[SPARC64]: Niagara-2 optimized copies.

The bzero/memset implementation stays the same as Niagara-1.

Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/arch/sparc64/lib/NG2memcpy.S b/arch/sparc64/lib/NG2memcpy.S
new file mode 100644
index 0000000..0aed756
--- /dev/null
+++ b/arch/sparc64/lib/NG2memcpy.S
@@ -0,0 +1,520 @@
+/* NG2memcpy.S: Niagara-2 optimized memcpy.
+ *
+ * Copyright (C) 2007 David S. Miller (davem@davemloft.net)
+ */
+
+#ifdef __KERNEL__
+#include <asm/visasm.h>
+#include <asm/asi.h>
+#define GLOBAL_SPARE	%g7
+#else
+#define ASI_PNF 0x82
+#define ASI_BLK_P 0xf0
+#define ASI_BLK_INIT_QUAD_LDD_P 0xe2
+#define FPRS_FEF  0x04
+#ifdef MEMCPY_DEBUG
+#define VISEntryHalf rd %fprs, %o5; wr %g0, FPRS_FEF, %fprs; \
+		     clr %g1; clr %g2; clr %g3; subcc %g0, %g0, %g0;
+#define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs
+#else
+#define VISEntryHalf rd %fprs, %o5; wr %g0, FPRS_FEF, %fprs
+#define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs
+#endif
+#define GLOBAL_SPARE	%g5
+#endif
+
+#ifndef STORE_ASI
+#ifndef SIMULATE_NIAGARA_ON_NON_NIAGARA
+#define STORE_ASI	ASI_BLK_INIT_QUAD_LDD_P
+#else
+#define STORE_ASI	0x80		/* ASI_P */
+#endif
+#endif
+
+#ifndef EX_LD
+#define EX_LD(x)	x
+#endif
+
+#ifndef EX_ST
+#define EX_ST(x)	x
+#endif
+
+#ifndef EX_RETVAL
+#define EX_RETVAL(x)	x
+#endif
+
+#ifndef LOAD
+#define LOAD(type,addr,dest)	type [addr], dest
+#endif
+
+#ifndef LOAD_BLK
+#define LOAD_BLK(addr,dest)	ldda [addr] ASI_BLK_P, dest
+#endif
+
+#ifndef STORE
+#ifndef MEMCPY_DEBUG
+#define STORE(type,src,addr)	type src, [addr]
+#else
+#define STORE(type,src,addr)	type##a src, [addr] 0x80
+#endif
+#endif
+
+#ifndef STORE_BLK
+#define STORE_BLK(src,addr)	stda src, [addr] ASI_BLK_P
+#endif
+
+#ifndef STORE_INIT
+#define STORE_INIT(src,addr)	stxa src, [addr] STORE_ASI
+#endif
+
+#ifndef FUNC_NAME
+#define FUNC_NAME	NG2memcpy
+#endif
+
+#ifndef PREAMBLE
+#define PREAMBLE
+#endif
+
+#ifndef XCC
+#define XCC xcc
+#endif
+
+#define FREG_FROB(x0, x1, x2, x3, x4, x5, x6, x7, x8) \
+	faligndata	%x0, %x1, %f0; \
+	faligndata	%x1, %x2, %f2; \
+	faligndata	%x2, %x3, %f4; \
+	faligndata	%x3, %x4, %f6; \
+	faligndata	%x4, %x5, %f8; \
+	faligndata	%x5, %x6, %f10; \
+	faligndata	%x6, %x7, %f12; \
+	faligndata	%x7, %x8, %f14;
+
+#define FREG_MOVE_1(x0) \
+	fmovd		%x0, %f0;
+#define FREG_MOVE_2(x0, x1) \
+	fmovd		%x0, %f0; \
+	fmovd		%x1, %f2;
+#define FREG_MOVE_3(x0, x1, x2) \
+	fmovd		%x0, %f0; \
+	fmovd		%x1, %f2; \
+	fmovd		%x2, %f4;
+#define FREG_MOVE_4(x0, x1, x2, x3) \
+	fmovd		%x0, %f0; \
+	fmovd		%x1, %f2; \
+	fmovd		%x2, %f4; \
+	fmovd		%x3, %f6;
+#define FREG_MOVE_5(x0, x1, x2, x3, x4) \
+	fmovd		%x0, %f0; \
+	fmovd		%x1, %f2; \
+	fmovd		%x2, %f4; \
+	fmovd		%x3, %f6; \
+	fmovd		%x4, %f8;
+#define FREG_MOVE_6(x0, x1, x2, x3, x4, x5) \
+	fmovd		%x0, %f0; \
+	fmovd		%x1, %f2; \
+	fmovd		%x2, %f4; \
+	fmovd		%x3, %f6; \
+	fmovd		%x4, %f8; \
+	fmovd		%x5, %f10;
+#define FREG_MOVE_7(x0, x1, x2, x3, x4, x5, x6) \
+	fmovd		%x0, %f0; \
+	fmovd		%x1, %f2; \
+	fmovd		%x2, %f4; \
+	fmovd		%x3, %f6; \
+	fmovd		%x4, %f8; \
+	fmovd		%x5, %f10; \
+	fmovd		%x6, %f12;
+#define FREG_MOVE_8(x0, x1, x2, x3, x4, x5, x6, x7) \
+	fmovd		%x0, %f0; \
+	fmovd		%x1, %f2; \
+	fmovd		%x2, %f4; \
+	fmovd		%x3, %f6; \
+	fmovd		%x4, %f8; \
+	fmovd		%x5, %f10; \
+	fmovd		%x6, %f12; \
+	fmovd		%x7, %f14;
+#define FREG_LOAD_1(base, x0) \
+	EX_LD(LOAD(ldd, base + 0x00, %x0))
+#define FREG_LOAD_2(base, x0, x1) \
+	EX_LD(LOAD(ldd, base + 0x00, %x0)); \
+	EX_LD(LOAD(ldd, base + 0x08, %x1));
+#define FREG_LOAD_3(base, x0, x1, x2) \
+	EX_LD(LOAD(ldd, base + 0x00, %x0)); \
+	EX_LD(LOAD(ldd, base + 0x08, %x1)); \
+	EX_LD(LOAD(ldd, base + 0x10, %x2));
+#define FREG_LOAD_4(base, x0, x1, x2, x3) \
+	EX_LD(LOAD(ldd, base + 0x00, %x0)); \
+	EX_LD(LOAD(ldd, base + 0x08, %x1)); \
+	EX_LD(LOAD(ldd, base + 0x10, %x2)); \
+	EX_LD(LOAD(ldd, base + 0x18, %x3));
+#define FREG_LOAD_5(base, x0, x1, x2, x3, x4) \
+	EX_LD(LOAD(ldd, base + 0x00, %x0)); \
+	EX_LD(LOAD(ldd, base + 0x08, %x1)); \
+	EX_LD(LOAD(ldd, base + 0x10, %x2)); \
+	EX_LD(LOAD(ldd, base + 0x18, %x3)); \
+	EX_LD(LOAD(ldd, base + 0x20, %x4));
+#define FREG_LOAD_6(base, x0, x1, x2, x3, x4, x5) \
+	EX_LD(LOAD(ldd, base + 0x00, %x0)); \
+	EX_LD(LOAD(ldd, base + 0x08, %x1)); \
+	EX_LD(LOAD(ldd, base + 0x10, %x2)); \
+	EX_LD(LOAD(ldd, base + 0x18, %x3)); \
+	EX_LD(LOAD(ldd, base + 0x20, %x4)); \
+	EX_LD(LOAD(ldd, base + 0x28, %x5));
+#define FREG_LOAD_7(base, x0, x1, x2, x3, x4, x5, x6) \
+	EX_LD(LOAD(ldd, base + 0x00, %x0)); \
+	EX_LD(LOAD(ldd, base + 0x08, %x1)); \
+	EX_LD(LOAD(ldd, base + 0x10, %x2)); \
+	EX_LD(LOAD(ldd, base + 0x18, %x3)); \
+	EX_LD(LOAD(ldd, base + 0x20, %x4)); \
+	EX_LD(LOAD(ldd, base + 0x28, %x5)); \
+	EX_LD(LOAD(ldd, base + 0x30, %x6));
+
+	.register	%g2,#scratch
+	.register	%g3,#scratch
+
+	.text
+	.align		64
+
+	.globl	FUNC_NAME
+	.type	FUNC_NAME,#function
+FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
+	srlx		%o2, 31, %g2
+	cmp		%g2, 0
+	tne		%xcc, 5
+	PREAMBLE
+	mov		%o0, GLOBAL_SPARE
+	cmp		%o2, 0
+	be,pn		%XCC, 85f
+	 or		%o0, %o1, %o3
+	cmp		%o2, 16
+	blu,a,pn	%XCC, 80f
+	 or		%o3, %o2, %o3
+
+	/* 2 blocks (128 bytes) is the minimum we can do the block
+	 * copy with.  We need to ensure that we'll iterate at least
+	 * once in the block copy loop.  At worst we'll need to align
+	 * the destination to a 64-byte boundary which can chew up
+	 * to (64 - 1) bytes from the length before we perform the
+	 * block copy loop.
+	 *
+	 * However, the cut-off point, performance wise, is around
+	 * 4 64-byte blocks.
+	 */
+	cmp		%o2, (4 * 64)
+	blu,pt		%XCC, 75f
+	 andcc		%o3, 0x7, %g0
+
+	/* %o0:	dst
+	 * %o1:	src
+	 * %o2:	len  (known to be >= 128)
+	 *
+	 * The block copy loops can use %o4, %g2, %g3 as
+	 * temporaries while copying the data.  %o5 must
+	 * be preserved between VISEntryHalf and VISExitHalf
+	 */
+
+	LOAD(prefetch, %o1 + 0x000, #one_read)
+	LOAD(prefetch, %o1 + 0x040, #one_read)
+	LOAD(prefetch, %o1 + 0x080, #one_read)
+
+	/* Align destination on 64-byte boundary.  */
+	andcc		%o0, (64 - 1), %o4
+	be,pt		%XCC, 2f
+	 sub		%o4, 64, %o4
+	sub		%g0, %o4, %o4	! bytes to align dst
+	sub		%o2, %o4, %o2
+1:	subcc		%o4, 1, %o4
+	EX_LD(LOAD(ldub, %o1, %g1))
+	EX_ST(STORE(stb, %g1, %o0))
+	add		%o1, 1, %o1
+	bne,pt		%XCC, 1b
+	add		%o0, 1, %o0
+
+2:
+	/* Clobbers o5/g1/g2/g3/g7/icc/xcc.  We must preserve
+	 * o5 from here until we hit VISExitHalf.
+	 */
+	VISEntryHalf
+
+	alignaddr	%o1, %g0, %g0
+
+	add		%o1, (64 - 1), %o4
+	andn		%o4, (64 - 1), %o4
+	andn		%o2, (64 - 1), %g1
+	sub		%o2, %g1, %o2
+
+	and		%o1, (64 - 1), %g2
+	add		%o1, %g1, %o1
+	sub		%o0, %o4, %g3
+	brz,pt		%g2, 190f
+	 cmp		%g2, 32
+	blu,a		5f
+	 cmp		%g2, 16
+	cmp		%g2, 48
+	blu,a		4f
+	 cmp		%g2, 40
+	cmp		%g2, 56
+	blu		170f
+	 nop
+	ba,a,pt		%xcc, 180f
+
+4:	/* 32 <= low bits < 48 */
+	blu		150f
+	 nop
+	ba,a,pt		%xcc, 160f
+5:	/* 0 < low bits < 32 */
+	blu,a		6f
+	 cmp		%g2, 8
+	cmp		%g2, 24
+	blu		130f
+	 nop
+	ba,a,pt		%xcc, 140f
+6:	/* 0 < low bits < 16 */
+	bgeu		120f
+	 nop
+	/* fall through for 0 < low bits < 8 */
+110:	sub		%o4, 64, %g2
+	EX_LD(LOAD_BLK(%g2, %f0))
+1:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
+	EX_LD(LOAD_BLK(%o4, %f16))
+	FREG_FROB(f0, f2, f4, f6, f8, f10, f12, f14, f16)
+	EX_ST(STORE_BLK(%f0, %o4 + %g3))
+	FREG_MOVE_8(f16, f18, f20, f22, f24, f26, f28, f30)
+	subcc		%g1, 64, %g1
+	add		%o4, 64, %o4
+	bne,pt		%xcc, 1b
+	 LOAD(prefetch, %o4 + 64, #one_read)
+	ba,pt		%xcc, 195f
+	 nop
+
+120:	sub		%o4, 56, %g2
+	FREG_LOAD_7(%g2, f0, f2, f4, f6, f8, f10, f12)
+1:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
+	EX_LD(LOAD_BLK(%o4, %f16))
+	FREG_FROB(f0, f2, f4, f6, f8, f10, f12, f16, f18)
+	EX_ST(STORE_BLK(%f0, %o4 + %g3))
+	FREG_MOVE_7(f18, f20, f22, f24, f26, f28, f30)
+	subcc		%g1, 64, %g1
+	add		%o4, 64, %o4
+	bne,pt		%xcc, 1b
+	 LOAD(prefetch, %o4 + 64, #one_read)
+	ba,pt		%xcc, 195f
+	 nop
+
+130:	sub		%o4, 48, %g2
+	FREG_LOAD_6(%g2, f0, f2, f4, f6, f8, f10)
+1:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
+	EX_LD(LOAD_BLK(%o4, %f16))
+	FREG_FROB(f0, f2, f4, f6, f8, f10, f16, f18, f20)
+	EX_ST(STORE_BLK(%f0, %o4 + %g3))
+	FREG_MOVE_6(f20, f22, f24, f26, f28, f30)
+	subcc		%g1, 64, %g1
+	add		%o4, 64, %o4
+	bne,pt		%xcc, 1b
+	 LOAD(prefetch, %o4 + 64, #one_read)
+	ba,pt		%xcc, 195f
+	 nop
+
+140:	sub		%o4, 40, %g2
+	FREG_LOAD_5(%g2, f0, f2, f4, f6, f8)
+1:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
+	EX_LD(LOAD_BLK(%o4, %f16))
+	FREG_FROB(f0, f2, f4, f6, f8, f16, f18, f20, f22)
+	EX_ST(STORE_BLK(%f0, %o4 + %g3))
+	FREG_MOVE_5(f22, f24, f26, f28, f30)
+	subcc		%g1, 64, %g1
+	add		%o4, 64, %o4
+	bne,pt		%xcc, 1b
+	 LOAD(prefetch, %o4 + 64, #one_read)
+	ba,pt		%xcc, 195f
+	 nop
+
+150:	sub		%o4, 32, %g2
+	FREG_LOAD_4(%g2, f0, f2, f4, f6)
+1:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
+	EX_LD(LOAD_BLK(%o4, %f16))
+	FREG_FROB(f0, f2, f4, f6, f16, f18, f20, f22, f24)
+	EX_ST(STORE_BLK(%f0, %o4 + %g3))
+	FREG_MOVE_4(f24, f26, f28, f30)
+	subcc		%g1, 64, %g1
+	add		%o4, 64, %o4
+	bne,pt		%xcc, 1b
+	 LOAD(prefetch, %o4 + 64, #one_read)
+	ba,pt		%xcc, 195f
+	 nop
+
+160:	sub		%o4, 24, %g2
+	FREG_LOAD_3(%g2, f0, f2, f4)
+1:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
+	EX_LD(LOAD_BLK(%o4, %f16))
+	FREG_FROB(f0, f2, f4, f16, f18, f20, f22, f24, f26)
+	EX_ST(STORE_BLK(%f0, %o4 + %g3))
+	FREG_MOVE_3(f26, f28, f30)
+	subcc		%g1, 64, %g1
+	add		%o4, 64, %o4
+	bne,pt		%xcc, 1b
+	 LOAD(prefetch, %o4 + 64, #one_read)
+	ba,pt		%xcc, 195f
+	 nop
+
+170:	sub		%o4, 16, %g2
+	FREG_LOAD_2(%g2, f0, f2)
+1:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
+	EX_LD(LOAD_BLK(%o4, %f16))
+	FREG_FROB(f0, f2, f16, f18, f20, f22, f24, f26, f28)
+	EX_ST(STORE_BLK(%f0, %o4 + %g3))
+	FREG_MOVE_2(f28, f30)
+	subcc		%g1, 64, %g1
+	add		%o4, 64, %o4
+	bne,pt		%xcc, 1b
+	 LOAD(prefetch, %o4 + 64, #one_read)
+	ba,pt		%xcc, 195f
+	 nop
+
+180:	sub		%o4, 8, %g2
+	FREG_LOAD_1(%g2, f0)
+1:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
+	EX_LD(LOAD_BLK(%o4, %f16))
+	FREG_FROB(f0, f16, f18, f20, f22, f24, f26, f28, f30)
+	EX_ST(STORE_BLK(%f0, %o4 + %g3))
+	FREG_MOVE_1(f30)
+	subcc		%g1, 64, %g1
+	add		%o4, 64, %o4
+	bne,pt		%xcc, 1b
+	 LOAD(prefetch, %o4 + 64, #one_read)
+	ba,pt		%xcc, 195f
+	 nop
+
+190:
+1:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
+	subcc		%g1, 64, %g1
+	EX_LD(LOAD_BLK(%o4, %f0))
+	EX_ST(STORE_BLK(%f0, %o4 + %g3))
+	add		%o4, 64, %o4
+	bne,pt		%xcc, 1b
+	 LOAD(prefetch, %o4 + 64, #one_read)
+
+195:
+	add		%o4, %g3, %o0
+	membar		#Sync
+
+	VISExitHalf
+
+	/* %o2 contains any final bytes still needed to be copied
+	 * over. If anything is left, we copy it one byte at a time.
+	 */
+	brz,pt		%o2, 85f
+	 sub		%o0, %o1, %o3
+	ba,a,pt		%XCC, 90f
+
+	.align		64
+75: /* 16 < len <= 64 */
+	bne,pn		%XCC, 75f
+	 sub		%o0, %o1, %o3
+
+72:
+	andn		%o2, 0xf, %o4
+	and		%o2, 0xf, %o2
+1:	subcc		%o4, 0x10, %o4
+	EX_LD(LOAD(ldx, %o1, %o5))
+	add		%o1, 0x08, %o1
+	EX_LD(LOAD(ldx, %o1, %g1))
+	sub		%o1, 0x08, %o1
+	EX_ST(STORE(stx, %o5, %o1 + %o3))
+	add		%o1, 0x8, %o1
+	EX_ST(STORE(stx, %g1, %o1 + %o3))
+	bgu,pt		%XCC, 1b
+	 add		%o1, 0x8, %o1
+73:	andcc		%o2, 0x8, %g0
+	be,pt		%XCC, 1f
+	 nop
+	sub		%o2, 0x8, %o2
+	EX_LD(LOAD(ldx, %o1, %o5))
+	EX_ST(STORE(stx, %o5, %o1 + %o3))
+	add		%o1, 0x8, %o1
+1:	andcc		%o2, 0x4, %g0
+	be,pt		%XCC, 1f
+	 nop
+	sub		%o2, 0x4, %o2
+	EX_LD(LOAD(lduw, %o1, %o5))
+	EX_ST(STORE(stw, %o5, %o1 + %o3))
+	add		%o1, 0x4, %o1
+1:	cmp		%o2, 0
+	be,pt		%XCC, 85f
+	 nop
+	ba,pt		%xcc, 90f
+	 nop
+
+75:
+	andcc		%o0, 0x7, %g1
+	sub		%g1, 0x8, %g1
+	be,pn		%icc, 2f
+	 sub		%g0, %g1, %g1
+	sub		%o2, %g1, %o2
+
+1:	subcc		%g1, 1, %g1
+	EX_LD(LOAD(ldub, %o1, %o5))
+	EX_ST(STORE(stb, %o5, %o1 + %o3))
+	bgu,pt		%icc, 1b
+	 add		%o1, 1, %o1
+
+2:	add		%o1, %o3, %o0
+	andcc		%o1, 0x7, %g1
+	bne,pt		%icc, 8f
+	 sll		%g1, 3, %g1
+
+	cmp		%o2, 16
+	bgeu,pt		%icc, 72b
+	 nop
+	ba,a,pt		%xcc, 73b
+
+8:	mov		64, %o3
+	andn		%o1, 0x7, %o1
+	EX_LD(LOAD(ldx, %o1, %g2))
+	sub		%o3, %g1, %o3
+	andn		%o2, 0x7, %o4
+	sllx		%g2, %g1, %g2
+1:	add		%o1, 0x8, %o1
+	EX_LD(LOAD(ldx, %o1, %g3))
+	subcc		%o4, 0x8, %o4
+	srlx		%g3, %o3, %o5
+	or		%o5, %g2, %o5
+	EX_ST(STORE(stx, %o5, %o0))
+	add		%o0, 0x8, %o0
+	bgu,pt		%icc, 1b
+	 sllx		%g3, %g1, %g2
+
+	srl		%g1, 3, %g1
+	andcc		%o2, 0x7, %o2
+	be,pn		%icc, 85f
+	 add		%o1, %g1, %o1
+	ba,pt		%xcc, 90f
+	 sub		%o0, %o1, %o3
+
+	.align		64
+80: /* 0 < len <= 16 */
+	andcc		%o3, 0x3, %g0
+	bne,pn		%XCC, 90f
+	 sub		%o0, %o1, %o3
+
+1:
+	subcc		%o2, 4, %o2
+	EX_LD(LOAD(lduw, %o1, %g1))
+	EX_ST(STORE(stw, %g1, %o1 + %o3))
+	bgu,pt		%XCC, 1b
+	 add		%o1, 4, %o1
+
+85:	retl
+	 mov		EX_RETVAL(GLOBAL_SPARE), %o0
+
+	.align		32
+90:
+	subcc		%o2, 1, %o2
+	EX_LD(LOAD(ldub, %o1, %g1))
+	EX_ST(STORE(stb, %g1, %o1 + %o3))
+	bgu,pt		%XCC, 90b
+	 add		%o1, 1, %o1
+	retl
+	 mov		EX_RETVAL(GLOBAL_SPARE), %o0
+
+	.size		FUNC_NAME, .-FUNC_NAME