src/gromacs/simd/tests/simd_floatingpoint.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2014, 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 #include "config.h"
  36
  37 #include <math.h>
  38 #include "gromacs/math/utilities.h"
  39
  40 #include "simd.h"
  41
  42 namespace gmx
  43 {
  44 namespace test
  45 {
  46 namespace
  47 {
  48
  49 /*! \cond internal */
  50 /*! \addtogroup module_simd */
  51 /*! \{ */
  52
  53 #ifdef GMX_SIMD_HAVE_REAL
  54
  55 /*! \brief Test fixture for floating-point tests (identical to the generic \ref SimdTest) */
  56 typedef SimdTest SimdFloatingpointTest;
  57
  58 TEST_F(SimdFloatingpointTest, gmxSimdSetZeroR)
  59 {
  60     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(0.0), gmx_simd_setzero_r());
  61 }
  62
  63 TEST_F(SimdFloatingpointTest, gmxSimdSet1R)
  64 {
  65     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(1.0), gmx_simd_set1_r(1.0));
  66 }
  67
  68 TEST_F(SimdFloatingpointTest, gmxSimdLoad1R)
  69 {
  70     real r = 2.0;
  71     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(r), gmx_simd_load1_r(&r));
  72 }
  73
  74 TEST_F(SimdFloatingpointTest, gmxSimdAddR)
  75 {
  76     GMX_EXPECT_SIMD_REAL_EQ(rSimd_5_7_9,
  77                             gmx_simd_add_r(rSimd_1_2_3, rSimd_4_5_6)); // 1+4=5, 2+5=7, 3+6=9
  78 }
  79
  80 TEST_F(SimdFloatingpointTest, gmxSimdSubR)
  81 {
  82     GMX_EXPECT_SIMD_REAL_EQ(rSimd_4_5_6,
  83                             gmx_simd_sub_r(rSimd_5_7_9, rSimd_1_2_3)); // 5-1=4, 7-2=5, 9-3=6
  84 }
  85
  86 TEST_F(SimdFloatingpointTest, gmxSimdMulR)
  87 {
  88     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(4, 10, 18),
  89                             gmx_simd_mul_r(rSimd_1_2_3, rSimd_4_5_6));
  90 }
  91
  92 TEST_F(SimdFloatingpointTest, gmxSimdFmaddR)
  93 {
  94     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(11, 18, 27),
  95                             gmx_simd_fmadd_r(rSimd_1_2_3, rSimd_4_5_6, rSimd_7_8_9)); // 1*4+7, etc.
  96 }
  97
  98 TEST_F(SimdFloatingpointTest, gmxSimdFmsubR)
  99 {
 100     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(-3, 2, 9),
 101                             gmx_simd_fmsub_r(rSimd_1_2_3, rSimd_4_5_6, rSimd_7_8_9)); // 1*4-7, etc.
 102 }
 103
 104 TEST_F(SimdFloatingpointTest, gmxSimdFnmaddR)
 105 {
 106     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(3, -2, -9),
 107                             gmx_simd_fnmadd_r(rSimd_1_2_3, rSimd_4_5_6, rSimd_7_8_9)); // -1*4+7, etc.
 108 }
 109
 110 TEST_F(SimdFloatingpointTest, gmxSimdFnmsubR)
 111 {
 112     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(-11, -18, -27),
 113                             gmx_simd_fnmsub_r(rSimd_1_2_3, rSimd_4_5_6, rSimd_7_8_9)); // -1*4-7, etc.
 114 }
 115
 116 TEST_F(SimdFloatingpointTest, gmxSimdFabsR)
 117 {
 118     GMX_EXPECT_SIMD_REAL_EQ(rSimd_1_2_3, gmx_simd_fabs_r(rSimd_1_2_3));    // fabs(x)=x
 119     GMX_EXPECT_SIMD_REAL_EQ(rSimd_1_2_3, gmx_simd_fabs_r(rSimd_m1_m2_m3)); // fabs(-x)=x
 120 }
 121
 122 TEST_F(SimdFloatingpointTest, gmxSimdFnegR)
 123 {
 124     GMX_EXPECT_SIMD_REAL_EQ(rSimd_m1_m2_m3, gmx_simd_fneg_r(rSimd_1_2_3));    // fneg(x)=-x
 125     GMX_EXPECT_SIMD_REAL_EQ(rSimd_1_2_3,    gmx_simd_fneg_r(rSimd_m1_m2_m3)); // fneg(-x)=x
 126 }
 127
 128 #ifdef GMX_SIMD_HAVE_LOGICAL
 129 TEST_F(SimdFloatingpointTest, gmxSimdAndR)
 130 {
 131     GMX_EXPECT_SIMD_REAL_EQ(rSimd_Bits3, gmx_simd_and_r(rSimd_Bits1, rSimd_Bits2)); // Bits1 & Bits2 = Bits3
 132 }
 133
 134 TEST_F(SimdFloatingpointTest, gmxSimdAndnotR)
 135 {
 136     GMX_EXPECT_SIMD_REAL_EQ(rSimd_Bits4, gmx_simd_andnot_r(rSimd_Bits1, rSimd_Bits2)); // (~Bits1) & Bits2 = Bits3
 137 }
 138
 139 TEST_F(SimdFloatingpointTest, gmxSimdOrR)
 140 {
 141     GMX_EXPECT_SIMD_REAL_EQ(rSimd_Bits5, gmx_simd_or_r(rSimd_Bits1, rSimd_Bits2)); // Bits1 | Bits2 = Bits3
 142 }
 143
 144 TEST_F(SimdFloatingpointTest, gmxSimdXorR)
 145 {
 146     GMX_EXPECT_SIMD_REAL_EQ(rSimd_Bits6, gmx_simd_xor_r(rSimd_Bits1, rSimd_Bits2)); // Bits1 ^ Bits2 = Bits3
 147 }
 148 #endif
 149
 150 TEST_F(SimdFloatingpointTest, gmxSimdMaxR)
 151 {
 152     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(3, 2, 4), gmx_simd_max_r(rSimd_1_2_3, rSimd_3_1_4));
 153     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(3, 2, 4), gmx_simd_max_r(rSimd_3_1_4, rSimd_1_2_3));
 154     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(-1, -1, -3), gmx_simd_max_r(rSimd_m1_m2_m3, rSimd_m3_m1_m4));
 155     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(-1, -1, -3), gmx_simd_max_r(rSimd_m3_m1_m4, rSimd_m1_m2_m3));
 156 }
 157
 158 TEST_F(SimdFloatingpointTest, gmxSimdMinR)
 159 {
 160     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(1, 1, 3), gmx_simd_min_r(rSimd_1_2_3, rSimd_3_1_4));
 161     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(1, 1, 3), gmx_simd_min_r(rSimd_3_1_4, rSimd_1_2_3));
 162     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(-3, -2, -4), gmx_simd_min_r(rSimd_m1_m2_m3, rSimd_m3_m1_m4));
 163     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(-3, -2, -4), gmx_simd_min_r(rSimd_m3_m1_m4, rSimd_m1_m2_m3));
 164 }
 165
 166 TEST_F(SimdFloatingpointTest, gmxSimdRoundR)
 167 {
 168     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(2), gmx_simd_round_r(gmx_simd_set1_r(2.25)));
 169     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(4), gmx_simd_round_r(gmx_simd_set1_r(3.75)));
 170     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(-2), gmx_simd_round_r(gmx_simd_set1_r(-2.25)));
 171     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(-4), gmx_simd_round_r(gmx_simd_set1_r(-3.75)));
 172 }
 173
 174 TEST_F(SimdFloatingpointTest, gmxSimdTruncR)
 175 {
 176     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(2), gmx_simd_trunc_r(rSimd_2p25));
 177     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(3), gmx_simd_trunc_r(rSimd_3p75));
 178     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(-2), gmx_simd_trunc_r(rSimd_m2p25));
 179     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(-3), gmx_simd_trunc_r(rSimd_m3p75));
 180 }
 181
 182 TEST_F(SimdFloatingpointTest, gmxSimdFractionR)
 183 {
 184     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(0.25), gmx_simd_fraction_r(rSimd_2p25));   // fract(2.25)=0.25
 185     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(0.75), gmx_simd_fraction_r(rSimd_3p75));   // fract(3.75)=0.75
 186     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(-0.25), gmx_simd_fraction_r(rSimd_m2p25)); // fract(-2.25)=-0.25
 187     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom1R(-0.75), gmx_simd_fraction_r(rSimd_m3p75)); // fract(-3.75)=-0.75
 188 }
 189
 190 // We explicitly test the exponent/mantissa routines with double precision data,
 191 // since these usually rely on direct manipulation and shift of the SIMD registers,
 192 // where it is easy to make mistakes with single vs double precision.
 193
 194 TEST_F(SimdFloatingpointTest, gmxSimdGetExponentR)
 195 {
 196     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(60.0, -41.0, 54.0), gmx_simd_get_exponent_r(rSimd_Exp));
 197 #if (defined GMX_SIMD_HAVE_DOUBLE) && (defined GMX_DOUBLE)
 198     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(587.0, -462.0, 672.0), gmx_simd_get_exponent_r(rSimd_ExpDouble));
 199 #endif
 200 }
 201
 202 TEST_F(SimdFloatingpointTest, gmxSimdGetMantissaR)
 203 {
 204     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(1.219097320577810839026256,
 205                                               1.166738027848349235071623,
 206                                               1.168904015004464724825084), gmx_simd_get_mantissa_r(rSimd_Exp));
 207 #if (defined GMX_SIMD_HAVE_DOUBLE) && (defined GMX_DOUBLE)
 208     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(1.241261238952345623563251,
 209                                               1.047294723759123852359232,
 210                                               1.856066204750275957395734), gmx_simd_get_mantissa_r(rSimd_ExpDouble));
 211 #endif
 212 }
 213
 214 TEST_F(SimdFloatingpointTest, gmxSimdSetExponentR)
 215 {
 216     gmx_simd_real_t x0 = setSimdRealFrom3R(0.5, 11.5, 99.5);
 217     gmx_simd_real_t x1 = setSimdRealFrom3R(-0.5, -11.5, -99.5);
 218
 219     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(pow(2.0, 60.0), pow(2.0, -41.0), pow(2.0, 54.0)),
 220                             gmx_simd_set_exponent_r(setSimdRealFrom3R(60.0, -41.0, 54.0)));
 221 #if (defined GMX_SIMD_HAVE_DOUBLE) && (defined GMX_DOUBLE)
 222     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(pow(2.0, 587.0), pow(2.0, -462.0), pow(2.0, 672.0)),
 223                             gmx_simd_set_exponent_r(setSimdRealFrom3R(587.0, -462.0, 672.0)));
 224 #endif
 225     /* Rounding mode in gmx_simd_set_exponent_r() must be consistent with gmx_simd_round_r() */
 226     GMX_EXPECT_SIMD_REAL_EQ(gmx_simd_set_exponent_r(gmx_simd_round_r(x0)), gmx_simd_set_exponent_r(x0));
 227     GMX_EXPECT_SIMD_REAL_EQ(gmx_simd_set_exponent_r(gmx_simd_round_r(x1)), gmx_simd_set_exponent_r(x1));
 228 }
 229
 230 /*
 231  * We do extensive 1/sqrt(x) and 1/x accuracy testing in the math module, so
 232  * we just make sure the lookup instructions appear to work here
 233  */
 234
 235 TEST_F(SimdFloatingpointTest, gmxSimdRsqrtR)
 236 {
 237     gmx_simd_real_t x      = setSimdRealFrom3R(4.0, M_PI, 1234567890.0);
 238     gmx_simd_real_t ref    = setSimdRealFrom3R(0.5, 1.0/sqrt(M_PI), 1.0/sqrt(1234567890.0));
 239
 240     /* Set the allowed ulp error as 2 to the power of the number of bits in
 241      * the mantissa that do not have to be correct after the table lookup.
 242      */
 243     setUlpTol(1LL << (std::numeric_limits<real>::digits-GMX_SIMD_RSQRT_BITS));
 244
 245     GMX_EXPECT_SIMD_REAL_NEAR(ref, gmx_simd_rsqrt_r(x));
 246 }
 247
 248 TEST_F(SimdFloatingpointTest, gmxSimdRcpR)
 249 {
 250     gmx_simd_real_t x      = setSimdRealFrom3R(4.0, M_PI, 1234567890.0);
 251     gmx_simd_real_t ref    = setSimdRealFrom3R(0.25, 1.0/M_PI, 1.0/1234567890.0);
 252
 253     /* Set the allowed ulp error as 2 to the power of the number of bits in
 254      * the mantissa that do not have to be correct after the table lookup.
 255      */
 256     setUlpTol(1LL << (std::numeric_limits<real>::digits-GMX_SIMD_RCP_BITS));
 257
 258     GMX_EXPECT_SIMD_REAL_NEAR(ref, gmx_simd_rcp_r(x));
 259 }
 260
 261 TEST_F(SimdFloatingpointTest, gmxSimdBoolCmpEqAndBlendZeroR)
 262 {
 263     gmx_simd_bool_t eq   = gmx_simd_cmpeq_r(rSimd_5_7_9, rSimd_7_8_9);
 264     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(0, 0, 3), gmx_simd_blendzero_r(rSimd_1_2_3, eq));
 265 }
 266
 267 TEST_F(SimdFloatingpointTest, gmxSimdBlendNotZeroR)
 268 {
 269     gmx_simd_bool_t eq   = gmx_simd_cmpeq_r(rSimd_5_7_9, rSimd_7_8_9);
 270     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(1, 2, 0), gmx_simd_blendnotzero_r(rSimd_1_2_3, eq));
 271 }
 272
 273 TEST_F(SimdFloatingpointTest, gmxSimdBoolCmpLER)
 274 {
 275     gmx_simd_bool_t le   = gmx_simd_cmple_r(rSimd_5_7_9, rSimd_7_8_9);
 276     GMX_EXPECT_SIMD_REAL_EQ(rSimd_1_2_3, gmx_simd_blendzero_r(rSimd_1_2_3, le));
 277 }
 278
 279 TEST_F(SimdFloatingpointTest, gmxSimdBoolCmpLTR)
 280 {
 281     gmx_simd_bool_t lt   = gmx_simd_cmplt_r(rSimd_5_7_9, rSimd_7_8_9);
 282     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(1, 2, 0), gmx_simd_blendzero_r(rSimd_1_2_3, lt));
 283 }
 284
 285 TEST_F(SimdFloatingpointTest, gmxSimdBoolAndB)
 286 {
 287     gmx_simd_bool_t eq   = gmx_simd_cmpeq_r(rSimd_5_7_9, rSimd_7_8_9);
 288     gmx_simd_bool_t le   = gmx_simd_cmple_r(rSimd_5_7_9, rSimd_7_8_9);
 289     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(0, 0, 3), gmx_simd_blendzero_r(rSimd_1_2_3, gmx_simd_and_b(eq, le)));
 290 }
 291
 292 TEST_F(SimdFloatingpointTest, gmxSimdBoolOrB)
 293 {
 294     gmx_simd_bool_t eq   = gmx_simd_cmpeq_r(rSimd_5_7_9, rSimd_7_8_9);
 295     gmx_simd_bool_t lt   = gmx_simd_cmplt_r(rSimd_5_7_9, rSimd_7_8_9);
 296     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(1, 2, 3), gmx_simd_blendzero_r(rSimd_1_2_3, gmx_simd_or_b(eq, lt)));
 297 }
 298
 299 TEST_F(SimdFloatingpointTest, gmxSimdAnytrueB)
 300 {
 301     gmx_simd_bool_t eq;
 302
 303     /* this test is a bit tricky since we don't know the simd width.
 304      * We cannot check for truth values for "any" element beyond the first,
 305      * since that part of the data will not be used if simd width is 1.
 306      */
 307     eq = gmx_simd_cmpeq_r(rSimd_5_7_9, setSimdRealFrom3R(5, 0, 0));
 308     EXPECT_NE(0, gmx_simd_anytrue_b(eq));
 309
 310     eq = gmx_simd_cmpeq_r(rSimd_1_2_3, rSimd_4_5_6);
 311     EXPECT_EQ(0, gmx_simd_anytrue_b(eq));
 312 }
 313
 314 TEST_F(SimdFloatingpointTest, gmxSimdBlendvR)
 315 {
 316     gmx_simd_bool_t lt   = gmx_simd_cmplt_r(rSimd_5_7_9, rSimd_7_8_9);
 317     GMX_EXPECT_SIMD_REAL_EQ(setSimdRealFrom3R(4, 5, 3), gmx_simd_blendv_r(rSimd_1_2_3, rSimd_4_5_6, lt));
 318 }
 319
 320 TEST_F(SimdFloatingpointTest, gmxSimdReduceR)
 321 {
 322     // The horizontal sum of the SIMD variable depends on the width, so
 323     // simply store it an extra time and calculate what the sum should be
 324     std::vector<real> v   = simdReal2Vector(rSimd_4_5_6);
 325     real              sum = 0.0;
 326
 327     for (int i = 0; i < GMX_SIMD_REAL_WIDTH; i++)
 328     {
 329         sum += v[i];
 330     }
 331
 332     EXPECT_EQ(sum, gmx_simd_reduce_r(rSimd_4_5_6));
 333 }
 334
 335 #endif      // GMX_SIMD_HAVE_REAL
 336
 337 /*! \} */
 338 /*! \endcond */
 339
 340 }      // namespace
 341 }      // namespace
 342 }      // namespace