sysdeps/sparc/sparc32/strcpy.S

   1 /* Copy SRC to DEST returning DEST.
   2    For SPARC v7.
   3    Copyright (C) 1996-2021 Free Software Foundation, Inc.
   4    This file is part of the GNU C Library.
   5    Contributed by Jakub Jelinek <jj@ultra.linux.cz>.
   6
   7    The GNU C Library is free software; you can redistribute it and/or
   8    modify it under the terms of the GNU Lesser General Public
   9    License as published by the Free Software Foundation; either
  10    version 2.1 of the License, or (at your option) any later version.
  11
  12    The GNU C Library is distributed in the hope that it will be useful,
  13    but WITHOUT ANY WARRANTY; without even the implied warranty of
  14    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15    Lesser General Public License for more details.
  16
  17    You should have received a copy of the GNU Lesser General Public
  18    License along with the GNU C Library; if not, see
  19    <https://www.gnu.org/licenses/>.  */
  20
  21 #include <sysdep.h>
  22
  23         /* Normally, this uses ((xword - 0x01010101) & 0x80808080) test
  24            to find out if any byte in xword could be zero. This is fast, but
  25            also gives false alarm for any byte in range 0x81-0xff. It does
  26            not matter for correctness, as if this test tells us there could
  27            be some zero byte, we check it byte by byte, but if bytes with
  28            high bits set are common in the strings, then this will give poor
  29            performance. You can #define EIGHTBIT_NOT_RARE and the algorithm
  30            will use one tick slower, but more precise test
  31            ((xword - 0x01010101) & (~xword) & 0x80808080),
  32            which does not give any false alarms (but if some bits are set,
  33            one cannot assume from it which bytes are zero and which are not).
  34            It is yet to be measured, what is the correct default for glibc
  35            in these days for an average user.
  36          */
  37
  38         .text
  39         .align          4
  40
  41 ENTRY(strcpy)
  42         mov             %o0, %g2
  43         andcc           %o1, 3, %g0
  44         be              10f
  45          sethi          %hi(0x80808080), %o4
  46
  47         ldub            [%o1], %o5
  48         stb             %o5, [%o0]
  49         cmp             %o5, 0
  50         add             %o0, 1, %o0
  51         be              0f
  52          add            %o1, 1, %o1
  53         andcc           %o1, 3, %g0
  54         be              4f
  55          or             %o4, %lo(0x80808080), %o3
  56         ldub            [%o1], %o5
  57         stb             %o5, [%o0]
  58         cmp             %o5, 0
  59         add             %o0, 1, %o0
  60         be              0f
  61          add            %o1, 1, %o1
  62         andcc           %o1, 3, %g0
  63         be              5f
  64          sethi          %hi(0x01010101), %o4
  65         ldub            [%o1], %o5
  66         stb             %o5, [%o0]
  67         cmp             %o5, 0
  68         add             %o0, 1, %o0
  69         be              0f
  70          add            %o1, 1, %o1
  71         b               6f
  72          andcc          %o0, 3, %g3
  73
  74 10:     or              %o4, %lo(0x80808080), %o3
  75 4:      sethi           %hi(0x01010101), %o4
  76 5:      andcc           %o0, 3, %g3
  77 6:      bne             10f
  78          or             %o4, %lo(0x01010101), %o2
  79 1:      ld              [%o1], %o5
  80         add             %o1, 4, %o1
  81         sub             %o5, %o2, %o4
  82         add             %o0, 4, %o0
  83         andcc           %o4, %o3, %g0
  84         be,a            1b
  85          st             %o5, [%o0 - 4]
  86
  87         srl             %o5, 24, %g5
  88         andcc           %g5, 0xff, %g0
  89         be              1f
  90          srl            %o5, 16, %g5
  91         andcc           %g5, 0xff, %g0
  92         be              2f
  93          srl            %o5, 8, %g5
  94         andcc           %g5, 0xff, %g0
  95         be              3f
  96          andcc          %o5, 0xff, %g0
  97         bne             1b
  98          st             %o5, [%o0 - 4]
  99         retl
 100          mov            %g2, %o0
 101 3:      srl             %o5, 16, %o5
 102         sth             %o5, [%o0 - 4]
 103         stb             %g0, [%o0 - 2]
 104         retl
 105          mov            %g2, %o0
 106 2:      srl             %o5, 16, %o5
 107         sth             %o5, [%o0 - 4]
 108         retl
 109          mov            %g2, %o0
 110 1:      stb             %g0, [%o0 - 4]
 111         retl
 112          mov            %g2, %o0
 113
 114 10:     ld              [%o1], %o5
 115         add             %o1, 4, %o1
 116         sub             %o5, %o2, %o4
 117         cmp             %g3, 2
 118         be              2f
 119          cmp            %g3, 3
 120         be              3f
 121          andcc          %o4, %o3, %g0
 122         bne             5f
 123          srl            %o5, 24, %g5
 124         stb             %g5, [%o0]
 125         sub             %o0, 1, %o0
 126         srl             %o5, 8, %g5
 127         sth             %g5, [%o0 + 2]
 128 1:      add             %o0, 4, %o0
 129 4:      sll             %o5, 24, %g6
 130         ld              [%o1], %o5
 131         add             %o1, 4, %o1
 132         srl             %o5, 8, %g5
 133         sub             %o5, %o2, %o4
 134         or              %g5, %g6, %g5
 135         andcc           %o4, %o3, %g0
 136         be,a            1b
 137          st             %g5, [%o0]
 138         srl             %o5, 24, %o4
 139         andcc           %o4, 0xff, %g0
 140         be              6f
 141          srl            %o5, 16, %o4
 142         andcc           %o4, 0xff, %g0
 143         be              7f
 144          srl            %o5, 8, %o4
 145         st              %g5, [%o0]
 146         andcc           %o4, 0xff, %g0
 147         be              0f
 148          andcc          %o5, 0xff, %g0
 149 1:      bne             4b
 150          add            %o0, 4, %o0
 151 9:      stb             %g0, [%o0]
 152 0:      retl
 153          mov            %g2, %o0
 154 6:      srl             %g5, 16, %g5
 155         sth             %g5, [%o0]
 156         retl
 157          mov            %g2, %o0
 158 7:      srl             %g5, 16, %g5
 159         sth             %g5, [%o0]
 160         stb             %g0, [%o0 + 2]
 161         retl
 162          mov            %g2, %o0
 163 5:      andcc           %g5, 0xff, %g4
 164         be              9b
 165          srl            %o5, 16, %g5
 166         andcc           %g5, 0xff, %g0
 167         be              7f
 168          srl            %o5, 8, %g5
 169         andcc           %g5, 0xff, %g0
 170         stb             %g4, [%o0]
 171         sth             %g5, [%o0 + 1]
 172         sub             %o0, 1, %o0
 173         bne             1b
 174          andcc          %o5, 0xff, %g0
 175         retl
 176          mov            %g2, %o0
 177 7:      stb             %g4, [%o0]
 178         stb             %g0, [%o0 + 1]
 179         retl
 180          mov            %g2, %o0
 181
 182 2:      andcc           %o4, %o3, %g0
 183         bne             5f
 184          srl            %o5, 16, %g5
 185         sth             %g5, [%o0]
 186         sub             %o0, 2, %o0
 187 1:      add             %o0, 4, %o0
 188 4:      sll             %o5, 16, %g6
 189         ld              [%o1], %o5
 190         add             %o1, 4, %o1
 191         srl             %o5, 16, %g5
 192         sub             %o5, %o2, %o4
 193         or              %g5, %g6, %g5
 194         andcc           %o4, %o3, %g0
 195         be,a            1b
 196          st             %g5, [%o0]
 197         srl             %o5, 24, %o4
 198         andcc           %o4, 0xff, %g0
 199         be              7f
 200          srl            %o5, 16, %o4
 201         st              %g5, [%o0]
 202         andcc           %o4, 0xff, %g0
 203         be              0b
 204          srl            %o5, 8, %o4
 205 1:      andcc           %o4, 0xff, %g0
 206         be              8f
 207          andcc          %o5, 0xff, %g0
 208         bne             4b
 209          add            %o0, 4, %o0
 210         sth             %o5, [%o0]
 211         retl
 212          mov            %g2, %o0
 213 7:      srl             %g5, 16, %g5
 214         sth             %g5, [%o0]
 215         stb             %g0, [%o0 + 2]
 216         retl
 217          mov            %g2, %o0
 218 8:      stb             %g0, [%o0 + 4]
 219         retl
 220          mov            %g2, %o0
 221 5:      srl             %o5, 24, %g5
 222         andcc           %g5, 0xff, %g0
 223         be              9b
 224          srl            %o5, 16, %g5
 225         andcc           %g5, 0xff, %g0
 226         sth             %g5, [%o0]
 227         sub             %o0, 2, %o0
 228         bne             1b
 229          srl            %o5, 8, %o4
 230         retl
 231          mov            %g2, %o0
 232
 233 3:      bne             5f
 234          srl            %o5, 24, %g5
 235         stb             %g5, [%o0]
 236         sub             %o0, 3, %o0
 237 1:      add             %o0, 4, %o0
 238 4:      sll             %o5, 8, %g6
 239         ld              [%o1], %o5
 240         add             %o1, 4, %o1
 241         srl             %o5, 24, %g5
 242         sub             %o5, %o2, %o4
 243         or              %g5, %g6, %g5
 244         andcc           %o4, %o3, %g0
 245         be              1b
 246          st             %g5, [%o0]
 247         srl             %o5, 24, %o4
 248         andcc           %o4, 0xff, %g0
 249         be              0b
 250          srl            %o5, 16, %o4
 251 1:      andcc           %o4, 0xff, %g0
 252         be              8b
 253          srl            %o5, 8, %o4
 254         andcc           %o4, 0xff, %g0
 255         be              9f
 256          andcc          %o5, 0xff, %g0
 257         bne             4b
 258          add            %o0, 4, %o0
 259         srl             %o5, 8, %o5
 260         sth             %o5, [%o0]
 261         stb             %g0, [%o0 + 2]
 262         retl
 263          mov            %g2, %o0
 264 9:      srl             %o5, 8, %o5
 265         sth             %o5, [%o0 + 4]
 266         retl
 267          mov            %g2, %o0
 268 5:      andcc           %g5, 0xff, %g0
 269         stb             %g5, [%o0]
 270         sub             %o0, 3, %o0
 271         bne             1b
 272          srl            %o5, 16, %o4
 273         retl
 274          mov            %g2, %o0
 275 END(strcpy)
 276 libc_hidden_builtin_def (strcpy)