n_sqrt.S revision 1.6 1 1.6 agc /* $NetBSD: n_sqrt.S,v 1.6 2003/08/07 16:44:45 agc Exp $ */
2 1.1 ragge /*
3 1.1 ragge * Copyright (c) 1985, 1993
4 1.1 ragge * The Regents of the University of California. All rights reserved.
5 1.1 ragge *
6 1.1 ragge * Redistribution and use in source and binary forms, with or without
7 1.1 ragge * modification, are permitted provided that the following conditions
8 1.1 ragge * are met:
9 1.1 ragge * 1. Redistributions of source code must retain the above copyright
10 1.1 ragge * notice, this list of conditions and the following disclaimer.
11 1.1 ragge * 2. Redistributions in binary form must reproduce the above copyright
12 1.1 ragge * notice, this list of conditions and the following disclaimer in the
13 1.1 ragge * documentation and/or other materials provided with the distribution.
14 1.6 agc * 3. Neither the name of the University nor the names of its contributors
15 1.1 ragge * may be used to endorse or promote products derived from this software
16 1.1 ragge * without specific prior written permission.
17 1.1 ragge *
18 1.1 ragge * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19 1.1 ragge * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 1.1 ragge * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 1.1 ragge * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22 1.1 ragge * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 1.1 ragge * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 1.1 ragge * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 1.1 ragge * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 1.1 ragge * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 1.1 ragge * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 1.1 ragge * SUCH DAMAGE.
29 1.1 ragge *
30 1.1 ragge * @(#)sqrt.s 8.1 (Berkeley) 6/4/93
31 1.1 ragge */
32 1.1 ragge
33 1.3 matt #include <machine/asm.h>
34 1.3 matt
35 1.1 ragge /*
36 1.1 ragge * double sqrt(arg) revised August 15,1982
37 1.1 ragge * double arg;
38 1.1 ragge * if(arg<0.0) { _errno = EDOM; return(<a reserved operand>); }
39 1.1 ragge * if arg is a reserved operand it is returned as it is
40 1.1 ragge * W. Kahan's magic square root
41 1.1 ragge * coded by Heidi Stettner and revised by Emile LeBlanc 8/18/82
42 1.1 ragge *
43 1.1 ragge * entry points:_d_sqrt address of double arg is on the stack
44 1.1 ragge * _sqrt double arg is on the stack
45 1.1 ragge */
46 1.1 ragge .set EDOM,33
47 1.1 ragge
48 1.5 matt ENTRY(d_sqrt, 0x003c) # save %r5,%r4,%r3,%r2
49 1.5 matt movq *4(%ap),%r0
50 1.3 matt jbr dsqrt2
51 1.3 matt
52 1.5 matt ENTRY(sqrt, 0x003c) # save %r5,%r4,%r3,%r2
53 1.5 matt movq 4(%ap),%r0
54 1.3 matt
55 1.5 matt dsqrt2: bicw3 $0x807f,%r0,%r2 # check exponent of input
56 1.1 ragge jeql noexp # biased exponent is zero -> 0.0 or reserved
57 1.4 matt bsbb __libm_dsqrt_r5_lcl+2
58 1.1 ragge noexp: ret
59 1.1 ragge
60 1.1 ragge /* **************************** internal procedure */
61 1.1 ragge
62 1.4 matt __libm_dsqrt_r5_lcl:
63 1.3 matt ALTENTRY(__libm_dsqrt_r5)
64 1.3 matt nop
65 1.3 matt nop
66 1.3 matt /* ENTRY POINT FOR cdabs and cdsqrt */
67 1.1 ragge /* returns double square root scaled by */
68 1.5 matt /* 2^%r6 */
69 1.1 ragge
70 1.5 matt movd %r0,%r4
71 1.1 ragge jleq nonpos # argument is not positive
72 1.5 matt movzwl %r4,%r2
73 1.5 matt ashl $-1,%r2,%r0
74 1.5 matt addw2 $0x203c,%r0 # %r0 has magic initial approximation
75 1.1 ragge /*
76 1.1 ragge * Do two steps of Heron's rule
77 1.1 ragge * ((arg/guess) + guess) / 2 = better guess
78 1.1 ragge */
79 1.5 matt divf3 %r0,%r4,%r2
80 1.5 matt addf2 %r2,%r0
81 1.5 matt subw2 $0x80,%r0 # divide by two
82 1.5 matt
83 1.5 matt divf3 %r0,%r4,%r2
84 1.5 matt addf2 %r2,%r0
85 1.5 matt subw2 $0x80,%r0 # divide by two
86 1.1 ragge
87 1.1 ragge /* Scale argument and approximation to prevent over/underflow */
88 1.1 ragge
89 1.5 matt bicw3 $0x807f,%r4,%r1
90 1.5 matt subw2 $0x4080,%r1 # %r1 contains scaling factor
91 1.5 matt subw2 %r1,%r4
92 1.5 matt movl %r0,%r2
93 1.5 matt subw2 %r1,%r2
94 1.1 ragge
95 1.1 ragge /* Cubic step
96 1.1 ragge *
97 1.1 ragge * b = a + 2*a*(n-a*a)/(n+3*a*a) where b is better approximation,
98 1.1 ragge * a is approximation, and n is the original argument.
99 1.1 ragge * (let s be scale factor in the following comments)
100 1.1 ragge */
101 1.5 matt clrl %r1
102 1.5 matt clrl %r3
103 1.5 matt muld2 %r0,%r2 # %r2:%r3 = a*a/s
104 1.5 matt subd2 %r2,%r4 # %r4:%r5 = n/s - a*a/s
105 1.5 matt addw2 $0x100,%r2 # %r2:%r3 = 4*a*a/s
106 1.5 matt addd2 %r4,%r2 # %r2:%r3 = n/s + 3*a*a/s
107 1.5 matt muld2 %r0,%r4 # %r4:%r5 = a*n/s - a*a*a/s
108 1.5 matt divd2 %r2,%r4 # %r4:%r5 = a*(n-a*a)/(n+3*a*a)
109 1.5 matt addw2 $0x80,%r4 # %r4:%r5 = 2*a*(n-a*a)/(n+3*a*a)
110 1.5 matt addd2 %r4,%r0 # %r0:%r1 = a + 2*a*(n-a*a)/(n+3*a*a)
111 1.1 ragge rsb # DONE!
112 1.1 ragge nonpos:
113 1.1 ragge jneq negarg
114 1.3 matt ret # argument and root are zero
115 1.1 ragge negarg:
116 1.1 ragge pushl $EDOM
117 1.3 matt calls $1,_C_LABEL(infnan) # generate the reserved op fault
118 1.1 ragge ret
119