[16269] | 1 | /*
|
---|
| 2 | * exp.h
|
---|
| 3 | * The basic idea is to exploit Pade polynomials.
|
---|
| 4 | * A lot of ideas were inspired by the cephes math library (by Stephen L. Moshier
|
---|
| 5 | * moshier@na-net.ornl.gov) as well as actual code.
|
---|
| 6 | * The Cephes library can be found here: http://www.netlib.org/cephes/
|
---|
| 7 | *
|
---|
| 8 | * Created on: Jun 23, 2012
|
---|
| 9 | * Author: Danilo Piparo, Thomas Hauth, Vincenzo Innocente
|
---|
| 10 | */
|
---|
| 11 |
|
---|
| 12 | /*
|
---|
| 13 | * VDT is free software: you can redistribute it and/or modify
|
---|
| 14 | * it under the terms of the GNU Lesser Public License as published by
|
---|
| 15 | * the Free Software Foundation, either version 3 of the License, or
|
---|
| 16 | * (at your option) any later version.
|
---|
| 17 | *
|
---|
| 18 | * This program is distributed in the hope that it will be useful,
|
---|
| 19 | * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
| 20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
---|
| 21 | * GNU Lesser Public License for more details.
|
---|
| 22 | *
|
---|
| 23 | * You should have received a copy of the GNU Lesser Public License
|
---|
| 24 | * along with this program. If not, see <http://www.gnu.org/licenses/>.
|
---|
| 25 | */
|
---|
| 26 |
|
---|
| 27 | #ifndef _VDT_EXP_
|
---|
| 28 | #define _VDT_EXP_
|
---|
| 29 |
|
---|
| 30 | #include "vdtcore_common.h"
|
---|
| 31 | #include <limits>
|
---|
| 32 |
|
---|
| 33 | namespace vdt{
|
---|
| 34 |
|
---|
| 35 | namespace details{
|
---|
| 36 |
|
---|
| 37 | const double EXP_LIMIT = 708;
|
---|
| 38 |
|
---|
| 39 | const double PX1exp = 1.26177193074810590878E-4;
|
---|
| 40 | const double PX2exp = 3.02994407707441961300E-2;
|
---|
| 41 | const double PX3exp = 9.99999999999999999910E-1;
|
---|
| 42 | const double QX1exp = 3.00198505138664455042E-6;
|
---|
| 43 | const double QX2exp = 2.52448340349684104192E-3;
|
---|
| 44 | const double QX3exp = 2.27265548208155028766E-1;
|
---|
| 45 | const double QX4exp = 2.00000000000000000009E0;
|
---|
| 46 |
|
---|
| 47 | const double LOG2E = 1.4426950408889634073599; // 1/log(2)
|
---|
| 48 |
|
---|
| 49 | const float MAXLOGF = 88.72283905206835f;
|
---|
| 50 | const float MINLOGF = -88.f;
|
---|
| 51 |
|
---|
| 52 | const float C1F = 0.693359375f;
|
---|
| 53 | const float C2F = -2.12194440e-4f;
|
---|
| 54 |
|
---|
| 55 | const float PX1expf = 1.9875691500E-4f;
|
---|
| 56 | const float PX2expf =1.3981999507E-3f;
|
---|
| 57 | const float PX3expf =8.3334519073E-3f;
|
---|
| 58 | const float PX4expf =4.1665795894E-2f;
|
---|
| 59 | const float PX5expf =1.6666665459E-1f;
|
---|
| 60 | const float PX6expf =5.0000001201E-1f;
|
---|
| 61 |
|
---|
| 62 | const float LOG2EF = 1.44269504088896341f;
|
---|
| 63 |
|
---|
| 64 | }
|
---|
| 65 |
|
---|
| 66 | // Exp double precision --------------------------------------------------------
|
---|
| 67 |
|
---|
| 68 |
|
---|
| 69 | /// Exponential Function double precision
|
---|
| 70 | inline double fast_exp(double initial_x){
|
---|
| 71 |
|
---|
| 72 | double x = initial_x;
|
---|
| 73 | double px=details::fpfloor(details::LOG2E * x +0.5);
|
---|
| 74 |
|
---|
| 75 | const int32_t n = int32_t(px);
|
---|
| 76 |
|
---|
| 77 | x -= px * 6.93145751953125E-1;
|
---|
| 78 | x -= px * 1.42860682030941723212E-6;
|
---|
| 79 |
|
---|
| 80 | const double xx = x * x;
|
---|
| 81 |
|
---|
| 82 | // px = x * P(x**2).
|
---|
| 83 | px = details::PX1exp;
|
---|
| 84 | px *= xx;
|
---|
| 85 | px += details::PX2exp;
|
---|
| 86 | px *= xx;
|
---|
| 87 | px += details::PX3exp;
|
---|
| 88 | px *= x;
|
---|
| 89 |
|
---|
| 90 | // Evaluate Q(x**2).
|
---|
| 91 | double qx = details::QX1exp;
|
---|
| 92 | qx *= xx;
|
---|
| 93 | qx += details::QX2exp;
|
---|
| 94 | qx *= xx;
|
---|
| 95 | qx += details::QX3exp;
|
---|
| 96 | qx *= xx;
|
---|
| 97 | qx += details::QX4exp;
|
---|
| 98 |
|
---|
| 99 | // e**x = 1 + 2x P(x**2)/( Q(x**2) - P(x**2) )
|
---|
| 100 | x = px / (qx - px);
|
---|
| 101 | x = 1.0 + 2.0 * x;
|
---|
| 102 |
|
---|
| 103 | // Build 2^n in double.
|
---|
| 104 | x *= details::uint642dp(( ((uint64_t)n) +1023)<<52);
|
---|
| 105 |
|
---|
| 106 | if (initial_x > details::EXP_LIMIT)
|
---|
| 107 | x = std::numeric_limits<double>::infinity();
|
---|
| 108 | if (initial_x < -details::EXP_LIMIT)
|
---|
| 109 | x = 0.;
|
---|
| 110 |
|
---|
| 111 | return x;
|
---|
| 112 |
|
---|
| 113 | }
|
---|
| 114 |
|
---|
| 115 | // Exp single precision --------------------------------------------------------
|
---|
| 116 |
|
---|
| 117 | /// Exponential Function single precision
|
---|
| 118 | inline float fast_expf(float initial_x) {
|
---|
| 119 |
|
---|
| 120 | float x = initial_x;
|
---|
| 121 |
|
---|
| 122 | float z = details::fpfloor( details::LOG2EF * x +0.5f ); /* floor() truncates toward -infinity. */
|
---|
| 123 |
|
---|
| 124 | x -= z * details::C1F;
|
---|
| 125 | x -= z * details::C2F;
|
---|
| 126 | const int32_t n = int32_t ( z );
|
---|
| 127 |
|
---|
| 128 | const float x2 = x * x;
|
---|
| 129 |
|
---|
| 130 | z = x*details::PX1expf;
|
---|
| 131 | z += details::PX2expf;
|
---|
| 132 | z *= x;
|
---|
| 133 | z += details::PX3expf;
|
---|
| 134 | z *= x;
|
---|
| 135 | z += details::PX4expf;
|
---|
| 136 | z *= x;
|
---|
| 137 | z += details::PX5expf;
|
---|
| 138 | z *= x;
|
---|
| 139 | z += details::PX6expf;
|
---|
| 140 | z *= x2;
|
---|
| 141 | z += x + 1.0f;
|
---|
| 142 |
|
---|
| 143 | /* multiply by power of 2 */
|
---|
| 144 | z *= details::uint322sp((n+0x7f)<<23);
|
---|
| 145 |
|
---|
| 146 | if (initial_x > details::MAXLOGF) z=std::numeric_limits<float>::infinity();
|
---|
| 147 | if (initial_x < details::MINLOGF) z=0.f;
|
---|
| 148 |
|
---|
| 149 | return z;
|
---|
| 150 |
|
---|
| 151 | }
|
---|
| 152 |
|
---|
| 153 | //------------------------------------------------------------------------------
|
---|
| 154 |
|
---|
| 155 | } // end namespace vdt
|
---|
| 156 |
|
---|
| 157 | #endif
|
---|