root/usr/src/lib/libm/common/R/besself.c
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
 */
/*
 * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

#pragma weak __j0f = j0f
#pragma weak __j1f = j1f
#pragma weak __jnf = jnf
#pragma weak __y0f = y0f
#pragma weak __y1f = y1f
#pragma weak __ynf = ynf

#include "libm.h"
#include <float.h>

#if defined(__i386) && !defined(__amd64)
extern int __swapRP(int);
#endif

static const float
        zerof   = 0.0f,
        onef    = 1.0f;

static const double C[] = {
        0.0,
        -0.125,
        0.25,
        0.375,
        0.5,
        1.0,
        2.0,
        8.0,
        0.5641895835477562869480794515607725858441,     /* 1/sqrt(pi) */
        0.636619772367581343075535053490057448, /* 2/pi */
        1.0e9,
};

#define zero    C[0]
#define neighth C[1]
#define quarter C[2]
#define three8  C[3]
#define half    C[4]
#define one     C[5]
#define two     C[6]
#define eight   C[7]
#define isqrtpi C[8]
#define tpi     C[9]
#define big     C[10]

static const double Cj0y0[] = {
        0.4861344183386052721391238447e5,       /* pr */
        0.1377662549407112278133438945e6,
        0.1222466364088289731869114004e6,
        0.4107070084315176135583353374e5,
        0.5026073801860637125889039915e4,
        0.1783193659125479654541542419e3,
        0.88010344055383421691677564e0,
        0.4861344183386052721414037058e5,       /* ps */
        0.1378196632630384670477582699e6,
        0.1223967185341006542748936787e6,
        0.4120150243795353639995862617e5,
        0.5068271181053546392490184353e4,
        0.1829817905472769960535671664e3,
        1.0,
        -0.1731210995701068539185611951e3,      /* qr */
        -0.5522559165936166961235240613e3,
        -0.5604935606637346590614529613e3,
        -0.2200430300226009379477365011e3,
        -0.323869355375648849771296746e2,
        -0.14294979207907956223499258e1,
        -0.834690374102384988158918e-2,
        0.1107975037248683865326709645e5,       /* qs */
        0.3544581680627082674651471873e5,
        0.3619118937918394132179019059e5,
        0.1439895563565398007471485822e5,
        0.2190277023344363955930226234e4,
        0.106695157020407986137501682e3,
        1.0,
};

#define pr      Cj0y0
#define ps      (Cj0y0+7)
#define qr      (Cj0y0+14)
#define qs      (Cj0y0+21)

static const double Cj0[] = {
        -2.500000000000003622131880894830476755537e-0001,       /* r0 */
        1.095597547334830263234433855932375353303e-0002,
        -1.819734750463320921799187258987098087697e-0004,
        9.977001946806131657544212501069893930846e-0007,
        1.0,                                                    /* s0 */
        1.867609810662950169966782360588199673741e-0002,
        1.590389206181565490878430827706972074208e-0004,
        6.520867386742583632375520147714499522721e-0007,
        9.999999999999999942156495584397047660949e-0001,        /* r1 */
        -2.389887722731319130476839836908143731281e-0001,
        1.293359476138939027791270393439493640570e-0002,
        -2.770985642343140122168852400228563364082e-0004,
        2.905241575772067678086738389169625218912e-0006,
        -1.636846356264052597969042009265043251279e-0008,
        5.072306160724884775085431059052611737827e-0011,
        -8.187060730684066824228914775146536139112e-0014,
        5.422219326959949863954297860723723423842e-0017,
        1.0,                                                    /* s1 */
        1.101122772686807702762104741932076228349e-0002,
        6.140169310641649223411427764669143978228e-0005,
        2.292035877515152097976946119293215705250e-0007,
        6.356910426504644334558832036362219583789e-0010,
        1.366626326900219555045096999553948891401e-0012,
        2.280399586866739522891837985560481180088e-0015,
        2.801559820648939665270492520004836611187e-0018,
        2.073101088320349159764410261466350732968e-0021,
};

#define r0      Cj0
#define s0      (Cj0+4)
#define r1      (Cj0+8)
#define s1      (Cj0+17)

static const double Cy0[] = {
        -7.380429510868722526754723020704317641941e-0002,       /* u0 */
        1.772607102684869924301459663049874294814e-0001,
        -1.524370666542713828604078090970799356306e-0002,
        4.650819100693891757143771557629924591915e-0004,
        -7.125768872339528975036316108718239946022e-0006,
        6.411017001656104598327565004771515257146e-0008,
        -3.694275157433032553021246812379258781665e-0010,
        1.434364544206266624252820889648445263842e-0012,
        -3.852064731859936455895036286874139896861e-0015,
        7.182052899726138381739945881914874579696e-0018,
        -9.060556574619677567323741194079797987200e-0021,
        7.124435467408860515265552217131230511455e-0024,
        -2.709726774636397615328813121715432044771e-0027,
        1.0,                                                    /* v0 */
        4.678678931512549002587702477349214886475e-0003,
        9.486828955529948534822800829497565178985e-0006,
        1.001495929158861646659010844136682454906e-0008,
        4.725338116256021660204443235685358593611e-0012,
};

#define u0      Cy0
#define v0      (Cy0+13)

static const double Cj1y1[] = {
        -0.4435757816794127857114720794e7,      /* pr0 */
        -0.9942246505077641195658377899e7,
        -0.6603373248364939109255245434e7,
        -0.1523529351181137383255105722e7,
        -0.1098240554345934672737413139e6,
        -0.1611616644324610116477412898e4,
        -0.4435757816794127856828016962e7,      /* ps0 */
        -0.9934124389934585658967556309e7,
        -0.6585339479723087072826915069e7,
        -0.1511809506634160881644546358e7,
        -0.1072638599110382011903063867e6,
        -0.1455009440190496182453565068e4,
        0.3322091340985722351859704442e5,       /* qr0 */
        0.8514516067533570196555001171e5,
        0.6617883658127083517939992166e5,
        0.1849426287322386679652009819e5,
        0.1706375429020768002061283546e4,
        0.3526513384663603218592175580e2,
        0.7087128194102874357377502472e6,       /* qs0 */
        0.1819458042243997298924553839e7,
        0.1419460669603720892855755253e7,
        0.4002944358226697511708610813e6,
        0.3789022974577220264142952256e5,
        0.8638367769604990967475517183e3,
};

#define pr0     Cj1y1
#define ps0     (Cj1y1+6)
#define qr0     (Cj1y1+12)
#define qs0     (Cj1y1+18)

static const double Cj1[] = {
        -6.250000000000002203053200981413218949548e-0002,       /* a0 */
        1.600998455640072901321605101981501263762e-0003,
        -1.963888815948313758552511884390162864930e-0005,
        8.263917341093549759781339713418201620998e-0008,
        1.0e0,                                                  /* b0 */
        1.605069137643004242395356851797873766927e-0002,
        1.149454623251299996428500249509098499383e-0004,
        3.849701673735260970379681807910852327825e-0007,
        4.999999999999999995517408894340485471724e-0001,
        -6.003825028120475684835384519945468075423e-0002,
        2.301719899263321828388344461995355419832e-0003,
        -4.208494869238892934859525221654040304068e-0005,
        4.377745135188837783031540029700282443388e-0007,
        -2.854106755678624335145364226735677754179e-0009,
        1.234002865443952024332943901323798413689e-0011,
        -3.645498437039791058951273508838177134310e-0014,
        7.404320596071797459925377103787837414422e-0017,
        -1.009457448277522275262808398517024439084e-0019,
        8.520158355824819796968771418801019930585e-0023,
        -3.458159926081163274483854614601091361424e-0026,
        1.0e0,                                                  /* b1 */
        4.923499437590484879081138588998986303306e-0003,
        1.054389489212184156499666953501976688452e-0005,
        1.180768373106166527048240364872043816050e-0008,
        5.942665743476099355323245707680648588540e-0012,
};

#define a0      Cj1
#define b0      (Cj1+4)
#define a1      (Cj1+8)
#define b1      (Cj1+20)

static const double Cy1[] = {
        -1.960570906462389461018983259589655961560e-0001,       /* c0 */
        4.931824118350661953459180060007970291139e-0002,
        -1.626975871565393656845930125424683008677e-0003,
        1.359657517926394132692884168082224258360e-0005,
        1.0e0,                                                  /* d0 */
        2.565807214838390835108224713630901653793e-0002,
        3.374175208978404268650522752520906231508e-0004,
        2.840368571306070719539936935220728843177e-0006,
        1.396387402048998277638900944415752207592e-0008,
        -1.960570906462389473336339614647555351626e-0001,       /* c1 */
        5.336268030335074494231369159933012844735e-0002,
        -2.684137504382748094149184541866332033280e-0003,
        5.737671618979185736981543498580051903060e-0005,
        -6.642696350686335339171171785557663224892e-0007,
        4.692417922568160354012347591960362101664e-0009,
        -2.161728635907789319335231338621412258355e-0011,
        6.727353419738316107197644431844194668702e-0014,
        -1.427502986803861372125234355906790573422e-0016,
        2.020392498726806769468143219616642940371e-0019,
        -1.761371948595104156753045457888272716340e-0022,
        7.352828391941157905175042420249225115816e-0026,
        1.0e0,                                                  /* d1 */
        5.029187436727947764916247076102283399442e-0003,
        1.102693095808242775074856548927801750627e-0005,
        1.268035774543174837829534603830227216291e-0008,
        6.579416271766610825192542295821308730206e-0012,
};

#define c0      Cy1
#define d0      (Cy1+4)
#define c1      (Cy1+9)
#define d1      (Cy1+21)


/* core of j0f computation; assumes fx is finite */
static double
__k_j0f(float fx)
{
        double  x, z, s, c, ss, cc, r, t, p0, q0;
        int     ix, i;

        ix = *(int *)&fx & ~0x80000000;
        x = fabs((double)fx);
        if (ix > 0x41000000) {
                /* x > 8; see comments in j0.c */
                s = sin(x);
                c = cos(x);
                if (signbit(s) != signbit(c)) {
                        ss = s - c;
                        cc = -cos(x + x) / ss;
                } else {
                        cc = s + c;
                        ss = -cos(x + x) / cc;
                }
                if (ix > 0x501502f9) {
                        /* x > 1.0e10 */
                        p0 = one;
                        q0 = neighth / x;
                } else {
                        t = eight / x;
                        z = t * t;
                        p0 = (pr[0] + z * (pr[1] + z * (pr[2] + z * (pr[3] +
                            z * (pr[4] + z * (pr[5] + z * pr[6])))))) /
                            (ps[0] + z * (ps[1] + z * (ps[2] + z * (ps[3] +
                            z * (ps[4] + z * (ps[5] + z))))));
                        q0 = ((qr[0] + z * (qr[1] + z * (qr[2] + z * (qr[3] +
                            z * (qr[4] + z * (qr[5] + z * qr[6])))))) /
                            (qs[0] + z * (qs[1] + z * (qs[2] + z * (qs[3] +
                            z * (qs[4] + z * (qs[5] + z))))))) * t;
                }
                return (isqrtpi * (p0 * cc - q0 * ss) / sqrt(x));
        }
        if (ix <= 0x3727c5ac) {
                /* x <= 1.0e-5 */
                if (ix <= 0x219392ef) /* x <= 1.0e-18 */
                        return (one - x);
                return (one - x * x * quarter);
        }
        z = x * x;
        if (ix <= 0x3fa3d70a) {
                /* x <= 1.28 */
                r = r0[0] + z * (r0[1] + z * (r0[2] + z * r0[3]));
                s = s0[0] + z * (s0[1] + z * (s0[2] + z * s0[3]));
                return (one + z * (r / s));
        }
        r = r1[8];
        s = s1[8];
        for (i = 7; i >= 0; i--) {
                r = r * z + r1[i];
                s = s * z + s1[i];
        }
        return (r / s);
}

float
j0f(float fx)
{
        float   f;
        int     ix;
#if defined(__i386) && !defined(__amd64)
        int     rp;
#endif

        ix = *(int *)&fx & ~0x80000000;
        if (ix >= 0x7f800000) {                 /* nan or inf */
                if (ix > 0x7f800000)
                        return (fx * fx);
                return (zerof);
        }

#if defined(__i386) && !defined(__amd64)
        rp = __swapRP(fp_extended);
#endif
        f = (float)__k_j0f(fx);
#if defined(__i386) && !defined(__amd64)
        if (rp != fp_extended)
                (void) __swapRP(rp);
#endif
        return (f);
}

/* core of y0f computation; assumes fx is finite and positive */
static double
__k_y0f(float fx)
{
        double  x, z, s, c, ss, cc, t, p0, q0, u, v;
        int     ix, i;

        ix = *(int *)&fx;
        x = (double)fx;
        if (ix > 0x41000000) {
                /* x > 8; see comments in j0.c */
                s = sin(x);
                c = cos(x);
                if (signbit(s) != signbit(c)) {
                        ss = s - c;
                        cc = -cos(x + x) / ss;
                } else {
                        cc = s + c;
                        ss = -cos(x + x) / cc;
                }
                if (ix > 0x501502f9) {
                        /* x > 1.0e10 */
                        p0 = one;
                        q0 = neighth / x;
                } else {
                        t = eight / x;
                        z = t * t;
                        p0 = (pr[0] + z * (pr[1] + z * (pr[2] + z * (pr[3] +
                            z * (pr[4] + z * (pr[5] + z * pr[6])))))) /
                            (ps[0] + z * (ps[1] + z * (ps[2] + z * (ps[3] +
                            z * (ps[4] + z * (ps[5] + z))))));
                        q0 = ((qr[0] + z * (qr[1] + z * (qr[2] + z * (qr[3] +
                            z * (qr[4] + z * (qr[5] + z * qr[6])))))) /
                            (qs[0] + z * (qs[1] + z * (qs[2] + z * (qs[3] +
                            z * (qs[4] + z * (qs[5] + z))))))) * t;
                }
                return (isqrtpi * (p0 * ss + q0 * cc) / sqrt(x));
        }
        if (ix <= 0x219392ef) /* x <= 1.0e-18 */
                return (u0[0] + tpi * log(x));
        z = x * x;
        u = u0[12];
        for (i = 11; i >= 0; i--)
                u = u * z + u0[i];
        v = v0[0] + z * (v0[1] + z * (v0[2] + z * (v0[3] + z * v0[4])));
        return (u / v + tpi * (__k_j0f(fx) * log(x)));
}

float
y0f(float fx)
{
        float   f;
        int     ix;
#if defined(__i386) && !defined(__amd64)
        int     rp;
#endif

        ix = *(int *)&fx;
        if ((ix & ~0x80000000) > 0x7f800000)    /* nan */
                return (fx * fx);
        if (ix <= 0) {                          /* zero or negative */
                if ((ix << 1) == 0)
                        return (-onef / zerof);
                return (zerof / zerof);
        }
        if (ix == 0x7f800000)                   /* +inf */
                return (zerof);

#if defined(__i386) && !defined(__amd64)
        rp = __swapRP(fp_extended);
#endif
        f = (float)__k_y0f(fx);
#if defined(__i386) && !defined(__amd64)
        if (rp != fp_extended)
                (void) __swapRP(rp);
#endif
        return (f);
}

/* core of j1f computation; assumes fx is finite */
static double
__k_j1f(float fx)
{
        double  x, z, s, c, ss, cc, r, t, p1, q1;
        int     i, ix, sgn;

        ix = *(int *)&fx;
        sgn = (unsigned)ix >> 31;
        ix &= ~0x80000000;
        x = fabs((double)fx);
        if (ix > 0x41000000) {
                /* x > 8; see comments in j1.c */
                s = sin(x);
                c = cos(x);
                if (signbit(s) != signbit(c)) {
                        cc = s - c;
                        ss = cos(x + x) / cc;
                } else {
                        ss = -s - c;
                        cc = cos(x + x) / ss;
                }
                if (ix > 0x501502f9) {
                        /* x > 1.0e10 */
                        p1 = one;
                        q1 = three8 / x;
                } else {
                        t = eight / x;
                        z = t * t;
                        p1 = (pr0[0] + z * (pr0[1] + z * (pr0[2] + z *
                            (pr0[3] + z * (pr0[4] + z * pr0[5]))))) /
                            (ps0[0] + z * (ps0[1] + z * (ps0[2] + z *
                            (ps0[3] + z * (ps0[4] + z * (ps0[5] + z))))));
                        q1 = ((qr0[0] + z * (qr0[1] + z * (qr0[2] + z *
                            (qr0[3] + z * (qr0[4] + z * qr0[5]))))) /
                            (qs0[0] + z * (qs0[1] + z * (qs0[2] + z *
                            (qs0[3] + z * (qs0[4] + z * (qs0[5] + z))))))) * t;
                }
                t = isqrtpi * (p1 * cc - q1 * ss) / sqrt(x);
                return ((sgn)? -t : t);
        }
        if (ix <= 0x3727c5ac) {
                /* x <= 1.0e-5 */
                if (ix <= 0x219392ef) /* x <= 1.0e-18 */
                        t = half * x;
                else
                        t = x * (half + neighth * x * x);
                return ((sgn)? -t : t);
        }
        z = x * x;
        if (ix < 0x3fa3d70a) {
                /* x < 1.28 */
                r = a0[0] + z * (a0[1] + z * (a0[2] + z * a0[3]));
                s = b0[0] + z * (b0[1] + z * (b0[2] + z * b0[3]));
                t = x * half + x * (z * (r / s));
        } else {
                r = a1[11];
                for (i = 10; i >= 0; i--)
                        r = r * z + a1[i];
                s = b1[0] + z * (b1[1] + z * (b1[2] + z * (b1[3] + z * b1[4])));
                t = x * (r / s);
        }
        return ((sgn)? -t : t);
}

float
j1f(float fx)
{
        float   f;
        int     ix;
#if defined(__i386) && !defined(__amd64)
        int     rp;
#endif

        ix = *(int *)&fx & ~0x80000000;
        if (ix >= 0x7f800000)                   /* nan or inf */
                return (onef / fx);

#if defined(__i386) && !defined(__amd64)
        rp = __swapRP(fp_extended);
#endif
        f = (float)__k_j1f(fx);
#if defined(__i386) && !defined(__amd64)
        if (rp != fp_extended)
                (void) __swapRP(rp);
#endif
        return (f);
}

/* core of y1f computation; assumes fx is finite and positive */
static double
__k_y1f(float fx)
{
        double  x, z, s, c, ss, cc, u, v, p1, q1, t;
        int     i, ix;

        ix = *(int *)&fx;
        x = (double)fx;
        if (ix > 0x41000000) {
                /* x > 8; see comments in j1.c */
                s = sin(x);
                c = cos(x);
                if (signbit(s) != signbit(c)) {
                        cc = s - c;
                        ss = cos(x + x) / cc;
                } else {
                        ss = -s - c;
                        cc = cos(x + x) / ss;
                }
                if (ix > 0x501502f9) {
                        /* x > 1.0e10 */
                        p1 = one;
                        q1 = three8 / x;
                } else {
                        t = eight / x;
                        z = t * t;
                        p1 = (pr0[0] + z * (pr0[1] + z * (pr0[2] + z *
                            (pr0[3] + z * (pr0[4] + z * pr0[5]))))) /
                            (ps0[0] + z * (ps0[1] + z * (ps0[2] + z *
                            (ps0[3] + z * (ps0[4] + z * (ps0[5] + z))))));
                        q1 = ((qr0[0] + z * (qr0[1] + z * (qr0[2] + z *
                            (qr0[3] + z * (qr0[4] + z * qr0[5]))))) /
                            (qs0[0] + z * (qs0[1] + z * (qs0[2] + z *
                            (qs0[3] + z * (qs0[4] + z * (qs0[5] + z))))))) * t;
                }
                return (isqrtpi * (p1 * ss + q1 * cc) / sqrt(x));
        }
        if (ix <= 0x219392ef) /* x <= 1.0e-18 */
                return (-tpi / x);
        z = x * x;
        if (ix < 0x3fa3d70a) {
                /* x < 1.28 */
                u = c0[0] + z * (c0[1] + z * (c0[2] + z * c0[3]));
                v = d0[0] + z * (d0[1] + z * (d0[2] + z * (d0[3] + z * d0[4])));
        } else {
                u = c1[11];
                for (i = 10; i >= 0; i--)
                        u = u * z + c1[i];
                v = d1[0] + z * (d1[1] + z * (d1[2] + z * (d1[3] + z * d1[4])));
        }
        return (x * (u / v) + tpi * (__k_j1f(fx) * log(x) - one / x));
}

float
y1f(float fx)
{
        float   f;
        int     ix;
#if defined(__i386) && !defined(__amd64)
        int     rp;
#endif

        ix = *(int *)&fx;
        if ((ix & ~0x80000000) > 0x7f800000)    /* nan */
                return (fx * fx);
        if (ix <= 0) {                          /* zero or negative */
                if ((ix << 1) == 0)
                        return (-onef / zerof);
                return (zerof / zerof);
        }
        if (ix == 0x7f800000)                   /* +inf */
                return (zerof);

#if defined(__i386) && !defined(__amd64)
        rp = __swapRP(fp_extended);
#endif
        f = (float)__k_y1f(fx);
#if defined(__i386) && !defined(__amd64)
        if (rp != fp_extended)
                (void) __swapRP(rp);
#endif
        return (f);
}

float
jnf(int n, float fx)
{
        double  a, b, temp, x, z, w, t, q0, q1, h;
        float   f;
        int     i, ix, sgn, m, k;
#if defined(__i386) && !defined(__amd64)
        int     rp;
#endif

        if (n < 0) {
                n = -n;
                fx = -fx;
        }
        if (n == 0)
                return (j0f(fx));
        if (n == 1)
                return (j1f(fx));

        ix = *(int *)&fx;
        sgn = (n & 1)? ((unsigned)ix >> 31) : 0;
        ix &= ~0x80000000;
        if (ix >= 0x7f800000) {         /* nan or inf */
                if (ix > 0x7f800000)
                        return (fx * fx);
                return ((sgn)? -zerof : zerof);
        }
        if ((ix << 1) == 0)
                return ((sgn)? -zerof : zerof);

#if defined(__i386) && !defined(__amd64)
        rp = __swapRP(fp_extended);
#endif
        fx = fabsf(fx);
        x = (double)fx;
        if ((double)n <= x) {
                /* safe to use J(n+1,x) = 2n/x * J(n,x) - J(n-1,x) */
                a = __k_j0f(fx);
                b = __k_j1f(fx);
                for (i = 1; i < n; i++) {
                        temp = b;
                        b = b * ((double)(i + i) / x) - a;
                        a = temp;
                }
                f = (float)b;
#if defined(__i386) && !defined(__amd64)
                if (rp != fp_extended)
                        (void) __swapRP(rp);
#endif
                return ((sgn)? -f : f);
        }
        if (ix < 0x3089705f) {
                /* x < 1.0e-9; use J(n,x) = 1/n! * (x / 2)^n */
                if (n > 6)
                        n = 6;  /* result underflows to zero for n >= 6 */
                b = t = half * x;
                a = one;
                for (i = 2; i <= n; i++) {
                        b *= t;
                        a *= (double)i;
                }
                b /= a;
        } else {
                /*
                 * Use the backward recurrence:
                 *
                 *                      x      x^2      x^2
                 *  J(n,x)/J(n-1,x) =  ---- - ------ - ------   .....
                 *                      2n    2(n+1)   2(n+2)
                 *
                 * Let w = 2n/x and h = 2/x.  Then the above quotient
                 * is equal to the continued fraction:
                 *                   1
                 *      = -----------------------
                 *                      1
                 *         w - -----------------
                 *                        1
                 *              w+h - ---------
                 *                      w+2h - ...
                 *
                 * To determine how many terms are needed, run the
                 * recurrence
                 *
                 *      Q(0) = w,
                 *      Q(1) = w(w+h) - 1,
                 *      Q(k) = (w+k*h)*Q(k-1) - Q(k-2).
                 *
                 * Then when Q(k) > 1e4, k is large enough for single
                 * precision.
                 */
/* XXX NOT DONE - rework this */
                w = (n + n) / x;
                h = two / x;
                q0 = w;
                z = w + h;
                q1 = w * z - one;
                k = 1;
                while (q1 < big) {
                        k++;
                        z += h;
                        temp = z * q1 - q0;
                        q0 = q1;
                        q1 = temp;
                }
                m = n + n;
                t = zero;
                for (i = (n + k) << 1; i >= m; i -= 2)
                        t = one / ((double)i / x - t);
                a = t;
                b = one;
                /*
                 * estimate log((2/x)^n*n!) = n*log(2/x)+n*ln(n)
                 * hence, if n*(log(2n/x)) > ...
                 *      single 8.8722839355e+01
                 *      double 7.09782712893383973096e+02
                 *      then recurrent value may overflow and the result is
                 *      likely underflow to zero
                 */
                temp = (double)n;
                temp *= log((two / x) * temp);
                if (temp < 7.09782712893383973096e+02) {
                        for (i = n - 1; i > 0; i--) {
                                temp = b;
                                b = b * ((double)(i + i) / x) - a;
                                a = temp;
                        }
                } else {
                        for (i = n - 1; i > 0; i--) {
                                temp = b;
                                b = b * ((double)(i + i) / x) - a;
                                a = temp;
                                if (b > 1.0e100) {
                                        a /= b;
                                        t /= b;
                                        b = one;
                                }
                        }
                }
                b = (t * __k_j0f(fx) / b);
        }
        f = (float)b;
#if defined(__i386) && !defined(__amd64)
        if (rp != fp_extended)
                (void) __swapRP(rp);
#endif
        return ((sgn)? -f : f);
}

float
ynf(int n, float fx)
{
        double  a, b, temp, x;
        float   f;
        int     i, sign, ix;
#if defined(__i386) && !defined(__amd64)
        int     rp;
#endif

        sign = 0;
        if (n < 0) {
                n = -n;
                if (n & 1)
                        sign = 1;
        }
        if (n == 0)
                return (y0f(fx));
        if (n == 1)
                return ((sign)? -y1f(fx) : y1f(fx));

        ix = *(int *)&fx;
        if ((ix & ~0x80000000) > 0x7f800000)    /* nan */
                return (fx * fx);
        if (ix <= 0) {                          /* zero or negative */
                if ((ix << 1) == 0)
                        return (-onef / zerof);
                return (zerof / zerof);
        }
        if (ix == 0x7f800000)                   /* +inf */
                return (zerof);

#if defined(__i386) && !defined(__amd64)
        rp = __swapRP(fp_extended);
#endif
        a = __k_y0f(fx);
        b = __k_y1f(fx);
        x = (double)fx;
        for (i = 1; i < n; i++) {
                temp = b;
                b *= (double)(i + i) / x;
                if (b <= -DBL_MAX)
                        break;
                b -= a;
                a = temp;
        }
        f = (float)b;
#if defined(__i386) && !defined(__amd64)
        if (rp != fp_extended)
                (void) __swapRP(rp);
#endif
        return ((sign)? -f : f);
}