mirror of
https://sourceware.org/git/glibc.git
synced 2024-11-22 04:50:07 +00:00
New generic cosf
The same logic used in s_cosf.S version for x86 and powerpc is used to create a generic s_cosf.c, so there is no performance improvement in x86_64 and powerpc64. * sysdeps/ieee754/flt-32/s_cosf.c: New implementation.
This commit is contained in:
Paul Clarke
Tulio Magno Quites Machado Filho
parent
c9cd7b0ce5
commit
f4b2aea6e1
@ -1,3 +1,7 @@
|
|||||||
|
2017-12-11 Paul A. Clarke <pc@us.ibm.com>
|
||||||
|
|
||||||
|
* sysdeps/ieee754/flt-32/s_cosf.c: New implementation.
|
||||||
|
|
||||||
2017-12-11 Adhemerval Zanella <azanella@linux.vnet.ibm.com>
|
2017-12-11 Adhemerval Zanella <azanella@linux.vnet.ibm.com>
|
||||||
Tulio Magno Quites Machado Filho <tuliom@linux.vnet.ibm.com>
|
Tulio Magno Quites Machado Filho <tuliom@linux.vnet.ibm.com>
|
||||||
|
|
||||||
|
|||||||
@ -1,21 +1,20 @@
|
|||||||
/* s_cosf.c -- float version of s_cos.c.
|
/* Compute cosine of argument.
|
||||||
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
|
Copyright (C) 2017 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
|
||||||
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
|
License as published by the Free Software Foundation; either
|
||||||
*
|
version 2.1 of the License, or (at your option) any later version.
|
||||||
* 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.
|
|
||||||
* ====================================================
|
|
||||||
*/
|
|
||||||
|
|
||||||
#if defined(LIBM_SCCS) && !defined(lint)
|
The GNU C Library is distributed in the hope that it will be useful,
|
||||||
static char rcsid[] = "$NetBSD: s_cosf.c,v 1.4 1995/05/10 20:47:03 jtc Exp $";
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||||
#endif
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
||||||
|
Lesser General Public License for more details.
|
||||||
|
|
||||||
|
You should have received a copy of the GNU Lesser General Public
|
||||||
|
License along with the GNU C Library; if not, see
|
||||||
|
<http://www.gnu.org/licenses/>. */
|
||||||
|
|
||||||
#include <errno.h>
|
#include <errno.h>
|
||||||
#include <math.h>
|
#include <math.h>
|
||||||
@ -28,35 +27,210 @@ static char rcsid[] = "$NetBSD: s_cosf.c,v 1.4 1995/05/10 20:47:03 jtc Exp $";
|
|||||||
# define COSF_FUNC COSF
|
# define COSF_FUNC COSF
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
float COSF_FUNC(float x)
|
/* Chebyshev constants for cos, range -PI/4 - PI/4. */
|
||||||
|
static const double C0 = -0x1.ffffffffe98aep-2;
|
||||||
|
static const double C1 = 0x1.55555545c50c7p-5;
|
||||||
|
static const double C2 = -0x1.6c16b348b6874p-10;
|
||||||
|
static const double C3 = 0x1.a00eb9ac43ccp-16;
|
||||||
|
static const double C4 = -0x1.23c97dd8844d7p-22;
|
||||||
|
|
||||||
|
/* Chebyshev constants for sin, range -PI/4 - PI/4. */
|
||||||
|
static const double S0 = -0x1.5555555551cd9p-3;
|
||||||
|
static const double S1 = 0x1.1111110c2688bp-7;
|
||||||
|
static const double S2 = -0x1.a019f8b4bd1f9p-13;
|
||||||
|
static const double S3 = 0x1.71d7264e6b5b4p-19;
|
||||||
|
static const double S4 = -0x1.a947e1674b58ap-26;
|
||||||
|
|
||||||
|
/* Chebyshev constants for cos, range 2^-27 - 2^-5. */
|
||||||
|
static const double CC0 = -0x1.fffffff5cc6fdp-2;
|
||||||
|
static const double CC1 = 0x1.55514b178dac5p-5;
|
||||||
|
|
||||||
|
/* PI/2 with 98 bits of accuracy. */
|
||||||
|
static const double PI_2_hi = 0x1.921fb544p+0;
|
||||||
|
static const double PI_2_lo = 0x1.0b4611a626332p-34;
|
||||||
|
|
||||||
|
static const double inv_PI_4 = 0x1.45f306dc9c883p+0; /* 4/PI. */
|
||||||
|
|
||||||
|
#define FLOAT_EXPONENT_SHIFT 23
|
||||||
|
#define FLOAT_EXPONENT_BIAS 127
|
||||||
|
|
||||||
|
static const double pio2_table[] = {
|
||||||
|
0 * M_PI_2,
|
||||||
|
1 * M_PI_2,
|
||||||
|
2 * M_PI_2,
|
||||||
|
3 * M_PI_2,
|
||||||
|
4 * M_PI_2,
|
||||||
|
5 * M_PI_2
|
||||||
|
};
|
||||||
|
|
||||||
|
static const double invpio4_table[] = {
|
||||||
|
0x0p+0,
|
||||||
|
0x1.45f306cp+0,
|
||||||
|
0x1.c9c882ap-28,
|
||||||
|
0x1.4fe13a8p-58,
|
||||||
|
0x1.f47d4dp-85,
|
||||||
|
0x1.bb81b6cp-112,
|
||||||
|
0x1.4acc9ep-142,
|
||||||
|
0x1.0e4107cp-169
|
||||||
|
};
|
||||||
|
|
||||||
|
static const double ones[] = { 1.0, -1.0 };
|
||||||
|
|
||||||
|
/* Compute the cosine value using Chebyshev polynomials where
|
||||||
|
THETA is the range reduced absolute value of the input
|
||||||
|
and it is less than Pi/4,
|
||||||
|
N is calculated as trunc(|x|/(Pi/4)) + 1 and it is used to decide
|
||||||
|
whether a sine or cosine approximation is more accurate and
|
||||||
|
the sign of the result. */
|
||||||
|
static inline float
|
||||||
|
reduced (double theta, unsigned int n)
|
||||||
{
|
{
|
||||||
float y[2],z=0.0;
|
double sign, cx;
|
||||||
int32_t n,ix;
|
const double theta2 = theta * theta;
|
||||||
|
|
||||||
GET_FLOAT_WORD(ix,x);
|
/* Determine positive or negative primary interval. */
|
||||||
|
n += 2;
|
||||||
|
sign = ones[(n >> 2) & 1];
|
||||||
|
|
||||||
/* |x| ~< pi/4 */
|
/* Are we in the primary interval of sin or cos? */
|
||||||
ix &= 0x7fffffff;
|
if ((n & 2) == 0)
|
||||||
if(ix <= 0x3f490fd8) return __kernel_cosf(x,z);
|
{
|
||||||
|
/* Here cosf() is calculated using sin Chebyshev polynomial:
|
||||||
|
x+x^3*(S0+x^2*(S1+x^2*(S2+x^2*(S3+x^2*S4)))). */
|
||||||
|
cx = S3 + theta2 * S4;
|
||||||
|
cx = S2 + theta2 * cx;
|
||||||
|
cx = S1 + theta2 * cx;
|
||||||
|
cx = S0 + theta2 * cx;
|
||||||
|
cx = theta + theta * theta2 * cx;
|
||||||
|
}
|
||||||
|
else
|
||||||
|
{
|
||||||
|
/* Here cosf() is calculated using cos Chebyshev polynomial:
|
||||||
|
1.0+x^2*(C0+x^2*(C1+x^2*(C2+x^2*(C3+x^2*C4)))). */
|
||||||
|
cx = C3 + theta2 * C4;
|
||||||
|
cx = C2 + theta2 * cx;
|
||||||
|
cx = C1 + theta2 * cx;
|
||||||
|
cx = C0 + theta2 * cx;
|
||||||
|
cx = 1. + theta2 * cx;
|
||||||
|
}
|
||||||
|
return sign * cx;
|
||||||
|
}
|
||||||
|
|
||||||
/* cos(Inf or NaN) is NaN */
|
float
|
||||||
else if (ix>=0x7f800000) {
|
COSF_FUNC (float x)
|
||||||
if (ix == 0x7f800000)
|
{
|
||||||
__set_errno (EDOM);
|
double theta = x;
|
||||||
return x-x;
|
double abstheta = fabs (theta);
|
||||||
|
if (isless (abstheta, M_PI_4))
|
||||||
|
{
|
||||||
|
double cx;
|
||||||
|
if (abstheta >= 0x1p-5)
|
||||||
|
{
|
||||||
|
const double theta2 = theta * theta;
|
||||||
|
/* Chebyshev polynomial of the form for cos:
|
||||||
|
* 1 + x^2 (C0 + x^2 (C1 + x^2 (C2 + x^2 (C3 + x^2 * C4)))). */
|
||||||
|
cx = C3 + theta2 * C4;
|
||||||
|
cx = C2 + theta2 * cx;
|
||||||
|
cx = C1 + theta2 * cx;
|
||||||
|
cx = C0 + theta2 * cx;
|
||||||
|
cx = 1. + theta2 * cx;
|
||||||
|
return cx;
|
||||||
}
|
}
|
||||||
|
else if (abstheta >= 0x1p-27)
|
||||||
/* argument reduction needed */
|
{
|
||||||
else {
|
/* A simpler Chebyshev approximation is close enough for this range:
|
||||||
n = __ieee754_rem_pio2f(x,y);
|
* 1 + x^2 (CC0 + x^3 * CC1). */
|
||||||
switch(n&3) {
|
const double theta2 = theta * theta;
|
||||||
case 0: return __kernel_cosf(y[0],y[1]);
|
cx = CC0 + theta * theta2 * CC1;
|
||||||
case 1: return -__kernel_sinf(y[0],y[1],1);
|
cx = 1.0 + theta2 * cx;
|
||||||
case 2: return -__kernel_cosf(y[0],y[1]);
|
return cx;
|
||||||
default:
|
}
|
||||||
return __kernel_sinf(y[0],y[1],1);
|
else
|
||||||
|
{
|
||||||
|
/* For small enough |theta|, this is close enough. */
|
||||||
|
return 1.0 - abstheta;
|
||||||
|
}
|
||||||
|
}
|
||||||
|
else /* |theta| >= Pi/4. */
|
||||||
|
{
|
||||||
|
if (isless (abstheta, 9 * M_PI_4))
|
||||||
|
{
|
||||||
|
/* There are cases where FE_UPWARD rounding mode can
|
||||||
|
produce a result of abstheta * inv_PI_4 == 9,
|
||||||
|
where abstheta < 9pi/4, so the domain for
|
||||||
|
pio2_table must go to 5 (9 / 2 + 1). */
|
||||||
|
unsigned int n = (abstheta * inv_PI_4) + 1;
|
||||||
|
theta = abstheta - pio2_table[n / 2];
|
||||||
|
return reduced (theta, n);
|
||||||
|
}
|
||||||
|
else if (isless (abstheta, INFINITY))
|
||||||
|
{
|
||||||
|
if (abstheta < 0x1p+23)
|
||||||
|
{
|
||||||
|
unsigned int n = ((unsigned int) (abstheta * inv_PI_4)) + 1;
|
||||||
|
double x = n / 2;
|
||||||
|
theta = (abstheta - x * PI_2_hi) - x * PI_2_lo;
|
||||||
|
/* Argument reduction needed. */
|
||||||
|
return reduced (theta, n);
|
||||||
|
}
|
||||||
|
else /* |theta| >= 2^23. */
|
||||||
|
{
|
||||||
|
x = fabsf (x);
|
||||||
|
int exponent;
|
||||||
|
GET_FLOAT_WORD (exponent, x);
|
||||||
|
exponent = (exponent >> FLOAT_EXPONENT_SHIFT)
|
||||||
|
- FLOAT_EXPONENT_BIAS;
|
||||||
|
exponent += 3;
|
||||||
|
exponent /= 28;
|
||||||
|
double a = invpio4_table[exponent] * x;
|
||||||
|
double b = invpio4_table[exponent + 1] * x;
|
||||||
|
double c = invpio4_table[exponent + 2] * x;
|
||||||
|
double d = invpio4_table[exponent + 3] * x;
|
||||||
|
uint64_t l = a;
|
||||||
|
l &= ~0x7;
|
||||||
|
a -= l;
|
||||||
|
double e = a + b;
|
||||||
|
l = e;
|
||||||
|
e = a - l;
|
||||||
|
if (l & 1)
|
||||||
|
{
|
||||||
|
e -= 1.0;
|
||||||
|
e += b;
|
||||||
|
e += c;
|
||||||
|
e += d;
|
||||||
|
e *= M_PI_4;
|
||||||
|
return reduced (e, l + 1);
|
||||||
|
}
|
||||||
|
else
|
||||||
|
{
|
||||||
|
e += b;
|
||||||
|
e += c;
|
||||||
|
e += d;
|
||||||
|
if (e <= 1.0)
|
||||||
|
{
|
||||||
|
e *= M_PI_4;
|
||||||
|
return reduced (e, l + 1);
|
||||||
|
}
|
||||||
|
else
|
||||||
|
{
|
||||||
|
l++;
|
||||||
|
e -= 2.0;
|
||||||
|
e *= M_PI_4;
|
||||||
|
return reduced (e, l + 1);
|
||||||
|
}
|
||||||
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
else
|
||||||
|
{
|
||||||
|
int32_t ix;
|
||||||
|
GET_FLOAT_WORD (ix, abstheta);
|
||||||
|
/* cos(Inf or NaN) is NaN. */
|
||||||
|
if (ix == 0x7f800000) /* Inf. */
|
||||||
|
__set_errno (EDOM);
|
||||||
|
return x - x;
|
||||||
|
}
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
#ifndef COSF
|
#ifndef COSF
|
||||||
|
|||||||
Loading…
Reference in New Issue
Block a user