source/libs/gmp/gmp-src/mpn/powerpc32/mod_34lsub1.asm

   1 dnl  PowerPC-32 mpn_mod_34lsub1 -- mpn remainder mod 2^24-1.
   2
   3 dnl  Copyright 2002, 2003, 2005 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
  34 C                cycles/limb
  35 C 603e:            ?
  36 C 604e:            3
  37 C 75x (G3):        3
  38 C 7400,7410 (G4):  3
  39 C 744x,745x (G4+): 3
  40 C power4/ppc970:   2.5
  41 C power5:          2.5
  42
  43 C mp_limb_t mpn_mod_34lsub1 (mp_srcptr src, mp_size_t size)
  44 C
  45 C There seems no need to schedule the loads back, the code is still 3.0 c/l
  46 C on 750/7400 no matter where they're placed.
  47 C
  48 C Alternatives:
  49 C
  50 C Fetching half words would allow add instead for accumulating, instead of
  51 C adde and its serialization.  An outer loop would be required though, since
  52 C 2^16 halfwords can overflow.  lhz+add would be 2.0 c/l, but if there's
  53 C also a bdz or bdnz for each and a pointer update say every three limbs
  54 C then the total would be 2.67 c/l which isn't much faster than the current
  55 C simpler code.
  56
  57 ASM_START()
  58 PROLOGUE(mpn_mod_34lsub1)
  59
  60         C r3    src
  61         C r4    size
  62
  63         mtctr   r4
  64         addic   r6, r3, 8               C &src[2], and clear CA
  65
  66         lwz     r3, 0(r3)               C acc0 = src[0]
  67         bdz     L(done)
  68
  69         lwz     r4, -4(r6)              C acc1 = src[1]
  70         bdz     L(two)
  71
  72         lwz     r5, 0(r6)               C acc2 = src[2]
  73         lis     r7, 0                   C no carry if just three limbs
  74
  75         bdz     L(three)
  76         lis     r7, 1                   C 0x10000 carry pos
  77
  78 L(top):
  79         C r3    acc0
  80         C r4    acc1
  81         C r5    acc2
  82         C r6    src, incrementing
  83         C r7    carry pos
  84
  85         lwz     r0, 4(r6)
  86         adde    r3, r3, r0
  87         bdz     L(end0)
  88
  89         lwz     r0, 8(r6)
  90         adde    r4, r4, r0
  91         bdz     L(end1)
  92
  93         lwzu    r0, 12(r6)
  94         adde    r5, r5, r0
  95         bdnz    L(top)
  96
  97
  98         srwi    r7, r7, 8
  99 L(end0):
 100         srwi    r7, r7, 8
 101 L(end1):
 102         subfe   r0, r0, r0              C -1 if not CA
 103
 104         andc    r7, r7, r0              C final carry, 0x10000, 0x100, 1 or 0
 105 L(three):
 106         rlwinm  r6, r3, 0,8,31          C acc0 low
 107
 108         add     r7, r7, r6
 109         rlwinm  r6, r3, 8,24,31         C acc0 high
 110
 111         add     r7, r7, r6
 112         rlwinm  r6, r4, 8,8,23          C acc1 low
 113
 114         add     r7, r7, r6
 115         rlwinm  r6, r4, 16,16,31        C acc1 high
 116
 117         add     r7, r7, r6
 118         rlwinm  r6, r5, 16,8,15         C acc2 low
 119
 120         add     r7, r7, r6
 121         rlwinm  r6, r5, 24,8,31         C acc2 high
 122
 123         add     r3, r7, r6
 124
 125 L(done):
 126         blr
 127
 128 L(two):
 129         C r3    acc0
 130         C r4    acc1
 131
 132         rlwinm  r5, r3, 8,24,31         C acc0 high
 133         rlwinm  r3, r3, 0,8,31          C acc0 low
 134
 135         add     r3, r3, r5              C acc0 high + low
 136         rlwinm  r5, r4, 16,16,31        C acc1 high
 137
 138         add     r3, r3, r5              C add acc1 high
 139         rlwinm  r5, r4, 8,8,23          C acc1 low
 140
 141         add     r3, r3, r5              C add acc1 low
 142
 143         blr
 144
 145 EPILOGUE()