runtime: replace runtime·rnd function with ROUND macro

It's sad to introduce a new macro, but rnd shows up consistently
in profiles, and the function call overwhelms the two arithmetic
instructions it performs.

R=r
CC=golang-dev
https://golang.org/cl/6260051

6 files changed, 35 insertions, 48 deletions

diff --git a/src/pkg/runtime/chan.c b/src/pkg/runtime/chan.c
index ef27144ef3..07ab431b43 100644
--- a/src/pkg/runtime/chan.c
+++ b/src/pkg/runtime/chan.c
@@ -86,7 +86,7 @@ runtime·makechan_c(ChanType *t, int64 hint)
 	Hchan *c;
 	int32 n;
 	Type *elem;
-	
+
 	elem = t->elem;
 
 	if(hint < 0 || (int32)hint != hint || (elem->size > 0 && hint > ((uintptr)-1) / elem->size))
@@ -180,7 +180,7 @@ runtime·chansend(ChanType *t, Hchan *c, byte *ep, bool *pres)
 	sg = dequeue(&c->recvq);
 	if(sg != nil) {
 		runtime·unlock(c);
-		
+
 		gp = sg->g;
 		gp->param = sg;
 		if(sg->elem != nil)
@@ -446,7 +446,7 @@ runtime·selectnbsend(ChanType *t, Hchan *c, ...)
 	byte *ae, *ap;
 
 	ae = (byte*)(&c + 1);
-	ap = ae + runtime·rnd(t->elem->size, Structrnd);
+	ap = ae + ROUND(t->elem->size, Structrnd);
 	runtime·chansend(t, c, ae, ap);
 }
 
@@ -474,7 +474,7 @@ void
 runtime·selectnbrecv(ChanType *t, byte *v, Hchan *c, bool selected)
 {
 	runtime·chanrecv(t, c, v, &selected, nil);
-}	
+}
 
 // func selectnbrecv2(elem *any, ok *bool, c chan any) bool
 //
@@ -500,7 +500,7 @@ void
 runtime·selectnbrecv2(ChanType *t, byte *v, bool *received, Hchan *c, bool selected)
 {
 	runtime·chanrecv(t, c, v, &selected, received);
-}	
+}
 
 // For reflect:
 //	func chansend(c chan, val iword, nb bool) (selected bool)
@@ -514,7 +514,7 @@ reflect·chansend(ChanType *t, Hchan *c, uintptr val, bool nb, uintptr selected)
 {
 	bool *sp;
 	byte *vp;
-	
+
 	if(nb) {
 		selected = false;
 		sp = (bool*)&selected;
@@ -571,7 +571,7 @@ runtime·newselect(int32 size, ...)
 	int32 o;
 	Select **selp;
 
-	o = runtime·rnd(sizeof(size), Structrnd);
+	o = ROUND(sizeof(size), Structrnd);
 	selp = (Select**)((byte*)&size + o);
 	newselect(size, selp);
 }
@@ -619,7 +619,7 @@ runtime·selectsend(Select *sel, Hchan *c, void *elem, bool selected)
 	// nil cases do not compete
 	if(c == nil)
 		return;
-	
+
 	selectsend(sel, c, runtime·getcallerpc(&sel), elem, (byte*)&selected - (byte*)&sel);
 }
 
@@ -628,7 +628,7 @@ selectsend(Select *sel, Hchan *c, void *pc, void *elem, int32 so)
 {
 	int32 i;
 	Scase *cas;
-	
+
 	i = sel->ncase;
 	if(i >= sel->tcase)
 		runtime·throw("selectsend: too many cases");
@@ -899,7 +899,7 @@ loop:
 		case CaseRecv:
 			enqueue(&c->recvq, sg);
 			break;
-		
+
 		case CaseSend:
 			enqueue(&c->sendq, sg);
 			break;
diff --git a/src/pkg/runtime/hashmap.c b/src/pkg/runtime/hashmap.c
index 63ed4e2a37..ea9887a19f 100644
--- a/src/pkg/runtime/hashmap.c
+++ b/src/pkg/runtime/hashmap.c
@@ -115,7 +115,7 @@ hash_init (Hmap *h, int32 datasize, int64 hint)
 
 	if(datasize < sizeof (void *))
 		datasize = sizeof (void *);
-	datasize = runtime·rnd(datasize, sizeof (void *));
+	datasize = ROUND(datasize, sizeof (void *));
 	init_sizes (hint, &init_power);
 	h->datasize = datasize;
 	assert (h->datasize == datasize);
@@ -273,7 +273,7 @@ hash_lookup (MapType *t, Hmap *h, void *data, void **pres)
 	struct hash_entry *end_e;
 	void *key;
 	bool eq;
-	
+
 	hash = h->hash0;
 	(*t->key->alg->hash) (&hash, t->key->size, data);
 	hash &= ~HASH_MASK;
@@ -462,7 +462,7 @@ hash_insert (MapType *t, Hmap *h, void *data, void **pres)
 {
 	uintptr hash;
 	int32 rc;
-	
+
 	hash = h->hash0;
 	(*t->key->alg->hash) (&hash, t->key->size, data);
 	rc = hash_insert_internal (t, &h->st, 0, hash, h, data, pres);
@@ -618,7 +618,7 @@ hash_iter_init (MapType *t, Hmap *h, struct hash_iter *it)
 	it->subtable_state[0].e = h->st->entry;
 	it->subtable_state[0].start = h->st->entry;
 	it->subtable_state[0].last = h->st->last;
-	
+
 	// fastrand1 returns 31 useful bits.
 	// We don't care about not having a bottom bit but we
 	// do want top bits.
@@ -731,7 +731,7 @@ runtime·makemap_c(MapType *typ, int64 hint)
 	Hmap *h;
 	Type *key, *val;
 	uintptr ksize, vsize;
-	
+
 	key = typ->key;
 	val = typ->elem;
 
@@ -744,8 +744,8 @@ runtime·makemap_c(MapType *typ, int64 hint)
 	h = runtime·mal(sizeof(*h));
 	h->flag |= CanFreeTable;  /* until reflect gets involved, free is okay */
 
-	ksize = runtime·rnd(key->size, sizeof(void*));
-	vsize = runtime·rnd(val->size, sizeof(void*));
+	ksize = ROUND(key->size, sizeof(void*));
+	vsize = ROUND(val->size, sizeof(void*));
 	if(ksize > MaxData || vsize > MaxData || ksize+vsize > MaxData) {
 		// Either key is too big, or value is, or combined they are.
 		// Prefer to keep the key if possible, because we look at
@@ -829,7 +829,7 @@ runtime·mapaccess1(MapType *t, Hmap *h, ...)
 	bool pres;
 
 	ak = (byte*)(&h + 1);
-	av = ak + runtime·rnd(t->key->size, Structrnd);
+	av = ak + ROUND(t->key->size, Structrnd);
 
 	runtime·mapaccess(t, h, ak, av, &pres);
 
@@ -854,7 +854,7 @@ runtime·mapaccess2(MapType *t, Hmap *h, ...)
 	byte *ak, *av, *ap;
 
 	ak = (byte*)(&h + 1);
-	av = ak + runtime·rnd(t->key->size, Structrnd);
+	av = ak + ROUND(t->key->size, Structrnd);
 	ap = av + t->elem->size;
 
 	runtime·mapaccess(t, h, ak, av, ap);
@@ -952,7 +952,7 @@ runtime·mapassign1(MapType *t, Hmap *h, ...)
 		runtime·panicstring("assignment to entry in nil map");
 
 	ak = (byte*)(&h + 1);
-	av = ak + runtime·rnd(t->key->size, t->elem->align);
+	av = ak + ROUND(t->key->size, t->elem->align);
 
 	runtime·mapassign(t, h, ak, av);
 }
@@ -1171,7 +1171,7 @@ runtime·mapiter2(struct hash_iter *it, ...)
 
 	t = it->t;
 	ak = (byte*)(&it + 1);
-	av = ak + runtime·rnd(t->key->size, t->elem->align);
+	av = ak + ROUND(t->key->size, t->elem->align);
 
 	res = it->data;
 	if(res == nil)
diff --git a/src/pkg/runtime/iface.c b/src/pkg/runtime/iface.c
index 2b60c4f23a..b7eb2c18d1 100644
--- a/src/pkg/runtime/iface.c
+++ b/src/pkg/runtime/iface.c
@@ -193,7 +193,7 @@ runtime·convT2I(Type *t, InterfaceType *inter, ...)
 
 	elem = (byte*)(&inter+1);
 	wid = t->size;
-	ret = (Iface*)(elem + runtime·rnd(wid, Structrnd));
+	ret = (Iface*)(elem + ROUND(wid, Structrnd));
 	ret->tab = itab(inter, t, 0);
 	copyin(t, elem, &ret->data);
 }
@@ -209,7 +209,7 @@ runtime·convT2E(Type *t, ...)
 
 	elem = (byte*)(&t+1);
 	wid = t->size;
-	ret = (Eface*)(elem + runtime·rnd(wid, Structrnd));
+	ret = (Eface*)(elem + ROUND(wid, Structrnd));
 	ret->type = t;
 	copyin(t, elem, &ret->data);
 }
@@ -387,7 +387,7 @@ void
 runtime·convI2I(InterfaceType* inter, Iface i, Iface ret)
 {
 	Itab *tab;
-	
+
 	ret.data = i.data;
 	if((tab = i.tab) == nil)
 		ret.tab = nil;
@@ -694,7 +694,7 @@ reflect·unsafe_NewArray(Eface typ, uint32 n, void *ret)
 	// We know that the pointer to the original
 	// type structure sits before the data pointer.
 	t = (Type*)((Eface*)typ.data-1);
-	
+
 	size = n*t->size;
 	if(t->kind&KindNoPointers)
 		ret = runtime·mallocgc(size, FlagNoPointers, 1, 1);
diff --git a/src/pkg/runtime/print.c b/src/pkg/runtime/print.c
index 6702c3cde7..b41e28b37a 100644
--- a/src/pkg/runtime/print.c
+++ b/src/pkg/runtime/print.c
@@ -18,10 +18,10 @@ gwrite(void *v, int32 n)
 		runtime·write(2, v, n);
 		return;
 	}
-	
+
 	if(g->writenbuf == 0)
 		return;
-	
+
 	if(n > g->writenbuf)
 		n = g->writenbuf;
 	runtime·memmove(g->writebuf, v, n);
@@ -88,36 +88,36 @@ vprintf(int8 *s, byte *base)
 			break;
 		case 'd':	// 32-bit
 		case 'x':
-			arg = runtime·rnd(arg, 4);
+			arg = ROUND(arg, 4);
 			narg = arg + 4;
 			break;
 		case 'D':	// 64-bit
 		case 'U':
 		case 'X':
 		case 'f':
-			arg = runtime·rnd(arg, sizeof(uintptr));
+			arg = ROUND(arg, sizeof(uintptr));
 			narg = arg + 8;
 			break;
 		case 'C':
-			arg = runtime·rnd(arg, sizeof(uintptr));
+			arg = ROUND(arg, sizeof(uintptr));
 			narg = arg + 16;
 			break;
 		case 'p':	// pointer-sized
 		case 's':
-			arg = runtime·rnd(arg, sizeof(uintptr));
+			arg = ROUND(arg, sizeof(uintptr));
 			narg = arg + sizeof(uintptr);
 			break;
 		case 'S':	// pointer-aligned but bigger
-			arg = runtime·rnd(arg, sizeof(uintptr));
+			arg = ROUND(arg, sizeof(uintptr));
 			narg = arg + sizeof(String);
 			break;
 		case 'a':	// pointer-aligned but bigger
-			arg = runtime·rnd(arg, sizeof(uintptr));
+			arg = ROUND(arg, sizeof(uintptr));
 			narg = arg + sizeof(Slice);
 			break;
 		case 'i':	// pointer-aligned but bigger
 		case 'e':
-			arg = runtime·rnd(arg, sizeof(uintptr));
+			arg = ROUND(arg, sizeof(uintptr));
 			narg = arg + sizeof(Eface);
 			break;
 		}
diff --git a/src/pkg/runtime/runtime.c b/src/pkg/runtime/runtime.c
index 2cb3501dd1..cca061be79 100644
--- a/src/pkg/runtime/runtime.c
+++ b/src/pkg/runtime/runtime.c
@@ -156,19 +156,6 @@ runtime·mchr(byte *p, byte c, byte *ep)
 	return nil;
 }
 
-uint32
-runtime·rnd(uint32 n, uint32 m)
-{
-	uint32 r;
-
-	if(m > maxround)
-		m = maxround;
-	r = n % m;
-	if(r)
-		n += m-r;
-	return n;
-}
-
 static int32	argc;
 static uint8**	argv;
 
diff --git a/src/pkg/runtime/runtime.h b/src/pkg/runtime/runtime.h
index 1f7819197a..665d477f7d 100644
--- a/src/pkg/runtime/runtime.h
+++ b/src/pkg/runtime/runtime.h
@@ -390,6 +390,7 @@ struct ParFor
 #define	nelem(x)	(sizeof(x)/sizeof((x)[0]))
 #define	nil		((void*)0)
 #define	offsetof(s,m)	(uint32)(&(((s*)0)->m))
+#define	ROUND(x, n)	(((x)+(n)-1)&~((n)-1)) /* all-caps to mark as macro: it evaluates n twice */
 
 /*
  * known to compiler
@@ -533,7 +534,6 @@ void	runtime·goenvs_unix(void);
 void*	runtime·getu(void);
 void	runtime·throw(int8*);
 void	runtime·panicstring(int8*);
-uint32	runtime·rnd(uint32, uint32);
 void	runtime·prints(int8*);
 void	runtime·printf(int8*, ...);
 byte*	runtime·mchr(byte*, byte, byte*);