[PATCH 5/5] arm: Implement armv6 optimized rawmemchr

[Richard Henderson <rth at twiddle dot net>]

---
 ports/sysdeps/arm/armv6/rawmemchr.S | 105 ++++++++++++++++++++++++++++++++++++
 1 file changed, 105 insertions(+)
 create mode 100644 ports/sysdeps/arm/armv6/rawmemchr.S

diff --git a/ports/sysdeps/arm/armv6/rawmemchr.S b/ports/sysdeps/arm/armv6/rawmemchr.S
new file mode 100644
index 0000000..7877bcf
--- /dev/null
+++ b/ports/sysdeps/arm/armv6/rawmemchr.S
@@ -0,0 +1,105 @@
+/* rawmemchr -- find a byte within an unsized memory block.
+   Copyright (C) 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/>.  */
+
+#include <sysdep.h>
+
+	.syntax unified
+	.text
+
+ENTRY (__rawmemchr)
+	@ r0 = start of string
+	@ r1 = character to match
+	@ returns a pointer to the match, which must be present.
+	ldrb	r2, [r0]		@ load first byte asap
+
+	@ To cater to long strings, we want to search through a few
+	@ characters until we reach an aligned pointer.  To cater to
+	@ small strings, we don't want to start doing word operations
+	@ immediately.  The compromise is a maximum of 16 bytes less
+	@ whatever is required to end with an aligned pointer.
+	@ r3 = number of characters to search in alignment loop
+	and	r3, r0, #7
+	uxtb	r1, r1
+	rsb	r3, r3, #15		@ 16 - 1 peeled loop iteration
+	cmp	r2, r1
+	it	eq
+	bxeq	lr
+
+	@ Loop until we find ...
+1:	ldrb	r2, [r0, #1]!
+	subs	r3, r3, #1		@ ... the alignment point
+	it	ne
+	cmpne	r2, r1			@ ... or C
+	bne	1b
+
+	@ Disambiguate the exit possibilites above
+	cmp	r2, r1			@ Found C
+	it	eq
+	bxeq	lr
+	add	r0, r0, #1
+
+	@ So now we're aligned.
+	ldrd	r2, r3, [r0], #8
+	orr	r1, r1, r1, lsl #8	@ Replicate C to all bytes
+#ifdef ARCH_HAS_T2
+	movw	ip, #0x0101
+	pld	[r0, #64]
+	movt	ip, #0x0101
+#else
+	ldr	ip, =0x01010101
+	pld	[r0, #64]
+#endif
+	orr	r1, r1, r1, lsl #16
+
+	@ Loop searching for C, 8 bytes at a time.
+	@ Subtracting (unsigned saturating) from 1 means result of 1 for
+	@ any byte that was originally zero and 0 otherwise.  Therefore
+	@ we consider the lsb of each byte the "found" bit.
+2:	eor	r2, r2, r1		@ Convert C bytes to 0
+	eor	r3, r3, r1
+	uqsub8	r2, ip, r2		@ Find C
+	uqsub8	r3, ip, r3
+	pld	[r0, #128]
+	orrs	r3, r3, r2		@ Test both words for found
+	it	eq
+	ldrdeq	r2, r3, [r0], #8
+	beq	2b
+
+	@ Found something.  Disambiguate between first and second words.
+	@ Adjust r0 to point to the word containing the match.
+	@ Adjust r2 to the found bits for the word containing the match.
+	cmp	r2, #0
+	sub	r0, r0, #4
+	ite	eq
+	moveq	r2, r3
+	subne	r0, r0, #4
+
+	@ Find the bit-offset of the match within the word.  Note that the
+	@ bit result from clz will be 7 higher than "true", but we'll
+	@ immediately discard those bits converting to a byte offset.
+#ifdef __ARMEL__
+	rev	r2, r2			@ For LE, count from the little end
+#endif
+	clz	r2, r2
+	add	r0, r0, r2, lsr #3	@ Adjust the pointer to the found byte
+	bx	lr
+
+END (__rawmemchr)
+
+weak_alias (__rawmemchr, rawmemchr)
+libc_hidden_def (__rawmemchr)
-- 
1.8.1.2