1
dnl PowerPC-32 mpn_mod_34lsub1 -- mpn remainder mod 2^24-1.
3
dnl Copyright 2002, 2003, 2005 Free Software Foundation, Inc.
5
dnl This file is part of the GNU MP Library.
7
dnl The GNU MP Library is free software; you can redistribute it and/or modify
8
dnl it under the terms of the GNU Lesser General Public License as published
9
dnl by the Free Software Foundation; either version 2.1 of the License, or (at
10
dnl your option) any later version.
12
dnl The GNU MP Library is distributed in the hope that it will be useful, but
13
dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
14
dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public
15
dnl License for more details.
17
dnl You should have received a copy of the GNU Lesser General Public License
18
dnl along with the GNU MP Library; see the file COPYING.LIB. If not, write
19
dnl to the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
20
dnl Boston, MA 02110-1301, USA.
22
include(`../config.m4')
30
C 744x,745x (G4+): 3.0
34
C mp_limb_t mpn_mod_34lsub1 (mp_srcptr src, mp_size_t size)
36
C There seems no need to schedule the loads back, the code is still 3.0 c/l
37
C on 750/7400 no matter where they're placed.
41
C Fetching half words would allow add instead for accumulating, instead of
42
C adde and its serialization. An outer loop would be required though, since
43
C 2^16 halfwords can overflow. lhz+add would be 2.0 c/l, but if there's
44
C also a bdz or bdnz for each and a pointer update say every three limbs
45
C then the total would be 2.67 c/l which isn't much faster than the current
49
PROLOGUE(mpn_mod_34lsub1)
55
addic r6, r3, 8 C &src[2], and clear CA
57
lwz r3, 0(r3) C acc0 = src[0]
60
lwz r4, -4(r6) C acc1 = src[1]
63
lwz r5, 0(r6) C acc2 = src[2]
64
lis r7, 0 C no carry if just three limbs
67
lis r7, 1 C 0x10000 carry pos
73
C r6 src, incrementing
93
subfe r0, r0, r0 C -1 if not CA
95
andc r7, r7, r0 C final carry, 0x10000, 0x100, 1 or 0
97
rlwinm r6, r3, 0,8,31 C acc0 low
100
rlwinm r6, r3, 8,24,31 C acc0 high
103
rlwinm r6, r4, 8,8,23 C acc1 low
106
rlwinm r6, r4, 16,16,31 C acc1 high
109
rlwinm r6, r5, 16,8,15 C acc2 low
112
rlwinm r6, r5, 24,8,31 C acc2 high
123
rlwinm r5, r3, 8,24,31 C acc0 high
124
rlwinm r3, r3, 0,8,31 C acc0 low
126
add r3, r3, r5 C acc0 high + low
127
rlwinm r5, r4, 16,16,31 C acc1 high
129
add r3, r3, r5 C add acc1 high
130
rlwinm r5, r4, 8,8,23 C acc1 low
132
add r3, r3, r5 C add acc1 low