AArch64: Implement optimized memcpy.

ports/ChangeLog.aarch64

@@ -1,3 +1,7 @@
+2013-01-17  Marcus Shawcroft  <marcus.shawcroft@linaro.org>
+
+        * sysdeps/aarch64/memcpy.S: New file.
+
 2013-01-17  Marcus Shawcroft  <marcus.shawcroft@linaro.org>
 
 	* sysdeps/aarch64/memset.S: New file.

ports/sysdeps/aarch64/memcpy.S

@@ -0,0 +1,176 @@
+/* Copyright (C) 2012-2013 Free Software Foundation, Inc.
+
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library.  If not, see
+   <http://www.gnu.org/licenses/>.  */
+
+/* Assumptions:
+ *
+ * ARMv8-a, AArch64
+ * Unaligned accesses
+ *
+ */
+
+#define dstin	x0
+#define src	x1
+#define count	x2
+#define tmp1	x3
+#define tmp1w	w3
+#define tmp2	x4
+#define tmp2w	w4
+#define tmp3	x5
+#define tmp3w	w5
+#define dst	x6
+
+#define A_l	x7
+#define A_h	x8
+#define B_l	x9
+#define B_h	x10
+#define C_l	x11
+#define C_h	x12
+#define D_l	x13
+#define D_h	x14
+
+#include <sysdep.h>
+
+ENTRY_ALIGN (memcpy, 6)
+
+	mov	dst, dstin
+	cmp	count, #64
+	b.ge	L(cpy_not_short)
+	cmp	count, #15
+	b.le	L(tail15tiny)
+
+	/* Deal with small copies quickly by dropping straight into the
+	 * exit block.  */
+L(tail63):
+	/* Copy up to 48 bytes of data.  At this point we only need the
+	 * bottom 6 bits of count to be accurate.  */
+	ands	tmp1, count, #0x30
+	b.eq	L(tail15)
+	add	dst, dst, tmp1
+	add	src, src, tmp1
+	cmp	tmp1w, #0x20
+	b.eq	1f
+	b.lt	2f
+	ldp	A_l, A_h, [src, #-48]
+	stp	A_l, A_h, [dst, #-48]
+1:
+	ldp	A_l, A_h, [src, #-32]
+	stp	A_l, A_h, [dst, #-32]
+2:
+	ldp	A_l, A_h, [src, #-16]
+	stp	A_l, A_h, [dst, #-16]
+
+L(tail15):
+	ands	count, count, #15
+	beq	1f
+	add	src, src, count
+	ldp	A_l, A_h, [src, #-16]
+	add	dst, dst, count
+	stp	A_l, A_h, [dst, #-16]
+1:
+	RET
+
+L(tail15tiny):
+	/* Copy up to 15 bytes of data.  Does not assume additional data
+	   being copied.  */
+	tbz	count, #3, 1f
+	ldr	tmp1, [src], #8
+	str	tmp1, [dst], #8
+1:
+	tbz	count, #2, 1f
+	ldr	tmp1w, [src], #4
+	str	tmp1w, [dst], #4
+1:
+	tbz	count, #1, 1f
+	ldrh	tmp1w, [src], #2
+	strh	tmp1w, [dst], #2
+1:
+	tbz	count, #0, 1f
+	ldrb	tmp1w, [src]
+	strb	tmp1w, [dst]
+1:
+	RET
+
+L(cpy_not_short):
+	/* We don't much care about the alignment of DST, but we want SRC
+	 * to be 128-bit (16 byte) aligned so that we don't cross cache line
+	 * boundaries on both loads and stores.  */
+	neg	tmp2, src
+	ands	tmp2, tmp2, #15		/* Bytes to reach alignment.  */
+	b.eq	2f
+	sub	count, count, tmp2
+	/* Copy more data than needed; it's faster than jumping
+	 * around copying sub-Quadword quantities.  We know that
+	 * it can't overrun.  */
+	ldp	A_l, A_h, [src]
+	add	src, src, tmp2
+	stp	A_l, A_h, [dst]
+	add	dst, dst, tmp2
+	/* There may be less than 63 bytes to go now.  */
+	cmp	count, #63
+	b.le	L(tail63)
+2:
+	subs	count, count, #128
+	b.ge	L(cpy_body_large)
+	/* Less than 128 bytes to copy, so handle 64 here and then jump
+	 * to the tail.  */
+	ldp	A_l, A_h, [src]
+	ldp	B_l, B_h, [src, #16]
+	ldp	C_l, C_h, [src, #32]
+	ldp	D_l, D_h, [src, #48]
+	stp	A_l, A_h, [dst]
+	stp	B_l, B_h, [dst, #16]
+	stp	C_l, C_h, [dst, #32]
+	stp	D_l, D_h, [dst, #48]
+	tst	count, #0x3f
+	add	src, src, #64
+	add	dst, dst, #64
+	b.ne	L(tail63)
+	RET
+
+	/* Critical loop.  Start at a new cache line boundary.  Assuming
+	 * 64 bytes per line this ensures the entire loop is in one line.  */
+	.p2align 6
+L(cpy_body_large):
+	/* There are at least 128 bytes to copy.  */
+	ldp	A_l, A_h, [src, #0]
+	sub	dst, dst, #16		/* Pre-bias.  */
+	ldp	B_l, B_h, [src, #16]
+	ldp	C_l, C_h, [src, #32]
+	ldp	D_l, D_h, [src, #48]!	/* src += 64 - Pre-bias.  */
+1:
+	stp	A_l, A_h, [dst, #16]
+	ldp	A_l, A_h, [src, #16]
+	stp	B_l, B_h, [dst, #32]
+	ldp	B_l, B_h, [src, #32]
+	stp	C_l, C_h, [dst, #48]
+	ldp	C_l, C_h, [src, #48]
+	stp	D_l, D_h, [dst, #64]!
+	ldp	D_l, D_h, [src, #64]!
+	subs	count, count, #64
+	b.ge	1b
+	stp	A_l, A_h, [dst, #16]
+	stp	B_l, B_h, [dst, #32]
+	stp	C_l, C_h, [dst, #48]
+	stp	D_l, D_h, [dst, #64]
+	add	src, src, #16
+	add	dst, dst, #64 + 16
+	tst	count, #0x3f
+	b.ne	L(tail63)
+	RET
+END (memcpy)
+libc_hidden_builtin_def (memcpy)