memset.S 3.1 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133
  1. /* SPDX-License-Identifier: GPL-2.0 */
  2. /*
  3. * linux/arch/alpha/lib/memset.S
  4. *
  5. * This is an efficient (and small) implementation of the C library "memset()"
  6. * function for the alpha.
  7. *
  8. * (C) Copyright 1996 Linus Torvalds
  9. *
  10. * This routine is "moral-ware": you are free to use it any way you wish, and
  11. * the only obligation I put on you is a moral one: if you make any improvements
  12. * to the routine, please send me your improvements for me to use similarly.
  13. *
  14. * The scheduling comments are according to the EV5 documentation (and done by
  15. * hand, so they might well be incorrect, please do tell me about it..)
  16. */
  17. #include <asm/export.h>
  18. .set noat
  19. .set noreorder
  20. .text
  21. .globl memset
  22. .globl __memset
  23. .globl ___memset
  24. .globl __memset16
  25. .globl __constant_c_memset
  26. .ent ___memset
  27. .align 5
  28. ___memset:
  29. .frame $30,0,$26,0
  30. .prologue 0
  31. and $17,255,$1 /* E1 */
  32. insbl $17,1,$17 /* .. E0 */
  33. bis $17,$1,$17 /* E0 (p-c latency, next cycle) */
  34. sll $17,16,$1 /* E1 (p-c latency, next cycle) */
  35. bis $17,$1,$17 /* E0 (p-c latency, next cycle) */
  36. sll $17,32,$1 /* E1 (p-c latency, next cycle) */
  37. bis $17,$1,$17 /* E0 (p-c latency, next cycle) */
  38. ldq_u $31,0($30) /* .. E1 */
  39. .align 5
  40. __constant_c_memset:
  41. addq $18,$16,$6 /* E0 */
  42. bis $16,$16,$0 /* .. E1 */
  43. xor $16,$6,$1 /* E0 */
  44. ble $18,end /* .. E1 */
  45. bic $1,7,$1 /* E0 */
  46. beq $1,within_one_quad /* .. E1 (note EV5 zero-latency forwarding) */
  47. and $16,7,$3 /* E0 */
  48. beq $3,aligned /* .. E1 (note EV5 zero-latency forwarding) */
  49. ldq_u $4,0($16) /* E0 */
  50. bis $16,$16,$5 /* .. E1 */
  51. insql $17,$16,$2 /* E0 */
  52. subq $3,8,$3 /* .. E1 */
  53. addq $18,$3,$18 /* E0 $18 is new count ($3 is negative) */
  54. mskql $4,$16,$4 /* .. E1 (and possible load stall) */
  55. subq $16,$3,$16 /* E0 $16 is new aligned destination */
  56. bis $2,$4,$1 /* .. E1 */
  57. bis $31,$31,$31 /* E0 */
  58. ldq_u $31,0($30) /* .. E1 */
  59. stq_u $1,0($5) /* E0 */
  60. bis $31,$31,$31 /* .. E1 */
  61. .align 4
  62. aligned:
  63. sra $18,3,$3 /* E0 */
  64. and $18,7,$18 /* .. E1 */
  65. bis $16,$16,$5 /* E0 */
  66. beq $3,no_quad /* .. E1 */
  67. .align 3
  68. loop:
  69. stq $17,0($5) /* E0 */
  70. subq $3,1,$3 /* .. E1 */
  71. addq $5,8,$5 /* E0 */
  72. bne $3,loop /* .. E1 */
  73. no_quad:
  74. bis $31,$31,$31 /* E0 */
  75. beq $18,end /* .. E1 */
  76. ldq $7,0($5) /* E0 */
  77. mskqh $7,$6,$2 /* .. E1 (and load stall) */
  78. insqh $17,$6,$4 /* E0 */
  79. bis $2,$4,$1 /* .. E1 */
  80. stq $1,0($5) /* E0 */
  81. ret $31,($26),1 /* .. E1 */
  82. .align 3
  83. within_one_quad:
  84. ldq_u $1,0($16) /* E0 */
  85. insql $17,$16,$2 /* E1 */
  86. mskql $1,$16,$4 /* E0 (after load stall) */
  87. bis $2,$4,$2 /* E0 */
  88. mskql $2,$6,$4 /* E0 */
  89. mskqh $1,$6,$2 /* .. E1 */
  90. bis $2,$4,$1 /* E0 */
  91. stq_u $1,0($16) /* E0 */
  92. end:
  93. ret $31,($26),1 /* E1 */
  94. .end ___memset
  95. EXPORT_SYMBOL(___memset)
  96. EXPORT_SYMBOL(__constant_c_memset)
  97. .align 5
  98. .ent __memset16
  99. __memset16:
  100. .prologue 0
  101. inswl $17,0,$1 /* E0 */
  102. inswl $17,2,$2 /* E0 */
  103. inswl $17,4,$3 /* E0 */
  104. or $1,$2,$1 /* .. E1 */
  105. inswl $17,6,$4 /* E0 */
  106. or $1,$3,$1 /* .. E1 */
  107. or $1,$4,$17 /* E0 */
  108. br __constant_c_memset /* .. E1 */
  109. .end __memset16
  110. EXPORT_SYMBOL(__memset16)
  111. memset = ___memset
  112. __memset = ___memset
  113. EXPORT_SYMBOL(memset)
  114. EXPORT_SYMBOL(__memset)