1 /* Euclidean distance function.  Double/Binary64 version.
2    Copyright (C) 2021-2022 Free Software Foundation, Inc.
3    This file is part of the GNU C Library.
4 
5    The GNU C Library is free software; you can redistribute it and/or
6    modify it under the terms of the GNU Lesser General Public
7    License as published by the Free Software Foundation; either
8    version 2.1 of the License, or (at your option) any later version.
9 
10    The GNU C Library is distributed in the hope that it will be useful,
11    but WITHOUT ANY WARRANTY; without even the implied warranty of
12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13    Lesser General Public License for more details.
14 
15    You should have received a copy of the GNU Lesser General Public
16    License along with the GNU C Library; if not, see
17    <https://www.gnu.org/licenses/>.  */
18 
19 /* The implementation uses a correction based on 'An Improved Algorithm for
20    hypot(a,b)' by Carlos F. Borges [1] usingthe MyHypot3 with the following
21    changes:
22 
23    - Handle qNaN and sNaN.
24    - Tune the 'widely varying operands' to avoid spurious underflow
25      due the multiplication and fix the return value for upwards
26      rounding mode.
27    - Handle required underflow exception for subnormal results.
28 
29    The expected ULP is ~0.792 or ~0.948 if FMA is used.  For FMA, the
30    correction is not used and the error of sqrt (x^2 + y^2) is below 1 ULP
31    if x^2 + y^2 is computed with less than 0.707 ULP error.  If |x| >= |2y|,
32    fma (x, x, y^2) has ~0.625 ULP.  If |x| < |2y|, fma (|2x|, |y|, (x - y)^2)
33    has ~0.625 ULP.
34 
35    [1] https://arxiv.org/pdf/1904.09481.pdf  */
36 
37 #include <errno.h>
38 #include <math.h>
39 #include <math_private.h>
40 #include <math-underflow.h>
41 #include <math-narrow-eval.h>
42 #include <math-use-builtins.h>
43 #include <math-svid-compat.h>
44 #include <libm-alias-finite.h>
45 #include <libm-alias-double.h>
46 #include "math_config.h"
47 
48 #define SCALE     0x1p-600
49 #define LARGE_VAL 0x1p+511
50 #define TINY_VAL  0x1p-459
51 #define EPS       0x1p-54
52 
53 static inline double
handle_errno(double r)54 handle_errno (double r)
55 {
56   if (isinf (r))
57     __set_errno (ERANGE);
58   return r;
59 }
60 
61 /* Hypot kernel. The inputs must be adjusted so that ax >= ay >= 0
62    and squaring ax, ay and (ax - ay) does not overflow or underflow.  */
63 static inline double
kernel(double ax,double ay)64 kernel (double ax, double ay)
65 {
66   double t1, t2;
67 #ifdef __FP_FAST_FMA
68   t1 = ay + ay;
69   t2 = ax - ay;
70 
71   if (t1 >= ax)
72     return sqrt (fma (t1, ax, t2 * t2));
73   else
74     return sqrt (fma (ax, ax, ay * ay));
75 
76 #else
77   double h = sqrt (ax * ax + ay * ay);
78   if (h <= 2.0 * ay)
79     {
80       double delta = h - ay;
81       t1 = ax * (2.0 * delta - ax);
82       t2 = (delta - 2.0 * (ax - ay)) * delta;
83     }
84   else
85     {
86       double delta = h - ax;
87       t1 = 2.0 * delta * (ax - 2.0 * ay);
88       t2 = (4.0 * delta - ay) * ay + delta * delta;
89     }
90 
91   h -= (t1 + t2) / (2.0 * h);
92   return h;
93 #endif
94 }
95 
96 double
__hypot(double x,double y)97 __hypot (double x, double y)
98 {
99   if (!isfinite(x) || !isfinite(y))
100     {
101       if ((isinf (x) || isinf (y))
102 	  && !issignaling_inline (x) && !issignaling_inline (y))
103 	return INFINITY;
104       return x + y;
105     }
106 
107   x = fabs (x);
108   y = fabs (y);
109 
110   double ax = USE_FMAX_BUILTIN ? fmax (x, y) : (x < y ? y : x);
111   double ay = USE_FMIN_BUILTIN ? fmin (x, y) : (x < y ? x : y);
112 
113   /* If ax is huge, scale both inputs down.  */
114   if (__glibc_unlikely (ax > LARGE_VAL))
115     {
116       if (__glibc_unlikely (ay <= ax * EPS))
117 	return handle_errno (math_narrow_eval (ax + ay));
118 
119       return handle_errno (math_narrow_eval (kernel (ax * SCALE, ay * SCALE)
120 					     / SCALE));
121     }
122 
123   /* If ay is tiny, scale both inputs up.  */
124   if (__glibc_unlikely (ay < TINY_VAL))
125     {
126       if (__glibc_unlikely (ax >= ay / EPS))
127 	return math_narrow_eval (ax + ay);
128 
129       ax = math_narrow_eval (kernel (ax / SCALE, ay / SCALE) * SCALE);
130       math_check_force_underflow_nonneg (ax);
131       return ax;
132     }
133 
134   /* Common case: ax is not huge and ay is not tiny.  */
135   if (__glibc_unlikely (ay <= ax * EPS))
136     return ax + ay;
137 
138   return kernel (ax, ay);
139 }
140 strong_alias (__hypot, __ieee754_hypot)
141 libm_alias_finite (__ieee754_hypot, __hypot)
142 #if LIBM_SVID_COMPAT
143 versioned_symbol (libm, __hypot, hypot, GLIBC_2_35);
144 libm_alias_double_other (__hypot, hypot)
145 #else
146 libm_alias_double (__hypot, hypot)
147 #endif
148