6l: function at a time code layout

Also change the span-dependent jump algorithm
to use fewer iterations:

* resolve forward jumps at their targets (comefrom list)
* mark jumps as small or big and only do small->big
* record whether a jump failed to be encodable

These changes mean that a function with only small
jumps can be laid out in a single iteration, and the
vast majority of functions take just two iterations.
I was seeing a maximum of 5 iterations before; the
max now is 3 and there are fewer that get even that far.

R=ken2
CC=golang-dev
https://golang.org/cl/2537041

src/cmd/5l/5.out.h

@@ -246,6 +246,7 @@ enum	as
 
 /* internal only */
 #define	D_SIZE		(D_NONE+40)
+#define	D_PCREL		(D_NONE+41)
 
 /*
  * this is the ranlib header

src/cmd/6l/6.out.h

@@ -824,6 +824,7 @@ enum
 	D_INDIR,	/* additive */
 
 	D_SIZE = D_INDIR + D_INDIR,	/* 6l internal */
+	D_PCREL,
 
 	T_TYPE		= 1<<0,
 	T_INDEX		= 1<<1,

src/cmd/6l/asm.c

@@ -344,10 +344,8 @@ phsh(ElfPhdr *ph, ElfShdr *sh)
 void
 asmb(void)
 {
-	Prog *p;
 	int32 v, magic;
 	int a, dynsym;
-	uchar *op1;
 	vlong vl, va, startva, fo, w, symo, elfsymo, elfstro, elfsymsize, machlink;
 	vlong symdatva = SYMDATVA;
 	ElfEhdr *eh;
@@ -366,35 +364,8 @@ asmb(void)
 	elfsymo = 0;
 	seek(cout, HEADR, 0);
 	pc = INITTEXT;
-
-	for(cursym = textp; cursym != nil; cursym = cursym->next) {
-		for(p = cursym->text; p != P; p = p->link) {
-			if(p->pc != pc) {
-				if(!debug['a'])
-					print("%P\n", curp);
-				diag("phase error %llux sb %llux in %s", p->pc, pc, TNAME);
-				pc = p->pc;
-			}
-			curp = p;
-			asmins(p);
-			a = (andptr - and);
-			if(cbc < a)
-				cflush();
-			if(debug['a']) {
-				Bprint(&bso, pcstr, pc);
-				for(op1 = and; op1 < andptr; op1++)
-					Bprint(&bso, "%.2ux", *op1);
-				for(; op1 < and+Maxand; op1++)
-					Bprint(&bso, "  ");
-				Bprint(&bso, "%P\n", curp);
-			}
-			memmove(cbp, and, a);
-			cbp += a;
-			pc += a;
-			cbc -= a;
-		}
-	}
-	cflush();
+	codeblk(pc, segtext.sect->len);
+	pc += segtext.sect->len;
 
 	/* output read-only data in text segment */
 	sect = segtext.sect->next;

src/cmd/6l/l.h

@@ -94,8 +94,8 @@ struct	Prog
 	Adr	from;
 	Adr	to;
 	Prog*	forwd;
+	Prog*	comefrom;
 	Prog*	link;
-	Prog*	dlink;
 	Prog*	pcond;	/* work on this */
 	vlong	pc;
 	int32	spadj;

src/cmd/8l/8.out.h

@@ -499,6 +499,7 @@ enum
 
 	D_CONST2 = D_INDIR+D_INDIR,
 	D_SIZE,	/* 8l internal */
+	D_PCREL,
 
 	T_TYPE		= 1<<0,
 	T_INDEX		= 1<<1,

src/cmd/ld/data.c

@@ -156,6 +156,9 @@ relocsym(Sym *s)
 		case D_ADDR:
 			o = symaddr(r->sym);
 			break;
+		case D_PCREL:
+			o = symaddr(r->sym) - (s->value + r->off + r->siz);
+			break;
 		case D_SIZE:
 			o = r->sym->size;
 			break;
@@ -190,11 +193,8 @@ reloc(void)
 
 	for(s=textp; s!=S; s=s->next)
 		relocsym(s);
-	for(s=datap; s!=S; s=s->next) {
-		if(!s->reachable)
-			diag("unerachable? %s", s->name);
+	for(s=datap; s!=S; s=s->next)
 		relocsym(s);
-	}
 }
 
 void
@@ -341,6 +341,70 @@ blk(Sym *allsym, int32 addr, int32 size)
 	cflush();
 }
 			
+void
+codeblk(int32 addr, int32 size)
+{
+	Sym *sym;
+	int32 eaddr, i, n, epc;
+	Prog *p;
+	uchar *q;
+
+	if(debug['a'])
+		Bprint(&bso, "codeblk [%#x,%#x) at offset %#llx\n", addr, addr+size, seek(cout, 0, 1));
+
+	blk(textp, addr, size);
+
+	/* again for printing */
+	if(!debug['a'])
+		return;
+
+	for(sym = textp; sym != nil; sym = sym->next) {
+		if(!sym->reachable)
+			continue;
+		if(sym->value >= addr)
+			break;
+	}
+
+	eaddr = addr + size;
+	for(; sym != nil; sym = sym->next) {
+		if(!sym->reachable)
+			continue;
+		if(sym->value >= eaddr)
+			break;
+
+		if(addr < sym->value) {
+			Bprint(&bso, "%-20s %.8llux|", "_", addr);
+			for(; addr < sym->value; addr++)
+				Bprint(&bso, " %.2ux", 0);
+			Bprint(&bso, "\n");
+		}
+		p = sym->text;
+		Bprint(&bso, "%-20s %.8llux| %P\n", sym->name, addr, p);
+		for(p = p->link; p != P; p = p->link) {
+			if(p->link != P)
+				epc = p->link->pc;
+			else
+				epc = sym->value + sym->size;
+			Bprint(&bso, "%.6ux\t", p->pc);
+			q = sym->p + p->pc - sym->value;
+			n = epc - p->pc;
+			for(i=0; i<n; i++)
+				Bprint(&bso, "%.2ux", *q++);
+			for(; i < 10; i++)
+				Bprint(&bso, "  ");
+			Bprint(&bso, " | %P\n", p);
+			addr += n;
+		}
+	}
+
+	if(addr < eaddr) {
+		Bprint(&bso, "%-20s %.8llux|", "_", addr);
+		for(; addr < eaddr; addr++)
+			Bprint(&bso, " %.2ux", 0);
+	}
+	Bflush(&bso);
+}
+			
 void
 datblk(int32 addr, int32 size)
 {
@@ -348,6 +412,9 @@ datblk(int32 addr, int32 size)
 	int32 eaddr;
 	uchar *p, *ep;
 
+	if(debug['a'])
+		Bprint(&bso, "datblk [%#x,%#x) at offset %#llx\n", addr, addr+size, seek(cout, 0, 1));
+
 	blk(datap, addr, size);
 
 	/* again for printing */
@@ -363,10 +430,8 @@ datblk(int32 addr, int32 size)
 		if(sym->value >= eaddr)
 			break;
 		if(addr < sym->value) {
-			Bprint(&bso, "%-20s %.8ux|", "(pre-pad)", addr);
-			for(; addr < sym->value; addr++)
-				Bprint(&bso, " %.2ux", 0);
-			Bprint(&bso, "\n");
+			Bprint(&bso, "%-20s %.8ux| 00 ...\n", "(pre-pad)", addr);
+			addr = sym->value;
 		}
 		Bprint(&bso, "%-20s %.8ux|", sym->name, addr);
 		p = sym->p;
@@ -379,11 +444,9 @@ datblk(int32 addr, int32 size)
 		Bprint(&bso, "\n");
 	}
 
-	if(addr < eaddr) {
-		Bprint(&bso, "%-20s %.8ux|", "(post-pad)", addr);
-		for(; addr < eaddr; addr++)
-			Bprint(&bso, " %.2ux", 0);
-	}
+	if(addr < eaddr)
+		Bprint(&bso, "%-20s %.8ux| 00 ...\n", "(post-pad)", addr);
+	Bprint(&bso, "%-20s %.8ux|\n", "", eaddr);
 }
 
 void

src/cmd/ld/lib.h

@@ -132,6 +132,7 @@ char*	expandpkg(char*, char*);
 void	deadcode(void);
 void	ewrite(int, void*, int);
 Reloc*	addrel(Sym*);
+void	codeblk(int32, int32);
 void	datblk(int32, int32);
 Sym*	datsort(Sym*);
 void	reloc(void);