source/libs/gmp/gmp-src/mpn/arm/v6/addmul_2.asm

   1 dnl  ARM mpn_addmul_2.
   2
   3 dnl  Contributed to the GNU project by Torbjörn Granlund.
   4
   5 dnl  Copyright 2012, 2013, 2015 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 StrongARM:     -
  37 C XScale         -
  38 C ARM11          4.68
  39 C Cortex-A7      3.625
  40 C Cortex-A8      4
  41 C Cortex-A9      2.25
  42 C Cortex-A15     2.5
  43
  44 define(`rp',`r0')
  45 define(`up',`r1')
  46 define(`n', `r2')
  47 define(`vp',`r3')
  48
  49 define(`v0',`r6')
  50 define(`v1',`r7')
  51 define(`u0',`r3')
  52 define(`u1',`r9')
  53
  54 define(`cya',`r8')
  55 define(`cyb',`r12')
  56
  57
  58 ASM_START()
  59 PROLOGUE(mpn_addmul_2)
  60         push    { r4-r9 }
  61
  62         ldrd    v0, v1, [vp, #0]
  63         mov     cya, #0
  64         mov     cyb, #0
  65
  66         tst     n, #1
  67         beq     L(evn)
  68
  69 L(odd): ldr     u1, [up, #0]
  70         ldr     r4, [rp, #0]
  71         tst     n, #2
  72         beq     L(fi1)
  73 L(fi3): sub     up, up, #8
  74         sub     rp, rp, #8
  75         b       L(lo3)
  76 L(fi1): sub     n, n, #1
  77         b       L(top)
  78
  79 L(evn): ldr     u0, [up, #0]
  80         ldr     r5, [rp, #0]
  81         tst     n, #2
  82         bne     L(fi2)
  83 L(fi0): sub     up, up, #4
  84         sub     rp, rp, #4
  85         b       L(lo0)
  86 L(fi2): sub     up, up, #12
  87         sub     rp, rp, #12
  88         b       L(lo2)
  89
  90         ALIGN(16)
  91 L(top): ldr     r5, [rp, #4]
  92         umaal   r4, cya, u1, v0
  93         ldr     u0, [up, #4]
  94         umaal   r5, cyb, u1, v1
  95         str     r4, [rp, #0]
  96 L(lo0): ldr     r4, [rp, #8]
  97         umaal   r5, cya, u0, v0
  98         ldr     u1, [up, #8]
  99         umaal   r4, cyb, u0, v1
 100         str     r5, [rp, #4]
 101 L(lo3): ldr     r5, [rp, #12]
 102         umaal   r4, cya, u1, v0
 103         ldr     u0, [up, #12]
 104         umaal   r5, cyb, u1, v1
 105         str     r4, [rp, #8]
 106 L(lo2): ldr     r4, [rp, #16]!
 107         umaal   r5, cya, u0, v0
 108         ldr     u1, [up, #16]!
 109         umaal   r4, cyb, u0, v1
 110         str     r5, [rp, #-4]
 111         subs    n, n, #4
 112         bhi     L(top)
 113
 114 L(end): umaal   r4, cya, u1, v0
 115         umaal   cya, cyb, u1, v1
 116         str     r4, [rp, #0]
 117         str     cya, [rp, #4]
 118         mov     r0, cyb
 119
 120         pop     { r4-r9 }
 121         bx      r14
 122 EPILOGUE()