glibc/math/k_casinh_template.c
Szabolcs Nagy 347a5b592c math: Fix float conversion regressions with gcc-12 [BZ #28713]
Converting double precision constants to float is now affected by the
runtime dynamic rounding mode instead of being evaluated at compile
time with default rounding mode (except static object initializers).

This can change the computed result and cause performance regression.
The known correctness issues (increased ulp errors) are already fixed,
this patch fixes remaining cases of unnecessary runtime conversions.

Add float M_* macros to math.h as new GNU extension API.  To avoid
conversions the new M_* macros are used and instead of casting double
literals to float, use float literals (only required if the conversion
is inexact).

The patch was tested on aarch64 where the following symbols had new
spurious conversion instructions that got fixed:

  __clog10f
  __gammaf_r_finite@GLIBC_2.17
  __j0f_finite@GLIBC_2.17
  __j1f_finite@GLIBC_2.17
  __jnf_finite@GLIBC_2.17
  __kernel_casinhf
  __lgamma_negf
  __log1pf
  __y0f_finite@GLIBC_2.17
  __y1f_finite@GLIBC_2.17
  cacosf
  cacoshf
  casinhf
  catanf
  catanhf
  clogf
  gammaf_positive

Fixes bug 28713.

Reviewed-by: Paul Zimmermann <Paul.Zimmermann@inria.fr>
2022-01-10 14:27:17 +00:00

207 lines
5.5 KiB
C

/* Return arc hyperbolic sine for a complex float type, with the
imaginary part of the result possibly adjusted for use in
computing other functions.
Copyright (C) 1997-2022 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<https://www.gnu.org/licenses/>. */
#include <complex.h>
#include <math.h>
#include <math_private.h>
#include <math-underflow.h>
#include <float.h>
/* Return the complex inverse hyperbolic sine of finite nonzero Z,
with the imaginary part of the result subtracted from pi/2 if ADJ
is nonzero. */
CFLOAT
M_DECL_FUNC (__kernel_casinh) (CFLOAT x, int adj)
{
CFLOAT res;
FLOAT rx, ix;
CFLOAT y;
/* Avoid cancellation by reducing to the first quadrant. */
rx = M_FABS (__real__ x);
ix = M_FABS (__imag__ x);
if (rx >= 1 / M_EPSILON || ix >= 1 / M_EPSILON)
{
/* For large x in the first quadrant, x + csqrt (1 + x * x)
is sufficiently close to 2 * x to make no significant
difference to the result; avoid possible overflow from
the squaring and addition. */
__real__ y = rx;
__imag__ y = ix;
if (adj)
{
FLOAT t = __real__ y;
__real__ y = M_COPYSIGN (__imag__ y, __imag__ x);
__imag__ y = t;
}
res = M_SUF (__clog) (y);
__real__ res += M_MLIT (M_LN2);
}
else if (rx >= M_LIT (0.5) && ix < M_EPSILON / 8)
{
FLOAT s = M_HYPOT (1, rx);
__real__ res = M_LOG (rx + s);
if (adj)
__imag__ res = M_ATAN2 (s, __imag__ x);
else
__imag__ res = M_ATAN2 (ix, s);
}
else if (rx < M_EPSILON / 8 && ix >= M_LIT (1.5))
{
FLOAT s = M_SQRT ((ix + 1) * (ix - 1));
__real__ res = M_LOG (ix + s);
if (adj)
__imag__ res = M_ATAN2 (rx, M_COPYSIGN (s, __imag__ x));
else
__imag__ res = M_ATAN2 (s, rx);
}
else if (ix > 1 && ix < M_LIT (1.5) && rx < M_LIT (0.5))
{
if (rx < M_EPSILON * M_EPSILON)
{
FLOAT ix2m1 = (ix + 1) * (ix - 1);
FLOAT s = M_SQRT (ix2m1);
__real__ res = M_LOG1P (2 * (ix2m1 + ix * s)) / 2;
if (adj)
__imag__ res = M_ATAN2 (rx, M_COPYSIGN (s, __imag__ x));
else
__imag__ res = M_ATAN2 (s, rx);
}
else
{
FLOAT ix2m1 = (ix + 1) * (ix - 1);
FLOAT rx2 = rx * rx;
FLOAT f = rx2 * (2 + rx2 + 2 * ix * ix);
FLOAT d = M_SQRT (ix2m1 * ix2m1 + f);
FLOAT dp = d + ix2m1;
FLOAT dm = f / dp;
FLOAT r1 = M_SQRT ((dm + rx2) / 2);
FLOAT r2 = rx * ix / r1;
__real__ res = M_LOG1P (rx2 + dp + 2 * (rx * r1 + ix * r2)) / 2;
if (adj)
__imag__ res = M_ATAN2 (rx + r1, M_COPYSIGN (ix + r2, __imag__ x));
else
__imag__ res = M_ATAN2 (ix + r2, rx + r1);
}
}
else if (ix == 1 && rx < M_LIT (0.5))
{
if (rx < M_EPSILON / 8)
{
__real__ res = M_LOG1P (2 * (rx + M_SQRT (rx))) / 2;
if (adj)
__imag__ res = M_ATAN2 (M_SQRT (rx), M_COPYSIGN (1, __imag__ x));
else
__imag__ res = M_ATAN2 (1, M_SQRT (rx));
}
else
{
FLOAT d = rx * M_SQRT (4 + rx * rx);
FLOAT s1 = M_SQRT ((d + rx * rx) / 2);
FLOAT s2 = M_SQRT ((d - rx * rx) / 2);
__real__ res = M_LOG1P (rx * rx + d + 2 * (rx * s1 + s2)) / 2;
if (adj)
__imag__ res = M_ATAN2 (rx + s1, M_COPYSIGN (1 + s2, __imag__ x));
else
__imag__ res = M_ATAN2 (1 + s2, rx + s1);
}
}
else if (ix < 1 && rx < M_LIT (0.5))
{
if (ix >= M_EPSILON)
{
if (rx < M_EPSILON * M_EPSILON)
{
FLOAT onemix2 = (1 + ix) * (1 - ix);
FLOAT s = M_SQRT (onemix2);
__real__ res = M_LOG1P (2 * rx / s) / 2;
if (adj)
__imag__ res = M_ATAN2 (s, __imag__ x);
else
__imag__ res = M_ATAN2 (ix, s);
}
else
{
FLOAT onemix2 = (1 + ix) * (1 - ix);
FLOAT rx2 = rx * rx;
FLOAT f = rx2 * (2 + rx2 + 2 * ix * ix);
FLOAT d = M_SQRT (onemix2 * onemix2 + f);
FLOAT dp = d + onemix2;
FLOAT dm = f / dp;
FLOAT r1 = M_SQRT ((dp + rx2) / 2);
FLOAT r2 = rx * ix / r1;
__real__ res = M_LOG1P (rx2 + dm + 2 * (rx * r1 + ix * r2)) / 2;
if (adj)
__imag__ res = M_ATAN2 (rx + r1, M_COPYSIGN (ix + r2,
__imag__ x));
else
__imag__ res = M_ATAN2 (ix + r2, rx + r1);
}
}
else
{
FLOAT s = M_HYPOT (1, rx);
__real__ res = M_LOG1P (2 * rx * (rx + s)) / 2;
if (adj)
__imag__ res = M_ATAN2 (s, __imag__ x);
else
__imag__ res = M_ATAN2 (ix, s);
}
math_check_force_underflow_nonneg (__real__ res);
}
else
{
__real__ y = (rx - ix) * (rx + ix) + 1;
__imag__ y = 2 * rx * ix;
y = M_SUF (__csqrt) (y);
__real__ y += rx;
__imag__ y += ix;
if (adj)
{
FLOAT t = __real__ y;
__real__ y = M_COPYSIGN (__imag__ y, __imag__ x);
__imag__ y = t;
}
res = M_SUF (__clog) (y);
}
/* Give results the correct sign for the original argument. */
__real__ res = M_COPYSIGN (__real__ res, __real__ x);
__imag__ res = M_COPYSIGN (__imag__ res, (adj ? 1 : __imag__ x));
return res;
}