source/libs/gmp/gmp-src/mpn/arm64/invert_limb.asm

   1 dnl  ARM64 mpn_invert_limb -- Invert a normalized limb.
   2
   3 dnl  Contributed to the GNU project by Torbjörn Granlund.
   4
   5 dnl  Copyright 2013 Free Software Foundation, Inc.
   6
   7 dnl  This file is part of the GNU MP Library.
   8 dnl
   9 dnl  The GNU MP Library is free software; you can redistribute it and/or modify
  10 dnl  it under the terms of either:
  11 dnl
  12 dnl    * the GNU Lesser General Public License as published by the Free
  13 dnl      Software Foundation; either version 3 of the License, or (at your
  14 dnl      option) any later version.
  15 dnl
  16 dnl  or
  17 dnl
  18 dnl    * the GNU General Public License as published by the Free Software
  19 dnl      Foundation; either version 2 of the License, or (at your option) any
  20 dnl      later version.
  21 dnl
  22 dnl  or both in parallel, as here.
  23 dnl
  24 dnl  The GNU MP Library is distributed in the hope that it will be useful, but
  25 dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  26 dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  27 dnl  for more details.
  28 dnl
  29 dnl  You should have received copies of the GNU General Public License and the
  30 dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
  31 dnl  see https://www.gnu.org/licenses/.
  32
  33 include(`../config.m4')
  34
  35 C            cycles/limb
  36 C Cortex-A53     ?
  37 C Cortex-A57     ?
  38
  39 C Compiler generated, mildly edited.  Could surely be further optimised.
  40
  41 ASM_START()
  42 PROLOGUE(mpn_invert_limb)
  43         lsr     x2, x0, #54
  44         adrp    x1, approx_tab
  45         and     x2, x2, #0x1fe
  46         add     x1, x1, :lo12:approx_tab
  47         ldrh    w3, [x1,x2]
  48         lsr     x4, x0, #24
  49         add     x4, x4, #1
  50         ubfiz   x2, x3, #11, #16
  51         umull   x3, w3, w3
  52         mul     x3, x3, x4
  53         sub     x2, x2, #1
  54         sub     x2, x2, x3, lsr #40
  55         lsl     x3, x2, #60
  56         mul     x1, x2, x2
  57         msub    x1, x1, x4, x3
  58         lsl     x2, x2, #13
  59         add     x1, x2, x1, lsr #47
  60         and     x2, x0, #1
  61         neg     x3, x2
  62         and     x3, x3, x1, lsr #1
  63         add     x2, x2, x0, lsr #1
  64         msub    x2, x1, x2, x3
  65         umulh   x2, x2, x1
  66         lsl     x1, x1, #31
  67         add     x1, x1, x2, lsr #1
  68         mul     x3, x1, x0
  69         umulh   x2, x1, x0
  70         adds    x4, x3, x0
  71         adc     x0, x2, x0
  72         sub     x0, x1, x0
  73         ret
  74 EPILOGUE()
  75
  76         RODATA
  77         ALIGN(2)
  78         TYPE(   approx_tab, object)
  79         SIZE(   approx_tab, 512)
  80 approx_tab:
  81 forloop(i,256,512-1,dnl
  82 `       .hword  eval(0x7fd00/i)
  83 ')dnl