mpn/alpha/copyd.asm

   1 dnl  Alpha mpn_copyd -- copy, decrementing.
   2
   3 dnl  Copyright 2002, 2003 Free Software Foundation, Inc.
   4
   5 dnl  This file is part of the GNU MP Library.
   6 dnl
   7 dnl  The GNU MP Library is free software; you can redistribute it and/or modify
   8 dnl  it under the terms of either:
   9 dnl
  10 dnl    * the GNU Lesser General Public License as published by the Free
  11 dnl      Software Foundation; either version 3 of the License, or (at your
  12 dnl      option) any later version.
  13 dnl
  14 dnl  or
  15 dnl
  16 dnl    * the GNU General Public License as published by the Free Software
  17 dnl      Foundation; either version 2 of the License, or (at your option) any
  18 dnl      later version.
  19 dnl
  20 dnl  or both in parallel, as here.
  21 dnl
  22 dnl  The GNU MP Library is distributed in the hope that it will be useful, but
  23 dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  24 dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  25 dnl  for more details.
  26 dnl
  27 dnl  You should have received copies of the GNU General Public License and the
  28 dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
  29 dnl  see https://www.gnu.org/licenses/.
  30
  31 include(`../config.m4')
  32
  33 C      cycles/limb
  34 C EV4:     4
  35 C EV5:     1.75
  36 C EV6:     1
  37
  38 C INPUT PARAMETERS
  39 C rp    r16
  40 C up    r17
  41 C n     r18
  42
  43
  44 ASM_START()
  45 PROLOGUE(mpn_copyd)
  46         s8addq  r18,r16,r16             C E0
  47         s8addq  r18,r17,r17             C E1
  48         lda     r18,-8(r18)             C E0
  49         blt     r18,$Lend               C E1
  50 $Loop:  ldq     r0,-8(r17)              C E0
  51         ldq     r1,-16(r17)             C E1
  52         ldq     r2,-24(r17)             C E0
  53         ldq     r3,-32(r17)             C E1
  54         ldq     r4,-40(r17)             C E0
  55         ldq     r5,-48(r17)             C E1
  56         ldq     r6,-56(r17)             C E0
  57         ldq     r7,-64(r17)             C E1
  58         stq     r0,-8(r16)              C E0
  59         lda     r17,-64(r17)            C E1
  60         stq     r1,-16(r16)             C E0
  61         bis     r31, r31, r31           C E1
  62         stq     r2,-24(r16)             C E0
  63         lda     r18,-8(r18)             C E1
  64         stq     r3,-32(r16)             C E0
  65         bis     r31, r31, r31           C E1
  66         stq     r4,-40(r16)             C E0
  67         bis     r31, r31, r31           C E1
  68         stq     r5,-48(r16)             C E0
  69         bis     r31, r31, r31           C E1
  70         stq     r6,-56(r16)             C E0
  71         bis     r31, r31, r31           C E1
  72         stq     r7,-64(r16)             C E0
  73         lda     r16,-64(r16)            C E1
  74         bge     r18,$Loop               C E1
  75 $Lend:  lda     r18,7(r18)              C E0
  76         blt     r18,$Lret               C E1
  77         ldq     r0,-8(r17)              C E0
  78         beq     r18,$Lend0              C E1
  79 $Loop0: stq     r0,-8(r16)              C E0
  80         lda     r16,-8(r16)             C E1
  81         ldq     r0,-16(r17)             C E0
  82         lda     r18,-1(r18)             C E1
  83         lda     r17,-8(r17)             C E0
  84         bgt     r18,$Loop0              C E1
  85 $Lend0: stq     r0,-8(r16)              C E0
  86 $Lret:  ret     r31,(r26),1             C E1
  87 EPILOGUE(mpn_copyd)
  88 ASM_END()