bcopy should handle overlapping copies like memmove, while the previous version worked like memcpy. Implement optimized assembly memmove.

1 files changed, 285 insertions, 8 deletions

diff --git a/sysdeps/sparc/sparc64/memcpy.S b/sysdeps/sparc/sparc64/memcpy.S
index 73057ff0a1..f825b909f7 100644
--- a/sysdeps/sparc/sparc64/memcpy.S
+++ b/sysdeps/sparc/sparc64/memcpy.S
@@ -3,7 +3,7 @@
    Copyright (C) 1996, 97, 98, 99 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by David S. Miller (davem@caip.rutgers.edu) and
-                  Jakub Jelinek (jj@ultra.linux.cz).
+		  Jakub Jelinek (jakub@redhat.com).
 
    The GNU C Library is free software; you can redistribute it and/or
    modify it under the terms of the GNU Library General Public License as
@@ -27,6 +27,7 @@
 	.register	%g2, #scratch
 	.register	%g3, #scratch
 	.register	%g7, #scratch
+#define XCC	xcc
 #endif
 #define FPRS_FEF	4
 
@@ -135,16 +136,76 @@
 	stx		%t0, [%dst - offset - 0x10]; 		\
 	stx		%t1, [%dst - offset - 0x08];
 
+	/* Macros for non-VIS memmove code. */
+#define RMOVE_BIGCHUNK(src, dst, offset, t0, t1, t2, t3)	\
+	ldx		[%src - offset - 0x20], %t0; 		\
+	ldx		[%src - offset - 0x18], %t1; 		\
+	ldx		[%src - offset - 0x10], %t2; 		\
+	ldx		[%src - offset - 0x08], %t3; 		\
+	stw		%t0, [%dst - offset - 0x1c]; 		\
+	srlx		%t0, 32, %t0;				\
+	stw		%t0, [%dst - offset - 0x20]; 		\
+	stw		%t1, [%dst - offset - 0x14]; 		\
+	srlx		%t1, 32, %t1;				\
+	stw		%t1, [%dst - offset - 0x18]; 		\
+	stw		%t2, [%dst - offset - 0x0c]; 		\
+	srlx		%t2, 32, %t2;				\
+	stw		%t2, [%dst - offset - 0x10]; 		\
+	stw		%t3, [%dst - offset - 0x04];		\
+	srlx		%t3, 32, %t3;				\
+	stw		%t3, [%dst - offset - 0x08];
+
+#define RMOVE_BIGALIGNCHUNK(src, dst, offset, t0, t1, t2, t3)	\
+	ldx		[%src - offset - 0x20], %t0; 		\
+	ldx		[%src - offset - 0x18], %t1; 		\
+	ldx		[%src - offset - 0x10], %t2; 		\
+	ldx		[%src - offset - 0x08], %t3; 		\
+	stx		%t0, [%dst - offset - 0x20]; 		\
+	stx		%t1, [%dst - offset - 0x18]; 		\
+	stx		%t2, [%dst - offset - 0x10]; 		\
+	stx		%t3, [%dst - offset - 0x08];		\
+	ldx		[%src - offset - 0x40], %t0; 		\
+	ldx		[%src - offset - 0x38], %t1; 		\
+	ldx		[%src - offset - 0x30], %t2; 		\
+	ldx		[%src - offset - 0x28], %t3; 		\
+	stx		%t0, [%dst - offset - 0x40]; 		\
+	stx		%t1, [%dst - offset - 0x38]; 		\
+	stx		%t2, [%dst - offset - 0x30]; 		\
+	stx		%t3, [%dst - offset - 0x28];
+
+#define RMOVE_LASTCHUNK(src, dst, offset, t0, t1, t2, t3)	\
+	ldx		[%src + offset + 0x00], %t0;		\
+	ldx		[%src + offset + 0x08], %t1; 		\
+	stw		%t0, [%dst + offset + 0x04]; 		\
+	srlx		%t0, 32, %t2;				\
+	stw		%t2, [%dst + offset + 0x00]; 		\
+	stw		%t1, [%dst + offset + 0x0c]; 		\
+	srlx		%t1, 32, %t3;				\
+	stw		%t3, [%dst + offset + 0x08];
+
+#define RMOVE_LASTALIGNCHUNK(src, dst, offset, t0, t1)		\
+	ldx		[%src + offset + 0x00], %t0; 		\
+	ldx		[%src + offset + 0x08], %t1; 		\
+	stx		%t0, [%dst + offset + 0x00]; 		\
+	stx		%t1, [%dst + offset + 0x08];
+
 	.text
 	.align		32
 
 ENTRY(bcopy)
-	or		%o0, 0, %g3			/* IEU0		Group		*/
-	addcc		%o1, 0, %o0			/* IEU1				*/
-	brgez,pt	%o2, 210f			/* CTI				*/
-	 or		%g3, 0, %o1			/* IEU0		Group		*/
-	retl						/* CTI		Group brk forced*/
-	 clr		%o0				/* IEU0				*/
+	sub		%o1, %o0, %o4			/* IEU0		Group		*/
+	mov		%o0, %g3			/* IEU1				*/
+	cmp		%o4, %o2			/* IEU1		Group		*/
+	mov		%o1, %o0			/* IEU0				*/
+	bgeu,pt		%XCC, 210f			/* CTI				*/
+	 mov		%g3, %o1			/* IEU0		Group		*/
+#ifndef USE_BPR
+	srl		%o2, 0, %o2			/* IEU1				*/
+#endif
+	brnz,pn		%o2, 220f			/* CTI		Group		*/
+	 add		%o0, %o2, %o0			/* IEU0				*/
+	retl
+	 nop
 END(bcopy)
 
 	.align		32
@@ -450,7 +511,7 @@ ENTRY(memcpy)
 #endif	
 	brz,pn		%o2, 209b			/* CTI		Group		*/
 	 mov		%o0, %g4			/* IEU0				*/
-	cmp		%o2, 15				/* IEU1		Group		*/
+218:	cmp		%o2, 15				/* IEU1		Group		*/
 	bleu,pn		%xcc, 208b			/* CTI				*/
 	 cmp		%o2, (64 * 6)			/* IEU1		Group		*/
 	bgeu,pn		%xcc, 200b			/* CTI				*/
@@ -636,6 +697,222 @@ ENTRY(memcpy)
 	 mov		%g4, %o0
 END(memcpy)
 
+	.align		32
+228:	andcc		%o2, 1, %g0			/* IEU1		Group		*/
+	be,pt		%icc, 2f+4			/* CTI				*/
+1:	 ldub		[%o1 - 1], %o5			/* LOAD		Group		*/
+	sub		%o1, 1, %o1			/* IEU0				*/
+	sub		%o0, 1, %o0			/* IEU1				*/
+	subcc		%o2, 1, %o2			/* IEU1		Group		*/
+	be,pn		%xcc, 229f			/* CTI				*/
+	 stb		%o5, [%o0]			/* Store			*/
+2:	ldub		[%o1 - 1], %o5			/* LOAD		Group		*/
+	sub		%o0, 2, %o0			/* IEU0				*/
+	ldub		[%o1 - 2], %g5			/* LOAD		Group		*/
+	sub		%o1, 2, %o1			/* IEU0				*/
+	subcc		%o2, 2, %o2			/* IEU1		Group		*/
+	stb		%o5, [%o0 + 1]			/* Store			*/
+	bne,pt		%xcc, 2b			/* CTI				*/
+	 stb		%g5, [%o0]			/* Store			*/
+229:	retl
+	 mov		%g4, %o0
+219:	retl
+	 nop
+
+	.align		32
+ENTRY(memmove)
+#ifndef USE_BPR
+	srl		%o2, 0, %o2			/* IEU1		Group		*/
+#endif
+	brz,pn		%o2, 219b			/* CTI		Group		*/
+	 sub		%o0, %o1, %o4			/* IEU0				*/
+	cmp		%o4, %o2			/* IEU1		Group		*/
+	bgeu,pt		%XCC, 218b			/* CTI				*/
+	 mov		%o0, %g4			/* IEU0				*/
+	add		%o0, %o2, %o0			/* IEU0		Group		*/
+220:	add		%o1, %o2, %o1			/* IEU1				*/
+	cmp		%o2, 15				/* IEU1		Group		*/
+	bleu,pn		%xcc, 228b			/* CTI				*/
+	 andcc		%o0, 7, %g2			/* IEU1		Group		*/
+	sub		%o0, %o1, %g5			/* IEU0				*/
+	andcc		%g5, 3, %o5			/* IEU1		Group		*/
+	bne,pn		%xcc, 232f			/* CTI				*/
+	 andcc		%o1, 3, %g0			/* IEU1		Group		*/
+	be,a,pt		%xcc, 236f			/* CTI				*/
+	 andcc		%o1, 4, %g0			/* IEU1		Group		*/
+	andcc		%o1, 1, %g0			/* IEU1		Group		*/
+	be,pn		%xcc, 4f			/* CTI				*/
+	 andcc		%o1, 2, %g0			/* IEU1		Group		*/
+	ldub		[%o1 - 1], %g2			/* Load		Group		*/
+	sub		%o1, 1, %o1			/* IEU0				*/
+	sub		%o0, 1, %o0			/* IEU1				*/
+	sub		%o2, 1, %o2			/* IEU0		Group		*/
+	be,pn		%xcc, 5f			/* CTI		Group		*/
+	 stb		%g2, [%o0]			/* Store			*/
+4:	lduh		[%o1 - 2], %g2			/* Load		Group		*/
+	sub		%o1, 2, %o1			/* IEU0				*/
+	sub		%o0, 2, %o0			/* IEU1				*/
+	sub		%o2, 2, %o2			/* IEU0				*/
+	sth		%g2, [%o0]			/* Store	Group + bubble	*/
+5:	andcc		%o1, 4, %g0			/* IEU1				*/
+236:	be,a,pn		%xcc, 2f			/* CTI				*/
+	 andcc		%o2, -128, %g7			/* IEU1		Group		*/
+	lduw		[%o1 - 4], %g5			/* Load		Group		*/
+	sub		%o1, 4, %o1			/* IEU0				*/
+	sub		%o0, 4, %o0			/* IEU1				*/
+	sub		%o2, 4, %o2			/* IEU0		Group		*/
+	stw		%g5, [%o0]			/* Store			*/
+	andcc		%o2, -128, %g7			/* IEU1		Group		*/
+2:	be,pn		%xcc, 235f			/* CTI				*/
+	 andcc		%o0, 4, %g0			/* IEU1		Group		*/
+	be,pn		%xcc, 282f + 4			/* CTI		Group		*/
+5:	RMOVE_BIGCHUNK(o1, o0, 0x00, g1, g3, g5, o5)
+	RMOVE_BIGCHUNK(o1, o0, 0x20, g1, g3, g5, o5)
+	RMOVE_BIGCHUNK(o1, o0, 0x40, g1, g3, g5, o5)
+	RMOVE_BIGCHUNK(o1, o0, 0x60, g1, g3, g5, o5)
+	subcc		%g7, 128, %g7			/* IEU1		Group		*/
+	sub		%o1, 128, %o1			/* IEU0				*/
+	bne,pt		%xcc, 5b			/* CTI				*/
+	 sub		%o0, 128, %o0			/* IEU0		Group		*/
+235:	andcc		%o2, 0x70, %g7			/* IEU1		Group		*/
+41:	be,pn		%xcc, 280f			/* CTI				*/
+	 andcc		%o2, 8, %g0			/* IEU1		Group		*/
+							/* Clk1 8-(			*/
+							/* Clk2 8-(			*/
+							/* Clk3 8-(			*/
+							/* Clk4 8-(			*/
+279:	rd		%pc, %o5			/* PDU		Group		*/
+	sll		%g7, 1, %g5			/* IEU0		Group		*/
+	sub		%o1, %g7, %o1			/* IEU1				*/
+	sub		%o5, %g5, %o5			/* IEU0  	Group		*/
+	jmpl		%o5 + %lo(280f - 279b), %g0	/* CTI		Group brk forced*/
+	 sub		%o0, %g7, %o0			/* IEU0		Group		*/
+	RMOVE_LASTCHUNK(o1, o0, 0x60, g2, g3, g5, o5)
+	RMOVE_LASTCHUNK(o1, o0, 0x50, g2, g3, g5, o5)
+	RMOVE_LASTCHUNK(o1, o0, 0x40, g2, g3, g5, o5)
+	RMOVE_LASTCHUNK(o1, o0, 0x30, g2, g3, g5, o5)
+	RMOVE_LASTCHUNK(o1, o0, 0x20, g2, g3, g5, o5)
+	RMOVE_LASTCHUNK(o1, o0, 0x10, g2, g3, g5, o5)
+	RMOVE_LASTCHUNK(o1, o0, 0x00, g2, g3, g5, o5)
+280:	be,pt		%xcc, 281f			/* CTI				*/
+	 andcc		%o2, 4, %g0			/* IEU1				*/
+	ldx		[%o1 - 8], %g2			/* Load		Group		*/
+	sub		%o0, 8, %o0			/* IEU0				*/
+	stw		%g2, [%o0 + 4]			/* Store	Group		*/
+	sub		%o1, 8, %o1			/* IEU1				*/
+	srlx		%g2, 32, %g2			/* IEU0		Group		*/
+	stw		%g2, [%o0]			/* Store			*/
+281:	be,pt		%xcc, 1f			/* CTI				*/
+	 andcc		%o2, 2, %g0			/* IEU1		Group		*/
+	lduw		[%o1 - 4], %g2			/* Load		Group		*/
+	sub		%o1, 4, %o1			/* IEU0				*/
+	stw		%g2, [%o0 - 4]			/* Store	Group		*/
+	sub		%o0, 4, %o0			/* IEU0				*/
+1:	be,pt		%xcc, 1f			/* CTI				*/
+	 andcc		%o2, 1, %g0			/* IEU1		Group		*/
+	lduh		[%o1 - 2], %g2			/* Load		Group		*/
+	sub		%o1, 2, %o1			/* IEU0				*/
+	sth		%g2, [%o0 - 2]			/* Store	Group		*/
+	sub		%o0, 2, %o0			/* IEU0				*/
+1:	be,pt		%xcc, 211f			/* CTI				*/
+	 nop						/* IEU1				*/
+	ldub		[%o1 - 1], %g2			/* Load		Group		*/
+	stb		%g2, [%o0 - 1]			/* Store	Group + bubble	*/
+211:	retl
+	 mov		%g4, %o0
+
+282:	RMOVE_BIGALIGNCHUNK(o1, o0, 0x00, g1, g3, g5, o5)
+	RMOVE_BIGALIGNCHUNK(o1, o0, 0x40, g1, g3, g5, o5)
+	subcc		%g7, 128, %g7			/* IEU1		Group		*/
+	sub		%o1, 128, %o1			/* IEU0				*/
+	bne,pt		%xcc, 282b			/* CTI				*/
+	 sub		%o0, 128, %o0			/* IEU0		Group		*/
+	andcc		%o2, 0x70, %g7			/* IEU1				*/
+	be,pn		%xcc, 284f			/* CTI				*/
+	 andcc		%o2, 8, %g0			/* IEU1		Group		*/
+							/* Clk1 8-(			*/
+							/* Clk2 8-(			*/
+							/* Clk3 8-(			*/
+							/* Clk4 8-(			*/
+283:	rd		%pc, %o5			/* PDU		Group		*/
+	sub		%o1, %g7, %o1			/* IEU0		Group		*/
+	sub		%o5, %g7, %o5			/* IEU1				*/
+	jmpl		%o5 + %lo(284f - 283b), %g0	/* CTI		Group brk forced*/
+	 sub		%o0, %g7, %o0			/* IEU0		Group		*/
+	RMOVE_LASTALIGNCHUNK(o1, o0, 0x60, g2, g3)
+	RMOVE_LASTALIGNCHUNK(o1, o0, 0x50, g2, g3)
+	RMOVE_LASTALIGNCHUNK(o1, o0, 0x40, g2, g3)
+	RMOVE_LASTALIGNCHUNK(o1, o0, 0x30, g2, g3)
+	RMOVE_LASTALIGNCHUNK(o1, o0, 0x20, g2, g3)
+	RMOVE_LASTALIGNCHUNK(o1, o0, 0x10, g2, g3)
+	RMOVE_LASTALIGNCHUNK(o1, o0, 0x00, g2, g3)
+284:	be,pt		%xcc, 285f			/* CTI		Group		*/
+	 andcc		%o2, 4, %g0			/* IEU1				*/
+	ldx		[%o1 - 8], %g2			/* Load		Group		*/
+	sub		%o0, 8, %o0			/* IEU0				*/
+	sub		%o1, 8, %o1			/* IEU0		Group		*/
+	stx		%g2, [%o0]			/* Store			*/
+285:	be,pt		%xcc, 1f			/* CTI				*/
+	 andcc		%o2, 2, %g0			/* IEU1		Group		*/
+	lduw		[%o1 - 4], %g2			/* Load		Group		*/
+	sub		%o0, 4, %o0			/* IEU0				*/
+	sub		%o1, 4, %o1			/* IEU0		Group		*/
+	stw		%g2, [%o0]			/* Store			*/
+1:	be,pt		%xcc, 1f			/* CTI				*/
+	 andcc		%o2, 1, %g0			/* IEU1		Group		*/
+	lduh		[%o1 - 2], %g2			/* Load		Group		*/
+	sub		%o0, 2, %o0			/* IEU0				*/
+	sub		%o1, 2, %o1			/* IEU0		Group		*/
+	sth		%g2, [%o0]			/* Store			*/
+1:	be,pt		%xcc, 1f			/* CTI				*/
+	 nop						/* IEU0		Group		*/
+	ldub		[%o1 - 1], %g2			/* Load		Group		*/
+	stb		%g2, [%o0 - 1]			/* Store	Group + bubble	*/
+1:	retl
+	 mov		%g4, %o0
+
+232:	brz,pt		%g2, 2f				/* CTI		Group		*/
+	 sub		%o2, %g2, %o2			/* IEU0		Group		*/
+1:	ldub		[%o1 - 1], %g5			/* Load		Group		*/
+	sub		%o1, 1, %o1			/* IEU0				*/
+	sub		%o0, 1, %o0			/* IEU1				*/
+	subcc		%g2, 1, %g2			/* IEU1		Group		*/
+	bne,pt		%xcc, 1b			/* CTI				*/
+	 stb		%g5, [%o0]			/* Store			*/
+2:	andn		%o2, 7, %g5 			/* IEU0		Group		*/
+	and		%o2, 7, %o2			/* IEU1				*/
+	fmovd		%f0, %f2			/* FPU				*/
+	alignaddr	%o1, %g0, %g1			/* GRU		Group		*/
+	ldd		[%g1], %f4			/* Load		Group		*/
+1:	ldd		[%g1 - 8], %f6			/* Load		Group		*/
+	sub		%g1, 8, %g1			/* IEU0		Group		*/
+	subcc		%g5, 8, %g5			/* IEU1				*/
+	faligndata	%f6, %f4, %f0			/* GRU		Group		*/
+	std		%f0, [%o0 - 8]			/* Store			*/
+	sub		%o1, 8, %o1			/* IEU0		Group		*/
+	be,pn		%xcc, 233f			/* CTI				*/
+	 sub		%o0, 8, %o0			/* IEU1				*/
+	ldd		[%g1 - 8], %f4			/* Load		Group		*/
+	sub		%g1, 8, %g1			/* IEU0				*/
+	subcc		%g5, 8, %g5			/* IEU1				*/
+	faligndata	%f4, %f6, %f0			/* GRU		Group		*/
+	std		%f0, [%o0 - 8]			/* Store			*/
+	sub		%o1, 8, %o1			/* IEU0				*/
+	bne,pn		%xcc, 1b			/* CTI		Group		*/
+	 sub		%o0, 8, %o0			/* IEU0				*/
+233:	brz,pn		%o2, 234f			/* CTI		Group		*/
+	 nop						/* IEU0				*/
+237:	ldub		[%o1 - 1], %g5			/* LOAD				*/
+	sub		%o1, 1, %o1			/* IEU0				*/
+	sub		%o0, 1, %o0			/* IEU1				*/
+	subcc		%o2, 1, %o2			/* IEU1				*/
+	bne,pt		%xcc, 237b			/* CTI				*/
+	 stb		%g5, [%o0]			/* Store	Group		*/
+234:	wr		%g0, FPRS_FEF, %fprs
+	retl
+	 mov		%g4, %o0
+END(memmove)
+
 #ifdef USE_BPR
 weak_alias(memcpy, __align_cpy_1)
 weak_alias(memcpy, __align_cpy_2)