const_soft_float/soft_f64/
sqrt.rs

1/* origin: FreeBSD /usr/src/lib/msun/src/e_sqrt.c */
2/*
3 * ====================================================
4 * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
5 *
6 * Developed at SunSoft, a Sun Microsystems, Inc. business.
7 * Permission to use, copy, modify, and distribute this
8 * software is freely granted, provided that this notice
9 * is preserved.
10 * ====================================================
11 */
12/* sqrt(x)
13 * Return correctly rounded sqrt.
14 *           ------------------------------------------
15 *           |  Use the hardware sqrt if you have one |
16 *           ------------------------------------------
17 * Method:
18 *   Bit by bit method using integer arithmetic. (Slow, but portable)
19 *   1. Normalization
20 *      Scale x to y in [1,4) with even powers of 2:
21 *      find an integer k such that  1 <= (y=x*2^(2k)) < 4, then
22 *              sqrt(x) = 2^k * sqrt(y)
23 *   2. Bit by bit computation
24 *      Let q  = sqrt(y) truncated to i bit after binary point (q = 1),
25 *           i                                                   0
26 *                                     i+1         2
27 *          s  = 2*q , and      y  =  2   * ( y - q  ).         (1)
28 *           i      i            i                 i
29 *
30 *      To compute q    from q , one checks whether
31 *                  i+1       i
32 *
33 *                            -(i+1) 2
34 *                      (q + 2      ) <= y.                     (2)
35 *                        i
36 *                                                            -(i+1)
37 *      If (2) is false, then q   = q ; otherwise q   = q  + 2      .
38 *                             i+1   i             i+1   i
39 *
40 *      With some algebraic manipulation, it is not difficult to see
41 *      that (2) is equivalent to
42 *                             -(i+1)
43 *                      s  +  2       <= y                      (3)
44 *                       i                i
45 *
46 *      The advantage of (3) is that s  and y  can be computed by
47 *                                    i      i
48 *      the following recurrence formula:
49 *          if (3) is false
50 *
51 *          s     =  s  ,       y    = y   ;                    (4)
52 *           i+1      i          i+1    i
53 *
54 *          otherwise,
55 *                         -i                     -(i+1)
56 *          s     =  s  + 2  ,  y    = y  -  s  - 2             (5)
57 *           i+1      i          i+1    i     i
58 *
59 *      One may easily use induction to prove (4) and (5).
60 *      Note. Since the left hand side of (3) contain only i+2 bits,
61 *            it does not necessary to do a full (53-bit) comparison
62 *            in (3).
63 *   3. Final rounding
64 *      After generating the 53 bits result, we compute one more bit.
65 *      Together with the remainder, we can decide whether the
66 *      result is exact, bigger than 1/2ulp, or less than 1/2ulp
67 *      (it will never equal to 1/2ulp).
68 *      The rounding mode can be detected by checking whether
69 *      huge + tiny is equal to huge, and whether huge - tiny is
70 *      equal to huge for some floating point number "huge" and "tiny".
71 *
72 * Special cases:
73 *      sqrt(+-0) = +-0         ... exact
74 *      sqrt(inf) = inf
75 *      sqrt(-ve) = NaN         ... with invalid signal
76 *      sqrt(NaN) = NaN         ... with invalid signal for signaling NaN
77 */
78
79use crate::soft_f64::{
80    helpers::{ge, gt},
81    SoftF64,
82};
83
84type F = SoftF64;
85
86pub(crate) const fn sqrt(x: F) -> F {
87    const TINY: F = SoftF64(1.0e-300);
88
89    let mut z: F;
90    let sign: u32 = 0x80000000;
91    let mut ix0: i32;
92    let mut s0: i32;
93    let mut q: i32;
94    let mut m: i32;
95    let mut t: i32;
96    let mut i: i32;
97    let mut r: u32;
98    let mut t1: u32;
99    let mut s1: u32;
100    let mut ix1: u32;
101    let mut q1: u32;
102
103    ix0 = (x.to_bits() >> 32) as i32;
104    ix1 = x.to_bits() as u32;
105
106    /* take care of Inf and NaN */
107    if (ix0 & 0x7ff00000) == 0x7ff00000 {
108        return x.mul(x).add(x); /* sqrt(NaN)=NaN, sqrt(+inf)=+inf, sqrt(-inf)=sNaN */
109    }
110    /* take care of zero */
111    if ix0 <= 0 {
112        if ((ix0 & !(sign as i32)) | ix1 as i32) == 0 {
113            return x; /* sqrt(+-0) = +-0 */
114        }
115        if ix0 < 0 {
116            return (x.sub(x)).div(x.sub(x)); /* sqrt(-ve) = sNaN */
117        }
118    }
119    /* normalize x */
120    m = ix0 >> 20;
121    if m == 0 {
122        /* subnormal x */
123        while ix0 == 0 {
124            m -= 21;
125            ix0 |= (ix1 >> 11) as i32;
126            ix1 <<= 21;
127        }
128        i = 0;
129        while (ix0 & 0x00100000) == 0 {
130            i += 1;
131            ix0 <<= 1;
132        }
133        m -= i - 1;
134        ix0 |= (ix1 as usize >> (32 - i) as usize) as i32;
135        ix1 = ix1 << i as usize;
136    }
137    m -= 1023; /* unbias exponent */
138    ix0 = (ix0 & 0x000fffff) | 0x00100000;
139    if (m & 1) == 1 {
140        /* odd m, double x to make it even */
141        ix0 += ix0 + ((ix1 & sign) >> 31) as i32;
142        ix1 = ix1.wrapping_add(ix1);
143    }
144    m >>= 1; /* m = [m/2] */
145
146    /* generate sqrt(x) bit by bit */
147    ix0 += ix0 + ((ix1 & sign) >> 31) as i32;
148    ix1 = ix1.wrapping_add(ix1);
149    q = 0; /* [q,q1] = sqrt(x) */
150    q1 = 0;
151    s0 = 0;
152    s1 = 0;
153    r = 0x00200000; /* r = moving bit from right to left */
154
155    while r != 0 {
156        t = s0 + r as i32;
157        if t <= ix0 {
158            s0 = t + r as i32;
159            ix0 -= t;
160            q += r as i32;
161        }
162        ix0 += ix0 + ((ix1 & sign) >> 31) as i32;
163        ix1 = ix1.wrapping_add(ix1);
164        r >>= 1;
165    }
166
167    r = sign;
168    while r != 0 {
169        t1 = s1.wrapping_add(r);
170        t = s0;
171        if t < ix0 || (t == ix0 && t1 <= ix1) {
172            s1 = t1.wrapping_add(r);
173            if (t1 & sign) == sign && (s1 & sign) == 0 {
174                s0 += 1;
175            }
176            ix0 -= t;
177            if ix1 < t1 {
178                ix0 -= 1;
179            }
180            ix1 = ix1.wrapping_sub(t1);
181            q1 += r;
182        }
183        ix0 += ix0 + ((ix1 & sign) >> 31) as i32;
184        ix1 = ix1.wrapping_add(ix1);
185        r >>= 1;
186    }
187
188    /* use floating add to find out rounding direction */
189    if (ix0 as u32 | ix1) != 0 {
190        z = SoftF64(1.0).sub(TINY); /* raise inexact flag */
191        if ge(z, SoftF64::ONE) {
192            z = SoftF64::ONE.add(TINY);
193            if q1 == 0xffffffff {
194                q1 = 0;
195                q += 1;
196            } else if gt(z, SoftF64::ONE) {
197                if q1 == 0xfffffffe {
198                    q += 1;
199                }
200                q1 = q1.wrapping_add(2);
201            } else {
202                q1 += q1 & 1;
203            }
204        }
205    }
206    ix0 = (q >> 1) + 0x3fe00000;
207    ix1 = q1 >> 1;
208    if (q & 1) == 1 {
209        ix1 |= sign;
210    }
211    ix0 += m << 20;
212    SoftF64::from_bits((ix0 as u64) << 32 | ix1 as u64)
213}
214
215#[cfg(test)]
216mod tests {
217    use super::*;
218    //  use core::f64::*;
219
220    #[test]
221    fn sanity_check() {
222        const SQRT_100: SoftF64 = sqrt(SoftF64(100.0));
223        assert_eq!(SQRT_100.0, 10.0);
224
225        const SQRT_4: SoftF64 = sqrt(SoftF64(4.0));
226        assert_eq!(SQRT_4.0, 2.0);
227    }
228
229    /// The spec: https://en.cppreference.com/w/cpp/numeric/math/sqrt
230    #[test]
231    fn spec_tests() {
232        // Not Asserted: FE_INVALID exception is raised if argument is negative.
233        assert!(sqrt(SoftF64(-1.0)).0.is_nan());
234        assert!(sqrt(SoftF64(f64::NAN)).0.is_nan());
235        for f in [0.0, -0.0, f64::INFINITY].iter().copied() {
236            assert_eq!(sqrt(SoftF64(f)).0, f);
237        }
238    }
239}