1 dnl Alpha mpn_lshift -- Shift a number left.
3 dnl Copyright (C) 1994, 1995, 2000 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 by
9 dnl the Free Software Foundation; either version 2.1 of the License, or (at your
10 dnl 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 to
19 dnl the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
20 dnl MA 02111-1307, USA.
22 include(`../config.m4')
30 dnl This code runs at 4.8 cycles/limb on the 21064. With infinite unrolling,
31 dnl it would take 4 cycles/limb. It should be possible to get down to 3
32 dnl cycles/limb since both ldq and stq can be paired with the other used
33 dnl instructions. But there are many restrictions in the 21064 pipeline that
34 dnl makes it hard, if not impossible, to get down to 3 cycles/limb:
36 dnl 1. ldq has a 3 cycle delay, srl and sll have a 2 cycle delay.
37 dnl 2. Only aligned instruction pairs can be paired.
38 dnl 3. The store buffer or silo might not be able to deal with the bandwidth.
42 s8addq r18,r17,r17 C make r17 point at end of s1
43 ldq r4,-8(r17) C load first limb
46 s8addq r18,r16,r16 C make r16 point at end of RES
48 and r18,4-1,r20 C number of limbs in first loop
49 srl r4,r7,r0 C compute function result