sysdeps/alpha/alphaev5/lshift.s

   1  # Alpha EV5 __mpn_lshift --
   2
   3  # Copyright (C) 1994, 1995 Free Software Foundation, Inc.
   4
   5  # This file is part of the GNU MP Library.
   6
   7  # The GNU MP Library is free software; you can redistribute it and/or modify
   8  # it under the terms of the GNU Lesser General Public License as published by
   9  # the Free Software Foundation; either version 2.1 of the License, or (at your
  10  # option) any later version.
  11
  12  # The GNU MP Library is distributed in the hope that it will be useful, but
  13  # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  14  # or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
  15  # License for more details.
  16
  17  # You should have received a copy of the GNU Lesser General Public License
  18  # along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
  19  # the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
  20  # MA 02111-1307, USA.
  21
  22
  23  # INPUT PARAMETERS
  24  # res_ptr      r16
  25  # s1_ptr       r17
  26  # size         r18
  27  # cnt          r19
  28
  29  # This code runs at 3.25 cycles/limb on the EV5.
  30
  31         .set    noreorder
  32         .set    noat
  33 .text
  34         .align  3
  35         .globl  __mpn_lshift
  36         .ent    __mpn_lshift
  37 __mpn_lshift:
  38         .frame  $30,0,$26,0
  39
  40         s8addq  $18,$17,$17     # make r17 point at end of s1
  41         ldq     $4,-8($17)      # load first limb
  42         subq    $31,$19,$20
  43         s8addq  $18,$16,$16     # make r16 point at end of RES
  44         subq    $18,1,$18
  45         and     $18,4-1,$28     # number of limbs in first loop
  46         srl     $4,$20,$0       # compute function result
  47
  48         beq     $28,.L0
  49         subq    $18,$28,$18
  50
  51         .align  3
  52 .Loop0: ldq     $3,-16($17)
  53         subq    $16,8,$16
  54         sll     $4,$19,$5
  55         subq    $17,8,$17
  56         subq    $28,1,$28
  57         srl     $3,$20,$6
  58         or      $3,$3,$4
  59         or      $5,$6,$8
  60         stq     $8,0($16)
  61         bne     $28,.Loop0
  62
  63 .L0:    sll     $4,$19,$24
  64         beq     $18,.Lend
  65  # warm up phase 1
  66         ldq     $1,-16($17)
  67         subq    $18,4,$18
  68         ldq     $2,-24($17)
  69         ldq     $3,-32($17)
  70         ldq     $4,-40($17)
  71         beq     $18,.Lend1
  72  # warm up phase 2
  73         srl     $1,$20,$7
  74         sll     $1,$19,$21
  75         srl     $2,$20,$8
  76         ldq     $1,-48($17)
  77         sll     $2,$19,$22
  78         ldq     $2,-56($17)
  79         srl     $3,$20,$5
  80         or      $7,$24,$7
  81         sll     $3,$19,$23
  82         or      $8,$21,$8
  83         srl     $4,$20,$6
  84         ldq     $3,-64($17)
  85         sll     $4,$19,$24
  86         ldq     $4,-72($17)
  87         subq    $18,4,$18
  88         beq     $18,.Lend2
  89         .align  4
  90  # main loop
  91 .Loop:  stq     $7,-8($16)
  92         or      $5,$22,$5
  93         stq     $8,-16($16)
  94         or      $6,$23,$6
  95
  96         srl     $1,$20,$7
  97         subq    $18,4,$18
  98         sll     $1,$19,$21
  99         unop    # ldq   $31,-96($17)
 100
 101         srl     $2,$20,$8
 102         ldq     $1,-80($17)
 103         sll     $2,$19,$22
 104         ldq     $2,-88($17)
 105
 106         stq     $5,-24($16)
 107         or      $7,$24,$7
 108         stq     $6,-32($16)
 109         or      $8,$21,$8
 110
 111         srl     $3,$20,$5
 112         unop    # ldq   $31,-96($17)
 113         sll     $3,$19,$23
 114         subq    $16,32,$16
 115
 116         srl     $4,$20,$6
 117         ldq     $3,-96($17)
 118         sll     $4,$19,$24
 119         ldq     $4,-104($17)
 120
 121         subq    $17,32,$17
 122         bne     $18,.Loop
 123  # cool down phase 2/1
 124 .Lend2: stq     $7,-8($16)
 125         or      $5,$22,$5
 126         stq     $8,-16($16)
 127         or      $6,$23,$6
 128         srl     $1,$20,$7
 129         sll     $1,$19,$21
 130         srl     $2,$20,$8
 131         sll     $2,$19,$22
 132         stq     $5,-24($16)
 133         or      $7,$24,$7
 134         stq     $6,-32($16)
 135         or      $8,$21,$8
 136         srl     $3,$20,$5
 137         sll     $3,$19,$23
 138         srl     $4,$20,$6
 139         sll     $4,$19,$24
 140  # cool down phase 2/2
 141         stq     $7,-40($16)
 142         or      $5,$22,$5
 143         stq     $8,-48($16)
 144         or      $6,$23,$6
 145         stq     $5,-56($16)
 146         stq     $6,-64($16)
 147  # cool down phase 2/3
 148         stq     $24,-72($16)
 149         ret     $31,($26),1
 150
 151  # cool down phase 1/1
 152 .Lend1: srl     $1,$20,$7
 153         sll     $1,$19,$21
 154         srl     $2,$20,$8
 155         sll     $2,$19,$22
 156         srl     $3,$20,$5
 157         or      $7,$24,$7
 158         sll     $3,$19,$23
 159         or      $8,$21,$8
 160         srl     $4,$20,$6
 161         sll     $4,$19,$24
 162  # cool down phase 1/2
 163         stq     $7,-8($16)
 164         or      $5,$22,$5
 165         stq     $8,-16($16)
 166         or      $6,$23,$6
 167         stq     $5,-24($16)
 168         stq     $6,-32($16)
 169         stq     $24,-40($16)
 170         ret     $31,($26),1
 171
 172 .Lend:  stq     $24,-8($16)
 173         ret     $31,($26),1
 174         .end    __mpn_lshift