glibc/sysdeps/ieee754/dbl-64/e_jn.c

348 lines
8.6 KiB
C
Raw Normal View History

/* @(#)e_jn.c 5.1 93/09/24 */
/*
* ====================================================
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
*
* Developed at SunPro, a Sun Microsystems, Inc. business.
* Permission to use, copy, modify, and distribute this
* software is freely granted, provided that this notice
* is preserved.
* ====================================================
*/
/*
* __ieee754_jn(n, x), __ieee754_yn(n, x)
* floating point Bessel's function of the 1st and 2nd kind
* of order n
*
* Special cases:
* y0(0)=y1(0)=yn(n,0) = -inf with overflow signal;
* y0(-ve)=y1(-ve)=yn(n,-ve) are NaN with invalid signal.
* Note 2. About jn(n,x), yn(n,x)
* For n=0, j0(x) is called,
* for n=1, j1(x) is called,
* for n<x, forward recursion us used starting
* from values of j0(x) and j1(x).
* for n>x, a continued fraction approximation to
* j(n,x)/j(n-1,x) is evaluated and then backward
* recursion is used starting from a supposed value
* for j(n,x). The resulting value of j(0,x) is
* compared with the actual value to correct the
* supposed value of j(n,x).
*
* yn(n,x) is similar in all respects, except
* that forward recursion is used for all
* values of n>1.
*
*/
2012-07-25 10:59:36 +00:00
#include <errno.h>
Fix yn overflow handling in non-default rounding modes (bug 16561, bug 16562). This patch fixes bugs 16561 and 16562, bad results of yn in overflow cases in non-default rounding modes, both because an intermediate overflow in the recurrence does not get detected if the result is not an infinity and because an overflowing result may occur in the wrong sign. The fix is to set FE_TONEAREST mode internally for the parts of the function where such overflows can occur (which includes the call to y1 - where yn is used to compute a Bessel function of order -1, negating the result of y1 isn't correct for overflowing results in directed rounding modes) and then compute an overflowing value in the original rounding mode if the to-nearest result was an infinity. Tested x86_64 and x86 and ulps updated accordingly. Also tested for mips64 and powerpc32 to test the ldbl-128 and ldbl-128ibm changes. (The tests for these bugs were added in my previous y1 patch, so the only thing this patch has to do with the testsuite is enable yn testing in all rounding modes.) [BZ #16561] [BZ #16562] * sysdeps/ieee754/dbl-64/e_jn.c: Include <float.h>. (__ieee754_yn): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/flt-32/e_jnf.c: Include <float.h>. (__ieee754_ynf): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-128/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-96/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/i386/fpu/fenv_private.h [!__SSE2_MATH__] (libc_feholdsetround_ctx): New macro. * math/libm-test.inc (yn_test): Use ALL_RM_TEST. * sysdeps/i386/fpu/libm-test-ulps: Update. * sysdeps/x86_64/fpu/libm-test-ulps : Likewise.
2014-06-27 14:52:13 +00:00
#include <float.h>
#include <math.h>
#include <math_private.h>
static const double
2013-10-17 14:03:24 +00:00
invsqrtpi = 5.64189583547756279280e-01, /* 0x3FE20DD7, 0x50429B6D */
two = 2.00000000000000000000e+00, /* 0x40000000, 0x00000000 */
one = 1.00000000000000000000e+00; /* 0x3FF00000, 0x00000000 */
2013-10-17 14:03:24 +00:00
static const double zero = 0.00000000000000000000e+00;
double
2013-10-17 14:03:24 +00:00
__ieee754_jn (int n, double x)
{
2013-10-17 14:03:24 +00:00
int32_t i, hx, ix, lx, sgn;
Use round-to-nearest internally in jn, test with ALL_RM_TEST (bug 18602). Some existing jn tests, if run in non-default rounding modes, produce errors above those accepted in glibc, which causes problems for moving tests of jn to use ALL_RM_TEST. This patch makes jn set rounding to-nearest internally, as was done for yn some time ago, then computes the appropriate underflowing value for results that underflowed to zero in to-nearest, and moves the tests to ALL_RM_TEST. It does nothing about the general inaccuracy of Bessel function implementations in glibc, though it should make jn more accurate on average in non-default rounding modes through reduced error accumulation. The recomputation of results that underflowed to zero should as a side-effect fix some cases of bug 16559, where jn just used an exact zero, but that is *not* the goal of this patch and other cases of that bug remain unfixed. (Most of the changes in the patch are reindentation to add new scopes for SET_RESTORE_ROUND*.) Tested for x86_64, x86, powerpc and mips64. [BZ #16559] [BZ #18602] * sysdeps/ieee754/dbl-64/e_jn.c (__ieee754_jn): Set round-to-nearest internally then recompute results that underflowed to zero in the original rounding mode. * sysdeps/ieee754/flt-32/e_jnf.c (__ieee754_jnf): Likewise. * sysdeps/ieee754/ldbl-128/e_jnl.c (__ieee754_jnl): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c (__ieee754_jnl): Likewise. * sysdeps/ieee754/ldbl-96/e_jnl.c (__ieee754_jnl): Likewise * math/libm-test.inc (jn_test): Use ALL_RM_TEST. * sysdeps/i386/fpu/libm-test-ulps: Update. * sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
2015-06-25 21:46:02 +00:00
double a, b, temp, di, ret;
2013-10-17 14:03:24 +00:00
double z, w;
2013-10-17 14:03:24 +00:00
/* J(-n,x) = (-1)^n * J(n, x), J(n, -x) = (-1)^n * J(n, x)
* Thus, J(-n,x) = J(n,-x)
*/
EXTRACT_WORDS (hx, lx, x);
ix = 0x7fffffff & hx;
/* if J(n,NaN) is NaN */
if (__glibc_unlikely ((ix | ((u_int32_t) (lx | -lx)) >> 31) > 0x7ff00000))
2013-10-17 14:03:24 +00:00
return x + x;
if (n < 0)
{
n = -n;
x = -x;
hx ^= 0x80000000;
}
if (n == 0)
return (__ieee754_j0 (x));
if (n == 1)
return (__ieee754_j1 (x));
sgn = (n & 1) & (hx >> 31); /* even n -- 0, odd n -- sign(x) */
x = fabs (x);
Use round-to-nearest internally in jn, test with ALL_RM_TEST (bug 18602). Some existing jn tests, if run in non-default rounding modes, produce errors above those accepted in glibc, which causes problems for moving tests of jn to use ALL_RM_TEST. This patch makes jn set rounding to-nearest internally, as was done for yn some time ago, then computes the appropriate underflowing value for results that underflowed to zero in to-nearest, and moves the tests to ALL_RM_TEST. It does nothing about the general inaccuracy of Bessel function implementations in glibc, though it should make jn more accurate on average in non-default rounding modes through reduced error accumulation. The recomputation of results that underflowed to zero should as a side-effect fix some cases of bug 16559, where jn just used an exact zero, but that is *not* the goal of this patch and other cases of that bug remain unfixed. (Most of the changes in the patch are reindentation to add new scopes for SET_RESTORE_ROUND*.) Tested for x86_64, x86, powerpc and mips64. [BZ #16559] [BZ #18602] * sysdeps/ieee754/dbl-64/e_jn.c (__ieee754_jn): Set round-to-nearest internally then recompute results that underflowed to zero in the original rounding mode. * sysdeps/ieee754/flt-32/e_jnf.c (__ieee754_jnf): Likewise. * sysdeps/ieee754/ldbl-128/e_jnl.c (__ieee754_jnl): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c (__ieee754_jnl): Likewise. * sysdeps/ieee754/ldbl-96/e_jnl.c (__ieee754_jnl): Likewise * math/libm-test.inc (jn_test): Use ALL_RM_TEST. * sysdeps/i386/fpu/libm-test-ulps: Update. * sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
2015-06-25 21:46:02 +00:00
{
SET_RESTORE_ROUND (FE_TONEAREST);
if (__glibc_unlikely ((ix | lx) == 0 || ix >= 0x7ff00000))
/* if x is 0 or inf */
return sgn == 1 ? -zero : zero;
else if ((double) n <= x)
{
/* Safe to use J(n+1,x)=2n/x *J(n,x)-J(n-1,x) */
if (ix >= 0x52D00000) /* x > 2**302 */
{ /* (x >> n**2)
2013-10-17 14:03:24 +00:00
* Jn(x) = cos(x-(2n+1)*pi/4)*sqrt(2/x*pi)
* Yn(x) = sin(x-(2n+1)*pi/4)*sqrt(2/x*pi)
* Let s=sin(x), c=cos(x),
* xn=x-(2n+1)*pi/4, sqt2 = sqrt(2),then
*
* n sin(xn)*sqt2 cos(xn)*sqt2
* ----------------------------------
* 0 s-c c+s
* 1 -s-c -c+s
* 2 -s+c -c-s
* 3 s+c c-s
*/
Use round-to-nearest internally in jn, test with ALL_RM_TEST (bug 18602). Some existing jn tests, if run in non-default rounding modes, produce errors above those accepted in glibc, which causes problems for moving tests of jn to use ALL_RM_TEST. This patch makes jn set rounding to-nearest internally, as was done for yn some time ago, then computes the appropriate underflowing value for results that underflowed to zero in to-nearest, and moves the tests to ALL_RM_TEST. It does nothing about the general inaccuracy of Bessel function implementations in glibc, though it should make jn more accurate on average in non-default rounding modes through reduced error accumulation. The recomputation of results that underflowed to zero should as a side-effect fix some cases of bug 16559, where jn just used an exact zero, but that is *not* the goal of this patch and other cases of that bug remain unfixed. (Most of the changes in the patch are reindentation to add new scopes for SET_RESTORE_ROUND*.) Tested for x86_64, x86, powerpc and mips64. [BZ #16559] [BZ #18602] * sysdeps/ieee754/dbl-64/e_jn.c (__ieee754_jn): Set round-to-nearest internally then recompute results that underflowed to zero in the original rounding mode. * sysdeps/ieee754/flt-32/e_jnf.c (__ieee754_jnf): Likewise. * sysdeps/ieee754/ldbl-128/e_jnl.c (__ieee754_jnl): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c (__ieee754_jnl): Likewise. * sysdeps/ieee754/ldbl-96/e_jnl.c (__ieee754_jnl): Likewise * math/libm-test.inc (jn_test): Use ALL_RM_TEST. * sysdeps/i386/fpu/libm-test-ulps: Update. * sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
2015-06-25 21:46:02 +00:00
double s;
double c;
__sincos (x, &s, &c);
switch (n & 3)
{
case 0: temp = c + s; break;
case 1: temp = -c + s; break;
case 2: temp = -c - s; break;
case 3: temp = c - s; break;
}
b = invsqrtpi * temp / __ieee754_sqrt (x);
}
else
{
a = __ieee754_j0 (x);
b = __ieee754_j1 (x);
for (i = 1; i < n; i++)
{
temp = b;
b = b * ((double) (i + i) / x) - a; /* avoid underflow */
a = temp;
}
}
}
else
{
if (ix < 0x3e100000) /* x < 2**-29 */
{ /* x is tiny, return the first Taylor expansion of J(n,x)
2013-10-17 14:03:24 +00:00
* J(n,x) = 1/n!*(x/2)^n - ...
*/
Use round-to-nearest internally in jn, test with ALL_RM_TEST (bug 18602). Some existing jn tests, if run in non-default rounding modes, produce errors above those accepted in glibc, which causes problems for moving tests of jn to use ALL_RM_TEST. This patch makes jn set rounding to-nearest internally, as was done for yn some time ago, then computes the appropriate underflowing value for results that underflowed to zero in to-nearest, and moves the tests to ALL_RM_TEST. It does nothing about the general inaccuracy of Bessel function implementations in glibc, though it should make jn more accurate on average in non-default rounding modes through reduced error accumulation. The recomputation of results that underflowed to zero should as a side-effect fix some cases of bug 16559, where jn just used an exact zero, but that is *not* the goal of this patch and other cases of that bug remain unfixed. (Most of the changes in the patch are reindentation to add new scopes for SET_RESTORE_ROUND*.) Tested for x86_64, x86, powerpc and mips64. [BZ #16559] [BZ #18602] * sysdeps/ieee754/dbl-64/e_jn.c (__ieee754_jn): Set round-to-nearest internally then recompute results that underflowed to zero in the original rounding mode. * sysdeps/ieee754/flt-32/e_jnf.c (__ieee754_jnf): Likewise. * sysdeps/ieee754/ldbl-128/e_jnl.c (__ieee754_jnl): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c (__ieee754_jnl): Likewise. * sysdeps/ieee754/ldbl-96/e_jnl.c (__ieee754_jnl): Likewise * math/libm-test.inc (jn_test): Use ALL_RM_TEST. * sysdeps/i386/fpu/libm-test-ulps: Update. * sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
2015-06-25 21:46:02 +00:00
if (n > 33) /* underflow */
b = zero;
else
{
temp = x * 0.5; b = temp;
for (a = one, i = 2; i <= n; i++)
{
a *= (double) i; /* a = n! */
b *= temp; /* b = (x/2)^n */
}
b = b / a;
}
}
else
{
/* use backward recurrence */
/* x x^2 x^2
* J(n,x)/J(n-1,x) = ---- ------ ------ .....
* 2n - 2(n+1) - 2(n+2)
*
* 1 1 1
* (for large x) = ---- ------ ------ .....
* 2n 2(n+1) 2(n+2)
* -- - ------ - ------ -
* x x x
*
* 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 needed, let
* Q(0) = w, Q(1) = w(w+h) - 1,
* Q(k) = (w+k*h)*Q(k-1) - Q(k-2),
* When Q(k) > 1e4 good for single
* When Q(k) > 1e9 good for double
* When Q(k) > 1e17 good for quadruple
*/
/* determine k */
double t, v;
double q0, q1, h, tmp; int32_t k, m;
w = (n + n) / (double) x; h = 2.0 / (double) x;
q0 = w; z = w + h; q1 = w * z - 1.0; k = 1;
while (q1 < 1.0e9)
{
k += 1; z += h;
tmp = z * q1 - q0;
q0 = q1;
q1 = tmp;
}
m = n + n;
for (t = zero, i = 2 * (n + k); i >= m; i -= 2)
t = one / (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
* long double 1.1356523406294143949491931077970765006170e+04
* then recurrent value may overflow and the result is
* likely underflow to zero
*/
tmp = n;
v = two / x;
tmp = tmp * __ieee754_log (fabs (v * tmp));
if (tmp < 7.09782712893383973096e+02)
{
for (i = n - 1, di = (double) (i + i); i > 0; i--)
{
temp = b;
b *= di;
b = b / x - a;
a = temp;
di -= two;
}
}
else
{
for (i = n - 1, di = (double) (i + i); i > 0; i--)
{
temp = b;
b *= di;
b = b / x - a;
a = temp;
di -= two;
/* scale b to avoid spurious overflow */
if (b > 1e100)
{
a /= b;
t /= b;
b = one;
}
}
}
/* j0() and j1() suffer enormous loss of precision at and
* near zero; however, we know that their zero points never
* coincide, so just choose the one further away from zero.
*/
z = __ieee754_j0 (x);
w = __ieee754_j1 (x);
if (fabs (z) >= fabs (w))
b = (t * z / b);
else
b = (t * w / a);
}
}
if (sgn == 1)
ret = -b;
else
ret = b;
ret = math_narrow_eval (ret);
Use round-to-nearest internally in jn, test with ALL_RM_TEST (bug 18602). Some existing jn tests, if run in non-default rounding modes, produce errors above those accepted in glibc, which causes problems for moving tests of jn to use ALL_RM_TEST. This patch makes jn set rounding to-nearest internally, as was done for yn some time ago, then computes the appropriate underflowing value for results that underflowed to zero in to-nearest, and moves the tests to ALL_RM_TEST. It does nothing about the general inaccuracy of Bessel function implementations in glibc, though it should make jn more accurate on average in non-default rounding modes through reduced error accumulation. The recomputation of results that underflowed to zero should as a side-effect fix some cases of bug 16559, where jn just used an exact zero, but that is *not* the goal of this patch and other cases of that bug remain unfixed. (Most of the changes in the patch are reindentation to add new scopes for SET_RESTORE_ROUND*.) Tested for x86_64, x86, powerpc and mips64. [BZ #16559] [BZ #18602] * sysdeps/ieee754/dbl-64/e_jn.c (__ieee754_jn): Set round-to-nearest internally then recompute results that underflowed to zero in the original rounding mode. * sysdeps/ieee754/flt-32/e_jnf.c (__ieee754_jnf): Likewise. * sysdeps/ieee754/ldbl-128/e_jnl.c (__ieee754_jnl): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c (__ieee754_jnl): Likewise. * sysdeps/ieee754/ldbl-96/e_jnl.c (__ieee754_jnl): Likewise * math/libm-test.inc (jn_test): Use ALL_RM_TEST. * sysdeps/i386/fpu/libm-test-ulps: Update. * sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
2015-06-25 21:46:02 +00:00
}
if (ret == 0)
{
ret = math_narrow_eval (__copysign (DBL_MIN, ret) * DBL_MIN);
__set_errno (ERANGE);
}
Refactor code forcing underflow exceptions. Various floating-point functions have code to force underflow exceptions if a tiny result was computed in a way that might not have resulted in such exceptions even though the result is inexact. This typically uses math_force_eval to ensure that the underflowing expression is evaluated, but sometimes uses volatile. This patch refactors such code to use three new macros math_check_force_underflow, math_check_force_underflow_nonneg and math_check_force_underflow_complex (which in turn use math_force_eval). In the limited number of cases not suited to a simple conversion to these macros, existing uses of volatile are changed to use math_force_eval instead. The converted code does not always execute exactly the same sequence of operations as the original code, but the overall effects should be the same. Tested for x86_64, x86, mips64 and powerpc. * sysdeps/generic/math_private.h (fabs_tg): New macro. (min_of_type): Likewise. (math_check_force_underflow): Likewise. (math_check_force_underflow_nonneg): Likewise. (math_check_force_underflow_complex): Likewise. * math/e_exp2l.c (__ieee754_exp2l): Use math_check_force_underflow_nonneg. * math/k_casinh.c (__kernel_casinh): Likewise. * math/k_casinhf.c (__kernel_casinhf): Likewise. * math/k_casinhl.c (__kernel_casinhl): Likewise. * math/s_catan.c (__catan): Use math_check_force_underflow_complex. * math/s_catanf.c (__catanf): Likewise. * math/s_catanh.c (__catanh): Likewise. * math/s_catanhf.c (__catanhf): Likewise. * math/s_catanhl.c (__catanhl): Likewise. * math/s_catanl.c (__catanl): Likewise. * math/s_ccosh.c (__ccosh): Likewise. * math/s_ccoshf.c (__ccoshf): Likewise. * math/s_ccoshl.c (__ccoshl): Likewise. * math/s_cexp.c (__cexp): Likewise. * math/s_cexpf.c (__cexpf): Likewise. * math/s_cexpl.c (__cexpl): Likewise. * math/s_clog.c (__clog): Use math_check_force_underflow_nonneg. * math/s_clog10.c (__clog10): Likewise. * math/s_clog10f.c (__clog10f): Likewise. * math/s_clog10l.c (__clog10l): Likewise. * math/s_clogf.c (__clogf): Likewise. * math/s_clogl.c (__clogl): Likewise. * math/s_csin.c (__csin): Use math_check_force_underflow_complex. * math/s_csinf.c (__csinf): Likewise. * math/s_csinh.c (__csinh): Likewise. * math/s_csinhf.c (__csinhf): Likewise. * math/s_csinhl.c (__csinhl): Likewise. * math/s_csinl.c (__csinl): Likewise. * math/s_csqrt.c (__csqrt): Use math_check_force_underflow. * math/s_csqrtf.c (__csqrtf): Likewise. * math/s_csqrtl.c (__csqrtl): Likewise. * math/s_ctan.c (__ctan): Use math_check_force_underflow_complex. * math/s_ctanf.c (__ctanf): Likewise. * math/s_ctanh.c (__ctanh): Likewise. * math/s_ctanhf.c (__ctanhf): Likewise. * math/s_ctanhl.c (__ctanhl): Likewise. * math/s_ctanl.c (__ctanl): Likewise. * stdlib/strtod_l.c (round_and_return): Use math_force_eval instead of volatile. * sysdeps/ieee754/dbl-64/e_asin.c (__ieee754_asin): Use math_check_force_underflow. * sysdeps/ieee754/dbl-64/e_atanh.c (__ieee754_atanh): Likewise. * sysdeps/ieee754/dbl-64/e_exp.c (__ieee754_exp): Do not use volatile when forcing underflow. * sysdeps/ieee754/dbl-64/e_exp2.c (__ieee754_exp2): Use math_check_force_underflow_nonneg. * sysdeps/ieee754/dbl-64/e_gamma_r.c (__ieee754_gamma_r): Likewise. * sysdeps/ieee754/dbl-64/e_j1.c (__ieee754_j1): Use math_check_force_underflow. * sysdeps/ieee754/dbl-64/e_jn.c (__ieee754_jn): Likewise. * sysdeps/ieee754/dbl-64/e_sinh.c (__ieee754_sinh): Likewise. * sysdeps/ieee754/dbl-64/s_asinh.c (__asinh): Likewise. * sysdeps/ieee754/dbl-64/s_atan.c (atan): Use math_check_force_underflow_nonneg. * sysdeps/ieee754/dbl-64/s_erf.c (__erf): Use math_check_force_underflow. * sysdeps/ieee754/dbl-64/s_expm1.c (__expm1): Likewise. * sysdeps/ieee754/dbl-64/s_fma.c (__fma): Use math_force_eval instead of volatile. * sysdeps/ieee754/dbl-64/s_log1p.c (__log1p): Use math_check_force_underflow. * sysdeps/ieee754/dbl-64/s_sin.c (__sin): Likewise. * sysdeps/ieee754/dbl-64/s_tan.c (tan): Use math_check_force_underflow_nonneg. * sysdeps/ieee754/dbl-64/s_tanh.c (__tanh): Use math_check_force_underflow. * sysdeps/ieee754/flt-32/e_asinf.c (__ieee754_asinf): Likewise. * sysdeps/ieee754/flt-32/e_atanhf.c (__ieee754_atanhf): Likewise. * sysdeps/ieee754/flt-32/e_exp2f.c (__ieee754_exp2f): Use math_check_force_underflow_nonneg. * sysdeps/ieee754/flt-32/e_gammaf_r.c (__ieee754_gammaf_r): Likewise. * sysdeps/ieee754/flt-32/e_j1f.c (__ieee754_j1f): Use math_check_force_underflow. * sysdeps/ieee754/flt-32/e_jnf.c (__ieee754_jnf): Likewise. * sysdeps/ieee754/flt-32/e_sinhf.c (__ieee754_sinhf): Likewise. * sysdeps/ieee754/flt-32/k_sinf.c (__kernel_sinf): Likewise. * sysdeps/ieee754/flt-32/k_tanf.c (__kernel_tanf): Likewise. * sysdeps/ieee754/flt-32/s_asinhf.c (__asinhf): Likewise. * sysdeps/ieee754/flt-32/s_atanf.c (__atanf): Likewise. * sysdeps/ieee754/flt-32/s_erff.c (__erff): Likewise. * sysdeps/ieee754/flt-32/s_expm1f.c (__expm1f): Likewise. * sysdeps/ieee754/flt-32/s_log1pf.c (__log1pf): Likewise. * sysdeps/ieee754/flt-32/s_tanhf.c (__tanhf): Likewise. * sysdeps/ieee754/ldbl-128/e_asinl.c (__ieee754_asinl): Likewise. * sysdeps/ieee754/ldbl-128/e_atanhl.c (__ieee754_atanhl): Likewise. * sysdeps/ieee754/ldbl-128/e_expl.c (__ieee754_expl): Use math_check_force_underflow_nonneg. * sysdeps/ieee754/ldbl-128/e_gammal_r.c (__ieee754_gammal_r): Likewise. * sysdeps/ieee754/ldbl-128/e_j1l.c (__ieee754_j1l): Use math_check_force_underflow. * sysdeps/ieee754/ldbl-128/e_jnl.c (__ieee754_jnl): Likewise. * sysdeps/ieee754/ldbl-128/e_sinhl.c (__ieee754_sinhl): Likewise. * sysdeps/ieee754/ldbl-128/k_sincosl.c (__kernel_sincosl): Likewise. * sysdeps/ieee754/ldbl-128/k_sinl.c (__kernel_sinl): Likewise. * sysdeps/ieee754/ldbl-128/k_tanl.c (__kernel_tanl): Likewise. * sysdeps/ieee754/ldbl-128/s_asinhl.c (__asinhl): Likewise. * sysdeps/ieee754/ldbl-128/s_atanl.c (__atanl): Likewise. * sysdeps/ieee754/ldbl-128/s_erfl.c (__erfl): Likewise. * sysdeps/ieee754/ldbl-128/s_expm1l.c (__expm1l): Likewise. * sysdeps/ieee754/ldbl-128/s_fmal.c (__fmal): Use math_force_eval instead of volatile. * sysdeps/ieee754/ldbl-128/s_log1pl.c (__log1pl): Use math_check_force_underflow. * sysdeps/ieee754/ldbl-128/s_tanhl.c (__tanhl): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_asinl.c (__ieee754_asinl): Use math_check_force_underflow. * sysdeps/ieee754/ldbl-128ibm/e_atanhl.c (__ieee754_atanhl): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_gammal_r.c (__ieee754_gammal_r): Use math_check_force_underflow_nonneg. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c (__ieee754_jnl): Use math_check_force_underflow. * sysdeps/ieee754/ldbl-128ibm/e_sinhl.c (__ieee754_sinhl): Likewise. * sysdeps/ieee754/ldbl-128ibm/k_sincosl.c (__kernel_sincosl): Likewise. * sysdeps/ieee754/ldbl-128ibm/k_sinl.c (__kernel_sinl): Likewise. * sysdeps/ieee754/ldbl-128ibm/k_tanl.c (__kernel_tanl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_asinhl.c (__asinhl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_atanl.c (__atanl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_erfl.c (__erfl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_tanhl.c (__tanhl): Likewise. * sysdeps/ieee754/ldbl-96/e_asinl.c (__ieee754_asinl): Likewise. * sysdeps/ieee754/ldbl-96/e_atanhl.c (__ieee754_atanhl): Likewise. * sysdeps/ieee754/ldbl-96/e_gammal_r.c (__ieee754_gammal_r): Use math_check_force_underflow_nonneg. * sysdeps/ieee754/ldbl-96/e_j1l.c (__ieee754_j1l): Use math_check_force_underflow. * sysdeps/ieee754/ldbl-96/e_jnl.c (__ieee754_jnl): Likewise. * sysdeps/ieee754/ldbl-96/e_sinhl.c (__ieee754_sinhl): Likewise. * sysdeps/ieee754/ldbl-96/k_sinl.c (__kernel_sinl): Likewise. * sysdeps/ieee754/ldbl-96/k_tanl.c (__kernel_tanl): Use math_check_force_underflow_nonneg. * sysdeps/ieee754/ldbl-96/s_asinhl.c (__asinhl): Use math_check_force_underflow. * sysdeps/ieee754/ldbl-96/s_erfl.c (__erfl): Likewise. * sysdeps/ieee754/ldbl-96/s_fmal.c (__fmal): Use math_force_eval instead of volatile. * sysdeps/ieee754/ldbl-96/s_tanhl.c (__tanhl): Use math_check_force_underflow.
2015-09-23 22:42:30 +00:00
else
math_check_force_underflow (ret);
Use round-to-nearest internally in jn, test with ALL_RM_TEST (bug 18602). Some existing jn tests, if run in non-default rounding modes, produce errors above those accepted in glibc, which causes problems for moving tests of jn to use ALL_RM_TEST. This patch makes jn set rounding to-nearest internally, as was done for yn some time ago, then computes the appropriate underflowing value for results that underflowed to zero in to-nearest, and moves the tests to ALL_RM_TEST. It does nothing about the general inaccuracy of Bessel function implementations in glibc, though it should make jn more accurate on average in non-default rounding modes through reduced error accumulation. The recomputation of results that underflowed to zero should as a side-effect fix some cases of bug 16559, where jn just used an exact zero, but that is *not* the goal of this patch and other cases of that bug remain unfixed. (Most of the changes in the patch are reindentation to add new scopes for SET_RESTORE_ROUND*.) Tested for x86_64, x86, powerpc and mips64. [BZ #16559] [BZ #18602] * sysdeps/ieee754/dbl-64/e_jn.c (__ieee754_jn): Set round-to-nearest internally then recompute results that underflowed to zero in the original rounding mode. * sysdeps/ieee754/flt-32/e_jnf.c (__ieee754_jnf): Likewise. * sysdeps/ieee754/ldbl-128/e_jnl.c (__ieee754_jnl): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c (__ieee754_jnl): Likewise. * sysdeps/ieee754/ldbl-96/e_jnl.c (__ieee754_jnl): Likewise * math/libm-test.inc (jn_test): Use ALL_RM_TEST. * sysdeps/i386/fpu/libm-test-ulps: Update. * sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
2015-06-25 21:46:02 +00:00
return ret;
}
strong_alias (__ieee754_jn, __jn_finite)
double
2013-10-17 14:03:24 +00:00
__ieee754_yn (int n, double x)
{
2013-10-17 14:03:24 +00:00
int32_t i, hx, ix, lx;
int32_t sign;
Fix yn overflow handling in non-default rounding modes (bug 16561, bug 16562). This patch fixes bugs 16561 and 16562, bad results of yn in overflow cases in non-default rounding modes, both because an intermediate overflow in the recurrence does not get detected if the result is not an infinity and because an overflowing result may occur in the wrong sign. The fix is to set FE_TONEAREST mode internally for the parts of the function where such overflows can occur (which includes the call to y1 - where yn is used to compute a Bessel function of order -1, negating the result of y1 isn't correct for overflowing results in directed rounding modes) and then compute an overflowing value in the original rounding mode if the to-nearest result was an infinity. Tested x86_64 and x86 and ulps updated accordingly. Also tested for mips64 and powerpc32 to test the ldbl-128 and ldbl-128ibm changes. (The tests for these bugs were added in my previous y1 patch, so the only thing this patch has to do with the testsuite is enable yn testing in all rounding modes.) [BZ #16561] [BZ #16562] * sysdeps/ieee754/dbl-64/e_jn.c: Include <float.h>. (__ieee754_yn): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/flt-32/e_jnf.c: Include <float.h>. (__ieee754_ynf): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-128/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-96/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/i386/fpu/fenv_private.h [!__SSE2_MATH__] (libc_feholdsetround_ctx): New macro. * math/libm-test.inc (yn_test): Use ALL_RM_TEST. * sysdeps/i386/fpu/libm-test-ulps: Update. * sysdeps/x86_64/fpu/libm-test-ulps : Likewise.
2014-06-27 14:52:13 +00:00
double a, b, temp, ret;
2013-10-17 14:03:24 +00:00
EXTRACT_WORDS (hx, lx, x);
ix = 0x7fffffff & hx;
/* if Y(n,NaN) is NaN */
if (__glibc_unlikely ((ix | ((u_int32_t) (lx | -lx)) >> 31) > 0x7ff00000))
2013-10-17 14:03:24 +00:00
return x + x;
if (__glibc_unlikely ((ix | lx) == 0))
2013-10-17 14:03:24 +00:00
return -HUGE_VAL + x;
/* -inf and overflow exception. */;
if (__glibc_unlikely (hx < 0))
2013-10-17 14:03:24 +00:00
return zero / (zero * x);
sign = 1;
if (n < 0)
{
n = -n;
sign = 1 - ((n & 1) << 1);
}
if (n == 0)
return (__ieee754_y0 (x));
Fix yn overflow handling in non-default rounding modes (bug 16561, bug 16562). This patch fixes bugs 16561 and 16562, bad results of yn in overflow cases in non-default rounding modes, both because an intermediate overflow in the recurrence does not get detected if the result is not an infinity and because an overflowing result may occur in the wrong sign. The fix is to set FE_TONEAREST mode internally for the parts of the function where such overflows can occur (which includes the call to y1 - where yn is used to compute a Bessel function of order -1, negating the result of y1 isn't correct for overflowing results in directed rounding modes) and then compute an overflowing value in the original rounding mode if the to-nearest result was an infinity. Tested x86_64 and x86 and ulps updated accordingly. Also tested for mips64 and powerpc32 to test the ldbl-128 and ldbl-128ibm changes. (The tests for these bugs were added in my previous y1 patch, so the only thing this patch has to do with the testsuite is enable yn testing in all rounding modes.) [BZ #16561] [BZ #16562] * sysdeps/ieee754/dbl-64/e_jn.c: Include <float.h>. (__ieee754_yn): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/flt-32/e_jnf.c: Include <float.h>. (__ieee754_ynf): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-128/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-96/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/i386/fpu/fenv_private.h [!__SSE2_MATH__] (libc_feholdsetround_ctx): New macro. * math/libm-test.inc (yn_test): Use ALL_RM_TEST. * sysdeps/i386/fpu/libm-test-ulps: Update. * sysdeps/x86_64/fpu/libm-test-ulps : Likewise.
2014-06-27 14:52:13 +00:00
{
SET_RESTORE_ROUND (FE_TONEAREST);
if (n == 1)
{
ret = sign * __ieee754_y1 (x);
goto out;
}
if (__glibc_unlikely (ix == 0x7ff00000))
return zero;
if (ix >= 0x52D00000) /* x > 2**302 */
{ /* (x >> n**2)
* Jn(x) = cos(x-(2n+1)*pi/4)*sqrt(2/x*pi)
* Yn(x) = sin(x-(2n+1)*pi/4)*sqrt(2/x*pi)
* Let s=sin(x), c=cos(x),
* xn=x-(2n+1)*pi/4, sqt2 = sqrt(2),then
*
* n sin(xn)*sqt2 cos(xn)*sqt2
* ----------------------------------
* 0 s-c c+s
* 1 -s-c -c+s
* 2 -s+c -c-s
* 3 s+c c-s
*/
double c;
double s;
__sincos (x, &s, &c);
switch (n & 3)
{
case 0: temp = s - c; break;
case 1: temp = -s - c; break;
case 2: temp = -s + c; break;
case 3: temp = s + c; break;
}
b = invsqrtpi * temp / __ieee754_sqrt (x);
}
else
{
u_int32_t high;
a = __ieee754_y0 (x);
b = __ieee754_y1 (x);
/* quit if b is -inf */
GET_HIGH_WORD (high, b);
for (i = 1; i < n && high != 0xfff00000; i++)
{
temp = b;
b = ((double) (i + i) / x) * b - a;
GET_HIGH_WORD (high, b);
a = temp;
}
/* If B is +-Inf, set up errno accordingly. */
if (!isfinite (b))
Fix yn overflow handling in non-default rounding modes (bug 16561, bug 16562). This patch fixes bugs 16561 and 16562, bad results of yn in overflow cases in non-default rounding modes, both because an intermediate overflow in the recurrence does not get detected if the result is not an infinity and because an overflowing result may occur in the wrong sign. The fix is to set FE_TONEAREST mode internally for the parts of the function where such overflows can occur (which includes the call to y1 - where yn is used to compute a Bessel function of order -1, negating the result of y1 isn't correct for overflowing results in directed rounding modes) and then compute an overflowing value in the original rounding mode if the to-nearest result was an infinity. Tested x86_64 and x86 and ulps updated accordingly. Also tested for mips64 and powerpc32 to test the ldbl-128 and ldbl-128ibm changes. (The tests for these bugs were added in my previous y1 patch, so the only thing this patch has to do with the testsuite is enable yn testing in all rounding modes.) [BZ #16561] [BZ #16562] * sysdeps/ieee754/dbl-64/e_jn.c: Include <float.h>. (__ieee754_yn): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/flt-32/e_jnf.c: Include <float.h>. (__ieee754_ynf): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-128/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-96/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/i386/fpu/fenv_private.h [!__SSE2_MATH__] (libc_feholdsetround_ctx): New macro. * math/libm-test.inc (yn_test): Use ALL_RM_TEST. * sysdeps/i386/fpu/libm-test-ulps: Update. * sysdeps/x86_64/fpu/libm-test-ulps : Likewise.
2014-06-27 14:52:13 +00:00
__set_errno (ERANGE);
}
if (sign > 0)
ret = b;
else
ret = -b;
}
out:
if (isinf (ret))
Fix yn overflow handling in non-default rounding modes (bug 16561, bug 16562). This patch fixes bugs 16561 and 16562, bad results of yn in overflow cases in non-default rounding modes, both because an intermediate overflow in the recurrence does not get detected if the result is not an infinity and because an overflowing result may occur in the wrong sign. The fix is to set FE_TONEAREST mode internally for the parts of the function where such overflows can occur (which includes the call to y1 - where yn is used to compute a Bessel function of order -1, negating the result of y1 isn't correct for overflowing results in directed rounding modes) and then compute an overflowing value in the original rounding mode if the to-nearest result was an infinity. Tested x86_64 and x86 and ulps updated accordingly. Also tested for mips64 and powerpc32 to test the ldbl-128 and ldbl-128ibm changes. (The tests for these bugs were added in my previous y1 patch, so the only thing this patch has to do with the testsuite is enable yn testing in all rounding modes.) [BZ #16561] [BZ #16562] * sysdeps/ieee754/dbl-64/e_jn.c: Include <float.h>. (__ieee754_yn): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/flt-32/e_jnf.c: Include <float.h>. (__ieee754_ynf): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-128/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/ieee754/ldbl-96/e_jnl.c: Include <float.h>. (__ieee754_ynl): Set FE_TONEAREST mode internally and then recompute overflowing results in original rounding mode. * sysdeps/i386/fpu/fenv_private.h [!__SSE2_MATH__] (libc_feholdsetround_ctx): New macro. * math/libm-test.inc (yn_test): Use ALL_RM_TEST. * sysdeps/i386/fpu/libm-test-ulps: Update. * sysdeps/x86_64/fpu/libm-test-ulps : Likewise.
2014-06-27 14:52:13 +00:00
ret = __copysign (DBL_MAX, ret) * DBL_MAX;
return ret;
}
strong_alias (__ieee754_yn, __yn_finite)