sysdeps/powerpc/powerpc64/power4/strncmp.S

   1 /* Optimized strcmp implementation for PowerPC64.
   2    Copyright (C) 2003, 2006, 2011 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, write to the Free
  17    Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA
  18    02110-1301 USA.  */
  19
  20 #include <sysdep.h>
  21 #include <bp-sym.h>
  22 #include <bp-asm.h>
  23
  24 /* See strlen.s for comments on how the end-of-string testing works.  */
  25
  26 /* int [r3] strncmp (const char *s1 [r3], const char *s2 [r4], size_t size [r5])  */
  27
  28 EALIGN (BP_SYM(strncmp), 4, 0)
  29         CALL_MCOUNT 3
  30
  31 #define rTMP    r0
  32 #define rRTN    r3
  33 #define rSTR1   r3      /* first string arg */
  34 #define rSTR2   r4      /* second string arg */
  35 #define rN      r5      /* max string length */
  36 /* Note:  The Bounded pointer support in this code is broken.  This code
  37    was inherited from PPC32 and and that support was never completed.
  38    Current PPC gcc does not support -fbounds-check or -fbounded-pointers.  */
  39 #define rWORD1  r6      /* current word in s1 */
  40 #define rWORD2  r7      /* current word in s2 */
  41 #define rWORD3  r10
  42 #define rWORD4  r11
  43 #define rFEFE   r8      /* constant 0xfefefefefefefeff (-0x0101010101010101) */
  44 #define r7F7F   r9      /* constant 0x7f7f7f7f7f7f7f7f */
  45 #define rNEG    r10     /* ~(word in s1 | 0x7f7f7f7f7f7f7f7f) */
  46 #define rBITDIF r11     /* bits that differ in s1 & s2 words */
  47
  48         dcbt    0,rSTR1
  49         or      rTMP, rSTR2, rSTR1
  50         lis     r7F7F, 0x7f7f
  51         dcbt    0,rSTR2
  52         clrldi. rTMP, rTMP, 61
  53         cmpldi  cr1, rN, 0
  54         lis     rFEFE, -0x101
  55         bne     L(unaligned)
  56 /* We are doubleword alligned so set up for two loops.  first a double word
  57    loop, then fall into the byte loop if any residual.  */
  58         srdi.   rTMP, rN, 3
  59         clrldi  rN, rN, 61
  60         addi    rFEFE, rFEFE, -0x101
  61         addi    r7F7F, r7F7F, 0x7f7f
  62         cmpldi  cr1, rN, 0
  63         beq     L(unaligned)
  64
  65         mtctr   rTMP    /* Power4 wants mtctr 1st in dispatch group.  */
  66         ld      rWORD1, 0(rSTR1)
  67         ld      rWORD2, 0(rSTR2)
  68         sldi    rTMP, rFEFE, 32
  69         insrdi  r7F7F, r7F7F, 32, 0
  70         add     rFEFE, rFEFE, rTMP
  71         b       L(g1)
  72
  73 L(g0):
  74         ldu     rWORD1, 8(rSTR1)
  75         bne-    cr1, L(different)
  76         ldu     rWORD2, 8(rSTR2)
  77 L(g1):  add     rTMP, rFEFE, rWORD1
  78         nor     rNEG, r7F7F, rWORD1
  79         bdz     L(tail)
  80         and.    rTMP, rTMP, rNEG
  81         cmpd    cr1, rWORD1, rWORD2
  82         beq+    L(g0)
  83
  84 /* OK. We've hit the end of the string. We need to be careful that
  85    we don't compare two strings as different because of gunk beyond
  86    the end of the strings...  */
  87
  88 L(endstring):
  89         and     rTMP, r7F7F, rWORD1
  90         beq     cr1, L(equal)
  91         add     rTMP, rTMP, r7F7F
  92         xor.    rBITDIF, rWORD1, rWORD2
  93
  94         andc    rNEG, rNEG, rTMP
  95         blt-    L(highbit)
  96         cntlzd  rBITDIF, rBITDIF
  97         cntlzd  rNEG, rNEG
  98         addi    rNEG, rNEG, 7
  99         cmpd    cr1, rNEG, rBITDIF
 100         sub     rRTN, rWORD1, rWORD2
 101         blt-    cr1, L(equal)
 102         sradi   rRTN, rRTN, 63
 103         ori     rRTN, rRTN, 1
 104         blr
 105 L(equal):
 106         li      rRTN, 0
 107         blr
 108
 109 L(different):
 110         ldu     rWORD1, -8(rSTR1)
 111         xor.    rBITDIF, rWORD1, rWORD2
 112         sub     rRTN, rWORD1, rWORD2
 113         blt-    L(highbit)
 114         sradi   rRTN, rRTN, 63
 115         ori     rRTN, rRTN, 1
 116         blr
 117 L(highbit):
 118         srdi    rWORD2, rWORD2, 56
 119         srdi    rWORD1, rWORD1, 56
 120         sub     rRTN, rWORD1, rWORD2
 121         blr
 122
 123
 124 /* Oh well.  In this case, we just do a byte-by-byte comparison.  */
 125         .align 4
 126 L(tail):
 127         and.    rTMP, rTMP, rNEG
 128         cmpd    cr1, rWORD1, rWORD2
 129         bne-    L(endstring)
 130         addi    rSTR1, rSTR1, 8
 131         bne-    cr1, L(different)
 132         addi    rSTR2, rSTR2, 8
 133         cmpldi  cr1, rN, 0
 134 L(unaligned):
 135         mtctr   rN      /* Power4 wants mtctr 1st in dispatch group */
 136         ble     cr1, L(ux)
 137 L(uz):
 138         lbz     rWORD1, 0(rSTR1)
 139         lbz     rWORD2, 0(rSTR2)
 140         .align 4
 141 L(u1):
 142         cmpdi   cr1, rWORD1, 0
 143         bdz     L(u4)
 144         cmpd    rWORD1, rWORD2
 145         beq-    cr1, L(u4)
 146         bne-    L(u4)
 147         lbzu    rWORD3, 1(rSTR1)
 148         lbzu    rWORD4, 1(rSTR2)
 149         cmpdi   cr1, rWORD3, 0
 150         bdz     L(u3)
 151         cmpd    rWORD3, rWORD4
 152         beq-    cr1, L(u3)
 153         bne-    L(u3)
 154         lbzu    rWORD1, 1(rSTR1)
 155         lbzu    rWORD2, 1(rSTR2)
 156         cmpdi   cr1, rWORD1, 0
 157         bdz     L(u4)
 158         cmpd    rWORD1, rWORD2
 159         beq-    cr1, L(u4)
 160         bne-    L(u4)
 161         lbzu    rWORD3, 1(rSTR1)
 162         lbzu    rWORD4, 1(rSTR2)
 163         cmpdi   cr1, rWORD3, 0
 164         bdz     L(u3)
 165         cmpd    rWORD3, rWORD4
 166         beq-    cr1, L(u3)
 167         bne-    L(u3)
 168         lbzu    rWORD1, 1(rSTR1)
 169         lbzu    rWORD2, 1(rSTR2)
 170         b       L(u1)
 171
 172 L(u3):  sub     rRTN, rWORD3, rWORD4
 173         blr
 174 L(u4):  sub     rRTN, rWORD1, rWORD2
 175         blr
 176 L(ux):
 177         li      rRTN, 0
 178         blr
 179 END (BP_SYM (strncmp))
 180 libc_hidden_builtin_def (strncmp)
 181