Exemplo n.º 1
0
Arquivo: ggen.go Projeto: achanda/go
func ginsnop() {
	p := gc.Prog(s390x.AOR)
	p.From.Type = obj.TYPE_REG
	p.From.Reg = int16(s390x.REG_R0)
	p.To.Type = obj.TYPE_REG
	p.To.Reg = int16(s390x.REG_R0)
}
Exemplo n.º 2
0
Arquivo: ggen.go Projeto: achanda/go
func ginsnop() {
	p := gc.Prog(x86.AXCHGL)
	p.From.Type = obj.TYPE_REG
	p.From.Reg = x86.REG_AX
	p.To.Type = obj.TYPE_REG
	p.To.Reg = x86.REG_AX
}
Exemplo n.º 3
0
Arquivo: 387.go Projeto: achanda/go
// push pushes v onto the floating-point stack.  v must be in a register.
func push(s *gc.SSAGenState, v *ssa.Value) {
	p := gc.Prog(x86.AFMOVD)
	p.From.Type = obj.TYPE_REG
	p.From.Reg = s.SSEto387[v.Reg()]
	p.To.Type = obj.TYPE_REG
	p.To.Reg = x86.REG_F0
}
Exemplo n.º 4
0
Arquivo: ggen.go Projeto: achanda/go
func ginsnop() {
	p := gc.Prog(ppc64.AOR)
	p.From.Type = obj.TYPE_REG
	p.From.Reg = ppc64.REG_R0
	p.To.Type = obj.TYPE_REG
	p.To.Reg = ppc64.REG_R0
}
Exemplo n.º 5
0
func ginsnop() {
	p := gc.Prog(mips.ANOR)
	p.From.Type = obj.TYPE_REG
	p.From.Reg = mips.REG_R0
	p.To.Type = obj.TYPE_REG
	p.To.Reg = mips.REG_R0
}
Exemplo n.º 6
0
// opregreg emits instructions for
//     dest := dest(To) op src(From)
// and also returns the created obj.Prog so it
// may be further adjusted (offset, scale, etc).
func opregreg(op obj.As, dest, src int16) *obj.Prog {
	p := gc.Prog(op)
	p.From.Type = obj.TYPE_REG
	p.To.Type = obj.TYPE_REG
	p.To.Reg = dest
	p.From.Reg = src
	return p
}
Exemplo n.º 7
0
Arquivo: ggen.go Projeto: achanda/go
func ginsnop() {
	p := gc.Prog(arm.AAND)
	p.From.Type = obj.TYPE_REG
	p.From.Reg = arm.REG_R0
	p.To.Type = obj.TYPE_REG
	p.To.Reg = arm.REG_R0
	p.Scond = arm.C_SCOND_EQ
}
Exemplo n.º 8
0
// opregregimm emits instructions for
//	dest := src(From) op off
// and also returns the created obj.Prog so it
// may be further adjusted (offset, scale, etc).
func opregregimm(op obj.As, dest, src int16, off int64) *obj.Prog {
	p := gc.Prog(op)
	p.From.Type = obj.TYPE_CONST
	p.From.Offset = off
	p.Reg = src
	p.To.Reg = dest
	p.To.Type = obj.TYPE_REG
	return p
}
Exemplo n.º 9
0
Arquivo: ggen.go Projeto: achanda/go
func ginsnop() {
	// This is actually not the x86 NOP anymore,
	// but at the point where it gets used, AX is dead
	// so it's okay if we lose the high bits.
	p := gc.Prog(x86.AXCHGL)
	p.From.Type = obj.TYPE_REG
	p.From.Reg = x86.REG_AX
	p.To.Type = obj.TYPE_REG
	p.To.Reg = x86.REG_AX
}
Exemplo n.º 10
0
Arquivo: ssa.go Projeto: achanda/go
func ssaGenISEL(v *ssa.Value, cr int64, r1, r2 int16) {
	r := v.Reg()
	p := gc.Prog(ppc64.AISEL)
	p.To.Type = obj.TYPE_REG
	p.To.Reg = r
	p.Reg = r1
	p.From3 = &obj.Addr{Type: obj.TYPE_REG, Reg: r2}
	p.From.Type = obj.TYPE_CONST
	p.From.Offset = cr
}
Exemplo n.º 11
0
Arquivo: 387.go Projeto: achanda/go
// flush387 removes all entries from the 387 floating-point stack.
func flush387(s *gc.SSAGenState) {
	for k := range s.SSEto387 {
		p := gc.Prog(x86.AFMOVDP)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = x86.REG_F0
		p.To.Type = obj.TYPE_REG
		p.To.Reg = x86.REG_F0
		delete(s.SSEto387, k)
	}
}
Exemplo n.º 12
0
Arquivo: gsubr.go Projeto: Greentor/go
// generate one instruction:
//	as f, t
func rawgins(as obj.As, f *gc.Node, t *gc.Node) *obj.Prog {
	// self move check
	// TODO(mundaym): use sized math and extend to MOVB, MOVWZ etc.
	switch as {
	case s390x.AMOVD, s390x.AFMOVS, s390x.AFMOVD:
		if f != nil && t != nil &&
			f.Op == gc.OREGISTER && t.Op == gc.OREGISTER &&
			f.Reg == t.Reg {
			return nil
		}
	}

	p := gc.Prog(as)
	gc.Naddr(&p.From, f)
	gc.Naddr(&p.To, t)

	switch as {
	// Bad things the front end has done to us. Crash to find call stack.
	case s390x.ACMP, s390x.ACMPU:
		if p.From.Type == obj.TYPE_MEM || p.To.Type == obj.TYPE_MEM {
			gc.Debug['h'] = 1
			gc.Fatalf("bad inst: %v", p)
		}
	}

	if gc.Debug['g'] != 0 {
		fmt.Printf("%v\n", p)
	}

	w := int32(0)
	switch as {
	case s390x.AMOVB, s390x.AMOVBZ:
		w = 1

	case s390x.AMOVH, s390x.AMOVHZ:
		w = 2

	case s390x.AMOVW, s390x.AMOVWZ:
		w = 4

	case s390x.AMOVD:
		if p.From.Type == obj.TYPE_CONST || p.From.Type == obj.TYPE_ADDR {
			break
		}
		w = 8
	}

	if w != 0 && ((f != nil && p.From.Width < int64(w)) || (t != nil && p.To.Type != obj.TYPE_REG && p.To.Width > int64(w))) {
		gc.Dump("f", f)
		gc.Dump("t", t)
		gc.Fatalf("bad width: %v (%d, %d)\n", p, p.From.Width, p.To.Width)
	}

	return p
}
Exemplo n.º 13
0
Arquivo: ssa.go Projeto: achanda/go
// genshift generates a Prog for r = r0 op (r1 shifted by s)
func genshift(as obj.As, r0, r1, r int16, typ int64, s int64) *obj.Prog {
	p := gc.Prog(as)
	p.From.Type = obj.TYPE_SHIFT
	p.From.Offset = int64(makeshift(r1, typ, s))
	p.Reg = r0
	if r != 0 {
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	}
	return p
}
Exemplo n.º 14
0
func ssaGenBlock(s *gc.SSAGenState, b, next *ssa.Block) {
	s.SetLineno(b.Line)

	switch b.Kind {
	case ssa.BlockCall:
		if b.Succs[0] != next {
			p := gc.Prog(obj.AJMP)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{p, b.Succs[0]})
		}
	case ssa.BlockRet:
		gc.Prog(obj.ARET)
	case ssa.BlockARMLT:
		p := gc.Prog(arm.ABGE)
		p.To.Type = obj.TYPE_BRANCH
		s.Branches = append(s.Branches, gc.Branch{p, b.Succs[0]})
		p = gc.Prog(obj.AJMP)
		p.To.Type = obj.TYPE_BRANCH
		s.Branches = append(s.Branches, gc.Branch{p, b.Succs[1]})
	}
}
Exemplo n.º 15
0
Arquivo: gsubr.go Projeto: wycharry/go
/*
 * generate one instruction:
 *	as f, t
 */
func rawgins(as int, f *gc.Node, t *gc.Node) *obj.Prog {
	// TODO(austin): Add self-move test like in 6g (but be careful
	// of truncation moves)

	p := gc.Prog(as)
	gc.Naddr(&p.From, f)
	gc.Naddr(&p.To, t)

	switch as {
	// Bad things the front end has done to us. Crash to find call stack.
	case s390x.AAND, s390x.AMULLD:
		if p.From.Type == obj.TYPE_CONST {
			gc.Debug['h'] = 1
			gc.Fatalf("bad inst: %v", p)
		}
	case s390x.ACMP, s390x.ACMPU:
		if p.From.Type == obj.TYPE_MEM || p.To.Type == obj.TYPE_MEM {
			gc.Debug['h'] = 1
			gc.Fatalf("bad inst: %v", p)
		}
	}

	if gc.Debug['g'] != 0 {
		fmt.Printf("%v\n", p)
	}

	w := int32(0)
	switch as {
	case s390x.AMOVB, s390x.AMOVBZ:
		w = 1

	case s390x.AMOVH, s390x.AMOVHZ:
		w = 2

	case s390x.AMOVW, s390x.AMOVWZ:
		w = 4

	case s390x.AMOVD:
		if p.From.Type == obj.TYPE_CONST || p.From.Type == obj.TYPE_ADDR {
			break
		}
		w = 8
	}

	if w != 0 && ((f != nil && p.From.Width < int64(w)) || (t != nil && p.To.Type != obj.TYPE_REG && p.To.Width > int64(w))) {
		gc.Dump("f", f)
		gc.Dump("t", t)
		gc.Fatalf("bad width: %v (%d, %d)\n", p, p.From.Width, p.To.Width)
	}

	return p
}
Exemplo n.º 16
0
Arquivo: gsubr.go Projeto: wycharry/go
// gmvc tries to move f to t using a mvc instruction.
// If successful it returns true, otherwise it returns false.
func gmvc(f, t *gc.Node) bool {
	ft := int(gc.Simsimtype(f.Type))
	tt := int(gc.Simsimtype(t.Type))

	if ft != tt {
		return false
	}

	if f.Op != gc.OINDREG || t.Op != gc.OINDREG {
		return false
	}

	if f.Xoffset < 0 || f.Xoffset >= 4096 {
		return false
	}

	if t.Xoffset < 0 || t.Xoffset >= 4096 {
		return false
	}

	var len int64
	switch ft {
	case gc.TUINT8, gc.TINT8, gc.TBOOL:
		len = 1
	case gc.TUINT16, gc.TINT16:
		len = 2
	case gc.TUINT32, gc.TINT32, gc.TFLOAT32:
		len = 4
	case gc.TUINT64, gc.TINT64, gc.TFLOAT64, gc.TPTR64:
		len = 8
	case gc.TUNSAFEPTR:
		len = int64(gc.Widthptr)
	default:
		return false
	}

	p := gc.Prog(s390x.AMVC)
	gc.Naddr(&p.From, f)
	gc.Naddr(&p.To, t)
	p.From3 = new(obj.Addr)
	p.From3.Offset = len
	p.From3.Type = obj.TYPE_CONST
	return true
}
Exemplo n.º 17
0
Arquivo: 387.go Projeto: achanda/go
// popAndSave pops a value off of the floating-point stack and stores
// it in the reigster assigned to v.
func popAndSave(s *gc.SSAGenState, v *ssa.Value) {
	r := v.Reg()
	if _, ok := s.SSEto387[r]; ok {
		// Pop value, write to correct register.
		p := gc.Prog(x86.AFMOVDP)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = x86.REG_F0
		p.To.Type = obj.TYPE_REG
		p.To.Reg = s.SSEto387[v.Reg()] + 1
	} else {
		// Don't actually pop value. This 387 register is now the
		// new home for the not-yet-assigned-a-home SSE register.
		// Increase the register mapping of all other registers by one.
		for rSSE, r387 := range s.SSEto387 {
			s.SSEto387[rSSE] = r387 + 1
		}
		s.SSEto387[r] = x86.REG_F0
	}
}
Exemplo n.º 18
0
Arquivo: gsubr.go Projeto: arnold8/go
/*
 * generate one instruction:
 *	as f, t
 */
func rawgins(as int, f *gc.Node, t *gc.Node) *obj.Prog {
	// TODO(austin): Add self-move test like in 6g (but be careful
	// of truncation moves)

	p := gc.Prog(as)
	gc.Naddr(&p.From, f)
	gc.Naddr(&p.To, t)

	switch as {
	case obj.ACALL:
		if p.To.Type == obj.TYPE_REG && p.To.Reg != ppc64.REG_CTR {
			// Allow front end to emit CALL REG, and rewrite into MOV REG, CTR; CALL CTR.
			if gc.Ctxt.Flag_dynlink {
				// Make sure function pointer is in R12 as well when
				// dynamically linking Go.
				// 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.
				q := gc.Prog(as)
				q.As = ppc64.AMOVD
				q.From = p.To
				q.To.Type = obj.TYPE_REG
				q.To.Reg = ppc64.REG_R12
			}
			pp := gc.Prog(as)
			pp.From = p.From
			pp.To.Type = obj.TYPE_REG
			pp.To.Reg = ppc64.REG_CTR

			p.As = ppc64.AMOVD
			p.From = p.To
			p.To.Type = obj.TYPE_REG
			p.To.Reg = ppc64.REG_CTR

			if gc.Debug['g'] != 0 {
				fmt.Printf("%v\n", p)
				fmt.Printf("%v\n", pp)
			}

			return pp
		}

	// Bad things the front end has done to us. Crash to find call stack.
	case ppc64.AAND, ppc64.AMULLD:
		if p.From.Type == obj.TYPE_CONST {
			gc.Debug['h'] = 1
			gc.Fatalf("bad inst: %v", p)
		}
	case ppc64.ACMP, ppc64.ACMPU:
		if p.From.Type == obj.TYPE_MEM || p.To.Type == obj.TYPE_MEM {
			gc.Debug['h'] = 1
			gc.Fatalf("bad inst: %v", p)
		}
	}

	if gc.Debug['g'] != 0 {
		fmt.Printf("%v\n", p)
	}

	w := int32(0)
	switch as {
	case ppc64.AMOVB,
		ppc64.AMOVBU,
		ppc64.AMOVBZ,
		ppc64.AMOVBZU:
		w = 1

	case ppc64.AMOVH,
		ppc64.AMOVHU,
		ppc64.AMOVHZ,
		ppc64.AMOVHZU:
		w = 2

	case ppc64.AMOVW,
		ppc64.AMOVWU,
		ppc64.AMOVWZ,
		ppc64.AMOVWZU:
		w = 4

	case ppc64.AMOVD,
		ppc64.AMOVDU:
		if p.From.Type == obj.TYPE_CONST || p.From.Type == obj.TYPE_ADDR {
			break
		}
		w = 8
	}

	if w != 0 && ((f != nil && p.From.Width < int64(w)) || (t != nil && p.To.Type != obj.TYPE_REG && p.To.Width > int64(w))) {
		gc.Dump("f", f)
		gc.Dump("t", t)
		gc.Fatalf("bad width: %v (%d, %d)\n", p, p.From.Width, p.To.Width)
	}

	return p
}
Exemplo n.º 19
0
Arquivo: gsubr.go Projeto: arnold8/go
/*
 * generate one instruction:
 *	as f, t
 */
func rawgins(as int, f *gc.Node, t *gc.Node) *obj.Prog {
	// TODO(austin): Add self-move test like in 6g (but be careful
	// of truncation moves)

	p := gc.Prog(as)
	gc.Naddr(&p.From, f)
	gc.Naddr(&p.To, t)

	switch as {
	case arm64.ACMP, arm64.AFCMPS, arm64.AFCMPD:
		if t != nil {
			if f.Op != gc.OREGISTER {
				gc.Fatalf("bad operands to gcmp")
			}
			p.From = p.To
			p.To = obj.Addr{}
			raddr(f, p)
		}
	}

	// Bad things the front end has done to us. Crash to find call stack.
	switch as {
	case arm64.AAND, arm64.AMUL:
		if p.From.Type == obj.TYPE_CONST {
			gc.Debug['h'] = 1
			gc.Fatalf("bad inst: %v", p)
		}
	case arm64.ACMP:
		if p.From.Type == obj.TYPE_MEM || p.To.Type == obj.TYPE_MEM {
			gc.Debug['h'] = 1
			gc.Fatalf("bad inst: %v", p)
		}
	}

	if gc.Debug['g'] != 0 {
		fmt.Printf("%v\n", p)
	}

	w := int32(0)
	switch as {
	case arm64.AMOVB,
		arm64.AMOVBU:
		w = 1

	case arm64.AMOVH,
		arm64.AMOVHU:
		w = 2

	case arm64.AMOVW,
		arm64.AMOVWU:
		w = 4

	case arm64.AMOVD:
		if p.From.Type == obj.TYPE_CONST || p.From.Type == obj.TYPE_ADDR {
			break
		}
		w = 8
	}

	if w != 0 && ((f != nil && p.From.Width < int64(w)) || (t != nil && p.To.Type != obj.TYPE_REG && p.To.Width > int64(w))) {
		gc.Dump("f", f)
		gc.Dump("t", t)
		gc.Fatalf("bad width: %v (%d, %d)\n", p, p.From.Width, p.To.Width)
	}

	return p
}
Exemplo n.º 20
0
Arquivo: gsubr.go Projeto: arnold8/go
/*
 * generate one instruction:
 *	as f, t
 */
func rawgins(as int, f *gc.Node, t *gc.Node) *obj.Prog {
	// TODO(austin): Add self-move test like in 6g (but be careful
	// of truncation moves)

	p := gc.Prog(as)
	gc.Naddr(&p.From, f)
	gc.Naddr(&p.To, t)

	switch as {
	case obj.ACALL:
		if p.To.Type == obj.TYPE_REG && p.To.Reg != ppc64.REG_CTR {
			// Allow front end to emit CALL REG, and rewrite into MOV REG, CTR; CALL CTR.
			pp := gc.Prog(as)
			pp.From = p.From
			pp.To.Type = obj.TYPE_REG
			pp.To.Reg = ppc64.REG_CTR

			p.As = ppc64.AMOVD
			p.From = p.To
			p.To.Type = obj.TYPE_REG
			p.To.Reg = ppc64.REG_CTR

			if gc.Debug['g'] != 0 {
				fmt.Printf("%v\n", p)
				fmt.Printf("%v\n", pp)
			}

			return pp
		}

	// Bad things the front end has done to us. Crash to find call stack.
	case ppc64.AAND, ppc64.AMULLD:
		if p.From.Type == obj.TYPE_CONST {
			gc.Debug['h'] = 1
			gc.Fatalf("bad inst: %v", p)
		}
	case ppc64.ACMP, ppc64.ACMPU:
		if p.From.Type == obj.TYPE_MEM || p.To.Type == obj.TYPE_MEM {
			gc.Debug['h'] = 1
			gc.Fatalf("bad inst: %v", p)
		}
	}

	if gc.Debug['g'] != 0 {
		fmt.Printf("%v\n", p)
	}

	w := int32(0)
	switch as {
	case ppc64.AMOVB,
		ppc64.AMOVBU,
		ppc64.AMOVBZ,
		ppc64.AMOVBZU:
		w = 1

	case ppc64.AMOVH,
		ppc64.AMOVHU,
		ppc64.AMOVHZ,
		ppc64.AMOVHZU:
		w = 2

	case ppc64.AMOVW,
		ppc64.AMOVWU,
		ppc64.AMOVWZ,
		ppc64.AMOVWZU:
		w = 4

	case ppc64.AMOVD,
		ppc64.AMOVDU:
		if p.From.Type == obj.TYPE_CONST || p.From.Type == obj.TYPE_ADDR {
			break
		}
		w = 8
	}

	if w != 0 && ((f != nil && p.From.Width < int64(w)) || (t != nil && p.To.Type != obj.TYPE_REG && p.To.Width > int64(w))) {
		gc.Dump("f", f)
		gc.Dump("t", t)
		gc.Fatalf("bad width: %v (%d, %d)\n", p, p.From.Width, p.To.Width)
	}

	return p
}
Exemplo n.º 21
0
Arquivo: gsubr.go Projeto: sreis/go
/*
 * generate one instruction:
 *	as f, t
 */
func gins(as int, f *gc.Node, t *gc.Node) *obj.Prog {
	if as == x86.AFMOVF && f != nil && f.Op == gc.OREGISTER && t != nil && t.Op == gc.OREGISTER {
		gc.Fatalf("gins MOVF reg, reg")
	}
	if as == x86.ACVTSD2SS && f != nil && f.Op == gc.OLITERAL {
		gc.Fatalf("gins CVTSD2SS const")
	}
	if as == x86.AMOVSD && t != nil && t.Op == gc.OREGISTER && t.Reg == x86.REG_F0 {
		gc.Fatalf("gins MOVSD into F0")
	}

	if as == x86.AMOVL && f != nil && f.Op == gc.OADDR && f.Left.Op == gc.ONAME && f.Left.Class != gc.PEXTERN && f.Left.Class != gc.PFUNC {
		// Turn MOVL $xxx(FP/SP) into LEAL xxx.
		// These should be equivalent but most of the backend
		// only expects to see LEAL, because that's what we had
		// historically generated. Various hidden assumptions are baked in by now.
		as = x86.ALEAL
		f = f.Left
	}

	switch as {
	case x86.AMOVB,
		x86.AMOVW,
		x86.AMOVL:
		if f != nil && t != nil && samaddr(f, t) {
			return nil
		}

	case x86.ALEAL:
		if f != nil && gc.Isconst(f, gc.CTNIL) {
			gc.Fatalf("gins LEAL nil %v", f.Type)
		}
	}

	p := gc.Prog(as)
	gc.Naddr(&p.From, f)
	gc.Naddr(&p.To, t)

	if gc.Debug['g'] != 0 {
		fmt.Printf("%v\n", p)
	}

	w := 0
	switch as {
	case x86.AMOVB:
		w = 1

	case x86.AMOVW:
		w = 2

	case x86.AMOVL:
		w = 4
	}

	if true && w != 0 && f != nil && (p.From.Width > int64(w) || p.To.Width > int64(w)) {
		gc.Dump("bad width from:", f)
		gc.Dump("bad width to:", t)
		gc.Fatalf("bad width: %v (%d, %d)\n", p, p.From.Width, p.To.Width)
	}

	if p.To.Type == obj.TYPE_ADDR && w > 0 {
		gc.Fatalf("bad use of addr: %v", p)
	}

	return p
}
Exemplo n.º 22
0
Arquivo: ssa.go Projeto: 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())
	}
}
Exemplo n.º 23
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())
	}
}
Exemplo n.º 24
0
func ssaGenBlock(s *gc.SSAGenState, b, next *ssa.Block) {
	s.SetLineno(b.Line)

	switch b.Kind {
	case ssa.BlockPlain, ssa.BlockCall, ssa.BlockCheck:
		if b.Succs[0] != next {
			p := gc.Prog(obj.AJMP)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{p, b.Succs[0]})
		}
	case ssa.BlockDefer:
		// defer returns in rax:
		// 0 if we should continue executing
		// 1 if we should jump to deferreturn call
		p := gc.Prog(x86.ATESTL)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = x86.REG_AX
		p.To.Type = obj.TYPE_REG
		p.To.Reg = x86.REG_AX
		p = gc.Prog(x86.AJNE)
		p.To.Type = obj.TYPE_BRANCH
		s.Branches = append(s.Branches, gc.Branch{p, b.Succs[1]})
		if b.Succs[0] != next {
			p := gc.Prog(obj.AJMP)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{p, b.Succs[0]})
		}
	case ssa.BlockExit:
		gc.Prog(obj.AUNDEF) // tell plive.go that we never reach here
	case ssa.BlockRet:
		gc.Prog(obj.ARET)
	case ssa.BlockRetJmp:
		p := gc.Prog(obj.AJMP)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(b.Aux.(*gc.Sym))

	case ssa.BlockAMD64EQF:
		gc.SSAGenFPJump(s, b, next, &eqfJumps)

	case ssa.BlockAMD64NEF:
		gc.SSAGenFPJump(s, b, next, &nefJumps)

	case ssa.BlockAMD64EQ, ssa.BlockAMD64NE,
		ssa.BlockAMD64LT, ssa.BlockAMD64GE,
		ssa.BlockAMD64LE, ssa.BlockAMD64GT,
		ssa.BlockAMD64ULT, ssa.BlockAMD64UGT,
		ssa.BlockAMD64ULE, ssa.BlockAMD64UGE:
		jmp := blockJump[b.Kind]
		likely := b.Likely
		var p *obj.Prog
		switch next {
		case b.Succs[0]:
			p = gc.Prog(jmp.invasm)
			likely *= -1
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{p, b.Succs[1]})
		case b.Succs[1]:
			p = gc.Prog(jmp.asm)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{p, b.Succs[0]})
		default:
			p = gc.Prog(jmp.asm)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{p, b.Succs[0]})
			q := gc.Prog(obj.AJMP)
			q.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{q, b.Succs[1]})
		}

		// liblink reorders the instruction stream as it sees fit.
		// Pass along what we know so liblink can make use of it.
		// TODO: Once we've fully switched to SSA,
		// make liblink leave our output alone.
		switch likely {
		case ssa.BranchUnlikely:
			p.From.Type = obj.TYPE_CONST
			p.From.Offset = 0
		case ssa.BranchLikely:
			p.From.Type = obj.TYPE_CONST
			p.From.Offset = 1
		}

	default:
		b.Unimplementedf("branch not implemented: %s. Control: %s", b.LongString(), b.Control.LongString())
	}
}
Exemplo n.º 25
0
/*
 * generate one instruction:
 *	as f, t
 */
func gins(as int, f *gc.Node, t *gc.Node) *obj.Prog {
	//	Node nod;
	//	int32 v;

	if f != nil && f.Op == gc.OINDEX {
		gc.Fatal("gins OINDEX not implemented")
	}

	//		gc.Regalloc(&nod, &regnode, Z);
	//		v = constnode.vconst;
	//		gc.Cgen(f->right, &nod);
	//		constnode.vconst = v;
	//		idx.reg = nod.reg;
	//		gc.Regfree(&nod);
	if t != nil && t.Op == gc.OINDEX {
		gc.Fatal("gins OINDEX not implemented")
	}

	//		gc.Regalloc(&nod, &regnode, Z);
	//		v = constnode.vconst;
	//		gc.Cgen(t->right, &nod);
	//		constnode.vconst = v;
	//		idx.reg = nod.reg;
	//		gc.Regfree(&nod);

	p := gc.Prog(as)
	gc.Naddr(&p.From, f)
	gc.Naddr(&p.To, t)

	switch as {
	case arm.ABL:
		if p.To.Type == obj.TYPE_REG {
			p.To.Type = obj.TYPE_MEM
		}

	case arm.ACMP, arm.ACMPF, arm.ACMPD:
		if t != nil {
			if f.Op != gc.OREGISTER {
				/* generate a comparison
				TODO(kaib): one of the args can actually be a small constant. relax the constraint and fix call sites.
				*/
				gc.Fatal("bad operands to gcmp")
			}
			p.From = p.To
			p.To = obj.Addr{}
			raddr(f, p)
		}

	case arm.AMULU:
		if f != nil && f.Op != gc.OREGISTER {
			gc.Fatal("bad operands to mul")
		}

	case arm.AMOVW:
		if (p.From.Type == obj.TYPE_MEM || p.From.Type == obj.TYPE_ADDR || p.From.Type == obj.TYPE_CONST) && (p.To.Type == obj.TYPE_MEM || p.To.Type == obj.TYPE_ADDR) {
			gc.Fatal("gins double memory")
		}

	case arm.AADD:
		if p.To.Type == obj.TYPE_MEM {
			gc.Fatal("gins arith to mem")
		}

	case arm.ARSB:
		if p.From.Type == obj.TYPE_NONE {
			gc.Fatal("rsb with no from")
		}
	}

	if gc.Debug['g'] != 0 {
		fmt.Printf("%v\n", p)
	}
	return p
}
Exemplo n.º 26
0
Arquivo: ssa.go Projeto: Harvey-OS/go
func ssaGenBlock(s *gc.SSAGenState, b, next *ssa.Block) {
	s.SetLineno(b.Line)

	switch b.Kind {
	case ssa.BlockPlain:
		if b.Succs[0].Block() != next {
			p := gc.Prog(obj.AJMP)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{P: p, B: b.Succs[0].Block()})
		}
	case ssa.BlockDefer:
		// defer returns in R1:
		// 0 if we should continue executing
		// 1 if we should jump to deferreturn call
		p := gc.Prog(mips.ABNE)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = mips.REGZERO
		p.Reg = mips.REG_R1
		p.To.Type = obj.TYPE_BRANCH
		s.Branches = append(s.Branches, gc.Branch{P: p, B: b.Succs[1].Block()})
		if b.Succs[0].Block() != next {
			p := gc.Prog(obj.AJMP)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{P: p, B: b.Succs[0].Block()})
		}
	case ssa.BlockExit:
		gc.Prog(obj.AUNDEF) // tell plive.go that we never reach here
	case ssa.BlockRet:
		gc.Prog(obj.ARET)
	case ssa.BlockRetJmp:
		p := gc.Prog(obj.ARET)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(b.Aux.(*gc.Sym))
	case ssa.BlockMIPSEQ, ssa.BlockMIPSNE,
		ssa.BlockMIPSLTZ, ssa.BlockMIPSGEZ,
		ssa.BlockMIPSLEZ, ssa.BlockMIPSGTZ,
		ssa.BlockMIPSFPT, ssa.BlockMIPSFPF:
		jmp := blockJump[b.Kind]
		var p *obj.Prog
		switch next {
		case b.Succs[0].Block():
			p = gc.Prog(jmp.invasm)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{P: p, B: b.Succs[1].Block()})
		case b.Succs[1].Block():
			p = gc.Prog(jmp.asm)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{P: p, B: b.Succs[0].Block()})
		default:
			p = gc.Prog(jmp.asm)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{P: p, B: b.Succs[0].Block()})
			q := gc.Prog(obj.AJMP)
			q.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{P: q, B: b.Succs[1].Block()})
		}
		if !b.Control.Type.IsFlags() {
			p.From.Type = obj.TYPE_REG
			p.From.Reg = b.Control.Reg()
		}
	default:
		b.Fatalf("branch not implemented: %s. Control: %s", b.LongString(), b.Control.LongString())
	}
}
Exemplo n.º 27
0
Arquivo: ssa.go Projeto: achanda/go
func ssaGenBlock(s *gc.SSAGenState, b, next *ssa.Block) {
	s.SetLineno(b.Line)

	switch b.Kind {
	case ssa.BlockPlain:
		if b.Succs[0].Block() != next {
			p := gc.Prog(obj.AJMP)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{P: p, B: b.Succs[0].Block()})
		}

	case ssa.BlockDefer:
		// defer returns in R0:
		// 0 if we should continue executing
		// 1 if we should jump to deferreturn call
		p := gc.Prog(arm.ACMP)
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = 0
		p.Reg = arm.REG_R0
		p = gc.Prog(arm.ABNE)
		p.To.Type = obj.TYPE_BRANCH
		s.Branches = append(s.Branches, gc.Branch{P: p, B: b.Succs[1].Block()})
		if b.Succs[0].Block() != next {
			p := gc.Prog(obj.AJMP)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{P: p, B: b.Succs[0].Block()})
		}

	case ssa.BlockExit:
		gc.Prog(obj.AUNDEF) // tell plive.go that we never reach here

	case ssa.BlockRet:
		gc.Prog(obj.ARET)

	case ssa.BlockRetJmp:
		p := gc.Prog(obj.ARET)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(b.Aux.(*gc.Sym))

	case ssa.BlockARMEQ, ssa.BlockARMNE,
		ssa.BlockARMLT, ssa.BlockARMGE,
		ssa.BlockARMLE, ssa.BlockARMGT,
		ssa.BlockARMULT, ssa.BlockARMUGT,
		ssa.BlockARMULE, ssa.BlockARMUGE:
		jmp := blockJump[b.Kind]
		var p *obj.Prog
		switch next {
		case b.Succs[0].Block():
			p = gc.Prog(jmp.invasm)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{P: p, B: b.Succs[1].Block()})
		case b.Succs[1].Block():
			p = gc.Prog(jmp.asm)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{P: p, B: b.Succs[0].Block()})
		default:
			p = gc.Prog(jmp.asm)
			p.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{P: p, B: b.Succs[0].Block()})
			q := gc.Prog(obj.AJMP)
			q.To.Type = obj.TYPE_BRANCH
			s.Branches = append(s.Branches, gc.Branch{P: q, B: b.Succs[1].Block()})
		}

	default:
		b.Fatalf("branch not implemented: %s. Control: %s", b.LongString(), b.Control.LongString())
	}
}
Exemplo n.º 28
0
Arquivo: ssa.go Projeto: achanda/go
func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
	s.SetLineno(v.Line)
	switch v.Op {
	case ssa.OpInitMem:
		// memory arg needs no code
	case ssa.OpArg:
		// input args need no code
	case ssa.OpSP, ssa.OpSB, ssa.OpGetG:
		// nothing to do
	case ssa.OpCopy, ssa.OpARMMOVWconvert, ssa.OpARMMOVWreg:
		if v.Type.IsMemory() {
			return
		}
		x := v.Args[0].Reg()
		y := v.Reg()
		if x == y {
			return
		}
		as := arm.AMOVW
		if v.Type.IsFloat() {
			switch v.Type.Size() {
			case 4:
				as = arm.AMOVF
			case 8:
				as = arm.AMOVD
			default:
				panic("bad float size")
			}
		}
		p := gc.Prog(as)
		p.From.Type = obj.TYPE_REG
		p.From.Reg = x
		p.To.Type = obj.TYPE_REG
		p.To.Reg = y
	case ssa.OpARMMOVWnop:
		if v.Reg() != v.Args[0].Reg() {
			v.Fatalf("input[0] and output not in same register %s", v.LongString())
		}
		// nothing to do
	case ssa.OpLoadReg:
		if v.Type.IsFlags() {
			v.Fatalf("load flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(loadByType(v.Type))
		gc.AddrAuto(&p.From, v.Args[0])
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpPhi:
		gc.CheckLoweredPhi(v)
	case ssa.OpStoreReg:
		if v.Type.IsFlags() {
			v.Fatalf("store flags not implemented: %v", v.LongString())
			return
		}
		p := gc.Prog(storeByType(v.Type))
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		gc.AddrAuto(&p.To, v)
	case ssa.OpARMUDIVrtcall:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = obj.Linklookup(gc.Ctxt, "udiv", 0)
	case ssa.OpARMADD,
		ssa.OpARMADC,
		ssa.OpARMSUB,
		ssa.OpARMSBC,
		ssa.OpARMRSB,
		ssa.OpARMAND,
		ssa.OpARMOR,
		ssa.OpARMXOR,
		ssa.OpARMBIC,
		ssa.OpARMMUL,
		ssa.OpARMADDF,
		ssa.OpARMADDD,
		ssa.OpARMSUBF,
		ssa.OpARMSUBD,
		ssa.OpARMMULF,
		ssa.OpARMMULD,
		ssa.OpARMDIVF,
		ssa.OpARMDIVD:
		r := v.Reg()
		r1 := v.Args[0].Reg()
		r2 := v.Args[1].Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r2
		p.Reg = r1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpARMADDS,
		ssa.OpARMSUBS:
		r := v.Reg0()
		r1 := v.Args[0].Reg()
		r2 := v.Args[1].Reg()
		p := gc.Prog(v.Op.Asm())
		p.Scond = arm.C_SBIT
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r2
		p.Reg = r1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpARMSLL,
		ssa.OpARMSRL,
		ssa.OpARMSRA:
		r := v.Reg()
		r1 := v.Args[0].Reg()
		r2 := v.Args[1].Reg()
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r2
		p.Reg = r1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpARMSRAcond:
		// ARM shift instructions uses only the low-order byte of the shift amount
		// generate conditional instructions to deal with large shifts
		// flag is already set
		// SRA.HS	$31, Rarg0, Rdst // shift 31 bits to get the sign bit
		// SRA.LO	Rarg1, Rarg0, Rdst
		r := v.Reg()
		r1 := v.Args[0].Reg()
		r2 := v.Args[1].Reg()
		p := gc.Prog(arm.ASRA)
		p.Scond = arm.C_SCOND_HS
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = 31
		p.Reg = r1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
		p = gc.Prog(arm.ASRA)
		p.Scond = arm.C_SCOND_LO
		p.From.Type = obj.TYPE_REG
		p.From.Reg = r2
		p.Reg = r1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = r
	case ssa.OpARMADDconst,
		ssa.OpARMADCconst,
		ssa.OpARMSUBconst,
		ssa.OpARMSBCconst,
		ssa.OpARMRSBconst,
		ssa.OpARMRSCconst,
		ssa.OpARMANDconst,
		ssa.OpARMORconst,
		ssa.OpARMXORconst,
		ssa.OpARMBICconst,
		ssa.OpARMSLLconst,
		ssa.OpARMSRLconst,
		ssa.OpARMSRAconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMADDSconst,
		ssa.OpARMSUBSconst,
		ssa.OpARMRSBSconst:
		p := gc.Prog(v.Op.Asm())
		p.Scond = arm.C_SBIT
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg0()
	case ssa.OpARMSRRconst:
		genshift(arm.AMOVW, 0, v.Args[0].Reg(), v.Reg(), arm.SHIFT_RR, v.AuxInt)
	case ssa.OpARMADDshiftLL,
		ssa.OpARMADCshiftLL,
		ssa.OpARMSUBshiftLL,
		ssa.OpARMSBCshiftLL,
		ssa.OpARMRSBshiftLL,
		ssa.OpARMRSCshiftLL,
		ssa.OpARMANDshiftLL,
		ssa.OpARMORshiftLL,
		ssa.OpARMXORshiftLL,
		ssa.OpARMBICshiftLL:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm.SHIFT_LL, v.AuxInt)
	case ssa.OpARMADDSshiftLL,
		ssa.OpARMSUBSshiftLL,
		ssa.OpARMRSBSshiftLL:
		p := genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg0(), arm.SHIFT_LL, v.AuxInt)
		p.Scond = arm.C_SBIT
	case ssa.OpARMADDshiftRL,
		ssa.OpARMADCshiftRL,
		ssa.OpARMSUBshiftRL,
		ssa.OpARMSBCshiftRL,
		ssa.OpARMRSBshiftRL,
		ssa.OpARMRSCshiftRL,
		ssa.OpARMANDshiftRL,
		ssa.OpARMORshiftRL,
		ssa.OpARMXORshiftRL,
		ssa.OpARMBICshiftRL:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm.SHIFT_LR, v.AuxInt)
	case ssa.OpARMADDSshiftRL,
		ssa.OpARMSUBSshiftRL,
		ssa.OpARMRSBSshiftRL:
		p := genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg0(), arm.SHIFT_LR, v.AuxInt)
		p.Scond = arm.C_SBIT
	case ssa.OpARMADDshiftRA,
		ssa.OpARMADCshiftRA,
		ssa.OpARMSUBshiftRA,
		ssa.OpARMSBCshiftRA,
		ssa.OpARMRSBshiftRA,
		ssa.OpARMRSCshiftRA,
		ssa.OpARMANDshiftRA,
		ssa.OpARMORshiftRA,
		ssa.OpARMXORshiftRA,
		ssa.OpARMBICshiftRA:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm.SHIFT_AR, v.AuxInt)
	case ssa.OpARMADDSshiftRA,
		ssa.OpARMSUBSshiftRA,
		ssa.OpARMRSBSshiftRA:
		p := genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg0(), arm.SHIFT_AR, v.AuxInt)
		p.Scond = arm.C_SBIT
	case ssa.OpARMXORshiftRR:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm.SHIFT_RR, v.AuxInt)
	case ssa.OpARMMVNshiftLL:
		genshift(v.Op.Asm(), 0, v.Args[0].Reg(), v.Reg(), arm.SHIFT_LL, v.AuxInt)
	case ssa.OpARMMVNshiftRL:
		genshift(v.Op.Asm(), 0, v.Args[0].Reg(), v.Reg(), arm.SHIFT_LR, v.AuxInt)
	case ssa.OpARMMVNshiftRA:
		genshift(v.Op.Asm(), 0, v.Args[0].Reg(), v.Reg(), arm.SHIFT_AR, v.AuxInt)
	case ssa.OpARMMVNshiftLLreg:
		genregshift(v.Op.Asm(), 0, v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm.SHIFT_LL)
	case ssa.OpARMMVNshiftRLreg:
		genregshift(v.Op.Asm(), 0, v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm.SHIFT_LR)
	case ssa.OpARMMVNshiftRAreg:
		genregshift(v.Op.Asm(), 0, v.Args[0].Reg(), v.Args[1].Reg(), v.Reg(), arm.SHIFT_AR)
	case ssa.OpARMADDshiftLLreg,
		ssa.OpARMADCshiftLLreg,
		ssa.OpARMSUBshiftLLreg,
		ssa.OpARMSBCshiftLLreg,
		ssa.OpARMRSBshiftLLreg,
		ssa.OpARMRSCshiftLLreg,
		ssa.OpARMANDshiftLLreg,
		ssa.OpARMORshiftLLreg,
		ssa.OpARMXORshiftLLreg,
		ssa.OpARMBICshiftLLreg:
		genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), v.Reg(), arm.SHIFT_LL)
	case ssa.OpARMADDSshiftLLreg,
		ssa.OpARMSUBSshiftLLreg,
		ssa.OpARMRSBSshiftLLreg:
		p := genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), v.Reg0(), arm.SHIFT_LL)
		p.Scond = arm.C_SBIT
	case ssa.OpARMADDshiftRLreg,
		ssa.OpARMADCshiftRLreg,
		ssa.OpARMSUBshiftRLreg,
		ssa.OpARMSBCshiftRLreg,
		ssa.OpARMRSBshiftRLreg,
		ssa.OpARMRSCshiftRLreg,
		ssa.OpARMANDshiftRLreg,
		ssa.OpARMORshiftRLreg,
		ssa.OpARMXORshiftRLreg,
		ssa.OpARMBICshiftRLreg:
		genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), v.Reg(), arm.SHIFT_LR)
	case ssa.OpARMADDSshiftRLreg,
		ssa.OpARMSUBSshiftRLreg,
		ssa.OpARMRSBSshiftRLreg:
		p := genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), v.Reg0(), arm.SHIFT_LR)
		p.Scond = arm.C_SBIT
	case ssa.OpARMADDshiftRAreg,
		ssa.OpARMADCshiftRAreg,
		ssa.OpARMSUBshiftRAreg,
		ssa.OpARMSBCshiftRAreg,
		ssa.OpARMRSBshiftRAreg,
		ssa.OpARMRSCshiftRAreg,
		ssa.OpARMANDshiftRAreg,
		ssa.OpARMORshiftRAreg,
		ssa.OpARMXORshiftRAreg,
		ssa.OpARMBICshiftRAreg:
		genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), v.Reg(), arm.SHIFT_AR)
	case ssa.OpARMADDSshiftRAreg,
		ssa.OpARMSUBSshiftRAreg,
		ssa.OpARMRSBSshiftRAreg:
		p := genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), v.Reg0(), arm.SHIFT_AR)
		p.Scond = arm.C_SBIT
	case ssa.OpARMHMUL,
		ssa.OpARMHMULU:
		// 32-bit high multiplication
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REGREG
		p.To.Reg = v.Reg()
		p.To.Offset = arm.REGTMP // throw away low 32-bit into tmp register
	case ssa.OpARMMULLU:
		// 32-bit multiplication, results 64-bit, high 32-bit in out0, low 32-bit in out1
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REGREG
		p.To.Reg = v.Reg0()           // high 32-bit
		p.To.Offset = int64(v.Reg1()) // low 32-bit
	case ssa.OpARMMULA:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_REGREG2
		p.To.Reg = v.Reg()                   // result
		p.To.Offset = int64(v.Args[2].Reg()) // addend
	case ssa.OpARMMOVWconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMMOVFconst,
		ssa.OpARMMOVDconst:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_FCONST
		p.From.Val = math.Float64frombits(uint64(v.AuxInt))
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMCMP,
		ssa.OpARMCMN,
		ssa.OpARMTST,
		ssa.OpARMTEQ,
		ssa.OpARMCMPF,
		ssa.OpARMCMPD:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		// Special layout in ARM assembly
		// Comparing to x86, the operands of ARM's CMP are reversed.
		p.From.Reg = v.Args[1].Reg()
		p.Reg = v.Args[0].Reg()
	case ssa.OpARMCMPconst,
		ssa.OpARMCMNconst,
		ssa.OpARMTSTconst,
		ssa.OpARMTEQconst:
		// Special layout in ARM assembly
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.Reg = v.Args[0].Reg()
	case ssa.OpARMCMPF0,
		ssa.OpARMCMPD0:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
	case ssa.OpARMCMPshiftLL:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), 0, arm.SHIFT_LL, v.AuxInt)
	case ssa.OpARMCMPshiftRL:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), 0, arm.SHIFT_LR, v.AuxInt)
	case ssa.OpARMCMPshiftRA:
		genshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), 0, arm.SHIFT_AR, v.AuxInt)
	case ssa.OpARMCMPshiftLLreg:
		genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), 0, arm.SHIFT_LL)
	case ssa.OpARMCMPshiftRLreg:
		genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), 0, arm.SHIFT_LR)
	case ssa.OpARMCMPshiftRAreg:
		genregshift(v.Op.Asm(), v.Args[0].Reg(), v.Args[1].Reg(), v.Args[2].Reg(), 0, arm.SHIFT_AR)
	case ssa.OpARMMOVWaddr:
		p := gc.Prog(arm.AMOVW)
		p.From.Type = obj.TYPE_ADDR
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()

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

	case ssa.OpARMMOVBload,
		ssa.OpARMMOVBUload,
		ssa.OpARMMOVHload,
		ssa.OpARMMOVHUload,
		ssa.OpARMMOVWload,
		ssa.OpARMMOVFload,
		ssa.OpARMMOVDload:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMMOVBstore,
		ssa.OpARMMOVHstore,
		ssa.OpARMMOVWstore,
		ssa.OpARMMOVFstore,
		ssa.OpARMMOVDstore:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[1].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = v.Args[0].Reg()
		gc.AddAux(&p.To, v)
	case ssa.OpARMMOVWloadidx:
		// this is just shift 0 bits
		fallthrough
	case ssa.OpARMMOVWloadshiftLL:
		p := genshift(v.Op.Asm(), 0, v.Args[1].Reg(), v.Reg(), arm.SHIFT_LL, v.AuxInt)
		p.From.Reg = v.Args[0].Reg()
	case ssa.OpARMMOVWloadshiftRL:
		p := genshift(v.Op.Asm(), 0, v.Args[1].Reg(), v.Reg(), arm.SHIFT_LR, v.AuxInt)
		p.From.Reg = v.Args[0].Reg()
	case ssa.OpARMMOVWloadshiftRA:
		p := genshift(v.Op.Asm(), 0, v.Args[1].Reg(), v.Reg(), arm.SHIFT_AR, v.AuxInt)
		p.From.Reg = v.Args[0].Reg()
	case ssa.OpARMMOVWstoreidx:
		// this is just shift 0 bits
		fallthrough
	case ssa.OpARMMOVWstoreshiftLL:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_SHIFT
		p.To.Reg = v.Args[0].Reg()
		p.To.Offset = int64(makeshift(v.Args[1].Reg(), arm.SHIFT_LL, v.AuxInt))
	case ssa.OpARMMOVWstoreshiftRL:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_SHIFT
		p.To.Reg = v.Args[0].Reg()
		p.To.Offset = int64(makeshift(v.Args[1].Reg(), arm.SHIFT_LR, v.AuxInt))
	case ssa.OpARMMOVWstoreshiftRA:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_SHIFT
		p.To.Reg = v.Args[0].Reg()
		p.To.Offset = int64(makeshift(v.Args[1].Reg(), arm.SHIFT_AR, v.AuxInt))
	case ssa.OpARMMOVBreg,
		ssa.OpARMMOVBUreg,
		ssa.OpARMMOVHreg,
		ssa.OpARMMOVHUreg:
		a := v.Args[0]
		for a.Op == ssa.OpCopy || a.Op == ssa.OpARMMOVWreg || a.Op == ssa.OpARMMOVWnop {
			a = a.Args[0]
		}
		if a.Op == ssa.OpLoadReg {
			t := a.Type
			switch {
			case v.Op == ssa.OpARMMOVBreg && t.Size() == 1 && t.IsSigned(),
				v.Op == ssa.OpARMMOVBUreg && t.Size() == 1 && !t.IsSigned(),
				v.Op == ssa.OpARMMOVHreg && t.Size() == 2 && t.IsSigned(),
				v.Op == ssa.OpARMMOVHUreg && t.Size() == 2 && !t.IsSigned():
				// arg is a proper-typed load, already zero/sign-extended, don't extend again
				if v.Reg() == v.Args[0].Reg() {
					return
				}
				p := gc.Prog(arm.AMOVW)
				p.From.Type = obj.TYPE_REG
				p.From.Reg = v.Args[0].Reg()
				p.To.Type = obj.TYPE_REG
				p.To.Reg = v.Reg()
				return
			default:
			}
		}
		fallthrough
	case ssa.OpARMMVN,
		ssa.OpARMCLZ,
		ssa.OpARMSQRTD,
		ssa.OpARMNEGF,
		ssa.OpARMNEGD,
		ssa.OpARMMOVWF,
		ssa.OpARMMOVWD,
		ssa.OpARMMOVFW,
		ssa.OpARMMOVDW,
		ssa.OpARMMOVFD,
		ssa.OpARMMOVDF:
		p := gc.Prog(v.Op.Asm())
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMMOVWUF,
		ssa.OpARMMOVWUD,
		ssa.OpARMMOVFWU,
		ssa.OpARMMOVDWU:
		p := gc.Prog(v.Op.Asm())
		p.Scond = arm.C_UBIT
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[0].Reg()
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMCMOVWHSconst:
		p := gc.Prog(arm.AMOVW)
		p.Scond = arm.C_SCOND_HS
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMCMOVWLSconst:
		p := gc.Prog(arm.AMOVW)
		p.Scond = arm.C_SCOND_LS
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = v.AuxInt
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpARMCALLstatic:
		if v.Aux.(*gc.Sym) == gc.Deferreturn.Sym {
			// Deferred calls will appear to be returning to
			// the CALL deferreturn(SB) that we are about to emit.
			// However, the stack trace code will show the line
			// of the instruction byte before the return PC.
			// To avoid that being an unrelated instruction,
			// insert an actual hardware NOP that will have the right line number.
			// This is different from obj.ANOP, which is a virtual no-op
			// that doesn't make it into the instruction stream.
			ginsnop()
		}
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(v.Aux.(*gc.Sym))
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARMCALLclosure:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Offset = 0
		p.To.Reg = v.Args[0].Reg()
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARMCALLdefer:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Deferproc.Sym)
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARMCALLgo:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Newproc.Sym)
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARMCALLinter:
		p := gc.Prog(obj.ACALL)
		p.To.Type = obj.TYPE_MEM
		p.To.Offset = 0
		p.To.Reg = v.Args[0].Reg()
		if gc.Maxarg < v.AuxInt {
			gc.Maxarg = v.AuxInt
		}
	case ssa.OpARMDUFFZERO:
		p := gc.Prog(obj.ADUFFZERO)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Pkglookup("duffzero", gc.Runtimepkg))
		p.To.Offset = v.AuxInt
	case ssa.OpARMDUFFCOPY:
		p := gc.Prog(obj.ADUFFCOPY)
		p.To.Type = obj.TYPE_MEM
		p.To.Name = obj.NAME_EXTERN
		p.To.Sym = gc.Linksym(gc.Pkglookup("duffcopy", gc.Runtimepkg))
		p.To.Offset = v.AuxInt
	case ssa.OpARMLoweredNilCheck:
		// Issue a load which will fault if arg is nil.
		p := gc.Prog(arm.AMOVB)
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = v.Args[0].Reg()
		gc.AddAux(&p.From, v)
		p.To.Type = obj.TYPE_REG
		p.To.Reg = arm.REGTMP
		if gc.Debug_checknil != 0 && v.Line > 1 { // v.Line==1 in generated wrappers
			gc.Warnl(v.Line, "generated nil check")
		}
	case ssa.OpARMLoweredZero:
		// MOVW.P	Rarg2, 4(R1)
		// CMP	Rarg1, R1
		// BLE	-2(PC)
		// arg1 is the address of the last element to zero
		// arg2 is known to be zero
		// auxint is alignment
		var sz int64
		var mov obj.As
		switch {
		case v.AuxInt%4 == 0:
			sz = 4
			mov = arm.AMOVW
		case v.AuxInt%2 == 0:
			sz = 2
			mov = arm.AMOVH
		default:
			sz = 1
			mov = arm.AMOVB
		}
		p := gc.Prog(mov)
		p.Scond = arm.C_PBIT
		p.From.Type = obj.TYPE_REG
		p.From.Reg = v.Args[2].Reg()
		p.To.Type = obj.TYPE_MEM
		p.To.Reg = arm.REG_R1
		p.To.Offset = sz
		p2 := gc.Prog(arm.ACMP)
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = v.Args[1].Reg()
		p2.Reg = arm.REG_R1
		p3 := gc.Prog(arm.ABLE)
		p3.To.Type = obj.TYPE_BRANCH
		gc.Patch(p3, p)
	case ssa.OpARMLoweredMove:
		// MOVW.P	4(R1), Rtmp
		// MOVW.P	Rtmp, 4(R2)
		// CMP	Rarg2, R1
		// BLE	-3(PC)
		// arg2 is the address of the last element of src
		// auxint is alignment
		var sz int64
		var mov obj.As
		switch {
		case v.AuxInt%4 == 0:
			sz = 4
			mov = arm.AMOVW
		case v.AuxInt%2 == 0:
			sz = 2
			mov = arm.AMOVH
		default:
			sz = 1
			mov = arm.AMOVB
		}
		p := gc.Prog(mov)
		p.Scond = arm.C_PBIT
		p.From.Type = obj.TYPE_MEM
		p.From.Reg = arm.REG_R1
		p.From.Offset = sz
		p.To.Type = obj.TYPE_REG
		p.To.Reg = arm.REGTMP
		p2 := gc.Prog(mov)
		p2.Scond = arm.C_PBIT
		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = arm.REGTMP
		p2.To.Type = obj.TYPE_MEM
		p2.To.Reg = arm.REG_R2
		p2.To.Offset = sz
		p3 := gc.Prog(arm.ACMP)
		p3.From.Type = obj.TYPE_REG
		p3.From.Reg = v.Args[2].Reg()
		p3.Reg = arm.REG_R1
		p4 := gc.Prog(arm.ABLE)
		p4.To.Type = obj.TYPE_BRANCH
		gc.Patch(p4, p)
	case ssa.OpVarDef:
		gc.Gvardef(v.Aux.(*gc.Node))
	case ssa.OpVarKill:
		gc.Gvarkill(v.Aux.(*gc.Node))
	case ssa.OpVarLive:
		gc.Gvarlive(v.Aux.(*gc.Node))
	case ssa.OpKeepAlive:
		gc.KeepAlive(v)
	case ssa.OpARMEqual,
		ssa.OpARMNotEqual,
		ssa.OpARMLessThan,
		ssa.OpARMLessEqual,
		ssa.OpARMGreaterThan,
		ssa.OpARMGreaterEqual,
		ssa.OpARMLessThanU,
		ssa.OpARMLessEqualU,
		ssa.OpARMGreaterThanU,
		ssa.OpARMGreaterEqualU:
		// generate boolean values
		// use conditional move
		p := gc.Prog(arm.AMOVW)
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = 0
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
		p = gc.Prog(arm.AMOVW)
		p.Scond = condBits[v.Op]
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = 1
		p.To.Type = obj.TYPE_REG
		p.To.Reg = v.Reg()
	case ssa.OpSelect0, ssa.OpSelect1:
		// nothing to do
	case ssa.OpARMLoweredGetClosurePtr:
		// Closure pointer is R7 (arm.REGCTXT).
		gc.CheckLoweredGetClosurePtr(v)
	case ssa.OpARMFlagEQ,
		ssa.OpARMFlagLT_ULT,
		ssa.OpARMFlagLT_UGT,
		ssa.OpARMFlagGT_ULT,
		ssa.OpARMFlagGT_UGT:
		v.Fatalf("Flag* ops should never make it to codegen %v", v.LongString())
	case ssa.OpARMInvertFlags:
		v.Fatalf("InvertFlags should never make it to codegen %v", v.LongString())
	default:
		v.Fatalf("genValue not implemented: %s", v.LongString())
	}
}
Exemplo n.º 29
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())
	}
}
Exemplo n.º 30
0
Arquivo: ssa.go Projeto: Harvey-OS/go
func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
	s.SetLineno(v.Line)
	switch v.Op {
	case ssa.OpInitMem:
		// memory arg needs no code
	case ssa.OpArg:
		// input args need no code
	case ssa.OpSP, ssa.OpSB, ssa.OpGetG:
		// nothing to do
	case ssa.OpSelect0, ssa.OpSelect1:
		// nothing to do
	case ssa.OpCopy, ssa.OpMIPSMOVWconvert, ssa.OpMIPSMOVWreg:
		t := v.Type
		if t.IsMemory() {
			return
		}
		x := v.Args[0].Reg()
		y := v.Reg()
		if x == y {
			return
		}
		as := mips.AMOVW
		if isFPreg(x) && isFPreg(y) {
			as = mips.AMOVF
			if t.Size() == 8 {
				as = mips.AMOVD
			}
		}

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

		gc.Prog(mips.ASYNC)

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

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

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

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

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

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

		gc.Prog(mips.ASYNC)

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

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

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

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

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

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

		gc.Prog(mips.ASYNC)

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

		gc.Prog(mips.ASYNC)

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

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

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

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

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

		gc.Prog(mips.ASYNC)

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

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

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

	case ssa.OpMIPSLoweredGetClosurePtr:
		// Closure pointer is R22 (mips.REGCTXT).
		gc.CheckLoweredGetClosurePtr(v)
	default:
		v.Fatalf("genValue not implemented: %s", v.LongString())
	}
}