source/libs/gmp/gmp-src/mpn/arm64/mul_1.asm

   1 dnl  ARM64 mpn_mul_1
   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 define(`rp', `x0')
  40 define(`up', `x1')
  41 define(`n',  `x2')
  42 define(`v0', `x3')
  43
  44 ASM_START()
  45 PROLOGUE(mpn_mul_1)
  46         ldr     x12, [up], #8
  47         and     x6, n, #3
  48         and     n, n, #-4
  49         cbz     x6, L(fi0)
  50         cmp     x6, #2
  51         b.cc    L(fi1)
  52         b.eq    L(fi2)
  53
  54 L(fi3): mul     x8, x12, v0
  55         umulh   x13, x12, v0
  56         cmn     xzr, xzr
  57         b       L(L3)
  58 L(fi2): mul     x7, x12, v0
  59         umulh   x5, x12, v0
  60         cmn     xzr, xzr
  61         b       L(L2)
  62 L(fi0): mul     x9, x12, v0
  63         umulh   x5, x12, v0
  64         sub     n, n, #4
  65         cmn     xzr, xzr
  66         b       L(L0)
  67 L(fi1): mul     x10, x12, v0
  68         umulh   x13, x12, v0
  69         cmn     xzr, xzr
  70         cbz     n, L(end)
  71
  72 L(top): sub     n, n, #4
  73         ldr     x12, [up], #8
  74         mul     x6, x12, v0
  75         umulh   x5, x12, v0
  76         str     x10, [rp], #8
  77         adcs    x9, x6, x13
  78 L(L0):  ldr     x12, [up], #8
  79         mul     x6, x12, v0
  80         umulh   x13, x12, v0
  81         str     x9, [rp] ,#8
  82         adcs    x8, x6, x5
  83 L(L3):  ldr     x12, [up], #8
  84         mul     x6, x12, v0
  85         umulh   x5, x12, v0
  86         str     x8, [rp], #8
  87         adcs    x7, x6, x13
  88 L(L2):  ldr     x12, [up], #8
  89         mul     x6, x12, v0
  90         umulh   x13, x12, v0
  91         str     x7, [rp], #8
  92         adcs    x10, x6, x5
  93         cbnz    n, L(top)
  94
  95 L(end): str     x10, [rp]
  96         adc     x0, x13, xzr
  97         ret
  98 EPILOGUE()