Home | History | Annotate | Download | only in src
      1 /*-
      2  * Copyright (c) 2011 David Schultz <das (at) FreeBSD.ORG>
      3  * All rights reserved.
      4  *
      5  * Redistribution and use in source and binary forms, with or without
      6  * modification, are permitted provided that the following conditions
      7  * are met:
      8  * 1. Redistributions of source code must retain the above copyright
      9  *    notice, this list of conditions and the following disclaimer.
     10  * 2. Redistributions in binary form must reproduce the above copyright
     11  *    notice, this list of conditions and the following disclaimer in the
     12  *    documentation and/or other materials provided with the distribution.
     13  *
     14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
     15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     24  * SUCH DAMAGE.
     25  */
     26 
     27 #include <sys/cdefs.h>
     28 __FBSDID("$FreeBSD$");
     29 
     30 #include <complex.h>
     31 
     32 #include "math.h"
     33 #include "math_private.h"
     34 
     35 static const uint32_t k = 1799;		/* constant for reduction */
     36 static const double kln2 =  1246.97177782734161156;	/* k * ln2 */
     37 
     38 /*
     39  * Compute exp(x), scaled to avoid spurious overflow.  An exponent is
     40  * returned separately in 'expt'.
     41  *
     42  * Input:  ln(DBL_MAX) <= x < ln(2 * DBL_MAX / DBL_MIN_DENORM) ~= 1454.91
     43  * Output: 2**1023 <= y < 2**1024
     44  */
     45 static double
     46 __frexp_exp(double x, int *expt)
     47 {
     48 	double exp_x;
     49 	uint32_t hx;
     50 
     51 	/*
     52 	 * We use exp(x) = exp(x - kln2) * 2**k, carefully chosen to
     53 	 * minimize |exp(kln2) - 2**k|.  We also scale the exponent of
     54 	 * exp_x to MAX_EXP so that the result can be multiplied by
     55 	 * a tiny number without losing accuracy due to denormalization.
     56 	 */
     57 	exp_x = exp(x - kln2);
     58 	GET_HIGH_WORD(hx, exp_x);
     59 	*expt = (hx >> 20) - (0x3ff + 1023) + k;
     60 	SET_HIGH_WORD(exp_x, (hx & 0xfffff) | ((0x3ff + 1023) << 20));
     61 	return (exp_x);
     62 }
     63 
     64 /*
     65  * __ldexp_exp(x, expt) and __ldexp_cexp(x, expt) compute exp(x) * 2**expt.
     66  * They are intended for large arguments (real part >= ln(DBL_MAX))
     67  * where care is needed to avoid overflow.
     68  *
     69  * The present implementation is narrowly tailored for our hyperbolic and
     70  * exponential functions.  We assume expt is small (0 or -1), and the caller
     71  * has filtered out very large x, for which overflow would be inevitable.
     72  */
     73 
     74 double
     75 __ldexp_exp(double x, int expt)
     76 {
     77 	double exp_x, scale;
     78 	int ex_expt;
     79 
     80 	exp_x = __frexp_exp(x, &ex_expt);
     81 	expt += ex_expt;
     82 	INSERT_WORDS(scale, (0x3ff + expt) << 20, 0);
     83 	return (exp_x * scale);
     84 }
     85 
     86 double complex
     87 __ldexp_cexp(double complex z, int expt)
     88 {
     89 	double x, y, exp_x, scale1, scale2;
     90 	int ex_expt, half_expt;
     91 
     92 	x = creal(z);
     93 	y = cimag(z);
     94 	exp_x = __frexp_exp(x, &ex_expt);
     95 	expt += ex_expt;
     96 
     97 	/*
     98 	 * Arrange so that scale1 * scale2 == 2**expt.  We use this to
     99 	 * compensate for scalbn being horrendously slow.
    100 	 */
    101 	half_expt = expt / 2;
    102 	INSERT_WORDS(scale1, (0x3ff + half_expt) << 20, 0);
    103 	half_expt = expt - half_expt;
    104 	INSERT_WORDS(scale2, (0x3ff + half_expt) << 20, 0);
    105 
    106 	return (cpack(cos(y) * exp_x * scale1 * scale2,
    107 	    sin(y) * exp_x * scale1 * scale2));
    108 }
    109