e_powf.c revision 369263
1164134Simp/* e_powf.c -- float version of e_pow.c.
2164134Simp * Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
3164134Simp */
4164134Simp
5164134Simp/*
6164134Simp * ====================================================
7164134Simp * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
8164134Simp *
9164134Simp * Developed at SunPro, a Sun Microsystems, Inc. business.
10164134Simp * Permission to use, copy, modify, and distribute this
11164134Simp * software is freely granted, provided that this notice
12164134Simp * is preserved.
13164134Simp * ====================================================
14164134Simp */
15164134Simp
16164134Simp#include <sys/cdefs.h>
17164134Simp__FBSDID("$FreeBSD: stable/11/lib/msun/src/e_powf.c 369263 2021-02-15 18:17:29Z dim $");
18164134Simp
19164134Simp#include "math.h"
20164134Simp#include "math_private.h"
21164134Simp
22164134Simpstatic const float
23164134Simpbp[] = {1.0, 1.5,},
24164134Simpdp_h[] = { 0.0, 5.84960938e-01,}, /* 0x3f15c000 */
25164134Simpdp_l[] = { 0.0, 1.56322085e-06,}, /* 0x35d1cfdc */
26164134Simpzero    =  0.0,
27164134Simphalf    =  0.5,
28164134Simpqrtr    =  0.25,
29171426Simpthrd    =  3.33333343e-01, /* 0x3eaaaaab */
30one	=  1.0,
31two	=  2.0,
32two24	=  16777216.0,	/* 0x4b800000 */
33huge	=  1.0e30,
34tiny    =  1.0e-30,
35	/* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */
36L1  =  6.0000002384e-01, /* 0x3f19999a */
37L2  =  4.2857143283e-01, /* 0x3edb6db7 */
38L3  =  3.3333334327e-01, /* 0x3eaaaaab */
39L4  =  2.7272811532e-01, /* 0x3e8ba305 */
40L5  =  2.3066075146e-01, /* 0x3e6c3255 */
41L6  =  2.0697501302e-01, /* 0x3e53f142 */
42P1   =  1.6666667163e-01, /* 0x3e2aaaab */
43P2   = -2.7777778450e-03, /* 0xbb360b61 */
44P3   =  6.6137559770e-05, /* 0x388ab355 */
45P4   = -1.6533901999e-06, /* 0xb5ddea0e */
46P5   =  4.1381369442e-08, /* 0x3331bb4c */
47lg2  =  6.9314718246e-01, /* 0x3f317218 */
48lg2_h  =  6.93145752e-01, /* 0x3f317200 */
49lg2_l  =  1.42860654e-06, /* 0x35bfbe8c */
50ovt =  4.2995665694e-08, /* -(128-log2(ovfl+.5ulp)) */
51cp    =  9.6179670095e-01, /* 0x3f76384f =2/(3ln2) */
52cp_h  =  9.6191406250e-01, /* 0x3f764000 =12b cp */
53cp_l  = -1.1736857402e-04, /* 0xb8f623c6 =tail of cp_h */
54ivln2    =  1.4426950216e+00, /* 0x3fb8aa3b =1/ln2 */
55ivln2_h  =  1.4426879883e+00, /* 0x3fb8aa00 =16b 1/ln2*/
56ivln2_l  =  7.0526075433e-06; /* 0x36eca570 =1/ln2 tail*/
57
58float
59__ieee754_powf(float x, float y)
60{
61	float z,ax,z_h,z_l,p_h,p_l;
62	float y1,t1,t2,r,s,sn,t,u,v,w;
63	int32_t i,j,k,yisint,n;
64	int32_t hx,hy,ix,iy,is;
65
66	GET_FLOAT_WORD(hx,x);
67	GET_FLOAT_WORD(hy,y);
68	ix = hx&0x7fffffff;  iy = hy&0x7fffffff;
69
70    /* y==zero: x**0 = 1 */
71	if(iy==0) return one;
72
73    /* x==1: 1**y = 1, even if y is NaN */
74	if (hx==0x3f800000) return one;
75
76    /* y!=zero: result is NaN if either arg is NaN */
77	if(ix > 0x7f800000 ||
78	   iy > 0x7f800000)
79		return (x+0.0F)+(y+0.0F);
80
81    /* determine if y is an odd int when x < 0
82     * yisint = 0	... y is not an integer
83     * yisint = 1	... y is an odd int
84     * yisint = 2	... y is an even int
85     */
86	yisint  = 0;
87	if(hx<0) {
88	    if(iy>=0x4b800000) yisint = 2; /* even integer y */
89	    else if(iy>=0x3f800000) {
90		k = (iy>>23)-0x7f;	   /* exponent */
91		j = iy>>(23-k);
92		if((j<<(23-k))==iy) yisint = 2-(j&1);
93	    }
94	}
95
96    /* special value of y */
97	if (iy==0x7f800000) {	/* y is +-inf */
98	    if (ix==0x3f800000)
99	        return  one;	/* (-1)**+-inf is NaN */
100	    else if (ix > 0x3f800000)/* (|x|>1)**+-inf = inf,0 */
101	        return (hy>=0)? y: zero;
102	    else			/* (|x|<1)**-,+inf = inf,0 */
103	        return (hy<0)?-y: zero;
104	}
105	if(iy==0x3f800000) {	/* y is  +-1 */
106	    if(hy<0) return one/x; else return x;
107	}
108	if(hy==0x40000000) return x*x; /* y is  2 */
109	if(hy==0x3f000000) {	/* y is  0.5 */
110	    if(hx>=0)	/* x >= +0 */
111	    return __ieee754_sqrtf(x);
112	}
113
114	ax   = fabsf(x);
115    /* special value of x */
116	if(ix==0x7f800000||ix==0||ix==0x3f800000){
117	    z = ax;			/*x is +-0,+-inf,+-1*/
118	    if(hy<0) z = one/z;	/* z = (1/|x|) */
119	    if(hx<0) {
120		if(((ix-0x3f800000)|yisint)==0) {
121		    z = (z-z)/(z-z); /* (-1)**non-int is NaN */
122		} else if(yisint==1)
123		    z = -z;		/* (x<0)**odd = -(|x|**odd) */
124	    }
125	    return z;
126	}
127
128	n = ((u_int32_t)hx>>31)-1;
129
130    /* (x<0)**(non-int) is NaN */
131	if((n|yisint)==0) return (x-x)/(x-x);
132
133	sn = one; /* s (sign of result -ve**odd) = -1 else = 1 */
134	if((n|(yisint-1))==0) sn = -one;/* (-ve)**(odd int) */
135
136    /* |y| is huge */
137	if(iy>0x4d000000) { /* if |y| > 2**27 */
138	/* over/underflow if x is not close to one */
139	    if(ix<0x3f7ffff7) return (hy<0)? sn*huge*huge:sn*tiny*tiny;
140	    if(ix>0x3f800007) return (hy>0)? sn*huge*huge:sn*tiny*tiny;
141	/* now |1-x| is tiny <= 2**-20, suffice to compute
142	   log(x) by x-x^2/2+x^3/3-x^4/4 */
143	    t = ax-1;		/* t has 20 trailing zeros */
144	    w = (t*t)*(half-t*(thrd-t*qrtr));
145	    u = ivln2_h*t;	/* ivln2_h has 16 sig. bits */
146	    v = t*ivln2_l-w*ivln2;
147	    t1 = u+v;
148	    GET_FLOAT_WORD(is,t1);
149	    SET_FLOAT_WORD(t1,is&0xfffff000);
150	    t2 = v-(t1-u);
151	} else {
152	    float s2,s_h,s_l,t_h,t_l;
153	    n = 0;
154	/* take care subnormal number */
155	    if(ix<0x00800000)
156		{ax *= two24; n -= 24; GET_FLOAT_WORD(ix,ax); }
157	    n  += ((ix)>>23)-0x7f;
158	    j  = ix&0x007fffff;
159	/* determine interval */
160	    ix = j|0x3f800000;		/* normalize ix */
161	    if(j<=0x1cc471) k=0;	/* |x|<sqrt(3/2) */
162	    else if(j<0x5db3d7) k=1;	/* |x|<sqrt(3)   */
163	    else {k=0;n+=1;ix -= 0x00800000;}
164	    SET_FLOAT_WORD(ax,ix);
165
166	/* compute s = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */
167	    u = ax-bp[k];		/* bp[0]=1.0, bp[1]=1.5 */
168	    v = one/(ax+bp[k]);
169	    s = u*v;
170	    s_h = s;
171	    GET_FLOAT_WORD(is,s_h);
172	    SET_FLOAT_WORD(s_h,is&0xfffff000);
173	/* t_h=ax+bp[k] High */
174	    is = ((ix>>1)&0xfffff000)|0x20000000;
175	    SET_FLOAT_WORD(t_h,is+0x00400000+(k<<21));
176	    t_l = ax - (t_h-bp[k]);
177	    s_l = v*((u-s_h*t_h)-s_h*t_l);
178	/* compute log(ax) */
179	    s2 = s*s;
180	    r = s2*s2*(L1+s2*(L2+s2*(L3+s2*(L4+s2*(L5+s2*L6)))));
181	    r += s_l*(s_h+s);
182	    s2  = s_h*s_h;
183	    t_h = 3+s2+r;
184	    GET_FLOAT_WORD(is,t_h);
185	    SET_FLOAT_WORD(t_h,is&0xfffff000);
186	    t_l = r-((t_h-3)-s2);
187	/* u+v = s*(1+...) */
188	    u = s_h*t_h;
189	    v = s_l*t_h+t_l*s;
190	/* 2/(3log2)*(s+...) */
191	    p_h = u+v;
192	    GET_FLOAT_WORD(is,p_h);
193	    SET_FLOAT_WORD(p_h,is&0xfffff000);
194	    p_l = v-(p_h-u);
195	    z_h = cp_h*p_h;		/* cp_h+cp_l = 2/(3*log2) */
196	    z_l = cp_l*p_h+p_l*cp+dp_l[k];
197	/* log2(ax) = (s+..)*2/(3*log2) = n + dp_h + z_h + z_l */
198	    t = n;
199	    t1 = (((z_h+z_l)+dp_h[k])+t);
200	    GET_FLOAT_WORD(is,t1);
201	    SET_FLOAT_WORD(t1,is&0xfffff000);
202	    t2 = z_l-(((t1-t)-dp_h[k])-z_h);
203	}
204
205    /* split up y into y1+y2 and compute (y1+y2)*(t1+t2) */
206	GET_FLOAT_WORD(is,y);
207	SET_FLOAT_WORD(y1,is&0xfffff000);
208	p_l = (y-y1)*t1+y*t2;
209	p_h = y1*t1;
210	z = p_l+p_h;
211	GET_FLOAT_WORD(j,z);
212	if (j>0x43000000)				/* if z > 128 */
213	    return sn*huge*huge;			/* overflow */
214	else if (j==0x43000000) {			/* if z == 128 */
215	    if(p_l+ovt>z-p_h) return sn*huge*huge;	/* overflow */
216	}
217	else if ((j&0x7fffffff)>0x43160000)		/* z <= -150 */
218	    return sn*tiny*tiny;			/* underflow */
219	else if (j==0xc3160000){			/* z == -150 */
220	    if(p_l<=z-p_h) return sn*tiny*tiny;		/* underflow */
221	}
222    /*
223     * compute 2**(p_h+p_l)
224     */
225	i = j&0x7fffffff;
226	k = (i>>23)-0x7f;
227	n = 0;
228	if(i>0x3f000000) {		/* if |z| > 0.5, set n = [z+0.5] */
229	    n = j+(0x00800000>>(k+1));
230	    k = ((n&0x7fffffff)>>23)-0x7f;	/* new k for n */
231	    SET_FLOAT_WORD(t,n&~(0x007fffff>>k));
232	    n = ((n&0x007fffff)|0x00800000)>>(23-k);
233	    if(j<0) n = -n;
234	    p_h -= t;
235	}
236	t = p_l+p_h;
237	GET_FLOAT_WORD(is,t);
238	SET_FLOAT_WORD(t,is&0xffff8000);
239	u = t*lg2_h;
240	v = (p_l-(t-p_h))*lg2+t*lg2_l;
241	z = u+v;
242	w = v-(z-u);
243	t  = z*z;
244	t1  = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5))));
245	r  = (z*t1)/(t1-two)-(w+z*w);
246	z  = one-(r-z);
247	GET_FLOAT_WORD(j,z);
248	j += (n<<23);
249	if((j>>23)<=0) z = scalbnf(z,n);	/* subnormal output */
250	else SET_FLOAT_WORD(z,j);
251	return sn*z;
252}
253