root/common/lib/libc/arch/aarch64/string/memcmp.S
/* $NetBSD: memcmp.S,v 1.3 2018/07/09 06:07:06 ryo Exp $ */

/*-
 * Copyright (c) 2014 The NetBSD Foundation, Inc.
 * All rights reserved.
 *
 * This code is derived from software contributed to The NetBSD Foundation
 * by Matt Thomas of 3am Software Foundry.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 */

#include <machine/asm.h>

RCSID("$NetBSD: memcmp.S,v 1.3 2018/07/09 06:07:06 ryo Exp $")

ENTRY(memcmp)
        mov     x9, x0
        mov     x10, x1
        mov     x0, xzr
        cbz     x2, .Lmemcmp_ret
#ifdef _KERNEL
        cmp     x2, #6
        b.eq    .Lmemcmp_6bytes
#endif
        cmp     x2, #8
        b.ls    .Lmemcmp_lessthan8

        ands    x3, x9, #7
        b.eq    .Lmemcmp_dword_loop

/*
 * The src1 address is not dword aligned.
 */
        add     x2, x2, x3              /* add unalignment to length */
        sub     x2, x2, #8              /* now subtract a dword */

        sub     x9, x9, x3              /* dword align src1 */

        ldr     x6, [x10], #8           /* load dword from src2 */
        sub     x10, x10, x3            /* src2 -= x3 */
        lsl     x3, x3, #3              /* convert bytes to bits */
        ldr     x4, [x9], #8            /* load dword from src1 */
#ifdef __AARCH64EB__
        lsl     x4, x4, x3              /* discard leading bytes from data1 */
        lsr     x6, x6, x3              /* discard leading bytes from data2 */
        lsl     x6, x6, x3              /* get back bit position */
#else
        lsr     x4, x4, x3              /* discard leading bytes from data1 */
        lsl     x6, x6, x3              /* discard leading bytes from data2 */
        lsr     x6, x6, x3              /* get back bit position */
#endif
        subs    x0, x4, x6              /* compare data */
        b.ne    .Lmemcmp_last_compare   /* difference.  find it */

.Lmemcmp_dword_loop:
        subs    x2, x2, #8
        b.mi    .Lmemcmp_finish_dword
        ldr     x4, [x9], #8
        ldr     x6, [x10], #8
        subs    x0, x4, x6
        b.eq    .Lmemcmp_dword_loop     /* no difference.  go to loop */
        b       .Lmemcmp_last_compare   /* go find the difference. */

.Lmemcmp_finish_dword:
        /*
         * we might have gotten here with nothing left.  If so, just bail.
         */
        tst     x2, #7
        b.eq    .Lmemcmp_ret
        mov     x4, xzr
        mov     x6, xzr
        /*
         *
         */
        tbz     x2, #2, .Lmemcmp_finish_word
        ldr     w4, [x9], #4
        ldr     w6, [x10], #4
#ifdef __AARCH64EB__
        lsl     x4, x4, #32             /* move to MSW */
        lsl     x6, x6, #32             /* move to MSW */
#endif

.Lmemcmp_finish_word:
        tbz     x2, #1, .Lmemcmp_finish_hword
        ldrh    w5, [x9], #2
        ldrh    w7, [x10], #2
#ifdef __AARCH64EB__
        orr     x4, x4, x5, lsl #16
        orr     x6, x6, x7, lsl #16
#else
        orr     x4, x4, x5, lsl #32
        orr     x6, x6, x7, lsl #32
#endif

.Lmemcmp_finish_hword:
        tbz     x2, #0, .Lmemcmp_last_compare0

        ldrb    w5, [x9]
        ldrb    w7, [x10]
#ifdef __AARCH64EB__
        orr     x4, x4, x5, lsl #8
        orr     x6, x6, x7, lsl #8
#else
        orr     x4, x4, x5, lsl #48
        orr     x6, x6, x7, lsl #48
#endif
        b       .Lmemcmp_last_compare0  /* go find the difference. */

/*
 * D
 */
.Lmemcmp_lessthan8:
        sub     x2, x2, #1
1:      ldrb    w4, [x9], #1
        ldrb    w5, [x10], #1
        subs    x2, x2, #1
        ccmp    x4, x5, #0, cs
        b.eq    1b
        sub     x0, x4, x5

.Lmemcmp_ret:
        ret

#ifdef _KERNEL
.Lmemcmp_6bytes:
        ldr     w4, [x9], #4
        ldrh    w5, [x9]
#if __AARCH64EB__
        orr     x4, x4, x5, lsl #48
        rev     x4, x4
#else
        orr     x4, x4, x5, lsl #32
#endif
        ldr     w6, [x10], #4
        ldrh    w7, [x10]
#if __AARCH64EB__
        orr     x6, x6, x7, lsl #48
        rev     x6, x6
#else
        orr     x6, x6, x7, lsl #32
#endif
#endif /* _KERNEL */

/*
 * We have loaded the final bytes in x4 and x6 in host-endian.  Now we have
 * to figure what the difference is (if any).  First we subtract.  Any bytes
 * that are the same will be 0. So to find the first non-zero byte we byterev
 * and then use clz to find that byte.
 * We mask the location to get the start of the byte.  We shift both
 * data dwords left to remove the equal part.  Then we shift right to discard
 * the trailing bytes.  Then we subtract and return.
 */
.Lmemcmp_last_compare0:
        subs    x0, x4, x6
        b.eq    .Lmemcmp_ret
.Lmemcmp_last_compare:
#if __AARCH64EB__
        clz     x1, x0          /* find first non-zero byte */
        rev     x0, x0
#else
        rev     x1, x0
        clz     x1, x1          /* find first non-zero byte */
#endif
        bfi     x1, xzr, #0, #3 /* make it byte aligned */
        lsr     x1, x0, x1      /* shift to LSB */
#if __AARCH64EL__
        rev     x4, x4          /* byte reverse */
        rev     x6, x6          /* byte reverse */
#endif
        subs    x0, x4, x6
        csetm   x0, cc          /* set mask bits as sign */
        bfm     x0, x1, #0, #7  /* extend with sign bit */
        ret
END(memcmp)