Golang Prog.From方法代码示例

func outgcode(a int, g1 *obj.Addr, reg int, g2, g3 *obj.Addr) {
	var p *obj.Prog
	var pl *obj.Plist

	if asm.Pass == 1 {
		goto out
	}

	p = asm.Ctxt.NewProg()
	p.As = int16(a)
	p.Lineno = stmtline
	if nosched != 0 {
		p.Mark |= ppc64.NOSCHED
	}
	p.From = *g1
	p.Reg = int16(reg)
	p.From3 = *g2
	p.To = *g3
	p.Pc = int64(asm.PC)

	if lastpc == nil {
		pl = obj.Linknewplist(asm.Ctxt)
		pl.Firstpc = p
	} else {
		lastpc.Link = p
	}
	lastpc = p

out:
	if a != obj.AGLOBL && a != obj.ADATA {
		asm.PC++
	}
}

func outcode(a int, g2 *Addr2) {
	var p *obj.Prog
	var pl *obj.Plist

	if asm.Pass == 1 {
		goto out
	}

	p = new(obj.Prog)
	*p = obj.Prog{}
	p.Ctxt = asm.Ctxt
	p.As = int16(a)
	p.Lineno = stmtline
	p.From = g2.from
	p.To = g2.to
	p.Pc = int64(asm.PC)

	if lastpc == nil {
		pl = obj.Linknewplist(asm.Ctxt)
		pl.Firstpc = p
	} else {

		lastpc.Link = p
	}
	lastpc = p

out:
	if a != obj.AGLOBL && a != obj.ADATA {
		asm.PC++
	}
}

func outcode(a int, g1 *obj.Addr, reg int, g2 *obj.Addr) {
	var p *obj.Prog
	var pl *obj.Plist

	if asm.Pass == 1 {
		goto out
	}

	if g1.Scale != 0 {
		if reg != 0 || g2.Scale != 0 {
			yyerror("bad addressing modes")
		}
		reg = int(g1.Scale)
	} else if g2.Scale != 0 {
		if reg != 0 {
			yyerror("bad addressing modes")
		}
		reg = int(g2.Scale)
	}

	p = asm.Ctxt.NewProg()
	p.As = int16(a)
	p.Lineno = stmtline
	if nosched != 0 {
		p.Mark |= ppc64.NOSCHED
	}
	p.From = *g1
	p.Reg = int16(reg)
	p.To = *g2
	p.Pc = int64(asm.PC)

	if lastpc == nil {
		pl = obj.Linknewplist(asm.Ctxt)
		pl.Firstpc = p
	} else {
		lastpc.Link = p
	}
	lastpc = p

out:
	if a != obj.AGLOBL && a != obj.ADATA {
		asm.PC++
	}
}

/*
 * The idea is to remove redundant constants.
 *	$c1->v1
 *	($c1->v2 s/$c1/v1)*
 *	set v1  return
 * The v1->v2 should be eliminated by copy propagation.
 */
func constprop(c1 *obj.Addr, v1 *obj.Addr, r *gc.Flow) {
	if gc.Debug['P'] != 0 {
		fmt.Printf("constprop %v->%v\n", gc.Ctxt.Dconv(c1), gc.Ctxt.Dconv(v1))
	}
	var p *obj.Prog
	for ; r != nil; r = r.S1 {
		p = r.Prog
		if gc.Debug['P'] != 0 {
			fmt.Printf("%v", p)
		}
		if gc.Uniqp(r) == nil {
			if gc.Debug['P'] != 0 {
				fmt.Printf("; merge; return\n")
			}
			return
		}

		if p.As == arm.AMOVW && copyas(&p.From, c1) {
			if gc.Debug['P'] != 0 {
				fmt.Printf("; sub%v/%v", gc.Ctxt.Dconv(&p.From), gc.Ctxt.Dconv(v1))
			}
			p.From = *v1
		} else if copyu(p, v1, nil) > 1 {
			if gc.Debug['P'] != 0 {
				fmt.Printf("; %vset; return\n", gc.Ctxt.Dconv(v1))
			}
			return
		}

		if gc.Debug['P'] != 0 {
			fmt.Printf("\n")
		}
		if r.S2 != nil {
			constprop(c1, v1, r.S2)
		}
	}
}

// If s==nil, copyu returns the set/use of v in p; otherwise, it
// modifies p to replace reads of v with reads of s and returns 0 for
// success or non-zero for failure.
//
// If s==nil, copy returns one of the following values:
// 	_Read           if v only used
//	_ReadWriteSame  if v is set and used in one address (read-alter-rewrite;
// 	                can't substitute)
//	_Write          if v is only set
//	_ReadWriteDiff  if v is set in one address and used in another (so addresses
// 	                can be rewritten independently)
//	_None           otherwise (not touched)
func copyu(p *obj.Prog, v *obj.Addr, s *obj.Addr) usage {
	if p.From3Type() != obj.TYPE_NONE && p.From3Type() != obj.TYPE_CONST {
		// Currently we never generate a From3 with anything other than a constant in it.
		fmt.Printf("copyu: From3 (%v) not implemented\n", gc.Ctxt.Dconv(p.From3))
	}

	switch p.As {
	default:
		fmt.Printf("copyu: can't find %v\n", p.As)
		return _ReadWriteSame

	case // read p.From, write p.To
		s390x.AMOVH,
		s390x.AMOVHZ,
		s390x.AMOVB,
		s390x.AMOVBZ,
		s390x.AMOVW,
		s390x.AMOVWZ,
		s390x.AMOVD,
		s390x.ANEG,
		s390x.AADDME,
		s390x.AADDZE,
		s390x.ASUBME,
		s390x.ASUBZE,
		s390x.AFMOVS,
		s390x.AFMOVD,
		s390x.ALEDBR,
		s390x.AFNEG,
		s390x.ALDEBR,
		s390x.ACLFEBR,
		s390x.ACLGEBR,
		s390x.ACLFDBR,
		s390x.ACLGDBR,
		s390x.ACFEBRA,
		s390x.ACGEBRA,
		s390x.ACFDBRA,
		s390x.ACGDBRA,
		s390x.ACELFBR,
		s390x.ACELGBR,
		s390x.ACDLFBR,
		s390x.ACDLGBR,
		s390x.ACEFBRA,
		s390x.ACEGBRA,
		s390x.ACDFBRA,
		s390x.ACDGBRA,
		s390x.AFSQRT:

		if s != nil {
			copysub(&p.From, v, s)

			// Update only indirect uses of v in p.To
			if !copyas(&p.To, v) {
				copysub(&p.To, v, s)
			}
			return _None
		}

		if copyas(&p.To, v) {
			// Fix up implicit from
			if p.From.Type == obj.TYPE_NONE {
				p.From = p.To
			}
			if copyau(&p.From, v) {
				return _ReadWriteDiff
			}
			return _Write
		}

		if copyau(&p.From, v) {
			return _Read
		}
		if copyau(&p.To, v) {
			// p.To only indirectly uses v
			return _Read
		}

		return _None

	// read p.From, read p.Reg, write p.To
	case s390x.AADD,
		s390x.AADDC,
		s390x.AADDE,
		s390x.ASUB,
		s390x.ASLW,
		s390x.ASRW,
		s390x.ASRAW,
		s390x.ASLD,
		s390x.ASRD,
//.........这里部分代码省略.........

func peep(firstp *obj.Prog) {
	g := (*gc.Graph)(gc.Flowstart(firstp, nil))
	if g == nil {
		return
	}
	gactive = 0

	// byte, word arithmetic elimination.
	elimshortmov(g)

	// constant propagation
	// find MOV $con,R followed by
	// another MOV $con,R without
	// setting R in the interim
	var p *obj.Prog
	for r := (*gc.Flow)(g.Start); r != nil; r = r.Link {
		p = r.Prog
		switch p.As {
		case x86.ALEAL,
			x86.ALEAQ:
			if regtyp(&p.To) {
				if p.From.Sym != nil {
					if p.From.Index == x86.REG_NONE {
						conprop(r)
					}
				}
			}

		case x86.AMOVB,
			x86.AMOVW,
			x86.AMOVL,
			x86.AMOVQ,
			x86.AMOVSS,
			x86.AMOVSD:
			if regtyp(&p.To) {
				if p.From.Type == obj.TYPE_CONST || p.From.Type == obj.TYPE_FCONST {
					conprop(r)
				}
			}
		}
	}

	var r *gc.Flow
	var r1 *gc.Flow
	var p1 *obj.Prog
	var t int
loop1:
	if gc.Debug['P'] != 0 && gc.Debug['v'] != 0 {
		gc.Dumpit("loop1", g.Start, 0)
	}

	t = 0
	for r = g.Start; r != nil; r = r.Link {
		p = r.Prog
		switch p.As {
		case x86.AMOVL,
			x86.AMOVQ,
			x86.AMOVSS,
			x86.AMOVSD:
			if regtyp(&p.To) {
				if regtyp(&p.From) {
					if copyprop(g, r) {
						excise(r)
						t++
					} else if subprop(r) && copyprop(g, r) {
						excise(r)
						t++
					}
				}
			}

		case x86.AMOVBLZX,
			x86.AMOVWLZX,
			x86.AMOVBLSX,
			x86.AMOVWLSX:
			if regtyp(&p.To) {
				r1 = rnops(gc.Uniqs(r))
				if r1 != nil {
					p1 = r1.Prog
					if p.As == p1.As && p.To.Type == p1.From.Type && p.To.Reg == p1.From.Reg {
						p1.As = x86.AMOVL
						t++
					}
				}
			}

		case x86.AMOVBQSX,
			x86.AMOVBQZX,
			x86.AMOVWQSX,
			x86.AMOVWQZX,
			x86.AMOVLQSX,
			x86.AMOVLQZX,
			x86.AMOVQL:
			if regtyp(&p.To) {
				r1 = rnops(gc.Uniqs(r))
				if r1 != nil {
					p1 = r1.Prog
					if p.As == p1.As && p.To.Type == p1.From.Type && p.To.Reg == p1.From.Reg {
						p1.As = x86.AMOVQ
						t++
//.........这里部分代码省略.........

func preprocess(ctxt *obj.Link, cursym *obj.LSym) {
	// TODO(minux): add morestack short-cuts with small fixed frame-size.
	ctxt.Cursym = cursym

	if cursym.Text == nil || cursym.Text.Link == nil {
		return
	}

	p := cursym.Text
	textstksiz := p.To.Offset

	cursym.Args = p.To.Val.(int32)
	cursym.Locals = int32(textstksiz)

	/*
	 * find leaf subroutines
	 * strip NOPs
	 * expand RET
	 * expand BECOME pseudo
	 */
	if ctxt.Debugvlog != 0 {
		fmt.Fprintf(ctxt.Bso, "%5.2f noops\n", obj.Cputime())
	}
	ctxt.Bso.Flush()

	var q *obj.Prog
	var q1 *obj.Prog
	for p := cursym.Text; p != nil; p = p.Link {
		switch p.As {
		/* too hard, just leave alone */
		case obj.ATEXT:
			q = p

			p.Mark |= LABEL | LEAF | SYNC
			if p.Link != nil {
				p.Link.Mark |= LABEL
			}

		case ANOR:
			q = p
			if p.To.Type == obj.TYPE_REG {
				if p.To.Reg == REGZERO {
					p.Mark |= LABEL | SYNC
				}
			}

		case ASYNC,
			AWORD:
			q = p
			p.Mark |= LABEL | SYNC
			continue

		case AMOVW, AMOVWZ, AMOVD:
			q = p
			if p.From.Reg >= REG_RESERVED || p.To.Reg >= REG_RESERVED {
				p.Mark |= LABEL | SYNC
			}
			continue

		case AFABS,
			AFADD,
			AFDIV,
			AFMADD,
			AFMOVD,
			/* case AFMOVDS: */
			AFMOVS,

			/* case AFMOVSD: */
			AFMSUB,
			AFMUL,
			AFNABS,
			AFNEG,
			AFNMADD,
			AFNMSUB,
			AFRSP,
			AFSUB:
			q = p

			p.Mark |= FLOAT
			continue

		case ABL,
			ABCL,
			obj.ADUFFZERO,
			obj.ADUFFCOPY:
			cursym.Text.Mark &^= LEAF
			fallthrough

		case ABC,
			ABEQ,
			ABGE,
			ABGT,
			ABLE,
			ABLT,
			ABNE,
			ABR,
			ABVC,
			ABVS,
			ACMPBEQ,
			ACMPBGE,
//.........这里部分代码省略.........

// If s==nil, copyu returns the set/use of v in p; otherwise, it
// modifies p to replace reads of v with reads of s and returns 0 for
// success or non-zero for failure.
//
// If s==nil, copy returns one of the following values:
// 	1 if v only used
//	2 if v is set and used in one address (read-alter-rewrite;
// 	  can't substitute)
//	3 if v is only set
//	4 if v is set in one address and used in another (so addresses
// 	  can be rewritten independently)
//	0 otherwise (not touched)
func copyu(p *obj.Prog, v *obj.Addr, s *obj.Addr) int {
	if p.From3Type() != obj.TYPE_NONE {
		// never generates a from3
		fmt.Printf("copyu: from3 (%v) not implemented\n", gc.Ctxt.Dconv(p.From3))
	}

	switch p.As {
	default:
		fmt.Printf("copyu: can't find %v\n", obj.Aconv(int(p.As)))
		return 2

	case obj.ANOP, /* read p->from, write p->to */
		mips.AMOVV,
		mips.AMOVF,
		mips.AMOVD,
		mips.AMOVH,
		mips.AMOVHU,
		mips.AMOVB,
		mips.AMOVBU,
		mips.AMOVW,
		mips.AMOVWU,
		mips.AMOVFD,
		mips.AMOVDF,
		mips.AMOVDW,
		mips.AMOVWD,
		mips.AMOVFW,
		mips.AMOVWF,
		mips.AMOVDV,
		mips.AMOVVD,
		mips.AMOVFV,
		mips.AMOVVF,
		mips.ATRUNCFV,
		mips.ATRUNCDV,
		mips.ATRUNCFW,
		mips.ATRUNCDW:
		if s != nil {
			if copysub(&p.From, v, s, 1) != 0 {
				return 1
			}

			// Update only indirect uses of v in p->to
			if !copyas(&p.To, v) {
				if copysub(&p.To, v, s, 1) != 0 {
					return 1
				}
			}
			return 0
		}

		if copyas(&p.To, v) {
			// Fix up implicit from
			if p.From.Type == obj.TYPE_NONE {
				p.From = p.To
			}
			if copyau(&p.From, v) {
				return 4
			}
			return 3
		}

		if copyau(&p.From, v) {
			return 1
		}
		if copyau(&p.To, v) {
			// p->to only indirectly uses v
			return 1
		}

		return 0

	case mips.ASGT, /* read p->from, read p->reg, write p->to */
		mips.ASGTU,

		mips.AADD,
		mips.AADDU,
		mips.ASUB,
		mips.ASUBU,
		mips.ASLL,
		mips.ASRL,
		mips.ASRA,
		mips.AOR,
		mips.ANOR,
		mips.AAND,
		mips.AXOR,

		mips.AADDV,
		mips.AADDVU,
		mips.ASUBV,
//.........这里部分代码省略.........

func progedit(ctxt *obj.Link, p *obj.Prog) {
	// Maintain information about code generation mode.
	if ctxt.Mode == 0 {
		ctxt.Mode = ctxt.Arch.Regsize * 8
	}
	p.Mode = int8(ctxt.Mode)

	switch p.As {
	case AMODE:
		if p.From.Type == obj.TYPE_CONST || (p.From.Type == obj.TYPE_MEM && p.From.Reg == REG_NONE) {
			switch int(p.From.Offset) {
			case 16, 32, 64:
				ctxt.Mode = int(p.From.Offset)
			}
		}
		obj.Nopout(p)
	}

	// Thread-local storage references use the TLS pseudo-register.
	// As a register, TLS refers to the thread-local storage base, and it
	// can only be loaded into another register:
	//
	//         MOVQ TLS, AX
	//
	// An offset from the thread-local storage base is written off(reg)(TLS*1).
	// Semantically it is off(reg), but the (TLS*1) annotation marks this as
	// indexing from the loaded TLS base. This emits a relocation so that
	// if the linker needs to adjust the offset, it can. For example:
	//
	//         MOVQ TLS, AX
	//         MOVQ 0(AX)(TLS*1), CX // load g into CX
	//
	// On systems that support direct access to the TLS memory, this
	// pair of instructions can be reduced to a direct TLS memory reference:
	//
	//         MOVQ 0(TLS), CX // load g into CX
	//
	// The 2-instruction and 1-instruction forms correspond to the two code
	// sequences for loading a TLS variable in the local exec model given in "ELF
	// Handling For Thread-Local Storage".
	//
	// We apply this rewrite on systems that support the 1-instruction form.
	// The decision is made using only the operating system and the -shared flag,
	// not the link mode. If some link modes on a particular operating system
	// require the 2-instruction form, then all builds for that operating system
	// will use the 2-instruction form, so that the link mode decision can be
	// delayed to link time.
	//
	// In this way, all supported systems use identical instructions to
	// access TLS, and they are rewritten appropriately first here in
	// liblink and then finally using relocations in the linker.
	//
	// When -shared is passed, we leave the code in the 2-instruction form but
	// assemble (and relocate) them in different ways to generate the initial
	// exec code sequence. It's a bit of a fluke that this is possible without
	// rewriting the instructions more comprehensively, and it only does because
	// we only support a single TLS variable (g).

	if CanUse1InsnTLS(ctxt) {
		// Reduce 2-instruction sequence to 1-instruction sequence.
		// Sequences like
		//	MOVQ TLS, BX
		//	... off(BX)(TLS*1) ...
		// become
		//	NOP
		//	... off(TLS) ...
		//
		// TODO(rsc): Remove the Hsolaris special case. It exists only to
		// guarantee we are producing byte-identical binaries as before this code.
		// But it should be unnecessary.
		if (p.As == AMOVQ || p.As == AMOVL) && p.From.Type == obj.TYPE_REG && p.From.Reg == REG_TLS && p.To.Type == obj.TYPE_REG && REG_AX <= p.To.Reg && p.To.Reg <= REG_R15 && ctxt.Headtype != obj.Hsolaris {
			obj.Nopout(p)
		}
		if p.From.Type == obj.TYPE_MEM && p.From.Index == REG_TLS && REG_AX <= p.From.Reg && p.From.Reg <= REG_R15 {
			p.From.Reg = REG_TLS
			p.From.Scale = 0
			p.From.Index = REG_NONE
		}

		if p.To.Type == obj.TYPE_MEM && p.To.Index == REG_TLS && REG_AX <= p.To.Reg && p.To.Reg <= REG_R15 {
			p.To.Reg = REG_TLS
			p.To.Scale = 0
			p.To.Index = REG_NONE
		}
	} else {
		// load_g_cx, below, always inserts the 1-instruction sequence. Rewrite it
		// as the 2-instruction sequence if necessary.
		//	MOVQ 0(TLS), BX
		// becomes
		//	MOVQ TLS, BX
		//	MOVQ 0(BX)(TLS*1), BX
		if (p.As == AMOVQ || p.As == AMOVL) && p.From.Type == obj.TYPE_MEM && p.From.Reg == REG_TLS && p.To.Type == obj.TYPE_REG && REG_AX <= p.To.Reg && p.To.Reg <= REG_R15 {
			q := obj.Appendp(ctxt, p)
			q.As = p.As
			q.From = p.From
			q.From.Type = obj.TYPE_MEM
			q.From.Reg = p.To.Reg
			q.From.Index = REG_TLS
			q.From.Scale = 2 // TODO: use 1
			q.To = p.To
//.........这里部分代码省略.........

func peep(firstp *obj.Prog) {
	g := gc.Flowstart(firstp, nil)
	if g == nil {
		return
	}
	gactive = 0

	var p *obj.Prog
	var r *gc.Flow
	var t int
loop1:
	if gc.Debug['P'] != 0 && gc.Debug['v'] != 0 {
		gc.Dumpit("loop1", g.Start, 0)
	}

	t = 0
	for r = g.Start; r != nil; r = r.Link {
		p = r.Prog

		// TODO(minux) Handle smaller moves. arm and amd64
		// distinguish between moves that *must* sign/zero
		// extend and moves that don't care so they
		// can eliminate moves that don't care without
		// breaking moves that do care. This might let us
		// simplify or remove the next peep loop, too.
		if p.As == arm64.AMOVD || p.As == arm64.AFMOVD {
			if regtyp(&p.To) {
				// Try to eliminate reg->reg moves
				if regtyp(&p.From) {
					if p.From.Type == p.To.Type {
						if copyprop(r) {
							excise(r)
							t++
						} else if subprop(r) && copyprop(r) {
							excise(r)
							t++
						}
					}
				}
			}
		}
	}

	if t != 0 {
		goto loop1
	}

	/*
	 * look for MOVB x,R; MOVB R,R (for small MOVs not handled above)
	 */
	var p1 *obj.Prog
	var r1 *gc.Flow
	for r := g.Start; r != nil; r = r.Link {
		p = r.Prog
		switch p.As {
		default:
			continue

		case arm64.AMOVH,
			arm64.AMOVHU,
			arm64.AMOVB,
			arm64.AMOVBU,
			arm64.AMOVW,
			arm64.AMOVWU:
			if p.To.Type != obj.TYPE_REG {
				continue
			}
		}

		r1 = r.Link
		if r1 == nil {
			continue
		}
		p1 = r1.Prog
		if p1.As != p.As {
			continue
		}
		if p1.From.Type != obj.TYPE_REG || p1.From.Reg != p.To.Reg {
			continue
		}
		if p1.To.Type != obj.TYPE_REG || p1.To.Reg != p.To.Reg {
			continue
		}
		excise(r1)
	}

	if gc.Debug['D'] > 1 {
		goto ret /* allow following code improvement to be suppressed */
	}

	// MOVD $c, R'; ADD R', R (R' unused) -> ADD $c, R
	for r := g.Start; r != nil; r = r.Link {
		p = r.Prog
		switch p.As {
		default:
			continue

		case arm64.AMOVD:
			if p.To.Type != obj.TYPE_REG {
				continue
//.........这里部分代码省略.........

// If s==nil, copyu returns the set/use of v in p; otherwise, it
// modifies p to replace reads of v with reads of s and returns 0 for
// success or non-zero for failure.
//
// If s==nil, copy returns one of the following values:
//	1 if v only used
//	2 if v is set and used in one address (read-alter-rewrite;
//	  can't substitute)
//	3 if v is only set
//	4 if v is set in one address and used in another (so addresses
//	  can be rewritten independently)
//	0 otherwise (not touched)
func copyu(p *obj.Prog, v *obj.Addr, s *obj.Addr) int {
	if p.From3.Type != obj.TYPE_NONE {
		// 7g never generates a from3
		fmt.Printf("copyu: from3 (%v) not implemented\n", gc.Ctxt.Dconv(&p.From3))
	}
	if p.To2.Type != obj.TYPE_NONE {
		// 7g never generates a to2
		fmt.Printf("copyu: to2 (%v) not implemented\n", gc.Ctxt.Dconv(&p.To2))
	}

	switch p.As {
	default:
		fmt.Printf("copyu: can't find %v\n", obj.Aconv(int(p.As)))
		return 2

	case obj.ANOP, /* read p->from, write p->to */
		arm64.ANEG,
		arm64.AFNEGD,
		arm64.AFNEGS,
		arm64.AFSQRTD,
		arm64.AFCVTZSD,
		arm64.AFCVTZSS,
		arm64.AFCVTZSDW,
		arm64.AFCVTZSSW,
		arm64.AFCVTZUD,
		arm64.AFCVTZUS,
		arm64.AFCVTZUDW,
		arm64.AFCVTZUSW,
		arm64.AFCVTSD,
		arm64.AFCVTDS,
		arm64.ASCVTFD,
		arm64.ASCVTFS,
		arm64.ASCVTFWD,
		arm64.ASCVTFWS,
		arm64.AUCVTFD,
		arm64.AUCVTFS,
		arm64.AUCVTFWD,
		arm64.AUCVTFWS,
		arm64.AMOVB,
		arm64.AMOVBU,
		arm64.AMOVH,
		arm64.AMOVHU,
		arm64.AMOVW,
		arm64.AMOVWU,
		arm64.AMOVD,
		arm64.AFMOVS,
		arm64.AFMOVD:
		if p.Scond == 0 {
			if s != nil {
				if copysub(&p.From, v, s, 1) != 0 {
					return 1
				}

				// Update only indirect uses of v in p->to
				if !copyas(&p.To, v) {
					if copysub(&p.To, v, s, 1) != 0 {
						return 1
					}
				}
				return 0
			}

			if copyas(&p.To, v) {
				// Fix up implicit from
				if p.From.Type == obj.TYPE_NONE {
					p.From = p.To
				}
				if copyau(&p.From, v) {
					return 4
				}
				return 3
			}

			if copyau(&p.From, v) {
				return 1
			}
			if copyau(&p.To, v) {
				// p->to only indirectly uses v
				return 1
			}

			return 0
		}

		/* rar p->from, write p->to or read p->from, rar p->to */
		if p.From.Type == obj.TYPE_MEM {
			if copyas(&p.From, v) {
				// No s!=nil check; need to fail
//.........这里部分代码省略.........

// If s==nil, copyu returns the set/use of v in p; otherwise, it
// modifies p to replace reads of v with reads of s and returns 0 for
// success or non-zero for failure.
//
// If s==nil, copy returns one of the following values:
// 	1 if v only used
//	2 if v is set and used in one address (read-alter-rewrite;
// 	  can't substitute)
//	3 if v is only set
//	4 if v is set in one address and used in another (so addresses
// 	  can be rewritten independently)
//	0 otherwise (not touched)
func copyu(p *obj.Prog, v *obj.Addr, s *obj.Addr) int {
	if p.From3Type() != obj.TYPE_NONE && p.From3Type() != obj.TYPE_CONST {
		// Currently we never generate a From3 with anything other than a constant in it.
		fmt.Printf("copyu: From3 (%v) not implemented\n", gc.Ctxt.Dconv(p.From3))
	}

	switch p.As {
	default:
		fmt.Printf("copyu: can't find %v\n", obj.Aconv(int(p.As)))
		return 2

	case obj.ANOP, /* read p->from, write p->to */
		s390x.AMOVH,
		s390x.AMOVHZ,
		s390x.AMOVB,
		s390x.AMOVBZ,
		s390x.AMOVW,
		s390x.AMOVWZ,
		s390x.AMOVD,
		s390x.ANEG,
		s390x.AADDME,
		s390x.AADDZE,
		s390x.ASUBME,
		s390x.ASUBZE,
		s390x.AFMOVS,
		s390x.AFMOVD,
		s390x.AFRSP,
		s390x.AFNEG,
		s390x.ALDEBR,
		s390x.ACLFEBR,
		s390x.ACLGEBR,
		s390x.ACLFDBR,
		s390x.ACLGDBR,
		s390x.ACFEBRA,
		s390x.ACGEBRA,
		s390x.ACFDBRA,
		s390x.ACGDBRA,
		s390x.ACELFBR,
		s390x.ACELGBR,
		s390x.ACDLFBR,
		s390x.ACDLGBR,
		s390x.ACEFBRA,
		s390x.ACEGBRA,
		s390x.ACDFBRA,
		s390x.ACDGBRA,
		s390x.AFSQRT:

		if s != nil {
			if copysub(&p.From, v, s, 1) != 0 {
				return 1
			}

			// Update only indirect uses of v in p->to
			if !copyas(&p.To, v) {
				if copysub(&p.To, v, s, 1) != 0 {
					return 1
				}
			}
			return 0
		}

		if copyas(&p.To, v) {
			// Fix up implicit from
			if p.From.Type == obj.TYPE_NONE {
				p.From = p.To
			}
			if copyau(&p.From, v) {
				return 4
			}
			return 3
		}

		if copyau(&p.From, v) {
			return 1
		}
		if copyau(&p.To, v) {
			// p->to only indirectly uses v
			return 1
		}

		return 0

	// read p->from, read p->reg, rar p->to
	case s390x.ARLWMI:
		if copyas(&p.To, v) {
			return 2
		}
		fallthrough
//.........这里部分代码省略.........

func preprocess(ctxt *obj.Link, cursym *obj.LSym) {
	// TODO(minux): add morestack short-cuts with small fixed frame-size.
	ctxt.Cursym = cursym

	if cursym.Text == nil || cursym.Text.Link == nil {
		return
	}

	p := cursym.Text
	textstksiz := p.To.Offset
	if textstksiz == -8 {
		// Compatibility hack.
		p.From3.Offset |= obj.NOFRAME
		textstksiz = 0
	}
	if textstksiz%8 != 0 {
		ctxt.Diag("frame size %d not a multiple of 8", textstksiz)
	}
	if p.From3.Offset&obj.NOFRAME != 0 {
		if textstksiz != 0 {
			ctxt.Diag("NOFRAME functions must have a frame size of 0, not %d", textstksiz)
		}
	}

	cursym.Args = p.To.Val.(int32)
	cursym.Locals = int32(textstksiz)

	/*
	 * find leaf subroutines
	 * strip NOPs
	 * expand RET
	 * expand BECOME pseudo
	 */
	if ctxt.Debugvlog != 0 {
		ctxt.Logf("%5.2f noops\n", obj.Cputime())
	}

	var q *obj.Prog
	var q1 *obj.Prog
	for p := cursym.Text; p != nil; p = p.Link {
		switch p.As {
		/* too hard, just leave alone */
		case obj.ATEXT:
			q = p

			p.Mark |= LABEL | LEAF | SYNC
			if p.Link != nil {
				p.Link.Mark |= LABEL
			}

		case ASYNC,
			AWORD:
			q = p
			p.Mark |= LABEL | SYNC
			continue

		case AMOVW, AMOVWZ, AMOVD:
			q = p
			if p.From.Reg >= REG_RESERVED || p.To.Reg >= REG_RESERVED {
				p.Mark |= LABEL | SYNC
			}
			continue

		case AFABS,
			AFADD,
			AFDIV,
			AFMADD,
			AFMOVD,
			AFMOVS,
			AFMSUB,
			AFMUL,
			AFNABS,
			AFNEG,
			AFNMADD,
			AFNMSUB,
			ALEDBR,
			ALDEBR,
			AFSUB:
			q = p

			p.Mark |= FLOAT
			continue

		case ABL,
			ABCL,
			obj.ADUFFZERO,
			obj.ADUFFCOPY:
			cursym.Text.Mark &^= LEAF
			fallthrough

		case ABC,
			ABEQ,
			ABGE,
			ABGT,
			ABLE,
			ABLT,
			ABLEU,
			ABLTU,
			ABNE,
			ABR,
//.........这里部分代码省略.........

func progedit(ctxt *obj.Link, p *obj.Prog) {
	// Maintain information about code generation mode.
	if ctxt.Mode == 0 {
		ctxt.Mode = ctxt.Arch.Regsize * 8
	}
	p.Mode = int8(ctxt.Mode)

	switch p.As {
	case AMODE:
		if p.From.Type == obj.TYPE_CONST || (p.From.Type == obj.TYPE_MEM && p.From.Reg == REG_NONE) {
			switch int(p.From.Offset) {
			case 16, 32, 64:
				ctxt.Mode = int(p.From.Offset)
			}
		}
		obj.Nopout(p)
	}

	// Thread-local storage references use the TLS pseudo-register.
	// As a register, TLS refers to the thread-local storage base, and it
	// can only be loaded into another register:
	//
	//         MOVQ TLS, AX
	//
	// An offset from the thread-local storage base is written off(reg)(TLS*1).
	// Semantically it is off(reg), but the (TLS*1) annotation marks this as
	// indexing from the loaded TLS base. This emits a relocation so that
	// if the linker needs to adjust the offset, it can. For example:
	//
	//         MOVQ TLS, AX
	//         MOVQ 8(AX)(TLS*1), CX // load m into CX
	//
	// On systems that support direct access to the TLS memory, this
	// pair of instructions can be reduced to a direct TLS memory reference:
	//
	//         MOVQ 8(TLS), CX // load m into CX
	//
	// The 2-instruction and 1-instruction forms correspond roughly to
	// ELF TLS initial exec mode and ELF TLS local exec mode, respectively.
	//
	// We applies this rewrite on systems that support the 1-instruction form.
	// The decision is made using only the operating system (and probably
	// the -shared flag, eventually), not the link mode. If some link modes
	// on a particular operating system require the 2-instruction form,
	// then all builds for that operating system will use the 2-instruction
	// form, so that the link mode decision can be delayed to link time.
	//
	// In this way, all supported systems use identical instructions to
	// access TLS, and they are rewritten appropriately first here in
	// liblink and then finally using relocations in the linker.

	if canuselocaltls(ctxt) {
		// Reduce TLS initial exec model to TLS local exec model.
		// Sequences like
		//	MOVQ TLS, BX
		//	... off(BX)(TLS*1) ...
		// become
		//	NOP
		//	... off(TLS) ...
		//
		// TODO(rsc): Remove the Hsolaris special case. It exists only to
		// guarantee we are producing byte-identical binaries as before this code.
		// But it should be unnecessary.
		if (p.As == AMOVQ || p.As == AMOVL) && p.From.Type == obj.TYPE_REG && p.From.Reg == REG_TLS && p.To.Type == obj.TYPE_REG && REG_AX <= p.To.Reg && p.To.Reg <= REG_R15 && ctxt.Headtype != obj.Hsolaris {
			obj.Nopout(p)
		}
		if p.From.Type == obj.TYPE_MEM && p.From.Index == REG_TLS && REG_AX <= p.From.Reg && p.From.Reg <= REG_R15 {
			p.From.Reg = REG_TLS
			p.From.Scale = 0
			p.From.Index = REG_NONE
		}

		if p.To.Type == obj.TYPE_MEM && p.To.Index == REG_TLS && REG_AX <= p.To.Reg && p.To.Reg <= REG_R15 {
			p.To.Reg = REG_TLS
			p.To.Scale = 0
			p.To.Index = REG_NONE
		}
	} else {
		// As a courtesy to the C compilers, rewrite TLS local exec load as TLS initial exec load.
		// The instruction
		//	MOVQ off(TLS), BX
		// becomes the sequence
		//	MOVQ TLS, BX
		//	MOVQ off(BX)(TLS*1), BX
		// This allows the C compilers to emit references to m and g using the direct off(TLS) form.
		if (p.As == AMOVQ || p.As == AMOVL) && p.From.Type == obj.TYPE_MEM && p.From.Reg == REG_TLS && p.To.Type == obj.TYPE_REG && REG_AX <= p.To.Reg && p.To.Reg <= REG_R15 {
			q := obj.Appendp(ctxt, p)
			q.As = p.As
			q.From = p.From
			q.From.Type = obj.TYPE_MEM
			q.From.Reg = p.To.Reg
			q.From.Index = REG_TLS
			q.From.Scale = 2 // TODO: use 1
			q.To = p.To
			p.From.Type = obj.TYPE_REG
			p.From.Reg = REG_TLS
			p.From.Index = REG_NONE
			p.From.Offset = 0
		}
	}
//.........这里部分代码省略.........

func mergetemp(firstp *obj.Prog) {
	const (
		debugmerge = 1
	)

	g := Flowstart(firstp, nil)
	if g == nil {
		return
	}

	// Build list of all mergeable variables.
	nvar := 0
	for l := Curfn.Dcl; l != nil; l = l.Next {
		if canmerge(l.N) {
			nvar++
		}
	}

	var_ := make([]TempVar, nvar)
	nvar = 0
	var n *Node
	var v *TempVar
	for l := Curfn.Dcl; l != nil; l = l.Next {
		n = l.N
		if canmerge(n) {
			v = &var_[nvar]
			nvar++
			n.Opt = v
			v.node = n
		}
	}

	// Build list of uses.
	// We assume that the earliest reference to a temporary is its definition.
	// This is not true of variables in general but our temporaries are all
	// single-use (that's why we have so many!).
	var p *obj.Prog
	var info ProgInfo
	for f := g.Start; f != nil; f = f.Link {
		p = f.Prog
		info = Thearch.Proginfo(p)

		if p.From.Node != nil && ((p.From.Node).(*Node)).Opt != nil && p.To.Node != nil && ((p.To.Node).(*Node)).Opt != nil {
			Fatal("double node %v", p)
		}
		v = nil
		n, _ = p.From.Node.(*Node)
		if n != nil {
			v, _ = n.Opt.(*TempVar)
		}
		if v == nil {
			n, _ = p.To.Node.(*Node)
			if n != nil {
				v, _ = n.Opt.(*TempVar)
			}
		}
		if v != nil {
			if v.def == nil {
				v.def = f
			}
			f.Data = v.use
			v.use = f
			if n == p.From.Node && (info.Flags&LeftAddr != 0) {
				v.addr = 1
			}
		}
	}

	if debugmerge > 1 && Debug['v'] != 0 {
		Dumpit("before", g.Start, 0)
	}

	nkill := 0

	// Special case.
	var p1 *obj.Prog
	var info1 ProgInfo
	var f *Flow
	for i := 0; i < len(var_); i++ {
		v = &var_[i]
		if v.addr != 0 {
			continue
		}

		// Used in only one instruction, which had better be a write.
		f = v.use
		if f != nil && f.Data.(*Flow) == nil {
			p = f.Prog
			info = Thearch.Proginfo(p)
			if p.To.Node == v.node && (info.Flags&RightWrite != 0) && info.Flags&RightRead == 0 {
				p.As = obj.ANOP
				p.To = obj.Addr{}
				v.removed = 1
				if debugmerge > 0 && Debug['v'] != 0 {
					fmt.Printf("drop write-only %v\n", Sconv(v.node.Sym, 0))
				}
			} else {
				Fatal("temp used and not set: %v", p)
			}
			nkill++
//.........这里部分代码省略.........