Deleted Added
full compact
e_jn.c (302408) e_jn.c (336196)
1
2/* @(#)e_jn.c 1.4 95/01/18 */
3/*
4 * ====================================================
5 * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
6 *
7 * Developed at SunSoft, a Sun Microsystems, Inc. business.
8 * Permission to use, copy, modify, and distribute this
1/* @(#)e_jn.c 1.4 95/01/18 */
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
9 * software is freely granted, provided that this notice
8 * software is freely granted, provided that this notice
10 * is preserved.
11 * ====================================================
12 */
13
14#include <sys/cdefs.h>
9 * is preserved.
10 * ====================================================
11 */
12
13#include <sys/cdefs.h>
15__FBSDID("$FreeBSD: stable/11/lib/msun/src/e_jn.c 279856 2015-03-10 17:10:54Z kargl $");
14__FBSDID("$FreeBSD: stable/11/lib/msun/src/e_jn.c 336196 2018-07-11 12:12:49Z markj $");
16
17/*
18 * __ieee754_jn(n, x), __ieee754_yn(n, x)
19 * floating point Bessel's function of the 1st and 2nd kind
20 * of order n
15
16/*
17 * __ieee754_jn(n, x), __ieee754_yn(n, x)
18 * floating point Bessel's function of the 1st and 2nd kind
19 * of order n
21 *
20 *
22 * Special cases:
23 * y0(0)=y1(0)=yn(n,0) = -inf with division by zero signal;
24 * y0(-ve)=y1(-ve)=yn(n,-ve) are NaN with invalid signal.
25 * Note 2. About jn(n,x), yn(n,x)
26 * For n=0, j0(x) is called,
27 * for n=1, j1(x) is called,
28 * for n<x, forward recursion us used starting
29 * from values of j0(x) and j1(x).
30 * for n>x, a continued fraction approximation to
31 * j(n,x)/j(n-1,x) is evaluated and then backward
32 * recursion is used starting from a supposed value
33 * for j(n,x). The resulting value of j(0,x) is
34 * compared with the actual value to correct the
35 * supposed value of j(n,x).
36 *
37 * yn(n,x) is similar in all respects, except
38 * that forward recursion is used for all
39 * values of n>1.
21 * Special cases:
22 * y0(0)=y1(0)=yn(n,0) = -inf with division by zero signal;
23 * y0(-ve)=y1(-ve)=yn(n,-ve) are NaN with invalid signal.
24 * Note 2. About jn(n,x), yn(n,x)
25 * For n=0, j0(x) is called,
26 * for n=1, j1(x) is called,
27 * for n<x, forward recursion us used starting
28 * from values of j0(x) and j1(x).
29 * for n>x, a continued fraction approximation to
30 * j(n,x)/j(n-1,x) is evaluated and then backward
31 * recursion is used starting from a supposed value
32 * for j(n,x). The resulting value of j(0,x) is
33 * compared with the actual value to correct the
34 * supposed value of j(n,x).
35 *
36 * yn(n,x) is similar in all respects, except
37 * that forward recursion is used for all
38 * values of n>1.
40 *
41 */
42
43#include "math.h"
44#include "math_private.h"
45
46static const volatile double vone = 1, vzero = 0;
47
48static const double

--- 12 unchanged lines hidden (view full) ---

61
62 /* J(-n,x) = (-1)^n * J(n, x), J(n, -x) = (-1)^n * J(n, x)
63 * Thus, J(-n,x) = J(n,-x)
64 */
65 EXTRACT_WORDS(hx,lx,x);
66 ix = 0x7fffffff&hx;
67 /* if J(n,NaN) is NaN */
68 if((ix|((u_int32_t)(lx|-lx))>>31)>0x7ff00000) return x+x;
39 */
40
41#include "math.h"
42#include "math_private.h"
43
44static const volatile double vone = 1, vzero = 0;
45
46static const double

--- 12 unchanged lines hidden (view full) ---

59
60 /* J(-n,x) = (-1)^n * J(n, x), J(n, -x) = (-1)^n * J(n, x)
61 * Thus, J(-n,x) = J(n,-x)
62 */
63 EXTRACT_WORDS(hx,lx,x);
64 ix = 0x7fffffff&hx;
65 /* if J(n,NaN) is NaN */
66 if((ix|((u_int32_t)(lx|-lx))>>31)>0x7ff00000) return x+x;
69 if(n<0){
67 if(n<0){
70 n = -n;
71 x = -x;
72 hx ^= 0x80000000;
73 }
74 if(n==0) return(__ieee754_j0(x));
75 if(n==1) return(__ieee754_j1(x));
76 sgn = (n&1)&(hx>>31); /* even n -- 0, odd n -- sign(x) */
77 x = fabs(x);
78 if((ix|lx)==0||ix>=0x7ff00000) /* if x is 0 or inf */
79 b = zero;
68 n = -n;
69 x = -x;
70 hx ^= 0x80000000;
71 }
72 if(n==0) return(__ieee754_j0(x));
73 if(n==1) return(__ieee754_j1(x));
74 sgn = (n&1)&(hx>>31); /* even n -- 0, odd n -- sign(x) */
75 x = fabs(x);
76 if((ix|lx)==0||ix>=0x7ff00000) /* if x is 0 or inf */
77 b = zero;
80 else if((double)n<=x) {
78 else if((double)n<=x) {
81 /* Safe to use J(n+1,x)=2n/x *J(n,x)-J(n-1,x) */
82 if(ix>=0x52D00000) { /* x > 2**302 */
79 /* Safe to use J(n+1,x)=2n/x *J(n,x)-J(n-1,x) */
80 if(ix>=0x52D00000) { /* x > 2**302 */
83 /* (x >> n**2)
81 /* (x >> n**2)
84 * Jn(x) = cos(x-(2n+1)*pi/4)*sqrt(2/x*pi)
85 * Yn(x) = sin(x-(2n+1)*pi/4)*sqrt(2/x*pi)
82 * Jn(x) = cos(x-(2n+1)*pi/4)*sqrt(2/x*pi)
83 * Yn(x) = sin(x-(2n+1)*pi/4)*sqrt(2/x*pi)
86 * Let s=sin(x), c=cos(x),
87 * xn=x-(2n+1)*pi/4, sqt2 = sqrt(2),then
84 * Let s=sin(x), c=cos(x),
85 * xn=x-(2n+1)*pi/4, sqt2 = sqrt(2), then
88 *
89 * n sin(xn)*sqt2 cos(xn)*sqt2
90 * ----------------------------------
91 * 0 s-c c+s
92 * 1 -s-c -c+s
93 * 2 -s+c -c-s
94 * 3 s+c c-s
95 */
96 switch(n&3) {
97 case 0: temp = cos(x)+sin(x); break;
98 case 1: temp = -cos(x)+sin(x); break;
99 case 2: temp = -cos(x)-sin(x); break;
100 case 3: temp = cos(x)-sin(x); break;
101 }
102 b = invsqrtpi*temp/sqrt(x);
86 *
87 * n sin(xn)*sqt2 cos(xn)*sqt2
88 * ----------------------------------
89 * 0 s-c c+s
90 * 1 -s-c -c+s
91 * 2 -s+c -c-s
92 * 3 s+c c-s
93 */
94 switch(n&3) {
95 case 0: temp = cos(x)+sin(x); break;
96 case 1: temp = -cos(x)+sin(x); break;
97 case 2: temp = -cos(x)-sin(x); break;
98 case 3: temp = cos(x)-sin(x); break;
99 }
100 b = invsqrtpi*temp/sqrt(x);
103 } else {
101 } else {
104 a = __ieee754_j0(x);
105 b = __ieee754_j1(x);
106 for(i=1;i<n;i++){
107 temp = b;
108 b = b*((double)(i+i)/x) - a; /* avoid underflow */
109 a = temp;
110 }
111 }
112 } else {
113 if(ix<0x3e100000) { /* x < 2**-29 */
102 a = __ieee754_j0(x);
103 b = __ieee754_j1(x);
104 for(i=1;i<n;i++){
105 temp = b;
106 b = b*((double)(i+i)/x) - a; /* avoid underflow */
107 a = temp;
108 }
109 }
110 } else {
111 if(ix<0x3e100000) { /* x < 2**-29 */
114 /* x is tiny, return the first Taylor expansion of J(n,x)
112 /* x is tiny, return the first Taylor expansion of J(n,x)
115 * J(n,x) = 1/n!*(x/2)^n - ...
116 */
117 if(n>33) /* underflow */
118 b = zero;
119 else {
120 temp = x*0.5; b = temp;
121 for (a=one,i=2;i<=n;i++) {
122 a *= (double)i; /* a = n! */
123 b *= temp; /* b = (x/2)^n */
124 }
125 b = b/a;
126 }
127 } else {
128 /* use backward recurrence */
113 * J(n,x) = 1/n!*(x/2)^n - ...
114 */
115 if(n>33) /* underflow */
116 b = zero;
117 else {
118 temp = x*0.5; b = temp;
119 for (a=one,i=2;i<=n;i++) {
120 a *= (double)i; /* a = n! */
121 b *= temp; /* b = (x/2)^n */
122 }
123 b = b/a;
124 }
125 } else {
126 /* use backward recurrence */
129 /* x x^2 x^2
127 /* x x^2 x^2
130 * J(n,x)/J(n-1,x) = ---- ------ ------ .....
131 * 2n - 2(n+1) - 2(n+2)
132 *
128 * J(n,x)/J(n-1,x) = ---- ------ ------ .....
129 * 2n - 2(n+1) - 2(n+2)
130 *
133 * 1 1 1
131 * 1 1 1
134 * (for large x) = ---- ------ ------ .....
135 * 2n 2(n+1) 2(n+2)
132 * (for large x) = ---- ------ ------ .....
133 * 2n 2(n+1) 2(n+2)
136 * -- - ------ - ------ -
134 * -- - ------ - ------ -
137 * x x x
138 *
139 * Let w = 2n/x and h=2/x, then the above quotient
140 * is equal to the continued fraction:
141 * 1
142 * = -----------------------
143 * 1
144 * w - -----------------
145 * 1
146 * w+h - ---------
147 * w+2h - ...
148 *
149 * To determine how many terms needed, let
150 * Q(0) = w, Q(1) = w(w+h) - 1,
151 * Q(k) = (w+k*h)*Q(k-1) - Q(k-2),
135 * x x x
136 *
137 * Let w = 2n/x and h=2/x, then the above quotient
138 * is equal to the continued fraction:
139 * 1
140 * = -----------------------
141 * 1
142 * w - -----------------
143 * 1
144 * w+h - ---------
145 * w+2h - ...
146 *
147 * To determine how many terms needed, let
148 * Q(0) = w, Q(1) = w(w+h) - 1,
149 * Q(k) = (w+k*h)*Q(k-1) - Q(k-2),
152 * When Q(k) > 1e4 good for single
153 * When Q(k) > 1e9 good for double
154 * When Q(k) > 1e17 good for quadruple
150 * When Q(k) > 1e4 good for single
151 * When Q(k) > 1e9 good for double
152 * When Q(k) > 1e17 good for quadruple
155 */
156 /* determine k */
157 double t,v;
158 double q0,q1,h,tmp; int32_t k,m;
159 w = (n+n)/(double)x; h = 2.0/(double)x;
160 q0 = w; z = w+h; q1 = w*z - 1.0; k=1;
161 while(q1<1.0e9) {
162 k += 1; z += h;

--- 69 unchanged lines hidden (view full) ---

232 if(n<0){
233 n = -n;
234 sign = 1 - ((n&1)<<1);
235 }
236 if(n==0) return(__ieee754_y0(x));
237 if(n==1) return(sign*__ieee754_y1(x));
238 if(ix==0x7ff00000) return zero;
239 if(ix>=0x52D00000) { /* x > 2**302 */
153 */
154 /* determine k */
155 double t,v;
156 double q0,q1,h,tmp; int32_t k,m;
157 w = (n+n)/(double)x; h = 2.0/(double)x;
158 q0 = w; z = w+h; q1 = w*z - 1.0; k=1;
159 while(q1<1.0e9) {
160 k += 1; z += h;

--- 69 unchanged lines hidden (view full) ---

230 if(n<0){
231 n = -n;
232 sign = 1 - ((n&1)<<1);
233 }
234 if(n==0) return(__ieee754_y0(x));
235 if(n==1) return(sign*__ieee754_y1(x));
236 if(ix==0x7ff00000) return zero;
237 if(ix>=0x52D00000) { /* x > 2**302 */
240 /* (x >> n**2)
238 /* (x >> n**2)
241 * Jn(x) = cos(x-(2n+1)*pi/4)*sqrt(2/x*pi)
242 * Yn(x) = sin(x-(2n+1)*pi/4)*sqrt(2/x*pi)
239 * Jn(x) = cos(x-(2n+1)*pi/4)*sqrt(2/x*pi)
240 * Yn(x) = sin(x-(2n+1)*pi/4)*sqrt(2/x*pi)
243 * Let s=sin(x), c=cos(x),
244 * xn=x-(2n+1)*pi/4, sqt2 = sqrt(2),then
241 * Let s=sin(x), c=cos(x),
242 * xn=x-(2n+1)*pi/4, sqt2 = sqrt(2), then
245 *
246 * n sin(xn)*sqt2 cos(xn)*sqt2
247 * ----------------------------------
248 * 0 s-c c+s
249 * 1 -s-c -c+s
250 * 2 -s+c -c-s
251 * 3 s+c c-s
252 */

--- 22 unchanged lines hidden ---
243 *
244 * n sin(xn)*sqt2 cos(xn)*sqt2
245 * ----------------------------------
246 * 0 s-c c+s
247 * 1 -s-c -c+s
248 * 2 -s+c -c-s
249 * 3 s+c c-s
250 */

--- 22 unchanged lines hidden ---