1
/* hppa1.1 mul_1 -- Multiply a limb vector with a limb and store
2
* the result in a second limb vector.
4
* Copyright (C) 1992, 1993, 1994, 1998,
5
* 2001 Free Software Foundation, Inc.
7
* This file is part of GnuPG.
9
* GnuPG is free software; you can redistribute it and/or modify
10
* it under the terms of the GNU General Public License as published by
11
* the Free Software Foundation; either version 2 of the License, or
12
* (at your option) any later version.
14
* GnuPG is distributed in the hope that it will be useful,
15
* but WITHOUT ANY WARRANTY; without even the implied warranty of
16
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17
* GNU General Public License for more details.
19
* You should have received a copy of the GNU General Public License
20
* along with this program; if not, write to the Free Software
21
* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA
23
* Note: This code is heavily based on the GNU MP Library.
24
* Actually it's the same code with only minor changes in the
25
* way the data is stored; this is to support the abstraction
26
* of an optional secure memory allocation which may be used
27
* to avoid revealing of sensitive data due to paging etc.
28
* The GNU MP Library itself is published under the LGPL;
29
* however I decided to publish this code under the plain GPL.
35
* mpihelp_mul_1( mpi_ptr_t res_ptr, (r26)
36
* mpi_ptr_t s1_ptr, (r25)
37
* mpi_size_t s1_size, (r24)
38
* mpi_limb_t s2_limb) (r23)
42
* This runs at 9 cycles/limb on a PA7000. With the used instructions, it can
43
* not become faster due to data cache contention after a store. On the
44
* PA7100 it runs at 7 cycles/limb, and that can not be improved either, since
45
* only the xmpyu does not need the integer pipeline, so the only dual-issue
46
* we will get are addc+xmpyu. Unrolling would not help either CPU.
48
* We could use fldds to read two limbs at a time from the S1 array, and that
49
* could bring down the times to 8.5 and 6.5 cycles/limb for the PA7000 and
50
* PA7100, respectively. We don't do that since it does not seem worth the
51
* (alignment) troubles...
53
* At least the PA7100 is rumored to be able to deal with cache-misses
54
* without stalling instruction issue. If this is true, and the cache is
55
* actually also lockup-free, we should use a deeper software pipeline, and
56
* load from S1 very early! (The loads and stores to -12(sp) will surely be
64
.callinfo frame=64,no_calls
69
stw %r23,-16(%r30) ; move s2_limb ...
70
addib,= -1,%r24,L$just_one_limb
71
fldws -16(%r30),%fr4 ; ... into fr4
72
add %r0,%r0,%r0 ; clear carry
77
ldw -12(%r30),%r19 ; least significant limb in product
92
addib,<> -1,%r24,L$loop
104
.label L$just_one_limb