arm: Implement armv6 optimized string routines

The strcpy and strchr (and related) functions are four times faster
than the byte-by-byte default versions.

The strlen function is twice as fast for long strings and 50% faster
for short strings over the armv4 version.

1 files changed, 143 insertions, 0 deletions

diff --git a/ports/sysdeps/arm/armv6/strchr.S b/ports/sysdeps/arm/armv6/strchr.S
new file mode 100644
index 0000000000..c856283d53
--- /dev/null
+++ b/ports/sysdeps/arm/armv6/strchr.S
@@ -0,0 +1,143 @@
+/* strchr -- find the first instance of C in a nul-terminated string.
+   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 (strchr)
+	@ r0 = start of string
+	@ r1 = character to match
+	@ returns NULL for no match, or a pointer to the match
+	ldrb	r2, [r0]		@ load the first byte asap
+	uxtb	r1, r1
+
+	@ 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
+	rsb	r3, r3, #15		@ 16 - 1 peeled loop iteration
+	cmp	r2, r1			@ Found C?
+	it	ne
+	cmpne	r2, #0			@ Found EOS?
+	beq	99f
+
+	@ Loop until we find ...
+1:	ldrb	r2, [r0, #1]!
+	subs	r3, r3, #1		@ ... the aligment point
+	it	ne
+	cmpne	r2, r1			@ ... or the character
+	it	ne
+	cmpne	r2, #0			@ ... or EOS
+	bne	1b
+
+	@ Disambiguate the exit possibilites above
+	cmp	r2, r1			@ Found the character
+	it	ne
+	cmpne	r2, #0			@ Found EOS
+	beq	99f
+	add	r0, r0, #1
+
+	@ So now we're aligned.  Now we actually need a stack frame.
+	push	{ r4, r5, r6, r7 }
+	cfi_adjust_cfa_offset (16)
+	cfi_rel_offset (r4, 0)
+	cfi_rel_offset (r5, 4)
+	cfi_rel_offset (r6, 8)
+	cfi_rel_offset (r7, 12)
+
+	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 EOS or C, 8 bytes at a time.
+2:
+	@ 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.
+	uqsub8	r4, ip, r2		@ Find EOS
+	eor	r6, r2, r1		@ Convert C bytes to 0
+	uqsub8	r5, ip, r3
+	eor	r7, r3, r1
+	uqsub8	r6, ip, r6		@ Find C
+	pld	[r0, #128]		@ Prefetch 2 lines ahead
+	uqsub8	r7, ip, r7
+	orr	r4, r4, r6		@ Combine found for EOS and C
+	orr	r5, r5, r7
+	orrs	r6, r4, r5		@ Combine the two words
+	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 contents of the word containing the match.
+	@ Adjust r4 to the found bits for the word containing the match.
+	cmp	r4, #0
+	sub	r0, r0, #4
+	itte	eq
+	moveq	r4, r5
+	moveq	r2, r3
+	subne	r0, r0, #4
+
+	@ Find the bit-offset of the match within the word.
+#if defined(__ARMEL__)
+	@ For LE, swap the found word so clz searches from the little end.
+	rev	r4, r4
+#else
+	@ For BE, byte swap the word to make it easier to extract the byte.
+	rev	r2, r2
+#endif
+	@ We're counting 0x01 (not 0x80), so the bit offset is 7 too high.
+	clz	r3, r4
+	sub	r3, r3, #7
+	lsr	r2, r2, r3		@ Shift down found byte
+	uxtb	r1, r1			@ Undo replication of C
+	uxtb	r2, r2			@ Extract found byte
+	add	r0, r0, r3, lsr #3	@ Adjust the pointer to the found byte
+
+	pop	{ r4, r5, r6, r7 }
+	cfi_adjust_cfa_offset (-16)
+	cfi_restore (r4)
+	cfi_restore (r5)
+	cfi_restore (r6)
+	cfi_restore (r7)
+
+	@ Disambiguate between EOS and C.
+99:
+	cmp	r2, r1
+	it	ne
+	movne	r0, #0			@ Found EOS, return NULL
+	bx	lr
+
+END (strchr)
+
+weak_alias (strchr, index)
+libc_hidden_builtin_def (strchr)