sysdeps/alpha/alphaev5/rshift.s

   1  # Alpha EV5 __mpn_rshift --
   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 Library General Public License as published by
   9  # the Free Software Foundation; either version 2 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 Library General Public
  15  # License for more details.
  16
  17  # You should have received a copy of the GNU Library 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., 675 Mass Ave, Cambridge, MA 02139, USA.
  20
  21
  22  # INPUT PARAMETERS
  23  # res_ptr      r16
  24  # s1_ptr       r17
  25  # size         r18
  26  # cnt          r19
  27
  28  # This code runs at 3.25 cycles/limb on the EV5.
  29
  30         .set    noreorder
  31         .set    noat
  32 .text
  33         .align  3
  34         .globl  __mpn_rshift
  35         .ent    __mpn_rshift
  36 __mpn_rshift:
  37         .frame  $30,0,$26,0
  38
  39         ldq     $4,0($17)       # load first limb
  40         subq    $31,$19,$20
  41         subq    $18,1,$18
  42         and     $18,4-1,$28     # number of limbs in first loop
  43         sll     $4,$20,$0       # compute function result
  44
  45         beq     $28,.L0
  46         subq    $18,$28,$18
  47
  48         .align  3
  49 .Loop0: ldq     $3,8($17)
  50         addq    $16,8,$16
  51         srl     $4,$19,$5
  52         addq    $17,8,$17
  53         subq    $28,1,$28
  54         sll     $3,$20,$6
  55         or      $3,$3,$4
  56         or      $5,$6,$8
  57         stq     $8,-8($16)
  58         bne     $28,.Loop0
  59
  60 .L0:    srl     $4,$19,$24
  61         beq     $18,.Lend
  62  # warm up phase 1
  63         ldq     $1,8($17)
  64         subq    $18,4,$18
  65         ldq     $2,16($17)
  66         ldq     $3,24($17)
  67         ldq     $4,32($17)
  68         beq     $18,.Lend1
  69  # warm up phase 2
  70         sll     $1,$20,$7
  71         srl     $1,$19,$21
  72         sll     $2,$20,$8
  73         ldq     $1,40($17)
  74         srl     $2,$19,$22
  75         ldq     $2,48($17)
  76         sll     $3,$20,$5
  77         or      $7,$24,$7
  78         srl     $3,$19,$23
  79         or      $8,$21,$8
  80         sll     $4,$20,$6
  81         ldq     $3,56($17)
  82         srl     $4,$19,$24
  83         ldq     $4,64($17)
  84         subq    $18,4,$18
  85         beq     $18,.Lend2
  86         .align  4
  87  # main loop
  88 .Loop:  stq     $7,0($16)
  89         or      $5,$22,$5
  90         stq     $8,8($16)
  91         or      $6,$23,$6
  92
  93         sll     $1,$20,$7
  94         subq    $18,4,$18
  95         srl     $1,$19,$21
  96         unop    # ldq   $31,-96($17)
  97
  98         sll     $2,$20,$8
  99         ldq     $1,72($17)
 100         srl     $2,$19,$22
 101         ldq     $2,80($17)
 102
 103         stq     $5,16($16)
 104         or      $7,$24,$7
 105         stq     $6,24($16)
 106         or      $8,$21,$8
 107
 108         sll     $3,$20,$5
 109         unop    # ldq   $31,-96($17)
 110         srl     $3,$19,$23
 111         addq    $16,32,$16
 112
 113         sll     $4,$20,$6
 114         ldq     $3,88($17)
 115         srl     $4,$19,$24
 116         ldq     $4,96($17)
 117
 118         addq    $17,32,$17
 119         bne     $18,.Loop
 120  # cool down phase 2/1
 121 .Lend2: stq     $7,0($16)
 122         or      $5,$22,$5
 123         stq     $8,8($16)
 124         or      $6,$23,$6
 125         sll     $1,$20,$7
 126         srl     $1,$19,$21
 127         sll     $2,$20,$8
 128         srl     $2,$19,$22
 129         stq     $5,16($16)
 130         or      $7,$24,$7
 131         stq     $6,24($16)
 132         or      $8,$21,$8
 133         sll     $3,$20,$5
 134         srl     $3,$19,$23
 135         sll     $4,$20,$6
 136         srl     $4,$19,$24
 137  # cool down phase 2/2
 138         stq     $7,32($16)
 139         or      $5,$22,$5
 140         stq     $8,40($16)
 141         or      $6,$23,$6
 142         stq     $5,48($16)
 143         stq     $6,56($16)
 144  # cool down phase 2/3
 145         stq     $24,64($16)
 146         ret     $31,($26),1
 147
 148  # cool down phase 1/1
 149 .Lend1: sll     $1,$20,$7
 150         srl     $1,$19,$21
 151         sll     $2,$20,$8
 152         srl     $2,$19,$22
 153         sll     $3,$20,$5
 154         or      $7,$24,$7
 155         srl     $3,$19,$23
 156         or      $8,$21,$8
 157         sll     $4,$20,$6
 158         srl     $4,$19,$24
 159  # cool down phase 1/2
 160         stq     $7,0($16)
 161         or      $5,$22,$5
 162         stq     $8,8($16)
 163         or      $6,$23,$6
 164         stq     $5,16($16)
 165         stq     $6,24($16)
 166         stq     $24,32($16)
 167         ret     $31,($26),1
 168
 169 .Lend:  stq     $24,0($16)
 170         ret     $31,($26),1
 171         .end    __mpn_rshift