From 3b4d792606bf6de962bed73e79c261d7ddd7266c Mon Sep 17 00:00:00 2001
From: Russ Cox <rsc@golang.org>
Date: Fri, 16 Aug 2013 21:45:59 -0400
Subject: [PATCH] cmd/gc: separate "has pointers" from "needs zeroing" in stack
 frame

When the new call site-specific frame bitmaps are available,
we can cut the zeroing to just those values that need it due
to scope escaping.

R=cshapiro, cshapiro
CC=golang-dev
https://golang.org/cl/13045043
---
 src/cmd/5g/ggen.c | 10 +++++-----
 src/cmd/6g/ggen.c | 10 +++++-----
 src/cmd/8g/ggen.c | 10 +++++-----
 src/cmd/gc/go.h   |  4 +++-
 src/cmd/gc/pgen.c | 38 +++++++++++++++++++++++++-------------
 5 files changed, 43 insertions(+), 29 deletions(-)

diff --git a/src/cmd/5g/ggen.c b/src/cmd/5g/ggen.c
index 52c2cd4395..f018c88aa8 100644
--- a/src/cmd/5g/ggen.c
+++ b/src/cmd/5g/ggen.c
@@ -35,11 +35,11 @@ defframe(Prog *ptxt, Bvec *bv)
 	p = ptxt;
 	while(p->link->as == AFUNCDATA || p->link->as == APCDATA || p->link->as == ATYPE)
 		p = p->link;
-	if(stkptrsize >= 8*widthptr) {
+	if(stkzerosize >= 8*widthptr) {
 		p = appendp(p, AMOVW, D_CONST, NREG, 0, D_REG, 0, 0);
-		p = appendp(p, AADD, D_CONST, NREG, 4+frame-stkptrsize, D_REG, 1, 0);
+		p = appendp(p, AADD, D_CONST, NREG, 4+frame-stkzerosize, D_REG, 1, 0);
 		p->reg = REGSP;
-		p = appendp(p, AADD, D_CONST, NREG, stkptrsize, D_REG, 2, 0);
+		p = appendp(p, AADD, D_CONST, NREG, stkzerosize, D_REG, 2, 0);
 		p->reg = 1;
 		p1 = p = appendp(p, AMOVW, D_REG, 0, 0, D_OREG, 1, 4);
 		p->scond |= C_PBIT;
@@ -49,13 +49,13 @@ defframe(Prog *ptxt, Bvec *bv)
 		patch(p, p1);
 	} else {
 		first = 1;
-		for(i=0, j=0; i<stkptrsize; i+=widthptr, j+=2) {
+		for(i=0, j=(stkptrsize-stkzerosize)/widthptr*2; i<stkzerosize; i+=widthptr, j+=2) {
 			if(bvget(bv, j) || bvget(bv, j+1)) {
 				if(first) {
 					p = appendp(p, AMOVW, D_CONST, NREG, 0, D_REG, 0, 0);
 					first = 0;
 				}
-				p = appendp(p, AMOVW, D_REG, 0, 0, D_OREG, REGSP, 4+frame-stkptrsize+i);
+				p = appendp(p, AMOVW, D_REG, 0, 0, D_OREG, REGSP, 4+frame-stkzerosize+i);
 			}
 		}
 	}
diff --git a/src/cmd/6g/ggen.c b/src/cmd/6g/ggen.c
index ec1d1ddc4c..6e2f0fd69e 100644
--- a/src/cmd/6g/ggen.c
+++ b/src/cmd/6g/ggen.c
@@ -30,16 +30,16 @@ defframe(Prog *ptxt, Bvec *bv)
 	// so that garbage collector only sees initialized values
 	// when it looks for pointers.
 	p = ptxt;
-	if(stkptrsize >= 8*widthptr) {
+	if(stkzerosize >= 8*widthptr) {
 		p = appendp(p, AMOVQ, D_CONST, 0, D_AX, 0);
-		p = appendp(p, AMOVQ, D_CONST, stkptrsize/widthptr, D_CX, 0);
-		p = appendp(p, ALEAQ, D_SP+D_INDIR, frame-stkptrsize, D_DI, 0);
+		p = appendp(p, AMOVQ, D_CONST, stkzerosize/widthptr, D_CX, 0);
+		p = appendp(p, ALEAQ, D_SP+D_INDIR, frame-stkzerosize, D_DI, 0);
 		p = appendp(p, AREP, D_NONE, 0, D_NONE, 0);
 		appendp(p, ASTOSQ, D_NONE, 0, D_NONE, 0);
 	} else {
-		for(i=0, j=0; i<stkptrsize; i+=widthptr, j+=2)
+		for(i=0, j=(stkptrsize-stkzerosize)/widthptr*2; i<stkzerosize; i+=widthptr, j+=2)
 			if(bvget(bv, j) || bvget(bv, j+1))
-				p = appendp(p, AMOVQ, D_CONST, 0, D_SP+D_INDIR, frame-stkptrsize+i);
+				p = appendp(p, AMOVQ, D_CONST, 0, D_SP+D_INDIR, frame-stkzerosize+i);
 	}
 }
 
diff --git a/src/cmd/8g/ggen.c b/src/cmd/8g/ggen.c
index 6b8c67e5bc..1f9e6e0593 100644
--- a/src/cmd/8g/ggen.c
+++ b/src/cmd/8g/ggen.c
@@ -32,16 +32,16 @@ defframe(Prog *ptxt, Bvec *bv)
 	// so that garbage collector only sees initialized values
 	// when it looks for pointers.
 	p = ptxt;
-	if(stkptrsize >= 8*widthptr) {
+	if(stkzerosize >= 8*widthptr) {
 		p = appendp(p, AMOVL, D_CONST, 0, D_AX, 0);
-		p = appendp(p, AMOVL, D_CONST, stkptrsize/widthptr, D_CX, 0);
-		p = appendp(p, ALEAL, D_SP+D_INDIR, frame-stkptrsize, D_DI, 0);
+		p = appendp(p, AMOVL, D_CONST, stkzerosize/widthptr, D_CX, 0);
+		p = appendp(p, ALEAL, D_SP+D_INDIR, frame-stkzerosize, D_DI, 0);
 		p = appendp(p, AREP, D_NONE, 0, D_NONE, 0);
 		appendp(p, ASTOSL, D_NONE, 0, D_NONE, 0);
 	} else {
-		for(i=0, j=0; i<stkptrsize; i+=widthptr, j+=2)
+		for(i=0, j=(stkptrsize-stkzerosize)/widthptr*2; i<stkzerosize; i+=widthptr, j+=2)
 			if(bvget(bv, j) || bvget(bv, j+1))
-				p = appendp(p, AMOVL, D_CONST, 0, D_SP+D_INDIR, frame-stkptrsize+i);
+				p = appendp(p, AMOVL, D_CONST, 0, D_SP+D_INDIR, frame-stkzerosize+i);
 	}
 }
 
diff --git a/src/cmd/gc/go.h b/src/cmd/gc/go.h
index c021e895ef..9232650c28 100644
--- a/src/cmd/gc/go.h
+++ b/src/cmd/gc/go.h
@@ -270,6 +270,7 @@ struct	Node
 	uchar	dupok;	// duplicate definitions ok (for func)
 	schar	likely; // likeliness of if statement
 	uchar	hasbreak;	// has break statement
+	uchar	needzero; // if it contains pointers, needs to be zeroed on function entry
 	uint	esc;		// EscXXX
 	int	funcdepth;
 
@@ -940,7 +941,8 @@ EXTERN	NodeList*	lastconst;
 EXTERN	Node*	lasttype;
 EXTERN	vlong	maxarg;
 EXTERN	vlong	stksize;		// stack size for current frame
-EXTERN	vlong	stkptrsize;		// prefix of stack containing pointers for current frame
+EXTERN	vlong	stkptrsize;		// prefix of stack containing pointers
+EXTERN	vlong	stkzerosize;		// prefix of stack that must be zeroed on entry
 EXTERN	int32	blockgen;		// max block number
 EXTERN	int32	block;			// current block number
 EXTERN	int	hasdefer;		// flag that curfn has defer statetment
diff --git a/src/cmd/gc/pgen.c b/src/cmd/gc/pgen.c
index edeaa06a69..f391634067 100644
--- a/src/cmd/gc/pgen.c
+++ b/src/cmd/gc/pgen.c
@@ -367,11 +367,11 @@ dumpgclocals(Node* fn, Sym *sym)
 
 // Sort the list of stack variables. Autos after anything else,
 // within autos, unused after used, within used, things with
-// pointers first, and then decreasing size.
+// pointers first, zeroed things first, and then decreasing size.
 // Because autos are laid out in decreasing addresses
-// on the stack, pointers first and decreasing size
-// really means, in memory, pointers near the top of the 
-// stack and increasing in size.
+// on the stack, pointers first, zeroed things first and decreasing size
+// really means, in memory, things with pointers needing zeroing at
+// the top of the stack and increasing in size.
 // Non-autos sort on offset.
 static int
 cmpstackvar(Node *a, Node *b)
@@ -394,6 +394,12 @@ cmpstackvar(Node *a, Node *b)
 	bp = haspointers(b->type);
 	if(ap != bp)
 		return bp - ap;
+
+	ap = a->needzero;
+	bp = b->needzero;
+	if(ap != bp)
+		return bp - ap;
+
 	if(a->type->width < b->type->width)
 		return +1;
 	if(a->type->width > b->type->width)
@@ -409,11 +415,12 @@ allocauto(Prog* ptxt)
 	Node* n;
 	vlong w;
 
-	if(curfn->dcl == nil) {
-		stksize = 0;
-		stkptrsize = 0;
+	stksize = 0;
+	stkptrsize = 0;
+	stkzerosize = 0;
+
+	if(curfn->dcl == nil)
 		return;
-	}
 
 	// Mark the PAUTO's unused.
 	for(ll=curfn->dcl; ll != nil; ll=ll->next)
@@ -421,6 +428,11 @@ allocauto(Prog* ptxt)
 			ll->n->used = 0;
 
 	markautoused(ptxt);
+	
+	// TODO: Remove when liveness analysis sets needzero instead.
+	for(ll=curfn->dcl; ll != nil; ll=ll->next)
+		if (ll->n->class == PAUTO)
+			ll->n->needzero = 1; // ll->n->addrtaken;
 
 	listsort(&curfn->dcl, cmpstackvar);
 
@@ -430,8 +442,6 @@ allocauto(Prog* ptxt)
 	if (n->class == PAUTO && n->op == ONAME && !n->used) {
 		// No locals used at all
 		curfn->dcl = nil;
-		stksize = 0;
-		stkptrsize = 0;
 		fixautoused(ptxt);
 		return;
 	}
@@ -446,8 +456,6 @@ allocauto(Prog* ptxt)
 	}
 
 	// Reassign stack offsets of the locals that are still there.
-	stksize = 0;
-	stkptrsize = 0;
 	for(ll = curfn->dcl; ll != nil; ll=ll->next) {
 		n = ll->n;
 		if (n->class != PAUTO || n->op != ONAME)
@@ -459,8 +467,11 @@ allocauto(Prog* ptxt)
 			fatal("bad width");
 		stksize += w;
 		stksize = rnd(stksize, n->type->align);
-		if(haspointers(n->type))
+		if(haspointers(n->type)) {
 			stkptrsize = stksize;
+			if(n->needzero)
+				stkzerosize = stksize;
+		}
 		if(thechar == '5')
 			stksize = rnd(stksize, widthptr);
 		if(stksize >= (1ULL<<31)) {
@@ -471,6 +482,7 @@ allocauto(Prog* ptxt)
 	}
 	stksize = rnd(stksize, widthptr);
 	stkptrsize = rnd(stkptrsize, widthptr);
+	stkzerosize = rnd(stkzerosize, widthptr);
 
 	fixautoused(ptxt);