• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 #include <stdint.h>
2 #include <math.h>
3 #include "libm.h"
4 #include "sqrt_data.h"
5 
6 #define FENV_SUPPORT 1
7 
mul32(uint32_t a,uint32_t b)8 static inline uint32_t mul32(uint32_t a, uint32_t b)
9 {
10 	return (uint64_t)a*b >> 32;
11 }
12 
13 /* see sqrt.c for more detailed comments.  */
14 
sqrtf(float x)15 float sqrtf(float x)
16 {
17 	uint32_t ix, m, m1, m0, even, ey;
18 
19 	ix = asuint(x);
20 	if (predict_false(ix - 0x00800000 >= 0x7f800000 - 0x00800000)) {
21 		/* x < 0x1p-126 or inf or nan.  */
22 		if (ix * 2 == 0)
23 			return x;
24 		if (ix == 0x7f800000)
25 			return x;
26 		if (ix > 0x7f800000)
27 			return __math_invalidf(x);
28 		/* x is subnormal, normalize it.  */
29 		ix = asuint(x * 0x1p23f);
30 		ix -= 23 << 23;
31 	}
32 
33 	/* x = 4^e m; with int e and m in [1, 4).  */
34 	even = ix & 0x00800000;
35 	m1 = (ix << 8) | 0x80000000;
36 	m0 = (ix << 7) & 0x7fffffff;
37 	m = even ? m0 : m1;
38 
39 	/* 2^e is the exponent part of the return value.  */
40 	ey = ix >> 1;
41 	ey += 0x3f800000 >> 1;
42 	ey &= 0x7f800000;
43 
44 	/* compute r ~ 1/sqrt(m), s ~ sqrt(m) with 2 goldschmidt iterations.  */
45 	static const uint32_t three = 0xc0000000;
46 	uint32_t r, s, d, u, i;
47 	i = (ix >> 17) % 128;
48 	r = (uint32_t)__rsqrt_tab[i] << 16;
49 	/* |r*sqrt(m) - 1| < 0x1p-8 */
50 	s = mul32(m, r);
51 	/* |s/sqrt(m) - 1| < 0x1p-8 */
52 	d = mul32(s, r);
53 	u = three - d;
54 	r = mul32(r, u) << 1;
55 	/* |r*sqrt(m) - 1| < 0x1.7bp-16 */
56 	s = mul32(s, u) << 1;
57 	/* |s/sqrt(m) - 1| < 0x1.7bp-16 */
58 	d = mul32(s, r);
59 	u = three - d;
60 	s = mul32(s, u);
61 	/* -0x1.03p-28 < s/sqrt(m) - 1 < 0x1.fp-31 */
62 	s = (s - 1)>>6;
63 	/* s < sqrt(m) < s + 0x1.08p-23 */
64 
65 	/* compute nearest rounded result.  */
66 	uint32_t d0, d1, d2;
67 	float y, t;
68 	d0 = (m << 16) - s*s;
69 	d1 = s - d0;
70 	d2 = d1 + s + 1;
71 	s += d1 >> 31;
72 	s &= 0x007fffff;
73 	s |= ey;
74 	y = asfloat(s);
75 	if (FENV_SUPPORT) {
76 		/* handle rounding and inexact exception. */
77 		uint32_t tiny = predict_false(d2==0) ? 0 : 0x01000000;
78 		tiny |= (d1^d2) & 0x80000000;
79 		t = asfloat(tiny);
80 		y = eval_as_float(y + t);
81 	}
82 	return y;
83 }
84