sysdeps/i386/strchr.S

   1 /* strchr (str, ch) -- Return pointer to first occurrence of CH in STR.
   2    For Intel 80x86, x>=3.
   3    Copyright (C) 1994-1997,1999,2000,2002,2003 Free Software Foundation, Inc.
   4    This file is part of the GNU C Library.
   5    Contributed by Ulrich Drepper <drepper@gnu.ai.mit.edu>
   6    Some optimisations by Alan Modra <Alan@SPRI.Levels.UniSA.Edu.Au>
   7
   8    The GNU C Library is free software; you can redistribute it and/or
   9    modify it under the terms of the GNU Lesser General Public
  10    License as published by the Free Software Foundation; either
  11    version 2.1 of the License, or (at your option) any later version.
  12
  13    The GNU C Library is distributed in the hope that it will be useful,
  14    but WITHOUT ANY WARRANTY; without even the implied warranty of
  15    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16    Lesser General Public License for more details.
  17
  18    You should have received a copy of the GNU Lesser General Public
  19    License along with the GNU C Library; if not, write to the Free
  20    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  21    02111-1307 USA.  */
  22
  23 #include <sysdep.h>
  24 #include "asm-syntax.h"
  25 #include "bp-sym.h"
  26 #include "bp-asm.h"
  27
  28 #define PARMS   LINKAGE+4               /* space for 1 saved reg */
  29 #define RTN     PARMS
  30 #define STR     RTN+RTN_SIZE
  31 #define CHR     STR+PTR_SIZE
  32
  33         .text
  34 ENTRY (BP_SYM (strchr))
  35         ENTER
  36
  37         pushl %edi              /* Save callee-safe registers used here.  */
  38         movl STR(%esp), %eax
  39         movl CHR(%esp), %edx
  40         CHECK_BOUNDS_LOW (%eax, STR(%esp))
  41
  42         /* At the moment %edx contains C.  What we need for the
  43            algorithm is C in all bytes of the dword.  Avoid
  44            operations on 16 bit words because these require an
  45            prefix byte (and one more cycle).  */
  46         movb %dl, %dh           /* now it is 0|0|c|c */
  47         movl %edx, %ecx
  48         shll $16, %edx          /* now it is c|c|0|0 */
  49         movw %cx, %dx           /* and finally c|c|c|c */
  50
  51         /* Before we start with the main loop we process single bytes
  52            until the source pointer is aligned.  This has two reasons:
  53            1. aligned 32-bit memory access is faster
  54            and (more important)
  55            2. we process in the main loop 32 bit in one step although
  56               we don't know the end of the string.  But accessing at
  57               4-byte alignment guarantees that we never access illegal
  58               memory if this would not also be done by the trivial
  59               implementation (this is because all processor inherent
  60               boundaries are multiples of 4.  */
  61
  62         testb $3, %al           /* correctly aligned ? */
  63         jz L(11)                /* yes => begin loop */
  64         movb (%eax), %cl        /* load byte in question (we need it twice) */
  65         cmpb %cl, %dl           /* compare byte */
  66         je L(6)                 /* target found => return */
  67         testb %cl, %cl          /* is NUL? */
  68         jz L(2)                 /* yes => return NULL */
  69         incl %eax               /* increment pointer */
  70
  71         testb $3, %al           /* correctly aligned ? */
  72         jz L(11)                /* yes => begin loop */
  73         movb (%eax), %cl        /* load byte in question (we need it twice) */
  74         cmpb %cl, %dl           /* compare byte */
  75         je L(6)                 /* target found => return */
  76         testb %cl, %cl          /* is NUL? */
  77         jz L(2)                 /* yes => return NULL */
  78         incl %eax               /* increment pointer */
  79
  80         testb $3, %al           /* correctly aligned ? */
  81         jz L(11)                /* yes => begin loop */
  82         movb (%eax), %cl        /* load byte in question (we need it twice) */
  83         cmpb %cl, %dl           /* compare byte */
  84         je L(6)                 /* target found => return */
  85         testb %cl, %cl          /* is NUL? */
  86         jz L(2)                 /* yes => return NULL */
  87         incl %eax               /* increment pointer */
  88
  89         /* No we have reached alignment.  */
  90         jmp L(11)               /* begin loop */
  91
  92       /* We exit the loop if adding MAGIC_BITS to LONGWORD fails to
  93          change any of the hole bits of LONGWORD.
  94
  95          1) Is this safe?  Will it catch all the zero bytes?
  96          Suppose there is a byte with all zeros.  Any carry bits
  97          propagating from its left will fall into the hole at its
  98          least significant bit and stop.  Since there will be no
  99          carry from its most significant bit, the LSB of the
 100          byte to the left will be unchanged, and the zero will be
 101          detected.
 102
 103          2) Is this worthwhile?  Will it ignore everything except
 104          zero bytes?  Suppose every byte of LONGWORD has a bit set
 105          somewhere.  There will be a carry into bit 8.  If bit 8
 106          is set, this will carry into bit 16.  If bit 8 is clear,
 107          one of bits 9-15 must be set, so there will be a carry
 108          into bit 16.  Similarly, there will be a carry into bit
 109          24.  If one of bits 24-31 is set, there will be a carry
 110          into bit 32 (=carry flag), so all of the hole bits will
 111          be changed.
 112
 113          3) But wait!  Aren't we looking for C, not zero?
 114          Good point.  So what we do is XOR LONGWORD with a longword,
 115          each of whose bytes is C.  This turns each byte that is C
 116          into a zero.  */
 117
 118         /* Each round the main loop processes 16 bytes.  */
 119
 120         ALIGN(4)
 121
 122 L(1):   addl $16, %eax          /* adjust pointer for whole round */
 123
 124 L(11):  movl (%eax), %ecx       /* get word (= 4 bytes) in question */
 125         xorl %edx, %ecx         /* XOR with word c|c|c|c => bytes of str == c
 126                                    are now 0 */
 127         movl $0xfefefeff, %edi  /* magic value */
 128         addl %ecx, %edi         /* add the magic value to the word.  We get
 129                                    carry bits reported for each byte which
 130                                    is *not* C */
 131
 132         /* According to the algorithm we had to reverse the effect of the
 133            XOR first and then test the overflow bits.  But because the
 134            following XOR would destroy the carry flag and it would (in a
 135            representation with more than 32 bits) not alter then last
 136            overflow, we can now test this condition.  If no carry is signaled
 137            no overflow must have occurred in the last byte => it was 0. */
 138         jnc L(7)
 139
 140         /* We are only interested in carry bits that change due to the
 141            previous add, so remove original bits */
 142         xorl %ecx, %edi         /* ((word^charmask)+magic)^(word^charmask) */
 143
 144         /* Now test for the other three overflow bits.  */
 145         orl $0xfefefeff, %edi   /* set all non-carry bits */
 146         incl %edi               /* add 1: if one carry bit was *not* set
 147                                    the addition will not result in 0.  */
 148
 149         /* If at least one byte of the word is C we don't get 0 in %edi.  */
 150         jnz L(7)                /* found it => return pointer */
 151
 152         /* Now we made sure the dword does not contain the character we are
 153            looking for.  But because we deal with strings we have to check
 154            for the end of string before testing the next dword.  */
 155
 156         xorl %edx, %ecx         /* restore original dword without reload */
 157         movl $0xfefefeff, %edi  /* magic value */
 158         addl %ecx, %edi         /* add the magic value to the word.  We get
 159                                    carry bits reported for each byte which
 160                                    is *not* 0 */
 161         jnc L(2)                /* highest byte is NUL => return NULL */
 162         xorl %ecx, %edi         /* (word+magic)^word */
 163         orl $0xfefefeff, %edi   /* set all non-carry bits */
 164         incl %edi               /* add 1: if one carry bit was *not* set
 165                                    the addition will not result in 0.  */
 166         jnz L(2)                /* found NUL => return NULL */
 167
 168         movl 4(%eax), %ecx      /* get word (= 4 bytes) in question */
 169         xorl %edx, %ecx         /* XOR with word c|c|c|c => bytes of str == c
 170                                    are now 0 */
 171         movl $0xfefefeff, %edi  /* magic value */
 172         addl %ecx, %edi         /* add the magic value to the word.  We get
 173                                    carry bits reported for each byte which
 174                                    is *not* C */
 175         jnc L(71)               /* highest byte is C => return pointer */
 176         xorl %ecx, %edi         /* ((word^charmask)+magic)^(word^charmask) */
 177         orl $0xfefefeff, %edi   /* set all non-carry bits */
 178         incl %edi               /* add 1: if one carry bit was *not* set
 179                                    the addition will not result in 0.  */
 180         jnz L(71)               /* found it => return pointer */
 181         xorl %edx, %ecx         /* restore original dword without reload */
 182         movl $0xfefefeff, %edi  /* magic value */
 183         addl %ecx, %edi         /* add the magic value to the word.  We get
 184                                    carry bits reported for each byte which
 185                                    is *not* 0 */
 186         jnc L(2)                /* highest byte is NUL => return NULL */
 187         xorl %ecx, %edi         /* (word+magic)^word */
 188         orl $0xfefefeff, %edi   /* set all non-carry bits */
 189         incl %edi               /* add 1: if one carry bit was *not* set
 190                                    the addition will not result in 0.  */
 191         jnz L(2)                /* found NUL => return NULL */
 192
 193         movl 8(%eax), %ecx      /* get word (= 4 bytes) in question */
 194         xorl %edx, %ecx         /* XOR with word c|c|c|c => bytes of str == c
 195                                    are now 0 */
 196         movl $0xfefefeff, %edi  /* magic value */
 197         addl %ecx, %edi         /* add the magic value to the word.  We get
 198                                    carry bits reported for each byte which
 199                                    is *not* C */
 200         jnc L(72)               /* highest byte is C => return pointer */
 201         xorl %ecx, %edi         /* ((word^charmask)+magic)^(word^charmask) */
 202         orl $0xfefefeff, %edi   /* set all non-carry bits */
 203         incl %edi               /* add 1: if one carry bit was *not* set
 204                                    the addition will not result in 0.  */
 205         jnz L(72)               /* found it => return pointer */
 206         xorl %edx, %ecx         /* restore original dword without reload */
 207         movl $0xfefefeff, %edi  /* magic value */
 208         addl %ecx, %edi         /* add the magic value to the word.  We get
 209                                    carry bits reported for each byte which
 210                                    is *not* 0 */
 211         jnc L(2)                /* highest byte is NUL => return NULL */
 212         xorl %ecx, %edi         /* (word+magic)^word */
 213         orl $0xfefefeff, %edi   /* set all non-carry bits */
 214         incl %edi               /* add 1: if one carry bit was *not* set
 215                                    the addition will not result in 0.  */
 216         jnz L(2)                /* found NUL => return NULL */
 217
 218         movl 12(%eax), %ecx     /* get word (= 4 bytes) in question */
 219         xorl %edx, %ecx         /* XOR with word c|c|c|c => bytes of str == c
 220                                    are now 0 */
 221         movl $0xfefefeff, %edi  /* magic value */
 222         addl %ecx, %edi         /* add the magic value to the word.  We get
 223                                    carry bits reported for each byte which
 224                                    is *not* C */
 225         jnc L(73)               /* highest byte is C => return pointer */
 226         xorl %ecx, %edi         /* ((word^charmask)+magic)^(word^charmask) */
 227         orl $0xfefefeff, %edi   /* set all non-carry bits */
 228         incl %edi               /* add 1: if one carry bit was *not* set
 229                                    the addition will not result in 0.  */
 230         jnz L(73)               /* found it => return pointer */
 231         xorl %edx, %ecx         /* restore original dword without reload */
 232         movl $0xfefefeff, %edi  /* magic value */
 233         addl %ecx, %edi         /* add the magic value to the word.  We get
 234                                    carry bits reported for each byte which
 235                                    is *not* 0 */
 236         jnc L(2)                /* highest byte is NUL => return NULL */
 237         xorl %ecx, %edi         /* (word+magic)^word */
 238         orl $0xfefefeff, %edi   /* set all non-carry bits */
 239         incl %edi               /* add 1: if one carry bit was *not* set
 240                                    the addition will not result in 0.  */
 241         jz L(1)                 /* no NUL found => restart loop */
 242
 243 L(2):   /* Return NULL.  */
 244         xorl %eax, %eax
 245         RETURN_NULL_BOUNDED_POINTER
 246         popl %edi               /* restore saved register content */
 247
 248         LEAVE
 249         RET_PTR
 250
 251 L(73):  addl $4, %eax           /* adjust pointer */
 252 L(72):  addl $4, %eax
 253 L(71):  addl $4, %eax
 254
 255         /* We now scan for the byte in which the character was matched.
 256            But we have to take care of the case that a NUL char is
 257            found before this in the dword.  Note that we XORed %ecx
 258            with the byte we're looking for, therefore the tests below look
 259            reversed.  */
 260
 261 L(7):   testb %cl, %cl          /* is first byte C? */
 262         jz L(6)                 /* yes => return pointer */
 263         cmpb %dl, %cl           /* is first byte NUL? */
 264         je L(2)                 /* yes => return NULL */
 265         incl %eax               /* it's not in the first byte */
 266
 267         testb %ch, %ch          /* is second byte C? */
 268         jz L(6)                 /* yes => return pointer */
 269         cmpb %dl, %ch           /* is second byte NUL? */
 270         je L(2)                 /* yes => return NULL? */
 271         incl %eax               /* it's not in the second byte */
 272
 273         shrl $16, %ecx          /* make upper byte accessible */
 274         testb %cl, %cl          /* is third byte C? */
 275         jz L(6)                 /* yes => return pointer */
 276         cmpb %dl, %cl           /* is third byte NUL? */
 277         je L(2)                 /* yes => return NULL */
 278
 279         /* It must be in the fourth byte and it cannot be NUL.  */
 280         incl %eax
 281
 282 L(6):
 283         CHECK_BOUNDS_HIGH (%eax, STR(%esp), jb)
 284         RETURN_BOUNDED_POINTER (STR(%esp))
 285         popl %edi               /* restore saved register content */
 286
 287         LEAVE
 288         RET_PTR
 289 END (BP_SYM (strchr))
 290
 291 weak_alias (BP_SYM (strchr), BP_SYM (index))
 292 libc_hidden_builtin_def (strchr)