sysdeps/powerpc/powerpc64/strcmp.S

   1 /* Optimized strcmp implementation for PowerPC64.
   2    Copyright (C) 1997-2015 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    <http://www.gnu.org/licenses/>.  */
  18
  19 #include <sysdep.h>
  20
  21 /* See strlen.s for comments on how the end-of-string testing works.  */
  22
  23 /* int [r3] strcmp (const char *s1 [r3], const char *s2 [r4])  */
  24
  25 EALIGN (strcmp, 4, 0)
  26         CALL_MCOUNT 2
  27
  28 #define rTMP2   r0
  29 #define rRTN    r3
  30 #define rSTR1   r3      /* first string arg */
  31 #define rSTR2   r4      /* second string arg */
  32 #define rWORD1  r5      /* current word in s1 */
  33 #define rWORD2  r6      /* current word in s2 */
  34 #define rFEFE   r7      /* constant 0xfefefefefefefeff (-0x0101010101010101) */
  35 #define r7F7F   r8      /* constant 0x7f7f7f7f7f7f7f7f */
  36 #define rNEG    r9      /* ~(word in s1 | 0x7f7f7f7f7f7f7f7f) */
  37 #define rBITDIF r10     /* bits that differ in s1 & s2 words */
  38 #define rTMP    r11
  39
  40         dcbt    0,rSTR1
  41         or      rTMP, rSTR2, rSTR1
  42         dcbt    0,rSTR2
  43         clrldi. rTMP, rTMP, 61
  44         lis     rFEFE, -0x101
  45         bne     L(unaligned)
  46
  47         ld      rWORD1, 0(rSTR1)
  48         ld      rWORD2, 0(rSTR2)
  49         lis     r7F7F, 0x7f7f
  50         addi    rFEFE, rFEFE, -0x101
  51         addi    r7F7F, r7F7F, 0x7f7f
  52         sldi    rTMP, rFEFE, 32
  53         insrdi  r7F7F, r7F7F, 32, 0
  54         add     rFEFE, rFEFE, rTMP
  55         b       L(g1)
  56
  57 L(g0):  ldu     rWORD1, 8(rSTR1)
  58         bne     cr1, L(different)
  59         ldu     rWORD2, 8(rSTR2)
  60 L(g1):  add     rTMP, rFEFE, rWORD1
  61         nor     rNEG, r7F7F, rWORD1
  62         and.    rTMP, rTMP, rNEG
  63         cmpd    cr1, rWORD1, rWORD2
  64         beq+    L(g0)
  65
  66 /* OK. We've hit the end of the string. We need to be careful that
  67    we don't compare two strings as different because of gunk beyond
  68    the end of the strings...  */
  69 #ifdef __LITTLE_ENDIAN__
  70 L(endstring):
  71         addi    rTMP2, rTMP, -1
  72         beq     cr1, L(equal)
  73         andc    rTMP2, rTMP2, rTMP
  74         rldimi  rTMP2, rTMP2, 1, 0
  75         and     rWORD2, rWORD2, rTMP2   /* Mask off gunk.  */
  76         and     rWORD1, rWORD1, rTMP2
  77         cmpd    cr1, rWORD1, rWORD2
  78         beq     cr1, L(equal)
  79         xor     rBITDIF, rWORD1, rWORD2 /* rBITDIF has bits that differ.  */
  80         neg     rNEG, rBITDIF
  81         and     rNEG, rNEG, rBITDIF     /* rNEG has LS bit that differs.  */
  82         cntlzd  rNEG, rNEG              /* bitcount of the bit.  */
  83         andi.   rNEG, rNEG, 56          /* bitcount to LS byte that differs. */
  84         sld     rWORD1, rWORD1, rNEG    /* shift left to clear MS bytes.  */
  85         sld     rWORD2, rWORD2, rNEG
  86         xor.    rBITDIF, rWORD1, rWORD2
  87         sub     rRTN, rWORD1, rWORD2
  88         blt-    L(highbit)
  89         sradi   rRTN, rRTN, 63          /* must return an int.  */
  90         ori     rRTN, rRTN, 1
  91         blr
  92 L(equal):
  93         li      rRTN, 0
  94         blr
  95
  96 L(different):
  97         ld      rWORD1, -8(rSTR1)
  98         xor     rBITDIF, rWORD1, rWORD2 /* rBITDIF has bits that differ.  */
  99         neg     rNEG, rBITDIF
 100         and     rNEG, rNEG, rBITDIF     /* rNEG has LS bit that differs.  */
 101         cntlzd  rNEG, rNEG              /* bitcount of the bit.  */
 102         andi.   rNEG, rNEG, 56          /* bitcount to LS byte that differs. */
 103         sld     rWORD1, rWORD1, rNEG    /* shift left to clear MS bytes.  */
 104         sld     rWORD2, rWORD2, rNEG
 105         xor.    rBITDIF, rWORD1, rWORD2
 106         sub     rRTN, rWORD1, rWORD2
 107         blt-    L(highbit)
 108         sradi   rRTN, rRTN, 63
 109         ori     rRTN, rRTN, 1
 110         blr
 111 L(highbit):
 112         sradi   rRTN, rWORD2, 63
 113         ori     rRTN, rRTN, 1
 114         blr
 115
 116 #else
 117 L(endstring):
 118         and     rTMP, r7F7F, rWORD1
 119         beq     cr1, L(equal)
 120         add     rTMP, rTMP, r7F7F
 121         xor.    rBITDIF, rWORD1, rWORD2
 122         andc    rNEG, rNEG, rTMP
 123         blt-    L(highbit)
 124         cntlzd  rBITDIF, rBITDIF
 125         cntlzd  rNEG, rNEG
 126         addi    rNEG, rNEG, 7
 127         cmpd    cr1, rNEG, rBITDIF
 128         sub     rRTN, rWORD1, rWORD2
 129         blt-    cr1, L(equal)
 130         sradi   rRTN, rRTN, 63          /* must return an int.  */
 131         ori     rRTN, rRTN, 1
 132         blr
 133 L(equal):
 134         li      rRTN, 0
 135         blr
 136
 137 L(different):
 138         ld      rWORD1, -8(rSTR1)
 139         xor.    rBITDIF, rWORD1, rWORD2
 140         sub     rRTN, rWORD1, rWORD2
 141         blt-    L(highbit)
 142         sradi   rRTN, rRTN, 63
 143         ori     rRTN, rRTN, 1
 144         blr
 145 L(highbit):
 146         sradi   rRTN, rWORD2, 63
 147         ori     rRTN, rRTN, 1
 148         blr
 149 #endif
 150
 151 /* Oh well.  In this case, we just do a byte-by-byte comparison.  */
 152         .align 4
 153 L(unaligned):
 154         lbz     rWORD1, 0(rSTR1)
 155         lbz     rWORD2, 0(rSTR2)
 156         b       L(u1)
 157
 158 L(u0):  lbzu    rWORD1, 1(rSTR1)
 159         bne-    L(u4)
 160         lbzu    rWORD2, 1(rSTR2)
 161 L(u1):  cmpwi   cr1, rWORD1, 0
 162         beq-    cr1, L(u3)
 163         cmpd    rWORD1, rWORD2
 164         bne-    L(u3)
 165         lbzu    rWORD1, 1(rSTR1)
 166         lbzu    rWORD2, 1(rSTR2)
 167         cmpdi   cr1, rWORD1, 0
 168         cmpd    rWORD1, rWORD2
 169         bne+    cr1, L(u0)
 170 L(u3):  sub     rRTN, rWORD1, rWORD2
 171         blr
 172 L(u4):  lbz     rWORD1, -1(rSTR1)
 173         sub     rRTN, rWORD1, rWORD2
 174         blr
 175 END (strcmp)
 176 libc_hidden_builtin_def (strcmp)