Subversion Repositories openrisc_me

/*******************************************************************************
 * 
 * Copyright (c) 1993 Intel Corporation
 * 
 * Intel hereby grants you permission to copy, modify, and distribute this
 * software and its documentation.  Intel grants this permission provided
 * that the above copyright notice appears in all copies and that both the
 * copyright notice and this permission notice appear in supporting
 * documentation.  In addition, Intel grants this permission provided that
 * you prominently mark as "not part of the original" any modifications
 * made to this software or documentation, and that the name of Intel
 * Corporation not be used in advertising or publicity pertaining to
 * distribution of the software or the documentation without specific,
 * written prior permission.
 * 
 * Intel Corporation provides this AS IS, WITHOUT ANY WARRANTY, EXPRESS OR
 * IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY WARRANTY OF MERCHANTABILITY
 * OR FITNESS FOR A PARTICULAR PURPOSE.  Intel makes no guarantee or
 * representations regarding the use of, or the results of the use of,
 * the software and documentation in terms of correctness, accuracy,
 * reliability, currentness, or otherwise; and you rely on the software,
 * documentation and results solely at your own risk.
 *
 * IN NO EVENT SHALL INTEL BE LIABLE FOR ANY LOSS OF USE, LOSS OF BUSINESS,
 * LOSS OF PROFITS, INDIRECT, INCIDENTAL, SPECIAL OR CONSEQUENTIAL DAMAGES
 * OF ANY KIND.  IN NO EVENT SHALL INTEL'S TOTAL LIABILITY EXCEED THE SUM
 * PAID TO INTEL FOR THE PRODUCT LICENSED HEREUNDER.
 * 
 ******************************************************************************/

        .file "mccpy_ca.s"
#ifdef  __PIC
        .pic
#endif
#ifdef  __PID
        .pid
#endif
/*
 * (c) copyright 1989,1993 Intel Corp., all rights reserved
 */

/*
        procedure memccpy  (optimized assembler version for the 80960CA)

        dest_addr = memccpy (dest_addr, src_addr, char, len)

        copy len bytes pointed to by src_addr to the space pointed to by 
        dest_addr, stopping if char is copied.  If char is copied,
        return address of byte after char in dest string;  else null.

        Undefined behavior will occur if the end of the source array is in 
        the last two words of the program's allocated memory space.  This 
        is so because the routine fetches ahead.  Disallowing the fetch 
        ahead would impose a severe performance penalty.

        Undefined behavior will also occur if the source and destination
        strings overlap.


        This program handles five cases:

        1) both arguments start on a word boundary
        2) neither are word aligned, but they are offset by the same amount
        3) source is word aligned, destination is not
        4) destination is word aligned, source is not
        5) neither is word aligned, and they are offset by differing amounts

        At the time of this writing, only g0 thru g7 and g13 are available 
        for use in this leafproc;  other registers would have to be saved and
        restored.  These nine registers, plus tricky use of g14 are sufficient
        to implement the routine.
*/

#if __i960_BIG_ENDIAN__
#define MSW g6
#define LSW g7
#else
#define LSW g6
#define MSW g7
#endif

        .globl  _memccpy
        .leafproc       _memccpy, __memccpy
        .align  2
_memccpy:
#ifndef __PIC
        lda     Lrett,g14
#else
        lda     Lrett-(.+8)(ip),g14
#endif
__memccpy:
        notand  g1,3,g5         # extract word addr of start of src
         lda    (g14),g13       # preserve return address
        cmpibge.f 0,g3,Lexit_char_not_found # Lexit if # of bytes to move is <= 0
        cmpo    g5,g1           # check alignment of src
         ld     (g5),LSW        # fetch word containing at least first byte
        notand  g0,3,g4         # extract word addr of start of dest
         lda    4(g5),g5        # advance src word addr
        shlo    24,g2,g2        # reduce char to single byte
         bne.f  Lcase_245       # branch if src is NOT word aligned

Lcase_13:
        cmpobe.t g0,g4,Lcase_1_setup    # branch if dest word aligned

Lcase_3:                                # src is word aligned; dest is not
        mov     LSW,MSW         # make copy of first word of src
        addo    4,g4,g1         # move dest word ptr to first word boundary
         lda    32,g14          # initialize shift count to zero

Lcase_25:
Lcase_3_cloop_at_start:         # character copying loop for start of dest str
        cmpdeci 0,g3,g3         # is max_bytes exhausted?
#if __i960_BIG_ENDIAN__
         lda    -8(g14),g14     # augment the shift counter
#else
         lda    8(g14),g14      # augment the shift counter
#endif
        be.f    Lexit_char_not_found    # Lexit if max_bytes is exhausted
#if __i960_BIG_ENDIAN__
        rotate  8,MSW,MSW       # move next byte into position for extraction
#endif
        shlo    24,MSW,g4
         stob   MSW,(g0)        # store the byte in dest
        cmpo    g4,g2
         lda    1(g0),g0        # post-increment dest ptr
#if ! __i960_BIG_ENDIAN__
        shro    8,MSW,MSW       # move next byte into position for extraction
#endif
         be.f   Lexit_char_found        # Lexit if char found
        cmpobne.t g1,g0,Lcase_3_cloop_at_start # branch if reached word boundary

        ld      (g5),MSW        # fetch msw of operand for double shift

Lcase_4:
        shro    8,g2,g4
        or      g4,g2,g1
        shro    16,g1,g4
        or      g4,g1,g4

#if __i960_BIG_ENDIAN__
        cmpobne 0,g14,Lcase_3_wloop

Lcase_3_wloop2:
        cmpi    g3,4            # less than four bytes to move?
         lda    4(g5),g5        # post-increment src word addr
        mov     LSW,g1          # extract 4 bytes of src
         bl.f   Lcase_13_cloop_setup    # branch if < four bytes left to move
        scanbyte g4,g1          # branch if word has char in it
         bo.f   Lcase_13_cloop_setup
        mov     MSW,LSW         # move msw to lsw
         ld     (g5),MSW        # pre-fetch msw of operand for double shift
        subi    4,g3,g3         # decrease max_byte count by the 4 bytes moved
         st     g1,(g0)         # store 4 bytes to dest
        addo    4,g0,g0         # post-increment dest ptr
         b      Lcase_3_wloop2
#endif
        
Lcase_3_wloop:
        cmpi    g3,4            # less than four bytes to move?
         lda    4(g5),g5        # post-increment src word addr
        eshro   g14,g6,g1       # extract 4 bytes of src
         bl.f   Lcase_13_cloop_setup    # branch if < four bytes left to move
        scanbyte g4,g1          # branch if word has char in it
         bo.f   Lcase_13_cloop_setup
        mov     MSW,LSW         # move msw to lsw
         ld     (g5),MSW        # pre-fetch msw of operand for double shift
        subi    4,g3,g3         # decrease max_byte count by the 4 bytes moved
         st     g1,(g0)         # store 4 bytes to dest
        addo    4,g0,g0         # post-increment dest ptr
         b      Lcase_3_wloop

Lcase_1_setup:
        subo    4,g0,g0         # store is pre-incrementing;  back up dest addr
        shro    8,g2,g4
        or      g4,g2,MSW
        shro    16,MSW,g4
        or      g4,MSW,g4
         b      Lcase_1
Lcase_1_wloop:                  # word copying loop
        subi    4,g3,g3         # decrease max_byte count by the 4 bytes moved
         ld     (g5),LSW        # pre-fetch next word of src
        addo    4,g5,g5         # post-increment src addr
         st     g1,(g0)         # store word in dest string
Lcase_1:                                # src and dest are word aligned
        cmpi    g3,4            # check for fewer than four bytes to move
        addo    4,g0,g0         # pre-increment dest addr
         lda    (LSW),g1        # keep a copy of the src word
         bl.f   Lcase_13_cloop_setup    # branch if less than four bytes to copy
        scanbyte LSW,g4         # branch if char is not in foursome
        bno.t   Lcase_1_wloop
Lcase_13_cloop_setup:
        cmpibe.f 0,g3,Lexit_char_not_found      # Lexit if max_bytes is exhausted

Lcase_1_cloop:
#if __i960_BIG_ENDIAN__
        rotate  8,g1,g1         # move next byte into position for extraction
#endif
        shlo    24,g1,g4
         stob   g1,(g0)         # store the byte in dest
        cmpo    g4,g2
         lda    1(g0),g0        # post-increment dest byte addr
        subi    1,g3,g3
         be.f   Lexit_char_found        # Lexit if char reached
        cmpi    0,g3
#if ! __i960_BIG_ENDIAN__
        shro    8,g1,g1         # move next byte into position for extraction
#endif
         bne.t  Lcase_1_cloop   # continue if len not exhausted

Lexit_char_not_found:
        mov     0,g0
Lexit_char_found:
        lda     0,g14
        bx      (g13)           # g0 = dest array address;  g14 = 0
Lrett:
        ret


Lcase_245:
        cmpo    g0,g4           # check alignment of dest
         ld     (g5),MSW        # pre-fetch second half
        and     3,g1,g1         # compute shift count
        shlo    3,g1,g14        
#if __i960_BIG_ENDIAN__
        subo    g14,0,g14       # adjust shift count for big endian
#endif
         be.t   Lcase_4         # branch if dest is word aligned
        or      g4,g1,g1        # is src earlier in word, later, or sync w/ dst
        cmpo    g0,g1           # < indicates first word of dest has more bytes
                                /* than first word of source. */
        eshro   g14,g6,g4       # extract four bytes
         lda    4(g0),g1        # move dest word addr to first word boundary
#if __i960_BIG_ENDIAN__
         bge.f  1f
#else
         bg.f   1f
#endif
        mov     MSW,LSW
         lda    4(g5),g5        # move src word addr to second word boundary
1:
        notand  g1,3,g1
         mov    g4,MSW
         b      Lcase_25

/* end of memccpy */