Update.
[glibc.git] / sysdeps / rs6000 / submul_1.s
blob8e5946fe18ca135d67f6416d297296671604bc7c
1 # IBM POWER __mpn_submul_1 -- Multiply a limb vector with a limb and subtract
2 # the result from a second limb vector.
4 # Copyright (C) 1992, 1994 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 Library General Public License as published by
10 # the Free Software Foundation; either version 2 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 Library General Public
16 # License for more details.
18 # You should have received a copy of the GNU Library General Public License
19 # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
20 # the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
21 # MA 02111-1307, USA.
24 # INPUT PARAMETERS
25 # res_ptr r3
26 # s1_ptr r4
27 # size r5
28 # s2_limb r6
30 # The RS/6000 has no unsigned 32x32->64 bit multiplication instruction. To
31 # obtain that operation, we have to use the 32x32->64 signed multiplication
32 # instruction, and add the appropriate compensation to the high limb of the
33 # result. We add the multiplicand if the multiplier has its most significant
34 # bit set, and we add the multiplier if the multiplicand has its most
35 # significant bit set. We need to preserve the carry flag between each
36 # iteration, so we have to compute the compensation carefully (the natural,
37 # srai+and doesn't work). Since the POWER architecture has a branch unit
38 # we can branch in zero cycles, so that's how we perform the additions.
40 .toc
41 .csect .__mpn_submul_1[PR]
42 .align 2
43 .globl __mpn_submul_1
44 .globl .__mpn_submul_1
45 .csect __mpn_submul_1[DS]
46 __mpn_submul_1:
47 .long .__mpn_submul_1[PR], TOC[tc0], 0
48 .csect .__mpn_submul_1[PR]
49 .__mpn_submul_1:
51 cal 3,-4(3)
52 l 0,0(4)
53 cmpi 0,6,0
54 mtctr 5
55 mul 9,0,6
56 srai 7,0,31
57 and 7,7,6
58 mfmq 11
59 cax 9,9,7
60 l 7,4(3)
61 sf 8,11,7 # add res_limb
62 a 11,8,11 # invert cy (r11 is junk)
63 blt Lneg
64 Lpos: bdz Lend
66 Lploop: lu 0,4(4)
67 stu 8,4(3)
68 cmpi 0,0,0
69 mul 10,0,6
70 mfmq 0
71 ae 11,0,9 # low limb + old_cy_limb + old cy
72 l 7,4(3)
73 aze 10,10 # propagate cy to new cy_limb
74 sf 8,11,7 # add res_limb
75 a 11,8,11 # invert cy (r11 is junk)
76 bge Lp0
77 cax 10,10,6 # adjust high limb for negative limb from s1
78 Lp0: bdz Lend0
79 lu 0,4(4)
80 stu 8,4(3)
81 cmpi 0,0,0
82 mul 9,0,6
83 mfmq 0
84 ae 11,0,10
85 l 7,4(3)
86 aze 9,9
87 sf 8,11,7
88 a 11,8,11 # invert cy (r11 is junk)
89 bge Lp1
90 cax 9,9,6 # adjust high limb for negative limb from s1
91 Lp1: bdn Lploop
93 b Lend
95 Lneg: cax 9,9,0
96 bdz Lend
97 Lnloop: lu 0,4(4)
98 stu 8,4(3)
99 cmpi 0,0,0
100 mul 10,0,6
101 mfmq 7
102 ae 11,7,9
103 l 7,4(3)
104 ae 10,10,0 # propagate cy to new cy_limb
105 sf 8,11,7 # add res_limb
106 a 11,8,11 # invert cy (r11 is junk)
107 bge Ln0
108 cax 10,10,6 # adjust high limb for negative limb from s1
109 Ln0: bdz Lend0
110 lu 0,4(4)
111 stu 8,4(3)
112 cmpi 0,0,0
113 mul 9,0,6
114 mfmq 7
115 ae 11,7,10
116 l 7,4(3)
117 ae 9,9,0 # propagate cy to new cy_limb
118 sf 8,11,7 # add res_limb
119 a 11,8,11 # invert cy (r11 is junk)
120 bge Ln1
121 cax 9,9,6 # adjust high limb for negative limb from s1
122 Ln1: bdn Lnloop
123 b Lend
125 Lend0: cal 9,0(10)
126 Lend: st 8,4(3)
127 aze 3,9