sysdeps/alpha/lshift.s

   1  # Alpha 21064 __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 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., 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 4.8 cycles/limb on the 21064.  With infinite unrolling,
  30  # it would take 4 cycles/limb.  It should be possible to get down to 3
  31  # cycles/limb since both ldq and stq can be paired with the other used
  32  # instructions.  But there are many restrictions in the 21064 pipeline that
  33  # makes it hard, if not impossible, to get down to 3 cycles/limb:
  34
  35  # 1. ldq has a 3 cycle delay, srl and sll have a 2 cycle delay.
  36  # 2. Only aligned instruction pairs can be paired.
  37  # 3. The store buffer or silo might not be able to deal with the bandwidth.
  38
  39         .set    noreorder
  40         .set    noat
  41 .text
  42         .align  3
  43         .globl  __mpn_lshift
  44         .ent    __mpn_lshift
  45 __mpn_lshift:
  46         .frame  $30,0,$26,0
  47
  48         s8addq  $18,$17,$17     # make r17 point at end of s1
  49         ldq     $4,-8($17)      # load first limb
  50         subq    $17,8,$17
  51         subq    $31,$19,$7
  52         s8addq  $18,$16,$16     # make r16 point at end of RES
  53         subq    $18,1,$18
  54         and     $18,4-1,$20     # number of limbs in first loop
  55         srl     $4,$7,$0        # compute function result
  56
  57         beq     $20,.L0
  58         subq    $18,$20,$18
  59
  60         .align  3
  61 .Loop0:
  62         ldq     $3,-8($17)
  63         subq    $16,8,$16
  64         subq    $17,8,$17
  65         subq    $20,1,$20
  66         sll     $4,$19,$5
  67         srl     $3,$7,$6
  68         bis     $3,$3,$4
  69         bis     $5,$6,$8
  70         stq     $8,0($16)
  71         bne     $20,.Loop0
  72
  73 .L0:    beq     $18,.Lend
  74
  75         .align  3
  76 .Loop:  ldq     $3,-8($17)
  77         subq    $16,32,$16
  78         subq    $18,4,$18
  79         sll     $4,$19,$5
  80         srl     $3,$7,$6
  81
  82         ldq     $4,-16($17)
  83         sll     $3,$19,$1
  84         bis     $5,$6,$8
  85         stq     $8,24($16)
  86         srl     $4,$7,$2
  87
  88         ldq     $3,-24($17)
  89         sll     $4,$19,$5
  90         bis     $1,$2,$8
  91         stq     $8,16($16)
  92         srl     $3,$7,$6
  93
  94         ldq     $4,-32($17)
  95         sll     $3,$19,$1
  96         bis     $5,$6,$8
  97         stq     $8,8($16)
  98         srl     $4,$7,$2
  99
 100         subq    $17,32,$17
 101         bis     $1,$2,$8
 102         stq     $8,0($16)
 103
 104         bgt     $18,.Loop
 105
 106 .Lend:  sll     $4,$19,$8
 107         stq     $8,-8($16)
 108         ret     $31,($26),1
 109         .end    __mpn_lshift