libc: update atomic ops and fp code for arm (from sources)

sys/src/libc/arm/atom.s

@@ -1,44 +1,47 @@
+#define	CLREX		WORD	$0xf57ff01f
+#define	LDREX(a,r)	WORD	$(0xe<<28|0x01900f9f | (a)<<16 | (r)<<12)
+/* `The order of operands is from left to right in dataflow order' - asm man */
+#define	STREX(v,a,r)	WORD	$(0xe<<28|0x01800f90 | (a)<<16 | (r)<<12 | (v)<<0)
 
 /*
  * int cas(ulong *p, ulong ov, ulong nv);
  */
 
-#define	LDREX(a,r)	WORD	$(0xe<<28|0x01900f9f | (a)<<16 | (r)<<12)
+TEXT	cas+0(SB),0,$0		/* r0 holds p */
-#define	STREX(a,v,r)	WORD	$(0xe<<28|0x01800f90 | (a)<<16 | (r)<<12 | (v)<<0)
+TEXT	casp+0(SB),0,$0		/* r0 holds p */
-
-TEXT	cas+0(SB),0,$12		/* r0 holds p */
-TEXT	casp+0(SB),0,$12	/* r0 holds p */
-TEXT	casl+0(SB),0,$12	/* r0 holds p */
 	MOVW	ov+4(FP), R1
 	MOVW	nv+8(FP), R2
 spincas:
 	LDREX(0,3)	/*	LDREX	0(R0),R3	*/
 	CMP.S	R3, R1
 	BNE	fail
-	STREX(0,2,4)	/*	STREX	0(R0),R2,R4	*/
+	STREX(2,0,4)	/*	STREX	0(R0),R2,R4	*/
 	CMP.S	$0, R4
 	BNE	spincas
 	MOVW	$1, R0
 	RET
 fail:
+	CLREX
 	MOVW	$0, R0
 	RET
 
+TEXT _xinc(SB), $0	/* void	_xinc(long *); */
 TEXT ainc(SB), $0	/* long ainc(long *); */
 spinainc:
 	LDREX(0,3)	/*	LDREX	0(R0),R3	*/
 	ADD	$1,R3
-	STREX(0,3,4)	/*	STREX	0(R0),R2,R4	*/
+	STREX(3,0,4)	/*	STREX	0(R0),R3,R4	*/
 	CMP.S	$0, R4
 	BNE	spinainc
 	MOVW	R3, R0
 	RET
 
-TEXT adec(SB), $0	/* long ainc(long *); */
+TEXT _xdec(SB), $0	/* long _xdec(long *); */
+TEXT adec(SB), $0	/* long adec(long *); */
 spinadec:
 	LDREX(0,3)	/*	LDREX	0(R0),R3	*/
 	SUB	$1,R3
-	STREX(0,3,4)	/*	STREX	0(R0),R3,R4	*/
+	STREX(3,0,4)	/*	STREX	0(R0),R3,R4	*/
 	CMP.S	$0, R4
 	BNE	spinadec
 	MOVW	R3, R0
@@ -50,6 +53,6 @@ TEXT loadlinked(SB), $0	/* long loadlinked(long *); */
 
 TEXT storecond(SB), $0	/* int storecond(long *, long); */
 	MOVW	ov+4(FP), R3
-	STREX(0,3,4)	/*	STREX	0(R0),R3,R0	*/
+	STREX(3,0,0)	/*	STREX	0(R0),R3,R0	*/
 	RSB	$1, R0
 	RET

sys/src/libc/arm/cycles.c

@@ -1,7 +1,10 @@
 #include <u.h>
 #include <libc.h>
 
-void	cycles(uvlong*u)
+#pragma profile off
+
+void
+cycles(uvlong*u)
 {
 	*u = 0LL;
 }

sys/src/libc/arm/getcallerpc.s

@@ -1,3 +1,3 @@
-TEXT getcallerpc(SB), $-4
+TEXT getcallerpc(SB), 1, $-4
 	MOVW	0(R13), R0
 	RET

sys/src/libc/arm/getfcr.vfp.S

@@ -0,0 +1,21 @@
+/* for VFP */
+#define VMRS(fp, cpu) WORD $(0xeef00a10 | (fp)<<16 | (cpu)<<12) /* FP → arm */
+#define VMSR(cpu, fp) WORD $(0xeee00a10 | (fp)<<16 | (cpu)<<12) /* arm → FP */
+
+#define Fpscr 1
+
+TEXT	setfcr(SB), $0
+	VMSR(0, Fpscr)
+	RET
+
+TEXT	getfcr(SB), $0
+	VMRS(Fpscr, 0)
+	RET
+
+TEXT	getfsr(SB), $0
+	VMSR(0, Fpscr)
+	RET
+
+TEXT	setfsr(SB), $0
+	VMRS(Fpscr, 0)
+	RET

sys/src/libc/arm/main9p.s

@@ -34,8 +34,8 @@ loop:
 TEXT	_savearg(SB), 1, $0
 	RET
 
-TEXT	_callpc(SB), 1, $0
+TEXT	_callpc(SB), 1, $-4
-	MOVW	argp-4(FP), R(arg)
+	MOVW	0(R13), R(arg)
 	RET
 
 DATA	_exitstr<>+0(SB)/4, $"main"

sys/src/libc/arm/memmove.s

@@ -5,9 +5,9 @@ N = 3
 TMP = 3					/* N and TMP don't overlap */
 TMP1 = 4
 
-TEXT memcpy(SB), $-4
+TEXT memcpy(SB), $0
 	B	_memmove
-TEXT memmove(SB), $-4
+TEXT memmove(SB), $0
 _memmove:
 	MOVW	R(TS), to+0(FP)		/* need to save for return value */
 	MOVW	from+4(FP), R(FROM)

sys/src/libc/arm/strchr.s

@@ -1,4 +1,4 @@
-TEXT strchr(SB), $-4
+TEXT strchr(SB), $0
 	MOVBU	c+4(FP), R1
 	CMP	$0, R1
 	BEQ	_null

sys/src/libc/arm/strcmp.s

@@ -1,4 +1,4 @@
-TEXT strcmp(SB), $-4
+TEXT strcmp(SB), $0
 	MOVW	R0, R1
 	MOVW	s2+4(FP), R2
 

sys/src/libc/arm/strcpy.s

@@ -1,4 +1,4 @@
-TEXT strcpy(SB), $-4
+TEXT strcpy(SB), $0
 	MOVW		R0, to+0(FP)	/* need to save for return value */
 	MOVW		from+4(FP), R1
 	MOVW		$0xFF, R2	/* mask */

sys/src/libc/arm/tas.s

@@ -1,4 +1,4 @@
-TEXT	_tas(SB), $-4
+TEXT	_tas(SB), 1, $-4
 	MOVW	R0,R1
 	MOVW	$1,R0
 	SWPW	R0,(R1)		/* fix: deprecated in armv7 */

sys/src/libc/arm/vlop.s

@@ -1,4 +1,4 @@
-TEXT	_mulv(SB), $0
+TEXT	_mulv(SB), 1, $0
 	MOVW	4(FP),R8	/* l0 */
 	MOVW	8(FP),R11	/* h0 */
 	MOVW	12(FP),R4	/* l1 */

sys/src/libc/arm/vlrt.c

@@ -128,6 +128,8 @@ _v2f(Vlong x)
 	return _v2d(x);
 }
 
+/* too many of these are also needed by profiler; leave them out */
+#pragma profile off
 
 static void
 dodiv(Vlong num, Vlong den, Vlong *q, Vlong *r)
@@ -503,6 +505,7 @@ _sl2v(Vlong *ret, long sl)
 	ret->hi = t >> 31;
 }
 
+
 void
 _ul2v(Vlong *ret, ulong ul)
 {