/*-
 * Copyright (c) 1990 The Regents of the University of California.
 * All rights reserved.
 *
 * This code is derived from locore.s.
 *
 * 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.
 * 3. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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>

#if defined(LIBC_SCCS)
        RCSID("$NetBSD: bcopy.S,v 1.5 2014/03/22 19:16:34 jakllsch Exp $")
#endif

        /*
         * (ov)bcopy (src,dst,cnt)
         *  ws@tools.de     (Wolfgang Solfrank, TooLs GmbH) +49-228-985800
         *
         * Hacked about by dsl@netbsd.org
         */

#ifdef MEMCOPY
ENTRY(memcpy)
#define NO_OVERLAP
#else
#ifdef MEMMOVE
ENTRY(memmove)
#else
ENTRY(bcopy)
#endif
#endif
        movq    %rdx,%rcx
#if defined(MEMCOPY) || defined(MEMMOVE)
        movq    %rdi,%rax       /* must return destination address */
        mov     %rdi,%r11       /* for misaligned check */
#else
        mov     %rsi,%r11       /* for misaligned check */
        xchgq   %rdi,%rsi       /* bcopy() has arg order reversed */
#endif

#if !defined(NO_OVERLAP)
        movq    %rdi,%r8
        subq    %rsi,%r8
#endif

        shrq    $3,%rcx         /* count for copy by words */
        jz      8f              /* j if less than 8 bytes */

        lea     -8(%rdi,%rdx),%r9       /* target address of last 8 */
        mov     -8(%rsi,%rdx),%r10      /* get last word */
#if !defined(NO_OVERLAP)
        cmpq    %rdx,%r8        /* overlapping? */
        jb      10f
#endif

/*
 * Non-overlaping, copy forwards.
 * Newer Intel cpus (Nehalem) will do 16byte read/write transfers
 * if %ecx is more than 76.
 * AMD might do something similar some day.
 */
        and     $7,%r11         /* destination misaligned ? */
        jnz     2f
        rep
        movsq
        mov     %r10,(%r9)      /* write last word */
        ret

/*
 * Destination misaligned
 * AMD say it is better to align the destination (not the source).
 * This will also re-align copies if the source and dest are both
 * misaligned by the same amount)
 * (I think Nehalem will use its accelerated copy if the source
 * and destination have the same alignment.)
 */
2:
        lea     -9(%r11,%rdx),%rcx      /* post re-alignment count */
        neg     %r11                    /* now -1 .. -7 */
        mov     (%rsi),%rdx             /* get first word */
        mov     %rdi,%r8                /* target for first word */
        lea     8(%rsi,%r11),%rsi
        lea     8(%rdi,%r11),%rdi
        shr     $3,%rcx
        rep
        movsq
        mov     %rdx,(%r8)              /* write first word */
        mov     %r10,(%r9)              /* write last word */
        ret

#if !defined(NO_OVERLAP)
/* Must copy backwards.
 * Reverse copy is probably easy to code faster than 'rep movds'
 * since that requires (IIRC) an extra clock every 3 iterations (AMD).
 * However I don't suppose anything cares that much!
 * The big cost is the std/cld pair - reputedly 50+ cycles on Netburst P4.
 * The copy is aligned with the buffer start (more likely to
 * be a multiple of 8 than the end).
 */
10:
        lea     -8(%rsi,%rcx,8),%rsi
        lea     -8(%rdi,%rcx,8),%rdi
        std
        rep
        movsq
        cld
        mov     %r10,(%r9)      /* write last bytes */
        ret
#endif

/* Less than 8 bytes to copy, copy by bytes */
/* Intel Nehalem optimise 'rep movsb' for <= 7 bytes (9-15 clocks).
 * For longer transfers it is 50+ !
 */
8:      mov     %rdx,%rcx

#if !defined(NO_OVERLAP)
        cmpq    %rdx,%r8        /* overlapping? */
        jb      81f
#endif

        /* nope, copy forwards. */
        rep
        movsb
        ret

#if !defined(NO_OVERLAP)
/* Must copy backwards */
81:
        lea     -1(%rsi,%rcx),%rsi
        lea     -1(%rdi,%rcx),%rdi
        std
        rep
        movsb
        cld
        ret
#endif

#ifdef MEMCOPY
END(memcpy)
#else
#ifdef MEMMOVE
END(memmove)
#else
END(bcopy)
#endif
#endif