sysdeps/powerpc/powerpc64/power7/memset.S

   1 /* Optimized memset implementation for PowerPC64/POWER7.
   2    Copyright (C) 2010-2019 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
  22 /* void * [r3] memset (void *s [r3], int c [r4], size_t n [r5]));
  23    Returns 's'.  */
  24
  25 #ifndef MEMSET
  26 # define MEMSET memset
  27 #endif
  28         .machine power7
  29 ENTRY_TOCLESS (MEMSET, 5)
  30         CALL_MCOUNT 3
  31
  32 L(_memset):
  33         cmpldi  cr7,5,31
  34         cmpldi  cr6,5,8
  35         mr      10,3
  36
  37         /* Replicate byte to word.  */
  38         insrdi  4,4,8,48
  39         insrdi  4,4,16,32
  40         ble     cr6,L(small)    /* If length <= 8, use short copy code.  */
  41
  42         neg     0,3
  43         ble     cr7,L(medium)   /* If length < 32, use medium copy code.  */
  44
  45         andi.   11,10,7         /* Check alignment of SRC.  */
  46         insrdi  4,4,32,0        /* Replicate word to double word.  */
  47
  48         mr      12,5
  49         beq     L(big_aligned)
  50
  51         clrldi  0,0,61
  52         mtocrf  0x01,0
  53         subf    5,0,5
  54
  55         /* Get DST aligned to 8 bytes.  */
  56 1:      bf      31,2f
  57
  58         stb     4,0(10)
  59         addi    10,10,1
  60 2:      bf      30,4f
  61
  62         sth     4,0(10)
  63         addi    10,10,2
  64 4:      bf      29,L(big_aligned)
  65
  66         stw     4,0(10)
  67         addi    10,10,4
  68
  69         .align  4
  70 L(big_aligned):
  71
  72         cmpldi  cr5,5,255
  73         li      0,32
  74         dcbtst  0,10
  75         cmpldi  cr6,4,0
  76         srdi    9,5,3   /* Number of full doublewords remaining.  */
  77         crand   27,26,21
  78         mtocrf  0x01,9
  79         bt      27,L(huge)
  80
  81         /* From this point on, we'll copy 32+ bytes and the value
  82            isn't 0 (so we can't use dcbz).  */
  83
  84         srdi    8,5,5
  85         clrldi  11,5,61
  86         cmpldi  cr6,11,0
  87         cmpldi  cr1,9,4
  88         mtctr   8
  89
  90         /* Copy 1~3 doublewords so the main loop starts
  91         at a multiple of 32 bytes.  */
  92
  93         bf      30,1f
  94
  95         std     4,0(10)
  96         std     4,8(10)
  97         addi    10,10,16
  98         bf      31,L(big_loop)
  99
 100         std     4,0(10)
 101         addi    10,10,8
 102         mr      12,10
 103         blt     cr1,L(tail_bytes)
 104         b       L(big_loop)
 105
 106         .align  4
 107 1:      /* Copy 1 doubleword.  */
 108         bf      31,L(big_loop)
 109
 110         std     4,0(10)
 111         addi    10,10,8
 112
 113         /* Main aligned copy loop.  Copies 32-bytes at a time and
 114            ping-pong through r10 and r12 to avoid AGEN delays.  */
 115         .align  4
 116 L(big_loop):
 117         addi    12,10,32
 118         std     4,0(10)
 119         std     4,8(10)
 120         std     4,16(10)
 121         std     4,24(10)
 122         bdz     L(tail_bytes)
 123
 124         addi    10,10,64
 125         std     4,0(12)
 126         std     4,8(12)
 127         std     4,16(12)
 128         std     4,24(12)
 129         bdnz    L(big_loop)
 130
 131         mr      12,10
 132         b       L(tail_bytes)
 133
 134         .align  4
 135 L(tail_bytes):
 136
 137         /* Check for tail bytes.  */
 138         beqlr   cr6
 139
 140         clrldi  0,5,61
 141         mtocrf  0x01,0
 142
 143         /*  At this point we have a tail of 0-7 bytes and we know that the
 144         destination is doubleword-aligned.  */
 145 4:      /* Copy 4 bytes.  */
 146         bf      29,2f
 147
 148         stw     4,0(12)
 149         addi    12,12,4
 150 2:      /* Copy 2 bytes.  */
 151         bf      30,1f
 152
 153         sth     4,0(12)
 154         addi    12,12,2
 155 1:      /* Copy 1 byte.  */
 156         bflr    31
 157
 158         stb     4,0(12)
 159         blr
 160
 161         /* Special case when value is 0 and we have a long length to deal
 162            with.  Use dcbz to zero out 128-bytes at a time.  Before using
 163            dcbz though, we need to get the destination 128-bytes aligned.  */
 164         .align  4
 165 L(huge):
 166         andi.   11,10,127
 167         neg     0,10
 168         beq     L(huge_aligned)
 169
 170         clrldi  0,0,57
 171         subf    5,0,5
 172         srdi    0,0,3
 173         mtocrf  0x01,0
 174
 175         /* Get DST aligned to 128 bytes.  */
 176 8:      bf      28,4f
 177
 178         std     4,0(10)
 179         std     4,8(10)
 180         std     4,16(10)
 181         std     4,24(10)
 182         std     4,32(10)
 183         std     4,40(10)
 184         std     4,48(10)
 185         std     4,56(10)
 186         addi    10,10,64
 187         .align  4
 188 4:      bf      29,2f
 189
 190         std     4,0(10)
 191         std     4,8(10)
 192         std     4,16(10)
 193         std     4,24(10)
 194         addi    10,10,32
 195         .align  4
 196 2:      bf      30,1f
 197
 198         std     4,0(10)
 199         std     4,8(10)
 200         addi    10,10,16
 201         .align  4
 202 1:      bf      31,L(huge_aligned)
 203
 204         std     4,0(10)
 205         addi    10,10,8
 206
 207
 208 L(huge_aligned):
 209         srdi    8,5,7
 210         clrldi  11,5,57
 211         cmpldi  cr6,11,0
 212         mtctr   8
 213
 214         .align  4
 215 L(huge_loop):
 216         dcbz    0,10
 217         addi    10,10,128
 218         bdnz    L(huge_loop)
 219
 220         /* Check how many bytes are still left.  */
 221         beqlr   cr6
 222
 223         subf    9,3,10
 224         subf    5,9,12
 225         srdi    8,5,3
 226         cmpldi  cr6,8,0
 227         mtocrf  0x01,8
 228
 229         /* We have a tail o 1~127 bytes.  Copy up to 15 doublewords for
 230         speed.  We'll handle the resulting tail bytes later.  */
 231         beq     cr6,L(tail)
 232
 233 8:      bf      28,4f
 234
 235         std     4,0(10)
 236         std     4,8(10)
 237         std     4,16(10)
 238         std     4,24(10)
 239         std     4,32(10)
 240         std     4,40(10)
 241         std     4,48(10)
 242         std     4,56(10)
 243         addi    10,10,64
 244         .align  4
 245 4:      bf      29,2f
 246
 247         std     4,0(10)
 248         std     4,8(10)
 249         std     4,16(10)
 250         std     4,24(10)
 251         addi    10,10,32
 252         .align  4
 253 2:      bf      30,1f
 254
 255         std     4,0(10)
 256         std     4,8(10)
 257         addi    10,10,16
 258         .align  4
 259 1:      bf      31,L(tail)
 260
 261         std     4,0(10)
 262         addi    10,10,8
 263
 264         /* Handle the rest of the tail bytes here.  */
 265 L(tail):
 266         mtocrf  0x01,5
 267
 268         .align  4
 269 4:      bf      29,2f
 270
 271         stw     4,0(10)
 272         addi    10,10,4
 273         .align  4
 274 2:      bf      30,1f
 275
 276         sth     4,0(10)
 277         addi    10,10,2
 278         .align  4
 279 1:      bflr    31
 280
 281         stb     4,0(10)
 282         blr
 283
 284         /* Expanded tree to copy tail bytes without increments.  */
 285         .align  4
 286 L(copy_tail):
 287         bf      29,L(FXX)
 288
 289         stw     4,0(10)
 290         bf      30,L(TFX)
 291
 292         sth     4,4(10)
 293         bflr    31
 294
 295         stb     4,6(10)
 296         blr
 297
 298         .align  4
 299 L(FXX): bf      30,L(FFX)
 300
 301         sth     4,0(10)
 302         bflr    31
 303
 304         stb     4,2(10)
 305         blr
 306
 307         .align  4
 308 L(TFX): bflr    31
 309
 310         stb     4,4(10)
 311         blr
 312
 313         .align  4
 314 L(FFX): bflr    31
 315
 316         stb     4,0(10)
 317         blr
 318
 319         /* Handle copies of 9~31 bytes.  */
 320         .align  4
 321 L(medium):
 322         /* At least 9 bytes to go.  */
 323         andi.   11,10,3
 324         clrldi  0,0,62
 325         beq     L(medium_aligned)
 326
 327         /* Force 4-bytes alignment for DST.  */
 328         mtocrf  0x01,0
 329         subf    5,0,5
 330 1:      /* Copy 1 byte.  */
 331         bf      31,2f
 332
 333         stb     4,0(10)
 334         addi    10,10,1
 335 2:      /* Copy 2 bytes.  */
 336         bf      30,L(medium_aligned)
 337
 338         sth     4,0(10)
 339         addi    10,10,2
 340
 341         .align  4
 342 L(medium_aligned):
 343         /* At least 6 bytes to go, and DST is word-aligned.  */
 344         cmpldi  cr1,5,16
 345         mtocrf  0x01,5
 346         blt     cr1,8f
 347
 348         /* Copy 16 bytes.  */
 349         stw     4,0(10)
 350         stw     4,4(10)
 351         stw     4,8(10)
 352         stw     4,12(10)
 353         addi    10,10,16
 354 8:      /* Copy 8 bytes.  */
 355         bf      28,4f
 356
 357         stw     4,0(10)
 358         stw     4,4(10)
 359         addi    10,10,8
 360 4:      /* Copy 4 bytes.  */
 361         bf      29,2f
 362
 363         stw     4,0(10)
 364         addi    10,10,4
 365 2:      /* Copy 2-3 bytes.  */
 366         bf      30,1f
 367
 368         sth     4,0(10)
 369         addi    10,10,2
 370 1:      /* Copy 1 byte.  */
 371         bflr    31
 372
 373         stb     4,0(10)
 374         blr
 375
 376         /* Handles copies of 0~8 bytes.  */
 377         .align  4
 378 L(small):
 379         mtocrf  0x01,5
 380         bne     cr6,L(copy_tail)
 381
 382         stw     4,0(10)
 383         stw     4,4(10)
 384         blr
 385
 386 END_GEN_TB (MEMSET,TB_TOCLESS)
 387 libc_hidden_builtin_def (memset)
 388
 389 /* Copied from bzero.S to prevent the linker from inserting a stub
 390    between bzero and memset.  */
 391 ENTRY_TOCLESS (__bzero)
 392         CALL_MCOUNT 3
 393         mr      r5,r4
 394         li      r4,0
 395         b       L(_memset)
 396 END (__bzero)
 397 #ifndef __bzero
 398 weak_alias (__bzero, bzero)
 399 #endif