1 ;! HP-PA-1.1 __mpn_submul_1 -- Multiply a limb vector with a limb and
2 ;! subtract the result from a second limb vector.
4 ;! Copyright (C) 1992-2024 Free Software Foundation, Inc.
6 ;! This file is part of the GNU MP Library.
8 ;! The GNU MP Library is free software; you can redistribute it and/or modify
9 ;! it under the terms of the GNU Lesser General Public License as published by
10 ;! the Free Software Foundation; either version 2.1 of the License, or (at your
11 ;! option) any later version.
13 ;! The GNU MP Library is distributed in the hope that it will be useful, but
14 ;! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
15 ;! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public
16 ;! License for more details.
18 ;! You should have received a copy of the GNU Lesser General Public License
19 ;! along with the GNU MP Library. If not, see
20 ;! <https://www.gnu.org/licenses/>.
29 ;! This runs at 12 cycles/limb on a PA7000. With the used instructions, it
30 ;! can not become faster due to data cache contention after a store. On the
31 ;! PA7100 it runs at 11 cycles/limb, and that can not be improved either,
32 ;! since only the xmpyu does not need the integer pipeline, so the only
33 ;! dual-issue we will get are addc+xmpyu. Unrolling could gain a cycle/limb
36 ;! There are some ideas described in mul_1.s that applies to this code too.
38 ;! It seems possible to make this run as fast as __mpn_addmul_1, if we use
39 ;! sub,>>= %r29,%r20,%r22
41 ;! but that requires reworking the hairy software pipeline...
44 .export __mpn_submul_1
47 .callinfo frame=64,no_calls
52 stw %r23,-16(%r30) ;! move s2_limb ...
53 addib,= -1,%r24,L$just_one_limb
54 fldws -16(%r30),%fr4 ;! ... into fr4
55 add %r0,%r0,%r0 ;! clear carry
60 ldw -12(%r30),%r20 ;! least significant limb in product
79 addib,<> -1,%r24,L$loop