Mercurial > hg > vamp-build-and-test

annotate DEPENDENCIES/generic/include/boost/math/special_functions/detail/lanczos_sse2.hpp @ 133:4acb5d8d80b6 tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Don't fail environmental check if README.md exists (but .txt and no-suffix don't)
author Chris Cannam
date Tue, 30 Jul 2019 12:25:44 +0100
parents c530137014c0
children
rev   line source
Chris@16 1 // (C) Copyright John Maddock 2006.
Chris@16 2 // Use, modification and distribution are subject to the
Chris@16 3 // Boost Software License, Version 1.0. (See accompanying file
Chris@16 4 // LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
Chris@16 5
Chris@16 6 #ifndef BOOST_MATH_SPECIAL_FUNCTIONS_LANCZOS_SSE2
Chris@16 7 #define BOOST_MATH_SPECIAL_FUNCTIONS_LANCZOS_SSE2
Chris@16 8
Chris@16 9 #ifdef _MSC_VER
Chris@16 10 #pragma once
Chris@16 11 #endif
Chris@16 12
Chris@16 13 #include <emmintrin.h>
Chris@16 14
Chris@16 15 #if defined(__GNUC__) || defined(__PGI)
Chris@16 16 #define ALIGN16 __attribute__((__aligned__(16)))
Chris@16 17 #else
Chris@16 18 #define ALIGN16 __declspec(align(16))
Chris@16 19 #endif
Chris@16 20
Chris@16 21 namespace boost{ namespace math{ namespace lanczos{
Chris@16 22
Chris@16 23 template <>
Chris@16 24 inline double lanczos13m53::lanczos_sum<double>(const double& x)
Chris@16 25 {
Chris@16 26 static const ALIGN16 double coeff[26] = {
Chris@16 27 static_cast<double>(2.506628274631000270164908177133837338626L),
Chris@16 28 static_cast<double>(1u),
Chris@16 29 static_cast<double>(210.8242777515793458725097339207133627117L),
Chris@16 30 static_cast<double>(66u),
Chris@16 31 static_cast<double>(8071.672002365816210638002902272250613822L),
Chris@16 32 static_cast<double>(1925u),
Chris@16 33 static_cast<double>(186056.2653952234950402949897160456992822L),
Chris@16 34 static_cast<double>(32670u),
Chris@16 35 static_cast<double>(2876370.628935372441225409051620849613599L),
Chris@16 36 static_cast<double>(357423u),
Chris@16 37 static_cast<double>(31426415.58540019438061423162831820536287L),
Chris@16 38 static_cast<double>(2637558u),
Chris@16 39 static_cast<double>(248874557.8620541565114603864132294232163L),
Chris@16 40 static_cast<double>(13339535u),
Chris@16 41 static_cast<double>(1439720407.311721673663223072794912393972L),
Chris@16 42 static_cast<double>(45995730u),
Chris@16 43 static_cast<double>(6039542586.35202800506429164430729792107L),
Chris@16 44 static_cast<double>(105258076u),
Chris@16 45 static_cast<double>(17921034426.03720969991975575445893111267L),
Chris@16 46 static_cast<double>(150917976u),
Chris@16 47 static_cast<double>(35711959237.35566804944018545154716670596L),
Chris@16 48 static_cast<double>(120543840u),
Chris@16 49 static_cast<double>(42919803642.64909876895789904700198885093L),
Chris@16 50 static_cast<double>(39916800u),
Chris@16 51 static_cast<double>(23531376880.41075968857200767445163675473L),
Chris@16 52 static_cast<double>(0u)
Chris@16 53 };
Chris@101 54 __m128d vx = _mm_load1_pd(&x);
Chris@101 55 __m128d sum_even = _mm_load_pd(coeff);
Chris@101 56 __m128d sum_odd = _mm_load_pd(coeff+2);
Chris@101 57 __m128d nc_odd, nc_even;
Chris@101 58 __m128d vx2 = _mm_mul_pd(vx, vx);
Chris@16 59
Chris@16 60 sum_even = _mm_mul_pd(sum_even, vx2);
Chris@16 61 nc_even = _mm_load_pd(coeff + 4);
Chris@16 62 sum_odd = _mm_mul_pd(sum_odd, vx2);
Chris@16 63 nc_odd = _mm_load_pd(coeff + 6);
Chris@16 64 sum_even = _mm_add_pd(sum_even, nc_even);
Chris@16 65 sum_odd = _mm_add_pd(sum_odd, nc_odd);
Chris@16 66
Chris@16 67 sum_even = _mm_mul_pd(sum_even, vx2);
Chris@16 68 nc_even = _mm_load_pd(coeff + 8);
Chris@16 69 sum_odd = _mm_mul_pd(sum_odd, vx2);
Chris@16 70 nc_odd = _mm_load_pd(coeff + 10);
Chris@16 71 sum_even = _mm_add_pd(sum_even, nc_even);
Chris@16 72 sum_odd = _mm_add_pd(sum_odd, nc_odd);
Chris@16 73
Chris@16 74 sum_even = _mm_mul_pd(sum_even, vx2);
Chris@16 75 nc_even = _mm_load_pd(coeff + 12);
Chris@16 76 sum_odd = _mm_mul_pd(sum_odd, vx2);
Chris@16 77 nc_odd = _mm_load_pd(coeff + 14);
Chris@16 78 sum_even = _mm_add_pd(sum_even, nc_even);
Chris@16 79 sum_odd = _mm_add_pd(sum_odd, nc_odd);
Chris@16 80
Chris@16 81 sum_even = _mm_mul_pd(sum_even, vx2);
Chris@16 82 nc_even = _mm_load_pd(coeff + 16);
Chris@16 83 sum_odd = _mm_mul_pd(sum_odd, vx2);
Chris@16 84 nc_odd = _mm_load_pd(coeff + 18);
Chris@16 85 sum_even = _mm_add_pd(sum_even, nc_even);
Chris@16 86 sum_odd = _mm_add_pd(sum_odd, nc_odd);
Chris@16 87
Chris@16 88 sum_even = _mm_mul_pd(sum_even, vx2);
Chris@16 89 nc_even = _mm_load_pd(coeff + 20);
Chris@16 90 sum_odd = _mm_mul_pd(sum_odd, vx2);
Chris@16 91 nc_odd = _mm_load_pd(coeff + 22);
Chris@16 92 sum_even = _mm_add_pd(sum_even, nc_even);
Chris@16 93 sum_odd = _mm_add_pd(sum_odd, nc_odd);
Chris@16 94
Chris@16 95 sum_even = _mm_mul_pd(sum_even, vx2);
Chris@16 96 nc_even = _mm_load_pd(coeff + 24);
Chris@16 97 sum_odd = _mm_mul_pd(sum_odd, vx);
Chris@16 98 sum_even = _mm_add_pd(sum_even, nc_even);
Chris@16 99 sum_even = _mm_add_pd(sum_even, sum_odd);
Chris@16 100
Chris@16 101
Chris@16 102 double ALIGN16 t[2];
Chris@16 103 _mm_store_pd(t, sum_even);
Chris@16 104
Chris@16 105 return t[0] / t[1];
Chris@16 106 }
Chris@16 107
Chris@16 108 template <>
Chris@16 109 inline double lanczos13m53::lanczos_sum_expG_scaled<double>(const double& x)
Chris@16 110 {
Chris@16 111 static const ALIGN16 double coeff[26] = {
Chris@16 112 static_cast<double>(0.006061842346248906525783753964555936883222L),
Chris@16 113 static_cast<double>(1u),
Chris@16 114 static_cast<double>(0.5098416655656676188125178644804694509993L),
Chris@16 115 static_cast<double>(66u),
Chris@16 116 static_cast<double>(19.51992788247617482847860966235652136208L),
Chris@16 117 static_cast<double>(1925u),
Chris@16 118 static_cast<double>(449.9445569063168119446858607650988409623L),
Chris@16 119 static_cast<double>(32670u),
Chris@16 120 static_cast<double>(6955.999602515376140356310115515198987526L),
Chris@16 121 static_cast<double>(357423u),
Chris@16 122 static_cast<double>(75999.29304014542649875303443598909137092L),
Chris@16 123 static_cast<double>(2637558u),
Chris@16 124 static_cast<double>(601859.6171681098786670226533699352302507L),
Chris@16 125 static_cast<double>(13339535u),
Chris@16 126 static_cast<double>(3481712.15498064590882071018964774556468L),
Chris@16 127 static_cast<double>(45995730u),
Chris@16 128 static_cast<double>(14605578.08768506808414169982791359218571L),
Chris@16 129 static_cast<double>(105258076u),
Chris@16 130 static_cast<double>(43338889.32467613834773723740590533316085L),
Chris@16 131 static_cast<double>(150917976u),
Chris@16 132 static_cast<double>(86363131.28813859145546927288977868422342L),
Chris@16 133 static_cast<double>(120543840u),
Chris@16 134 static_cast<double>(103794043.1163445451906271053616070238554L),
Chris@16 135 static_cast<double>(39916800u),
Chris@16 136 static_cast<double>(56906521.91347156388090791033559122686859L),
Chris@16 137 static_cast<double>(0u)
Chris@16 138 };
Chris@101 139 __m128d vx = _mm_load1_pd(&x);
Chris@101 140 __m128d sum_even = _mm_load_pd(coeff);
Chris@101 141 __m128d sum_odd = _mm_load_pd(coeff+2);
Chris@101 142 __m128d nc_odd, nc_even;
Chris@101 143 __m128d vx2 = _mm_mul_pd(vx, vx);
Chris@16 144
Chris@16 145 sum_even = _mm_mul_pd(sum_even, vx2);
Chris@16 146 nc_even = _mm_load_pd(coeff + 4);
Chris@16 147 sum_odd = _mm_mul_pd(sum_odd, vx2);
Chris@16 148 nc_odd = _mm_load_pd(coeff + 6);
Chris@16 149 sum_even = _mm_add_pd(sum_even, nc_even);
Chris@16 150 sum_odd = _mm_add_pd(sum_odd, nc_odd);
Chris@16 151
Chris@16 152 sum_even = _mm_mul_pd(sum_even, vx2);
Chris@16 153 nc_even = _mm_load_pd(coeff + 8);
Chris@16 154 sum_odd = _mm_mul_pd(sum_odd, vx2);
Chris@16 155 nc_odd = _mm_load_pd(coeff + 10);
Chris@16 156 sum_even = _mm_add_pd(sum_even, nc_even);
Chris@16 157 sum_odd = _mm_add_pd(sum_odd, nc_odd);
Chris@16 158
Chris@16 159 sum_even = _mm_mul_pd(sum_even, vx2);
Chris@16 160 nc_even = _mm_load_pd(coeff + 12);
Chris@16 161 sum_odd = _mm_mul_pd(sum_odd, vx2);
Chris@16 162 nc_odd = _mm_load_pd(coeff + 14);
Chris@16 163 sum_even = _mm_add_pd(sum_even, nc_even);
Chris@16 164 sum_odd = _mm_add_pd(sum_odd, nc_odd);
Chris@16 165
Chris@16 166 sum_even = _mm_mul_pd(sum_even, vx2);
Chris@16 167 nc_even = _mm_load_pd(coeff + 16);
Chris@16 168 sum_odd = _mm_mul_pd(sum_odd, vx2);
Chris@16 169 nc_odd = _mm_load_pd(coeff + 18);
Chris@16 170 sum_even = _mm_add_pd(sum_even, nc_even);
Chris@16 171 sum_odd = _mm_add_pd(sum_odd, nc_odd);
Chris@16 172
Chris@16 173 sum_even = _mm_mul_pd(sum_even, vx2);
Chris@16 174 nc_even = _mm_load_pd(coeff + 20);
Chris@16 175 sum_odd = _mm_mul_pd(sum_odd, vx2);
Chris@16 176 nc_odd = _mm_load_pd(coeff + 22);
Chris@16 177 sum_even = _mm_add_pd(sum_even, nc_even);
Chris@16 178 sum_odd = _mm_add_pd(sum_odd, nc_odd);
Chris@16 179
Chris@16 180 sum_even = _mm_mul_pd(sum_even, vx2);
Chris@16 181 nc_even = _mm_load_pd(coeff + 24);
Chris@16 182 sum_odd = _mm_mul_pd(sum_odd, vx);
Chris@16 183 sum_even = _mm_add_pd(sum_even, nc_even);
Chris@16 184 sum_even = _mm_add_pd(sum_even, sum_odd);
Chris@16 185
Chris@16 186
Chris@16 187 double ALIGN16 t[2];
Chris@16 188 _mm_store_pd(t, sum_even);
Chris@16 189
Chris@16 190 return t[0] / t[1];
Chris@16 191 }
Chris@16 192
Chris@16 193 } // namespace lanczos
Chris@16 194 } // namespace math
Chris@16 195 } // namespace boost
Chris@16 196
Chris@16 197 #undef ALIGN16
Chris@16 198
Chris@16 199 #endif // BOOST_MATH_SPECIAL_FUNCTIONS_LANCZOS
Chris@16 200
Chris@16 201
Chris@16 202
Chris@16 203
Chris@16 204