sysdeps/powerpc/powerpc32/power7/rawmemchr.S

   1 /* Optimized rawmemchr implementation for PowerPC32/POWER7 using cmpb insn.
   2    Copyright (C) 2010-2021 Free Software Foundation, Inc.
   3    This file is part of the GNU C Library.
   4
   5    The GNU C Library is free software; you can redistribute it and/or
   6    modify it under the terms of the GNU Lesser General Public
   7    License as published by the Free Software Foundation; either
   8    version 2.1 of the License, or (at your option) any later version.
   9
  10    The GNU C Library is distributed in the hope that it will be useful,
  11    but WITHOUT ANY WARRANTY; without even the implied warranty of
  12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13    Lesser General Public License for more details.
  14
  15    You should have received a copy of the GNU Lesser General Public
  16    License along with the GNU C Library; if not, see
  17    <https://www.gnu.org/licenses/>.  */
  18
  19 #include <sysdep.h>
  20
  21 /* int [r3] rawmemchr (void *s [r3], int c [r4])  */
  22         .machine  power7
  23 ENTRY (__rawmemchr)
  24         CALL_MCOUNT
  25         dcbt    0,r3
  26         clrrwi  r8,r3,2       /* Align the address to word boundary.  */
  27
  28         /* Replicate byte to word.  */
  29         insrwi  r4,r4,8,16
  30         insrwi  r4,r4,16,0
  31
  32         /* Now r4 has a word of c bytes.  */
  33
  34         rlwinm  r6,r3,3,27,28 /* Calculate padding.  */
  35         lwz     r12,0(r8)     /* Load word from memory.  */
  36         cmpb    r5,r12,r4     /* Compare each byte against c byte.  */
  37 #ifdef __LITTLE_ENDIAN__
  38         srw     r5,r5,r6
  39         slw     r5,r5,r6
  40 #else
  41         slw     r5,r5,r6      /* Move left to discard ignored bits.  */
  42         srw     r5,r5,r6      /* Bring the bits back as zeros.  */
  43 #endif
  44         cmpwi   cr7,r5,0      /* If r5 == 0, no c bytes have been found.  */
  45         bne     cr7,L(done)
  46
  47         mtcrf   0x01,r8
  48
  49         /* Are we now aligned to a doubleword boundary?  If so, skip to
  50            the main loop.  Otherwise, go through the alignment code.  */
  51
  52         bt      29,L(loop)
  53
  54         /* Handle WORD2 of pair.  */
  55         lwzu    r12,4(r8)
  56         cmpb    r5,r12,r4
  57         cmpwi   cr7,r5,0
  58         bne     cr7,L(done)
  59         b       L(loop)       /* We branch here (rather than falling through)
  60                                  to skip the nops due to heavy alignment
  61                                  of the loop below.  */
  62
  63         /* Main loop to look for the end of the string.  Since it's a
  64            small loop (< 8 instructions), align it to 32-bytes.  */
  65         .p2align  5
  66 L(loop):
  67         /* Load two words, compare and merge in a
  68            single register for speed.  This is an attempt
  69            to speed up the byte-checking process for bigger strings.  */
  70         lwz     r12,4(r8)
  71         lwzu    r11,8(r8)
  72         cmpb    r5,r12,r4
  73         cmpb    r6,r11,r4
  74         or      r7,r5,r6
  75         cmpwi   cr7,r7,0
  76         beq     cr7,L(loop)
  77
  78         /* OK, one (or both) of the words contains a 'c' byte.  Check
  79            the first word and decrement the address in case the first
  80            word really contains a c byte.  */
  81
  82         cmpwi   cr6,r5,0
  83         addi    r8,r8,-4
  84         bne     cr6,L(done)
  85
  86         /* The 'c' byte must be in the second word.  Adjust the address
  87            again and move the result of cmpb to r10 so we can calculate the
  88            pointer.  */
  89         mr      r5,r6
  90         addi    r8,r8,4
  91
  92         /* r5 has the output of the cmpb instruction, that is, it contains
  93            0xff in the same position as the 'c' byte in the original
  94            word from the string.  Use that fact to find out what is
  95            the position of the byte inside the string.  */
  96 L(done):
  97 #ifdef __LITTLE_ENDIAN__
  98         addi    r0,r5,-1
  99         andc    r0,r0,r5
 100         popcntw r0,r0
 101 #else
 102         cntlzw  r0,r5         /* Count leading zeros before the match.  */
 103 #endif
 104         srwi    r0,r0,3       /* Convert leading zeros to bytes.  */
 105         add     r3,r8,r0      /* Return address of the matching char.  */
 106         blr
 107 END (__rawmemchr)
 108 weak_alias (__rawmemchr,rawmemchr)
 109 libc_hidden_builtin_def (__rawmemchr)