sysdeps/powerpc/powerpc64/power7/memrchr.S

   1 /* Optimized memrchr implementation for PowerPC64/POWER7 using cmpb insn.
   2    Copyright (C) 2010 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 #include <bp-sym.h>
  22 #include <bp-asm.h>
  23
  24 /* int [r3] memrchr (char *s [r3], int byte [r4], int size [r5])  */
  25         .machine  power7
  26 ENTRY (BP_SYM (__memrchr))
  27         CALL_MCOUNT
  28         dcbt    0,r3
  29         mr      r7,r3
  30         add     r3,r7,r5      /* Calculate the last acceptable address.  */
  31         cmpld   cr7,r3,r7     /* Is the address equal or less than r3?  */
  32
  33         /* Replicate BYTE to doubleword.  */
  34         rlwimi  r4,r4,8,16,23
  35         rlwimi  r4,r4,16,0,15
  36         insrdi  r4,r4,32,0
  37         bge     cr7,L(proceed)
  38
  39         li      r3,-1         /* Make r11 the biggest if r4 <= 0.  */
  40 L(proceed):
  41         li      r6,-8
  42         addi    r9,r3,-1
  43         clrrdi  r8,r9,3
  44         addi    r8,r8,8
  45         neg     r0,r3
  46         rlwinm  r0,r0,3,26,28 /* Calculate padding.  */
  47
  48         cmpldi  r5,32
  49         ble     L(small_range)
  50
  51         ldbrx   r12,r8,r6     /* Load reversed doubleword from memory.  */
  52         cmpb    r10,r12,r4    /* Check for BYTE in DWORD1.  */
  53         sld     r10,r10,r0
  54         srd     r10,r10,r0
  55         cmpldi  cr7,r10,0     /* If r10 == 0, no BYTE's have been found.  */
  56         bne     cr7,L(done)
  57
  58         /* Are we done already?  */
  59         addi    r9,r8,-8
  60         cmpld   cr6,r9,r7
  61         ble     cr6,L(null)
  62
  63         mtcrf   0x01,r8
  64         /* Are we now aligned to a doubleword boundary?  If so, skip to
  65            the main loop.  Otherwise, go through the alignment code.  */
  66         mr      r8,r9
  67         bt      28,L(loop_setup)
  68
  69         /* Handle DWORD2 of pair.  */
  70         ldbrx   r12,r8,r6
  71         cmpb    r10,r12,r4
  72         cmpldi  cr7,r10,0
  73         bne     cr7,L(done)
  74
  75         /* Are we done already.  */
  76         addi    r8,r8,-8
  77         cmpld   cr6,r8,r7
  78         ble     cr6,L(null)
  79
  80 L(loop_setup):
  81         li      r0,-16
  82         sub     r5,r8,r7
  83         srdi    r9,r5,4       /* Number of loop iterations.  */
  84         mtctr   r9            /* Setup the counter.  */
  85         b       L(loop)
  86         /* Main loop to look for BYTE backwards in the string.  Since it's a
  87            small loop (< 8 instructions), align it to 32-bytes.  */
  88         .p2align  5
  89 L(loop):
  90         /* Load two doublewords, compare and merge in a
  91            single register for speed.  This is an attempt
  92            to speed up the byte-checking process for bigger strings.  */
  93
  94         ldbrx   r12,r8,r6
  95         ldbrx   r11,r8,r0
  96         addi    r8,r8,-8
  97         cmpb    r10,r12,r4
  98         cmpb    r9,r11,r4
  99         or      r5,r9,r10     /* Merge everything in one doubleword.  */
 100         cmpldi  cr7,r5,0
 101         bne     cr7,L(found)
 102         addi    r8,r8,-8
 103         bdnz    L(loop)
 104         /* We're here because the counter reached 0, and that means we
 105            didn't have any matches for BYTE in the whole range.  Just return
 106            the original range.  */
 107         addi    r9,r8,8
 108         cmpld   cr6,r9,r7
 109         bgt     cr6,L(loop_small)
 110         b       L(null)
 111
 112         /* OK, one (or both) of the words contains BYTE.  Check
 113            the first word and decrement the address in case the first
 114            word really contains BYTE.  */
 115         .align  4
 116 L(found):
 117         cmpldi  cr6,r10,0
 118         addi    r8,r8,8
 119         bne     cr6,L(done)
 120
 121         /* BYTE must be in the second word.  Adjust the address
 122            again and move the result of cmpb to r10 so we can calculate the
 123            pointer.  */
 124
 125         mr      r10,r9
 126         addi    r8,r8,-8
 127
 128         /* r10 has the output of the cmpb instruction, that is, it contains
 129            0xff in the same position as the BYTE in the original
 130            word from the string.  Use that to calculate the pointer.
 131            We need to make sure BYTE is *before* the end of the
 132            range.  */
 133 L(done):
 134         cntlzd  r0,r10        /* Count leading zeroes before the match.  */
 135         srdi    r6,r0,3       /* Convert leading zeroes to bytes.  */
 136         addi    r0,r6,1
 137         sub     r3,r8,r0
 138         cmpld   r3,r7
 139         blt     L(null)
 140         blr
 141
 142         .align  4
 143 L(null):
 144         li      r3,0
 145         blr
 146
 147 /* Deals with size <= 32.  */
 148         .align  4
 149 L(small_range):
 150         cmpldi  r5,0
 151         beq     L(null)
 152
 153         ldbrx   r12,r8,r6     /* Load reversed doubleword from memory.  */
 154         cmpb    r10,r12,r4    /* Check for BYTE in DWORD1.  */
 155         sld     r10,r10,r0
 156         srd     r10,r10,r0
 157         cmpldi  cr7,r10,0
 158         bne     cr7,L(done)
 159
 160         /* Are we done already?  */
 161         addi    r8,r8,-8
 162         cmpld   r8,r7
 163         ble     L(null)
 164         b       L(loop_small)
 165
 166         .p2align  5
 167 L(loop_small):
 168         ldbrx   r12,r8,r6
 169         cmpb    r10,r12,r4
 170         cmpldi  cr6,r10,0
 171         bne     cr6,L(done)
 172         addi    r8,r8,-8
 173         cmpld   r8,r7
 174         ble     L(null)
 175         b       L(loop_small)
 176
 177 END (BP_SYM (__memrchr))
 178 weak_alias (BP_SYM (__memrchr), BP_SYM(memrchr))
 179 libc_hidden_builtin_def (memrchr)