sysdeps/sparc/sparc64/rawmemchr.S

   1 /* rawmemchr (str, ch) -- Return pointer to first occurrence of CH in STR.
   2    For SPARC v9.
   3    Copyright (C) 1999-2015 Free Software Foundation, Inc.
   4    This file is part of the GNU C Library.
   5    Contributed by Jakub Jelinek <jj@ultra.linux.cz>.
   6    This version is developed using the same algorithm as the fast C
   7    version which carries the following introduction:
   8    Based on strlen implementation by Torbjorn Granlund (tege@sics.se),
   9    with help from Dan Sahlin (dan@sics.se) and
  10    commentary by Jim Blandy (jimb@ai.mit.edu);
  11    adaptation to memchr suggested by Dick Karpinski (dick@cca.ucsf.edu),
  12    and implemented by Roland McGrath (roland@ai.mit.edu).
  13
  14    The GNU C Library is free software; you can redistribute it and/or
  15    modify it under the terms of the GNU Lesser General Public
  16    License as published by the Free Software Foundation; either
  17    version 2.1 of the License, or (at your option) any later version.
  18
  19    The GNU C Library is distributed in the hope that it will be useful,
  20    but WITHOUT ANY WARRANTY; without even the implied warranty of
  21    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  22    Lesser General Public License for more details.
  23
  24    You should have received a copy of the GNU Lesser General Public
  25    License along with the GNU C Library; if not, see
  26    <http://www.gnu.org/licenses/>.  */
  27
  28 #include <sysdep.h>
  29 #include <asm/asi.h>
  30 #ifndef XCC
  31 #define XCC xcc
  32 #define USE_BPR
  33         .register       %g2, #scratch
  34         .register       %g3, #scratch
  35 #endif
  36
  37         /* Normally, this uses
  38            ((xword - 0x0101010101010101) & 0x8080808080808080) test
  39            to find out if any byte in xword could be zero. This is fast, but
  40            also gives false alarm for any byte in range 0x81-0xff. It does
  41            not matter for correctness, as if this test tells us there could
  42            be some zero byte, we check it byte by byte, but if bytes with
  43            high bits set are common in the strings, then this will give poor
  44            performance. You can #define EIGHTBIT_NOT_RARE and the algorithm
  45            will use one tick slower, but more precise test
  46            ((xword - 0x0101010101010101) & (~xword) & 0x8080808080808080),
  47            which does not give any false alarms (but if some bits are set,
  48            one cannot assume from it which bytes are zero and which are not).
  49            It is yet to be measured, what is the correct default for glibc
  50            in these days for an average user.
  51          */
  52
  53         .text
  54         .align          32
  55 ENTRY(__rawmemchr)
  56         and             %o1, 0xff, %o1                  /* IEU0         Group           */
  57         sethi           %hi(0x01010101), %g1            /* IEU1                         */
  58         ldub            [%o0], %o3                      /* Load                         */
  59         sll             %o1, 8, %o4                     /* IEU0         Group           */
  60
  61         or              %g1, %lo(0x01010101), %g1       /* IEU1                         */
  62         sllx            %g1, 32, %g2                    /* IEU0         Group           */
  63         or              %o4, %o1, %o4                   /* IEU1                         */
  64         andcc           %o0, 7, %g0                     /* IEU1         Group           */
  65
  66         sll             %o4, 16, %g5                    /* IEU0                         */
  67         or              %o4, %g5, %o4                   /* IEU0         Group           */
  68         or              %g1, %g2, %g1                   /* IEU1                         */
  69         bne,pn          %icc, 32f                       /* CTI                          */
  70
  71          sllx           %o4, 32, %g5                    /* IEU0         Group           */
  72         cmp             %o3, %o1                        /* IEU1                         */
  73         be,pn           %icc, 30f                       /* CTI                          */
  74          sllx           %g1, 7, %g2                     /* IEU0         Group           */
  75
  76 18:     ldx             [%o0], %o3                      /* Load                         */
  77         or              %o4, %g5, %o4                   /* IEU1                         */
  78         add             %o0, 8, %o0                     /* IEU0         Group           */
  79 19:     xor             %o3, %o4, %o3                   /* IEU0         Group           */
  80
  81         sub             %o3, %g1, %o2                   /* IEU0         Group           */
  82 #ifdef EIGHTBIT_NOT_RARE
  83         andn            %o2, %o3, %o5                   /* IEU0         Group           */
  84         ldxa            [%o0] ASI_PNF, %o3              /* Load                         */
  85         andcc           %o5, %g2, %g0                   /* IEU1         Group           */
  86 #else
  87         ldxa            [%o0] ASI_PNF, %o3              /* Load                         */
  88         andcc           %o2, %g2, %g0                   /* IEU1         Group           */
  89 #endif
  90         be,pt           %xcc, 19b                       /* CTI                          */
  91
  92          add            %o0, 8, %o0                     /* IEU0                         */
  93         addcc           %o2, %g1, %g3                   /* IEU1         Group           */
  94         srlx            %o2, 32, %o2                    /* IEU0                         */
  95 20:     andcc           %o2, %g2, %g0                   /* IEU1         Group           */
  96
  97         be,pn           %xcc, 21f                       /* CTI                          */
  98          srlx           %g3, 56, %o2                    /* IEU0                         */
  99         andcc           %o2, 0xff, %g0                  /* IEU1         Group           */
 100         be,pn           %icc, 29f                       /* CTI                          */
 101
 102          srlx           %g3, 48, %o2                    /* IEU0                         */
 103         andcc           %o2, 0xff, %g0                  /* IEU1         Group           */
 104         be,pn           %icc, 28f                       /* CTI                          */
 105          srlx           %g3, 40, %o2                    /* IEU0                         */
 106
 107         andcc           %o2, 0xff, %g0                  /* IEU1         Group           */
 108         be,pn           %icc, 27f                       /* CTI                          */
 109          srlx           %g3, 32, %o2                    /* IEU0                         */
 110         andcc           %o2, 0xff, %g0                  /* IEU1         Group           */
 111
 112         be,pn           %icc, 26f                       /* CTI                          */
 113 21:      srlx           %g3, 24, %o2                    /* IEU0                         */
 114         andcc           %o2, 0xff, %g0                  /* IEU1         Group           */
 115         be,pn           %icc, 25f                       /* CTI                          */
 116
 117          srlx           %g3, 16, %o2                    /* IEU0                         */
 118         andcc           %o2, 0xff, %g0                  /* IEU1         Group           */
 119         be,pn           %icc, 24f                       /* CTI                          */
 120          srlx           %g3, 8, %o2                     /* IEU0                         */
 121
 122         andcc           %o2, 0xff, %g0                  /* IEU1         Group           */
 123         be,pn           %icc, 23f                       /* CTI                          */
 124          xor            %o3, %o4, %o3                   /* IEU0                         */
 125         andcc           %g3, 0xff, %g0                  /* IEU1         Group           */
 126
 127         be,pn           %icc, 22f                       /* CTI                          */
 128          sub            %o3, %g1, %o2                   /* IEU0                         */
 129         ldxa            [%o0] ASI_PNF, %o3              /* Load                         */
 130         andcc           %o2, %g2, %g0                   /* IEU1         Group           */
 131
 132         be,pt           %xcc, 19b                       /* CTI                          */
 133          add            %o0, 8, %o0                     /* IEU0                         */
 134         addcc           %o2, %g1, %g3                   /* IEU1         Group           */
 135         ba,pt           %xcc, 20b                       /* CTI                          */
 136
 137          srlx           %o2, 32, %o2                    /* IEU0                         */
 138
 139         .align          16
 140 22:     retl                                            /* CTI+IEU1     Group           */
 141          add            %o0, -9, %o0                    /* IEU0                         */
 142 23:     retl                                            /* CTI+IEU1     Group           */
 143          add            %o0, -10, %o0                   /* IEU0                         */
 144
 145 24:     retl                                            /* CTI+IEU1     Group           */
 146          add            %o0, -11, %o0                   /* IEU0                         */
 147 25:     retl                                            /* CTI+IEU1     Group           */
 148          add            %o0, -12, %o0                   /* IEU0                         */
 149
 150 26:     retl                                            /* CTI+IEU1     Group           */
 151          add            %o0, -13, %o0                   /* IEU0                         */
 152 27:     retl                                            /* CTI+IEU1     Group           */
 153          add            %o0, -14, %o0                   /* IEU0                         */
 154
 155 28:     retl                                            /* CTI+IEU1     Group           */
 156          add            %o0, -15, %o0                   /* IEU0                         */
 157 29:     retl                                            /* CTI+IEU1     Group           */
 158          add            %o0, -16, %o0                   /* IEU0                         */
 159
 160 30:     retl                                            /* CTI+IEU1     Group           */
 161          nop                                            /* IEU0                         */
 162
 163         .align          16
 164 32:     andcc           %o0, 7, %g0                     /* IEU1         Group           */
 165         be,a,pn         %icc, 18b                       /* CTI                          */
 166          sllx           %g1, 7, %g2                     /* IEU0                         */
 167         add             %o0, 1, %o0                     /* IEU0         Group           */
 168
 169         cmp             %o3, %o1                        /* IEU1                         */
 170         bne,a,pt        %icc, 32b                       /* CTI                          */
 171          lduba          [%o0] ASI_PNF, %o3              /* Load                         */
 172         retl                                            /* CTI+IEU1     Group           */
 173
 174          add            %o0, -1, %o0                    /* IEU0                         */
 175 END(__rawmemchr)
 176
 177 libc_hidden_def (__rawmemchr)
 178 weak_alias (__rawmemchr, rawmemchr)