xref: /freebsd/lib/msun/src/s_csqrtl.c (revision 61e21613)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2007-2008 David Schultz <das@FreeBSD.ORG>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 #include <complex.h>
31 #include <float.h>
32 #include <math.h>
33 
34 #include "math_private.h"
35 
36 /*
37  * Several thresholds require a 15-bit exponent and also the usual bias.
38  * s_logl.c and e_hypotl have less hard-coding but end up requiring the
39  * same for the exponent and more for the mantissa.
40  */
41 #if LDBL_MAX_EXP != 0x4000
42 #error "Unsupported long double format"
43 #endif
44 
45 /*
46  * Overflow must be avoided for components >= LDBL_MAX / (1 + sqrt(2)).
47  * The precise threshold is nontrivial to determine and spell, so use a
48  * lower threshold of approximaely LDBL_MAX / 4, and don't use LDBL_MAX
49  * to spell this since LDBL_MAX is broken on i386 (it overflows in 53-bit
50  * precision).
51  */
52 #define	THRESH	0x1p16382L
53 
54 long double complex
55 csqrtl(long double complex z)
56 {
57 	long double complex result;
58 	long double a, b, rx, ry, scale, t;
59 
60 	a = creall(z);
61 	b = cimagl(z);
62 
63 	/* Handle special cases. */
64 	if (z == 0)
65 		return (CMPLXL(0, b));
66 	if (isinf(b))
67 		return (CMPLXL(INFINITY, b));
68 	if (isnan(a)) {
69 		t = (b - b) / (b - b);	/* raise invalid if b is not a NaN */
70 		return (CMPLXL(a + 0.0L + t, a + 0.0L + t)); /* NaN + NaN i */
71 	}
72 	if (isinf(a)) {
73 		/*
74 		 * csqrt(inf + NaN i)  = inf +  NaN i
75 		 * csqrt(inf + y i)    = inf +  0 i
76 		 * csqrt(-inf + NaN i) = NaN +- inf i
77 		 * csqrt(-inf + y i)   = 0   +  inf i
78 		 */
79 		if (signbit(a))
80 			return (CMPLXL(fabsl(b - b), copysignl(a, b)));
81 		else
82 			return (CMPLXL(a, copysignl(b - b, b)));
83 	}
84 	if (isnan(b)) {
85 		t = (a - a) / (a - a);	/* raise invalid */
86 		return (CMPLXL(b + 0.0L + t, b + 0.0L + t)); /* NaN + NaN i */
87 	}
88 
89 	/* Scale to avoid overflow. */
90 	if (fabsl(a) >= THRESH || fabsl(b) >= THRESH) {
91 		/*
92 		 * Don't scale a or b if this might give (spurious)
93 		 * underflow.  Then the unscaled value is an equivalent
94 		 * infinitesmal (or 0).
95 		 */
96 		if (fabsl(a) >= 0x1p-16380L)
97 			a *= 0.25;
98 		if (fabsl(b) >= 0x1p-16380L)
99 			b *= 0.25;
100 		scale = 2;
101 	} else {
102 		scale = 1;
103 	}
104 
105 	/* Scale to reduce inaccuracies when both components are denormal. */
106 	if (fabsl(a) < 0x1p-16382L && fabsl(b) < 0x1p-16382L) {
107 		a *= 0x1p64;
108 		b *= 0x1p64;
109 		scale = 0x1p-32;
110 	}
111 
112 	/* Algorithm 312, CACM vol 10, Oct 1967. */
113 	if (a >= 0) {
114 		t = sqrtl((a + hypotl(a, b)) * 0.5);
115 		rx = scale * t;
116 		ry = scale * b / (2 * t);
117 	} else {
118 		t = sqrtl((-a + hypotl(a, b)) * 0.5);
119 		rx = scale * fabsl(b) / (2 * t);
120 		ry = copysignl(scale * t, b);
121 	}
122 
123 	return (CMPLXL(rx, ry));
124 }
125