cmd/gc: capture variables by value

Language specification says that variables are captured by reference.
And that is what gc compiler does. However, in lots of cases it is
possible to capture variables by value under the hood without
affecting visible behavior of programs. For example, consider
the following typical pattern:

	func (o *Obj) requestMany(urls []string) []Result {
		wg := new(sync.WaitGroup)
		wg.Add(len(urls))
		res := make([]Result, len(urls))
		for i := range urls {
			i := i
			go func() {
				res[i] = o.requestOne(urls[i])
				wg.Done()
			}()
		}
		wg.Wait()
		return res
	}

Currently o, wg, res, and i are captured by reference causing 3+len(urls)
allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap).
But all of them can be captured by value without changing behavior.

This change implements simple strategy for capturing by value:
if a captured variable is not addrtaken and never assigned to,
then it is captured by value (it is effectively const).
This simple strategy turned out to be very effective:
~80% of all captures in std lib are turned into value captures.
The remaining 20% are mostly in defers and non-escaping closures,
that is, they do not cause allocations anyway.

benchmark                                    old allocs     new allocs     delta
BenchmarkCompressedZipGarbage                153            126            -17.65%
BenchmarkEncodeDigitsSpeed1e4                91             69             -24.18%
BenchmarkEncodeDigitsSpeed1e5                178            129            -27.53%
BenchmarkEncodeDigitsSpeed1e6                1510           1051           -30.40%
BenchmarkEncodeDigitsDefault1e4              100            75             -25.00%
BenchmarkEncodeDigitsDefault1e5              193            139            -27.98%
BenchmarkEncodeDigitsDefault1e6              1420           985            -30.63%
BenchmarkEncodeDigitsCompress1e4             100            75             -25.00%
BenchmarkEncodeDigitsCompress1e5             193            139            -27.98%
BenchmarkEncodeDigitsCompress1e6             1420           985            -30.63%
BenchmarkEncodeTwainSpeed1e4                 109            81             -25.69%
BenchmarkEncodeTwainSpeed1e5                 211            151            -28.44%
BenchmarkEncodeTwainSpeed1e6                 1588           1097           -30.92%
BenchmarkEncodeTwainDefault1e4               103            77             -25.24%
BenchmarkEncodeTwainDefault1e5               199            143            -28.14%
BenchmarkEncodeTwainDefault1e6               1324           917            -30.74%
BenchmarkEncodeTwainCompress1e4              103            77             -25.24%
BenchmarkEncodeTwainCompress1e5              190            137            -27.89%
BenchmarkEncodeTwainCompress1e6              1327           919            -30.75%
BenchmarkConcurrentDBExec                    16223          16220          -0.02%
BenchmarkConcurrentStmtQuery                 17687          16182          -8.51%
BenchmarkConcurrentStmtExec                  5191           5186           -0.10%
BenchmarkConcurrentTxQuery                   17665          17661          -0.02%
BenchmarkConcurrentTxExec                    15154          15150          -0.03%
BenchmarkConcurrentTxStmtQuery               17661          16157          -8.52%
BenchmarkConcurrentTxStmtExec                3677           3673           -0.11%
BenchmarkConcurrentRandom                    14000          13614          -2.76%
BenchmarkManyConcurrentQueries               25             22             -12.00%
BenchmarkDecodeComplex128Slice               318            252            -20.75%
BenchmarkDecodeFloat64Slice                  318            252            -20.75%
BenchmarkDecodeInt32Slice                    318            252            -20.75%
BenchmarkDecodeStringSlice                   2318           2252           -2.85%
BenchmarkDecode                              11             8              -27.27%
BenchmarkEncodeGray                          64             56             -12.50%
BenchmarkEncodeNRGBOpaque                    64             56             -12.50%
BenchmarkEncodeNRGBA                         67             58             -13.43%
BenchmarkEncodePaletted                      68             60             -11.76%
BenchmarkEncodeRGBOpaque                     64             56             -12.50%
BenchmarkGoLookupIP                          153            139            -9.15%
BenchmarkGoLookupIPNoSuchHost                508            466            -8.27%
BenchmarkGoLookupIPWithBrokenNameServer      245            226            -7.76%
BenchmarkClientServer                        62             59             -4.84%
BenchmarkClientServerParallel4               62             59             -4.84%
BenchmarkClientServerParallel64              62             59             -4.84%
BenchmarkClientServerParallelTLS4            79             76             -3.80%
BenchmarkClientServerParallelTLS64           112            109            -2.68%
BenchmarkCreateGoroutinesCapture             10             6              -40.00%
BenchmarkAfterFunc                           1006           1005           -0.10%

Fixes #6632.

Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca
Reviewed-on: https://go-review.googlesource.com/3166Reviewed-by: Russ Cox <rsc@golang.org>

src/cmd/gc/closure.c

@@ -70,7 +70,7 @@ closurebody(NodeList *body)
 	for(l=func->cvars; l; l=l->next) {
 		v = l->n;
 		v->closure->closure = v->outer;
-		v->heapaddr = nod(OADDR, oldname(v->sym), N);
+		v->outerexpr = oldname(v->sym);
 	}
 
 	return func;
@@ -81,48 +81,46 @@ static Node* makeclosure(Node *func);
 void
 typecheckclosure(Node *func, int top)
 {
-	Node *oldfn;
+	Node *oldfn, *n;
 	NodeList *l;
-	Node *v;
+	int olddd;
+
+	for(l=func->cvars; l; l=l->next) {
+		n = l->n->closure;
+		if(!n->captured) {
+			n->captured = 1;
+			if(n->decldepth == 0)
+				fatal("typecheckclosure: var %hN does not have decldepth assigned", n);
+			// Ignore assignments to the variable in straightline code
+			// preceding the first capturing by a closure.
+			if(n->decldepth == decldepth)
+				n->assigned = 0;
+		}
+	}
+
+	for(l=func->dcl; l; l=l->next)
+		if(l->n->op == ONAME && (l->n->class == PPARAM || l->n->class == PPARAMOUT))
+			l->n->decldepth = 1;
 
 	oldfn = curfn;
 	typecheck(&func->ntype, Etype);
 	func->type = func->ntype->type;
-	
+
 	// Type check the body now, but only if we're inside a function.
 	// At top level (in a variable initialization: curfn==nil) we're not
 	// ready to type check code yet; we'll check it later, because the
 	// underlying closure function we create is added to xtop.
 	if(curfn && func->type != T) {
 		curfn = func;
+		olddd = decldepth;
+		decldepth = 1;
 		typechecklist(func->nbody, Etop);
+		decldepth = olddd;
 		curfn = oldfn;
 	}
 
-	// type check the & of closed variables outside the closure,
-	// so that the outer frame also grabs them and knows they
-	// escape.
-	func->enter = nil;
-	for(l=func->cvars; l; l=l->next) {
-		v = l->n;
-		if(v->type == T) {
-			// if v->type is nil, it means v looked like it was
-			// going to be used in the closure but wasn't.
-			// this happens because when parsing a, b, c := f()
-			// the a, b, c gets parsed as references to older
-			// a, b, c before the parser figures out this is a
-			// declaration.
-			v->op = 0;
-			continue;
-		}
-		// For a closure that is called in place, but not
-		// inside a go statement, avoid moving variables to the heap.
-		if ((top & (Ecall|Eproc)) == Ecall)
-			v->heapaddr->etype = 1;
-		typecheck(&v->heapaddr, Erv);
-		func->enter = list(func->enter, v->heapaddr);
-		v->heapaddr = N;
-	}
+	// Remember closure context for capturevars.
+	func->etype = (top & (Ecall|Eproc)) == Ecall;
 
 	// Create top-level function 
 	xtop = list(xtop, makeclosure(func));
@@ -131,11 +129,8 @@ typecheckclosure(Node *func, int top)
 static Node*
 makeclosure(Node *func)
 {
-	Node *xtype, *v, *addr, *xfunc, *cv;
-	NodeList *l, *body;
+	Node *xtype, *xfunc;
 	static int closgen;
-	char *p;
-	vlong offset;
 
 	/*
 	 * wrap body in external function
@@ -156,16 +151,81 @@ makeclosure(Node *func)
 	xfunc->nname->funcdepth = func->funcdepth;
 	xfunc->funcdepth = func->funcdepth;
 	xfunc->endlineno = func->endlineno;
+
+	xfunc->nbody = func->nbody;
+	xfunc->dcl = concat(func->dcl, xfunc->dcl);
+	if(xfunc->nbody == nil)
+		fatal("empty body - won't generate any code");
+	typecheck(&xfunc, Etop);
+
+	xfunc->closure = func;
+	func->closure = xfunc;
 	
-	// declare variables holding addresses taken from closure
-	// and initialize in entry prologue.
+	func->nbody = nil;
+	func->list = nil;
+	func->rlist = nil;
+
+	return xfunc;
+}
+
+// capturevars is called in a separate phase after all typechecking is done.
+// It decides whether each variable captured by a closure should be captured
+// by value or by reference.
+// We use value capturing for values <= 128 bytes that are never reassigned
+// after declaration.
+void
+capturevars(Node *xfunc)
+{
+	Node *func, *v, *addr, *cv, *outer;
+	NodeList *l, *body;
+	char *p;
+	vlong offset;
+	int nvar, lno;
+
+	lno = lineno;
+	lineno = xfunc->lineno;
+
+	nvar = 0;
 	body = nil;
 	offset = widthptr;
-	xfunc->needctxt = func->cvars != nil;
+	func = xfunc->closure;
+	func->enter = nil;
 	for(l=func->cvars; l; l=l->next) {
 		v = l->n;
-		if(v->op == 0)
+		if(v->type == T) {
+			// if v->type is nil, it means v looked like it was
+			// going to be used in the closure but wasn't.
+			// this happens because when parsing a, b, c := f()
+			// the a, b, c gets parsed as references to older
+			// a, b, c before the parser figures out this is a
+			// declaration.
+			v->op = OXXX;
 			continue;
+		}
+		nvar++;
+
+		// type check the & of closed variables outside the closure,
+		// so that the outer frame also grabs them and knows they escape.
+		dowidth(v->type);
+		outer = v->outerexpr;
+		v->outerexpr = N;
+		if(!v->closure->addrtaken && !v->closure->assigned && v->type->width <= 128)
+			v->byval = 1;
+		else {
+			outer = nod(OADDR, outer, N);
+			// For a closure that is called in place, but not
+			// inside a go statement, avoid moving variables to the heap.
+			outer->etype = func->etype;
+		}
+		if(debug['m'] > 1)
+			warnl(v->lineno, "%S capturing by %s: %S (addr=%d assign=%d width=%d)",
+				(v->curfn && v->curfn->nname) ? v->curfn->nname->sym : S, v->byval ? "value" : "ref",
+				v->sym, v->closure->addrtaken, v->closure->assigned, (int32)v->type->width);
+		typecheck(&outer, Erv);
+		func->enter = list(func->enter, outer);
+
+		// declare variables holding addresses taken from closure
+		// and initialize in entry prologue.
 		addr = nod(ONAME, N, N);
 		p = smprint("&%s", v->sym->name);
 		addr->sym = lookup(p);
@@ -179,38 +239,35 @@ makeclosure(Node *func)
 		xfunc->dcl = list(xfunc->dcl, addr);
 		v->heapaddr = addr;
 		cv = nod(OCLOSUREVAR, N, N);
-		cv->type = ptrto(v->type);
-		cv->xoffset = offset;
-		body = list(body, nod(OAS, addr, cv));
-		offset += widthptr;
+		if(v->byval) {
+			cv->type = v->type;
+			offset = rnd(offset, v->type->align);
+			cv->xoffset = offset;
+			offset += v->type->width;
+			body = list(body, nod(OAS, addr, nod(OADDR, cv, N)));
+		} else {
+			v->closure->addrtaken = 1;
+			cv->type = ptrto(v->type);
+			offset = rnd(offset, widthptr);
+			cv->xoffset = offset;
+			offset += widthptr;
+			body = list(body, nod(OAS, addr, cv));
+		}
 	}
 	typechecklist(body, Etop);
 	walkstmtlist(body);
 	xfunc->enter = body;
+	xfunc->needctxt = nvar > 0;
+	func->etype = 0;
 
-	xfunc->nbody = func->nbody;
-	xfunc->dcl = concat(func->dcl, xfunc->dcl);
-	if(xfunc->nbody == nil)
-		fatal("empty body - won't generate any code");
-	typecheck(&xfunc, Etop);
-
-	xfunc->closure = func;
-	func->closure = xfunc;
-	
-	func->nbody = nil;
-	func->list = nil;
-	func->rlist = nil;
-
-	return xfunc;
+	lineno = lno;
 }
 
 Node*
 walkclosure(Node *func, NodeList **init)
 {
-	Node *clos, *typ;
+	Node *clos, *typ, *typ1, *v;
 	NodeList *l;
-	char buf[20];
-	int narg;
 
 	// If no closure vars, don't bother wrapping.
 	if(func->cvars == nil)
@@ -230,14 +287,16 @@ walkclosure(Node *func, NodeList **init)
 	// the struct is unnamed so that closures in multiple packages with the
 	// same struct type can share the descriptor.
 
-	narg = 0;
 	typ = nod(OTSTRUCT, N, N);
 	typ->list = list1(nod(ODCLFIELD, newname(lookup("F")), typenod(types[TUINTPTR])));
 	for(l=func->cvars; l; l=l->next) {
-		if(l->n->op == 0)
+		v = l->n;
+		if(v->op == OXXX)
 			continue;
-		snprint(buf, sizeof buf, "A%d", narg++);
-		typ->list = list(typ->list, nod(ODCLFIELD, newname(lookup(buf)), l->n->heapaddr->ntype));
+		typ1 = typenod(v->type);
+		if(!v->byval)
+			typ1 = nod(OIND, typ1, N);
+		typ->list = list(typ->list, nod(ODCLFIELD, newname(v->sym), typ1));
 	}
 
 	clos = nod(OCOMPLIT, N, nod(OIND, typ, N));

src/cmd/gc/dcl.c

@@ -452,7 +452,6 @@ oldname(Sym *s)
 			c->funcdepth = funcdepth;
 			c->outer = n->closure;
 			n->closure = c;
-			n->addrtaken = 1;
 			c->closure = n;
 			c->xoffset = 0;
 			curfn->cvars = list(curfn->cvars, c);

src/cmd/gc/esc.c

@@ -429,7 +429,7 @@ esc(EscState *e, Node *n, Node *up)
 {
 	int lno;
 	NodeList *ll, *lr;
-	Node *a;
+	Node *a, *v;
 
 	if(n == N)
 		return;
@@ -676,12 +676,16 @@ esc(EscState *e, Node *n, Node *up)
 	case OCLOSURE:
 		// Link addresses of captured variables to closure.
 		for(ll=n->cvars; ll; ll=ll->next) {
-			if(ll->n->op == OXXX)  // unnamed out argument; see dcl.c:/^funcargs
+			v = ll->n;
+			if(v->op == OXXX)  // unnamed out argument; see dcl.c:/^funcargs
 				continue;
-			a = nod(OADDR, ll->n->closure, N);
-			a->lineno = ll->n->lineno;
-			a->escloopdepth = e->loopdepth;
-			typecheck(&a, Erv);
+			a = v->closure;
+			if(!v->byval) {
+				a = nod(OADDR, a, N);
+				a->lineno = v->lineno;
+				a->escloopdepth = e->loopdepth;
+				typecheck(&a, Erv);
+			}
 			escassign(e, n, a);
 		}
 		// fallthrough

src/cmd/gc/fmt.c

@@ -278,6 +278,12 @@ Jconv(Fmt *fp)
 	if(n->embedded != 0)
 		fmtprint(fp, " embedded(%d)", n->embedded);
 
+	if(n->addrtaken != 0)
+		fmtprint(fp, " addrtaken");
+
+	if(n->assigned != 0)
+		fmtprint(fp, " assigned");
+
 	if(!c && n->used != 0)
 		fmtprint(fp, " used(%d)", n->used);
 	return 0;

src/cmd/gc/go.h

@@ -284,6 +284,9 @@ struct	Node
 	uchar	readonly;
 	uchar	implicit;
 	uchar	addrtaken;	// address taken, even if not moved to heap
+	uchar	assigned;	// is the variable ever assigned to
+	uchar	captured;	// is the variable captured by a closure
+	uchar	byval;		// is the variable captured by value or by reference
 	uchar	dupok;	// duplicate definitions ok (for func)
 	uchar	wrapper;	// is method wrapper (for func)
 	uchar	reslice;	// this is a reslice x = x[0:y] or x = append(x, ...)
@@ -317,9 +320,11 @@ struct	Node
 	Node*	pack;	// real package for import . names
 	Node*	curfn;	// function for local variables
 	Type*	paramfld; // TFIELD for this PPARAM; also for ODOT, curfn
+	int	decldepth;	// declaration loop depth, increased for every loop or label
 
 	// ONAME func param with PHEAP
 	Node*	heapaddr;	// temp holding heap address of param
+	Node*	outerexpr;	// expression copied into closure for variable
 	Node*	stackparam;	// OPARAM node referring to stack copy of param
 	Node*	alloc;	// allocation call
 
@@ -874,6 +879,7 @@ EXTERN	Biobuf*	bout;
 EXTERN	int	nerrors;
 EXTERN	int	nsavederrors;
 EXTERN	int	nsyntaxerrors;
+EXTERN	int	decldepth;
 EXTERN	int	safemode;
 EXTERN	int	nolocalimports;
 EXTERN	char	namebuf[NSYMB];
@@ -943,7 +949,6 @@ EXTERN	Mpflt*	maxfltval[NTYPE];
 
 EXTERN	NodeList*	xtop;
 EXTERN	NodeList*	externdcl;
-EXTERN	NodeList*	closures;
 EXTERN	NodeList*	exportlist;
 EXTERN	NodeList*	importlist;	// imported functions and methods with inlinable bodies
 EXTERN	NodeList*	funcsyms;
@@ -1068,6 +1073,7 @@ void	bvset(Bvec *bv, int32 i);
 Node*	closurebody(NodeList *body);
 void	closurehdr(Node *ntype);
 void	typecheckclosure(Node *func, int top);
+void	capturevars(Node *func);
 Node*	walkclosure(Node *func, NodeList **init);
 void	typecheckpartialcall(Node*, Node*);
 Node*	walkpartialcall(Node*, NodeList**);
@@ -1454,7 +1460,7 @@ void	typechecklist(NodeList *l, int top);
 Node*	typecheckdef(Node *n);
 void	copytype(Node *n, Type *t);
 void	checkreturn(Node*);
-void	checkassign(Node*);
+void	checkassign(Node *stmt, Node*);
 void	queuemethod(Node *n);
 
 /*

src/cmd/gc/lex.c

@@ -391,6 +391,7 @@ gcmain(int argc, char *argv[])
 	for(l=xtop; l; l=l->next) {
 		if(l->n->op == ODCLFUNC || l->n->op == OCLOSURE) {
 			curfn = l->n;
+			decldepth = 1;
 			saveerrors();
 			typechecklist(l->n->nbody, Etop);
 			checkreturn(l->n);
@@ -399,12 +400,21 @@ gcmain(int argc, char *argv[])
 		}
 	}
 
+	// Phase 4: Decide how to capture variables
+	// and transform closure bodies accordingly.
+	for(l=xtop; l; l=l->next) {
+		if(l->n->op == ODCLFUNC && l->n->closure) {
+			curfn = l->n;
+			capturevars(l->n);
+		}
+	}
+
 	curfn = nil;
 	
 	if(nsavederrors+nerrors)
 		errorexit();
 
-	// Phase 4: Inlining
+	// Phase 5: Inlining
 	if(debug['l'] > 1) {
 		// Typecheck imported function bodies if debug['l'] > 1,
 		// otherwise lazily when used or re-exported.
@@ -430,7 +440,7 @@ gcmain(int argc, char *argv[])
 				inlcalls(l->n);
 	}
 
-	// Phase 5: Escape analysis.
+	// Phase 6: Escape analysis.
 	// Required for moving heap allocations onto stack,
 	// which in turn is required by the closure implementation,
 	// which stores the addresses of stack variables into the closure.
@@ -442,7 +452,7 @@ gcmain(int argc, char *argv[])
 	// Move large values off stack too.
 	movelarge(xtop);
 
-	// Phase 6: Compile top level functions.
+	// Phase 7: Compile top level functions.
 	for(l=xtop; l; l=l->next)
 		if(l->n->op == ODCLFUNC)
 			funccompile(l->n, 0);
@@ -450,7 +460,7 @@ gcmain(int argc, char *argv[])
 	if(nsavederrors+nerrors == 0)
 		fninit(xtop);
 
-	// Phase 7: Check external declarations.
+	// Phase 8: Check external declarations.
 	for(l=externdcl; l; l=l->next)
 		if(l->n->op == ONAME)
 			typecheck(&l->n, Erv);

src/cmd/gc/range.c

@@ -89,14 +89,14 @@ typecheckrange(Node *n)
 			v1->type = t1;
 		else if(v1->type != T && assignop(t1, v1->type, &why) == 0)
 			yyerror("cannot assign type %T to %lN in range%s", t1, v1, why);
-		checkassign(v1);
+		checkassign(n, v1);
 	}
 	if(v2) {
 		if(v2->defn == n)
 			v2->type = t2;
 		else if(v2->type != T && assignop(t2, v2->type, &why) == 0)
 			yyerror("cannot assign type %T to %lN in range%s", t2, v2, why);
-		checkassign(v2);
+		checkassign(n, v2);
 	}
 
 out:

src/cmd/gc/typecheck.c

@@ -27,7 +27,7 @@ static void	typecheckas2(Node*);
 static void	typecheckas(Node*);
 static void	typecheckfunc(Node*);
 static void	checklvalue(Node*, char*);
-static void	checkassignlist(NodeList*);
+static void	checkassignlist(Node*, NodeList*);
 static void	stringtoarraylit(Node**);
 static Node*	resolve(Node*);
 static void	checkdefergo(Node*);
@@ -348,6 +348,8 @@ reswitch:
 		goto ret;
 
 	case ONAME:
+		if(n->decldepth == 0)
+			n->decldepth = decldepth;
 		if(n->etype != 0) {
 			ok |= Ecall;
 			goto ret;
@@ -521,8 +523,8 @@ reswitch:
 	case OASOP:
 		ok |= Etop;
 		l = typecheck(&n->left, Erv);
-		checkassign(n->left);
 		r = typecheck(&n->right, Erv);
+		checkassign(n, n->left);
 		if(l->type == T || r->type == T)
 			goto error;
 		op = n->etype;
@@ -741,11 +743,16 @@ reswitch:
 			goto error;
 		checklvalue(n->left, "take the address of");
 		r = outervalue(n->left);
-		for(l = n->left; l != r; l = l->left)
+		for(l = n->left; l != r; l = l->left) {
 			l->addrtaken = 1;
+			if(l->closure)
+				l->closure->addrtaken = 1;
+		}
 		if(l->orig != l && l->op == ONAME)
 			fatal("found non-orig name node %N", l);
 		l->addrtaken = 1;
+		if(l->closure)
+			l->closure->addrtaken = 1;
 		defaultlit(&n->left, T);
 		l = n->left;
 		if((t = l->type) == T)
@@ -1680,12 +1687,16 @@ reswitch:
 	case ODCL:
 	case OEMPTY:
 	case OGOTO:
-	case OLABEL:
 	case OXFALL:
 	case OVARKILL:
 		ok |= Etop;
 		goto ret;
 
+	case OLABEL:
+		ok |= Etop;
+		decldepth++;
+		goto ret;
+
 	case ODEFER:
 		ok |= Etop;
 		typecheck(&n->left, Etop|Erv);
@@ -1702,11 +1713,13 @@ reswitch:
 	case OFOR:
 		ok |= Etop;
 		typechecklist(n->ninit, Etop);
+		decldepth++;
 		typecheck(&n->ntest, Erv);
 		if(n->ntest != N && (t = n->ntest->type) != T && t->etype != TBOOL)
 			yyerror("non-bool %lN used as for condition", n->ntest);
 		typecheck(&n->nincr, Etop);
 		typechecklist(n->nbody, Etop);
+		decldepth--;
 		goto ret;
 
 	case OIF:
@@ -2811,8 +2824,22 @@ checklvalue(Node *n, char *verb)
 }
 
 void
-checkassign(Node *n)
+checkassign(Node *stmt, Node *n)
 {
+	Node *r, *l;
+
+	if(n->defn != stmt) {
+		r = outervalue(n);
+		for(l = n; l != r; l = l->left) {
+			l->assigned = 1;
+			if(l->closure)
+				l->closure->assigned = 1;
+		}
+		l->assigned = 1;
+		if(l->closure)
+			l->closure->assigned = 1;
+	}
+
 	if(islvalue(n))
 		return;
 	if(n->op == OINDEXMAP) {
@@ -2828,10 +2855,10 @@ checkassign(Node *n)
 }
 
 static void
-checkassignlist(NodeList *l)
+checkassignlist(Node *stmt, NodeList *l)
 {
 	for(; l; l=l->next)
-		checkassign(l->n);
+		checkassign(stmt, l->n);
 }
 
 // Check whether l and r are the same side effect-free expression,
@@ -2881,8 +2908,8 @@ typecheckas(Node *n)
 	if(n->left->defn != n || n->left->ntype)
 		typecheck(&n->left, Erv | Easgn);
 
-	checkassign(n->left);
 	typecheck(&n->right, Erv);
+	checkassign(n, n->left);
 	if(n->right && n->right->type != T) {
 		if(n->left->type != T)
 			n->right = assignconv(n->right, n->left->type, "assignment");
@@ -2953,11 +2980,11 @@ typecheckas2(Node *n)
 	}
 	cl = count(n->list);
 	cr = count(n->rlist);
-	checkassignlist(n->list);
 	if(cl > 1 && cr == 1)
 		typecheck(&n->rlist->n, Erv | Efnstruct);
 	else
 		typechecklist(n->rlist, Erv);
+	checkassignlist(n, n->list);
 
 	if(cl == cr) {
 		// easy
@@ -3048,6 +3075,7 @@ static void
 typecheckfunc(Node *n)
 {
 	Type *t, *rcvr;
+	NodeList *l;
 
 	typecheck(&n->nname, Erv | Easgn);
 	if((t = n->nname->type) == T)
@@ -3057,6 +3085,10 @@ typecheckfunc(Node *n)
 	rcvr = getthisx(t)->type;
 	if(rcvr != nil && n->shortname != N && !isblank(n->shortname))
 		addmethod(n->shortname->sym, t, 1, n->nname->nointerface);
+
+	for(l=n->dcl; l; l=l->next)
+		if(l->n->op == ONAME && (l->n->class == PPARAM || l->n->class == PPARAMOUT))
+			l->n->decldepth = 1;
 }
 
 static void

src/cmd/gc/walk.c

@@ -1973,7 +1973,7 @@ isglobal(Node *n)
 {
 	while(n->op == ODOT || n->op == OPAREN || n->op == OCONVNOP || n->op == OINDEX && isfixedarray(n->left->type))
 		n = n->left;
-	
+
 	switch(n->op) {
 	case ONAME:
 		switch(n->class) {

src/runtime/proc_test.go

@@ -7,6 +7,7 @@ package runtime_test
 import (
 	"math"
 	"runtime"
+	"sync"
 	"sync/atomic"
 	"syscall"
 	"testing"
@@ -415,6 +416,25 @@ func benchmarkCreateGoroutines(b *testing.B, procs int) {
 	}
 }
 
+func BenchmarkCreateGoroutinesCapture(b *testing.B) {
+	b.ReportAllocs()
+	for i := 0; i < b.N; i++ {
+		const N = 4
+		var wg sync.WaitGroup
+		wg.Add(N)
+		for i := 0; i < N; i++ {
+			i := i
+			go func() {
+				if i >= N {
+					b.Logf("bad") // just to capture b
+				}
+				wg.Done()
+			}()
+		}
+		wg.Wait()
+	}
+}
+
 type Matrix [][]float64
 
 func BenchmarkMatmult(b *testing.B) {

test/closure1.go

+// run
+
+// Copyright 2015 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+func main() {
+	x := 0
+	func() {
+		x = 1
+	}()
+	func() {
+		if x != 1 {
+			panic("x != 1")
+		}
+	}()
+}
\ No newline at end of file

test/closure2.go

+// run
+
+// Copyright 2015 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Check that these do not use "by value" capturing,
+// because changes are made to the value during the closure.
+
+package main
+
+func main() {
+	type X struct {
+		v int
+	}
+	var x X
+	func() {
+		x.v++
+	}()
+	if x.v != 1 {
+		panic("x.v != 1")
+	}
+
+	type Y struct {
+		X
+	}
+	var y Y
+	func() {
+		y.v = 1
+	}()
+	if y.v != 1 {
+		panic("y.v != 1")
+	}
+
+	type Z struct {
+		a [3]byte
+	}
+	var z Z
+	func() {
+		i := 0
+		for z.a[1] = 1; i < 10; i++ {
+		}
+	}()
+	if z.a[1] != 1 {
+		panic("z.a[1] != 1")
+	}
+
+	w := 0
+	tmp := 0
+	f := func() {
+		if w != 1 {
+			panic("w != 1")
+		}
+	}
+	func() {
+		tmp = w // force capture of w, but do not write to it yet
+		func() {
+			func() {
+				w++ // write in a nested closure
+			}()
+		}()
+	}()
+	f()
+}

test/escape2.go

@@ -203,9 +203,17 @@ func (b *Bar2) LeakSelf2() { // ERROR "leaking param: b"
 }
 
 func foo21() func() int {
+	x := 42
+	return func() int { // ERROR "func literal escapes to heap"
+		return x
+	}
+}
+
+func foo21a() func() int {
 	x := 42             // ERROR "moved to heap: x"
 	return func() int { // ERROR "func literal escapes to heap"
-		return x // ERROR "&x escapes to heap"
+		x++  // ERROR "&x escapes to heap"
+		return x
 	}
 }
 
@@ -216,24 +224,31 @@ func foo22() int {
 	}()
 }
 
-func foo23(x int) func() int { // ERROR "moved to heap: x"
+func foo23(x int) func() int {
 	return func() int { // ERROR "func literal escapes to heap"
-		return x // ERROR "&x escapes to heap"
+		return x
 	}
 }
 
-func foo23a(x int) func() int { // ERROR "moved to heap: x"
+func foo23a(x int) func() int {
 	f := func() int { // ERROR "func literal escapes to heap"
-		return x // ERROR "&x escapes to heap"
+		return x
 	}
 	return f
 }
 
-func foo23b(x int) *(func() int) { // ERROR "moved to heap: x"
-	f := func() int { return x } // ERROR "moved to heap: f" "func literal escapes to heap" "&x escapes to heap"
+func foo23b(x int) *(func() int) {
+	f := func() int { return x } // ERROR "moved to heap: f" "func literal escapes to heap"
 	return &f                    // ERROR "&f escapes to heap"
 }
 
+func foo23c(x int) func() int { // ERROR "moved to heap: x"
+	return func() int { // ERROR "func literal escapes to heap"
+		x++ // ERROR "&x escapes to heap"
+		return x
+	}
+}
+
 func foo24(x int) int {
 	return func() int { // ERROR "func literal does not escape"
 		return x
@@ -523,23 +538,48 @@ func foo72b() [10]*int {
 
 // issue 2145
 func foo73() {
+	s := []int{3, 2, 1} // ERROR "\[\]int literal does not escape"
+	for _, v := range s {
+		vv := v
+		// actually just escapes its scope
+		defer func() { // ERROR "func literal escapes to heap"
+			println(vv)
+		}()
+	}
+}
+
+func foo731() {
 	s := []int{3, 2, 1} // ERROR "\[\]int literal does not escape"
 	for _, v := range s {
 		vv := v // ERROR "moved to heap: vv"
 		// actually just escapes its scope
 		defer func() { // ERROR "func literal escapes to heap"
-			println(vv) // ERROR "&vv escapes to heap"
+			vv = 42 // ERROR "&vv escapes to heap"
+			println(vv)
 		}()
 	}
 }
 
 func foo74() {
+	s := []int{3, 2, 1} // ERROR "\[\]int literal does not escape"
+	for _, v := range s {
+		vv := v
+		// actually just escapes its scope
+		fn := func() { // ERROR "func literal escapes to heap"
+			println(vv)
+		}
+		defer fn()
+	}
+}
+
+func foo74a() {
 	s := []int{3, 2, 1} // ERROR "\[\]int literal does not escape"
 	for _, v := range s {
 		vv := v // ERROR "moved to heap: vv"
 		// actually just escapes its scope
 		fn := func() { // ERROR "func literal escapes to heap"
-			println(vv) // ERROR "&vv escapes to heap"
+			vv += 1 // ERROR "&vv escapes to heap"
+			println(vv)
 		}
 		defer fn()
 	}
@@ -547,13 +587,25 @@ func foo74() {
 
 // issue 3975
 func foo74b() {
+	var array [3]func()
+	s := []int{3, 2, 1} // ERROR "\[\]int literal does not escape"
+	for i, v := range s {
+		vv := v
+		// actually just escapes its scope
+		array[i] = func() { // ERROR "func literal escapes to heap"
+			println(vv)
+		}
+	}
+}
+
+func foo74c() {
 	var array [3]func()
 	s := []int{3, 2, 1} // ERROR "\[\]int literal does not escape"
 	for i, v := range s {
 		vv := v // ERROR "moved to heap: vv"
 		// actually just escapes its scope
 		array[i] = func() { // ERROR "func literal escapes to heap"
-			println(vv) // ERROR "&vv escapes to heap"
+			println(&vv) // ERROR "&vv escapes to heap" "&vv does not escape"
 		}
 	}
 }
@@ -1213,9 +1265,9 @@ func foo134() {
 
 func foo135() {
 	var i int   // ERROR "moved to heap: i"
-	p := &i     // ERROR "&i escapes to heap" "moved to heap: p"
+	p := &i     // ERROR "&i escapes to heap"
 	go func() { // ERROR "func literal escapes to heap"
-		q := p   // ERROR "&p escapes to heap"
+		q := p
 		func() { // ERROR "func literal does not escape"
 			r := q
 			_ = r
@@ -1225,9 +1277,9 @@ func foo135() {
 
 func foo136() {
 	var i int   // ERROR "moved to heap: i"
-	p := &i     // ERROR "&i escapes to heap" "moved to heap: p"
+	p := &i     // ERROR "&i escapes to heap"
 	go func() { // ERROR "func literal escapes to heap"
-		q := p   // ERROR "&p escapes to heap" "leaking closure reference p"
+		q := p   // ERROR "leaking closure reference p"
 		func() { // ERROR "func literal does not escape"
 			r := q // ERROR "leaking closure reference q"
 			px = r
@@ -1239,9 +1291,9 @@ func foo137() {
 	var i int // ERROR "moved to heap: i"
 	p := &i   // ERROR "&i escapes to heap"
 	func() {  // ERROR "func literal does not escape"
-		q := p      // ERROR "leaking closure reference p" "moved to heap: q"
+		q := p      // ERROR "leaking closure reference p"
 		go func() { // ERROR "func literal escapes to heap"
-			r := q // ERROR "&q escapes to heap"
+			r := q
 			_ = r
 		}()
 	}()

test/escape2n.go

@@ -203,9 +203,17 @@ func (b *Bar2) LeakSelf2() { // ERROR "leaking param: b"
 }
 
 func foo21() func() int {
+	x := 42
+	return func() int { // ERROR "func literal escapes to heap"
+		return x
+	}
+}
+
+func foo21a() func() int {
 	x := 42             // ERROR "moved to heap: x"
 	return func() int { // ERROR "func literal escapes to heap"
-		return x // ERROR "&x escapes to heap"
+		x++  // ERROR "&x escapes to heap"
+		return x
 	}
 }
 
@@ -216,24 +224,31 @@ func foo22() int {
 	}()
 }
 
-func foo23(x int) func() int { // ERROR "moved to heap: x"
+func foo23(x int) func() int {
 	return func() int { // ERROR "func literal escapes to heap"
-		return x // ERROR "&x escapes to heap"
+		return x
 	}
 }
 
-func foo23a(x int) func() int { // ERROR "moved to heap: x"
+func foo23a(x int) func() int {
 	f := func() int { // ERROR "func literal escapes to heap"
-		return x // ERROR "&x escapes to heap"
+		return x
 	}
 	return f
 }
 
-func foo23b(x int) *(func() int) { // ERROR "moved to heap: x"
-	f := func() int { return x } // ERROR "moved to heap: f" "func literal escapes to heap" "&x escapes to heap"
+func foo23b(x int) *(func() int) {
+	f := func() int { return x } // ERROR "moved to heap: f" "func literal escapes to heap"
 	return &f                    // ERROR "&f escapes to heap"
 }
 
+func foo23c(x int) func() int { // ERROR "moved to heap: x"
+	return func() int { // ERROR "func literal escapes to heap"
+		x++ // ERROR "&x escapes to heap"
+		return x
+	}
+}
+
 func foo24(x int) int {
 	return func() int { // ERROR "func literal does not escape"
 		return x
@@ -523,23 +538,48 @@ func foo72b() [10]*int {
 
 // issue 2145
 func foo73() {
+	s := []int{3, 2, 1} // ERROR "\[\]int literal does not escape"
+	for _, v := range s {
+		vv := v
+		// actually just escapes its scope
+		defer func() { // ERROR "func literal escapes to heap"
+			println(vv)
+		}()
+	}
+}
+
+func foo731() {
 	s := []int{3, 2, 1} // ERROR "\[\]int literal does not escape"
 	for _, v := range s {
 		vv := v // ERROR "moved to heap: vv"
 		// actually just escapes its scope
 		defer func() { // ERROR "func literal escapes to heap"
-			println(vv) // ERROR "&vv escapes to heap"
+			vv = 42 // ERROR "&vv escapes to heap"
+			println(vv)
 		}()
 	}
 }
 
 func foo74() {
+	s := []int{3, 2, 1} // ERROR "\[\]int literal does not escape"
+	for _, v := range s {
+		vv := v
+		// actually just escapes its scope
+		fn := func() { // ERROR "func literal escapes to heap"
+			println(vv)
+		}
+		defer fn()
+	}
+}
+
+func foo74a() {
 	s := []int{3, 2, 1} // ERROR "\[\]int literal does not escape"
 	for _, v := range s {
 		vv := v // ERROR "moved to heap: vv"
 		// actually just escapes its scope
 		fn := func() { // ERROR "func literal escapes to heap"
-			println(vv) // ERROR "&vv escapes to heap"
+			vv += 1 // ERROR "&vv escapes to heap"
+			println(vv)
 		}
 		defer fn()
 	}
@@ -547,13 +587,25 @@ func foo74() {
 
 // issue 3975
 func foo74b() {
+	var array [3]func()
+	s := []int{3, 2, 1} // ERROR "\[\]int literal does not escape"
+	for i, v := range s {
+		vv := v
+		// actually just escapes its scope
+		array[i] = func() { // ERROR "func literal escapes to heap"
+			println(vv)
+		}
+	}
+}
+
+func foo74c() {
 	var array [3]func()
 	s := []int{3, 2, 1} // ERROR "\[\]int literal does not escape"
 	for i, v := range s {
 		vv := v // ERROR "moved to heap: vv"
 		// actually just escapes its scope
 		array[i] = func() { // ERROR "func literal escapes to heap"
-			println(vv) // ERROR "&vv escapes to heap"
+			println(&vv) // ERROR "&vv escapes to heap" "&vv does not escape"
 		}
 	}
 }
@@ -1213,9 +1265,9 @@ func foo134() {
 
 func foo135() {
 	var i int   // ERROR "moved to heap: i"
-	p := &i     // ERROR "&i escapes to heap" "moved to heap: p"
+	p := &i     // ERROR "&i escapes to heap"
 	go func() { // ERROR "func literal escapes to heap"
-		q := p   // ERROR "&p escapes to heap"
+		q := p
 		func() { // ERROR "func literal does not escape"
 			r := q
 			_ = r
@@ -1225,9 +1277,9 @@ func foo135() {
 
 func foo136() {
 	var i int   // ERROR "moved to heap: i"
-	p := &i     // ERROR "&i escapes to heap" "moved to heap: p"
+	p := &i     // ERROR "&i escapes to heap"
 	go func() { // ERROR "func literal escapes to heap"
-		q := p   // ERROR "&p escapes to heap" "leaking closure reference p"
+		q := p   // ERROR "leaking closure reference p"
 		func() { // ERROR "func literal does not escape"
 			r := q // ERROR "leaking closure reference q"
 			px = r
@@ -1239,9 +1291,9 @@ func foo137() {
 	var i int // ERROR "moved to heap: i"
 	p := &i   // ERROR "&i escapes to heap"
 	func() {  // ERROR "func literal does not escape"
-		q := p      // ERROR "leaking closure reference p" "moved to heap: q"
+		q := p      // ERROR "leaking closure reference p"
 		go func() { // ERROR "func literal escapes to heap"
-			r := q // ERROR "&q escapes to heap"
+			r := q
 			_ = r
 		}()
 	}()