[dev.power64] gc: convert Bits to a uint64 array

So far all of our architectures have had at most 32 registers,
so we've been able to use entry 0 in the Bits uint32 array
directly as a register mask.  Power64 has 64 registers, so
this converts Bits to a uint64 array so we can continue to use
entry 0 directly as a register mask on Power64.

LGTM=rsc
R=rsc
CC=golang-codereviews
https://codereview.appspot.com/169060043

8 files changed, 119 insertions, 109 deletions

diff --git a/src/cmd/5g/opt.h b/src/cmd/5g/opt.h
index 1946c1d33..5016d1cc8 100644
--- a/src/cmd/5g/opt.h
+++ b/src/cmd/5g/opt.h
@@ -63,8 +63,8 @@ enum
 
 uint32 BLOAD(Reg*);
 uint32 BSTORE(Reg*);
-uint32 LOAD(Reg*);
-uint32 STORE(Reg*);
+uint64 LOAD(Reg*);
+uint64 STORE(Reg*);
 */
 
 // A Reg is a wrapper around a single Prog (one instruction) that holds
@@ -145,7 +145,7 @@ void	synch(Reg*, Bits);
 uint32	allreg(uint32, Rgn*);
 void	paint1(Reg*, int);
 uint32	paint2(Reg*, int);
-void	paint3(Reg*, int, int32, int);
+void	paint3(Reg*, int, uint32, int);
 void	addreg(Adr*, int);
 void	dumpit(char *str, Flow *r0, int);
 
@@ -156,10 +156,10 @@ void	peep(Prog*);
 void	excise(Flow*);
 int	copyu(Prog*, Adr*, Adr*);
 
-int32	RtoB(int);
-int32	FtoB(int);
-int	BtoR(int32);
-int	BtoF(int32);
+uint32	RtoB(int);
+uint32	FtoB(int);
+int	BtoR(uint32);
+int	BtoF(uint32);
 
 /*
  * prog.c
diff --git a/src/cmd/5g/reg.c b/src/cmd/5g/reg.c
index 8e49a2d9c..5b25adaf4 100644
--- a/src/cmd/5g/reg.c
+++ b/src/cmd/5g/reg.c
@@ -35,7 +35,7 @@
 #include "opt.h"
 
 #define	NREGVAR	32
-#define	REGBITS	((uint32)0xffffffff)
+#define	REGBITS	((uint64)0xffffffffull)
 /*c2go enum {
 	NREGVAR = 32,
 	REGBITS = 0xffffffff,
@@ -86,7 +86,7 @@ setaddrs(Bits bit)
 		i = bnum(bit);
 		node = var[i].node;
 		n = var[i].name;
-		bit.b[i/32] &= ~(1L<<(i%32));
+		biclr(&bit, i);
 
 		// disable all pieces of that variable
 		for(i=0; i<nvar; i++) {
@@ -425,7 +425,7 @@ loop2:
 			if(debug['R'] > 1)
 				print("\n");
 			paint1(r, i);
-			bit.b[i/32] &= ~(1L<<(i%32));
+			biclr(&bit, i);
 			if(change <= 0) {
 				if(debug['R'])
 					print("%L $%d: %Q\n",
@@ -570,7 +570,7 @@ walkvardef(Node *n, Reg *r, int active)
 			break;
 		for(v=n->opt; v!=nil; v=v->nextinnode) {
 			bn = v - var;
-			r1->act.b[bn/32] |= 1L << (bn%32);
+			biset(&r1->act, bn);
 		}
 		if(r1->f.prog->as == ABL)
 			break;
@@ -606,7 +606,7 @@ addsplits(void)
 					~(r->calahead.b[z] & addrs.b[z]);
 			while(bany(&bit)) {
 				i = bnum(bit);
-				bit.b[i/32] &= ~(1L << (i%32));
+				biclr(&bit, i);
 			}
 		}
 	}
@@ -972,10 +972,10 @@ prop(Reg *r, Bits ref, Bits cal)
 			for(z=0; z<BITS; z++) {
 				if(cal.b[z] == 0)
 					continue;
-				for(i=0; i<32; i++) {
-					if(z*32+i >= nvar || ((cal.b[z]>>i)&1) == 0)
+				for(i=0; i<64; i++) {
+					if(z*64+i >= nvar || ((cal.b[z]>>i)&1) == 0)
 						continue;
-					v = var+z*32+i;
+					v = var+z*64+i;
 					if(v->node->opt == nil) // v represents fixed register, not Go variable
 						continue;
 
@@ -991,10 +991,10 @@ prop(Reg *r, Bits ref, Bits cal)
 					// This will set the bits at most twice, keeping the overall loop linear.
 					v1 = v->node->opt;
 					j = v1 - var;
-					if(v == v1 || ((cal.b[j/32]>>(j&31))&1) == 0) {
+					if(v == v1 || !btest(&cal, j)) {
 						for(; v1 != nil; v1 = v1->nextinnode) {
 							j = v1 - var;
-							cal.b[j/32] |= 1<<(j&31);
+							biset(&cal, j);
 						}
 					}
 				}
@@ -1115,10 +1115,10 @@ paint1(Reg *r, int bn)
 	Reg *r1;
 	Prog *p;
 	int z;
-	uint32 bb;
+	uint64 bb;
 
-	z = bn/32;
-	bb = 1L<<(bn%32);
+	z = bn/64;
+	bb = 1LL<<(bn%64);
 	if(r->act.b[z] & bb)
 		return;
 	for(;;) {
@@ -1193,10 +1193,10 @@ paint2(Reg *r, int bn)
 {
 	Reg *r1;
 	int z;
-	uint32 bb, vreg;
+	uint64 bb, vreg;
 
-	z = bn/32;
-	bb = 1L << (bn%32);
+	z = bn/64;
+	bb = 1LL << (bn%64);
 	vreg = regbits;
 	if(!(r->act.b[z] & bb))
 		return vreg;
@@ -1240,15 +1240,15 @@ paint2(Reg *r, int bn)
 }
 
 void
-paint3(Reg *r, int bn, int32 rb, int rn)
+paint3(Reg *r, int bn, uint32 rb, int rn)
 {
 	Reg *r1;
 	Prog *p;
 	int z;
-	uint32 bb;
+	uint64 bb;
 
-	z = bn/32;
-	bb = 1L << (bn%32);
+	z = bn/64;
+	bb = 1LL << (bn%64);
 	if(r->act.b[z] & bb)
 		return;
 	for(;;) {
@@ -1333,7 +1333,7 @@ addreg(Adr *a, int rn)
  *	10	R10
  *	12  R12
  */
-int32
+uint32
 RtoB(int r)
 {
 	if(r >= REGTMP-2 && r != 12)	// excluded R9 and R10 for m and g, but not R12
@@ -1342,7 +1342,7 @@ RtoB(int r)
 }
 
 int
-BtoR(int32 b)
+BtoR(uint32 b)
 {
 	b &= 0x11fcL;	// excluded R9 and R10 for m and g, but not R12
 	if(b == 0)
@@ -1357,7 +1357,7 @@ BtoR(int32 b)
  *	...	...
  *	31	F15
  */
-int32
+uint32
 FtoB(int f)
 {
 
@@ -1367,7 +1367,7 @@ FtoB(int f)
 }
 
 int
-BtoF(int32 b)
+BtoF(uint32 b)
 {
 
 	b &= 0xfffc0000L;
diff --git a/src/cmd/6g/opt.h b/src/cmd/6g/opt.h
index dbd039d89..4c9bb89fc 100644
--- a/src/cmd/6g/opt.h
+++ b/src/cmd/6g/opt.h
@@ -63,8 +63,8 @@ enum
 
 uint32 BLOAD(Reg*);
 uint32 BSTORE(Reg*);
-uint32 LOAD(Reg*);
-uint32 STORE(Reg*);
+uint64 LOAD(Reg*);
+uint64 STORE(Reg*);
 */
 
 // A Reg is a wrapper around a single Prog (one instruction) that holds
@@ -141,7 +141,7 @@ void	synch(Reg*, Bits);
 uint32	allreg(uint32, Rgn*);
 void	paint1(Reg*, int);
 uint32	paint2(Reg*, int);
-void	paint3(Reg*, int, int32, int);
+void	paint3(Reg*, int, uint32, int);
 void	addreg(Adr*, int);
 void	dumpone(Flow*, int);
 void	dumpit(char*, Flow*, int);
@@ -153,10 +153,10 @@ void	peep(Prog*);
 void	excise(Flow*);
 int	copyu(Prog*, Adr*, Adr*);
 
-int32	RtoB(int);
-int32	FtoB(int);
-int	BtoR(int32);
-int	BtoF(int32);
+uint32	RtoB(int);
+uint32	FtoB(int);
+int	BtoR(uint32);
+int	BtoF(uint32);
 
 /*
  * prog.c
diff --git a/src/cmd/6g/reg.c b/src/cmd/6g/reg.c
index 1f757e197..8d600d643 100644
--- a/src/cmd/6g/reg.c
+++ b/src/cmd/6g/reg.c
@@ -34,7 +34,7 @@
 #include "opt.h"
 
 #define	NREGVAR	32	/* 16 general + 16 floating */
-#define	REGBITS	((uint32)0xffffffff)
+#define	REGBITS	((uint64)0xffffffffull)
 /*c2go enum {
 	NREGVAR = 32,
 	REGBITS = 0xffffffff,
@@ -71,7 +71,7 @@ setaddrs(Bits bit)
 		i = bnum(bit);
 		node = var[i].node;
 		n = var[i].name;
-		bit.b[i/32] &= ~(1L<<(i%32));
+		biclr(&bit, i);
 
 		// disable all pieces of that variable
 		for(i=0; i<nvar; i++) {
@@ -364,7 +364,7 @@ loop2:
 			rgp->varno = i;
 			change = 0;
 			paint1(r, i);
-			bit.b[i/32] &= ~(1L<<(i%32));
+			biclr(&bit, i);
 			if(change <= 0)
 				continue;
 			rgp->cost = change;
@@ -477,7 +477,7 @@ walkvardef(Node *n, Reg *r, int active)
 			break;
 		for(v=n->opt; v!=nil; v=v->nextinnode) {
 			bn = v - var;
-			r1->act.b[bn/32] |= 1L << (bn%32);
+			biset(&r1->act, bn);
 		}
 		if(r1->f.prog->as == ACALL)
 			break;
@@ -822,10 +822,10 @@ prop(Reg *r, Bits ref, Bits cal)
 			for(z=0; z<BITS; z++) {
 				if(cal.b[z] == 0)
 					continue;
-				for(i=0; i<32; i++) {
-					if(z*32+i >= nvar || ((cal.b[z]>>i)&1) == 0)
+				for(i=0; i<64; i++) {
+					if(z*64+i >= nvar || ((cal.b[z]>>i)&1) == 0)
 						continue;
-					v = var+z*32+i;
+					v = var+z*64+i;
 					if(v->node->opt == nil) // v represents fixed register, not Go variable
 						continue;
 
@@ -841,10 +841,10 @@ prop(Reg *r, Bits ref, Bits cal)
 					// This will set the bits at most twice, keeping the overall loop linear.
 					v1 = v->node->opt;
 					j = v1 - var;
-					if(v == v1 || ((cal.b[j/32]>>(j&31))&1) == 0) {
+					if(v == v1 || !btest(&cal, j)) {
 						for(; v1 != nil; v1 = v1->nextinnode) {
 							j = v1 - var;
-							cal.b[j/32] |= 1UL<<(j&31);
+							biset(&cal, j);
 						}
 					}
 				}
@@ -959,10 +959,10 @@ paint1(Reg *r, int bn)
 {
 	Reg *r1;
 	int z;
-	uint32 bb;
+	uint64 bb;
 
-	z = bn/32;
-	bb = 1L<<(bn%32);
+	z = bn/64;
+	bb = 1LL<<(bn%64);
 	if(r->act.b[z] & bb)
 		return;
 	for(;;) {
@@ -1061,10 +1061,10 @@ paint2(Reg *r, int bn)
 {
 	Reg *r1;
 	int z;
-	uint32 bb, vreg, x;
+	uint64 bb, vreg, x;
 
-	z = bn/32;
-	bb = 1L << (bn%32);
+	z = bn/64;
+	bb = 1LL << (bn%64);
 	vreg = regbits;
 	if(!(r->act.b[z] & bb))
 		return vreg;
@@ -1117,15 +1117,15 @@ paint2(Reg *r, int bn)
 }
 
 void
-paint3(Reg *r, int bn, int32 rb, int rn)
+paint3(Reg *r, int bn, uint32 rb, int rn)
 {
 	Reg *r1;
 	Prog *p;
 	int z;
-	uint32 bb;
+	uint64 bb;
 
-	z = bn/32;
-	bb = 1L << (bn%32);
+	z = bn/64;
+	bb = 1LL << (bn%64);
 	if(r->act.b[z] & bb)
 		return;
 	for(;;) {
@@ -1198,7 +1198,7 @@ addreg(Adr *a, int rn)
 	ostats.ncvtreg++;
 }
 
-int32
+uint32
 RtoB(int r)
 {
 
@@ -1208,7 +1208,7 @@ RtoB(int r)
 }
 
 int
-BtoR(int32 b)
+BtoR(uint32 b)
 {
 	b &= 0xffffL;
 	if(nacl)
@@ -1224,7 +1224,7 @@ BtoR(int32 b)
  *	...
  *	31	X15
  */
-int32
+uint32
 FtoB(int f)
 {
 	if(f < D_X0 || f > D_X15)
@@ -1233,7 +1233,7 @@ FtoB(int f)
 }
 
 int
-BtoF(int32 b)
+BtoF(uint32 b)
 {
 
 	b &= 0xFFFF0000L;
diff --git a/src/cmd/8g/opt.h b/src/cmd/8g/opt.h
index 09f58c40a..0e2d165b1 100644
--- a/src/cmd/8g/opt.h
+++ b/src/cmd/8g/opt.h
@@ -63,8 +63,8 @@ enum
 
 uint32 BLOAD(Reg*);
 uint32 BSTORE(Reg*);
-uint32 LOAD(Reg*);
-uint32 STORE(Reg*);
+uint64 LOAD(Reg*);
+uint64 STORE(Reg*);
 */
 
 // A Reg is a wrapper around a single Prog (one instruction) that holds
@@ -159,7 +159,7 @@ void	synch(Reg*, Bits);
 uint32	allreg(uint32, Rgn*);
 void	paint1(Reg*, int);
 uint32	paint2(Reg*, int);
-void	paint3(Reg*, int, int32, int);
+void	paint3(Reg*, int, uint32, int);
 void	addreg(Adr*, int);
 void	dumpone(Flow*, int);
 void	dumpit(char*, Flow*, int);
@@ -171,10 +171,10 @@ void	peep(Prog*);
 void	excise(Flow*);
 int	copyu(Prog*, Adr*, Adr*);
 
-int32	RtoB(int);
-int32	FtoB(int);
-int	BtoR(int32);
-int	BtoF(int32);
+uint32	RtoB(int);
+uint32	FtoB(int);
+int	BtoR(uint32);
+int	BtoF(uint32);
 
 /*
  * prog.c
diff --git a/src/cmd/8g/reg.c b/src/cmd/8g/reg.c
index 302b273a1..0fbe68482 100644
--- a/src/cmd/8g/reg.c
+++ b/src/cmd/8g/reg.c
@@ -34,7 +34,7 @@
 #include "opt.h"
 
 #define	NREGVAR	16	/* 8 integer + 8 floating */
-#define	REGBITS	((uint32)0xffff)
+#define	REGBITS	((uint64)0xffffull)
 /*c2go enum {
 	NREGVAR = 16,
 	REGBITS = (1<<NREGVAR) - 1,
@@ -71,7 +71,7 @@ setaddrs(Bits bit)
 		i = bnum(bit);
 		node = var[i].node;
 		n = var[i].name;
-		bit.b[i/32] &= ~(1L<<(i%32));
+		biclr(&bit, i);
 
 		// disable all pieces of that variable
 		for(i=0; i<nvar; i++) {
@@ -336,7 +336,7 @@ loop2:
 			rgp->varno = i;
 			change = 0;
 			paint1(r, i);
-			bit.b[i/32] &= ~(1L<<(i%32));
+			biclr(&bit, i);
 			if(change <= 0)
 				continue;
 			rgp->cost = change;
@@ -446,7 +446,7 @@ walkvardef(Node *n, Reg *r, int active)
 			break;
 		for(v=n->opt; v!=nil; v=v->nextinnode) {
 			bn = v - var;
-			r1->act.b[bn/32] |= 1L << (bn%32);
+			biset(&r1->act, bn);
 		}
 		if(r1->f.prog->as == ACALL)
 			break;
@@ -788,10 +788,10 @@ prop(Reg *r, Bits ref, Bits cal)
 			for(z=0; z<BITS; z++) {
 				if(cal.b[z] == 0)
 					continue;
-				for(i=0; i<32; i++) {
-					if(z*32+i >= nvar || ((cal.b[z]>>i)&1) == 0)
+				for(i=0; i<64; i++) {
+					if(z*64+i >= nvar || ((cal.b[z]>>i)&1) == 0)
 						continue;
-					v = var+z*32+i;
+					v = var+z*64+i;
 					if(v->node->opt == nil) // v represents fixed register, not Go variable
 						continue;
 
@@ -807,10 +807,10 @@ prop(Reg *r, Bits ref, Bits cal)
 					// This will set the bits at most twice, keeping the overall loop linear.
 					v1 = v->node->opt;
 					j = v1 - var;
-					if(v == v1 || ((cal.b[j/32]>>(j&31))&1) == 0) {
+					if(v == v1 || !btest(&cal, j)) {
 						for(; v1 != nil; v1 = v1->nextinnode) {
 							j = v1 - var;
-							cal.b[j/32] |= 1<<(j&31);
+							biset(&cal, j);
 						}
 					}
 				}
@@ -926,10 +926,10 @@ paint1(Reg *r, int bn)
 	Reg *r1;
 	Prog *p;
 	int z;
-	uint32 bb;
+	uint64 bb;
 
-	z = bn/32;
-	bb = 1L<<(bn%32);
+	z = bn/64;
+	bb = 1LL<<(bn%64);
 	if(r->act.b[z] & bb)
 		return;
 	for(;;) {
@@ -1038,10 +1038,10 @@ paint2(Reg *r, int bn)
 {
 	Reg *r1;
 	int z;
-	uint32 bb, vreg, x;
+	uint64 bb, vreg, x;
 
-	z = bn/32;
-	bb = 1L << (bn%32);
+	z = bn/64;
+	bb = 1LL << (bn%64);
 	vreg = regbits;
 	if(!(r->act.b[z] & bb))
 		return vreg;
@@ -1094,15 +1094,15 @@ paint2(Reg *r, int bn)
 }
 
 void
-paint3(Reg *r, int bn, int32 rb, int rn)
+paint3(Reg *r, int bn, uint32 rb, int rn)
 {
 	Reg *r1;
 	Prog *p;
 	int z;
-	uint32 bb;
+	uint64 bb;
 
-	z = bn/32;
-	bb = 1L << (bn%32);
+	z = bn/64;
+	bb = 1LL << (bn%64);
 	if(r->act.b[z] & bb)
 		return;
 	for(;;) {
@@ -1175,7 +1175,7 @@ addreg(Adr *a, int rn)
 	ostats.ncvtreg++;
 }
 
-int32
+uint32
 RtoB(int r)
 {
 
@@ -1185,7 +1185,7 @@ RtoB(int r)
 }
 
 int
-BtoR(int32 b)
+BtoR(uint32 b)
 {
 
 	b &= 0xffL;
@@ -1194,7 +1194,7 @@ BtoR(int32 b)
 	return bitno(b) + D_AX;
 }
 
-int32
+uint32
 FtoB(int f)
 {
 	if(f < D_X0 || f > D_X7)
@@ -1203,7 +1203,7 @@ FtoB(int f)
 }
 
 int
-BtoF(int32 b)
+BtoF(uint32 b)
 {
 	b &= 0xFF00L;
 	if(b == 0)
diff --git a/src/cmd/gc/bits.c b/src/cmd/gc/bits.c
index 2e79f6f1d..fe9a168dc 100644
--- a/src/cmd/gc/bits.c
+++ b/src/cmd/gc/bits.c
@@ -95,11 +95,11 @@ int
 bnum(Bits a)
 {
 	int i;
-	int32 b;
+	uint64 b;
 
 	for(i=0; i<BITS; i++)
 		if(b = a.b[i])
-			return 32*i + bitno(b);
+			return 64*i + bitno(b);
 	fatal("bad in bnum");
 	return 0;
 }
@@ -110,27 +110,35 @@ blsh(uint n)
 	Bits c;
 
 	c = zbits;
-	c.b[n/32] = 1L << (n%32);
+	c.b[n/64] = 1LL << (n%64);
 	return c;
 }
 
-/*
 int
-bset(Bits a, uint n)
+btest(Bits *a, uint n)
 {
-	if(a.b[n/32] & (1L << (n%32)))
-		return 1;
-	return 0;
+	return (a->b[n/64] & (1LL << (n%64))) != 0;
+}
+
+void
+biset(Bits *a, uint n)
+{
+	a->b[n/64] |= 1LL << (n%64);
+}
+
+void
+biclr(Bits *a, uint n)
+{
+	a->b[n/64] &= ~(1LL << (n%64));
 }
-*/
 
 int
-bitno(int32 b)
+bitno(uint64 b)
 {
 	int i;
 
-	for(i=0; i<32; i++)
-		if(b & (1L<<i))
+	for(i=0; i<64; i++)
+		if(b & (1LL<<i))
 			return i;
 	fatal("bad in bitno");
 	return 0;
@@ -157,7 +165,7 @@ Qconv(Fmt *fp)
 			if(var[i].offset != 0)
 				fmtprint(fp, "%+lld", (vlong)var[i].offset);
 		}
-		bits.b[i/32] &= ~(1L << (i%32));
+		biclr(&bits, i);
 	}
 	return 0;
 }
diff --git a/src/cmd/gc/go.h b/src/cmd/gc/go.h
index 965a0550d..d3c4193b5 100644
--- a/src/cmd/gc/go.h
+++ b/src/cmd/gc/go.h
@@ -704,13 +704,13 @@ enum
 	Ecomplit = 1<<11,	// type in composite literal
 };
 
-#define	BITS	5
-#define	NVAR	(BITS*sizeof(uint32)*8)
+#define	BITS	3
+#define	NVAR	(BITS*sizeof(uint64)*8)
 
 typedef	struct	Bits	Bits;
 struct	Bits
 {
-	uint32	b[BITS];
+	uint64	b[BITS];
 };
 
 EXTERN	Bits	zbits;
@@ -1027,12 +1027,14 @@ int	Qconv(Fmt *fp);
 Bits	band(Bits a, Bits b);
 int	bany(Bits *a);
 int	beq(Bits a, Bits b);
-int	bitno(int32 b);
+int	bitno(uint64 b);
 Bits	blsh(uint n);
 Bits	bnot(Bits a);
 int	bnum(Bits a);
 Bits	bor(Bits a, Bits b);
-int	bset(Bits a, uint n);
+int	btest(Bits *a, uint n);
+void	biset(Bits *a, uint n);
+void	biclr(Bits *a, uint n);
 
 /*
  *	bv.c