divu.S 2.5 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586
  1. ;; SPDX-License-Identifier: GPL-2.0-or-later
  2. ;; Copyright 2010 Free Software Foundation, Inc.
  3. ;; Contributed by Bernd Schmidt <bernds@codesourcery.com>.
  4. ;;
  5. #include <linux/linkage.h>
  6. ;; ABI considerations for the divide functions
  7. ;; The following registers are call-used:
  8. ;; __c6xabi_divi A0,A1,A2,A4,A6,B0,B1,B2,B4,B5
  9. ;; __c6xabi_divu A0,A1,A2,A4,A6,B0,B1,B2,B4
  10. ;; __c6xabi_remi A1,A2,A4,A5,A6,B0,B1,B2,B4
  11. ;; __c6xabi_remu A1,A4,A5,A7,B0,B1,B2,B4
  12. ;;
  13. ;; In our implementation, divu and remu are leaf functions,
  14. ;; while both divi and remi call into divu.
  15. ;; A0 is not clobbered by any of the functions.
  16. ;; divu does not clobber B2 either, which is taken advantage of
  17. ;; in remi.
  18. ;; divi uses B5 to hold the original return address during
  19. ;; the call to divu.
  20. ;; remi uses B2 and A5 to hold the input values during the
  21. ;; call to divu. It stores B3 in on the stack.
  22. .text
  23. ENTRY(__c6xabi_divu)
  24. ;; We use a series of up to 31 subc instructions. First, we find
  25. ;; out how many leading zero bits there are in the divisor. This
  26. ;; gives us both a shift count for aligning (shifting) the divisor
  27. ;; to the, and the number of times we have to execute subc.
  28. ;; At the end, we have both the remainder and most of the quotient
  29. ;; in A4. The top bit of the quotient is computed first and is
  30. ;; placed in A2.
  31. ;; Return immediately if the dividend is zero.
  32. mv .s2x A4, B1
  33. [B1] lmbd .l2 1, B4, B1
  34. || [!B1] b .s2 B3 ; RETURN A
  35. || [!B1] mvk .d2 1, B4
  36. mv .l1x B1, A6
  37. || shl .s2 B4, B1, B4
  38. ;; The loop performs a maximum of 28 steps, so we do the
  39. ;; first 3 here.
  40. cmpltu .l1x A4, B4, A2
  41. [!A2] sub .l1x A4, B4, A4
  42. || shru .s2 B4, 1, B4
  43. || xor .s1 1, A2, A2
  44. shl .s1 A2, 31, A2
  45. || [B1] subc .l1x A4,B4,A4
  46. || [B1] add .s2 -1, B1, B1
  47. [B1] subc .l1x A4,B4,A4
  48. || [B1] add .s2 -1, B1, B1
  49. ;; RETURN A may happen here (note: must happen before the next branch)
  50. _divu_loop:
  51. cmpgt .l2 B1, 7, B0
  52. || [B1] subc .l1x A4,B4,A4
  53. || [B1] add .s2 -1, B1, B1
  54. [B1] subc .l1x A4,B4,A4
  55. || [B1] add .s2 -1, B1, B1
  56. || [B0] b .s1 _divu_loop
  57. [B1] subc .l1x A4,B4,A4
  58. || [B1] add .s2 -1, B1, B1
  59. [B1] subc .l1x A4,B4,A4
  60. || [B1] add .s2 -1, B1, B1
  61. [B1] subc .l1x A4,B4,A4
  62. || [B1] add .s2 -1, B1, B1
  63. [B1] subc .l1x A4,B4,A4
  64. || [B1] add .s2 -1, B1, B1
  65. [B1] subc .l1x A4,B4,A4
  66. || [B1] add .s2 -1, B1, B1
  67. ;; loop backwards branch happens here
  68. ret .s2 B3
  69. || mvk .s1 32, A1
  70. sub .l1 A1, A6, A6
  71. shl .s1 A4, A6, A4
  72. shru .s1 A4, 1, A4
  73. || sub .l1 A6, 1, A6
  74. or .l1 A2, A4, A4
  75. shru .s1 A4, A6, A4
  76. nop
  77. ENDPROC(__c6xabi_divu)