sysdeps/sh/memcpy.S

   1 /* Copyright (C) 1999-2019 Free Software Foundation, Inc.
   2    This file is part of the GNU C Library.
   3    Contributed by Kazumoto Kojima <kkojima@rr.iij4u.or.jp>
   4    Optimized by Toshiyasu Morita <toshiyasu.morita@hsa.hitachi.com>
   5
   6    The GNU C Library is free software; you can redistribute it and/or
   7    modify it under the terms of the GNU Lesser General Public
   8    License as published by the Free Software Foundation; either
   9    version 2.1 of the License, or (at your option) any later version.
  10
  11    The GNU C Library is distributed in the hope that it will be useful,
  12    but WITHOUT ANY WARRANTY; without even the implied warranty of
  13    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14    Lesser General Public License for more details.
  15
  16    You should have received a copy of the GNU Lesser General Public
  17    License along with the GNU C Library; if not, see
  18    <https://www.gnu.org/licenses/>.  */
  19
  20 #include <sysdep.h>
  21
  22 /* void *memcpy(void *dst, const void *src, size_t n);
  23     No overlap between the memory of DST and of SRC are assumed.  */
  24
  25 ENTRY(memcpy)
  26         mov     r4,r3           /* Save destination.  */
  27
  28         /* If less than 11 bytes, just do a byte copy.  */
  29         mov     #11,r0
  30         cmp/gt  r6,r0
  31         bt      L_byteloop_init
  32
  33         /* Check if we need to word-align source.  */
  34         mov     r5,r0
  35         tst     #1,r0
  36         bt      L_wordalign
  37
  38         mov.b   @r0+,r1         /* Copy one byte.  */
  39         add     #-1,r6
  40         mov.b   r1,@r4
  41         add     #1,r4
  42
  43         .balignw 4,0x0009
  44 L_wordalign:
  45         /* Check if we need to longword-align source.  */
  46         tst     #2,r0
  47         bt      L_copy
  48
  49         mov.w   @r0+,r1         /* Copy one word.  */
  50         add     #-2,r6
  51 #ifdef __BIG_ENDIAN__
  52         add     #1,r4
  53         mov.b   r1,@r4
  54         shlr8   r1
  55         mov.b   r1,@-r4
  56         add     #2,r4
  57 #else
  58         mov.b   r1,@r4
  59         add     #1,r4
  60         shlr8   r1
  61         mov.b   r1,@r4
  62         add     #1,r4
  63 #endif
  64 L_copy:
  65         mov     r0,r5
  66
  67         /* Calculate the correct routine to handle the destination
  68            alignment and simultaneously calculate the loop counts for
  69            both the 2 word copy loop and byte copy loop.  */
  70         mova    L_jumptable,r0
  71         mov     r0,r1
  72         mov     r4,r0
  73         mov     r6,r7
  74         and     #3,r0
  75         shlr2   r7
  76         shll    r0
  77         shlr    r7
  78         mov.w   @(r0,r1),r2
  79         mov     #7,r0
  80         braf    r2
  81         and     r0,r6
  82 L_base:
  83
  84         .balign 4
  85 L_jumptable:
  86         .word   L_copydest0 - L_base
  87         .word   L_copydest1_or_3 - L_base
  88         .word   L_copydest2 - L_base
  89         .word   L_copydest1_or_3 - L_base
  90
  91         .balign 4
  92         /* Copy routine for (dest mod 4) == 1 or == 3.  */
  93 L_copydest1_or_3:
  94         add     #-1,r4
  95         .balignw 4,0x0009
  96 L_copydest1_or_3_loop:
  97         mov.l   @r5+,r0         /* Read first longword.  */
  98         dt      r7
  99         mov.l   @r5+,r1         /* Read second longword.  */
 100 #ifdef __BIG_ENDIAN__
 101         /* Write first longword as byte, word, byte.  */
 102         mov.b   r0,@(4,r4)
 103         shlr8   r0
 104         mov.w   r0,@(2,r4)
 105         shlr16  r0
 106         mov.b   r0,@(1,r4)
 107         mov     r1,r0
 108         /* Write second longword as byte, word, byte.  */
 109         mov.b   r0,@(8,r4)
 110         shlr8   r0
 111         mov.w   r0,@(6,r4)
 112         shlr16  r0
 113         mov.b   r0,@(5,r4)
 114 #else
 115         /* Write first longword as byte, word, byte.  */
 116         mov.b   r0,@(1,r4)
 117         shlr8   r0
 118         mov.w   r0,@(2,r4)
 119         shlr16  r0
 120         mov.b   r0,@(4,r4)
 121         mov     r1,r0
 122         /* Write second longword as byte, word, byte.  */
 123         mov.b   r0,@(5,r4)
 124         shlr8   r0
 125         mov.w   r0,@(6,r4)
 126         shlr16  r0
 127         mov.b   r0,@(8,r4)
 128 #endif
 129         bf/s    L_copydest1_or_3_loop
 130         add     #8,r4
 131
 132         bra     L_byteloop_init
 133         add     #1,r4
 134
 135         .balign 4
 136         /* Copy routine for (dest mod 4) == 2.  */
 137 L_copydest2:
 138 L_copydest2_loop:
 139         mov.l   @r5+,r0
 140         dt      r7
 141         mov.l   @r5+,r1
 142 #ifdef __BIG_ENDIAN__
 143         mov.w   r0,@(2,r4)
 144         shlr16  r0
 145         mov.w   r0,@r4
 146         mov     r1,r0
 147         mov.w   r0,@(6,r4)
 148         shlr16  r0
 149         mov.w   r0,@(4,r4)
 150 #else
 151         mov.w   r0,@r4
 152         shlr16  r0
 153         mov.w   r0,@(2,r4)
 154         mov     r1,r0
 155         mov.w   r0,@(4,r4)
 156         shlr16  r0
 157         mov.w   r0,@(6,r4)
 158 #endif
 159         bf/s    L_copydest2_loop
 160         add     #8,r4
 161
 162         bra     L_byteloop_init
 163         nop
 164
 165         .balign 4
 166         /* Copy routine for (dest mod 4) == 0.  */
 167 L_copydest0:
 168         add     #-8,r4
 169         .balignw 4,0x0009
 170 L_copydest0_loop:
 171         mov.l   @r5+,r0
 172         dt      r7
 173         mov.l   @r5+,r1
 174         add     #8,r4
 175         mov.l   r0,@r4
 176         bf/s    L_copydest0_loop
 177         mov.l   r1,@(4,r4)
 178
 179         add     #8,r4           /* Fall through.  */
 180
 181 L_byteloop_init:
 182         tst     r6,r6
 183         bt      L_exit
 184
 185         .balignw 4,0x0009
 186         /* Copy remaining bytes.  */
 187 L_byteloop:
 188         mov.b   @r5+,r0
 189         dt      r6
 190         mov.b   r0,@r4
 191         bf/s    L_byteloop
 192         add     #1,r4
 193
 194 L_exit:
 195         rts
 196         mov     r3,r0           /* Return destination.  */
 197 END(memcpy)
 198 libc_hidden_builtin_def (memcpy)