cmd/gc: enable inlining in generated method wrappers.

Method calls on interfaces with large stored values
will call the pointer receiver method which may be
a wrapper over a method with value receiver.

This is particularly inefficient for very small bodies.
Inlining the wrapped method body saves a potentially expensive
function call.

benchmark                old ns/op    new ns/op    delta
BenchmarkSortString1K       802295       641387  -20.06%
BenchmarkSortInt1K          359914       238234  -33.81%
BenchmarkSortInt64K       35764226     22803078  -36.24%

Fixes #4707.

R=golang-dev, daniel.morsing, rsc
CC=golang-dev
https://golang.org/cl/7214044

src/cmd/gc/inl.c

@@ -565,21 +565,28 @@ mkinlcall1(Node **np, Node *fn, int isddd)
 	inlretvars = nil;
 	i = 0;
 	// Make temp names to use instead of the originals
-	for(ll = dcl; ll; ll=ll->next)
+	for(ll = dcl; ll; ll=ll->next) {
+		if(ll->n->class == PPARAMOUT)  // return values handled below.
+			continue;
 		if(ll->n->op == ONAME) {
 			ll->n->inlvar = inlvar(ll->n);
 			// Typecheck because inlvar is not necessarily a function parameter.
 			typecheck(&ll->n->inlvar, Erv);
 			if ((ll->n->class&~PHEAP) != PAUTO)
 				ninit = list(ninit, nod(ODCL, ll->n->inlvar, N));  // otherwise gen won't emit the allocations for heapallocs
-			if (ll->n->class == PPARAMOUT)  // we rely on the order being correct here
-				inlretvars = list(inlretvars, ll->n->inlvar);
+		}
 	}
 
-	// anonymous return values, synthesize names for use in assignment that replaces return
-	if(inlretvars == nil && fn->type->outtuple > 0)
+	// temporaries for return values.
 	for(t = getoutargx(fn->type)->type; t; t = t->down) {
+		if(t != T && t->nname != N && !isblank(t->nname)) {
+			m = inlvar(t->nname);
+			typecheck(&m, Erv);
+			t->nname->inlvar = m;
+		} else {
+			// anonymous return values, synthesize names for use in assignment that replaces return
 			m = retvar(t, i++);
+		}
 		ninit = list(ninit, nod(ODCL, m, N));
 		inlretvars = list(inlretvars, m);
 	}

src/cmd/gc/obj.c

@@ -16,6 +16,8 @@ static	void	dumpglobls(void);
 void
 dumpobj(void)
 {
+	NodeList *externs, *tmp;
+
 	bout = Bopen(outfile, OWRITE);
 	if(bout == nil) {
 		flusherrors();
@@ -31,8 +33,20 @@ dumpobj(void)
 
 	outhist(bout);
 
+	externs = nil;
+	if(externdcl != nil)
+		externs = externdcl->end;
+
 	dumpglobls();
 	dumptypestructs();
+
+	// Dump extra globals.
+	tmp = externdcl;
+	if(externs != nil)
+		externdcl = externs->next;
+	dumpglobls();
+	externdcl = tmp;
+
 	dumpdata();
 	dumpfuncs();
 

src/cmd/gc/subr.c

@@ -2565,6 +2565,7 @@ genwrapper(Type *rcvr, Type *method, Sym *newnam, int iface)
 		fn->dupok = 1;
 	typecheck(&fn, Etop);
 	typechecklist(fn->nbody, Etop);
+	inlcalls(fn);
 	curfn = nil;
 	funccompile(fn, 0);
 }