AArch64: Implement optimized strcmp

ports/ChangeLog.aarch64

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

ports/sysdeps/aarch64/strcmp.S

@@ -0,0 +1,155 @@
+/* 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
+ */
+
+#include <sysdep.h>
+
+#define REP8_01 0x0101010101010101
+#define REP8_7f 0x7f7f7f7f7f7f7f7f
+#define REP8_80 0x8080808080808080
+
+/* Parameters and result.  */
+#define src1		x0
+#define src2		x1
+#define result		x0
+
+/* Internal variables.  */
+#define data1		x2
+#define data1w		w2
+#define data2		x3
+#define data2w		w3
+#define has_nul		x4
+#define diff		x5
+#define syndrome	x6
+#define tmp1		x7
+#define tmp2		x8
+#define tmp3		x9
+#define zeroones	x10
+#define pos		x11
+
+	/* Start of performance-critical section  -- one 64B cache line.  */
+ENTRY_ALIGN(strcmp, 6)
+
+	eor	tmp1, src1, src2
+	mov	zeroones, #REP8_01
+	tst	tmp1, #7
+	b.ne	L(misaligned8)
+	ands	tmp1, src1, #7
+	b.ne	L(mutual_align)
+	/* NUL detection works on the principle that (X - 1) & (~X) & 0x80
+	   (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and
+	   can be done in parallel across the entire word.  */
+L(loop_aligned):
+	ldr	data1, [src1], #8
+	ldr	data2, [src2], #8
+L(start_realigned):
+	sub	tmp1, data1, zeroones
+	orr	tmp2, data1, #REP8_7f
+	eor	diff, data1, data2	/* Non-zero if differences found.  */
+	bic	has_nul, tmp1, tmp2	/* Non-zero if NUL terminator.  */
+	orr	syndrome, diff, has_nul
+	cbz	syndrome, L(loop_aligned)
+	/* End of performance-critical section  -- one 64B cache line.  */
+
+#ifndef	__AARCH64EB__
+	rev	syndrome, syndrome
+	rev	data1, data1
+	/* The MS-non-zero bit of the syndrome marks either the first bit
+	   that is different, or the top bit of the first zero byte.
+	   Shifting left now will bring the critical information into the
+	   top bits.  */
+	clz	pos, syndrome
+	rev	data2, data2
+	lsl	data1, data1, pos
+	lsl	data2, data2, pos
+	/* But we need to zero-extend (char is unsigned) the value and then
+	   perform a signed 32-bit subtraction.  */
+	lsr	data1, data1, #56
+	sub	result, data1, data2, lsr #56
+	RET
+#else
+	/* For big-endian we cannot use the trick with the syndrome value
+	   as carry-propagation can corrupt the upper bits if the trailing
+	   bytes in the string contain 0x01.  */
+	/* However, if there is no NUL byte in the dword, we can generate
+	   the result directly.  We can't just subtract the bytes as the
+	   MSB might be significant.  */
+	cbnz	has_nul, 1f
+	cmp	data1, data2
+	cset	result, ne
+	cneg	result, result, lo
+	RET
+1:
+	/* Re-compute the NUL-byte detection, using a byte-reversed value.  */
+	rev	tmp3, data1
+	sub	tmp1, tmp3, zeroones
+	orr	tmp2, tmp3, #REP8_7f
+	bic	has_nul, tmp1, tmp2
+	rev	has_nul, has_nul
+	orr	syndrome, diff, has_nul
+	clz	pos, syndrome
+	/* The MS-non-zero bit of the syndrome marks either the first bit
+	   that is different, or the top bit of the first zero byte.
+	   Shifting left now will bring the critical information into the
+	   top bits.  */
+	lsl	data1, data1, pos
+	lsl	data2, data2, pos
+	/* But we need to zero-extend (char is unsigned) the value and then
+	   perform a signed 32-bit subtraction.  */
+	lsr	data1, data1, #56
+	sub	result, data1, data2, lsr #56
+	RET
+#endif
+
+L(mutual_align):
+	/* Sources are mutually aligned, but are not currently at an
+	   alignment boundary.  Round down the addresses and then mask off
+	   the bytes that preceed the start point.  */
+	bic	src1, src1, #7
+	bic	src2, src2, #7
+	lsl	tmp1, tmp1, #3		/* Bytes beyond alignment -> bits.  */
+	ldr	data1, [src1], #8
+	neg	tmp1, tmp1		/* Bits to alignment -64.  */
+	ldr	data2, [src2], #8
+	mov	tmp2, #~0
+#ifdef __AARCH64EB__
+	/* Big-endian.  Early bytes are at MSB.  */
+	lsl	tmp2, tmp2, tmp1	/* Shift (tmp1 & 63).  */
+#else
+	/* Little-endian.  Early bytes are at LSB.  */
+	lsr	tmp2, tmp2, tmp1	/* Shift (tmp1 & 63).  */
+#endif
+	orr	data1, data1, tmp2
+	orr	data2, data2, tmp2
+	b	L(start_realigned)
+
+L(misaligned8):
+	/* We can do better than this.  */
+	ldrb	data1w, [src1], #1
+	ldrb	data2w, [src2], #1
+	cmp	data1w, #1
+	ccmp	data1w, data2w, #0, cs	/* NZCV = 0b0000.  */
+	b.eq	L(misaligned8)
+	sub	result, data1, data2
+	RET
+END(strcmp)
+libc_hidden_builtin_def (strcmp)