patch-2.1.44 linux/arch/sparc64/lib/VISbzero.S

• Lines: 247
• Date: Mon Jul 7 08:18:55 1997
• Orig file: v2.1.43/linux/arch/sparc64/lib/VISbzero.S
• Orig date: Wed Dec 31 16:00:00 1969

diff -u --recursive --new-file v2.1.43/linux/arch/sparc64/lib/VISbzero.S linux/arch/sparc64/lib/VISbzero.S
@@ -0,0 +1,246 @@
+/* $Id: VISbzero.S,v 1.4 1997/06/28 17:21:21 jj Exp $
+ * VISbzero.S: High speed clear operations utilizing the UltraSparc
+ *        Visual Instruction Set.
+ *
+ * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
+ * Copyright (C) 1996, 1997 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
+ */
+
+#include "VIS.h"
+
+#ifdef __KERNEL__
+#define EXN(x,y,a,b,z) 				\
+98: 	x,y;					\
+	.section .fixup;			\
+	.align	4;				\
+99:	ba	VISbzerofixup_ret##z;		\
+	 a, b, %o0;				\
+	.section __ex_table;			\
+	.align	8;				\
+	.xword	98b, 99b;			\
+	.text;					\
+	.align	4;
+#define EXC(x,y,a,b,c...) 			\
+98: 	x,y;					\
+	.section .fixup;			\
+	.align	4;				\
+99:	c;					\
+	ba	VISbzerofixup_ret0;		\
+	 a, b, %o0;				\
+	.section __ex_table;			\
+	.align	8;				\
+	.xword	98b, 99b;			\
+	.text;					\
+	.align	4;
+#define EXO1(x,y) 				\
+98: 	x,y;					\
+	.section __ex_table;			\
+	.align	8;				\
+	.xword	98b, VISbzerofixup_reto1;	\
+	.text;					\
+	.align	4;
+#define EX(x,y,a,b) EXN(x,y,a,b,0)
+#define EX1(x,y,a,b) EXN(x,y,a,b,1)
+#define EX2(x,y,a,b) EXN(x,y,a,b,2)
+#define EXT(start,end,handler) 			\
+	.section __ex_table;			\
+	.align	8;				\
+	.xword	start, 0, end, handler;		\
+	.text;					\
+	.align	4
+#else
+#define EX(x,y,a,b)		x,y
+#define EX1(x,y,a,b)		x,y
+#define EX2(x,y,a,b)		x,y
+#define EXC(x,y,a,b,c...)	x,y
+#define EXO1(x,y)		x,y
+#define EXT(a,b,c)
+#endif
+
+#define ZERO_BLOCKS(base, offset, source)			\
+	STX	source, [base - offset - 0x38] ASINORMAL;	\
+	STX	source, [base - offset - 0x30] ASINORMAL;	\
+	STX	source, [base - offset - 0x28] ASINORMAL;	\
+	STX	source, [base - offset - 0x20] ASINORMAL;	\
+	STX	source, [base - offset - 0x18] ASINORMAL;	\
+	STX	source, [base - offset - 0x10] ASINORMAL;	\
+	STX	source, [base - offset - 0x08] ASINORMAL;	\
+	STX	source, [base - offset - 0x00] ASINORMAL;
+
+#ifdef __KERNEL__
+#define RETL	clr %o0
+#else
+#define RETL	mov %g3, %o0
+#endif
+
+	/* Well, bzero is a lot easier to get right than bcopy... */
+#ifdef __KERNEL__
+	.section	__ex_table,#alloc
+	.section	.fixup,#alloc,#execinstr
+#endif
+	.text
+	.align		32
+#ifdef __KERNEL__
+	.globl		__bzero, __bzero_noasi
+__bzero:
+	wr		%g0, ASI_P, %asi		! LSU	Group
+__bzero_noasi:
+#else
+	.globl		bzero
+bzero_private:
+bzero:
+#ifndef REGS_64BIT
+	srl		%o1, 0, %o1
+#endif
+	mov		%o0, %g3
+#endif
+	cmp		%o1, 7
+	bleu,pn		%xcc, 17f
+	 andcc		%o0, 3, %o2
+	be,a,pt		%xcc, 4f
+	 andcc		%o0, 4, %g0
+	cmp		%o2, 3
+	be,pn		%xcc, 2f
+	 EXO1(STB	%g0, [%o0 + 0x00] ASINORMAL)
+	cmp		%o2, 2
+	be,pt		%xcc, 2f
+	 EX(STB		%g0, [%o0 + 0x01] ASINORMAL, sub %o1, 1)
+	EX(STB		%g0, [%o0 + 0x02] ASINORMAL, sub %o1, 2)
+2:	sub		%o2, 4, %o2
+	sub		%o0, %o2, %o0
+	add		%o1, %o2, %o1
+	andcc		%o0, 4, %g0
+4:	be,pt		%xcc, 2f
+	 cmp		%o1, 128
+	EXO1(STW	%g0, [%o0] ASINORMAL)
+	sub		%o1, 4, %o1
+	add		%o0, 4, %o0
+2:	blu,pn		%xcc, 9f
+	 andcc		%o0, 0x38, %o2
+	be,pn		%icc, 6f
+	 mov		64, %o5
+	andcc		%o0, 8, %g0
+	be,pn		%icc, 1f
+	 sub		%o5, %o2, %o5
+	EX(STX		%g0, [%o0] ASINORMAL, sub %o1, 0)
+	add		%o0, 8, %o0
+1:	andcc		%o5, 16, %g0
+	be,pn		%icc, 1f
+	 sub		%o1, %o5, %o1
+	EX1(STX		%g0, [%o0] ASINORMAL, add %g0, 0)
+	EX1(STX		%g0, [%o0 + 8] ASINORMAL, sub %g0, 8)
+	add		%o0, 16, %o0
+1:	andcc		%o5, 32, %g0
+	be,pn		%icc, 7f
+	 andncc		%o1, 0x3f, %o3
+	EX(STX		%g0, [%o0] ASINORMAL, add %o1, 32)
+	EX(STX		%g0, [%o0 + 8] ASINORMAL, add %o1, 24)
+	EX(STX		%g0, [%o0 + 16] ASINORMAL, add %o1, 16)
+	EX(STX		%g0, [%o0 + 24] ASINORMAL, add %o1, 8)
+	add		%o0, 32, %o0
+6:	andncc		%o1, 0x3f, %o3
+7:	be,pn		%xcc, 9f
+#ifdef __KERNEL__
+	 rd		%asi, %g7
+	wr		%g0, FPRS_FEF, %fprs
+	wr		%g7, ASI_BLK_XOR, %asi
+#else
+	 wr		%g0, ASI_BLK_P, %asi
+#endif
+	membar		#StoreStore | #LoadStore
+	fzero		%f0
+	andcc		%o3, 0xc0, %o2
+	and		%o1, 0x3f, %o1
+	fzero		%f2
+	andn		%o3, 0xff, %o3
+	faddd		%f0, %f2, %f4
+	fmuld		%f0, %f2, %f6
+	cmp		%o2, 64
+	faddd		%f0, %f2, %f8
+	fmuld		%f0, %f2, %f10
+	faddd		%f0, %f2, %f12
+	brz,pn		%o2, 10f
+	 fmuld		%f0, %f2, %f14
+	be,pn		%icc, 2f
+	 EXC(STBLK	%f0, [%o0 + 0x00] ASIBLK, add %o3, %o2, add %o2, %o1, %o2)
+	cmp		%o2, 128
+	be,pn		%icc, 2f
+	 EXC(STBLK	%f0, [%o0 + 0x40] ASIBLK, add %o3, %o2, add %o2, %o1, %o2; sub %o2, 64, %o2)
+	EXC(STBLK	%f0, [%o0 + 0x80] ASIBLK, add %o3, %o2, add %o2, %o1, %o2; sub %o2, 128, %o2)
+2:	brz,pn		%o3, 12f
+	 add		%o0, %o2, %o0
+10:	EX(STBLK	%f0, [%o0 + 0x00] ASIBLK, add %o3, %o1)
+	EXC(STBLK	%f0, [%o0 + 0x40] ASIBLK, add %o3, %o1, sub %o1, 64, %o1)
+	EXC(STBLK	%f0, [%o0 + 0x80] ASIBLK, add %o3, %o1, sub %o1, 128, %o1)
+	EXC(STBLK	%f0, [%o0 + 0xc0] ASIBLK, add %o3, %o1, sub %o1, 192, %o1)
+11:	subcc		%o3, 256, %o3
+	bne,pt		%xcc, 10b
+	 add		%o0, 256, %o0
+12:
+#ifdef __KERNEL__
+	wr		%g0, 0, %fprs
+	wr		%g7, 0x0, %asi
+#endif
+	membar		#Sync
+9:	andcc		%o1, 0xf8, %o2
+	be,pn		%xcc, 13f
+	 andcc		%o1, 7, %o1
+14:	rd		%pc, %o4
+	srl		%o2, 1, %o3
+	sub		%o4, %o3, %o4
+	jmpl		%o4 + (13f - 14b), %g0
+	 add		%o0, %o2, %o0
+12:	ZERO_BLOCKS(%o0, 0xc8, %g0)
+	ZERO_BLOCKS(%o0, 0x88, %g0)
+	ZERO_BLOCKS(%o0, 0x48, %g0)
+	ZERO_BLOCKS(%o0, 0x08, %g0)
+	EXT(12b,13f,VISbzerofixup_zb)
+13:	be,pn		%xcc, 8f
+	 andcc		%o1, 4, %g0
+	be,pn		%xcc, 1f
+	 andcc		%o1, 2, %g0
+	EX(STW		%g0, [%o0] ASINORMAL, and %o1, 7)
+	add		%o0, 4, %o0
+1:	be,pn		%xcc, 1f
+	 andcc		%o1, 1, %g0
+	EX(STH		%g0, [%o0] ASINORMAL, and %o1, 3)
+	add		%o0, 2, %o0
+1:	bne,a,pn	%xcc, 8f
+	 EX(STB		%g0, [%o0] ASINORMAL, add %g0, 1)
+8:	retl
+	 RETL
+17:	be,pn		%xcc, 13b
+	 orcc		%o1, 0, %g0
+	be,pn		%xcc, 0f
+8:	 add		%o0, 1, %o0
+	subcc		%o1, 1, %o1
+	bne,pt		%xcc, 8b
+	 EX(STB		%g0, [%o0 - 1] ASINORMAL, add %o1, 1)
+0:	retl
+	 RETL
+
+#ifdef __KERNEL__
+	.section	.fixup
+	.align		4
+VISbzerofixup_reto1:
+	mov		%o1, %o0
+VISbzerofixup_ret0:
+	retl
+	 wr		%g0, 0, %fprs
+VISbzerofixup_ret1:
+	and		%o5, 0x30, %o5
+	add		%o5, %o1, %o5
+	ba,pt		%xcc, VISbzerofixup_ret0
+	 add		%o0, %o5, %o0
+VISbzerofixup_ret2:
+	and		%o5, 0x20, %o5
+	add		%o5, %o1, %o5
+	ba,pt		%xcc, VISbzerofixup_ret0
+	 add		%o0, %o5, %o0
+VISbzerofixup_zb:
+	andcc		%o1, 7, %o1
+	sll		%g2, 3, %g2
+	add		%o1, 256, %o1
+	ba,pt		%xcc, VISbzerofixup_ret0
+	 sub		%o1, %g2, %o0
+#endif