strcpy.S 3.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150
  1. /* Optimized strcpy for Xtensa.
  2. Copyright (C) 2001, 2007 Free Software Foundation, Inc.
  3. This file is part of the GNU C Library.
  4. The GNU C Library is free software; you can redistribute it and/or
  5. modify it under the terms of the GNU Lesser General Public
  6. License as published by the Free Software Foundation; either
  7. version 2.1 of the License, or (at your option) any later version.
  8. The GNU C Library is distributed in the hope that it will be useful,
  9. but WITHOUT ANY WARRANTY; without even the implied warranty of
  10. MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11. Lesser General Public License for more details.
  12. You should have received a copy of the GNU Lesser General Public
  13. License along with the GNU C Library; if not, write to the Free
  14. Software Foundation, Inc., 51 Franklin Street - Fifth Floor,
  15. Boston, MA 02110-1301, USA. */
  16. #include "../../sysdeps/linux/xtensa/sysdep.h"
  17. #include <bits/xtensa-config.h>
  18. #ifdef __XTENSA_EB__
  19. #define MASK0 0xff000000
  20. #define MASK1 0x00ff0000
  21. #define MASK2 0x0000ff00
  22. #define MASK3 0x000000ff
  23. #else
  24. #define MASK0 0x000000ff
  25. #define MASK1 0x0000ff00
  26. #define MASK2 0x00ff0000
  27. #define MASK3 0xff000000
  28. #endif
  29. .text
  30. ENTRY (strcpy)
  31. /* a2 = dst, a3 = src */
  32. mov a10, a2 /* leave dst in return value register */
  33. movi a4, MASK0
  34. movi a5, MASK1
  35. movi a6, MASK2
  36. movi a7, MASK3
  37. bbsi.l a3, 0, .Lsrc1mod2
  38. bbsi.l a3, 1, .Lsrc2mod4
  39. .Lsrcaligned:
  40. /* Check if the destination is aligned. */
  41. movi a8, 3
  42. bnone a10, a8, .Laligned
  43. j .Ldstunaligned
  44. .Lsrc1mod2: /* src address is odd */
  45. l8ui a8, a3, 0 /* get byte 0 */
  46. addi a3, a3, 1 /* advance src pointer */
  47. s8i a8, a10, 0 /* store byte 0 */
  48. beqz a8, 1f /* if byte 0 is zero */
  49. addi a10, a10, 1 /* advance dst pointer */
  50. bbci.l a3, 1, .Lsrcaligned /* if src is now word-aligned */
  51. .Lsrc2mod4: /* src address is 2 mod 4 */
  52. l8ui a8, a3, 0 /* get byte 0 */
  53. /* 1-cycle interlock */
  54. s8i a8, a10, 0 /* store byte 0 */
  55. beqz a8, 1f /* if byte 0 is zero */
  56. l8ui a8, a3, 1 /* get byte 0 */
  57. addi a3, a3, 2 /* advance src pointer */
  58. s8i a8, a10, 1 /* store byte 0 */
  59. addi a10, a10, 2 /* advance dst pointer */
  60. bnez a8, .Lsrcaligned
  61. 1: retw
  62. /* dst is word-aligned; src is word-aligned. */
  63. .align 4
  64. #if XCHAL_HAVE_LOOPS
  65. /* (2 mod 4) alignment for loop instruction */
  66. .Laligned:
  67. _movi.n a8, 0 /* set up for the maximum loop count */
  68. loop a8, .Lz3 /* loop forever (almost anyway) */
  69. l32i a8, a3, 0 /* get word from src */
  70. addi a3, a3, 4 /* advance src pointer */
  71. bnone a8, a4, .Lz0 /* if byte 0 is zero */
  72. bnone a8, a5, .Lz1 /* if byte 1 is zero */
  73. bnone a8, a6, .Lz2 /* if byte 2 is zero */
  74. s32i a8, a10, 0 /* store word to dst */
  75. bnone a8, a7, .Lz3 /* if byte 3 is zero */
  76. addi a10, a10, 4 /* advance dst pointer */
  77. #else /* !XCHAL_HAVE_LOOPS */
  78. 1: addi a10, a10, 4 /* advance dst pointer */
  79. .Laligned:
  80. l32i a8, a3, 0 /* get word from src */
  81. addi a3, a3, 4 /* advance src pointer */
  82. bnone a8, a4, .Lz0 /* if byte 0 is zero */
  83. bnone a8, a5, .Lz1 /* if byte 1 is zero */
  84. bnone a8, a6, .Lz2 /* if byte 2 is zero */
  85. s32i a8, a10, 0 /* store word to dst */
  86. bany a8, a7, 1b /* if byte 3 is zero */
  87. #endif /* !XCHAL_HAVE_LOOPS */
  88. .Lz3: /* Byte 3 is zero. */
  89. retw
  90. .Lz0: /* Byte 0 is zero. */
  91. #ifdef __XTENSA_EB__
  92. movi a8, 0
  93. #endif
  94. s8i a8, a10, 0
  95. retw
  96. .Lz1: /* Byte 1 is zero. */
  97. #ifdef __XTENSA_EB__
  98. extui a8, a8, 16, 16
  99. #endif
  100. s16i a8, a10, 0
  101. retw
  102. .Lz2: /* Byte 2 is zero. */
  103. #ifdef __XTENSA_EB__
  104. extui a8, a8, 16, 16
  105. #endif
  106. s16i a8, a10, 0
  107. movi a8, 0
  108. s8i a8, a10, 2
  109. retw
  110. .align 4
  111. /* (2 mod 4) alignment for loop instruction */
  112. .Ldstunaligned:
  113. #if XCHAL_HAVE_LOOPS
  114. _movi.n a8, 0 /* set up for the maximum loop count */
  115. loop a8, 2f /* loop forever (almost anyway) */
  116. #endif
  117. 1: l8ui a8, a3, 0
  118. addi a3, a3, 1
  119. s8i a8, a10, 0
  120. addi a10, a10, 1
  121. #if XCHAL_HAVE_LOOPS
  122. beqz a8, 2f
  123. #else
  124. bnez a8, 1b
  125. #endif
  126. 2: retw
  127. libc_hidden_def (strcpy)