n_cabs.S revision 1.4 1 1.4 matt /* $NetBSD: n_cabs.S,v 1.4 2002/02/24 01:06:21 matt 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 1.1 ragge * 3. All advertising materials mentioning features or use of this software
15 1.1 ragge * must display the following acknowledgement:
16 1.1 ragge * This product includes software developed by the University of
17 1.1 ragge * California, Berkeley and its contributors.
18 1.1 ragge * 4. Neither the name of the University nor the names of its contributors
19 1.1 ragge * may be used to endorse or promote products derived from this software
20 1.1 ragge * without specific prior written permission.
21 1.1 ragge *
22 1.1 ragge * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 1.1 ragge * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 1.1 ragge * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 1.1 ragge * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 1.1 ragge * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 1.1 ragge * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 1.1 ragge * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 1.1 ragge * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 1.1 ragge * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 1.1 ragge * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 1.1 ragge * SUCH DAMAGE.
33 1.1 ragge *
34 1.1 ragge * @(#)cabs.s 8.1 (Berkeley) 6/4/93
35 1.1 ragge */
36 1.1 ragge
37 1.3 matt #include <machine/asm.h>
38 1.3 matt
39 1.3 matt .globl _C_LABEL(__libm_dsqrt_r5)
40 1.1 ragge /*
41 1.1 ragge * double precision complex absolute value
42 1.1 ragge * CABS by W. Kahan, 9/7/80.
43 1.1 ragge * Revised for reserved operands by E. LeBlanc, 8/18/82
44 1.4 matt * argument for complex absolute value by reference, *4(%ap)
45 1.4 matt * argument for cabs and hypot (C fcns) by value, 4(%ap)
46 1.4 matt * output is in %r0:%r1 (error less than 0.86 ulps)
47 1.1 ragge */
48 1.1 ragge
49 1.1 ragge /* entry for c functions cabs and hypot */
50 1.3 matt ALTENTRY(cabs)
51 1.4 matt ENTRY(hypot, 0x8040) # save %r6, enable floating overflow
52 1.4 matt movq 4(%ap),%r0 # %r0:1 = x
53 1.4 matt movq 12(%ap),%r2 # %r2:3 = y
54 1.3 matt jbr cabs2
55 1.3 matt
56 1.1 ragge /* entry for Fortran use, call by: d = abs(z) */
57 1.4 matt ENTRY(z_abs, 0x8040) # save %r6, enable floating overflow
58 1.4 matt movl 4(%ap),%r2 # indirect addressing is necessary here
59 1.4 matt movq (%r2)+,%r0 # %r0:1 = x
60 1.4 matt movq (%r2),%r2 # %r2:3 = y
61 1.1 ragge
62 1.1 ragge cabs2:
63 1.4 matt bicw3 $0x7f,%r0,%r4 # %r4 has signed biased exp of x
64 1.4 matt cmpw $0x8000,%r4
65 1.1 ragge jeql return # x is a reserved operand, so return it
66 1.4 matt bicw3 $0x7f,%r2,%r5 # %r5 has signed biased exp of y
67 1.4 matt cmpw $0x8000,%r5
68 1.1 ragge jneq cont /* y isn't a reserved operand */
69 1.4 matt movq %r2,%r0 /* return y if it's reserved */
70 1.1 ragge ret
71 1.1 ragge
72 1.1 ragge cont:
73 1.4 matt bsbb regs_set # %r0:1 = dsqrt(x^2+y^2)/2^%r6
74 1.4 matt addw2 %r6,%r0 # unscaled cdabs in %r0:1
75 1.1 ragge jvc return # unless it overflows
76 1.4 matt subw2 $0x80,%r0 # halve %r0 to get meaningful overflow
77 1.4 matt addd2 %r0,%r0 # overflow; %r0 is half of true abs value
78 1.1 ragge return:
79 1.1 ragge ret
80 1.1 ragge
81 1.3 matt ENTRY(__libm_cdabs_r6,0) # ENTRY POINT for cdsqrt
82 1.4 matt # calculates a scaled (factor in %r6)
83 1.1 ragge # complex absolute value
84 1.1 ragge
85 1.4 matt movq (%r4)+,%r0 # %r0:%r1 = x via indirect addressing
86 1.4 matt movq (%r4),%r2 # %r2:%r3 = y via indirect addressing
87 1.1 ragge
88 1.4 matt bicw3 $0x7f,%r0,%r5 # %r5 has signed biased exp of x
89 1.4 matt cmpw $0x8000,%r5
90 1.1 ragge jeql cdreserved # x is a reserved operand
91 1.4 matt bicw3 $0x7f,%r2,%r5 # %r5 has signed biased exp of y
92 1.4 matt cmpw $0x8000,%r5
93 1.1 ragge jneq regs_set /* y isn't a reserved operand either? */
94 1.1 ragge
95 1.1 ragge cdreserved:
96 1.4 matt movl *4(%ap),%r4 # %r4 -> (u,v), if x or y is reserved
97 1.4 matt movq %r0,(%r4)+ # copy u and v as is and return
98 1.4 matt movq %r2,(%r4) # (again addressing is indirect)
99 1.1 ragge ret
100 1.1 ragge
101 1.1 ragge regs_set:
102 1.4 matt bicw2 $0x8000,%r0 # %r0:%r1 = dabs(x)
103 1.4 matt bicw2 $0x8000,%r2 # %r2:%r3 = dabs(y)
104 1.4 matt cmpw %r0,%r2
105 1.1 ragge jgeq ordered
106 1.4 matt movq %r0,%r4
107 1.4 matt movq %r2,%r0
108 1.4 matt movq %r4,%r2 # force y's exp <= x's exp
109 1.1 ragge ordered:
110 1.4 matt bicw3 $0x7f,%r0,%r6 # %r6 = exponent(x) + bias(129)
111 1.1 ragge jeql retsb # if x = y = 0 then cdabs(x,y) = 0
112 1.4 matt subw2 $0x4780,%r6 # %r6 = exponent(x) - 14
113 1.4 matt subw2 %r6,%r0 # 2^14 <= scaled x < 2^15
114 1.4 matt bitw $0xff80,%r2
115 1.1 ragge jeql retsb # if y = 0 return dabs(x)
116 1.4 matt subw2 %r6,%r2
117 1.4 matt cmpw $0x3780,%r2 # if scaled y < 2^-18
118 1.1 ragge jgtr retsb # return dabs(x)
119 1.4 matt emodd %r0,$0,%r0,%r4,%r0 # %r4 + %r0:1 = scaled x^2
120 1.4 matt emodd %r2,$0,%r2,%r5,%r2 # %r5 + %r2:3 = scaled y^2
121 1.4 matt addd2 %r2,%r0
122 1.4 matt addl2 %r5,%r4
123 1.4 matt cvtld %r4,%r2
124 1.4 matt addd2 %r2,%r0 # %r0:1 = scaled x^2 + y^2
125 1.3 matt jmp _C_LABEL(__libm_dsqrt_r5)+2
126 1.4 matt # %r0:1 = dsqrt(x^2+y^2)/2^%r6
127 1.1 ragge retsb:
128 1.1 ragge rsb # error < 0.86 ulp
129