src/gromacs/simd/impl_x86_avx2_256/impl_x86_avx2_256_simd_float.h

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2014,2015, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35
  36 #ifndef GMX_SIMD_IMPL_X86_AVX2_256_SIMD_FLOAT_H
  37 #define GMX_SIMD_IMPL_X86_AVX2_256_SIMD_FLOAT_H
  38
  39 #include "config.h"
  40
  41 #include <immintrin.h>
  42
  43 #include "gromacs/simd/impl_x86_avx_256/impl_x86_avx_256_simd_float.h"
  44
  45 namespace gmx
  46 {
  47
  48 class SimdFIBool
  49 {
  50     public:
  51         SimdFIBool() {}
  52
  53         SimdFIBool(bool b) : simdInternal_(_mm256_set1_epi32( b ? 0xFFFFFFFF : 0)) {}
  54
  55         // Internal utility constructor to simplify return statements
  56         SimdFIBool(__m256i simd) : simdInternal_(simd) {}
  57
  58         __m256i  simdInternal_;
  59 };
  60
  61 static inline SimdFloat gmx_simdcall
  62 fma(SimdFloat a, SimdFloat b, SimdFloat c)
  63 {
  64     return {
  65                _mm256_fmadd_ps(a.simdInternal_, b.simdInternal_, c.simdInternal_)
  66     };
  67 }
  68
  69 static inline SimdFloat gmx_simdcall
  70 fms(SimdFloat a, SimdFloat b, SimdFloat c)
  71 {
  72     return {
  73                _mm256_fmsub_ps(a.simdInternal_, b.simdInternal_, c.simdInternal_)
  74     };
  75 }
  76
  77 static inline SimdFloat gmx_simdcall
  78 fnma(SimdFloat a, SimdFloat b, SimdFloat c)
  79 {
  80     return {
  81                _mm256_fnmadd_ps(a.simdInternal_, b.simdInternal_, c.simdInternal_)
  82     };
  83 }
  84
  85 static inline SimdFloat gmx_simdcall
  86 fnms(SimdFloat a, SimdFloat b, SimdFloat c)
  87 {
  88     return {
  89                _mm256_fnmsub_ps(a.simdInternal_, b.simdInternal_, c.simdInternal_)
  90     };
  91 }
  92
  93 static inline SimdFBool gmx_simdcall
  94 testBits(SimdFloat a)
  95 {
  96     __m256i ia  = _mm256_castps_si256(a.simdInternal_);
  97     __m256i res = _mm256_andnot_si256( _mm256_cmpeq_epi32(ia, _mm256_setzero_si256()), _mm256_cmpeq_epi32(ia, ia));
  98
  99     return {
 100                _mm256_castsi256_ps(res)
 101     };
 102 }
 103
 104 static inline SimdFloat gmx_simdcall
 105 frexp(SimdFloat value, SimdFInt32 * exponent)
 106 {
 107     const __m256  exponentMask = _mm256_castsi256_ps(_mm256_set1_epi32(0x7F800000));
 108     const __m256  mantissaMask = _mm256_castsi256_ps(_mm256_set1_epi32(0x807FFFFF));
 109     const __m256i exponentBias = _mm256_set1_epi32(126); // add 1 to make our definition identical to frexp()
 110     const __m256  half         = _mm256_set1_ps(0.5);
 111     __m256i       iExponent;
 112
 113     iExponent               = _mm256_castps_si256(_mm256_and_ps(value.simdInternal_, exponentMask));
 114     exponent->simdInternal_ = _mm256_sub_epi32(_mm256_srli_epi32(iExponent, 23), exponentBias);
 115
 116     return {
 117                _mm256_or_ps(_mm256_and_ps(value.simdInternal_, mantissaMask), half)
 118     };
 119 }
 120
 121 static inline SimdFloat gmx_simdcall
 122 ldexp(SimdFloat value, SimdFInt32 exponent)
 123 {
 124     const __m256i  exponentBias = _mm256_set1_epi32(127);
 125     __m256i        iExponent;
 126
 127     iExponent = _mm256_slli_epi32(_mm256_add_epi32(exponent.simdInternal_, exponentBias), 23);
 128     return {
 129                _mm256_mul_ps(value.simdInternal_, _mm256_castsi256_ps(iExponent))
 130     };
 131 }
 132
 133 static inline SimdFInt32 gmx_simdcall
 134 operator<<(SimdFInt32 a, int n)
 135 {
 136     return {
 137                _mm256_slli_epi32(a.simdInternal_, n)
 138     };
 139 }
 140
 141 static inline SimdFInt32 gmx_simdcall
 142 operator>>(SimdFInt32 a, int n)
 143 {
 144     return {
 145                _mm256_srli_epi32(a.simdInternal_, n)
 146     };
 147 }
 148
 149 static inline SimdFInt32 gmx_simdcall
 150 operator&(SimdFInt32 a, SimdFInt32 b)
 151 {
 152     return {
 153                _mm256_and_si256(a.simdInternal_, b.simdInternal_)
 154     };
 155 }
 156
 157 static inline SimdFInt32 gmx_simdcall
 158 andNot(SimdFInt32 a, SimdFInt32 b)
 159 {
 160     return {
 161                _mm256_andnot_si256(a.simdInternal_, b.simdInternal_)
 162     };
 163 }
 164
 165 static inline SimdFInt32 gmx_simdcall
 166 operator|(SimdFInt32 a, SimdFInt32 b)
 167 {
 168     return {
 169                _mm256_or_si256(a.simdInternal_, b.simdInternal_)
 170     };
 171 }
 172
 173 static inline SimdFInt32 gmx_simdcall
 174 operator^(SimdFInt32 a, SimdFInt32 b)
 175 {
 176     return {
 177                _mm256_xor_si256(a.simdInternal_, b.simdInternal_)
 178     };
 179 }
 180
 181 static inline SimdFInt32 gmx_simdcall
 182 operator+(SimdFInt32 a, SimdFInt32 b)
 183 {
 184     return {
 185                _mm256_add_epi32(a.simdInternal_, b.simdInternal_)
 186     };
 187 }
 188
 189 static inline SimdFInt32 gmx_simdcall
 190 operator-(SimdFInt32 a, SimdFInt32 b)
 191 {
 192     return {
 193                _mm256_sub_epi32(a.simdInternal_, b.simdInternal_)
 194     };
 195 }
 196
 197 static inline SimdFInt32 gmx_simdcall
 198 operator*(SimdFInt32 a, SimdFInt32 b)
 199 {
 200     return {
 201                _mm256_mullo_epi32(a.simdInternal_, b.simdInternal_)
 202     };
 203 }
 204
 205 static inline SimdFIBool gmx_simdcall
 206 operator==(SimdFInt32 a, SimdFInt32 b)
 207 {
 208     return {
 209                _mm256_cmpeq_epi32(a.simdInternal_, b.simdInternal_)
 210     };
 211 }
 212
 213 static inline SimdFIBool gmx_simdcall
 214 testBits(SimdFInt32 a)
 215 {
 216     return {
 217                _mm256_andnot_si256(_mm256_cmpeq_epi32(a.simdInternal_, _mm256_setzero_si256()),
 218                                    _mm256_cmpeq_epi32(a.simdInternal_, a.simdInternal_))
 219     };
 220 }
 221
 222 static inline SimdFIBool gmx_simdcall
 223 operator<(SimdFInt32 a, SimdFInt32 b)
 224 {
 225     return {
 226                _mm256_cmpgt_epi32(b.simdInternal_, a.simdInternal_)
 227     };
 228 }
 229
 230 static inline SimdFIBool gmx_simdcall
 231 operator&&(SimdFIBool a, SimdFIBool b)
 232 {
 233     return {
 234                _mm256_and_si256(a.simdInternal_, b.simdInternal_)
 235     };
 236 }
 237
 238 static inline SimdFIBool gmx_simdcall
 239 operator||(SimdFIBool a, SimdFIBool b)
 240 {
 241     return {
 242                _mm256_or_si256(a.simdInternal_, b.simdInternal_)
 243     };
 244 }
 245
 246 static inline bool gmx_simdcall
 247 anyTrue(SimdFIBool a) { return _mm256_movemask_epi8(a.simdInternal_) != 0; }
 248
 249 static inline SimdFInt32 gmx_simdcall
 250 selectByMask(SimdFInt32 a, SimdFIBool mask)
 251 {
 252     return {
 253                _mm256_and_si256(a.simdInternal_, mask.simdInternal_)
 254     };
 255 }
 256
 257 static inline SimdFInt32 gmx_simdcall
 258 selectByNotMask(SimdFInt32 a, SimdFIBool mask)
 259 {
 260     return {
 261                _mm256_andnot_si256(mask.simdInternal_, a.simdInternal_)
 262     };
 263 }
 264
 265 static inline SimdFInt32 gmx_simdcall
 266 blend(SimdFInt32 a, SimdFInt32 b, SimdFIBool sel)
 267 {
 268     return {
 269                _mm256_blendv_epi8(a.simdInternal_, b.simdInternal_, sel.simdInternal_)
 270     };
 271 }
 272
 273 static inline SimdFIBool gmx_simdcall
 274 cvtB2IB(SimdFBool a)
 275 {
 276     return {
 277                _mm256_castps_si256(a.simdInternal_)
 278     };
 279 }
 280
 281 static inline SimdFBool gmx_simdcall
 282 cvtIB2B(SimdFIBool a)
 283 {
 284     return {
 285                _mm256_castsi256_ps(a.simdInternal_)
 286     };
 287 }
 288
 289 }      // namespace gmx
 290
 291 #endif // GMX_SIMD_IMPL_X86_AVX2_256_SIMD_FLOAT_H