memset.S 3.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147
  1. /* memset/bzero -- set memory area to CH/0
  2. Optimized version for x86-64.
  3. Copyright (C) 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
  4. This file is part of the GNU C Library.
  5. Contributed by Andreas Jaeger <aj@suse.de>.
  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. 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. You should have received a copy of the GNU Lesser General Public
  15. License along with the GNU C Library; if not, write to the Free
  16. Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  17. 02111-1307 USA. */
  18. #include "_glibc_inc.h"
  19. /* BEWARE: `#ifdef memset' means that memset is redefined as `bzero' */
  20. #define BZERO_P (defined memset)
  21. /* This is somehow experimental and could made dependend on the cache
  22. size. */
  23. #define LARGE $120000
  24. .text
  25. #if defined __PIC__ && !defined NOT_IN_libc && defined __UCLIBC_HAS_FORTIFY__
  26. ENTRY (__memset_chk)
  27. cmpq %rdx, %rcx
  28. jb HIDDEN_JUMPTARGET (__chk_fail)
  29. END (__memset_chk)
  30. #endif
  31. ENTRY (memset)
  32. #if BZERO_P
  33. mov %rsi,%rdx /* Adjust parameter. */
  34. xorl %esi,%esi /* Fill with 0s. */
  35. #endif
  36. cmp $0x7,%rdx /* Check for small length. */
  37. mov %rdi,%rcx /* Save ptr as return value. */
  38. jbe 7f
  39. #if BZERO_P
  40. mov %rsi,%r8 /* Just copy 0. */
  41. #else
  42. /* Populate 8 bit data to full 64-bit. */
  43. movabs $0x0101010101010101,%r8
  44. movzbl %sil,%eax
  45. imul %rax,%r8
  46. #endif
  47. test $0x7,%edi /* Check for alignment. */
  48. jz 2f
  49. /* Next 3 insns are 9 bytes total, make sure we decode them in one go */
  50. .p2align 4,,9
  51. 1:
  52. /* Align ptr to 8 byte. */
  53. mov %sil,(%rcx)
  54. dec %rdx
  55. inc %rcx
  56. test $0x7,%cl
  57. jnz 1b
  58. 2: /* Check for really large regions. */
  59. mov %rdx,%rax
  60. shr $0x6,%rax
  61. je 4f
  62. cmp LARGE, %rdx
  63. jae 11f
  64. /* Next 3 insns are 11 bytes total, make sure we decode them in one go */
  65. .p2align 4,,11
  66. 3:
  67. /* Fill 64 bytes. */
  68. mov %r8,(%rcx)
  69. mov %r8,0x8(%rcx)
  70. mov %r8,0x10(%rcx)
  71. mov %r8,0x18(%rcx)
  72. mov %r8,0x20(%rcx)
  73. mov %r8,0x28(%rcx)
  74. mov %r8,0x30(%rcx)
  75. mov %r8,0x38(%rcx)
  76. add $0x40,%rcx
  77. dec %rax
  78. jne 3b
  79. 4: /* Fill final bytes. */
  80. and $0x3f,%edx
  81. mov %rdx,%rax
  82. shr $0x3,%rax
  83. je 6f
  84. 5: /* First in chunks of 8 bytes. */
  85. mov %r8,(%rcx)
  86. add $0x8,%rcx
  87. dec %rax
  88. jne 5b
  89. 6:
  90. and $0x7,%edx
  91. 7:
  92. test %rdx,%rdx
  93. je 9f
  94. 8: /* And finally as bytes (up to 7). */
  95. mov %sil,(%rcx)
  96. inc %rcx
  97. dec %rdx
  98. jne 8b
  99. 9:
  100. #if BZERO_P
  101. /* nothing */
  102. #else
  103. /* Load result (only if used as memset). */
  104. mov %rdi,%rax /* start address of destination is result */
  105. #endif
  106. retq
  107. /* Next 3 insns are 14 bytes total, make sure we decode them in one go */
  108. .p2align 4,,14
  109. 11:
  110. /* Fill 64 bytes without polluting the cache. */
  111. /* We could use movntdq %xmm0,(%rcx) here to further
  112. speed up for large cases but let's not use XMM registers. */
  113. movnti %r8,(%rcx)
  114. movnti %r8,0x8(%rcx)
  115. movnti %r8,0x10(%rcx)
  116. movnti %r8,0x18(%rcx)
  117. movnti %r8,0x20(%rcx)
  118. movnti %r8,0x28(%rcx)
  119. movnti %r8,0x30(%rcx)
  120. movnti %r8,0x38(%rcx)
  121. add $0x40,%rcx
  122. dec %rax
  123. jne 11b
  124. jmp 4b
  125. END (memset)
  126. #if !BZERO_P
  127. libc_hidden_def(memset)
  128. #endif
  129. #if !BZERO_P && defined __PIC__ && !defined NOT_IN_libc && defined __UCLIBC_HAS_FORTIFY__
  130. strong_alias (__memset_chk, __memset_zero_constant_len_parameter)
  131. #endif