Beispiel #1
0
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:
		// nothing to do
	case ssa.OpCopy:
	case ssa.OpLoadReg:
		// TODO: by type
		p := gc.Prog(arm.AMOVW)
		n, off := gc.AutoVar(v.Args[0])
		p.From.Type = obj.TYPE_MEM
		p.From.Node = n
		p.From.Sym = gc.Linksym(n.Sym)
		p.From.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.From.Name = obj.NAME_PARAM
			p.From.Offset += n.Xoffset
		} else {
			p.From.Name = obj.NAME_AUTO
		}
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

	case ssa.OpStoreReg:
		// TODO: by type
		p := gc.Prog(arm.AMOVW)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		n, off := gc.AutoVar(v)
		p.To.Type = obj.TYPE_MEM
		p.To.Node = n
		p.To.Sym = gc.Linksym(n.Sym)
		p.To.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.To.Name = obj.NAME_PARAM
			p.To.Offset += n.Xoffset
		} else {
			p.To.Name = obj.NAME_AUTO
		}
	case ssa.OpARMADD:
		r := gc.SSARegNum(v)
		r1 := gc.SSARegNum(v.Args[0])
		r2 := gc.SSARegNum(v.Args[1])
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r1
		p.Reg = r2
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpARMADDconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		if v.Aux != nil {
			panic("can't handle symbolic constant yet")
		}
		p.Reg = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpARMMOVWconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt2Int64()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpARMCMP:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.Reg = gc.SSARegNum(v.Args[1])
	case ssa.OpARMMOVWload:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpARMMOVWstore:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.To, v)
	case ssa.OpARMCALLstatic:
		// TODO: deferreturn
		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.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.OpARMLessThan:
		v.Fatalf("pseudo-op made it to output: %s", v.LongString())
	default:
		v.Unimplementedf("genValue not implemented: %s", v.LongString())
	}
}
Beispiel #2
0
Datei: ssa.go Projekt: Mokolea/go
func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
	s.SetLineno(v.Line)
	switch v.Op {
	case ssa.OpAMD64ADDQ, ssa.OpAMD64ADDL:
		r := gc.SSARegNum(v)
		r1 := gc.SSARegNum(v.Args[0])
		r2 := gc.SSARegNum(v.Args[1])
		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 := gc.SSARegNum(v)
		if r != gc.SSARegNum(v.Args[0]) {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		opregreg(v.Op.Asm(), r, gc.SSARegNum(v.Args[1]))

	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 := gc.SSARegNum(v.Args[1])

		// 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 := gc.SSARegNum(v.Args[1])

		// 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 = gc.SSARegNum(v.Args[1])

		// 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.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 := gc.SSARegNum(v)
		if r != gc.SSARegNum(v.Args[0]) {
			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 = gc.SSARegNum(v.Args[1])
		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 := gc.SSARegNum(v)
		a := gc.SSARegNum(v.Args[0])
		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.OpAMD64CMOVQEQconst, ssa.OpAMD64CMOVLEQconst, ssa.OpAMD64CMOVWEQconst,
		ssa.OpAMD64CMOVQNEconst, ssa.OpAMD64CMOVLNEconst, ssa.OpAMD64CMOVWNEconst:
		r := gc.SSARegNum(v)
		if r != gc.SSARegNum(v.Args[0]) {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}

		// Constant into AX
		p := gc.Prog(moveByType(v.Type))
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = x86.REG_AX

		p = gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = x86.REG_AX
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r

	case ssa.OpAMD64MULQconst, ssa.OpAMD64MULLconst:
		r := gc.SSARegNum(v)
		if r != gc.SSARegNum(v.Args[0]) {
			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 = gc.SSARegNum(v.Args[0])

	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 := gc.SSARegNum(v)
		if r != gc.SSARegNum(v.Args[0]) {
			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 := gc.SSARegNum(v)
		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 := gc.SSARegNum(v.Args[0])
		i := gc.SSARegNum(v.Args[1])
		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 = gc.SSARegNum(v)
	case ssa.OpAMD64LEAQ, ssa.OpAMD64LEAL:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpAMD64CMPQ, ssa.OpAMD64CMPL, ssa.OpAMD64CMPW, ssa.OpAMD64CMPB,
		ssa.OpAMD64TESTQ, ssa.OpAMD64TESTL, ssa.OpAMD64TESTW, ssa.OpAMD64TESTB:
		opregreg(v.Op.Asm(), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v.Args[0]))
	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(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]))
	case ssa.OpAMD64CMPQconst, ssa.OpAMD64CMPLconst, ssa.OpAMD64CMPWconst, ssa.OpAMD64CMPBconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		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 = gc.SSARegNum(v.Args[0])
	case ssa.OpAMD64MOVLconst, ssa.OpAMD64MOVQconst:
		x := gc.SSARegNum(v)
		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 := gc.SSARegNum(v)
		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 = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpAMD64MOVQloadidx8, ssa.OpAMD64MOVSDloadidx8:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.From.Scale = 8
		p.From.Index = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpAMD64MOVLloadidx4, ssa.OpAMD64MOVSSloadidx4:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.From.Scale = 4
		p.From.Index = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpAMD64MOVWloadidx2:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.From.Scale = 2
		p.From.Index = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpAMD64MOVBloadidx1, ssa.OpAMD64MOVWloadidx1, ssa.OpAMD64MOVLloadidx1, ssa.OpAMD64MOVQloadidx1, ssa.OpAMD64MOVSSloadidx1, ssa.OpAMD64MOVSDloadidx1:
		r := gc.SSARegNum(v.Args[0])
		i := gc.SSARegNum(v.Args[1])
		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 = gc.SSARegNum(v)
	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 = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		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 = gc.SSARegNum(v.Args[2])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Scale = 8
		p.To.Index = gc.SSARegNum(v.Args[1])
		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 = gc.SSARegNum(v.Args[2])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Scale = 4
		p.To.Index = gc.SSARegNum(v.Args[1])
		gc.AddAux(&p.To, v)
	case ssa.OpAMD64MOVWstoreidx2:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[2])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Scale = 2
		p.To.Index = gc.SSARegNum(v.Args[1])
		gc.AddAux(&p.To, v)
	case ssa.OpAMD64MOVBstoreidx1, ssa.OpAMD64MOVWstoreidx1, ssa.OpAMD64MOVLstoreidx1, ssa.OpAMD64MOVQstoreidx1, ssa.OpAMD64MOVSSstoreidx1, ssa.OpAMD64MOVSDstoreidx1:
		r := gc.SSARegNum(v.Args[0])
		i := gc.SSARegNum(v.Args[1])
		if i == x86.REG_SP {
			r, i = i, r
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[2])
		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 = gc.SSARegNum(v.Args[0])
		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 := gc.SSARegNum(v.Args[0])
		i := gc.SSARegNum(v.Args[1])
		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.OpAMD64CVTSL2SS, ssa.OpAMD64CVTSL2SD, ssa.OpAMD64CVTSQ2SS, ssa.OpAMD64CVTSQ2SD,
		ssa.OpAMD64CVTTSS2SL, ssa.OpAMD64CVTTSD2SL, ssa.OpAMD64CVTTSS2SQ, ssa.OpAMD64CVTTSD2SQ,
		ssa.OpAMD64CVTSS2SD, ssa.OpAMD64CVTSD2SS:
		opregreg(v.Op.Asm(), gc.SSARegNum(v), gc.SSARegNum(v.Args[0]))
	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.Unimplementedf("MOVOconst can only do constant=0")
		}
		r := gc.SSARegNum(v)
		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 := gc.SSARegNum(v.Args[0])
		y := gc.SSARegNum(v)
		if x != y {
			opregreg(moveByType(v.Type), y, x)
		}
	case ssa.OpLoadReg:
		if v.Type.IsFlags() {
			v.Unimplementedf("load flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(loadByType(v.Type))
		n, off := gc.AutoVar(v.Args[0])
		p.From.Type = obj.TYPE_MEM
		p.From.Node = n
		p.From.Sym = gc.Linksym(n.Sym)
		p.From.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.From.Name = obj.NAME_PARAM
			p.From.Offset += n.Xoffset
		} else {
			p.From.Name = obj.NAME_AUTO
		}
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

	case ssa.OpStoreReg:
		if v.Type.IsFlags() {
			v.Unimplementedf("store flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(storeByType(v.Type))
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		n, off := gc.AutoVar(v)
		p.To.Type = obj.TYPE_MEM
		p.To.Node = n
		p.To.Sym = gc.Linksym(n.Sym)
		p.To.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.To.Name = obj.NAME_PARAM
			p.To.Offset += n.Xoffset
		} else {
			p.To.Name = obj.NAME_AUTO
		}
	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 := gc.SSARegNum(v)
		// 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 = gc.SSARegNum(v.Args[0])
		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 = gc.SSARegNum(v.Args[0])
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpAMD64NEGQ, ssa.OpAMD64NEGL,
		ssa.OpAMD64BSWAPQ, ssa.OpAMD64BSWAPL,
		ssa.OpAMD64NOTQ, ssa.OpAMD64NOTL:
		r := gc.SSARegNum(v)
		if r != gc.SSARegNum(v.Args[0]) {
			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, ssa.OpAMD64BSFW,
		ssa.OpAMD64BSRQ, ssa.OpAMD64BSRL, ssa.OpAMD64BSRW,
		ssa.OpAMD64SQRTSD:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	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 = gc.SSARegNum(v)

	case ssa.OpAMD64SETNEF:
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
		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, gc.SSARegNum(v), x86.REG_AX)

	case ssa.OpAMD64SETEQF:
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
		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, gc.SSARegNum(v), 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.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:
		if !v.Args[0].Type.IsPtrShaped() {
			v.Fatalf("keeping non-pointer alive %v", v.Args[0])
		}
		n, off := gc.AutoVar(v.Args[0])
		if n == nil {
			v.Fatalf("KeepLive with non-spilled value %s %s", v, v.Args[0])
		}
		if off != 0 {
			v.Fatalf("KeepLive with non-zero offset spill location %s:%d", n, off)
		}
		gc.Gvarlive(n)
	case ssa.OpAMD64LoweredNilCheck:
		// Optimization - if the subsequent block has a load or store
		// at the same address, we don't need to issue this instruction.
		mem := v.Args[1]
		for _, w := range v.Block.Succs[0].Block().Values {
			if w.Op == ssa.OpPhi {
				if w.Type.IsMemory() {
					mem = w
				}
				continue
			}
			if len(w.Args) == 0 || !w.Args[len(w.Args)-1].Type.IsMemory() {
				// w doesn't use a store - can't be a memory op.
				continue
			}
			if w.Args[len(w.Args)-1] != mem {
				v.Fatalf("wrong store after nilcheck v=%s w=%s", v, w)
			}
			switch w.Op {
			case ssa.OpAMD64MOVQload, ssa.OpAMD64MOVLload, ssa.OpAMD64MOVWload, ssa.OpAMD64MOVBload,
				ssa.OpAMD64MOVQstore, ssa.OpAMD64MOVLstore, ssa.OpAMD64MOVWstore, ssa.OpAMD64MOVBstore,
				ssa.OpAMD64MOVBQSXload, ssa.OpAMD64MOVWQSXload, ssa.OpAMD64MOVLQSXload,
				ssa.OpAMD64MOVSSload, ssa.OpAMD64MOVSDload, ssa.OpAMD64MOVOload,
				ssa.OpAMD64MOVSSstore, ssa.OpAMD64MOVSDstore, ssa.OpAMD64MOVOstore:
				if w.Args[0] == v.Args[0] && w.Aux == nil && w.AuxInt >= 0 && w.AuxInt < minZeroPage {
					if gc.Debug_checknil != 0 && int(v.Line) > 1 {
						gc.Warnl(v.Line, "removed nil check")
					}
					return
				}
			case ssa.OpAMD64MOVQstoreconst, ssa.OpAMD64MOVLstoreconst, ssa.OpAMD64MOVWstoreconst, ssa.OpAMD64MOVBstoreconst:
				off := ssa.ValAndOff(v.AuxInt).Off()
				if w.Args[0] == v.Args[0] && w.Aux == nil && off >= 0 && off < minZeroPage {
					if gc.Debug_checknil != 0 && int(v.Line) > 1 {
						gc.Warnl(v.Line, "removed nil check")
					}
					return
				}
			}
			if w.Type.IsMemory() {
				if w.Op == ssa.OpVarDef || w.Op == ssa.OpVarKill || w.Op == ssa.OpVarLive {
					// these ops are OK
					mem = w
					continue
				}
				// We can't delay the nil check past the next store.
				break
			}
		}
		// 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 = gc.SSARegNum(v.Args[0])
		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")
		}
	default:
		v.Unimplementedf("genValue not implemented: %s", v.LongString())
	}
}
Beispiel #3
0
func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
	s.SetLineno(v.Line)
	switch v.Op {
	case ssa.OpAMD64ADDQ, ssa.OpAMD64ADDL, ssa.OpAMD64ADDW:
		r := gc.SSARegNum(v)
		r1 := gc.SSARegNum(v.Args[0])
		r2 := gc.SSARegNum(v.Args[1])
		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
			switch v.Op {
			case ssa.OpAMD64ADDQ:
				asm = x86.ALEAQ
			case ssa.OpAMD64ADDL:
				asm = x86.ALEAL
			case ssa.OpAMD64ADDW:
				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, symmetric
	case ssa.OpAMD64ADDB, ssa.OpAMD64ADDSS, ssa.OpAMD64ADDSD,
		ssa.OpAMD64ANDQ, ssa.OpAMD64ANDL, ssa.OpAMD64ANDW, ssa.OpAMD64ANDB,
		ssa.OpAMD64ORQ, ssa.OpAMD64ORL, ssa.OpAMD64ORW, ssa.OpAMD64ORB,
		ssa.OpAMD64XORQ, ssa.OpAMD64XORL, ssa.OpAMD64XORW, ssa.OpAMD64XORB,
		ssa.OpAMD64MULQ, ssa.OpAMD64MULL, ssa.OpAMD64MULW, ssa.OpAMD64MULB,
		ssa.OpAMD64MULSS, ssa.OpAMD64MULSD, ssa.OpAMD64PXOR:
		r := gc.SSARegNum(v)
		x := gc.SSARegNum(v.Args[0])
		y := gc.SSARegNum(v.Args[1])
		if x != r && y != r {
			opregreg(moveByType(v.Type), r, x)
			x = r
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
		if x == r {
			p.From.Reg = y
		} else {
			p.From.Reg = x
		}
	// 2-address opcode arithmetic, not symmetric
	case ssa.OpAMD64SUBQ, ssa.OpAMD64SUBL, ssa.OpAMD64SUBW, ssa.OpAMD64SUBB:
		r := gc.SSARegNum(v)
		x := gc.SSARegNum(v.Args[0])
		y := gc.SSARegNum(v.Args[1])
		var neg bool
		if y == r {
			// compute -(y-x) instead
			x, y = y, x
			neg = true
		}
		if x != r {
			opregreg(moveByType(v.Type), r, x)
		}
		opregreg(v.Op.Asm(), r, y)

		if neg {
			if v.Op == ssa.OpAMD64SUBQ {
				p := gc.Prog(x86.ANEGQ)
				p.To.Type = obj.TYPE_REG
				p.To.Reg = r
			} else { // Avoids partial registers write
				p := gc.Prog(x86.ANEGL)
				p.To.Type = obj.TYPE_REG
				p.To.Reg = r
			}
		}
	case ssa.OpAMD64SUBSS, ssa.OpAMD64SUBSD, ssa.OpAMD64DIVSS, ssa.OpAMD64DIVSD:
		r := gc.SSARegNum(v)
		x := gc.SSARegNum(v.Args[0])
		y := gc.SSARegNum(v.Args[1])
		if y == r && x != r {
			// r/y := x op r/y, need to preserve x and rewrite to
			// r/y := r/y op x15
			x15 := int16(x86.REG_X15)
			// register move y to x15
			// register move x to y
			// rename y with x15
			opregreg(moveByType(v.Type), x15, y)
			opregreg(moveByType(v.Type), r, x)
			y = x15
		} else if x != r {
			opregreg(moveByType(v.Type), r, x)
		}
		opregreg(v.Op.Asm(), r, y)

	case ssa.OpAMD64DIVQ, ssa.OpAMD64DIVL, ssa.OpAMD64DIVW,
		ssa.OpAMD64DIVQU, ssa.OpAMD64DIVLU, ssa.OpAMD64DIVWU,
		ssa.OpAMD64MODQ, ssa.OpAMD64MODL, ssa.OpAMD64MODW,
		ssa.OpAMD64MODQU, ssa.OpAMD64MODLU, ssa.OpAMD64MODWU:

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

		// CPU faults upon signed overflow, which occurs when most
		// negative int is divided by -1.
		var j *obj.Prog
		if v.Op == ssa.OpAMD64DIVQ || v.Op == ssa.OpAMD64DIVL ||
			v.Op == ssa.OpAMD64DIVW || v.Op == ssa.OpAMD64MODQ ||
			v.Op == ssa.OpAMD64MODL || v.Op == ssa.OpAMD64MODW {

			var c *obj.Prog
			switch v.Op {
			case ssa.OpAMD64DIVQ, ssa.OpAMD64MODQ:
				c = gc.Prog(x86.ACMPQ)
				j = gc.Prog(x86.AJEQ)
				// go ahead and sign extend to save doing it later
				gc.Prog(x86.ACQO)

			case ssa.OpAMD64DIVL, ssa.OpAMD64MODL:
				c = gc.Prog(x86.ACMPL)
				j = gc.Prog(x86.AJEQ)
				gc.Prog(x86.ACDQ)

			case ssa.OpAMD64DIVW, ssa.OpAMD64MODW:
				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.OpAMD64DIVQU || v.Op == ssa.OpAMD64MODQU ||
			v.Op == ssa.OpAMD64DIVLU || v.Op == ssa.OpAMD64MODLU ||
			v.Op == ssa.OpAMD64DIVWU || v.Op == ssa.OpAMD64MODWU {
			c := gc.Prog(x86.AXORQ)
			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.OpAMD64DIVQ || v.Op == ssa.OpAMD64DIVL ||
				v.Op == ssa.OpAMD64DIVW {
				// n * -1 = -n
				n = gc.Prog(x86.ANEGQ)
				n.To.Type = obj.TYPE_REG
				n.To.Reg = x86.REG_AX
			} else {
				// n % -1 == 0
				n = gc.Prog(x86.AXORQ)
				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.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 = gc.SSARegNum(v.Args[1])

		// 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.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 := gc.SSARegNum(v)
		x := gc.SSARegNum(v.Args[0])
		y := gc.SSARegNum(v.Args[1])
		if x != r && y != r {
			opregreg(moveByType(v.Type), r, x)
			x = r
		}
		p := gc.Prog(x86.AADDQ)
		p.From.Type = obj.TYPE_REG
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
		if x == r {
			p.From.Reg = y
		} else {
			p.From.Reg = x
		}
		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.OpAMD64SHLQ, ssa.OpAMD64SHLL, ssa.OpAMD64SHLW, ssa.OpAMD64SHLB,
		ssa.OpAMD64SHRQ, ssa.OpAMD64SHRL, ssa.OpAMD64SHRW, ssa.OpAMD64SHRB,
		ssa.OpAMD64SARQ, ssa.OpAMD64SARL, ssa.OpAMD64SARW, ssa.OpAMD64SARB:
		x := gc.SSARegNum(v.Args[0])
		r := gc.SSARegNum(v)
		if x != r {
			if r == x86.REG_CX {
				v.Fatalf("can't implement %s, target and shift both in CX", v.LongString())
			}
			p := gc.Prog(moveByType(v.Type))
			p.From.Type = obj.TYPE_REG
			p.From.Reg = x
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[1]) // should be CX
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpAMD64ADDQconst, ssa.OpAMD64ADDLconst, ssa.OpAMD64ADDWconst:
		r := gc.SSARegNum(v)
		a := gc.SSARegNum(v.Args[0])
		if r == a {
			if v.AuxInt2Int64() == 1 {
				var asm obj.As
				switch v.Op {
				// 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.
				case ssa.OpAMD64ADDQconst:
					asm = x86.AINCQ
				case ssa.OpAMD64ADDLconst:
					asm = x86.AINCL
				case ssa.OpAMD64ADDWconst:
					asm = x86.AINCL
				}
				p := gc.Prog(asm)
				p.To.Type = obj.TYPE_REG
				p.To.Reg = r
				return
			} else if v.AuxInt2Int64() == -1 {
				var asm obj.As
				switch v.Op {
				case ssa.OpAMD64ADDQconst:
					asm = x86.ADECQ
				case ssa.OpAMD64ADDLconst:
					asm = x86.ADECL
				case ssa.OpAMD64ADDWconst:
					asm = x86.ADECL
				}
				p := gc.Prog(asm)
				p.To.Type = obj.TYPE_REG
				p.To.Reg = r
				return
			} else {
				p := gc.Prog(v.Op.Asm())
				p.From.Type = obj.TYPE_CONST
				p.From.Offset = v.AuxInt2Int64()
				p.To.Type = obj.TYPE_REG
				p.To.Reg = r
				return
			}
		}
		var asm obj.As
		switch v.Op {
		case ssa.OpAMD64ADDQconst:
			asm = x86.ALEAQ
		case ssa.OpAMD64ADDLconst:
			asm = x86.ALEAL
		case ssa.OpAMD64ADDWconst:
			asm = x86.ALEAL
		}
		p := gc.Prog(asm)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = a
		p.From.Offset = v.AuxInt2Int64()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r

	case ssa.OpAMD64CMOVQEQconst, ssa.OpAMD64CMOVLEQconst, ssa.OpAMD64CMOVWEQconst,
		ssa.OpAMD64CMOVQNEconst, ssa.OpAMD64CMOVLNEconst, ssa.OpAMD64CMOVWNEconst:
		r := gc.SSARegNum(v)
		x := gc.SSARegNum(v.Args[0])
		// Arg0 is in/out, move in to out if not already same
		if r != x {
			p := gc.Prog(moveByType(v.Type))
			p.From.Type = obj.TYPE_REG
			p.From.Reg = x
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		}

		// Constant into AX, after arg0 movement in case arg0 is in AX
		p := gc.Prog(moveByType(v.Type))
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt2Int64()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = x86.REG_AX

		p = gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = x86.REG_AX
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r

	case ssa.OpAMD64MULQconst, ssa.OpAMD64MULLconst, ssa.OpAMD64MULWconst, ssa.OpAMD64MULBconst:
		r := gc.SSARegNum(v)
		x := gc.SSARegNum(v.Args[0])
		if r != x {
			p := gc.Prog(moveByType(v.Type))
			p.From.Type = obj.TYPE_REG
			p.From.Reg = x
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt2Int64()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
		// TODO: Teach doasm to compile the three-address multiply imul $c, r1, r2
		// instead of using the MOVQ above.
		//p.From3 = new(obj.Addr)
		//p.From3.Type = obj.TYPE_REG
		//p.From3.Reg = gc.SSARegNum(v.Args[0])
	case ssa.OpAMD64SUBQconst, ssa.OpAMD64SUBLconst, ssa.OpAMD64SUBWconst:
		x := gc.SSARegNum(v.Args[0])
		r := gc.SSARegNum(v)
		// We have 3-op add (lea), so transforming a = b - const into
		// a = b + (- const), saves us 1 instruction. We can't fit
		// - (-1 << 31) into  4 bytes offset in lea.
		// We handle 2-address just fine below.
		if v.AuxInt2Int64() == -1<<31 || x == r {
			if x != r {
				// This code compensates for the fact that the register allocator
				// doesn't understand 2-address instructions yet. TODO: fix that.
				p := gc.Prog(moveByType(v.Type))
				p.From.Type = obj.TYPE_REG
				p.From.Reg = x
				p.To.Type = obj.TYPE_REG
				p.To.Reg = r
			}
			p := gc.Prog(v.Op.Asm())
			p.From.Type = obj.TYPE_CONST
			p.From.Offset = v.AuxInt2Int64()
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		} else if x == r && v.AuxInt2Int64() == -1 {
			var asm obj.As
			// x = x - (-1) is the same as x++
			// See OpAMD64ADDQconst comments about inc vs add $1,reg
			switch v.Op {
			case ssa.OpAMD64SUBQconst:
				asm = x86.AINCQ
			case ssa.OpAMD64SUBLconst:
				asm = x86.AINCL
			case ssa.OpAMD64SUBWconst:
				asm = x86.AINCL
			}
			p := gc.Prog(asm)
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		} else if x == r && v.AuxInt2Int64() == 1 {
			var asm obj.As
			switch v.Op {
			case ssa.OpAMD64SUBQconst:
				asm = x86.ADECQ
			case ssa.OpAMD64SUBLconst:
				asm = x86.ADECL
			case ssa.OpAMD64SUBWconst:
				asm = x86.ADECL
			}
			p := gc.Prog(asm)
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		} else {
			var asm obj.As
			switch v.Op {
			case ssa.OpAMD64SUBQconst:
				asm = x86.ALEAQ
			case ssa.OpAMD64SUBLconst:
				asm = x86.ALEAL
			case ssa.OpAMD64SUBWconst:
				asm = x86.ALEAL
			}
			p := gc.Prog(asm)
			p.From.Type = obj.TYPE_MEM
			p.From.Reg = x
			p.From.Offset = -v.AuxInt2Int64()
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		}

	case ssa.OpAMD64ADDBconst,
		ssa.OpAMD64ANDQconst, ssa.OpAMD64ANDLconst, ssa.OpAMD64ANDWconst, ssa.OpAMD64ANDBconst,
		ssa.OpAMD64ORQconst, ssa.OpAMD64ORLconst, ssa.OpAMD64ORWconst, ssa.OpAMD64ORBconst,
		ssa.OpAMD64XORQconst, ssa.OpAMD64XORLconst, ssa.OpAMD64XORWconst, ssa.OpAMD64XORBconst,
		ssa.OpAMD64SUBBconst, ssa.OpAMD64SHLQconst, ssa.OpAMD64SHLLconst, ssa.OpAMD64SHLWconst,
		ssa.OpAMD64SHLBconst, ssa.OpAMD64SHRQconst, ssa.OpAMD64SHRLconst, ssa.OpAMD64SHRWconst,
		ssa.OpAMD64SHRBconst, ssa.OpAMD64SARQconst, ssa.OpAMD64SARLconst, ssa.OpAMD64SARWconst,
		ssa.OpAMD64SARBconst, ssa.OpAMD64ROLQconst, ssa.OpAMD64ROLLconst, ssa.OpAMD64ROLWconst,
		ssa.OpAMD64ROLBconst:
		// This code compensates for the fact that the register allocator
		// doesn't understand 2-address instructions yet. TODO: fix that.
		x := gc.SSARegNum(v.Args[0])
		r := gc.SSARegNum(v)
		if x != r {
			p := gc.Prog(moveByType(v.Type))
			p.From.Type = obj.TYPE_REG
			p.From.Reg = x
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt2Int64()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpAMD64SBBQcarrymask, ssa.OpAMD64SBBLcarrymask:
		r := gc.SSARegNum(v)
		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:
		p := gc.Prog(x86.ALEAQ)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		switch v.Op {
		case ssa.OpAMD64LEAQ1:
			p.From.Scale = 1
		case ssa.OpAMD64LEAQ2:
			p.From.Scale = 2
		case ssa.OpAMD64LEAQ4:
			p.From.Scale = 4
		case ssa.OpAMD64LEAQ8:
			p.From.Scale = 8
		}
		p.From.Index = gc.SSARegNum(v.Args[1])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpAMD64LEAQ:
		p := gc.Prog(x86.ALEAQ)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpAMD64CMPQ, ssa.OpAMD64CMPL, ssa.OpAMD64CMPW, ssa.OpAMD64CMPB,
		ssa.OpAMD64TESTQ, ssa.OpAMD64TESTL, ssa.OpAMD64TESTW, ssa.OpAMD64TESTB:
		opregreg(v.Op.Asm(), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v.Args[0]))
	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(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]))
	case ssa.OpAMD64CMPQconst, ssa.OpAMD64CMPLconst, ssa.OpAMD64CMPWconst, ssa.OpAMD64CMPBconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_CONST
		p.To.Offset = v.AuxInt2Int64()
	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.AuxInt2Int64()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v.Args[0])
	case ssa.OpAMD64MOVBconst, ssa.OpAMD64MOVWconst, ssa.OpAMD64MOVLconst, ssa.OpAMD64MOVQconst:
		x := gc.SSARegNum(v)
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt2Int64()
		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 := gc.SSARegNum(v)
		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 = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpAMD64MOVQloadidx1:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.From.Scale = 1
		p.From.Index = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpAMD64MOVQloadidx8, ssa.OpAMD64MOVSDloadidx8:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.From.Scale = 8
		p.From.Index = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpAMD64MOVLloadidx1:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.From.Scale = 1
		p.From.Index = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpAMD64MOVLloadidx4, ssa.OpAMD64MOVSSloadidx4:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.From.Scale = 4
		p.From.Index = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpAMD64MOVWloadidx1:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.From.Scale = 1
		p.From.Index = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpAMD64MOVWloadidx2:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.From.Scale = 2
		p.From.Index = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpAMD64MOVBloadidx1:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.From.Scale = 1
		p.From.Index = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	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 = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		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 = gc.SSARegNum(v.Args[2])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Scale = 8
		p.To.Index = gc.SSARegNum(v.Args[1])
		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 = gc.SSARegNum(v.Args[2])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Scale = 4
		p.To.Index = gc.SSARegNum(v.Args[1])
		gc.AddAux(&p.To, v)
	case ssa.OpAMD64MOVWstoreidx2:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[2])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Scale = 2
		p.To.Index = gc.SSARegNum(v.Args[1])
		gc.AddAux(&p.To, v)
	case ssa.OpAMD64MOVBstoreidx1:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[2])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Scale = 1
		p.To.Index = gc.SSARegNum(v.Args[1])
		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()
		i := sc.Val()
		switch v.Op {
		case ssa.OpAMD64MOVBstoreconst:
			i = int64(int8(i))
		case ssa.OpAMD64MOVWstoreconst:
			i = int64(int16(i))
		case ssa.OpAMD64MOVLstoreconst:
			i = int64(int32(i))
		case ssa.OpAMD64MOVQstoreconst:
		}
		p.From.Offset = i
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux2(&p.To, v, sc.Off())
	case ssa.OpAMD64MOVQstoreconstidx8, ssa.OpAMD64MOVLstoreconstidx4, ssa.OpAMD64MOVWstoreconstidx2, ssa.OpAMD64MOVBstoreconstidx1:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		sc := v.AuxValAndOff()
		switch v.Op {
		case ssa.OpAMD64MOVBstoreconstidx1:
			p.From.Offset = int64(int8(sc.Val()))
			p.To.Scale = 1
		case ssa.OpAMD64MOVWstoreconstidx2:
			p.From.Offset = int64(int16(sc.Val()))
			p.To.Scale = 2
		case ssa.OpAMD64MOVLstoreconstidx4:
			p.From.Offset = int64(int32(sc.Val()))
			p.To.Scale = 4
		case ssa.OpAMD64MOVQstoreconstidx8:
			p.From.Offset = sc.Val()
			p.To.Scale = 8
		}
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Index = gc.SSARegNum(v.Args[1])
		gc.AddAux2(&p.To, v, sc.Off())
	case ssa.OpAMD64MOVLQSX, ssa.OpAMD64MOVWQSX, ssa.OpAMD64MOVBQSX, ssa.OpAMD64MOVLQZX, ssa.OpAMD64MOVWQZX, ssa.OpAMD64MOVBQZX,
		ssa.OpAMD64CVTSL2SS, ssa.OpAMD64CVTSL2SD, ssa.OpAMD64CVTSQ2SS, ssa.OpAMD64CVTSQ2SD,
		ssa.OpAMD64CVTTSS2SL, ssa.OpAMD64CVTTSD2SL, ssa.OpAMD64CVTTSS2SQ, ssa.OpAMD64CVTTSD2SQ,
		ssa.OpAMD64CVTSS2SD, ssa.OpAMD64CVTSD2SS:
		opregreg(v.Op.Asm(), gc.SSARegNum(v), gc.SSARegNum(v.Args[0]))
	case ssa.OpAMD64DUFFZERO:
		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.OpAMD64MOVOconst:
		if v.AuxInt != 0 {
			v.Unimplementedf("MOVOconst can only do constant=0")
		}
		r := gc.SSARegNum(v)
		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: // TODO: use MOVQreg for reg->reg copies instead of OpCopy?
		if v.Type.IsMemory() {
			return
		}
		x := gc.SSARegNum(v.Args[0])
		y := gc.SSARegNum(v)
		if x != y {
			opregreg(moveByType(v.Type), y, x)
		}
	case ssa.OpLoadReg:
		if v.Type.IsFlags() {
			v.Unimplementedf("load flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(loadByType(v.Type))
		n, off := gc.AutoVar(v.Args[0])
		p.From.Type = obj.TYPE_MEM
		p.From.Node = n
		p.From.Sym = gc.Linksym(n.Sym)
		p.From.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.From.Name = obj.NAME_PARAM
			p.From.Offset += n.Xoffset
		} else {
			p.From.Name = obj.NAME_AUTO
		}
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

	case ssa.OpStoreReg:
		if v.Type.IsFlags() {
			v.Unimplementedf("store flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(storeByType(v.Type))
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		n, off := gc.AutoVar(v)
		p.To.Type = obj.TYPE_MEM
		p.To.Node = n
		p.To.Sym = gc.Linksym(n.Sym)
		p.To.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.To.Name = obj.NAME_PARAM
			p.To.Offset += n.Xoffset
		} else {
			p.To.Name = obj.NAME_AUTO
		}
	case ssa.OpPhi:
		// just check to make sure regalloc and stackalloc did it right
		if v.Type.IsMemory() {
			return
		}
		f := v.Block.Func
		loc := f.RegAlloc[v.ID]
		for _, a := range v.Args {
			if aloc := f.RegAlloc[a.ID]; aloc != loc { // TODO: .Equal() instead?
				v.Fatalf("phi arg at different location than phi: %v @ %v, but arg %v @ %v\n%s\n", v, loc, a, aloc, v.Block.Func)
			}
		}
	case ssa.OpInitMem:
		// memory arg needs no code
	case ssa.OpArg:
		// input args need no code
	case ssa.OpAMD64LoweredGetClosurePtr:
		// Output is hardwired to DX only,
		// and DX contains the closure pointer on
		// closure entry, and this "instruction"
		// is scheduled to the very beginning
		// of the entry block.
	case ssa.OpAMD64LoweredGetG:
		r := gc.SSARegNum(v)
		// 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 = gc.SSARegNum(v.Args[0])
		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 = gc.SSARegNum(v.Args[0])
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpAMD64NEGQ, ssa.OpAMD64NEGL, ssa.OpAMD64NEGW, ssa.OpAMD64NEGB,
		ssa.OpAMD64BSWAPQ, ssa.OpAMD64BSWAPL,
		ssa.OpAMD64NOTQ, ssa.OpAMD64NOTL, ssa.OpAMD64NOTW, ssa.OpAMD64NOTB:
		x := gc.SSARegNum(v.Args[0])
		r := gc.SSARegNum(v)
		if x != r {
			p := gc.Prog(moveByType(v.Type))
			p.From.Type = obj.TYPE_REG
			p.From.Reg = x
			p.To.Type = obj.TYPE_REG
			p.To.Reg = r
		}
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpAMD64BSFQ, ssa.OpAMD64BSFL, ssa.OpAMD64BSFW,
		ssa.OpAMD64BSRQ, ssa.OpAMD64BSRL, ssa.OpAMD64BSRW,
		ssa.OpAMD64SQRTSD:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpSP, ssa.OpSB:
		// 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 = gc.SSARegNum(v)

	case ssa.OpAMD64SETNEF:
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
		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, gc.SSARegNum(v), x86.REG_AX)

	case ssa.OpAMD64SETEQF:
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
		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, gc.SSARegNum(v), 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.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.OpAMD64LoweredNilCheck:
		// Optimization - if the subsequent block has a load or store
		// at the same address, we don't need to issue this instruction.
		mem := v.Args[1]
		for _, w := range v.Block.Succs[0].Values {
			if w.Op == ssa.OpPhi {
				if w.Type.IsMemory() {
					mem = w
				}
				continue
			}
			if len(w.Args) == 0 || !w.Args[len(w.Args)-1].Type.IsMemory() {
				// w doesn't use a store - can't be a memory op.
				continue
			}
			if w.Args[len(w.Args)-1] != mem {
				v.Fatalf("wrong store after nilcheck v=%s w=%s", v, w)
			}
			switch w.Op {
			case ssa.OpAMD64MOVQload, ssa.OpAMD64MOVLload, ssa.OpAMD64MOVWload, ssa.OpAMD64MOVBload,
				ssa.OpAMD64MOVQstore, ssa.OpAMD64MOVLstore, ssa.OpAMD64MOVWstore, ssa.OpAMD64MOVBstore,
				ssa.OpAMD64MOVBQSXload, ssa.OpAMD64MOVWQSXload, ssa.OpAMD64MOVLQSXload,
				ssa.OpAMD64MOVSSload, ssa.OpAMD64MOVSDload, ssa.OpAMD64MOVOload,
				ssa.OpAMD64MOVSSstore, ssa.OpAMD64MOVSDstore, ssa.OpAMD64MOVOstore:
				if w.Args[0] == v.Args[0] && w.Aux == nil && w.AuxInt >= 0 && w.AuxInt < minZeroPage {
					if gc.Debug_checknil != 0 && int(v.Line) > 1 {
						gc.Warnl(v.Line, "removed nil check")
					}
					return
				}
			case ssa.OpAMD64MOVQstoreconst, ssa.OpAMD64MOVLstoreconst, ssa.OpAMD64MOVWstoreconst, ssa.OpAMD64MOVBstoreconst:
				off := ssa.ValAndOff(v.AuxInt).Off()
				if w.Args[0] == v.Args[0] && w.Aux == nil && off >= 0 && off < minZeroPage {
					if gc.Debug_checknil != 0 && int(v.Line) > 1 {
						gc.Warnl(v.Line, "removed nil check")
					}
					return
				}
			}
			if w.Type.IsMemory() {
				if w.Op == ssa.OpVarDef || w.Op == ssa.OpVarKill || w.Op == ssa.OpVarLive {
					// these ops are OK
					mem = w
					continue
				}
				// We can't delay the nil check past the next store.
				break
			}
		}
		// 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 = gc.SSARegNum(v.Args[0])
		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")
		}
	default:
		v.Unimplementedf("genValue not implemented: %s", v.LongString())
	}
}
Beispiel #4
0
Datei: ssa.go Projekt: hurkgu/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 := gc.SSARegNum(v)
		r1 := gc.SSARegNum(v.Args[0])
		r2 := gc.SSARegNum(v.Args[1])
		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 := gc.SSARegNum(v)
		if r != gc.SSARegNum(v.Args[0]) {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		opregreg(v.Op.Asm(), r, gc.SSARegNum(v.Args[1]))

	case ssa.Op386ADDLcarry, ssa.Op386SUBLcarry:
		// output 0 is carry/borrow, output 1 is the low 32 bits.
		r := gc.SSARegNum1(v)
		if r != gc.SSARegNum(v.Args[0]) {
			v.Fatalf("input[0] and output[1] not in same register %s", v.LongString())
		}
		opregreg(v.Op.Asm(), r, gc.SSARegNum(v.Args[1]))

	case ssa.Op386ADDLconstcarry, ssa.Op386SUBLconstcarry:
		// output 0 is carry/borrow, output 1 is the low 32 bits.
		r := gc.SSARegNum1(v)
		if r != gc.SSARegNum(v.Args[0]) {
			v.Fatalf("input[0] and output[1] 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 := gc.SSARegNum(v.Args[1])

		// 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 = gc.SSARegNum(v.Args[1])

		// 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 = gc.SSARegNum(v.Args[1])

	case ssa.Op386ADDLconst:
		r := gc.SSARegNum(v)
		a := gc.SSARegNum(v.Args[0])
		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 := gc.SSARegNum(v)
		if r != gc.SSARegNum(v.Args[0]) {
			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 = gc.SSARegNum(v.Args[0])

	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 := gc.SSARegNum(v)
		if r != gc.SSARegNum(v.Args[0]) {
			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 := gc.SSARegNum(v)
		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 := gc.SSARegNum(v.Args[0])
		i := gc.SSARegNum(v.Args[1])
		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 = gc.SSARegNum(v)
	case ssa.Op386LEAL:
		p := gc.Prog(x86.ALEAL)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.Op386CMPL, ssa.Op386CMPW, ssa.Op386CMPB,
		ssa.Op386TESTL, ssa.Op386TESTW, ssa.Op386TESTB:
		opregreg(v.Op.Asm(), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v.Args[0]))
	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(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]))
	case ssa.Op386CMPLconst, ssa.Op386CMPWconst, ssa.Op386CMPBconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		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 = gc.SSARegNum(v.Args[0])
	case ssa.Op386MOVLconst:
		x := gc.SSARegNum(v)
		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 $0,AX -> XOR AX,AX optimization.
		if v.Aux != nil {
			p.Mark |= x86.PRESERVEFLAGS
		}
	case ssa.Op386MOVSSconst, ssa.Op386MOVSDconst:
		x := gc.SSARegNum(v)
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_FCONST
		p.From.Val = math.Float64frombits(uint64(v.AuxInt))
		p.To.Type = obj.TYPE_REG
		p.To.Reg = x
	case ssa.Op386MOVSSconst1, ssa.Op386MOVSDconst1:
		var literal string
		if v.Op == ssa.Op386MOVSDconst1 {
			literal = fmt.Sprintf("$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.Local = true
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.Op386MOVSSconst2, ssa.Op386MOVSDconst2:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

	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 = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.Op386MOVSDloadidx8:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.From.Scale = 8
		p.From.Index = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.Op386MOVLloadidx4, ssa.Op386MOVSSloadidx4:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.From.Scale = 4
		p.From.Index = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.Op386MOVWloadidx2:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.From.Scale = 2
		p.From.Index = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.Op386MOVBloadidx1, ssa.Op386MOVWloadidx1, ssa.Op386MOVLloadidx1, ssa.Op386MOVSSloadidx1, ssa.Op386MOVSDloadidx1:
		r := gc.SSARegNum(v.Args[0])
		i := gc.SSARegNum(v.Args[1])
		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 = gc.SSARegNum(v)
	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 = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.To, v)
	case ssa.Op386MOVSDstoreidx8:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[2])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Scale = 8
		p.To.Index = gc.SSARegNum(v.Args[1])
		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 = gc.SSARegNum(v.Args[2])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Scale = 4
		p.To.Index = gc.SSARegNum(v.Args[1])
		gc.AddAux(&p.To, v)
	case ssa.Op386MOVWstoreidx2:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[2])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Scale = 2
		p.To.Index = gc.SSARegNum(v.Args[1])
		gc.AddAux(&p.To, v)
	case ssa.Op386MOVBstoreidx1, ssa.Op386MOVWstoreidx1, ssa.Op386MOVLstoreidx1, ssa.Op386MOVSSstoreidx1, ssa.Op386MOVSDstoreidx1:
		r := gc.SSARegNum(v.Args[0])
		i := gc.SSARegNum(v.Args[1])
		if i == x86.REG_SP {
			r, i = i, r
		}
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[2])
		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 = gc.SSARegNum(v.Args[0])
		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 := gc.SSARegNum(v.Args[0])
		i := gc.SSARegNum(v.Args[1])
		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(), gc.SSARegNum(v), gc.SSARegNum(v.Args[0]))
	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 := gc.SSARegNum(v.Args[0])
		y := gc.SSARegNum(v)
		if x != y {
			opregreg(moveByType(v.Type), y, x)
		}
	case ssa.OpLoadReg:
		if v.Type.IsFlags() {
			v.Unimplementedf("load flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(loadByType(v.Type))
		n, off := gc.AutoVar(v.Args[0])
		p.From.Type = obj.TYPE_MEM
		p.From.Node = n
		p.From.Sym = gc.Linksym(n.Sym)
		p.From.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.From.Name = obj.NAME_PARAM
			p.From.Offset += n.Xoffset
		} else {
			p.From.Name = obj.NAME_AUTO
		}
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

	case ssa.OpStoreReg:
		if v.Type.IsFlags() {
			v.Unimplementedf("store flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(storeByType(v.Type))
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		n, off := gc.AutoVar(v)
		p.To.Type = obj.TYPE_MEM
		p.To.Node = n
		p.To.Sym = gc.Linksym(n.Sym)
		p.To.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.To.Name = obj.NAME_PARAM
			p.To.Offset += n.Xoffset
		} else {
			p.To.Name = obj.NAME_AUTO
		}
	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 := gc.SSARegNum(v)
		// 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 = gc.SSARegNum(v.Args[0])
		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 = gc.SSARegNum(v.Args[0])
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.Op386NEGL,
		ssa.Op386BSWAPL,
		ssa.Op386NOTL:
		r := gc.SSARegNum(v)
		if r != gc.SSARegNum(v.Args[0]) {
			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 = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	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 = gc.SSARegNum(v)

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

	case ssa.Op386SETEQF:
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
		q := gc.Prog(x86.ASETPC)
		q.To.Type = obj.TYPE_REG
		q.To.Reg = x86.REG_AX
		opregreg(x86.AANDL, gc.SSARegNum(v), 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:
		if !v.Args[0].Type.IsPtrShaped() {
			v.Fatalf("keeping non-pointer alive %v", v.Args[0])
		}
		n, off := gc.AutoVar(v.Args[0])
		if n == nil {
			v.Fatalf("KeepLive with non-spilled value %s %s", v, v.Args[0])
		}
		if off != 0 {
			v.Fatalf("KeepLive with non-zero offset spill location %s:%d", n, off)
		}
		gc.Gvarlive(n)
	case ssa.Op386LoweredNilCheck:
		// Optimization - if the subsequent block has a load or store
		// at the same address, we don't need to issue this instruction.
		mem := v.Args[1]
		for _, w := range v.Block.Succs[0].Block().Values {
			if w.Op == ssa.OpPhi {
				if w.Type.IsMemory() {
					mem = w
				}
				continue
			}
			if len(w.Args) == 0 || !w.Args[len(w.Args)-1].Type.IsMemory() {
				// w doesn't use a store - can't be a memory op.
				continue
			}
			if w.Args[len(w.Args)-1] != mem {
				v.Fatalf("wrong store after nilcheck v=%s w=%s", v, w)
			}
			switch w.Op {
			case ssa.Op386MOVLload, ssa.Op386MOVWload, ssa.Op386MOVBload,
				ssa.Op386MOVLstore, ssa.Op386MOVWstore, ssa.Op386MOVBstore,
				ssa.Op386MOVBLSXload, ssa.Op386MOVWLSXload,
				ssa.Op386MOVSSload, ssa.Op386MOVSDload,
				ssa.Op386MOVSSstore, ssa.Op386MOVSDstore:
				if w.Args[0] == v.Args[0] && w.Aux == nil && w.AuxInt >= 0 && w.AuxInt < minZeroPage {
					if gc.Debug_checknil != 0 && int(v.Line) > 1 {
						gc.Warnl(v.Line, "removed nil check")
					}
					return
				}
			case ssa.Op386MOVLstoreconst, ssa.Op386MOVWstoreconst, ssa.Op386MOVBstoreconst:
				off := ssa.ValAndOff(v.AuxInt).Off()
				if w.Args[0] == v.Args[0] && w.Aux == nil && off >= 0 && off < minZeroPage {
					if gc.Debug_checknil != 0 && int(v.Line) > 1 {
						gc.Warnl(v.Line, "removed nil check")
					}
					return
				}
			}
			if w.Type.IsMemory() {
				if w.Op == ssa.OpVarDef || w.Op == ssa.OpVarKill || w.Op == ssa.OpVarLive {
					// these ops are OK
					mem = w
					continue
				}
				// We can't delay the nil check past the next store.
				break
			}
		}
		// 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 = gc.SSARegNum(v.Args[0])
		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.Unimplementedf("genValue not implemented: %s", v.LongString())
	}
}
Beispiel #5
0
Datei: ssa.go Projekt: hurkgu/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.OpMIPS64MOVVconvert, ssa.OpMIPS64MOVVreg:
		if v.Type.IsMemory() {
			return
		}
		x := gc.SSARegNum(v.Args[0])
		y := gc.SSARegNum(v)
		if x == y {
			return
		}
		as := mips.AMOVV
		if v.Type.IsFloat() {
			switch v.Type.Size() {
			case 4:
				as = mips.AMOVF
			case 8:
				as = mips.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.OpMIPS64MOVVnop:
		if gc.SSARegNum(v) != gc.SSARegNum(v.Args[0]) {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		// nothing to do
	case ssa.OpLoadReg:
		if v.Type.IsFlags() {
			v.Unimplementedf("load flags not implemented: %v", v.LongString())
			return
		}
		r := gc.SSARegNum(v)
		p := gc.Prog(loadByType(v.Type, r))
		n, off := gc.AutoVar(v.Args[0])
		p.From.Type = obj.TYPE_MEM
		p.From.Node = n
		p.From.Sym = gc.Linksym(n.Sym)
		p.From.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.From.Name = obj.NAME_PARAM
			p.From.Offset += n.Xoffset
		} else {
			p.From.Name = obj.NAME_AUTO
		}
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpPhi:
		gc.CheckLoweredPhi(v)
	case ssa.OpStoreReg:
		if v.Type.IsFlags() {
			v.Unimplementedf("store flags not implemented: %v", v.LongString())
			return
		}
		r := gc.SSARegNum(v.Args[0])
		p := gc.Prog(storeByType(v.Type, r))
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r
		n, off := gc.AutoVar(v)
		p.To.Type = obj.TYPE_MEM
		p.To.Node = n
		p.To.Sym = gc.Linksym(n.Sym)
		p.To.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.To.Name = obj.NAME_PARAM
			p.To.Offset += n.Xoffset
		} else {
			p.To.Name = obj.NAME_AUTO
		}
	case ssa.OpMIPS64ADDV,
		ssa.OpMIPS64SUBV,
		ssa.OpMIPS64AND,
		ssa.OpMIPS64OR,
		ssa.OpMIPS64XOR,
		ssa.OpMIPS64NOR,
		ssa.OpMIPS64SLLV,
		ssa.OpMIPS64SRLV,
		ssa.OpMIPS64SRAV,
		ssa.OpMIPS64ADDF,
		ssa.OpMIPS64ADDD,
		ssa.OpMIPS64SUBF,
		ssa.OpMIPS64SUBD,
		ssa.OpMIPS64MULF,
		ssa.OpMIPS64MULD,
		ssa.OpMIPS64DIVF,
		ssa.OpMIPS64DIVD:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[1])
		p.Reg = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpMIPS64SGT,
		ssa.OpMIPS64SGTU:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.Reg = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpMIPS64ADDVconst,
		ssa.OpMIPS64SUBVconst,
		ssa.OpMIPS64ANDconst,
		ssa.OpMIPS64ORconst,
		ssa.OpMIPS64XORconst,
		ssa.OpMIPS64NORconst,
		ssa.OpMIPS64SLLVconst,
		ssa.OpMIPS64SRLVconst,
		ssa.OpMIPS64SRAVconst,
		ssa.OpMIPS64SGTconst,
		ssa.OpMIPS64SGTUconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.Reg = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpMIPS64MULV,
		ssa.OpMIPS64MULVU,
		ssa.OpMIPS64DIVV,
		ssa.OpMIPS64DIVVU:
		// result in hi,lo
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[1])
		p.Reg = gc.SSARegNum(v.Args[0])
	case ssa.OpMIPS64MOVVconst:
		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 = gc.SSARegNum(v)
	case ssa.OpMIPS64MOVFconst,
		ssa.OpMIPS64MOVDconst:
		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 = gc.SSARegNum(v)
	case ssa.OpMIPS64CMPEQF,
		ssa.OpMIPS64CMPEQD,
		ssa.OpMIPS64CMPGEF,
		ssa.OpMIPS64CMPGED,
		ssa.OpMIPS64CMPGTF,
		ssa.OpMIPS64CMPGTD:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.Reg = gc.SSARegNum(v.Args[1])
	case ssa.OpMIPS64MOVVaddr:
		p := gc.Prog(mips.AMOVV)
		p.From.Type = obj.TYPE_ADDR
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

		var wantreg string
		// MOVV $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 MOVV $off(SP), R
			wantreg = "SP"
			p.From.Reg = mips.REGSP
			p.From.Offset = v.AuxInt
		}
		if reg := gc.SSAReg(v.Args[0]); reg.Name() != wantreg {
			v.Fatalf("bad reg %s for symbol type %T, want %s", reg.Name(), v.Aux, wantreg)
		}
	case ssa.OpMIPS64MOVBload,
		ssa.OpMIPS64MOVBUload,
		ssa.OpMIPS64MOVHload,
		ssa.OpMIPS64MOVHUload,
		ssa.OpMIPS64MOVWload,
		ssa.OpMIPS64MOVWUload,
		ssa.OpMIPS64MOVVload,
		ssa.OpMIPS64MOVFload,
		ssa.OpMIPS64MOVDload:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpMIPS64MOVBstore,
		ssa.OpMIPS64MOVHstore,
		ssa.OpMIPS64MOVWstore,
		ssa.OpMIPS64MOVVstore,
		ssa.OpMIPS64MOVFstore,
		ssa.OpMIPS64MOVDstore:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.To, v)
	case ssa.OpMIPS64MOVBstorezero,
		ssa.OpMIPS64MOVHstorezero,
		ssa.OpMIPS64MOVWstorezero,
		ssa.OpMIPS64MOVVstorezero:
		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 = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.To, v)
	case ssa.OpMIPS64MOVBreg,
		ssa.OpMIPS64MOVBUreg,
		ssa.OpMIPS64MOVHreg,
		ssa.OpMIPS64MOVHUreg,
		ssa.OpMIPS64MOVWreg,
		ssa.OpMIPS64MOVWUreg:
		// TODO: remove extension if after proper load
		fallthrough
	case ssa.OpMIPS64MOVWF,
		ssa.OpMIPS64MOVWD,
		ssa.OpMIPS64MOVFW,
		ssa.OpMIPS64MOVDW,
		ssa.OpMIPS64MOVVF,
		ssa.OpMIPS64MOVVD,
		ssa.OpMIPS64MOVFV,
		ssa.OpMIPS64MOVDV,
		ssa.OpMIPS64MOVFD,
		ssa.OpMIPS64MOVDF,
		ssa.OpMIPS64NEGF,
		ssa.OpMIPS64NEGD:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpMIPS64NEGV:
		// SUB from REGZERO
		p := gc.Prog(mips.ASUBVU)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.Reg = mips.REGZERO
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpMIPS64CALLstatic:
		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.OpMIPS64CALLclosure:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Offset = 0
		p.To.Reg = gc.SSARegNum(v.Args[0])
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpMIPS64CALLdefer:
		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.OpMIPS64CALLgo:
		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.OpMIPS64CALLinter:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Offset = 0
		p.To.Reg = gc.SSARegNum(v.Args[0])
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpMIPS64LoweredNilCheck:
		// TODO: optimization
		// Issue a load which will fault if arg is nil.
		p := gc.Prog(mips.AMOVB)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = mips.REGZERO
		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:
		if !v.Args[0].Type.IsPtrShaped() {
			v.Fatalf("keeping non-pointer alive %v", v.Args[0])
		}
		n, off := gc.AutoVar(v.Args[0])
		if n == nil {
			v.Fatalf("KeepLive with non-spilled value %s %s", v, v.Args[0])
		}
		if off != 0 {
			v.Fatalf("KeepLive with non-zero offset spill location %s:%d", n, off)
		}
		gc.Gvarlive(n)
	case ssa.OpMIPS64FPFlagTrue,
		ssa.OpMIPS64FPFlagFalse:
		// MOVV	$0, r
		// BFPF	2(PC)
		// MOVV	$1, r
		branch := mips.ABFPF
		if v.Op == ssa.OpMIPS64FPFlagFalse {
			branch = mips.ABFPT
		}
		p := gc.Prog(mips.AMOVV)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = mips.REGZERO
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
		p2 := gc.Prog(branch)
		p2.To.Type = obj.TYPE_BRANCH
		p3 := gc.Prog(mips.AMOVV)
		p3.From.Type = obj.TYPE_CONST
		p3.From.Offset = 1
		p3.To.Type = obj.TYPE_REG
		p3.To.Reg = gc.SSARegNum(v)
		p4 := gc.Prog(obj.ANOP) // not a machine instruction, for branch to land
		gc.Patch(p2, p4)
	case ssa.OpSelect0, ssa.OpSelect1:
		// nothing to do
	case ssa.OpMIPS64LoweredGetClosurePtr:
		// Closure pointer is R22 (mips.REGCTXT).
		gc.CheckLoweredGetClosurePtr(v)
	default:
		v.Unimplementedf("genValue not implemented: %s", v.LongString())
	}
}
Beispiel #6
0
Datei: ssa.go Projekt: hurkgu/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 := gc.SSARegNum(v.Args[0])
		y := gc.SSARegNum(v)
		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 gc.SSARegNum(v) != gc.SSARegNum(v.Args[0]) {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		// nothing to do
	case ssa.OpLoadReg:
		if v.Type.IsFlags() {
			v.Unimplementedf("load flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(loadByType(v.Type))
		n, off := gc.AutoVar(v.Args[0])
		p.From.Type = obj.TYPE_MEM
		p.From.Node = n
		p.From.Sym = gc.Linksym(n.Sym)
		p.From.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.From.Name = obj.NAME_PARAM
			p.From.Offset += n.Xoffset
		} else {
			p.From.Name = obj.NAME_AUTO
		}
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpPhi:
		gc.CheckLoweredPhi(v)
	case ssa.OpStoreReg:
		if v.Type.IsFlags() {
			v.Unimplementedf("store flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(storeByType(v.Type))
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		n, off := gc.AutoVar(v)
		p.To.Type = obj.TYPE_MEM
		p.To.Node = n
		p.To.Sym = gc.Linksym(n.Sym)
		p.To.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.To.Name = obj.NAME_PARAM
			p.To.Offset += n.Xoffset
		} else {
			p.To.Name = obj.NAME_AUTO
		}
	case ssa.OpARMDIV,
		ssa.OpARMDIVU,
		ssa.OpARMMOD,
		ssa.OpARMMODU:
		// Note: for software division the assembler rewrite these
		// instructions to sequence of instructions:
		// - it puts numerator in R11 and denominator in g.m.divmod
		//	and call (say) _udiv
		// - _udiv saves R0-R3 on stack and call udiv, restores R0-R3
		//	before return
		// - udiv does the actual work
		//TODO: set approperiate regmasks and call udiv directly?
		// need to be careful for negative case
		// Or, as soft div is already expensive, we don't care?
		fallthrough
	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 := gc.SSARegNum(v)
		r1 := gc.SSARegNum(v.Args[0])
		r2 := gc.SSARegNum(v.Args[1])
		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 := gc.SSARegNum1(v)
		r1 := gc.SSARegNum(v.Args[0])
		r2 := gc.SSARegNum(v.Args[1])
		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 := gc.SSARegNum(v)
		r1 := gc.SSARegNum(v.Args[0])
		r2 := gc.SSARegNum(v.Args[1])
		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 := gc.SSARegNum(v)
		r1 := gc.SSARegNum(v.Args[0])
		r2 := gc.SSARegNum(v.Args[1])
		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 = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	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 = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum1(v)
	case ssa.OpARMSRRconst:
		genshift(arm.AMOVW, 0, gc.SSARegNum(v.Args[0]), gc.SSARegNum(v), 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(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v), arm.SHIFT_LL, v.AuxInt)
	case ssa.OpARMADDSshiftLL,
		ssa.OpARMSUBSshiftLL,
		ssa.OpARMRSBSshiftLL:
		p := genshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum1(v), 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(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v), arm.SHIFT_LR, v.AuxInt)
	case ssa.OpARMADDSshiftRL,
		ssa.OpARMSUBSshiftRL,
		ssa.OpARMRSBSshiftRL:
		p := genshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum1(v), 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(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v), arm.SHIFT_AR, v.AuxInt)
	case ssa.OpARMADDSshiftRA,
		ssa.OpARMSUBSshiftRA,
		ssa.OpARMRSBSshiftRA:
		p := genshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum1(v), arm.SHIFT_AR, v.AuxInt)
		p.Scond = arm.C_SBIT
	case ssa.OpARMMVNshiftLL:
		genshift(v.Op.Asm(), 0, gc.SSARegNum(v.Args[0]), gc.SSARegNum(v), arm.SHIFT_LL, v.AuxInt)
	case ssa.OpARMMVNshiftRL:
		genshift(v.Op.Asm(), 0, gc.SSARegNum(v.Args[0]), gc.SSARegNum(v), arm.SHIFT_LR, v.AuxInt)
	case ssa.OpARMMVNshiftRA:
		genshift(v.Op.Asm(), 0, gc.SSARegNum(v.Args[0]), gc.SSARegNum(v), arm.SHIFT_AR, v.AuxInt)
	case ssa.OpARMMVNshiftLLreg:
		genregshift(v.Op.Asm(), 0, gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v), arm.SHIFT_LL)
	case ssa.OpARMMVNshiftRLreg:
		genregshift(v.Op.Asm(), 0, gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v), arm.SHIFT_LR)
	case ssa.OpARMMVNshiftRAreg:
		genregshift(v.Op.Asm(), 0, gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v), 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(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v.Args[2]), gc.SSARegNum(v), arm.SHIFT_LL)
	case ssa.OpARMADDSshiftLLreg,
		ssa.OpARMSUBSshiftLLreg,
		ssa.OpARMRSBSshiftLLreg:
		p := genregshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v.Args[2]), gc.SSARegNum1(v), 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(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v.Args[2]), gc.SSARegNum(v), arm.SHIFT_LR)
	case ssa.OpARMADDSshiftRLreg,
		ssa.OpARMSUBSshiftRLreg,
		ssa.OpARMRSBSshiftRLreg:
		p := genregshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v.Args[2]), gc.SSARegNum1(v), 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(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v.Args[2]), gc.SSARegNum(v), arm.SHIFT_AR)
	case ssa.OpARMADDSshiftRAreg,
		ssa.OpARMSUBSshiftRAreg,
		ssa.OpARMRSBSshiftRAreg:
		p := genregshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v.Args[2]), gc.SSARegNum1(v), 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 = gc.SSARegNum(v.Args[0])
		p.Reg = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REGREG
		p.To.Reg = gc.SSARegNum(v)
		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 = gc.SSARegNum(v.Args[0])
		p.Reg = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REGREG
		p.To.Reg = gc.SSARegNum0(v)           // high 32-bit
		p.To.Offset = int64(gc.SSARegNum1(v)) // low 32-bit
	case ssa.OpARMMULA:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.Reg = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_REGREG2
		p.To.Reg = gc.SSARegNum(v)                   // result
		p.To.Offset = int64(gc.SSARegNum(v.Args[2])) // 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 = gc.SSARegNum(v)
	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 = gc.SSARegNum(v)
	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 = gc.SSARegNum(v.Args[1])
		p.Reg = gc.SSARegNum(v.Args[0])
	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 = gc.SSARegNum(v.Args[0])
	case ssa.OpARMCMPF0,
		ssa.OpARMCMPD0:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
	case ssa.OpARMCMPshiftLL:
		genshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), 0, arm.SHIFT_LL, v.AuxInt)
	case ssa.OpARMCMPshiftRL:
		genshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), 0, arm.SHIFT_LR, v.AuxInt)
	case ssa.OpARMCMPshiftRA:
		genshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), 0, arm.SHIFT_AR, v.AuxInt)
	case ssa.OpARMCMPshiftLLreg:
		genregshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v.Args[2]), 0, arm.SHIFT_LL)
	case ssa.OpARMCMPshiftRLreg:
		genregshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v.Args[2]), 0, arm.SHIFT_LR)
	case ssa.OpARMCMPshiftRAreg:
		genregshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v.Args[2]), 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 = gc.SSARegNum(v)

		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 := gc.SSAReg(v.Args[0]); reg.Name() != wantreg {
			v.Fatalf("bad reg %s for symbol type %T, want %s", reg.Name(), 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 = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	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 = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.To, v)
	case ssa.OpARMMOVWloadidx:
		// this is just shift 0 bits
		fallthrough
	case ssa.OpARMMOVWloadshiftLL:
		p := genshift(v.Op.Asm(), 0, gc.SSARegNum(v.Args[1]), gc.SSARegNum(v), arm.SHIFT_LL, v.AuxInt)
		p.From.Reg = gc.SSARegNum(v.Args[0])
	case ssa.OpARMMOVWloadshiftRL:
		p := genshift(v.Op.Asm(), 0, gc.SSARegNum(v.Args[1]), gc.SSARegNum(v), arm.SHIFT_LR, v.AuxInt)
		p.From.Reg = gc.SSARegNum(v.Args[0])
	case ssa.OpARMMOVWloadshiftRA:
		p := genshift(v.Op.Asm(), 0, gc.SSARegNum(v.Args[1]), gc.SSARegNum(v), arm.SHIFT_AR, v.AuxInt)
		p.From.Reg = gc.SSARegNum(v.Args[0])
	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 = gc.SSARegNum(v.Args[2])
		p.To.Type = obj.TYPE_SHIFT
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Offset = int64(makeshift(gc.SSARegNum(v.Args[1]), arm.SHIFT_LL, v.AuxInt))
	case ssa.OpARMMOVWstoreshiftRL:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[2])
		p.To.Type = obj.TYPE_SHIFT
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Offset = int64(makeshift(gc.SSARegNum(v.Args[1]), arm.SHIFT_LR, v.AuxInt))
	case ssa.OpARMMOVWstoreshiftRA:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[2])
		p.To.Type = obj.TYPE_SHIFT
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Offset = int64(makeshift(gc.SSARegNum(v.Args[1]), 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 gc.SSARegNum(v) == gc.SSARegNum(v.Args[0]) {
					return
				}
				p := gc.Prog(arm.AMOVW)
				p.From.Type = obj.TYPE_REG
				p.From.Reg = gc.SSARegNum(v.Args[0])
				p.To.Type = obj.TYPE_REG
				p.To.Reg = gc.SSARegNum(v)
				return
			default:
			}
		}
		fallthrough
	case ssa.OpARMMVN,
		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 = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	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 = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	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 = gc.SSARegNum(v)
	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 = gc.SSARegNum(v)
	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 = gc.SSARegNum(v.Args[0])
		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 = gc.SSARegNum(v.Args[0])
		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:
		// Optimization - if the subsequent block has a load or store
		// at the same address, we don't need to issue this instruction.
		mem := v.Args[1]
		for _, w := range v.Block.Succs[0].Block().Values {
			if w.Op == ssa.OpPhi {
				if w.Type.IsMemory() {
					mem = w
				}
				continue
			}
			if len(w.Args) == 0 || !w.Args[len(w.Args)-1].Type.IsMemory() {
				// w doesn't use a store - can't be a memory op.
				continue
			}
			if w.Args[len(w.Args)-1] != mem {
				v.Fatalf("wrong store after nilcheck v=%s w=%s", v, w)
			}
			switch w.Op {
			case ssa.OpARMMOVBload, ssa.OpARMMOVBUload, ssa.OpARMMOVHload, ssa.OpARMMOVHUload,
				ssa.OpARMMOVWload, ssa.OpARMMOVFload, ssa.OpARMMOVDload,
				ssa.OpARMMOVBstore, ssa.OpARMMOVHstore, ssa.OpARMMOVWstore,
				ssa.OpARMMOVFstore, ssa.OpARMMOVDstore:
				// arg0 is ptr, auxint is offset
				if w.Args[0] == v.Args[0] && w.Aux == nil && w.AuxInt >= 0 && w.AuxInt < minZeroPage {
					if gc.Debug_checknil != 0 && int(v.Line) > 1 {
						gc.Warnl(v.Line, "removed nil check")
					}
					return
				}
			case ssa.OpARMDUFFZERO, ssa.OpARMLoweredZero:
				// arg0 is ptr
				if w.Args[0] == v.Args[0] {
					if gc.Debug_checknil != 0 && int(v.Line) > 1 {
						gc.Warnl(v.Line, "removed nil check")
					}
					return
				}
			case ssa.OpARMDUFFCOPY, ssa.OpARMLoweredMove:
				// arg0 is dst ptr, arg1 is src ptr
				if w.Args[0] == v.Args[0] || w.Args[1] == v.Args[0] {
					if gc.Debug_checknil != 0 && int(v.Line) > 1 {
						gc.Warnl(v.Line, "removed nil check")
					}
					return
				}
			default:
			}
			if w.Type.IsMemory() {
				if w.Op == ssa.OpVarDef || w.Op == ssa.OpVarKill || w.Op == ssa.OpVarLive {
					// these ops are OK
					mem = w
					continue
				}
				// We can't delay the nil check past the next store.
				break
			}
		}
		// Issue a load which will fault if arg is nil.
		p := gc.Prog(arm.AMOVB)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		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 = gc.SSARegNum(v.Args[2])
		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 = gc.SSARegNum(v.Args[1])
		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 = gc.SSARegNum(v.Args[2])
		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:
		if !v.Args[0].Type.IsPtrShaped() {
			v.Fatalf("keeping non-pointer alive %v", v.Args[0])
		}
		n, off := gc.AutoVar(v.Args[0])
		if n == nil {
			v.Fatalf("KeepLive with non-spilled value %s %s", v, v.Args[0])
		}
		if off != 0 {
			v.Fatalf("KeepLive with non-zero offset spill location %s:%d", n, off)
		}
		gc.Gvarlive(n)
	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 = gc.SSARegNum(v)
		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 = gc.SSARegNum(v)
	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.Unimplementedf("genValue not implemented: %s", v.LongString())
	}
}
Beispiel #7
0
Datei: ssa.go Projekt: hurkgu/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.OpPPC64MOVDconvert:
		t := v.Type
		if t.IsMemory() {
			return
		}
		x := gc.SSARegNum(v.Args[0])
		y := gc.SSARegNum(v)
		if x != y {
			rt := obj.TYPE_REG
			op := ppc64.AMOVD

			if t.IsFloat() {
				op = ppc64.AFMOVD
			}
			p := gc.Prog(op)
			p.From.Type = rt
			p.From.Reg = x
			p.To.Type = rt
			p.To.Reg = y
		}

	case ssa.OpPPC64Xf2i64:
		{
			x := gc.SSARegNum(v.Args[0])
			y := gc.SSARegNum(v)
			p := gc.Prog(ppc64.AFMOVD)
			p.From.Type = obj.TYPE_REG
			p.From.Reg = x
			scratchFpMem(s, &p.To)
			p = gc.Prog(ppc64.AMOVD)
			p.To.Type = obj.TYPE_REG
			p.To.Reg = y
			scratchFpMem(s, &p.From)
		}
	case ssa.OpPPC64Xi2f64:
		{
			x := gc.SSARegNum(v.Args[0])
			y := gc.SSARegNum(v)
			p := gc.Prog(ppc64.AMOVD)
			p.From.Type = obj.TYPE_REG
			p.From.Reg = x
			scratchFpMem(s, &p.To)
			p = gc.Prog(ppc64.AFMOVD)
			p.To.Type = obj.TYPE_REG
			p.To.Reg = y
			scratchFpMem(s, &p.From)
		}

	case ssa.OpPPC64LoweredGetClosurePtr:
		// Closure pointer is R11 (already)
		gc.CheckLoweredGetClosurePtr(v)

	case ssa.OpLoadReg:
		loadOp := loadByType(v.Type)
		n, off := gc.AutoVar(v.Args[0])
		p := gc.Prog(loadOp)
		p.From.Type = obj.TYPE_MEM
		p.From.Node = n
		p.From.Sym = gc.Linksym(n.Sym)
		p.From.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.From.Name = obj.NAME_PARAM
			p.From.Offset += n.Xoffset
		} else {
			p.From.Name = obj.NAME_AUTO
		}
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

	case ssa.OpStoreReg:
		storeOp := storeByType(v.Type)
		n, off := gc.AutoVar(v)
		p := gc.Prog(storeOp)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_MEM
		p.To.Node = n
		p.To.Sym = gc.Linksym(n.Sym)
		p.To.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.To.Name = obj.NAME_PARAM
			p.To.Offset += n.Xoffset
		} else {
			p.To.Name = obj.NAME_AUTO
		}

	case ssa.OpPPC64DIVD:
		// For now,
		//
		// cmp arg1, -1
		// be  ahead
		// v = arg0 / arg1
		// b over
		// ahead: v = - arg0
		// over: nop
		r := gc.SSARegNum(v)
		r0 := gc.SSARegNum(v.Args[0])
		r1 := gc.SSARegNum(v.Args[1])

		p := gc.Prog(ppc64.ACMP)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r1
		p.To.Type = obj.TYPE_CONST
		p.To.Offset = -1

		pbahead := gc.Prog(ppc64.ABEQ)
		pbahead.To.Type = obj.TYPE_BRANCH

		p = gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r1
		p.Reg = r0
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r

		pbover := gc.Prog(obj.AJMP)
		pbover.To.Type = obj.TYPE_BRANCH

		p = gc.Prog(ppc64.ANEG)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r0
		gc.Patch(pbahead, p)

		p = gc.Prog(obj.ANOP)
		gc.Patch(pbover, p)

	case ssa.OpPPC64DIVW:
		// word-width version of above
		r := gc.SSARegNum(v)
		r0 := gc.SSARegNum(v.Args[0])
		r1 := gc.SSARegNum(v.Args[1])

		p := gc.Prog(ppc64.ACMPW)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r1
		p.To.Type = obj.TYPE_CONST
		p.To.Offset = -1

		pbahead := gc.Prog(ppc64.ABEQ)
		pbahead.To.Type = obj.TYPE_BRANCH

		p = gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r1
		p.Reg = r0
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r

		pbover := gc.Prog(obj.AJMP)
		pbover.To.Type = obj.TYPE_BRANCH

		p = gc.Prog(ppc64.ANEG)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r0
		gc.Patch(pbahead, p)

		p = gc.Prog(obj.ANOP)
		gc.Patch(pbover, p)

	case ssa.OpPPC64ADD, ssa.OpPPC64FADD, ssa.OpPPC64FADDS, ssa.OpPPC64SUB, ssa.OpPPC64FSUB, ssa.OpPPC64FSUBS,
		ssa.OpPPC64MULLD, ssa.OpPPC64MULLW, ssa.OpPPC64DIVDU, ssa.OpPPC64DIVWU,
		ssa.OpPPC64SRAD, ssa.OpPPC64SRAW, ssa.OpPPC64SRD, ssa.OpPPC64SRW, ssa.OpPPC64SLD, ssa.OpPPC64SLW,
		ssa.OpPPC64MULHD, ssa.OpPPC64MULHW, ssa.OpPPC64MULHDU, ssa.OpPPC64MULHWU,
		ssa.OpPPC64FMUL, ssa.OpPPC64FMULS, ssa.OpPPC64FDIV, ssa.OpPPC64FDIVS,
		ssa.OpPPC64AND, ssa.OpPPC64OR, ssa.OpPPC64ANDN, ssa.OpPPC64ORN, ssa.OpPPC64XOR, ssa.OpPPC64EQV:
		r := gc.SSARegNum(v)
		r1 := gc.SSARegNum(v.Args[0])
		r2 := gc.SSARegNum(v.Args[1])
		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.OpPPC64MaskIfNotCarry:
		r := gc.SSARegNum(v)
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = ppc64.REGZERO
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r

	case ssa.OpPPC64ADDconstForCarry:
		r1 := gc.SSARegNum(v.Args[0])
		p := gc.Prog(v.Op.Asm())
		p.Reg = r1
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = ppc64.REGTMP // Ignored; this is for the carry effect.

	case ssa.OpPPC64NEG, ssa.OpPPC64FNEG, ssa.OpPPC64FSQRT, ssa.OpPPC64FSQRTS, ssa.OpPPC64FCTIDZ, ssa.OpPPC64FCTIWZ, ssa.OpPPC64FCFID, ssa.OpPPC64FRSP:
		r := gc.SSARegNum(v)
		p := gc.Prog(v.Op.Asm())
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])

	case ssa.OpPPC64ADDconst, ssa.OpPPC64ANDconst, ssa.OpPPC64ORconst, ssa.OpPPC64XORconst,
		ssa.OpPPC64SRADconst, ssa.OpPPC64SRAWconst, ssa.OpPPC64SRDconst, ssa.OpPPC64SRWconst, ssa.OpPPC64SLDconst, ssa.OpPPC64SLWconst:
		p := gc.Prog(v.Op.Asm())
		p.Reg = gc.SSARegNum(v.Args[0])

		if v.Aux != nil {
			p.From.Type = obj.TYPE_CONST
			p.From.Offset = gc.AuxOffset(v)
		} else {
			p.From.Type = obj.TYPE_CONST
			p.From.Offset = v.AuxInt
		}

		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

	case ssa.OpPPC64MOVDaddr:
		p := gc.Prog(ppc64.AMOVD)
		p.From.Type = obj.TYPE_ADDR
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

		var wantreg string
		// Suspect comment, copied from ARM code
		// MOVD $sym+off(base), R
		// the assembler expands it as the following:
		// - base is SP: add constant offset to SP
		//               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 = ppc64.REGSP
			p.From.Offset = v.AuxInt
		}
		if reg := gc.SSAReg(v.Args[0]); reg.Name() != wantreg {
			v.Fatalf("bad reg %s for symbol type %T, want %s", reg.Name(), v.Aux, wantreg)
		}

	case ssa.OpPPC64MOVDconst, ssa.OpPPC64MOVWconst:
		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 = gc.SSARegNum(v)

	case ssa.OpPPC64FMOVDconst, ssa.OpPPC64FMOVSconst:
		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 = gc.SSARegNum(v)

	case ssa.OpPPC64FCMPU, ssa.OpPPC64CMP, ssa.OpPPC64CMPW, ssa.OpPPC64CMPU, ssa.OpPPC64CMPWU:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v.Args[1])

	case ssa.OpPPC64CMPconst, ssa.OpPPC64CMPUconst, ssa.OpPPC64CMPWconst, ssa.OpPPC64CMPWUconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_CONST
		p.To.Offset = v.AuxInt

	case ssa.OpPPC64MOVBreg, ssa.OpPPC64MOVBZreg, ssa.OpPPC64MOVHreg, ssa.OpPPC64MOVHZreg, ssa.OpPPC64MOVWreg, ssa.OpPPC64MOVWZreg:
		// Shift in register to required size
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.To.Reg = gc.SSARegNum(v)
		p.To.Type = obj.TYPE_REG

	case ssa.OpPPC64MOVDload, ssa.OpPPC64MOVWload, ssa.OpPPC64MOVBload, ssa.OpPPC64MOVHload, ssa.OpPPC64MOVWZload, ssa.OpPPC64MOVBZload, ssa.OpPPC64MOVHZload:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

	case ssa.OpPPC64FMOVDload, ssa.OpPPC64FMOVSload:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

	case ssa.OpPPC64MOVDstorezero, ssa.OpPPC64MOVWstorezero, ssa.OpPPC64MOVHstorezero, ssa.OpPPC64MOVBstorezero:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = ppc64.REGZERO
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.To, v)

	case ssa.OpPPC64MOVDstore, ssa.OpPPC64MOVWstore, ssa.OpPPC64MOVHstore, ssa.OpPPC64MOVBstore:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.To, v)
	case ssa.OpPPC64FMOVDstore, ssa.OpPPC64FMOVSstore:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.To, v)

	case ssa.OpPPC64Equal,
		ssa.OpPPC64NotEqual,
		ssa.OpPPC64LessThan,
		ssa.OpPPC64FLessThan,
		ssa.OpPPC64LessEqual,
		ssa.OpPPC64GreaterThan,
		ssa.OpPPC64FGreaterThan,
		ssa.OpPPC64GreaterEqual:
		// On Power7 or later, can use isel instruction:
		// for a < b, a > b, a = b:
		//   rt := 1
		//   isel rt,rt,r0,cond

		// for  a >= b, a <= b, a != b:
		//   rt := 1
		//   isel rt,0,rt,!cond

		// However, PPCbe support is for older machines than that,
		// and isel (which looks a lot like fsel) isn't recognized
		// yet by the Go assembler.  So for now, use the old instruction
		// sequence, which we'll need anyway.
		// TODO: add support for isel on PPCle and use it.

		// generate boolean values
		// use conditional move

		p := gc.Prog(ppc64.AMOVW)
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = 1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

		pb := gc.Prog(condOps[v.Op])
		pb.To.Type = obj.TYPE_BRANCH

		p = gc.Prog(ppc64.AMOVW)
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = 0
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

		p = gc.Prog(obj.ANOP)
		gc.Patch(pb, p)

	case ssa.OpPPC64FLessEqual, // These include a second branch for EQ -- dealing with NaN prevents REL= to !REL conversion
		ssa.OpPPC64FGreaterEqual:

		p := gc.Prog(ppc64.AMOVW)
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = 1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

		pb0 := gc.Prog(condOps[v.Op])
		pb0.To.Type = obj.TYPE_BRANCH
		pb1 := gc.Prog(ppc64.ABEQ)
		pb1.To.Type = obj.TYPE_BRANCH

		p = gc.Prog(ppc64.AMOVW)
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = 0
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)

		p = gc.Prog(obj.ANOP)
		gc.Patch(pb0, p)
		gc.Patch(pb1, p)

	case ssa.OpPPC64LoweredZero:
		// Similar to how this is done on ARM,
		// except that PPC MOVDU x,off(y) is *(y+off) = x; y=y+off
		// not store-and-increment.
		// Therefore R3 should be dest-align
		// and arg1 should be dest+size-align
		// HOWEVER, the input dest address cannot be dest-align because
		// that does not necessarily address valid memory and it's not
		// known how that might be optimized.  Therefore, correct it in
		// in the expansion:
		//
		// ADD    -8,R3,R3
		// MOVDU  R0, 8(R3)
		// CMP	  R3, Rarg1
		// BL	  -2(PC)
		// arg1 is the address of the last element to zero
		// auxint is alignment
		var sz int64
		var movu obj.As
		switch {
		case v.AuxInt%8 == 0:
			sz = 8
			movu = ppc64.AMOVDU
		case v.AuxInt%4 == 0:
			sz = 4
			movu = ppc64.AMOVWZU // MOVWU instruction not implemented
		case v.AuxInt%2 == 0:
			sz = 2
			movu = ppc64.AMOVHU
		default:
			sz = 1
			movu = ppc64.AMOVBU
		}

		p := gc.Prog(ppc64.AADD)
		p.Reg = gc.SSARegNum(v.Args[0])
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = -sz
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v.Args[0])

		p = gc.Prog(movu)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = ppc64.REG_R0
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		p.To.Offset = sz

		p2 := gc.Prog(ppc64.ACMPU)
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = gc.SSARegNum(v.Args[0])
		p2.To.Reg = gc.SSARegNum(v.Args[1])
		p2.To.Type = obj.TYPE_REG

		p3 := gc.Prog(ppc64.ABLT)
		p3.To.Type = obj.TYPE_BRANCH
		gc.Patch(p3, p)

	case ssa.OpPPC64LoweredMove:
		// Similar to how this is done on ARM,
		// except that PPC MOVDU x,off(y) is *(y+off) = x; y=y+off,
		// not store-and-increment.
		// Inputs must be valid pointers to memory,
		// so adjust arg0 and arg1 as part of the expansion.
		// arg2 should be src+size-align,
		//
		// ADD    -8,R3,R3
		// ADD    -8,R4,R4
		// MOVDU	8(R4), Rtmp
		// MOVDU 	Rtmp, 8(R3)
		// CMP	R4, Rarg2
		// BL	-3(PC)
		// arg2 is the address of the last element of src
		// auxint is alignment
		var sz int64
		var movu obj.As
		switch {
		case v.AuxInt%8 == 0:
			sz = 8
			movu = ppc64.AMOVDU
		case v.AuxInt%4 == 0:
			sz = 4
			movu = ppc64.AMOVWZU // MOVWU instruction not implemented
		case v.AuxInt%2 == 0:
			sz = 2
			movu = ppc64.AMOVHU
		default:
			sz = 1
			movu = ppc64.AMOVBU
		}

		p := gc.Prog(ppc64.AADD)
		p.Reg = gc.SSARegNum(v.Args[0])
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = -sz
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v.Args[0])

		p = gc.Prog(ppc64.AADD)
		p.Reg = gc.SSARegNum(v.Args[1])
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = -sz
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v.Args[1])

		p = gc.Prog(movu)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[1])
		p.From.Offset = sz
		p.To.Type = obj.TYPE_REG
		p.To.Reg = ppc64.REGTMP

		p2 := gc.Prog(movu)
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = ppc64.REGTMP
		p2.To.Type = obj.TYPE_MEM
		p2.To.Reg = gc.SSARegNum(v.Args[0])
		p2.To.Offset = sz

		p3 := gc.Prog(ppc64.ACMPU)
		p3.From.Reg = gc.SSARegNum(v.Args[1])
		p3.From.Type = obj.TYPE_REG
		p3.To.Reg = gc.SSARegNum(v.Args[2])
		p3.To.Type = obj.TYPE_REG

		p4 := gc.Prog(ppc64.ABLT)
		p4.To.Type = obj.TYPE_BRANCH
		gc.Patch(p4, p)

	case ssa.OpPPC64CALLstatic:
		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 two actual hardware NOPs 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.
			// PPC64 is unusual because TWO nops are required
			// (see gc/cgen.go, gc/plive.go -- copy of comment below)
			//
			// On ppc64, when compiling Go into position
			// independent code on ppc64le we insert an
			// instruction to reload the TOC pointer from the
			// stack as well. See the long comment near
			// jmpdefer in runtime/asm_ppc64.s for why.
			// If the MOVD is not needed, insert a hardware NOP
			// so that the same number of instructions are used
			// on ppc64 in both shared and non-shared modes.
			ginsnop()
			if gc.Ctxt.Flag_shared {
				p := gc.Prog(ppc64.AMOVD)
				p.From.Type = obj.TYPE_MEM
				p.From.Offset = 24
				p.From.Reg = ppc64.REGSP
				p.To.Type = obj.TYPE_REG
				p.To.Reg = ppc64.REG_R2
			} else {
				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.OpPPC64CALLclosure, ssa.OpPPC64CALLinter:
		p := gc.Prog(ppc64.AMOVD)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = ppc64.REG_CTR

		if gc.Ctxt.Flag_shared && p.From.Reg != ppc64.REG_R12 {
			// Make sure function pointer is in R12 as well when
			// compiling Go into PIC.
			// TODO(mwhudson): it would obviously be better to
			// change the register allocation to put the value in
			// R12 already, but I don't know how to do that.
			// TODO: We have the technology now to implement TODO above.
			q := gc.Prog(ppc64.AMOVD)
			q.From = p.From
			q.To.Type = obj.TYPE_REG
			q.To.Reg = ppc64.REG_R12
		}

		pp := gc.Prog(obj.ACALL)
		pp.To.Type = obj.TYPE_REG
		pp.To.Reg = ppc64.REG_CTR

		if gc.Ctxt.Flag_shared {
			// When compiling Go into PIC, the function we just
			// called via pointer might have been implemented in
			// a separate module and so overwritten the TOC
			// pointer in R2; reload it.
			q := gc.Prog(ppc64.AMOVD)
			q.From.Type = obj.TYPE_MEM
			q.From.Offset = 24
			q.From.Reg = ppc64.REGSP
			q.To.Type = obj.TYPE_REG
			q.To.Reg = ppc64.REG_R2
		}

		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}

	case ssa.OpPPC64CALLdefer:
		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.OpPPC64CALLgo:
		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.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:
		if !v.Args[0].Type.IsPtrShaped() {
			v.Fatalf("keeping non-pointer alive %v", v.Args[0])
		}
		n, off := gc.AutoVar(v.Args[0])
		if n == nil {
			v.Fatalf("KeepLive with non-spilled value %s %s", v, v.Args[0])
		}
		if off != 0 {
			v.Fatalf("KeepLive with non-zero offset spill location %s:%d", n, off)
		}
		gc.Gvarlive(n)

	case ssa.OpPhi:
		// just check to make sure regalloc and stackalloc did it right
		if v.Type.IsMemory() {
			return
		}
		f := v.Block.Func
		loc := f.RegAlloc[v.ID]
		for _, a := range v.Args {
			if aloc := f.RegAlloc[a.ID]; aloc != loc { // TODO: .Equal() instead?
				v.Fatalf("phi arg at different location than phi: %v @ %v, but arg %v @ %v\n%s\n", v, loc, a, aloc, v.Block.Func)
			}
		}

	case ssa.OpPPC64LoweredNilCheck:
		// Optimization - if the subsequent block has a load or store
		// at the same address, we don't need to issue this instruction.
		// mem := v.Args[1]
		// for _, w := range v.Block.Succs[0].Block().Values {
		// 	if w.Op == ssa.OpPhi {
		// 		if w.Type.IsMemory() {
		// 			mem = w
		// 		}
		// 		continue
		// 	}
		// 	if len(w.Args) == 0 || !w.Args[len(w.Args)-1].Type.IsMemory() {
		// 		// w doesn't use a store - can't be a memory op.
		// 		continue
		// 	}
		// 	if w.Args[len(w.Args)-1] != mem {
		// 		v.Fatalf("wrong store after nilcheck v=%s w=%s", v, w)
		// 	}
		// 	switch w.Op {
		// 	case ssa.OpPPC64MOVBload, ssa.OpPPC64MOVBUload, ssa.OpPPC64MOVHload, ssa.OpPPC64MOVHUload,
		// 		ssa.OpPPC64MOVWload, ssa.OpPPC64MOVFload, ssa.OpPPC64MOVDload,
		// 		ssa.OpPPC64MOVBstore, ssa.OpPPC64MOVHstore, ssa.OpPPC64MOVWstore,
		// 		ssa.OpPPC64MOVFstore, ssa.OpPPC64MOVDstore:
		// 		// arg0 is ptr, auxint is offset
		// 		if w.Args[0] == v.Args[0] && w.Aux == nil && w.AuxInt >= 0 && w.AuxInt < minZeroPage {
		// 			if gc.Debug_checknil != 0 && int(v.Line) > 1 {
		// 				gc.Warnl(v.Line, "removed nil check")
		// 			}
		// 			return
		// 		}
		// 	case ssa.OpPPC64DUFFZERO, ssa.OpPPC64LoweredZero, ssa.OpPPC64LoweredZeroU:
		// 		// arg0 is ptr
		// 		if w.Args[0] == v.Args[0] {
		// 			if gc.Debug_checknil != 0 && int(v.Line) > 1 {
		// 				gc.Warnl(v.Line, "removed nil check")
		// 			}
		// 			return
		// 		}
		// 	case ssa.OpPPC64DUFFCOPY, ssa.OpPPC64LoweredMove, ssa.OpPPC64LoweredMoveU:
		// 		// arg0 is dst ptr, arg1 is src ptr
		// 		if w.Args[0] == v.Args[0] || w.Args[1] == v.Args[0] {
		// 			if gc.Debug_checknil != 0 && int(v.Line) > 1 {
		// 				gc.Warnl(v.Line, "removed nil check")
		// 			}
		// 			return
		// 		}
		// 	default:
		// 	}
		// 	if w.Type.IsMemory() {
		// 		if w.Op == ssa.OpVarDef || w.Op == ssa.OpVarKill || w.Op == ssa.OpVarLive {
		// 			// these ops are OK
		// 			mem = w
		// 			continue
		// 		}
		// 		// We can't delay the nil check past the next store.
		// 		break
		// 	}
		// }
		// Issue a load which will fault if arg is nil.
		p := gc.Prog(ppc64.AMOVB)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = ppc64.REGTMP
		if gc.Debug_checknil != 0 && v.Line > 1 { // v.Line==1 in generated wrappers
			gc.Warnl(v.Line, "generated nil check")
		}

	case ssa.OpPPC64InvertFlags:
		v.Fatalf("InvertFlags should never make it to codegen %v", v.LongString())
	case ssa.OpPPC64FlagEQ, ssa.OpPPC64FlagLT, ssa.OpPPC64FlagGT:
		v.Fatalf("Flag* ops should never make it to codegen %v", v.LongString())

	default:
		v.Unimplementedf("genValue not implemented: %s", v.LongString())
	}
}
Beispiel #8
0
Datei: ssa.go Projekt: hurkgu/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 := gc.SSARegNum(v.Args[0])
		y := gc.SSARegNum(v)
		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 gc.SSARegNum(v) != gc.SSARegNum(v.Args[0]) {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		// nothing to do
	case ssa.OpLoadReg:
		if v.Type.IsFlags() {
			v.Unimplementedf("load flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(loadByType(v.Type))
		n, off := gc.AutoVar(v.Args[0])
		p.From.Type = obj.TYPE_MEM
		p.From.Node = n
		p.From.Sym = gc.Linksym(n.Sym)
		p.From.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.From.Name = obj.NAME_PARAM
			p.From.Offset += n.Xoffset
		} else {
			p.From.Name = obj.NAME_AUTO
		}
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpPhi:
		gc.CheckLoweredPhi(v)
	case ssa.OpStoreReg:
		if v.Type.IsFlags() {
			v.Unimplementedf("store flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(storeByType(v.Type))
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		n, off := gc.AutoVar(v)
		p.To.Type = obj.TYPE_MEM
		p.To.Node = n
		p.To.Sym = gc.Linksym(n.Sym)
		p.To.Offset = off
		if n.Class == gc.PPARAM || n.Class == gc.PPARAMOUT {
			p.To.Name = obj.NAME_PARAM
			p.To.Offset += n.Xoffset
		} else {
			p.To.Name = obj.NAME_AUTO
		}
	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 := gc.SSARegNum(v)
		r1 := gc.SSARegNum(v.Args[0])
		r2 := gc.SSARegNum(v.Args[1])
		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 = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpARM64ADDshiftLL,
		ssa.OpARM64SUBshiftLL,
		ssa.OpARM64ANDshiftLL,
		ssa.OpARM64ORshiftLL,
		ssa.OpARM64XORshiftLL,
		ssa.OpARM64BICshiftLL:
		genshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v), arm64.SHIFT_LL, v.AuxInt)
	case ssa.OpARM64ADDshiftRL,
		ssa.OpARM64SUBshiftRL,
		ssa.OpARM64ANDshiftRL,
		ssa.OpARM64ORshiftRL,
		ssa.OpARM64XORshiftRL,
		ssa.OpARM64BICshiftRL:
		genshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v), arm64.SHIFT_LR, v.AuxInt)
	case ssa.OpARM64ADDshiftRA,
		ssa.OpARM64SUBshiftRA,
		ssa.OpARM64ANDshiftRA,
		ssa.OpARM64ORshiftRA,
		ssa.OpARM64XORshiftRA,
		ssa.OpARM64BICshiftRA:
		genshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), gc.SSARegNum(v), 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 = gc.SSARegNum(v)
	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 = gc.SSARegNum(v)
	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 = gc.SSARegNum(v.Args[1])
		p.Reg = gc.SSARegNum(v.Args[0])
	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 = gc.SSARegNum(v.Args[0])
	case ssa.OpARM64CMPshiftLL:
		genshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), 0, arm64.SHIFT_LL, v.AuxInt)
	case ssa.OpARM64CMPshiftRL:
		genshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), 0, arm64.SHIFT_LR, v.AuxInt)
	case ssa.OpARM64CMPshiftRA:
		genshift(v.Op.Asm(), gc.SSARegNum(v.Args[0]), gc.SSARegNum(v.Args[1]), 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 = gc.SSARegNum(v)

		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 := gc.SSAReg(v.Args[0]); reg.Name() != wantreg {
			v.Fatalf("bad reg %s for symbol type %T, want %s", reg.Name(), 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 = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpARM64MOVBstore,
		ssa.OpARM64MOVHstore,
		ssa.OpARM64MOVWstore,
		ssa.OpARM64MOVDstore,
		ssa.OpARM64FMOVSstore,
		ssa.OpARM64FMOVDstore:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[1])
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = gc.SSARegNum(v.Args[0])
		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 = gc.SSARegNum(v.Args[0])
		gc.AddAux(&p.To, v)
	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 gc.SSARegNum(v) == gc.SSARegNum(v.Args[0]) {
					return
				}
				p := gc.Prog(arm64.AMOVD)
				p.From.Type = obj.TYPE_REG
				p.From.Reg = gc.SSARegNum(v.Args[0])
				p.To.Type = obj.TYPE_REG
				p.To.Reg = gc.SSARegNum(v)
				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:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = gc.SSARegNum(v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	case ssa.OpARM64CSELULT,
		ssa.OpARM64CSELULT0:
		r1 := int16(arm64.REGZERO)
		if v.Op == ssa.OpARM64CSELULT {
			r1 = gc.SSARegNum(v.Args[1])
		}
		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 = gc.SSARegNum(v.Args[0])
		p.From3 = &obj.Addr{Type: obj.TYPE_REG, Reg: r1}
		p.To.Type = obj.TYPE_REG
		p.To.Reg = gc.SSARegNum(v)
	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 = gc.SSARegNum(v.Args[0])
		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 = gc.SSARegNum(v.Args[1])
		p2.Reg = arm64.REG_R16
		p3 := gc.Prog(arm64.ABLE)
		p3.To.Type = obj.TYPE_BRANCH
		gc.Patch(p3, p)
	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 = gc.SSARegNum(v.Args[2])
		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 = gc.SSARegNum(v.Args[0])
		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 = gc.SSARegNum(v.Args[0])
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARM64LoweredNilCheck:
		// Optimization - if the subsequent block has a load or store
		// at the same address, we don't need to issue this instruction.
		mem := v.Args[1]
		for _, w := range v.Block.Succs[0].Block().Values {
			if w.Op == ssa.OpPhi {
				if w.Type.IsMemory() {
					mem = w
				}
				continue
			}
			if len(w.Args) == 0 || !w.Args[len(w.Args)-1].Type.IsMemory() {
				// w doesn't use a store - can't be a memory op.
				continue
			}
			if w.Args[len(w.Args)-1] != mem {
				v.Fatalf("wrong store after nilcheck v=%s w=%s", v, w)
			}
			switch w.Op {
			case ssa.OpARM64MOVBload, ssa.OpARM64MOVBUload, ssa.OpARM64MOVHload, ssa.OpARM64MOVHUload,
				ssa.OpARM64MOVWload, ssa.OpARM64MOVWUload, ssa.OpARM64MOVDload,
				ssa.OpARM64FMOVSload, ssa.OpARM64FMOVDload,
				ssa.OpARM64MOVBstore, ssa.OpARM64MOVHstore, ssa.OpARM64MOVWstore, ssa.OpARM64MOVDstore,
				ssa.OpARM64FMOVSstore, ssa.OpARM64FMOVDstore:
				// arg0 is ptr, auxint is offset
				if w.Args[0] == v.Args[0] && w.Aux == nil && w.AuxInt >= 0 && w.AuxInt < minZeroPage {
					if gc.Debug_checknil != 0 && int(v.Line) > 1 {
						gc.Warnl(v.Line, "removed nil check")
					}
					return
				}
			case ssa.OpARM64DUFFZERO, ssa.OpARM64LoweredZero:
				// arg0 is ptr
				if w.Args[0] == v.Args[0] {
					if gc.Debug_checknil != 0 && int(v.Line) > 1 {
						gc.Warnl(v.Line, "removed nil check")
					}
					return
				}
			case ssa.OpARM64LoweredMove:
				// arg0 is dst ptr, arg1 is src ptr
				if w.Args[0] == v.Args[0] || w.Args[1] == v.Args[0] {
					if gc.Debug_checknil != 0 && int(v.Line) > 1 {
						gc.Warnl(v.Line, "removed nil check")
					}
					return
				}
			default:
			}
			if w.Type.IsMemory() {
				if w.Op == ssa.OpVarDef || w.Op == ssa.OpVarKill || w.Op == ssa.OpVarLive {
					// these ops are OK
					mem = w
					continue
				}
				// We can't delay the nil check past the next store.
				break
			}
		}
		// Issue a load which will fault if arg is nil.
		p := gc.Prog(arm64.AMOVB)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = gc.SSARegNum(v.Args[0])
		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:
		if !v.Args[0].Type.IsPtrShaped() {
			v.Fatalf("keeping non-pointer alive %v", v.Args[0])
		}
		n, off := gc.AutoVar(v.Args[0])
		if n == nil {
			v.Fatalf("KeepLive with non-spilled value %s %s", v, v.Args[0])
		}
		if off != 0 {
			v.Fatalf("KeepLive with non-zero offset spill location %s:%d", n, off)
		}
		gc.Gvarlive(n)
	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 = gc.SSARegNum(v)
	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.Unimplementedf("genValue not implemented: %s", v.LongString())
	}
}