.
		// TODO: We currently use the 2-byte instruction TESTB AX, (reg).
		// Should we use the 3-byte TESTB $0, (reg) instead?  It is larger
		// but it doesn't have false dependency on AX.
		// Or maybe allocate an output register and use MOVL (reg),reg2 ?
		// That trades clobbering flags for clobbering a register.
		p := gc.Prog(x86.ATESTB)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = x86.REG_AX
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
		if gc.Debug_checknil != 0 && v.Line > 1 { // v.Line==1 in generated wrappers
			gc.Warnl(v.Line, "generated nil check")
		}
	case ssa.OpAMD64MOVLatomicload, ssa.OpAMD64MOVQatomicload:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg0()
	case ssa.OpAMD64XCHGL, ssa.OpAMD64XCHGQ:
		r := v.Reg0()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output[0] not in same register %s", v.LongString())
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[1].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpAMD64XADDLlock, ssa.OpAMD64XADDQlock:
		r := v.Reg0()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output[0] not in same register %s", v.LongString())
		}
		gc.Prog(x86.ALOCK)
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[1].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpAMD64CMPXCHGLlock, ssa.OpAMD64CMPXCHGQlock:
		if v.Args[1].Reg() != x86.REG_AX {
			v.Fatalf("input[1] not in AX %s", v.LongString())
		}
		gc.Prog(x86.ALOCK)
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
		p = gc.Prog(x86.ASETEQ)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg0()
	case ssa.OpAMD64ANDBlock, ssa.OpAMD64ORBlock:
		gc.Prog(x86.ALOCK)
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
	default:
		v.Fatalf("genValue not implemented: %s", v.LongString())
	}
}
예제 #5
0
파일: ssa.go 프로젝트: achanda/go
func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
	s.SetLineno(v.Line)
	switch v.Op {
	case ssa.OpInitMem:
		// memory arg needs no code
	case ssa.OpArg:
		// input args need no code
	case ssa.OpSP, ssa.OpSB, ssa.OpGetG:
		// nothing to do
	case ssa.OpCopy, ssa.OpARM64MOVDconvert, ssa.OpARM64MOVDreg:
		if v.Type.IsMemory() {
			return
		}
		x := v.Args[0].Reg()
		y := v.Reg()
		if x == y {
			return
		}
		as := arm64.AMOVD
		if v.Type.IsFloat() {
			switch v.Type.Size() {
			case 4:
				as = arm64.AFMOVS
			case 8:
				as = arm64.AFMOVD
			default:
				panic("bad float size")
			}
		}
		p := gc.Prog(as)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = x
		p.To.Type = obj.TYPE_REG
		p.To.Reg = y
	case ssa.OpARM64MOVDnop:
		if v.Reg() != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		// nothing to do
	case ssa.OpLoadReg:
		if v.Type.IsFlags() {
			v.Fatalf("load flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(loadByType(v.Type))
		gc.AddrAuto(&p.From, v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpPhi:
		gc.CheckLoweredPhi(v)
	case ssa.OpStoreReg:
		if v.Type.IsFlags() {
			v.Fatalf("store flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(storeByType(v.Type))
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		gc.AddrAuto(&p.To, v)
	case ssa.OpARM64ADD,
		ssa.OpARM64SUB,
		ssa.OpARM64AND,
		ssa.OpARM64OR,
		ssa.OpARM64XOR,
		ssa.OpARM64BIC,
		ssa.OpARM64MUL,
		ssa.OpARM64MULW,
		ssa.OpARM64MULH,
		ssa.OpARM64UMULH,
		ssa.OpARM64MULL,
		ssa.OpARM64UMULL,
		ssa.OpARM64DIV,
		ssa.OpARM64UDIV,
		ssa.OpARM64DIVW,
		ssa.OpARM64UDIVW,
		ssa.OpARM64MOD,
		ssa.OpARM64UMOD,
		ssa.OpARM64MODW,
		ssa.OpARM64UMODW,
		ssa.OpARM64SLL,
		ssa.OpARM64SRL,
		ssa.OpARM64SRA,
		ssa.OpARM64FADDS,
		ssa.OpARM64FADDD,
		ssa.OpARM64FSUBS,
		ssa.OpARM64FSUBD,
		ssa.OpARM64FMULS,
		ssa.OpARM64FMULD,
		ssa.OpARM64FDIVS,
		ssa.OpARM64FDIVD:
		r := v.Reg()
		r1 := v.Args[0].Reg()
		r2 := v.Args[1].Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r2
		p.Reg = r1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpARM64ADDconst,
		ssa.OpARM64SUBconst,
		ssa.OpARM64ANDconst,
		ssa.OpARM64ORconst,
		ssa.OpARM64XORconst,
		ssa.OpARM64BICconst,
		ssa.OpARM64SLLconst,
		ssa.OpARM64SRLconst,
		ssa.OpARM64SRAconst,
		ssa.OpARM64RORconst,
		ssa.OpARM64RORWconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARM64ADDshiftLL,
		ssa.OpARM64SUBshiftLL,
		ssa.OpARM64ANDshiftLL,
		ssa.OpARM64ORshiftLL,
		ssa.OpARM64XORshiftLL,
		ssa.OpARM64BICshiftLL:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm64.SHIFT_LL, v.AuxInt)
	case ssa.OpARM64ADDshiftRL,
		ssa.OpARM64SUBshiftRL,
		ssa.OpARM64ANDshiftRL,
		ssa.OpARM64ORshiftRL,
		ssa.OpARM64XORshiftRL,
		ssa.OpARM64BICshiftRL:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm64.SHIFT_LR, v.AuxInt)
	case ssa.OpARM64ADDshiftRA,
		ssa.OpARM64SUBshiftRA,
		ssa.OpARM64ANDshiftRA,
		ssa.OpARM64ORshiftRA,
		ssa.OpARM64XORshiftRA,
		ssa.OpARM64BICshiftRA:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm64.SHIFT_AR, v.AuxInt)
	case ssa.OpARM64MOVDconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARM64FMOVSconst,
		ssa.OpARM64FMOVDconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_FCONST
		p.From.Val = math.Float64frombits(uint64(v.AuxInt))
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARM64CMP,
		ssa.OpARM64CMPW,
		ssa.OpARM64CMN,
		ssa.OpARM64CMNW,
		ssa.OpARM64FCMPS,
		ssa.OpARM64FCMPD:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.Reg = v.Args[0].Reg()
	case ssa.OpARM64CMPconst,
		ssa.OpARM64CMPWconst,
		ssa.OpARM64CMNconst,
		ssa.OpARM64CMNWconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.Reg = v.Args[0].Reg()
	case ssa.OpARM64CMPshiftLL:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), 0, arm64.SHIFT_LL, v.AuxInt)
	case ssa.OpARM64CMPshiftRL:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), 0, arm64.SHIFT_LR, v.AuxInt)
	case ssa.OpARM64CMPshiftRA:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), 0, arm64.SHIFT_AR, v.AuxInt)
	case ssa.OpARM64MOVDaddr:
		p := gc.Prog(arm64.AMOVD)
		p.From.Type = obj.TYPE_ADDR
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()

		var wantreg string
		// MOVD $sym+off(base), R
		// the assembler expands it as the following:
		// - base is SP: add constant offset to SP (R13)
		//               when constant is large, tmp register (R11) may be used
		// - base is SB: load external address from constant pool (use relocation)
		switch v.Aux.(type) {
		default:
			v.Fatalf("aux is of unknown type %T", v.Aux)
		case *ssa.ExternSymbol:
			wantreg = "SB"
			gc.AddAux(&p.From, v)
		case *ssa.ArgSymbol, *ssa.AutoSymbol:
			wantreg = "SP"
			gc.AddAux(&p.From, v)
		case nil:
			// No sym, just MOVD $off(SP), R
			wantreg = "SP"
			p.From.Reg = arm64.REGSP
			p.From.Offset = v.AuxInt
		}
		if reg := v.Args[0].RegName(); reg != wantreg {
			v.Fatalf("bad reg %s for symbol type %T, want %s", reg, v.Aux, wantreg)
		}
	case ssa.OpARM64MOVBload,
		ssa.OpARM64MOVBUload,
		ssa.OpARM64MOVHload,
		ssa.OpARM64MOVHUload,
		ssa.OpARM64MOVWload,
		ssa.OpARM64MOVWUload,
		ssa.OpARM64MOVDload,
		ssa.OpARM64FMOVSload,
		ssa.OpARM64FMOVDload:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARM64LDAR,
		ssa.OpARM64LDARW:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg0()
	case ssa.OpARM64MOVBstore,
		ssa.OpARM64MOVHstore,
		ssa.OpARM64MOVWstore,
		ssa.OpARM64MOVDstore,
		ssa.OpARM64FMOVSstore,
		ssa.OpARM64FMOVDstore,
		ssa.OpARM64STLR,
		ssa.OpARM64STLRW:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpARM64MOVBstorezero,
		ssa.OpARM64MOVHstorezero,
		ssa.OpARM64MOVWstorezero,
		ssa.OpARM64MOVDstorezero:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = arm64.REGZERO
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpARM64LoweredAtomicExchange64,
		ssa.OpARM64LoweredAtomicExchange32:
		// LDAXR	(Rarg0), Rout
		// STLXR	Rarg1, (Rarg0), Rtmp
		// CBNZ		Rtmp, -2(PC)
		ld := arm64.ALDAXR
		st := arm64.ASTLXR
		if v.Op == ssa.OpARM64LoweredAtomicExchange32 {
			ld = arm64.ALDAXRW
			st = arm64.ASTLXRW
		}
		r0 := v.Args[0].Reg()
		r1 := v.Args[1].Reg()
		out := v.Reg0()
		p := gc.Prog(ld)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = r0
		p.To.Type = obj.TYPE_REG
		p.To.Reg = out
		p1 := gc.Prog(st)
		p1.From.Type = obj.TYPE_REG
		p1.From.Reg = r1
		p1.To.Type = obj.TYPE_MEM
		p1.To.Reg = r0
		p1.RegTo2 = arm64.REGTMP
		p2 := gc.Prog(arm64.ACBNZ)
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = arm64.REGTMP
		p2.To.Type = obj.TYPE_BRANCH
		gc.Patch(p2, p)
	case ssa.OpARM64LoweredAtomicAdd64,
		ssa.OpARM64LoweredAtomicAdd32:
		// LDAXR	(Rarg0), Rout
		// ADD		Rarg1, Rout
		// STLXR	Rout, (Rarg0), Rtmp
		// CBNZ		Rtmp, -3(PC)
		ld := arm64.ALDAXR
		st := arm64.ASTLXR
		if v.Op == ssa.OpARM64LoweredAtomicAdd32 {
			ld = arm64.ALDAXRW
			st = arm64.ASTLXRW
		}
		r0 := v.Args[0].Reg()
		r1 := v.Args[1].Reg()
		out := v.Reg0()
		p := gc.Prog(ld)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = r0
		p.To.Type = obj.TYPE_REG
		p.To.Reg = out
		p1 := gc.Prog(arm64.AADD)
		p1.From.Type = obj.TYPE_REG
		p1.From.Reg = r1
		p1.To.Type = obj.TYPE_REG
		p1.To.Reg = out
		p2 := gc.Prog(st)
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = out
		p2.To.Type = obj.TYPE_MEM
		p2.To.Reg = r0
		p2.RegTo2 = arm64.REGTMP
		p3 := gc.Prog(arm64.ACBNZ)
		p3.From.Type = obj.TYPE_REG
		p3.From.Reg = arm64.REGTMP
		p3.To.Type = obj.TYPE_BRANCH
		gc.Patch(p3, p)
	case ssa.OpARM64LoweredAtomicCas64,
		ssa.OpARM64LoweredAtomicCas32:
		// LDAXR	(Rarg0), Rtmp
		// CMP		Rarg1, Rtmp
		// BNE		3(PC)
		// STLXR	Rarg2, (Rarg0), Rtmp
		// CBNZ		Rtmp, -4(PC)
		// CSET		EQ, Rout
		ld := arm64.ALDAXR
		st := arm64.ASTLXR
		cmp := arm64.ACMP
		if v.Op == ssa.OpARM64LoweredAtomicCas32 {
			ld = arm64.ALDAXRW
			st = arm64.ASTLXRW
			cmp = arm64.ACMPW
		}
		r0 := v.Args[0].Reg()
		r1 := v.Args[1].Reg()
		r2 := v.Args[2].Reg()
		out := v.Reg0()
		p := gc.Prog(ld)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = r0
		p.To.Type = obj.TYPE_REG
		p.To.Reg = arm64.REGTMP
		p1 := gc.Prog(cmp)
		p1.From.Type = obj.TYPE_REG
		p1.From.Reg = r1
		p1.Reg = arm64.REGTMP
		p2 := gc.Prog(arm64.ABNE)
		p2.To.Type = obj.TYPE_BRANCH
		p3 := gc.Prog(st)
		p3.From.Type = obj.TYPE_REG
		p3.From.Reg = r2
		p3.To.Type = obj.TYPE_MEM
		p3.To.Reg = r0
		p3.RegTo2 = arm64.REGTMP
		p4 := gc.Prog(arm64.ACBNZ)
		p4.From.Type = obj.TYPE_REG
		p4.From.Reg = arm64.REGTMP
		p4.To.Type = obj.TYPE_BRANCH
		gc.Patch(p4, p)
		p5 := gc.Prog(arm64.ACSET)
		p5.From.Type = obj.TYPE_REG // assembler encodes conditional bits in Reg
		p5.From.Reg = arm64.COND_EQ
		p5.To.Type = obj.TYPE_REG
		p5.To.Reg = out
		gc.Patch(p2, p5)
	case ssa.OpARM64LoweredAtomicAnd8,
		ssa.OpARM64LoweredAtomicOr8:
		// LDAXRB	(Rarg0), Rtmp
		// AND/OR	Rarg1, Rtmp
		// STLXRB	Rtmp, (Rarg0), Rtmp
		// CBNZ		Rtmp, -3(PC)
		r0 := v.Args[0].Reg()
		r1 := v.Args[1].Reg()
		p := gc.Prog(arm64.ALDAXRB)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = r0
		p.To.Type = obj.TYPE_REG
		p.To.Reg = arm64.REGTMP
		p1 := gc.Prog(v.Op.Asm())
		p1.From.Type = obj.TYPE_REG
		p1.From.Reg = r1
		p1.To.Type = obj.TYPE_REG
		p1.To.Reg = arm64.REGTMP
		p2 := gc.Prog(arm64.ASTLXRB)
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = arm64.REGTMP
		p2.To.Type = obj.TYPE_MEM
		p2.To.Reg = r0
		p2.RegTo2 = arm64.REGTMP
		p3 := gc.Prog(arm64.ACBNZ)
		p3.From.Type = obj.TYPE_REG
		p3.From.Reg = arm64.REGTMP
		p3.To.Type = obj.TYPE_BRANCH
		gc.Patch(p3, p)
	case ssa.OpARM64MOVBreg,
		ssa.OpARM64MOVBUreg,
		ssa.OpARM64MOVHreg,
		ssa.OpARM64MOVHUreg,
		ssa.OpARM64MOVWreg,
		ssa.OpARM64MOVWUreg:
		a := v.Args[0]
		for a.Op == ssa.OpCopy || a.Op == ssa.OpARM64MOVDreg {
			a = a.Args[0]
		}
		if a.Op == ssa.OpLoadReg {
			t := a.Type
			switch {
			case v.Op == ssa.OpARM64MOVBreg && t.Size() == 1 && t.IsSigned(),
				v.Op == ssa.OpARM64MOVBUreg && t.Size() == 1 && !t.IsSigned(),
				v.Op == ssa.OpARM64MOVHreg && t.Size() == 2 && t.IsSigned(),
				v.Op == ssa.OpARM64MOVHUreg && t.Size() == 2 && !t.IsSigned(),
				v.Op == ssa.OpARM64MOVWreg && t.Size() == 4 && t.IsSigned(),
				v.Op == ssa.OpARM64MOVWUreg && t.Size() == 4 && !t.IsSigned():
				// arg is a proper-typed load, already zero/sign-extended, don't extend again
				if v.Reg() == v.Args[0].Reg() {
					return
				}
				p := gc.Prog(arm64.AMOVD)
				p.From.Type = obj.TYPE_REG
				p.From.Reg = v.Args[0].Reg()
				p.To.Type = obj.TYPE_REG
				p.To.Reg = v.Reg()
				return
			default:
			}
		}
		fallthrough
	case ssa.OpARM64MVN,
		ssa.OpARM64NEG,
		ssa.OpARM64FNEGS,
		ssa.OpARM64FNEGD,
		ssa.OpARM64FSQRTD,
		ssa.OpARM64FCVTZSSW,
		ssa.OpARM64FCVTZSDW,
		ssa.OpARM64FCVTZUSW,
		ssa.OpARM64FCVTZUDW,
		ssa.OpARM64FCVTZSS,
		ssa.OpARM64FCVTZSD,
		ssa.OpARM64FCVTZUS,
		ssa.OpARM64FCVTZUD,
		ssa.OpARM64SCVTFWS,
		ssa.OpARM64SCVTFWD,
		ssa.OpARM64SCVTFS,
		ssa.OpARM64SCVTFD,
		ssa.OpARM64UCVTFWS,
		ssa.OpARM64UCVTFWD,
		ssa.OpARM64UCVTFS,
		ssa.OpARM64UCVTFD,
		ssa.OpARM64FCVTSD,
		ssa.OpARM64FCVTDS,
		ssa.OpARM64REV,
		ssa.OpARM64REVW,
		ssa.OpARM64REV16W,
		ssa.OpARM64RBIT,
		ssa.OpARM64RBITW,
		ssa.OpARM64CLZ,
		ssa.OpARM64CLZW:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARM64CSELULT,
		ssa.OpARM64CSELULT0:
		r1 := int16(arm64.REGZERO)
		if v.Op == ssa.OpARM64CSELULT {
			r1 = v.Args[1].Reg()
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG // assembler encodes conditional bits in Reg
		p.From.Reg = arm64.COND_LO
		p.Reg = v.Args[0].Reg()
		p.From3 = &obj.Addr{Type: obj.TYPE_REG, Reg: r1}
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARM64DUFFZERO:
		// runtime.duffzero expects start address - 8 in R16
		p := gc.Prog(arm64.ASUB)
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = 8
		p.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = arm64.REG_R16
		p = gc.Prog(obj.ADUFFZERO)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Pkglookup("duffzero", gc.Runtimepkg))
		p.To.Offset = v.AuxInt
	case ssa.OpARM64LoweredZero:
		// MOVD.P	ZR, 8(R16)
		// CMP	Rarg1, R16
		// BLE	-2(PC)
		// arg1 is the address of the last element to zero
		p := gc.Prog(arm64.AMOVD)
		p.Scond = arm64.C_XPOST
		p.From.Type = obj.TYPE_REG
		p.From.Reg = arm64.REGZERO
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = arm64.REG_R16
		p.To.Offset = 8
		p2 := gc.Prog(arm64.ACMP)
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = v.Args[1].Reg()
		p2.Reg = arm64.REG_R16
		p3 := gc.Prog(arm64.ABLE)
		p3.To.Type = obj.TYPE_BRANCH
		gc.Patch(p3, p)
	case ssa.OpARM64DUFFCOPY:
		p := gc.Prog(obj.ADUFFCOPY)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Pkglookup("duffcopy", gc.Runtimepkg))
		p.To.Offset = v.AuxInt
	case ssa.OpARM64LoweredMove:
		// MOVD.P	8(R16), Rtmp
		// MOVD.P	Rtmp, 8(R17)
		// CMP	Rarg2, R16
		// BLE	-3(PC)
		// arg2 is the address of the last element of src
		p := gc.Prog(arm64.AMOVD)
		p.Scond = arm64.C_XPOST
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = arm64.REG_R16
		p.From.Offset = 8
		p.To.Type = obj.TYPE_REG
		p.To.Reg = arm64.REGTMP
		p2 := gc.Prog(arm64.AMOVD)
		p2.Scond = arm64.C_XPOST
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = arm64.REGTMP
		p2.To.Type = obj.TYPE_MEM
		p2.To.Reg = arm64.REG_R17
		p2.To.Offset = 8
		p3 := gc.Prog(arm64.ACMP)
		p3.From.Type = obj.TYPE_REG
		p3.From.Reg = v.Args[2].Reg()
		p3.Reg = arm64.REG_R16
		p4 := gc.Prog(arm64.ABLE)
		p4.To.Type = obj.TYPE_BRANCH
		gc.Patch(p4, p)
	case ssa.OpARM64CALLstatic:
		if v.Aux.(*gc.Sym) == gc.Deferreturn.Sym {
			// Deferred calls will appear to be returning to
			// the CALL deferreturn(SB) that we are about to emit.
			// However, the stack trace code will show the line
			// of the instruction byte before the return PC.
			// To avoid that being an unrelated instruction,
			// insert an actual hardware NOP that will have the right line number.
			// This is different from obj.ANOP, which is a virtual no-op
			// that doesn't make it into the instruction stream.
			ginsnop()
		}
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(v.Aux.(*gc.Sym))
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARM64CALLclosure:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Offset = 0
		p.To.Reg = v.Args[0].Reg()
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARM64CALLdefer:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Deferproc.Sym)
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARM64CALLgo:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Newproc.Sym)
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARM64CALLinter:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Offset = 0
		p.To.Reg = v.Args[0].Reg()
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARM64LoweredNilCheck:
		// Issue a load which will fault if arg is nil.
		p := gc.Prog(arm64.AMOVB)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = arm64.REGTMP
		if gc.Debug_checknil != 0 && v.Line > 1 { // v.Line==1 in generated wrappers
			gc.Warnl(v.Line, "generated nil check")
		}
	case ssa.OpVarDef:
		gc.Gvardef(v.Aux.(*gc.Node))
	case ssa.OpVarKill:
		gc.Gvarkill(v.Aux.(*gc.Node))
	case ssa.OpVarLive:
		gc.Gvarlive(v.Aux.(*gc.Node))
	case ssa.OpKeepAlive:
		gc.KeepAlive(v)
	case ssa.OpARM64Equal,
		ssa.OpARM64NotEqual,
		ssa.OpARM64LessThan,
		ssa.OpARM64LessEqual,
		ssa.OpARM64GreaterThan,
		ssa.OpARM64GreaterEqual,
		ssa.OpARM64LessThanU,
		ssa.OpARM64LessEqualU,
		ssa.OpARM64GreaterThanU,
		ssa.OpARM64GreaterEqualU:
		// generate boolean values using CSET
		p := gc.Prog(arm64.ACSET)
		p.From.Type = obj.TYPE_REG // assembler encodes conditional bits in Reg
		p.From.Reg = condBits[v.Op]
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpSelect0, ssa.OpSelect1:
		// nothing to do
	case ssa.OpARM64LoweredGetClosurePtr:
		// Closure pointer is R26 (arm64.REGCTXT).
		gc.CheckLoweredGetClosurePtr(v)
	case ssa.OpARM64FlagEQ,
		ssa.OpARM64FlagLT_ULT,
		ssa.OpARM64FlagLT_UGT,
		ssa.OpARM64FlagGT_ULT,
		ssa.OpARM64FlagGT_UGT:
		v.Fatalf("Flag* ops should never make it to codegen %v", v.LongString())
	case ssa.OpARM64InvertFlags:
		v.Fatalf("InvertFlags should never make it to codegen %v", v.LongString())
	default:
		v.Fatalf("genValue not implemented: %s", v.LongString())
	}
}
예제 #6
0
파일: ssa.go 프로젝트: achanda/go
func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
	s.SetLineno(v.Line)
	switch v.Op {
	case ssa.OpS390XSLD, ssa.OpS390XSLW,
		ssa.OpS390XSRD, ssa.OpS390XSRW,
		ssa.OpS390XSRAD, ssa.OpS390XSRAW:
		r := v.Reg()
		r1 := v.Args[0].Reg()
		r2 := v.Args[1].Reg()
		if r2 == s390x.REG_R0 {
			v.Fatalf("cannot use R0 as shift value %s", v.LongString())
		}
		p := opregreg(v.Op.Asm(), r, r2)
		if r != r1 {
			p.Reg = r1
		}
	case ssa.OpS390XADD, ssa.OpS390XADDW,
		ssa.OpS390XSUB, ssa.OpS390XSUBW,
		ssa.OpS390XAND, ssa.OpS390XANDW,
		ssa.OpS390XOR, ssa.OpS390XORW,
		ssa.OpS390XXOR, ssa.OpS390XXORW:
		r := v.Reg()
		r1 := v.Args[0].Reg()
		r2 := v.Args[1].Reg()
		p := opregreg(v.Op.Asm(), r, r2)
		if r != r1 {
			p.Reg = r1
		}
	// 2-address opcode arithmetic
	case ssa.OpS390XMULLD, ssa.OpS390XMULLW,
		ssa.OpS390XMULHD, ssa.OpS390XMULHDU,
		ssa.OpS390XFADDS, ssa.OpS390XFADD, ssa.OpS390XFSUBS, ssa.OpS390XFSUB,
		ssa.OpS390XFMULS, ssa.OpS390XFMUL, ssa.OpS390XFDIVS, ssa.OpS390XFDIV:
		r := v.Reg()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		opregreg(v.Op.Asm(), r, v.Args[1].Reg())
	case ssa.OpS390XDIVD, ssa.OpS390XDIVW,
		ssa.OpS390XDIVDU, ssa.OpS390XDIVWU,
		ssa.OpS390XMODD, ssa.OpS390XMODW,
		ssa.OpS390XMODDU, ssa.OpS390XMODWU:

		// TODO(mundaym): use the temp registers every time like x86 does with AX?
		dividend := v.Args[0].Reg()
		divisor := v.Args[1].Reg()

		// CPU faults upon signed overflow, which occurs when most
		// negative int is divided by -1.
		var j *obj.Prog
		if v.Op == ssa.OpS390XDIVD || v.Op == ssa.OpS390XDIVW ||
			v.Op == ssa.OpS390XMODD || v.Op == ssa.OpS390XMODW {

			var c *obj.Prog
			c = gc.Prog(s390x.ACMP)
			j = gc.Prog(s390x.ABEQ)

			c.From.Type = obj.TYPE_REG
			c.From.Reg = divisor
			c.To.Type = obj.TYPE_CONST
			c.To.Offset = -1

			j.To.Type = obj.TYPE_BRANCH

		}

		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = divisor
		p.Reg = 0
		p.To.Type = obj.TYPE_REG
		p.To.Reg = dividend

		// signed division, rest of the check for -1 case
		if j != nil {
			j2 := gc.Prog(s390x.ABR)
			j2.To.Type = obj.TYPE_BRANCH

			var n *obj.Prog
			if v.Op == ssa.OpS390XDIVD || v.Op == ssa.OpS390XDIVW {
				// n * -1 = -n
				n = gc.Prog(s390x.ANEG)
				n.To.Type = obj.TYPE_REG
				n.To.Reg = dividend
			} else {
				// n % -1 == 0
				n = gc.Prog(s390x.AXOR)
				n.From.Type = obj.TYPE_REG
				n.From.Reg = dividend
				n.To.Type = obj.TYPE_REG
				n.To.Reg = dividend
			}

			j.To.Val = n
			j2.To.Val = s.Pc()
		}
	case ssa.OpS390XADDconst, ssa.OpS390XADDWconst:
		opregregimm(v.Op.Asm(), v.Reg(), v.Args[0].Reg(), v.AuxInt)
	case ssa.OpS390XMULLDconst, ssa.OpS390XMULLWconst,
		ssa.OpS390XSUBconst, ssa.OpS390XSUBWconst,
		ssa.OpS390XANDconst, ssa.OpS390XANDWconst,
		ssa.OpS390XORconst, ssa.OpS390XORWconst,
		ssa.OpS390XXORconst, ssa.OpS390XXORWconst:
		r := v.Reg()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpS390XSLDconst, ssa.OpS390XSLWconst,
		ssa.OpS390XSRDconst, ssa.OpS390XSRWconst,
		ssa.OpS390XSRADconst, ssa.OpS390XSRAWconst,
		ssa.OpS390XRLLGconst, ssa.OpS390XRLLconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		r := v.Reg()
		r1 := v.Args[0].Reg()
		if r != r1 {
			p.Reg = r1
		}
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpS390XSUBEcarrymask, ssa.OpS390XSUBEWcarrymask:
		r := v.Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpS390XMOVDaddridx:
		r := v.Args[0].Reg()
		i := v.Args[1].Reg()
		p := gc.Prog(s390x.AMOVD)
		p.From.Scale = 1
		if i == s390x.REGSP {
			r, i = i, r
		}
		p.From.Type = obj.TYPE_ADDR
		p.From.Reg = r
		p.From.Index = i
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpS390XMOVDaddr:
		p := gc.Prog(s390x.AMOVD)
		p.From.Type = obj.TYPE_ADDR
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpS390XCMP, ssa.OpS390XCMPW, ssa.OpS390XCMPU, ssa.OpS390XCMPWU:
		opregreg(v.Op.Asm(), v.Args[1].Reg(), v.Args[0].Reg())
	case ssa.OpS390XFCMPS, ssa.OpS390XFCMP:
		opregreg(v.Op.Asm(), v.Args[1].Reg(), v.Args[0].Reg())
	case ssa.OpS390XCMPconst, ssa.OpS390XCMPWconst, ssa.OpS390XCMPUconst, ssa.OpS390XCMPWUconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_CONST
		p.To.Offset = v.AuxInt
	case ssa.OpS390XMOVDconst:
		x := v.Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = x
	case ssa.OpS390XFMOVSconst, ssa.OpS390XFMOVDconst:
		x := v.Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_FCONST
		p.From.Val = math.Float64frombits(uint64(v.AuxInt))
		p.To.Type = obj.TYPE_REG
		p.To.Reg = x
	case ssa.OpS390XADDWload, ssa.OpS390XADDload,
		ssa.OpS390XMULLWload, ssa.OpS390XMULLDload,
		ssa.OpS390XSUBWload, ssa.OpS390XSUBload,
		ssa.OpS390XANDWload, ssa.OpS390XANDload,
		ssa.OpS390XORWload, ssa.OpS390XORload,
		ssa.OpS390XXORWload, ssa.OpS390XXORload:
		r := v.Reg()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[1].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpS390XMOVDload,
		ssa.OpS390XMOVWZload, ssa.OpS390XMOVHZload, ssa.OpS390XMOVBZload,
		ssa.OpS390XMOVDBRload, ssa.OpS390XMOVWBRload, ssa.OpS390XMOVHBRload,
		ssa.OpS390XMOVBload, ssa.OpS390XMOVHload, ssa.OpS390XMOVWload,
		ssa.OpS390XFMOVSload, ssa.OpS390XFMOVDload:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpS390XMOVBZloadidx, ssa.OpS390XMOVHZloadidx, ssa.OpS390XMOVWZloadidx, ssa.OpS390XMOVDloadidx,
		ssa.OpS390XMOVHBRloadidx, ssa.OpS390XMOVWBRloadidx, ssa.OpS390XMOVDBRloadidx,
		ssa.OpS390XFMOVSloadidx, ssa.OpS390XFMOVDloadidx:
		r := v.Args[0].Reg()
		i := v.Args[1].Reg()
		if i == s390x.REGSP {
			r, i = i, r
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = r
		p.From.Scale = 1
		p.From.Index = i
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpS390XMOVBstore, ssa.OpS390XMOVHstore, ssa.OpS390XMOVWstore, ssa.OpS390XMOVDstore,
		ssa.OpS390XMOVHBRstore, ssa.OpS390XMOVWBRstore, ssa.OpS390XMOVDBRstore,
		ssa.OpS390XFMOVSstore, ssa.OpS390XFMOVDstore:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpS390XMOVBstoreidx, ssa.OpS390XMOVHstoreidx, ssa.OpS390XMOVWstoreidx, ssa.OpS390XMOVDstoreidx,
		ssa.OpS390XMOVHBRstoreidx, ssa.OpS390XMOVWBRstoreidx, ssa.OpS390XMOVDBRstoreidx,
		ssa.OpS390XFMOVSstoreidx, ssa.OpS390XFMOVDstoreidx:
		r := v.Args[0].Reg()
		i := v.Args[1].Reg()
		if i == s390x.REGSP {
			r, i = i, r
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = r
		p.To.Scale = 1
		p.To.Index = i
		gc.AddAux(&p.To, v)
	case ssa.OpS390XMOVDstoreconst, ssa.OpS390XMOVWstoreconst, ssa.OpS390XMOVHstoreconst, ssa.OpS390XMOVBstoreconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		sc := v.AuxValAndOff()
		p.From.Offset = sc.Val()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux2(&p.To, v, sc.Off())
	case ssa.OpS390XMOVBreg, ssa.OpS390XMOVHreg, ssa.OpS390XMOVWreg,
		ssa.OpS390XMOVBZreg, ssa.OpS390XMOVHZreg, ssa.OpS390XMOVWZreg,
		ssa.OpS390XCEFBRA, ssa.OpS390XCDFBRA, ssa.OpS390XCEGBRA, ssa.OpS390XCDGBRA,
		ssa.OpS390XCFEBRA, ssa.OpS390XCFDBRA, ssa.OpS390XCGEBRA, ssa.OpS390XCGDBRA,
		ssa.OpS390XLDEBR, ssa.OpS390XLEDBR,
		ssa.OpS390XFNEG, ssa.OpS390XFNEGS:
		opregreg(v.Op.Asm(), v.Reg(), v.Args[0].Reg())
	case ssa.OpS390XCLEAR:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		sc := v.AuxValAndOff()
		p.From.Offset = sc.Val()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux2(&p.To, v, sc.Off())
	case ssa.OpCopy, ssa.OpS390XMOVDconvert:
		if v.Type.IsMemory() {
			return
		}
		x := v.Args[0].Reg()
		y := v.Reg()
		if x != y {
			opregreg(moveByType(v.Type), y, x)
		}
	case ssa.OpLoadReg:
		if v.Type.IsFlags() {
			v.Fatalf("load flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(loadByType(v.Type))
		gc.AddrAuto(&p.From, v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpStoreReg:
		if v.Type.IsFlags() {
			v.Fatalf("store flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(storeByType(v.Type))
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		gc.AddrAuto(&p.To, v)
	case ssa.OpPhi:
		gc.CheckLoweredPhi(v)
	case ssa.OpInitMem:
		// memory arg needs no code
	case ssa.OpArg:
		// input args need no code
	case ssa.OpS390XLoweredGetClosurePtr:
		// Closure pointer is R12 (already)
		gc.CheckLoweredGetClosurePtr(v)
	case ssa.OpS390XLoweredGetG:
		r := v.Reg()
		p := gc.Prog(s390x.AMOVD)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = s390x.REGG
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpS390XCALLstatic:
		if v.Aux.(*gc.Sym) == gc.Deferreturn.Sym {
			// Deferred calls will appear to be returning to
			// the CALL deferreturn(SB) that we are about to emit.
			// However, the stack trace code will show the line
			// of the instruction byte before the return PC.
			// To avoid that being an unrelated instruction,
			// insert an actual hardware NOP that will have the right line number.
			// This is different from obj.ANOP, which is a virtual no-op
			// that doesn't make it into the instruction stream.
			ginsnop()
		}
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(v.Aux.(*gc.Sym))
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpS390XCALLclosure:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Args[0].Reg()
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpS390XCALLdefer:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Deferproc.Sym)
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpS390XCALLgo:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Newproc.Sym)
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpS390XCALLinter:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Args[0].Reg()
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpS390XFLOGR, ssa.OpS390XNEG, ssa.OpS390XNEGW,
		ssa.OpS390XMOVWBR, ssa.OpS390XMOVDBR:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpS390XNOT, ssa.OpS390XNOTW:
		v.Fatalf("NOT/NOTW generated %s", v.LongString())
	case ssa.OpS390XMOVDEQ, ssa.OpS390XMOVDNE,
		ssa.OpS390XMOVDLT, ssa.OpS390XMOVDLE,
		ssa.OpS390XMOVDGT, ssa.OpS390XMOVDGE,
		ssa.OpS390XMOVDGTnoinv, ssa.OpS390XMOVDGEnoinv:
		r := v.Reg()
		if r != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpS390XFSQRT:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpSP, ssa.OpSB:
		// nothing to do
	case ssa.OpSelect0, ssa.OpSelect1:
		// nothing to do
	case ssa.OpVarDef:
		gc.Gvardef(v.Aux.(*gc.Node))
	case ssa.OpVarKill:
		gc.Gvarkill(v.Aux.(*gc.Node))
	case ssa.OpVarLive:
		gc.Gvarlive(v.Aux.(*gc.Node))
	case ssa.OpKeepAlive:
		gc.KeepAlive(v)
	case ssa.OpS390XInvertFlags:
		v.Fatalf("InvertFlags should never make it to codegen %v", v.LongString())
	case ssa.OpS390XFlagEQ, ssa.OpS390XFlagLT, ssa.OpS390XFlagGT:
		v.Fatalf("Flag* ops should never make it to codegen %v", v.LongString())
	case ssa.OpS390XAddTupleFirst32, ssa.OpS390XAddTupleFirst64:
		v.Fatalf("AddTupleFirst* should never make it to codegen %v", v.LongString())
	case ssa.OpS390XLoweredNilCheck:
		// Issue a load which will fault if the input is nil.
		p := gc.Prog(s390x.AMOVBZ)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = s390x.REGTMP
		if gc.Debug_checknil != 0 && v.Line > 1 { // v.Line==1 in generated wrappers
			gc.Warnl(v.Line, "generated nil check")
		}
	case ssa.OpS390XMVC:
		vo := v.AuxValAndOff()
		p := gc.Prog(s390x.AMVC)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[1].Reg()
		p.From.Offset = vo.Off()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		p.To.Offset = vo.Off()
		p.From3 = new(obj.Addr)
		p.From3.Type = obj.TYPE_CONST
		p.From3.Offset = vo.Val()
	case ssa.OpS390XSTMG2, ssa.OpS390XSTMG3, ssa.OpS390XSTMG4,
		ssa.OpS390XSTM2, ssa.OpS390XSTM3, ssa.OpS390XSTM4:
		for i := 2; i < len(v.Args)-1; i++ {
			if v.Args[i].Reg() != v.Args[i-1].Reg()+1 {
				v.Fatalf("invalid store multiple %s", v.LongString())
			}
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.Reg = v.Args[len(v.Args)-2].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpS390XLoweredMove:
		// Inputs must be valid pointers to memory,
		// so adjust arg0 and arg1 as part of the expansion.
		// arg2 should be src+size,
		//
		// mvc: MVC  $256, 0(R2), 0(R1)
		//      MOVD $256(R1), R1
		//      MOVD $256(R2), R2
		//      CMP  R2, Rarg2
		//      BNE  mvc
		//      MVC  $rem, 0(R2), 0(R1) // if rem > 0
		// arg2 is the last address to move in the loop + 256
		mvc := gc.Prog(s390x.AMVC)
		mvc.From.Type = obj.TYPE_MEM
		mvc.From.Reg = v.Args[1].Reg()
		mvc.To.Type = obj.TYPE_MEM
		mvc.To.Reg = v.Args[0].Reg()
		mvc.From3 = new(obj.Addr)
		mvc.From3.Type = obj.TYPE_CONST
		mvc.From3.Offset = 256

		for i := 0; i < 2; i++ {
			movd := gc.Prog(s390x.AMOVD)
			movd.From.Type = obj.TYPE_ADDR
			movd.From.Reg = v.Args[i].Reg()
			movd.From.Offset = 256
			movd.To.Type = obj.TYPE_REG
			movd.To.Reg = v.Args[i].Reg()
		}

		cmpu := gc.Prog(s390x.ACMPU)
		cmpu.From.Reg = v.Args[1].Reg()
		cmpu.From.Type = obj.TYPE_REG
		cmpu.To.Reg = v.Args[2].Reg()
		cmpu.To.Type = obj.TYPE_REG

		bne := gc.Prog(s390x.ABLT)
		bne.To.Type = obj.TYPE_BRANCH
		gc.Patch(bne, mvc)

		if v.AuxInt > 0 {
			mvc := gc.Prog(s390x.AMVC)
			mvc.From.Type = obj.TYPE_MEM
			mvc.From.Reg = v.Args[1].Reg()
			mvc.To.Type = obj.TYPE_MEM
			mvc.To.Reg = v.Args[0].Reg()
			mvc.From3 = new(obj.Addr)
			mvc.From3.Type = obj.TYPE_CONST
			mvc.From3.Offset = v.AuxInt
		}
	case ssa.OpS390XLoweredZero:
		// Input must be valid pointers to memory,
		// so adjust arg0 as part of the expansion.
		// arg1 should be src+size,
		//
		// clear: CLEAR $256, 0(R1)
		//        MOVD  $256(R1), R1
		//        CMP   R1, Rarg1
		//        BNE   clear
		//        CLEAR $rem, 0(R1) // if rem > 0
		// arg1 is the last address to zero in the loop + 256
		clear := gc.Prog(s390x.ACLEAR)
		clear.From.Type = obj.TYPE_CONST
		clear.From.Offset = 256
		clear.To.Type = obj.TYPE_MEM
		clear.To.Reg = v.Args[0].Reg()

		movd := gc.Prog(s390x.AMOVD)
		movd.From.Type = obj.TYPE_ADDR
		movd.From.Reg = v.Args[0].Reg()
		movd.From.Offset = 256
		movd.To.Type = obj.TYPE_REG
		movd.To.Reg = v.Args[0].Reg()

		cmpu := gc.Prog(s390x.ACMPU)
		cmpu.From.Reg = v.Args[0].Reg()
		cmpu.From.Type = obj.TYPE_REG
		cmpu.To.Reg = v.Args[1].Reg()
		cmpu.To.Type = obj.TYPE_REG

		bne := gc.Prog(s390x.ABLT)
		bne.To.Type = obj.TYPE_BRANCH
		gc.Patch(bne, clear)

		if v.AuxInt > 0 {
			clear := gc.Prog(s390x.ACLEAR)
			clear.From.Type = obj.TYPE_CONST
			clear.From.Offset = v.AuxInt
			clear.To.Type = obj.TYPE_MEM
			clear.To.Reg = v.Args[0].Reg()
		}
	case ssa.OpS390XMOVWZatomicload, ssa.OpS390XMOVDatomicload:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg0()
	case ssa.OpS390XMOVWatomicstore, ssa.OpS390XMOVDatomicstore:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpS390XLAA, ssa.OpS390XLAAG:
		p := gc.Prog(v.Op.Asm())
		p.Reg = v.Reg0()
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpS390XLoweredAtomicCas32, ssa.OpS390XLoweredAtomicCas64:
		// Convert the flags output of CS{,G} into a bool.
		//    CS{,G} arg1, arg2, arg0
		//    MOVD   $0, ret
		//    BNE    2(PC)
		//    MOVD   $1, ret
		//    NOP (so the BNE has somewhere to land)

		// CS{,G} arg1, arg2, arg0
		cs := gc.Prog(v.Op.Asm())
		cs.From.Type = obj.TYPE_REG
		cs.From.Reg = v.Args[1].Reg() // old
		cs.Reg = v.Args[2].Reg()      // new
		cs.To.Type = obj.TYPE_MEM
		cs.To.Reg = v.Args[0].Reg()
		gc.AddAux(&cs.To, v)

		// MOVD $0, ret
		movd := gc.Prog(s390x.AMOVD)
		movd.From.Type = obj.TYPE_CONST
		movd.From.Offset = 0
		movd.To.Type = obj.TYPE_REG
		movd.To.Reg = v.Reg0()

		// BNE 2(PC)
		bne := gc.Prog(s390x.ABNE)
		bne.To.Type = obj.TYPE_BRANCH

		// MOVD $1, ret
		movd = gc.Prog(s390x.AMOVD)
		movd.From.Type = obj.TYPE_CONST
		movd.From.Offset = 1
		movd.To.Type = obj.TYPE_REG
		movd.To.Reg = v.Reg0()

		// NOP (so the BNE has somewhere to land)
		nop := gc.Prog(obj.ANOP)
		gc.Patch(bne, nop)
	case ssa.OpS390XLoweredAtomicExchange32, ssa.OpS390XLoweredAtomicExchange64:
		// Loop until the CS{,G} succeeds.
		//     MOV{WZ,D} arg0, ret
		// cs: CS{,G}    ret, arg1, arg0
		//     BNE       cs

		// MOV{WZ,D} arg0, ret
		load := gc.Prog(loadByType(v.Type.FieldType(0)))
		load.From.Type = obj.TYPE_MEM
		load.From.Reg = v.Args[0].Reg()
		load.To.Type = obj.TYPE_REG
		load.To.Reg = v.Reg0()
		gc.AddAux(&load.From, v)

		// CS{,G} ret, arg1, arg0
		cs := gc.Prog(v.Op.Asm())
		cs.From.Type = obj.TYPE_REG
		cs.From.Reg = v.Reg0()   // old
		cs.Reg = v.Args[1].Reg() // new
		cs.To.Type = obj.TYPE_MEM
		cs.To.Reg = v.Args[0].Reg()
		gc.AddAux(&cs.To, v)

		// BNE cs
		bne := gc.Prog(s390x.ABNE)
		bne.To.Type = obj.TYPE_BRANCH
		gc.Patch(bne, cs)
	default:
		v.Fatalf("genValue not implemented: %s", v.LongString())
	}
}