bde 76448b8654 Oops, on amd64 (and probably on all non-i386 systems), the previous ...

commit broke the 2**24 cases where |x| > DBL_MAX/2.  There are exponent
range problems not just for denormals (underflow) but for large values
(overflow).  Doubles have more than enough exponent range to avoid the
problems, but I forgot to convert enough terms to double, so there was
an x+x term which was sometimes evaluated in float precision.

Unfortunately, this is a pessimization with some combinations of systems
and compilers (it makes no difference on Athlon XP's, but on Athlon64's
it gives a 5% pessimization with gcc-3.4 but not with gcc-3.3).

Exlain the problem better in comments.

2006-01-05 09:18:48 +00:00

..

e_acos.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_acosf.c

…

e_acosh.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_acoshf.c

…

e_asin.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_asinf.c

Fixed the approximation to pio4. pio4_hi must be pio2_hi/2 since it

2005-12-04 13:52:46 +00:00

e_atan2.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_atan2f.c

Fixed lots of 1 ULP errors caused by a broken approximation for pi/2.

2004-06-02 17:09:05 +00:00

e_atanh.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_atanhf.c

…

e_cosh.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_coshf.c

Fixed some magic numbers.

2005-11-13 00:08:23 +00:00

e_exp.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_expf.c

Fixed the hi+lo approximation to log(2). The normal 17+24 bit decomposition

2005-11-30 04:56:49 +00:00

e_fmod.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_fmodf.c

…

e_gamma_r.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_gamma.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_gammaf_r.c

…

e_gammaf.c

…

e_hypot.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_hypotf.c

…

e_j0.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_j0f.c

…

e_j1.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_j1f.c

…

e_jn.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_jnf.c

…

e_lgamma_r.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_lgamma.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_lgammaf_r.c

Fixed about 50 million errors of infinity ulps and about 3 million errors

2005-11-28 08:32:15 +00:00

e_lgammaf.c

…

e_log10.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_log10f.c

…

e_log.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_logf.c

As for the float trig functions, use a minimax polynomial that is

2005-11-12 18:20:09 +00:00

e_pow.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_powf.c

Merged from double precision case (e_pow.c 1.10: sign fixes).

2004-06-01 19:33:30 +00:00

e_rem_pio2.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_rem_pio2f.c

Simplified setiing up args for __kernel_rem_pio2(). We already have x

2005-11-23 03:03:09 +00:00

e_remainder.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_remainderf.c

…

e_scalb.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_scalbf.c

"Create" ldexpf for non-i386 architectures.

2005-12-06 20:12:38 +00:00

e_sinh.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_sinhf.c

Fixed some magic numbers.

2005-11-13 00:41:46 +00:00

e_sqrt.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

e_sqrtf.c

…

k_cos.c

Use a better algorithm for reducing the error in __kernel_cos[f]().

2005-10-26 12:36:18 +00:00

k_cosf.c

Rearranged the polynomial evaluation to reduce dependencies, as in

2005-11-30 11:51:17 +00:00

k_rem_pio2.c

Reduce diffs against vendor source (Sun fdlibm 5.3).

2005-02-04 18:26:06 +00:00

k_rem_pio2f.c

Fixed range reduction for large multiples of pi/2 on systems with

2005-10-11 07:56:05 +00:00

k_sin.c

Updated the comment about the optimization for tiny x (the previous

2005-11-02 13:06:49 +00:00

k_sinf.c

Rearranged the polynomial evaluation to reduce dependencies, as in

2005-11-30 11:51:17 +00:00

k_tan.c

Moved the optimization for tiny x from __kernel_tan[f](x) to tan[f](x)

2005-11-02 14:01:45 +00:00

k_tanf.c

Rearranged the polynomial evaluation some more to reduce dependencies.

2005-11-28 11:46:20 +00:00

math_private.h

Use only double precision for "kernel" cosf and sinf (except for

2005-11-28 04:58:57 +00:00

math.h

Implement truncl() based on floorl().

2005-04-16 21:12:47 +00:00

s_asinh.c

…

s_asinhf.c

…

s_atan.c

Only provide one copy of the math functions. If we provide a MD function,

2003-07-23 04:53:47 +00:00

s_atanf.c

…

s_cbrt.c

Extract the high and low words together. With gcc-3.4 on uniformly

2005-12-20 01:21:30 +00:00

s_cbrtf.c

Oops, on amd64 (and probably on all non-i386 systems), the previous

2006-01-05 09:18:48 +00:00

s_ceil.c

Only provide one copy of the math functions. If we provide a MD function,

2003-07-23 04:53:47 +00:00

s_ceilf.c

…

s_ceill.c

Revert the last change, the conversion from long double to double can raise

2005-04-28 19:45:55 +00:00

s_cimag.c

Add implementations for cimag{,f,l}, creal{,f,l} and conj{,f,l}. They are

2004-05-30 09:21:56 +00:00

s_cimagf.c

Add implementations for cimag{,f,l}, creal{,f,l} and conj{,f,l}. They are

2004-05-30 09:21:56 +00:00

s_cimagl.c

Add implementations for cimag{,f,l}, creal{,f,l} and conj{,f,l}. They are

2004-05-30 09:21:56 +00:00

s_conj.c

Add implementations for cimag{,f,l}, creal{,f,l} and conj{,f,l}. They are

2004-05-30 09:21:56 +00:00

s_conjf.c

Add implementations for cimag{,f,l}, creal{,f,l} and conj{,f,l}. They are

2004-05-30 09:21:56 +00:00

s_conjl.c

Add implementations for cimag{,f,l}, creal{,f,l} and conj{,f,l}. They are

2004-05-30 09:21:56 +00:00

s_copysign.c

Only provide one copy of the math functions. If we provide a MD function,

2003-07-23 04:53:47 +00:00

s_copysignf.c

Only provide one copy of the math functions. If we provide a MD function,

2003-07-23 04:53:47 +00:00

s_copysignl.c

Add an implementation of copysignl(), a long double version of copysign().

2004-05-07 18:56:31 +00:00

s_cos.c

Moved the optimization for tiny x from __kernel_{cos,sin}[f](x) to

2005-10-24 14:08:36 +00:00

s_cosf.c

Fixed cosf(x) when x is a "negative" NaNs. I broke this in rev.1.10.

2005-11-30 06:47:18 +00:00

s_creal.c

Add implementations for cimag{,f,l}, creal{,f,l} and conj{,f,l}. They are

2004-05-30 09:21:56 +00:00

s_crealf.c

Add implementations for cimag{,f,l}, creal{,f,l} and conj{,f,l}. They are

2004-05-30 09:21:56 +00:00

s_creall.c

Add implementations for cimag{,f,l}, creal{,f,l} and conj{,f,l}. They are

2004-05-30 09:21:56 +00:00

s_erf.c

…

s_erff.c

…

s_exp2.c

Implement exp2() and exp2f().

2005-04-05 02:57:15 +00:00

s_exp2f.c

Implement exp2() and exp2f().

2005-04-05 02:57:15 +00:00

s_expm1.c

…

s_expm1f.c

…

s_fabs.c

…

s_fabsf.c

…

s_fabsl.c

Better safe than clever.

2003-10-25 19:53:28 +00:00

s_fdim.c

Implement and document fdim{,f,l}, fmax{,f,l}, and fmin{,f,l}.

2004-06-30 07:04:01 +00:00

s_finite.c

Only provide one copy of the math functions. If we provide a MD function,

2003-07-23 04:53:47 +00:00

s_finitef.c

…

s_floor.c

Only provide one copy of the math functions. If we provide a MD function,

2003-07-23 04:53:47 +00:00

s_floorf.c

…

s_floorl.c

Revert the last change, the conversion from long double to double can raise

2005-04-28 19:45:55 +00:00

s_fma.c

Fix the double rounding problem with subnormals, and

2005-03-18 02:27:59 +00:00

s_fmaf.c

Add fma() and fmaf(), which implement a fused multiply-add operation.

2005-01-22 09:53:18 +00:00

s_fmal.c

Fix the double rounding problem with subnormals, and

2005-03-18 02:27:59 +00:00

s_fmax.c

Implement and document fdim{,f,l}, fmax{,f,l}, and fmin{,f,l}.

2004-06-30 07:04:01 +00:00

s_fmaxf.c

Implement and document fdim{,f,l}, fmax{,f,l}, and fmin{,f,l}.

2004-06-30 07:04:01 +00:00

s_fmaxl.c

Implement and document fdim{,f,l}, fmax{,f,l}, and fmin{,f,l}.

2004-06-30 07:04:01 +00:00

s_fmin.c

Implement and document fdim{,f,l}, fmax{,f,l}, and fmin{,f,l}.

2004-06-30 07:04:01 +00:00

s_fminf.c

Implement and document fdim{,f,l}, fmax{,f,l}, and fmin{,f,l}.

2004-06-30 07:04:01 +00:00

s_fminl.c

Implement and document fdim{,f,l}, fmax{,f,l}, and fmin{,f,l}.

2004-06-30 07:04:01 +00:00

s_frexp.c

Replace strong references with weak references. There's no

2005-03-07 21:27:37 +00:00

s_frexpf.c

…

s_frexpl.c

Implement frexpl.

2005-03-07 04:54:51 +00:00

s_ilogb.c

Use the FP_ILOG macros from <math.h> rather than hardcoded return values.

2004-10-09 17:14:28 +00:00

s_ilogbf.c

Use the FP_ILOG macros from <math.h> rather than hardcoded return values.

2004-10-09 17:14:28 +00:00

s_ilogbl.c

Add and document ilogbl(), a long double version of ilogb().

2004-10-11 18:13:52 +00:00

s_isfinite.c

Implement the classification macros isfinite(), isinf(), isnan(), and

2004-07-09 03:32:40 +00:00

s_isnan.c

Replace s_isnan.c and s_isnanf.c with the more compact s_isnan.c from

2004-08-05 01:46:11 +00:00

s_isnormal.c

Implement the classification macros isfinite(), isinf(), isnan(), and

2004-07-09 03:32:40 +00:00

s_llrint.c

Add MI implementations of [l]lrint[f]() and [l]lround[f]().

2005-01-11 23:12:55 +00:00

s_llrintf.c

Add MI implementations of [l]lrint[f]() and [l]lround[f]().

2005-01-11 23:12:55 +00:00

s_llround.c

These files should include s_lround.c instead of s_lrint.c.

2005-04-08 00:52:27 +00:00

s_llroundf.c

These files should include s_lround.c instead of s_lrint.c.

2005-04-08 00:52:27 +00:00

s_llroundl.c

Add roundl(), lroundl(), and llroundl().

2005-04-08 01:24:08 +00:00

s_log1p.c

For log1pf(), fixed the approximations to sqrt(2), sqrt(2)-1 and

2005-12-04 12:30:44 +00:00

s_log1pf.c

For log1pf(), fixed the approximations to sqrt(2), sqrt(2)-1 and

2005-12-04 12:30:44 +00:00

s_logb.c

Fixed fdlibm[+cygnus] logbf() and logb() on denormals. Adjustment

2005-12-03 11:57:19 +00:00

s_logbf.c

Fixed fdlibm[+cygnus] logbf() and logb() on denormals. Adjustment

2005-12-03 11:57:19 +00:00

s_lrint.c

Add MI implementations of [l]lrint[f]() and [l]lround[f]().

2005-01-11 23:12:55 +00:00

s_lrintf.c

Add MI implementations of [l]lrint[f]() and [l]lround[f]().

2005-01-11 23:12:55 +00:00

s_lround.c

Fix a (coincidentally harmless) bug.

2005-04-08 00:52:16 +00:00

s_lroundf.c

These files should include s_lround.c instead of s_lrint.c.

2005-04-08 00:52:27 +00:00

s_lroundl.c

Add roundl(), lroundl(), and llroundl().

2005-04-08 01:24:08 +00:00

s_modf.c

…

s_modff.c

…

s_nearbyint.c

Add C99's nearbyint{,f}() functions as wrappers around rint().

2004-07-06 04:46:08 +00:00

s_nextafter.c

Replace strong references with weak references. There's no

2005-03-07 21:27:37 +00:00

s_nextafterf.c

- Try harder to trick gcc into not optimizing away statements

2005-03-07 04:55:58 +00:00

s_nextafterl.c

Implement nexttoward and nextafterl; the latter is also known as

2005-03-07 04:56:46 +00:00

s_nexttoward.c

Implement nexttoward and nextafterl; the latter is also known as

2005-03-07 04:56:46 +00:00

s_nexttowardf.c

Implement nexttowardf. This is used on both platforms with 11-bit

2005-03-07 04:57:38 +00:00

s_remquo.c

Implement and document remquo() and remquof().

2005-03-25 04:40:44 +00:00

s_remquof.c

Implement and document remquo() and remquof().

2005-03-25 04:40:44 +00:00

s_rint.c

Simplified the fix in rev.1.3. Instead of using long double for

2005-12-03 07:38:35 +00:00

s_rintf.c

Restored removal of the special handling needed for a result of +-0.

2005-12-03 09:00:29 +00:00

s_round.c

Fixed roundf(). The following cases never worked in FreeBSD:

2005-12-02 13:45:06 +00:00

s_roundf.c

Fixed roundf(). The following cases never worked in FreeBSD:

2005-12-02 13:45:06 +00:00

s_roundl.c

Fixed roundf(). The following cases never worked in FreeBSD:

2005-12-02 13:45:06 +00:00

s_scalbln.c

- Implement scalblnl.

2005-03-07 04:57:50 +00:00

s_scalbn.c

Replace strong references with weak references. There's no

2005-03-07 21:27:37 +00:00

s_scalbnf.c

Alias scalbnf as ldexpf. The two are identical in binary

2005-03-07 04:52:43 +00:00

s_scalbnl.c

Add scalbnl, also known as as ldexpl.

2005-03-07 04:52:58 +00:00

s_signbit.c

Fix two bugs in the signbit() macro, which was implemented last year:

2004-07-19 08:16:10 +00:00

s_signgam.c

…

s_significand.c

Only provide one copy of the math functions. If we provide a MD function,

2003-07-23 04:53:47 +00:00

s_significandf.c

…

s_sin.c

Moved the optimization for tiny x from __kernel_{cos,sin}[f](x) to

2005-10-24 14:08:36 +00:00

s_sinf.c

Exploit skew-symmetry to avoid an operation: -sin(x-A) = sin(A-x). This

2005-11-28 06:15:10 +00:00

s_tan.c

Moved the optimization for tiny x from __kernel_tan[f](x) to tan[f](x)

2005-11-02 14:01:45 +00:00

s_tanf.c

Changed spelling of the request-to-inline macro name to match the change

2005-11-28 05:35:32 +00:00

s_tanh.c

…

s_tanhf.c

Fixed some magic numbers.

2005-12-11 11:40:55 +00:00

s_trunc.c

Implement trunc() and truncf().

2004-06-20 09:25:43 +00:00

s_truncf.c

Implement trunc() and truncf().

2004-06-20 09:25:43 +00:00

s_truncl.c

Revert the last change, the conversion from long double to double can raise

2005-04-28 19:45:55 +00:00

w_cabs.c

…

w_cabsf.c

…

w_drem.c

…

w_dremf.c

Work around known GCC 3.4.x problem and use ANSI prototype for dremf().

2004-07-28 05:53:18 +00:00