ports/sysdeps/arm/memcpy.S

   1 /* Copyright (C) 2006-2014 Free Software Foundation, Inc.
   2    This file is part of the GNU C Library.
   3
   4    Contributed by MontaVista Software, Inc. (written by Nicolas Pitre)
   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 /* Thumb requires excessive IT insns here.  */
  21 #define NO_THUMB
  22 #include <sysdep.h>
  23 #include <arm-features.h>
  24
  25 /*
  26  * Data preload for architectures that support it (ARM V5TE and above)
  27  */
  28 #if (!defined (__ARM_ARCH_2__) && !defined (__ARM_ARCH_3__) \
  29      && !defined (__ARM_ARCH_3M__) && !defined (__ARM_ARCH_4__) \
  30      && !defined (__ARM_ARCH_4T__) && !defined (__ARM_ARCH_5__) \
  31      && !defined (__ARM_ARCH_5T__))
  32 #define PLD(code...)    code
  33 #else
  34 #define PLD(code...)
  35 #endif
  36
  37 /*
  38  * This can be used to enable code to cacheline align the source pointer.
  39  * Experiments on tested architectures (StrongARM and XScale) didn't show
  40  * this a worthwhile thing to do.  That might be different in the future.
  41  */
  42 //#define CALGN(code...)        code
  43 #define CALGN(code...)
  44
  45 /*
  46  * Endian independent macros for shifting bytes within registers.
  47  */
  48 #ifndef __ARMEB__
  49 #define PULL            lsr
  50 #define PUSH            lsl
  51 #else
  52 #define PULL            lsl
  53 #define PUSH            lsr
  54 #endif
  55
  56                 .text
  57                 .syntax unified
  58
  59 /* Prototype: void *memcpy(void *dest, const void *src, size_t n); */
  60
  61 ENTRY(memcpy)
  62
  63                 push    {r0, r4, lr}
  64                 cfi_adjust_cfa_offset (12)
  65                 cfi_rel_offset (r4, 4)
  66                 cfi_rel_offset (lr, 8)
  67
  68                 cfi_remember_state
  69
  70                 subs    r2, r2, #4
  71                 blt     8f
  72                 ands    ip, r0, #3
  73         PLD(    sfi_pld r1, #0                  )
  74                 bne     9f
  75                 ands    ip, r1, #3
  76                 bne     10f
  77
  78 1:              subs    r2, r2, #(28)
  79                 push    {r5 - r8}
  80                 cfi_adjust_cfa_offset (16)
  81                 cfi_rel_offset (r5, 0)
  82                 cfi_rel_offset (r6, 4)
  83                 cfi_rel_offset (r7, 8)
  84                 cfi_rel_offset (r8, 12)
  85                 blt     5f
  86
  87         CALGN(  ands    ip, r1, #31             )
  88         CALGN(  rsb     r3, ip, #32             )
  89         CALGN(  sbcsne  r4, r3, r2              )  @ C is always set here
  90         CALGN(  bcs     2f                      )
  91         CALGN(  adr     r4, 6f                  )
  92         CALGN(  subs    r2, r2, r3              )  @ C gets set
  93 #ifndef ARM_ALWAYS_BX
  94         CALGN(  add     pc, r4, ip, lsl #(ARM_BX_ALIGN_LOG2 - 2))
  95 #else
  96         CALGN(  add     r4, r4, ip, lsl #(ARM_BX_ALIGN_LOG2 - 2))
  97         CALGN(  bx      r4                      )
  98 #endif
  99
 100         PLD(    sfi_pld r1, #0                  )
 101 2:      PLD(    subs    r2, r2, #96             )
 102         PLD(    sfi_pld r1, #28                 )
 103         PLD(    blt     4f                      )
 104         PLD(    sfi_pld r1, #60                 )
 105         PLD(    sfi_pld r1, #92                 )
 106
 107 3:      PLD(    sfi_pld r1, #124                )
 108 4:              sfi_breg r1, \
 109                 ldmia   \B!, {r3, r4, r5, r6, r7, r8, ip, lr}
 110                 subs    r2, r2, #32
 111                 sfi_breg r0, \
 112                 stmia   \B!, {r3, r4, r5, r6, r7, r8, ip, lr}
 113                 bge     3b
 114         PLD(    cmn     r2, #96                 )
 115         PLD(    bge     4b                      )
 116
 117 5:              ands    ip, r2, #28
 118                 rsb     ip, ip, #32
 119 #ifndef ARM_ALWAYS_BX
 120                 /* C is always clear here.  */
 121                 addne   pc, pc, ip, lsl #(ARM_BX_ALIGN_LOG2 - 2)
 122                 b       7f
 123 #else
 124                 beq     7f
 125                 push    {r10}
 126                 cfi_adjust_cfa_offset (4)
 127                 cfi_rel_offset (r10, 0)
 128                 add     r10, pc, ip, lsl #(ARM_BX_ALIGN_LOG2 - 2)
 129                 bx      r10
 130 #endif
 131                 .p2align ARM_BX_ALIGN_LOG2
 132 6:              nop
 133                 .p2align ARM_BX_ALIGN_LOG2
 134                 sfi_breg r1, \
 135                 ldr     r3, [\B], #4
 136                 .p2align ARM_BX_ALIGN_LOG2
 137                 sfi_breg r1, \
 138                 ldr     r4, [\B], #4
 139                 .p2align ARM_BX_ALIGN_LOG2
 140                 sfi_breg r1, \
 141                 ldr     r5, [\B], #4
 142                 .p2align ARM_BX_ALIGN_LOG2
 143                 sfi_breg r1, \
 144                 ldr     r6, [\B], #4
 145                 .p2align ARM_BX_ALIGN_LOG2
 146                 sfi_breg r1, \
 147                 ldr     r7, [\B], #4
 148                 .p2align ARM_BX_ALIGN_LOG2
 149                 sfi_breg r1, \
 150                 ldr     r8, [\B], #4
 151                 .p2align ARM_BX_ALIGN_LOG2
 152                 sfi_breg r1, \
 153                 ldr     lr, [\B], #4
 154
 155 #ifndef ARM_ALWAYS_BX
 156                 add     pc, pc, ip, lsl #(ARM_BX_ALIGN_LOG2 - 2)
 157                 nop
 158 #else
 159                 add     r10, pc, ip, lsl #(ARM_BX_ALIGN_LOG2 - 2)
 160                 bx      r10
 161 #endif
 162                 .p2align ARM_BX_ALIGN_LOG2
 163                 nop
 164                 .p2align ARM_BX_ALIGN_LOG2
 165                 sfi_breg r0, \
 166                 str     r3, [\B], #4
 167                 .p2align ARM_BX_ALIGN_LOG2
 168                 sfi_breg r0, \
 169                 str     r4, [\B], #4
 170                 .p2align ARM_BX_ALIGN_LOG2
 171                 sfi_breg r0, \
 172                 str     r5, [\B], #4
 173                 .p2align ARM_BX_ALIGN_LOG2
 174                 sfi_breg r0, \
 175                 str     r6, [\B], #4
 176                 .p2align ARM_BX_ALIGN_LOG2
 177                 sfi_breg r0, \
 178                 str     r7, [\B], #4
 179                 .p2align ARM_BX_ALIGN_LOG2
 180                 sfi_breg r0, \
 181                 str     r8, [\B], #4
 182                 .p2align ARM_BX_ALIGN_LOG2
 183                 sfi_breg r0, \
 184                 str     lr, [\B], #4
 185
 186 #ifdef ARM_ALWAYS_BX
 187                 pop     {r10}
 188                 cfi_adjust_cfa_offset (-4)
 189                 cfi_restore (r10)
 190 #endif
 191
 192         CALGN(  bcs     2b                      )
 193
 194 7:              pop     {r5 - r8}
 195                 cfi_adjust_cfa_offset (-16)
 196                 cfi_restore (r5)
 197                 cfi_restore (r6)
 198                 cfi_restore (r7)
 199                 cfi_restore (r8)
 200
 201 8:              movs    r2, r2, lsl #31
 202                 sfi_breg r1, \
 203                 ldrbne  r3, [\B], #1
 204                 sfi_breg r1, \
 205                 ldrbcs  r4, [\B], #1
 206                 sfi_breg r1, \
 207                 ldrbcs  ip, [\B]
 208                 sfi_breg r0, \
 209                 strbne  r3, [\B], #1
 210                 sfi_breg r0, \
 211                 strbcs  r4, [\B], #1
 212                 sfi_breg r0, \
 213                 strbcs  ip, [\B]
 214
 215 #if ((defined (__ARM_ARCH_4T__) && defined(__THUMB_INTERWORK__)) \
 216      || defined (ARM_ALWAYS_BX))
 217                 pop     {r0, r4, lr}
 218                 cfi_adjust_cfa_offset (-12)
 219                 cfi_restore (r4)
 220                 cfi_restore (lr)
 221                 bx      lr
 222 #else
 223                 pop     {r0, r4, pc}
 224 #endif
 225
 226                 cfi_restore_state
 227
 228 9:              rsb     ip, ip, #4
 229                 cmp     ip, #2
 230                 sfi_breg r1, \
 231                 ldrbgt  r3, [\B], #1
 232                 sfi_breg r1, \
 233                 ldrbge  r4, [\B], #1
 234                 sfi_breg r1, \
 235                 ldrb    lr, [\B], #1
 236                 sfi_breg r0, \
 237                 strbgt  r3, [\B], #1
 238                 sfi_breg r0, \
 239                 strbge  r4, [\B], #1
 240                 subs    r2, r2, ip
 241                 sfi_breg r0, \
 242                 strb    lr, [\B], #1
 243                 blt     8b
 244                 ands    ip, r1, #3
 245                 beq     1b
 246
 247 10:             bic     r1, r1, #3
 248                 cmp     ip, #2
 249                 sfi_breg r1, \
 250                 ldr     lr, [\B], #4
 251                 beq     17f
 252                 bgt     18f
 253
 254
 255                 .macro  forward_copy_shift pull push
 256
 257                 subs    r2, r2, #28
 258                 blt     14f
 259
 260         CALGN(  ands    ip, r1, #31             )
 261         CALGN(  rsb     ip, ip, #32             )
 262         CALGN(  sbcsne  r4, ip, r2              )  @ C is always set here
 263         CALGN(  subcc   r2, r2, ip              )
 264         CALGN(  bcc     15f                     )
 265
 266 11:             push    {r5 - r8, r10}
 267                 cfi_adjust_cfa_offset (20)
 268                 cfi_rel_offset (r5, 0)
 269                 cfi_rel_offset (r6, 4)
 270                 cfi_rel_offset (r7, 8)
 271                 cfi_rel_offset (r8, 12)
 272                 cfi_rel_offset (r10, 16)
 273
 274         PLD(    sfi_pld r1, #0                  )
 275         PLD(    subs    r2, r2, #96             )
 276         PLD(    sfi_pld r1, #28                 )
 277         PLD(    blt     13f                     )
 278         PLD(    sfi_pld r1, #60                 )
 279         PLD(    sfi_pld r1, #92                 )
 280
 281 12:     PLD(    sfi_pld r1, #124                )
 282 13:             sfi_breg r1, \
 283                 ldmia   \B!, {r4, r5, r6, r7}
 284                 mov     r3, lr, PULL #\pull
 285                 subs    r2, r2, #32
 286                 sfi_breg r1, \
 287                 ldmia   \B!, {r8, r10, ip, lr}
 288                 orr     r3, r3, r4, PUSH #\push
 289                 mov     r4, r4, PULL #\pull
 290                 orr     r4, r4, r5, PUSH #\push
 291                 mov     r5, r5, PULL #\pull
 292                 orr     r5, r5, r6, PUSH #\push
 293                 mov     r6, r6, PULL #\pull
 294                 orr     r6, r6, r7, PUSH #\push
 295                 mov     r7, r7, PULL #\pull
 296                 orr     r7, r7, r8, PUSH #\push
 297                 mov     r8, r8, PULL #\pull
 298                 orr     r8, r8, r10, PUSH #\push
 299                 mov     r10, r10, PULL #\pull
 300                 orr     r10, r10, ip, PUSH #\push
 301                 mov     ip, ip, PULL #\pull
 302                 orr     ip, ip, lr, PUSH #\push
 303                 sfi_breg r0, \
 304                 stmia   \B!, {r3, r4, r5, r6, r7, r8, r10, ip}
 305                 bge     12b
 306         PLD(    cmn     r2, #96                 )
 307         PLD(    bge     13b                     )
 308
 309                 pop     {r5 - r8, r10}
 310                 cfi_adjust_cfa_offset (-20)
 311                 cfi_restore (r5)
 312                 cfi_restore (r6)
 313                 cfi_restore (r7)
 314                 cfi_restore (r8)
 315                 cfi_restore (r10)
 316
 317 14:             ands    ip, r2, #28
 318                 beq     16f
 319
 320 15:             mov     r3, lr, PULL #\pull
 321                 sfi_breg r1, \
 322                 ldr     lr, [\B], #4
 323                 subs    ip, ip, #4
 324                 orr     r3, r3, lr, PUSH #\push
 325                 sfi_breg r0, \
 326                 str     r3, [\B], #4
 327                 bgt     15b
 328         CALGN(  cmp     r2, #0                  )
 329         CALGN(  bge     11b                     )
 330
 331 16:             sub     r1, r1, #(\push / 8)
 332                 b       8b
 333
 334                 .endm
 335
 336
 337                 forward_copy_shift      pull=8  push=24
 338
 339 17:             forward_copy_shift      pull=16 push=16
 340
 341 18:             forward_copy_shift      pull=24 push=8
 342
 343 END(memcpy)
 344 libc_hidden_builtin_def (memcpy)