| 1 |
207 |
jeremybenn |
/* -------------------------------------------------------------- */
|
| 2 |
|
|
/* (C)Copyright 2006,2008, */
|
| 3 |
|
|
/* International Business Machines Corporation */
|
| 4 |
|
|
/* All Rights Reserved. */
|
| 5 |
|
|
/* */
|
| 6 |
|
|
/* Redistribution and use in source and binary forms, with or */
|
| 7 |
|
|
/* without modification, are permitted provided that the */
|
| 8 |
|
|
/* following conditions are met: */
|
| 9 |
|
|
/* */
|
| 10 |
|
|
/* - Redistributions of source code must retain the above copyright*/
|
| 11 |
|
|
/* notice, this list of conditions and the following disclaimer. */
|
| 12 |
|
|
/* */
|
| 13 |
|
|
/* - Redistributions in binary form must reproduce the above */
|
| 14 |
|
|
/* copyright notice, this list of conditions and the following */
|
| 15 |
|
|
/* disclaimer in the documentation and/or other materials */
|
| 16 |
|
|
/* provided with the distribution. */
|
| 17 |
|
|
/* */
|
| 18 |
|
|
/* - Neither the name of IBM Corporation nor the names of its */
|
| 19 |
|
|
/* contributors may be used to endorse or promote products */
|
| 20 |
|
|
/* derived from this software without specific prior written */
|
| 21 |
|
|
/* permission. */
|
| 22 |
|
|
/* */
|
| 23 |
|
|
/* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND */
|
| 24 |
|
|
/* CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, */
|
| 25 |
|
|
/* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */
|
| 26 |
|
|
/* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */
|
| 27 |
|
|
/* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR */
|
| 28 |
|
|
/* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, */
|
| 29 |
|
|
/* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT */
|
| 30 |
|
|
/* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; */
|
| 31 |
|
|
/* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) */
|
| 32 |
|
|
/* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN */
|
| 33 |
|
|
/* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR */
|
| 34 |
|
|
/* OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, */
|
| 35 |
|
|
/* EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
|
| 36 |
|
|
/* -------------------------------------------------------------- */
|
| 37 |
|
|
/* PROLOG END TAG zYx */
|
| 38 |
|
|
|
| 39 |
|
|
#ifdef __SPU__
|
| 40 |
|
|
|
| 41 |
|
|
#ifndef _ASIND2_H_
|
| 42 |
|
|
#define _ASIND2_H_ 1
|
| 43 |
|
|
|
| 44 |
|
|
#include "simdmath.h"
|
| 45 |
|
|
#include <spu_intrinsics.h>
|
| 46 |
|
|
#include "sqrtd2.h"
|
| 47 |
|
|
#include "divd2.h"
|
| 48 |
|
|
|
| 49 |
|
|
|
| 50 |
|
|
|
| 51 |
|
|
/*
|
| 52 |
|
|
* FUNCTION
|
| 53 |
|
|
* vector double _asind2(vector double x)
|
| 54 |
|
|
*
|
| 55 |
|
|
* DESCRIPTION
|
| 56 |
|
|
* Compute the arc sine of the vector of double precision elements
|
| 57 |
|
|
* specified by x, returning the resulting angles in radians. The input
|
| 58 |
|
|
* elements are to be in the closed interval [-1, 1]. Values outside
|
| 59 |
|
|
* this range result in a invalid operation execption being latched in
|
| 60 |
|
|
* the FPSCR register and a NAN is returned.
|
| 61 |
|
|
*
|
| 62 |
|
|
* The basic algorithm computes the arc sine using a rational polynomial
|
| 63 |
|
|
* of the form x + x^3 * P(x^2) / Q(x^2) for inputs |x| in the interval
|
| 64 |
|
|
* [0, 0.5]. Values outsize this range are transformed as by:
|
| 65 |
|
|
*
|
| 66 |
|
|
* asin(x) = PI/2 - 2*asin(sqrt((1-x)/2)) for x in the range (0.5, 1.0]
|
| 67 |
|
|
*
|
| 68 |
|
|
* asin(x) = -PI/2 + 2*asin(sqrt((1+x)/2)) for x in the range [-1.0, -0.5)
|
| 69 |
|
|
*
|
| 70 |
|
|
* This yields the basic algorithm of:
|
| 71 |
|
|
*
|
| 72 |
|
|
* absx = (x < 0.0) ? -x : x;
|
| 73 |
|
|
*
|
| 74 |
|
|
* if (absx > 0.5) {
|
| 75 |
|
|
* if (x < 0) {
|
| 76 |
|
|
* addend = -SM_PI_2;
|
| 77 |
|
|
* multiplier = -2.0;
|
| 78 |
|
|
* } else {
|
| 79 |
|
|
* addend = SM_PI_2;
|
| 80 |
|
|
* multiplier = 2.0;
|
| 81 |
|
|
* }
|
| 82 |
|
|
*
|
| 83 |
|
|
* x = sqrt(-0.5 * absx + 0.5);
|
| 84 |
|
|
* } else {
|
| 85 |
|
|
* addend = 0.0;
|
| 86 |
|
|
* multiplier = 1.0;
|
| 87 |
|
|
* }
|
| 88 |
|
|
*
|
| 89 |
|
|
* x2 = x * x;
|
| 90 |
|
|
* x3 = x2 * x;
|
| 91 |
|
|
*
|
| 92 |
|
|
* p = ((((P5 * x2 + P4)*x2 + P3)*x2 + P2)*x2 + P1)*x2 + P0;
|
| 93 |
|
|
*
|
| 94 |
|
|
* q = ((((Q5 * x2 + Q4)*x2 + Q3)*x2 + Q2)*x2 + Q1)*x2 + Q0;;
|
| 95 |
|
|
*
|
| 96 |
|
|
* pq = p / q;
|
| 97 |
|
|
*
|
| 98 |
|
|
* result = addend - (x3*pq + x)*multiplier;
|
| 99 |
|
|
*
|
| 100 |
|
|
* Where P5-P0 and Q5-Q0 are the polynomial coeficients.
|
| 101 |
|
|
*/
|
| 102 |
|
|
static __inline vector double _asind2(vector double x)
|
| 103 |
|
|
{
|
| 104 |
|
|
vec_uint4 x_gt_half, x_eq_half;
|
| 105 |
|
|
vec_double2 x_abs; // absolute value of x
|
| 106 |
|
|
vec_double2 x_trans; // transformed x when |x| > 0.5
|
| 107 |
|
|
vec_double2 x2, x3; // x squared and x cubed, respectively.
|
| 108 |
|
|
vec_double2 result;
|
| 109 |
|
|
vec_double2 multiplier, addend;
|
| 110 |
|
|
vec_double2 p, q, pq;
|
| 111 |
|
|
vec_double2 half = spu_splats(0.5);
|
| 112 |
|
|
vec_double2 sign = (vec_double2)spu_splats(0x8000000000000000ULL);
|
| 113 |
|
|
vec_uchar16 splat_hi = ((vec_uchar16){0,1,2,3, 0,1,2,3, 8,9,10,11, 8,9,10,11});
|
| 114 |
|
|
|
| 115 |
|
|
// Compute the absolute value of x
|
| 116 |
|
|
x_abs = spu_andc(x, sign);
|
| 117 |
|
|
|
| 118 |
|
|
// Perform transformation for the case where |x| > 0.5. We rely on
|
| 119 |
|
|
// sqrtd2 producing a NAN is |x| > 1.0.
|
| 120 |
|
|
x_trans = _sqrtd2(spu_nmsub(x_abs, half, half));
|
| 121 |
|
|
|
| 122 |
|
|
// Determine the correct addend and multiplier.
|
| 123 |
|
|
x_gt_half = spu_cmpgt((vec_uint4)x_abs, (vec_uint4)half);
|
| 124 |
|
|
x_eq_half = spu_cmpeq((vec_uint4)x_abs, (vec_uint4)half);
|
| 125 |
|
|
x_gt_half = spu_or(x_gt_half, spu_and(x_eq_half, spu_rlqwbyte(x_gt_half, 4)));
|
| 126 |
|
|
x_gt_half = spu_shuffle(x_gt_half, x_gt_half, splat_hi);
|
| 127 |
|
|
|
| 128 |
|
|
addend = spu_and(spu_sel(spu_splats((double)SM_PI_2), x, (vec_ullong2)sign), (vec_double2)x_gt_half);
|
| 129 |
|
|
|
| 130 |
|
|
multiplier = spu_sel(spu_splats(-1.0), spu_sel(spu_splats(2.0), x, (vec_ullong2)sign), (vec_ullong2)x_gt_half);
|
| 131 |
|
|
|
| 132 |
|
|
// Select whether to use the x or the transformed x for the polygon evaluation.
|
| 133 |
|
|
// if |x| > 0.5 use x_trans
|
| 134 |
|
|
// else use x
|
| 135 |
|
|
|
| 136 |
|
|
x = spu_sel(x, x_trans, (vec_ullong2)x_gt_half);
|
| 137 |
|
|
|
| 138 |
|
|
// Compute the polynomials.
|
| 139 |
|
|
|
| 140 |
|
|
x2 = spu_mul(x, x);
|
| 141 |
|
|
x3 = spu_mul(x2, x);
|
| 142 |
|
|
|
| 143 |
|
|
p = spu_madd(spu_splats(0.004253011369004428248960), x2, spu_splats(-0.6019598008014123785661));
|
| 144 |
|
|
p = spu_madd(p, x2, spu_splats(5.444622390564711410273));
|
| 145 |
|
|
p = spu_madd(p, x2, spu_splats(-16.26247967210700244449));
|
| 146 |
|
|
p = spu_madd(p, x2, spu_splats(19.56261983317594739197));
|
| 147 |
|
|
p = spu_madd(p, x2, spu_splats(-8.198089802484824371615));
|
| 148 |
|
|
|
| 149 |
|
|
q = spu_add(x2, spu_splats(-14.74091372988853791896));
|
| 150 |
|
|
q = spu_madd(q, x2, spu_splats(70.49610280856842141659));
|
| 151 |
|
|
q = spu_madd(q, x2, spu_splats(-147.1791292232726029859));
|
| 152 |
|
|
q = spu_madd(q, x2, spu_splats(139.5105614657485689735));
|
| 153 |
|
|
q = spu_madd(q, x2, spu_splats(-49.18853881490881290097));
|
| 154 |
|
|
|
| 155 |
|
|
// Compute the rational solution p/q and final multiplication and addend
|
| 156 |
|
|
// correction.
|
| 157 |
|
|
pq = _divd2(p, q);
|
| 158 |
|
|
|
| 159 |
|
|
result = spu_nmsub(spu_madd(x3, pq, x), multiplier, addend);
|
| 160 |
|
|
|
| 161 |
|
|
return (result);
|
| 162 |
|
|
}
|
| 163 |
|
|
|
| 164 |
|
|
#endif /* _ASIND2_H_ */
|
| 165 |
|
|
#endif /* __SPU__ */
|
