sysdeps/powerpc/powerpc64/power7/strnlen.S

   1 /* Optimized strnlen implementation for PowerPC64/POWER7 using cmpb insn.
   2    Copyright (C) 2010-2016 Free Software Foundation, Inc.
   3    Contributed by Luis Machado <luisgpm@br.ibm.com>.
   4    This file is part of the GNU C Library.
   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    <http://www.gnu.org/licenses/>.  */
  19
  20 #include <sysdep.h>
  21
  22 /* int [r3] strnlen (char *s [r3], int size [r4])  */
  23         .machine  power7
  24 ENTRY (__strnlen)
  25         CALL_MCOUNT 2
  26         dcbt    0,r3
  27         clrrdi  r8,r3,3
  28         add     r7,r3,r4      /* Calculate the last acceptable address.  */
  29         cmpldi  r4,32
  30         li      r0,0          /* Doubleword with null chars.  */
  31         addi    r7,r7,-1
  32
  33         /* If we have less than 33 bytes to search, skip to a faster code.  */
  34         ble     L(small_range)
  35
  36         rlwinm  r6,r3,3,26,28 /* Calculate padding.  */
  37         ld      r12,0(r8)     /* Load doubleword from memory.  */
  38         cmpb    r10,r12,r0    /* Check for null bytes in DWORD1.  */
  39 #ifdef __LITTLE_ENDIAN__
  40         srd     r10,r10,r6
  41         sld     r10,r10,r6
  42 #else
  43         sld     r10,r10,r6
  44         srd     r10,r10,r6
  45 #endif
  46         cmpldi  cr7,r10,0     /* If r10 == 0, no null's have been found.  */
  47         bne     cr7,L(done)
  48
  49         clrrdi  r7,r7,3       /* Address of last doubleword.  */
  50         mtcrf   0x01,r8
  51         /* Are we now aligned to a quadword boundary?  If so, skip to
  52            the main loop.  Otherwise, go through the alignment code.  */
  53
  54         bt      28,L(loop_setup)
  55
  56         /* Handle DWORD2 of pair.  */
  57         ldu     r12,8(r8)
  58         cmpb    r10,r12,r0
  59         cmpldi  cr7,r10,0
  60         bne     cr7,L(done)
  61
  62 L(loop_setup):
  63         /* The last dword we want to read in the loop below is the one
  64            containing the last byte of the string, ie. the dword at
  65            (s + size - 1) & ~7, or r7.  The first dword read is at
  66            r8 + 8, we read 2 * cnt dwords, so the last dword read will
  67            be at r8 + 8 + 16 * cnt - 8.  Solving for cnt gives
  68            cnt = (r7 - r8) / 16  */
  69         sub     r5,r7,r8
  70         srdi    r6,r5,4       /* Number of loop iterations.  */
  71         mtctr   r6            /* Setup the counter.  */
  72
  73         /* Main loop to look for the null byte in the string.  Since
  74            it's a small loop (< 8 instructions), align it to 32-bytes.  */
  75         .p2align  5
  76 L(loop):
  77         /* Load two doublewords, compare and merge in a
  78            single register for speed.  This is an attempt
  79            to speed up the null-checking process for bigger strings.  */
  80
  81         ld      r12,8(r8)
  82         ldu     r11,16(r8)
  83         cmpb    r10,r12,r0
  84         cmpb    r9,r11,r0
  85         or      r5,r9,r10     /* Merge everything in one doubleword.  */
  86         cmpldi  cr7,r5,0
  87         bne     cr7,L(found)
  88         bdnz    L(loop)
  89
  90         /* We may have one more dword to read.  */
  91         cmpld   cr6,r8,r7
  92         beq     cr6,L(end_max)
  93
  94         ldu     r12,8(r8)
  95         cmpb    r10,r12,r0
  96         cmpldi  cr6,r10,0
  97         bne     cr6,L(done)
  98
  99 L(end_max):
 100         mr      r3,r4
 101         blr
 102
 103         /* OK, one (or both) of the doublewords contains a null byte.  Check
 104            the first doubleword and decrement the address in case the first
 105            doubleword really contains a null byte.  */
 106         .align  4
 107 L(found):
 108         cmpldi  cr6,r10,0
 109         addi    r8,r8,-8
 110         bne     cr6,L(done)
 111
 112         /* The null byte must be in the second doubleword.  Adjust the address
 113            again and move the result of cmpb to r10 so we can calculate the
 114            length.  */
 115
 116         mr      r10,r9
 117         addi    r8,r8,8
 118
 119         /* r10 has the output of the cmpb instruction, that is, it contains
 120            0xff in the same position as the null byte in the original
 121            doubleword from the string.  Use that to calculate the length.
 122            We need to make sure the null char is *before* the end of the
 123            range.  */
 124 L(done):
 125 #ifdef __LITTLE_ENDIAN__
 126         addi    r0,r10,-1
 127         andc    r0,r0,r10
 128         popcntd r0,r0
 129 #else
 130         cntlzd  r0,r10        /* Count leading zeros before the match.  */
 131 #endif
 132         sub     r3,r8,r3
 133         srdi    r0,r0,3       /* Convert leading/trailing zeros to bytes.  */
 134         add     r3,r3,r0      /* Length until the match.  */
 135         cmpld   r3,r4
 136         blelr
 137         mr      r3,r4
 138         blr
 139
 140 /* Deals with size <= 32.  */
 141         .align  4
 142 L(small_range):
 143         cmpldi  r4,0
 144         beq     L(end_max)
 145
 146         clrrdi  r7,r7,3       /* Address of last doubleword.  */
 147
 148         rlwinm  r6,r3,3,26,28 /* Calculate padding.  */
 149         ld      r12,0(r8)     /* Load doubleword from memory.  */
 150         cmpb    r10,r12,r0    /* Check for null bytes in DWORD1.  */
 151 #ifdef __LITTLE_ENDIAN__
 152         srd     r10,r10,r6
 153         sld     r10,r10,r6
 154 #else
 155         sld     r10,r10,r6
 156         srd     r10,r10,r6
 157 #endif
 158         cmpldi  cr7,r10,0
 159         bne     cr7,L(done)
 160
 161         cmpld   r8,r7
 162         beq     L(end_max)
 163
 164         .p2align  5
 165 L(loop_small):
 166         ldu     r12,8(r8)
 167         cmpb    r10,r12,r0
 168         cmpldi  cr6,r10,0
 169         bne     cr6,L(done)
 170         cmpld   r8,r7
 171         bne     L(loop_small)
 172         mr      r3,r4
 173         blr
 174
 175 END (__strnlen)
 176 libc_hidden_def (__strnlen)
 177 weak_alias (__strnlen, strnlen)
 178 libc_hidden_def (strnlen)