Golang obj.Prog类代码示例

func Prog(as int) *obj.Prog {
	var p *obj.Prog

	if as == obj.ADATA || as == obj.AGLOBL {
		if ddumped != 0 {
			Fatal("already dumped data")
		}
		if dpc == nil {
			dpc = Ctxt.NewProg()
			dfirst = dpc
		}

		p = dpc
		dpc = Ctxt.NewProg()
		p.Link = dpc
	} else {
		p = Pc
		Pc = Ctxt.NewProg()
		Clearp(Pc)
		p.Link = Pc
	}

	if lineno == 0 {
		if Debug['K'] != 0 {
			Warn("prog: line 0")
		}
	}

	p.As = int16(as)
	p.Lineno = lineno
	return p
}

// Called after regopt and peep have run.
// Expand CHECKNIL pseudo-op into actual nil pointer check.
func expandchecks(firstp *obj.Prog) {
	var p1 *obj.Prog
	var p2 *obj.Prog

	for p := firstp; p != nil; p = p.Link {
		if p.As != obj.ACHECKNIL {
			continue
		}
		if gc.Debug_checknil != 0 && p.Lineno > 1 { // p->lineno==1 in generated wrappers
			gc.Warnl(int(p.Lineno), "generated nil check")
		}

		// check is
		//	CMP arg, $0
		//	JNE 2(PC) (likely)
		//	MOV AX, 0
		p1 = gc.Ctxt.NewProg()

		p2 = gc.Ctxt.NewProg()
		gc.Clearp(p1)
		gc.Clearp(p2)
		p1.Link = p2
		p2.Link = p.Link
		p.Link = p1
		p1.Lineno = p.Lineno
		p2.Lineno = p.Lineno
		p1.Pc = 9999
		p2.Pc = 9999
		p.As = int16(cmpptr)
		p.To.Type = obj.TYPE_CONST
		p.To.Offset = 0
		p1.As = x86.AJNE
		p1.From.Type = obj.TYPE_CONST
		p1.From.Offset = 1 // likely
		p1.To.Type = obj.TYPE_BRANCH
		p1.To.Val = p2.Link

		// crash by write to memory address 0.
		// if possible, since we know arg is 0, use 0(arg),
		// which will be shorter to encode than plain 0.
		p2.As = x86.AMOVL

		p2.From.Type = obj.TYPE_REG
		p2.From.Reg = x86.REG_AX
		if regtyp(&p.From) {
			p2.To.Type = obj.TYPE_MEM
			p2.To.Reg = p.From.Reg
		} else {
			p2.To.Type = obj.TYPE_MEM
			p2.To.Reg = x86.REG_NONE
		}

		p2.To.Offset = 0
	}
}

func appendpp(p *obj.Prog, as int, ftype int, freg int, foffset int64, ttype int, treg int, toffset int64) *obj.Prog {
	q := gc.Ctxt.NewProg()
	gc.Clearp(q)
	q.As = int16(as)
	q.Lineno = p.Lineno
	q.From.Type = int16(ftype)
	q.From.Reg = int16(freg)
	q.From.Offset = foffset
	q.To.Type = int16(ttype)
	q.To.Reg = int16(treg)
	q.To.Offset = toffset
	q.Link = p.Link
	p.Link = q
	return q
}

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 preprocess(ctxt *obj.Link, cursym *obj.LSym) {
	if ctxt.Tlsg == nil {
		ctxt.Tlsg = obj.Linklookup(ctxt, "runtime.tlsg", 0)
	}
	if ctxt.Symmorestack[0] == nil {
		ctxt.Symmorestack[0] = obj.Linklookup(ctxt, "runtime.morestack", 0)
		ctxt.Symmorestack[1] = obj.Linklookup(ctxt, "runtime.morestack_noctxt", 0)
	}

	if ctxt.Headtype == obj.Hplan9 && ctxt.Plan9privates == nil {
		ctxt.Plan9privates = obj.Linklookup(ctxt, "_privates", 0)
	}

	ctxt.Cursym = cursym

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

	p := cursym.Text
	autoffset := int32(p.To.Offset)
	if autoffset < 0 {
		autoffset = 0
	}

	var bpsize int
	if p.Mode == 64 && obj.Framepointer_enabled != 0 && autoffset > 0 {
		// Make room for to save a base pointer.  If autoffset == 0,
		// this might do something special like a tail jump to
		// another function, so in that case we omit this.
		bpsize = ctxt.Arch.Ptrsize

		autoffset += int32(bpsize)
		p.To.Offset += int64(bpsize)
	} else {
		bpsize = 0
	}

	textarg := int64(p.To.Val.(int32))
	cursym.Args = int32(textarg)
	cursym.Locals = int32(p.To.Offset)

	// TODO(rsc): Remove.
	if p.Mode == 32 && cursym.Locals < 0 {
		cursym.Locals = 0
	}

	// TODO(rsc): Remove 'p.Mode == 64 &&'.
	if p.Mode == 64 && autoffset < obj.StackSmall && p.From3.Offset&obj.NOSPLIT == 0 {
		for q := p; q != nil; q = q.Link {
			if q.As == obj.ACALL {
				goto noleaf
			}
			if (q.As == obj.ADUFFCOPY || q.As == obj.ADUFFZERO) && autoffset >= obj.StackSmall-8 {
				goto noleaf
			}
		}

		p.From3.Offset |= obj.NOSPLIT
	noleaf:
	}

	if p.From3.Offset&obj.NOSPLIT == 0 || (p.From3.Offset&obj.WRAPPER != 0) {
		p = obj.Appendp(ctxt, p)
		p = load_g_cx(ctxt, p) // load g into CX
	}

	var q *obj.Prog
	if cursym.Text.From3.Offset&obj.NOSPLIT == 0 {
		p = stacksplit(ctxt, p, autoffset, int32(textarg), cursym.Text.From3.Offset&obj.NEEDCTXT == 0, &q) // emit split check
	}

	if autoffset != 0 {
		if autoffset%int32(ctxt.Arch.Regsize) != 0 {
			ctxt.Diag("unaligned stack size %d", autoffset)
		}
		p = obj.Appendp(ctxt, p)
		p.As = AADJSP
		p.From.Type = obj.TYPE_CONST
		p.From.Offset = int64(autoffset)
		p.Spadj = autoffset
	} else {
		// zero-byte stack adjustment.
		// Insert a fake non-zero adjustment so that stkcheck can
		// recognize the end of the stack-splitting prolog.
		p = obj.Appendp(ctxt, p)

		p.As = obj.ANOP
		p.Spadj = int32(-ctxt.Arch.Ptrsize)
		p = obj.Appendp(ctxt, p)
		p.As = obj.ANOP
		p.Spadj = int32(ctxt.Arch.Ptrsize)
	}

	if q != nil {
		q.Pcond = p
	}
	deltasp := autoffset

	if bpsize > 0 {
//.........这里部分代码省略.........

func fixjmp(firstp *obj.Prog) {
	if Debug['R'] != 0 && Debug['v'] != 0 {
		fmt.Printf("\nfixjmp\n")
	}

	// pass 1: resolve jump to jump, mark all code as dead.
	jmploop := 0

	for p := firstp; p != nil; p = p.Link {
		if Debug['R'] != 0 && Debug['v'] != 0 {
			fmt.Printf("%v\n", p)
		}
		if p.As != obj.ACALL && p.To.Type == obj.TYPE_BRANCH && p.To.Val.(*obj.Prog) != nil && p.To.Val.(*obj.Prog).As == obj.AJMP {
			p.To.Val = chasejmp(p.To.Val.(*obj.Prog), &jmploop)
			if Debug['R'] != 0 && Debug['v'] != 0 {
				fmt.Printf("->%v\n", p)
			}
		}

		p.Opt = dead
	}

	if Debug['R'] != 0 && Debug['v'] != 0 {
		fmt.Printf("\n")
	}

	// pass 2: mark all reachable code alive
	mark(firstp)

	// pass 3: delete dead code (mostly JMPs).
	var last *obj.Prog

	for p := firstp; p != nil; p = p.Link {
		if p.Opt == dead {
			if p.Link == nil && p.As == obj.ARET && last != nil && last.As != obj.ARET {
				// This is the final ARET, and the code so far doesn't have one.
				// Let it stay. The register allocator assumes that all live code in
				// the function can be traversed by starting at all the RET instructions
				// and following predecessor links. If we remove the final RET,
				// this assumption will not hold in the case of an infinite loop
				// at the end of a function.
				// Keep the RET but mark it dead for the liveness analysis.
				p.Mode = 1
			} else {
				if Debug['R'] != 0 && Debug['v'] != 0 {
					fmt.Printf("del %v\n", p)
				}
				continue
			}
		}

		if last != nil {
			last.Link = p
		}
		last = p
	}

	last.Link = nil

	// pass 4: elide JMP to next instruction.
	// only safe if there are no jumps to JMPs anymore.
	if jmploop == 0 {
		var last *obj.Prog
		for p := firstp; p != nil; p = p.Link {
			if p.As == obj.AJMP && p.To.Type == obj.TYPE_BRANCH && p.To.Val == p.Link {
				if Debug['R'] != 0 && Debug['v'] != 0 {
					fmt.Printf("del %v\n", p)
				}
				continue
			}

			if last != nil {
				last.Link = p
			}
			last = p
		}

		last.Link = nil
	}

	if Debug['R'] != 0 && Debug['v'] != 0 {
		fmt.Printf("\n")
		for p := firstp; p != nil; p = p.Link {
			fmt.Printf("%v\n", p)
		}
		fmt.Printf("\n")
	}
}

func Clearp(p *obj.Prog) {
	obj.Nopout(p)
	p.As = obj.AEND
	p.Pc = int64(pcloc)
	pcloc++
}

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.........

// movb elimination.
// movb is simulated by the linker
// when a register other than ax, bx, cx, dx
// is used, so rewrite to other instructions
// when possible.  a movb into a register
// can smash the entire 32-bit register without
// causing any trouble.
//
// TODO: Using the Q forms here instead of the L forms
// seems unnecessary, and it makes the instructions longer.
func elimshortmov(g *gc.Graph) {
	var p *obj.Prog

	for r := (*gc.Flow)(g.Start); r != nil; r = r.Link {
		p = r.Prog
		if regtyp(&p.To) {
			switch p.As {
			case x86.AINCB,
				x86.AINCW:
				p.As = x86.AINCQ

			case x86.ADECB,
				x86.ADECW:
				p.As = x86.ADECQ

			case x86.ANEGB,
				x86.ANEGW:
				p.As = x86.ANEGQ

			case x86.ANOTB,
				x86.ANOTW:
				p.As = x86.ANOTQ
			}

			if regtyp(&p.From) || p.From.Type == obj.TYPE_CONST {
				// move or artihmetic into partial register.
				// from another register or constant can be movl.
				// we don't switch to 64-bit arithmetic if it can
				// change how the carry bit is set (and the carry bit is needed).
				switch p.As {
				case x86.AMOVB,
					x86.AMOVW:
					p.As = x86.AMOVQ

				case x86.AADDB,
					x86.AADDW:
					if !needc(p.Link) {
						p.As = x86.AADDQ
					}

				case x86.ASUBB,
					x86.ASUBW:
					if !needc(p.Link) {
						p.As = x86.ASUBQ
					}

				case x86.AMULB,
					x86.AMULW:
					p.As = x86.AMULQ

				case x86.AIMULB,
					x86.AIMULW:
					p.As = x86.AIMULQ

				case x86.AANDB,
					x86.AANDW:
					p.As = x86.AANDQ

				case x86.AORB,
					x86.AORW:
					p.As = x86.AORQ

				case x86.AXORB,
					x86.AXORW:
					p.As = x86.AXORQ

				case x86.ASHLB,
					x86.ASHLW:
					p.As = x86.ASHLQ
				}
			} else if p.From.Type != obj.TYPE_REG {
				// explicit zero extension, but don't
				// do that if source is a byte register
				// (only AH can occur and it's forbidden).
				switch p.As {
				case x86.AMOVB:
					p.As = x86.AMOVBQZX

				case x86.AMOVW:
					p.As = x86.AMOVWQZX
				}
			}
		}
	}
}