sysdeps/powerpc/powerpc64/strcpy.S

   1 /* Optimized strcpy 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 /* char * [r3] strcpy (char *dest [r3], const char *src [r4])  */
  24
  25 #ifdef USE_AS_STPCPY
  26 # define FUNC_NAME __stpcpy
  27 #else
  28 # define FUNC_NAME strcpy
  29 #endif
  30
  31 EALIGN (FUNC_NAME, 4, 0)
  32         CALL_MCOUNT 2
  33
  34 #define rTMP    r0
  35 #ifdef USE_AS_STPCPY
  36 #define rRTN    r3      /* pointer to previous word/doubleword in dest */
  37 #else
  38 #define rRTN    r12     /* pointer to previous word/doubleword in dest */
  39 #endif
  40 #define rSRC    r4      /* pointer to previous word/doubleword in src */
  41 #define rWORD   r6      /* current word from src */
  42 #define rFEFE   r7      /* constant 0xfefefeff | 0xfefefefefefefeff */
  43 #define r7F7F   r8      /* constant 0x7f7f7f7f | 0x7f7f7f7f7f7f7f7f */
  44 #define rNEG    r9      /* ~(word in s1 | r7F7F) */
  45 #define rALT    r10     /* alternate word from src */
  46
  47 #ifndef USE_AS_STPCPY
  48 /* Save the dst pointer to use as return value.  */
  49         mr      rRTN, r3
  50 #endif
  51         or      rTMP, rSRC, rRTN
  52         clrldi. rTMP, rTMP, 61
  53         bne     L(check_word_alignment)
  54
  55 /* For doubleword aligned memory, operate using doubleword load and stores.  */
  56         addi    rRTN, rRTN, -8
  57
  58         lis     rFEFE, -0x101
  59         lis     r7F7F, 0x7f7f
  60         ld      rWORD, 0(rSRC)
  61         addi    rFEFE, rFEFE, -0x101
  62         addi    r7F7F, r7F7F, 0x7f7f
  63         sldi    rTMP, rFEFE, 32
  64         insrdi  r7F7F, r7F7F, 32, 0
  65         add     rFEFE, rFEFE, rTMP
  66         b       L(g2)
  67
  68 L(g0):  ldu     rALT, 8(rSRC)
  69         stdu    rWORD, 8(rRTN)
  70         add     rTMP, rFEFE, rALT
  71         nor     rNEG, r7F7F, rALT
  72         and.    rTMP, rTMP, rNEG
  73         bne-    L(g1)
  74         ldu     rWORD, 8(rSRC)
  75         stdu    rALT, 8(rRTN)
  76 L(g2):  add     rTMP, rFEFE, rWORD
  77         nor     rNEG, r7F7F, rWORD
  78         and.    rTMP, rTMP, rNEG
  79         beq+    L(g0)
  80
  81         mr      rALT, rWORD
  82 /* We've hit the end of the string.  Do the rest byte-by-byte.  */
  83 L(g1):
  84 #ifdef __LITTLE_ENDIAN__
  85         extrdi. rTMP, rALT, 8, 56
  86         stbu    rALT, 8(rRTN)
  87         beqlr-
  88         extrdi. rTMP, rALT, 8, 48
  89         stbu    rTMP, 1(rRTN)
  90         beqlr-
  91         extrdi. rTMP, rALT, 8, 40
  92         stbu    rTMP, 1(rRTN)
  93         beqlr-
  94         extrdi. rTMP, rALT, 8, 32
  95         stbu    rTMP, 1(rRTN)
  96         beqlr-
  97         extrdi. rTMP, rALT, 8, 24
  98         stbu    rTMP, 1(rRTN)
  99         beqlr-
 100         extrdi. rTMP, rALT, 8, 16
 101         stbu    rTMP, 1(rRTN)
 102         beqlr-
 103         extrdi. rTMP, rALT, 8, 8
 104         stbu    rTMP, 1(rRTN)
 105         beqlr-
 106         extrdi  rTMP, rALT, 8, 0
 107         stbu    rTMP, 1(rRTN)
 108 #else
 109         extrdi. rTMP, rALT, 8, 0
 110         stbu    rTMP, 8(rRTN)
 111         beqlr-
 112         extrdi. rTMP, rALT, 8, 8
 113         stbu    rTMP, 1(rRTN)
 114         beqlr-
 115         extrdi. rTMP, rALT, 8, 16
 116         stbu    rTMP, 1(rRTN)
 117         beqlr-
 118         extrdi. rTMP, rALT, 8, 24
 119         stbu    rTMP, 1(rRTN)
 120         beqlr-
 121         extrdi. rTMP, rALT, 8, 32
 122         stbu    rTMP, 1(rRTN)
 123         beqlr
 124         extrdi. rTMP, rALT, 8, 40
 125         stbu    rTMP, 1(rRTN)
 126         beqlr-
 127         extrdi. rTMP, rALT, 8, 48
 128         stbu    rTMP, 1(rRTN)
 129         beqlr-
 130         stbu    rALT, 1(rRTN)
 131 #endif
 132         blr
 133
 134 L(check_word_alignment):
 135         clrldi. rTMP, rTMP, 62
 136         bne     L(unaligned)
 137
 138 /* For word aligned memory, operate using word load and stores.  */
 139         addi    rRTN, rRTN, -4
 140
 141         lis     rFEFE, -0x101
 142         lis     r7F7F, 0x7f7f
 143         lwz     rWORD, 0(rSRC)
 144         addi    rFEFE, rFEFE, -0x101
 145         addi    r7F7F, r7F7F, 0x7f7f
 146         b       L(g5)
 147
 148 L(g3):  lwzu    rALT, 4(rSRC)
 149         stwu    rWORD, 4(rRTN)
 150         add     rTMP, rFEFE, rALT
 151         nor     rNEG, r7F7F, rALT
 152         and.    rTMP, rTMP, rNEG
 153         bne-    L(g4)
 154         lwzu    rWORD, 4(rSRC)
 155         stwu    rALT, 4(rRTN)
 156 L(g5):  add     rTMP, rFEFE, rWORD
 157         nor     rNEG, r7F7F, rWORD
 158         and.    rTMP, rTMP, rNEG
 159         beq+    L(g3)
 160
 161         mr      rALT, rWORD
 162 /* We've hit the end of the string.  Do the rest byte-by-byte.  */
 163 L(g4):
 164 #ifdef __LITTLE_ENDIAN__
 165         rlwinm. rTMP, rALT, 0, 24, 31
 166         stbu    rALT, 4(rRTN)
 167         beqlr-
 168         rlwinm. rTMP, rALT, 24, 24, 31
 169         stbu    rTMP, 1(rRTN)
 170         beqlr-
 171         rlwinm. rTMP, rALT, 16, 24, 31
 172         stbu    rTMP, 1(rRTN)
 173         beqlr-
 174         rlwinm  rTMP, rALT, 8, 24, 31
 175         stbu    rTMP, 1(rRTN)
 176 #else
 177         rlwinm. rTMP, rALT, 8, 24, 31
 178         stbu    rTMP, 4(rRTN)
 179         beqlr-
 180         rlwinm. rTMP, rALT, 16, 24, 31
 181         stbu    rTMP, 1(rRTN)
 182         beqlr-
 183         rlwinm. rTMP, rALT, 24, 24, 31
 184         stbu    rTMP, 1(rRTN)
 185         beqlr-
 186         stbu    rALT, 1(rRTN)
 187 #endif
 188         blr
 189
 190 /* Oh well.  In this case, we just do a byte-by-byte copy.  */
 191         .align 4
 192         nop
 193 L(unaligned):
 194         lbz     rWORD, 0(rSRC)
 195         addi    rRTN, rRTN, -1
 196         cmpwi   rWORD, 0
 197         beq-    L(u2)
 198
 199 L(u0):  lbzu    rALT, 1(rSRC)
 200         stbu    rWORD, 1(rRTN)
 201         cmpwi   rALT, 0
 202         beq-    L(u1)
 203         nop             /* Let 601 load start of loop.  */
 204         lbzu    rWORD, 1(rSRC)
 205         stbu    rALT, 1(rRTN)
 206         cmpwi   rWORD, 0
 207         bne+    L(u0)
 208 L(u2):  stbu    rWORD, 1(rRTN)
 209         blr
 210 L(u1):  stbu    rALT, 1(rRTN)
 211         blr
 212 END (FUNC_NAME)
 213
 214 #ifndef USE_AS_STPCPY
 215 libc_hidden_builtin_def (strcpy)
 216 #endif