[dev.garbage] all: merge default (f38460037b72) into dev.garbage

This is the revision that dev.cc is branched from.

LGTM=austin
R=austin
CC=golang-codereviews
https://codereview.appspot.com/169590043

9 files changed, 140 insertions, 36 deletions

diff --git a/src/cmd/gc/builtin.c b/src/cmd/gc/builtin.c
index fbca4ee5f..aeeadedca 100644
--- a/src/cmd/gc/builtin.c
+++ b/src/cmd/gc/builtin.c
@@ -24,6 +24,8 @@ char *runtimeimport =
 	"func @\"\".printslice (? any)\n"
 	"func @\"\".printnl ()\n"
 	"func @\"\".printsp ()\n"
+	"func @\"\".printlock ()\n"
+	"func @\"\".printunlock ()\n"
 	"func @\"\".concatstring2 (? string, ? string) (? string)\n"
 	"func @\"\".concatstring3 (? string, ? string, ? string) (? string)\n"
 	"func @\"\".concatstring4 (? string, ? string, ? string, ? string) (? string)\n"
@@ -86,10 +88,33 @@ char *runtimeimport =
 	"func @\"\".writebarrierstring (@\"\".dst·1 *any, @\"\".src·2 any)\n"
 	"func @\"\".writebarrierslice (@\"\".dst·1 *any, @\"\".src·2 any)\n"
 	"func @\"\".writebarrieriface (@\"\".dst·1 *any, @\"\".src·2 any)\n"
-	"func @\"\".writebarrierfat2 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
-	"func @\"\".writebarrierfat3 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
-	"func @\"\".writebarrierfat4 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat01 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat10 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat11 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat001 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat010 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat011 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat100 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat101 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat110 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat111 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat0001 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat0010 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat0011 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat0100 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat0101 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat0110 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat0111 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat1000 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat1001 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat1010 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat1011 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat1100 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat1101 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat1110 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
+	"func @\"\".writebarrierfat1111 (@\"\".dst·1 *any, _ *byte, @\"\".src·3 any)\n"
 	"func @\"\".writebarrierfat (@\"\".typ·1 *byte, @\"\".dst·2 *any, @\"\".src·3 *any)\n"
+	"func @\"\".writebarriercopy (@\"\".typ·2 *byte, @\"\".dst·3 any, @\"\".src·4 any) (? int)\n"
 	"func @\"\".selectnbsend (@\"\".chanType·2 *byte, @\"\".hchan·3 chan<- any, @\"\".elem·4 *any) (? bool)\n"
 	"func @\"\".selectnbrecv (@\"\".chanType·2 *byte, @\"\".elem·3 *any, @\"\".hchan·4 <-chan any) (? bool)\n"
 	"func @\"\".selectnbrecv2 (@\"\".chanType·2 *byte, @\"\".elem·3 *any, @\"\".received·4 *bool, @\"\".hchan·5 <-chan any) (? bool)\n"
diff --git a/src/cmd/gc/go.h b/src/cmd/gc/go.h
index 965a0550d..cc590416b 100644
--- a/src/cmd/gc/go.h
+++ b/src/cmd/gc/go.h
@@ -1464,6 +1464,7 @@ void	walk(Node *fn);
 void	walkexpr(Node **np, NodeList **init);
 void	walkexprlist(NodeList *l, NodeList **init);
 void	walkexprlistsafe(NodeList *l, NodeList **init);
+void	walkexprlistcheap(NodeList *l, NodeList **init);
 void	walkstmt(Node **np);
 void	walkstmtlist(NodeList *l);
 Node*	conv(Node*, Type*);
diff --git a/src/cmd/gc/pgen.c b/src/cmd/gc/pgen.c
index 39028e3f8..259cec85a 100644
--- a/src/cmd/gc/pgen.c
+++ b/src/cmd/gc/pgen.c
@@ -471,7 +471,7 @@ allocauto(Prog* ptxt)
 		stksize = rnd(stksize, n->type->align);
 		if(haspointers(n->type))
 			stkptrsize = stksize;
-		if(thechar == '5')
+		if(thechar == '5' || thechar == '9')
 			stksize = rnd(stksize, widthptr);
 		if(stksize >= (1ULL<<31)) {
 			setlineno(curfn);
@@ -528,7 +528,7 @@ cgen_checknil(Node *n)
 		dump("checknil", n);
 		fatal("bad checknil");
 	}
-	if((thechar == '5' && n->op != OREGISTER) || !n->addable || n->op == OLITERAL) {
+	if(((thechar == '5' || thechar == '9') && n->op != OREGISTER) || !n->addable || n->op == OLITERAL) {
 		regalloc(&reg, types[tptr], n);
 		cgen(n, &reg);
 		gins(ACHECKNIL, &reg, N);
diff --git a/src/cmd/gc/plive.c b/src/cmd/gc/plive.c
index 0feb2c710..3bfa69b1f 100644
--- a/src/cmd/gc/plive.c
+++ b/src/cmd/gc/plive.c
@@ -1092,7 +1092,7 @@ twobitwalktype1(Type *t, vlong *xoffset, Bvec *bv)
 	case TCOMPLEX64:
 	case TCOMPLEX128:
 		for(i = 0; i < t->width; i++) {
-			bvset(bv, ((*xoffset + i) / widthptr) * BitsPerPointer); // 1 = live scalar
+			bvset(bv, ((*xoffset + i) / widthptr) * BitsPerPointer); // 1 = live scalar (BitsScalar)
 		}
 		*xoffset += t->width;
 		break;
@@ -1105,7 +1105,7 @@ twobitwalktype1(Type *t, vlong *xoffset, Bvec *bv)
 	case TMAP:
 		if((*xoffset & (widthptr-1)) != 0)
 			fatal("twobitwalktype1: invalid alignment, %T", t);
-		bvset(bv, (*xoffset / widthptr) * BitsPerPointer + 1); // 2 = live ptr
+		bvset(bv, (*xoffset / widthptr) * BitsPerPointer + 1); // 2 = live ptr (BitsPointer)
 		*xoffset += t->width;
 		break;
 
@@ -1113,7 +1113,7 @@ twobitwalktype1(Type *t, vlong *xoffset, Bvec *bv)
 		// struct { byte *str; intgo len; }
 		if((*xoffset & (widthptr-1)) != 0)
 			fatal("twobitwalktype1: invalid alignment, %T", t);
-		bvset(bv, (*xoffset / widthptr) * BitsPerPointer + 1); // 2 = live ptr in first slot
+		bvset(bv, (*xoffset / widthptr) * BitsPerPointer + 1); // 2 = live ptr in first slot (BitsPointer)
 		*xoffset += t->width;
 		break;
 
@@ -1123,15 +1123,8 @@ twobitwalktype1(Type *t, vlong *xoffset, Bvec *bv)
 		// struct { Type *type; union { void *ptr, uintptr val } data; }
 		if((*xoffset & (widthptr-1)) != 0)
 			fatal("twobitwalktype1: invalid alignment, %T", t);
-		bvset(bv, ((*xoffset / widthptr) * BitsPerPointer) + 0);
-		bvset(bv, ((*xoffset / widthptr) * BitsPerPointer) + 1); // 3 = multiword
-		// next word contains 2 = Iface, 3 = Eface
-		if(isnilinter(t)) {
-			bvset(bv, ((*xoffset / widthptr) * BitsPerPointer) + 2);
-			bvset(bv, ((*xoffset / widthptr) * BitsPerPointer) + 3);
-		} else {
-			bvset(bv, ((*xoffset / widthptr) * BitsPerPointer) + 3);
-		}
+		bvset(bv, (*xoffset / widthptr) * BitsPerPointer + 1); // 2 = live ptr in first slot (BitsPointer)
+		bvset(bv, (*xoffset / widthptr) * BitsPerPointer + 3); // 2 = live ptr in second slot (BitsPointer)
 		*xoffset += t->width;
 		break;
 
@@ -1144,7 +1137,7 @@ twobitwalktype1(Type *t, vlong *xoffset, Bvec *bv)
 			// struct { byte *array; uintgo len; uintgo cap; }
 			if((*xoffset & (widthptr-1)) != 0)
 				fatal("twobitwalktype1: invalid TARRAY alignment, %T", t);
-			bvset(bv, (*xoffset / widthptr) * BitsPerPointer + 1); // 2 = live ptr in first slot
+			bvset(bv, (*xoffset / widthptr) * BitsPerPointer + 1); // 2 = live ptr in first slot (BitsPointer)
 			*xoffset += t->width;
 		} else
 			for(i = 0; i < t->bound; i++)
diff --git a/src/cmd/gc/popt.c b/src/cmd/gc/popt.c
index 993bb2482..6e6db88ef 100644
--- a/src/cmd/gc/popt.c
+++ b/src/cmd/gc/popt.c
@@ -847,6 +847,10 @@ nilopt(Prog *firstp)
 	Graph *g;
 	int ncheck, nkill;
 
+	// TODO(minux): nilopt on power64 throw away seemly random segment of code.
+	if(thechar == '9')
+		return;
+
 	g = flowstart(firstp, sizeof(NilFlow));
 	if(g == nil)
 		return;
diff --git a/src/cmd/gc/reflect.c b/src/cmd/gc/reflect.c
index b2ff2fbc5..0f8802abc 100644
--- a/src/cmd/gc/reflect.c
+++ b/src/cmd/gc/reflect.c
@@ -1525,11 +1525,9 @@ gengcprog1(ProgGen *g, Type *t, vlong *xoffset)
 		*xoffset += t->width;
 		break;
 	case TINTER:
-		proggendata(g, BitsMultiWord);
-		if(isnilinter(t))
-			proggendata(g, BitsEface);
-		else
-			proggendata(g, BitsIface);
+		// Assuming IfacePointerOnly=1.
+		proggendata(g, BitsPointer);
+		proggendata(g, BitsPointer);
 		*xoffset += t->width;
 		break;
 	case TARRAY:
diff --git a/src/cmd/gc/runtime.go b/src/cmd/gc/runtime.go
index 0fb15c265..c6007714c 100644
--- a/src/cmd/gc/runtime.go
+++ b/src/cmd/gc/runtime.go
@@ -36,6 +36,8 @@ func printeface(any)
 func printslice(any)
 func printnl()
 func printsp()
+func printlock()
+func printunlock()
 
 func concatstring2(string, string) string
 func concatstring3(string, string, string) string
@@ -115,10 +117,35 @@ func writebarrieriface(dst *any, src any)
 // The unused *byte argument makes sure that src is 2-pointer-aligned,
 // which is the maximum alignment on NaCl amd64p32
 // (and possibly on 32-bit systems if we start 64-bit aligning uint64s).
-func writebarrierfat2(dst *any, _ *byte, src any)
-func writebarrierfat3(dst *any, _ *byte, src any)
-func writebarrierfat4(dst *any, _ *byte, src any)
+// The bitmap in the name tells which words being copied are pointers.
+func writebarrierfat01(dst *any, _ *byte, src any)
+func writebarrierfat10(dst *any, _ *byte, src any)
+func writebarrierfat11(dst *any, _ *byte, src any)
+func writebarrierfat001(dst *any, _ *byte, src any)
+func writebarrierfat010(dst *any, _ *byte, src any)
+func writebarrierfat011(dst *any, _ *byte, src any)
+func writebarrierfat100(dst *any, _ *byte, src any)
+func writebarrierfat101(dst *any, _ *byte, src any)
+func writebarrierfat110(dst *any, _ *byte, src any)
+func writebarrierfat111(dst *any, _ *byte, src any)
+func writebarrierfat0001(dst *any, _ *byte, src any)
+func writebarrierfat0010(dst *any, _ *byte, src any)
+func writebarrierfat0011(dst *any, _ *byte, src any)
+func writebarrierfat0100(dst *any, _ *byte, src any)
+func writebarrierfat0101(dst *any, _ *byte, src any)
+func writebarrierfat0110(dst *any, _ *byte, src any)
+func writebarrierfat0111(dst *any, _ *byte, src any)
+func writebarrierfat1000(dst *any, _ *byte, src any)
+func writebarrierfat1001(dst *any, _ *byte, src any)
+func writebarrierfat1010(dst *any, _ *byte, src any)
+func writebarrierfat1011(dst *any, _ *byte, src any)
+func writebarrierfat1100(dst *any, _ *byte, src any)
+func writebarrierfat1101(dst *any, _ *byte, src any)
+func writebarrierfat1110(dst *any, _ *byte, src any)
+func writebarrierfat1111(dst *any, _ *byte, src any)
+
 func writebarrierfat(typ *byte, dst *any, src *any)
+func writebarriercopy(typ *byte, dst any, src any) int
 
 func selectnbsend(chanType *byte, hchan chan<- any, elem *any) bool
 func selectnbrecv(chanType *byte, elem *any, hchan <-chan any) bool
diff --git a/src/cmd/gc/typecheck.c b/src/cmd/gc/typecheck.c
index 714c66268..f05d8022d 100644
--- a/src/cmd/gc/typecheck.c
+++ b/src/cmd/gc/typecheck.c
@@ -2891,7 +2891,8 @@ typecheckas(Node *n)
 		case OSLICE3:
 		case OSLICESTR:
 			// For x = x[0:y], x can be updated in place, without touching pointer.
-			if(samesafeexpr(n->left, n->right->left) && (n->right->right->left == N || iszero(n->right->right->left)))
+			// TODO(rsc): Reenable once it is actually updated in place without touching the pointer.
+			if(0 && samesafeexpr(n->left, n->right->left) && (n->right->right->left == N || iszero(n->right->right->left)))
 				n->right->reslice = 1;
 			break;
 		
@@ -2899,7 +2900,8 @@ typecheckas(Node *n)
 			// For x = append(x, ...), x can be updated in place when there is capacity,
 			// without touching the pointer; otherwise the emitted code to growslice
 			// can take care of updating the pointer, and only in that case.
-			if(n->right->list != nil && samesafeexpr(n->left, n->right->list->n))
+			// TODO(rsc): Reenable once the emitted code does update the pointer.
+			if(0 && n->right->list != nil && samesafeexpr(n->left, n->right->list->n))
 				n->right->reslice = 1;
 			break;
 		}
diff --git a/src/cmd/gc/walk.c b/src/cmd/gc/walk.c
index ff9b36208..37bd62dea 100644
--- a/src/cmd/gc/walk.c
+++ b/src/cmd/gc/walk.c
@@ -6,6 +6,7 @@
 #include	<libc.h>
 #include	"go.h"
 #include	"../ld/textflag.h"
+#include	"../../runtime/mgc0.h"
 
 static	Node*	walkprint(Node*, NodeList**);
 static	Node*	writebarrierfn(char*, Type*, Type*);
@@ -363,6 +364,15 @@ walkexprlistsafe(NodeList *l, NodeList **init)
 }
 
 void
+walkexprlistcheap(NodeList *l, NodeList **init)
+{
+	for(; l; l=l->next) {
+		l->n = cheapexpr(l->n, init);
+		walkexpr(&l->n, init);
+	}
+}
+
+void
 walkexpr(Node **np, NodeList **init)
 {
 	Node *r, *l, *var, *a;
@@ -1772,6 +1782,11 @@ walkprint(Node *nn, NodeList **init)
 	calls = nil;
 	notfirst = 0;
 
+	// Hoist all the argument evaluation up before the lock.
+	walkexprlistcheap(all, init);
+
+	calls = list(calls, mkcall("printlock", T, init));
+
 	for(l=all; l; l=l->next) {
 		if(notfirst) {
 			calls = list(calls, mkcall("printsp", T, init));
@@ -1852,6 +1867,9 @@ walkprint(Node *nn, NodeList **init)
 
 	if(op == OPRINTN)
 		calls = list(calls, mkcall("printnl", T, nil));
+
+	calls = list(calls, mkcall("printunlock", T, init));
+
 	typechecklist(calls, Etop);
 	walkexprlist(calls, init);
 
@@ -1988,6 +2006,9 @@ applywritebarrier(Node *n, NodeList **init)
 {
 	Node *l, *r;
 	Type *t;
+	vlong x;
+	static Bvec *bv;
+	char name[32];
 
 	if(n->left && n->right && needwritebarrier(n->left, n->right)) {
 		t = n->left->type;
@@ -2005,14 +2026,35 @@ applywritebarrier(Node *n, NodeList **init)
 		} else if(isinter(t)) {
 			n = mkcall1(writebarrierfn("writebarrieriface", t, n->right->type), T, init,
 				l, n->right);
-		} else if(t->width == 2*widthptr) {
-			n = mkcall1(writebarrierfn("writebarrierfat2", t, n->right->type), T, init,
-				l, nodnil(), n->right);
-		} else if(t->width == 3*widthptr) {
-			n = mkcall1(writebarrierfn("writebarrierfat3", t, n->right->type), T, init,
-				l, nodnil(), n->right);
-		} else if(t->width == 4*widthptr) {
-			n = mkcall1(writebarrierfn("writebarrierfat4", t, n->right->type), T, init,
+		} else if(t->width <= 4*widthptr) {
+			x = 0;
+			if(bv == nil)
+				bv = bvalloc(BitsPerPointer*4);
+			bvresetall(bv);
+			twobitwalktype1(t, &x, bv);
+			// The bvgets are looking for BitsPointer in successive slots.
+			enum {
+				PtrBit = 1,
+			};
+			if(BitsPointer != (1<<PtrBit))
+				fatal("wrong PtrBit");
+			switch(t->width/widthptr) {
+			default:
+				fatal("found writebarrierfat for %d-byte object of type %T", (int)t->width, t);
+			case 2:
+				snprint(name, sizeof name, "writebarrierfat%d%d",
+					bvget(bv, PtrBit), bvget(bv, BitsPerPointer+PtrBit));
+				break;
+			case 3:
+				snprint(name, sizeof name, "writebarrierfat%d%d%d",
+					bvget(bv, PtrBit), bvget(bv, BitsPerPointer+PtrBit), bvget(bv, 2*BitsPerPointer+PtrBit));
+				break;
+			case 4:
+				snprint(name, sizeof name, "writebarrierfat%d%d%d%d",
+					bvget(bv, PtrBit), bvget(bv, BitsPerPointer+PtrBit), bvget(bv, 2*BitsPerPointer+PtrBit), bvget(bv, 3*BitsPerPointer+PtrBit));
+				break;
+			}
+			n = mkcall1(writebarrierfn(name, t, n->right->type), T, init,
 				l, nodnil(), n->right);
 		} else {
 			r = n->right;
@@ -2874,6 +2916,11 @@ copyany(Node *n, NodeList **init, int runtimecall)
 {
 	Node *nl, *nr, *nfrm, *nto, *nif, *nlen, *nwid, *fn;
 	NodeList *l;
+	
+	if(haspointers(n->left->type->type)) {
+		fn = writebarrierfn("writebarriercopy", n->left->type, n->right->type);
+		return mkcall1(fn, n->type, init, typename(n->left->type->type), n->left, n->right);
+	}
 
 	if(runtimecall) {
 		if(n->right->type->etype == TSTRING)
@@ -3301,6 +3348,9 @@ walkrotate(Node **np)
 	int w, sl, sr, s;
 	Node *l, *r;
 	Node *n;
+
+	if(thechar == '9')
+		return;
 	
 	n = *np;
 
@@ -3426,6 +3476,10 @@ walkdiv(Node **np, NodeList **init)
 	Type *twide;
 	Magic m;
 
+	// TODO(minux)
+	if(thechar == '9')
+		return;
+
 	n = *np;
 	if(n->right->op != OLITERAL)
 		return;