source/libs/gmp/gmp-src/mpn/sparc64/lshiftc.asm

   1 dnl  SPARC v9 mpn_lshiftc
   2
   3 dnl  Contributed to the GNU project by David Miller.
   4
   5 dnl  Copyright 2013 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 UltraSPARC 1&2:        3
  37 C UltraSPARC 3:          3
  38 C UltraSPARC T1:        17
  39 C UltraSPARC T3:        10
  40 C UltraSPARC T4:         3.5
  41
  42 C INPUT PARAMETERS
  43 define(`rp',     `%i0')
  44 define(`up',     `%i1')
  45 define(`n',      `%i2')
  46 define(`cnt',    `%i3')
  47
  48 define(`tcnt',   `%i4')
  49 define(`retval', `%i5')
  50 define(`u0',     `%l0')
  51 define(`u1',     `%l1')
  52 define(`r0',     `%l6')
  53 define(`r1',     `%l7')
  54 define(`u0_off', `%o0')
  55 define(`u1_off', `%o1')
  56 define(`r0_off', `%o2')
  57 define(`r1_off', `%o3')
  58
  59 ASM_START()
  60         REGISTER(%g2,#scratch)
  61         REGISTER(%g3,#scratch)
  62 PROLOGUE(mpn_lshiftc)
  63         save    %sp, -176, %sp
  64
  65         sllx    n, 3, n
  66         sub     %g0, cnt, tcnt
  67
  68         sub     up, 8, u1_off
  69         add     rp, (5 * 8), r1_off
  70
  71         ldx     [n + u1_off], u1        C WAS: up - 8
  72         add     u1_off, (3 * 8), u1_off
  73
  74         sub     r1_off, 8, r0_off
  75         sub     u1_off, 8, u0_off
  76
  77         subcc   n, (3 * 8), n
  78         srlx    u1, tcnt, retval
  79
  80         bl,pn   %xcc, L(end12)
  81          sllx   u1, cnt, %l3
  82
  83         ldx     [n + u0_off], u0        C WAS: up - 16
  84         subcc   n, (2 * 8), n
  85
  86         ldx     [n + u1_off], u1        C WAS: up - 24
  87
  88         bl,pn   %xcc, L(end34)
  89          srlx   u0, tcnt, %l4
  90
  91         b,a     L(top)
  92         ALIGN(16)
  93 L(top):
  94         not     %l3, %l3
  95         sllx    u0, cnt, %l2
  96
  97         andn    %l3, %l4, r0
  98         ldx     [n + u0_off], u0        C WAS: up - 16
  99
 100         srlx    u1, tcnt, %l5
 101         stx     r0, [n + r0_off]        C WAS: rp - 8
 102
 103         subcc   n, (2 * 8), n
 104         not     %l2, %l2
 105
 106         sllx    u1, cnt, %l3
 107         andn    %l2, %l5, r1
 108
 109         ldx     [n + u1_off], u1        C WAS: up - 24
 110         srlx    u0, tcnt, %l4
 111
 112         bge,pt  %xcc, L(top)
 113          stx    r1, [n + r1_off]        C WAS: rp - 16
 114
 115 L(end34):
 116         not     %l3, %l3
 117         sllx    u0, cnt, %l2
 118
 119         andn    %l3, %l4, r0
 120         srlx    u1, tcnt, %l5
 121
 122         stx     r0, [n + r0_off]        C WAS: rp - 8
 123         not     %l2, %l2
 124
 125         andn    %l2, %l5, r1
 126         sub     n, (2 * 8), %o5
 127
 128         sllx    u1, cnt, %l3
 129         stx     r1, [%o5 + r1_off]      C WAS: rp - 16
 130
 131 L(end12):
 132         andcc   n, 8, %g0
 133         bz      %xcc, L(done)+4
 134          not    %l3, %l3
 135
 136         ldx     [n + u0_off], u1
 137         srlx    u1, tcnt, %l4
 138         andn    %l3, %l4, r0
 139         stx     r0, [r0_off - 24]
 140         sllx    u1, cnt, %l3
 141 L(done):
 142         not     %l3, %l3
 143         stx     %l3, [r0_off - 32]
 144
 145         ret
 146         restore retval, 0, %o0
 147 EPILOGUE()