1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
|
/*
libffm - Free, pretty fast replacement for some math (libm) routines
on Linux/AXP, optimized for the 21164
Copyright (C) 1998 Joachim Wesner <joachim.wesner@frankfurt.netsurf.de>
and Kazushige Goto <goto@statabo.rim.or.jp>
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Library General Public
License as published by the Free Software Foundation; either
version 2 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Library General Public License for more details.
You should have received a copy of the GNU Library General Public
License along with this library (see file COPYING.LIB); if not, write
to the Free Software Foundation, Inc., 675 Mass Ave, Cambridge,
MA 02139, USA.
*/
/*
Simple, fast powr(x, y) helper function for the fast log2/exp2 routines,
by Joachim Wesner <joachim.wesner@frankfurt.netsurf.de>, July 1998.
This routine is "quick&dirty" as in the Cray powr function, i.e. NO
extended precision is used in calculating log(x)*y, so the relative
error in the final result will grow to ~ 0.4*|ld(x)|*|y| ulp.
However, for integral values of y, repeated multiplications will be used
to calculate the result. So, for "random" x, the rel. error seems to
be in the order of 0.4*|y| ulp and for (not too large) integral values of
x, the final result should be EXACT.
In contrast to the Cray powr function, a different multiplication scheme
is used, where the number of multiplications only increases with log2(|y|).
No special handling of illegal arguments or NANs, yet !!!
*/
/* modified and improved by Kazushige Goto <goto@statabo.rim.or.jp> */
.set noat
.set noreorder
.text
.align 5
.globl powr
.ent powr
powr:
fabs $f17, $f10 # $f10 = fabs($f17)
mov 1, $1
lda $30,-16($30)
fbeq $f16,$x_eq_0 # x == 0
stq $1, 0($30)
ldgp $29,.-powr($27)
.frame $30,16,$26,0
.prologue 1
nop
#ifdef PROF
lda $28, _mcount
jsr $28, ($28), _mcount
unop
unop
#endif
ldt $f0, 0($30)
cvttqc $f10, $f11 # double -> int (f)
fbeq $f17,$y_eq_0 # y == 0
cvtqt $f0, $f0 # $f0 = 1.0
cvtqt $f11, $f1 # int -> double
stt $f11,8($30)
cmpteq $f1,$f10,$f1 # if (n==f)
ldq $2, 8($30)
fmov $f0, $f10 # z = 1.0
fbeq $f1,$L8 # Skip
unop
beq $2,$L10 # while(n)
.align 4
$loop: # This loop is very tight.
blbc $2,$L12 # if (n&1)
mult $f10,$f16,$f10 # z = z * x
$L12:
sra $2,1,$2 # n >>= 1
mult $f16,$f16,$f16 # x = x * x
unop
unop
unop
bne $2,$loop
$L10:
fblt $f17, $L14
fmov $f10, $f0 # return z
addq $30,16,$30
ret $31,($26),1
.align 4
$L14:
divt $f0,$f10,$f0
addq $30,16,$30
ret $31,($26),1
.align 4
$L8:
lda $27, log2
stt $f2,8($30)
fmov $f17, $f2
stq $26,0($30)
jsr $26, ($27), log2
mult $f0,$f2,$f16
ldgp $29,4($26)
jsr $26,exp2
ldq $26,0($30)
ldt $f2,8($30)
$L17:
addq $30,16,$30
ret $31,($26),1
.align 4
$x_eq_0:
fclr $f0
addq $30,16,$30
ret $31,($26),1
.align 4
$y_eq_0:
cvtqt $f0, $f0 # $f0 = 1.0
addq $30,16,$30
ret $31,($26),1
.end powr
#ifdef POW
.globl pow
pow = powr
#endif
|
