2 * This file is part of the GROMACS molecular simulation package.
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.
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.
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.
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.
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.
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.
39 #include "gromacs/math/utilities.h"
51 /*! \addtogroup module_simd */
54 #ifdef GMX_SIMD4_HAVE_REAL
56 /*! \brief Test fixture for SIMD4 floating-point operations (identical to the SIMD4 \ref Simd4Test) */
57 typedef Simd4Test Simd4FloatingpointTest;
59 TEST_F(Simd4FloatingpointTest, gmxSimd4SetZeroR)
61 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(0.0), gmx_simd4_setzero_r());
64 TEST_F(Simd4FloatingpointTest, gmxSimd4Set1R)
66 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(1.0), gmx_simd4_set1_r(1.0));
69 TEST_F(Simd4FloatingpointTest, gmxSimd4Load1R)
72 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(r), gmx_simd4_load1_r(&r));
75 TEST_F(Simd4FloatingpointTest, gmxSimd4AddR)
77 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_5_7_9, gmx_simd4_add_r(rSimd4_1_2_3, rSimd4_4_5_6)); // 1+4=5, 2+5=7, 3+6=9
80 TEST_F(Simd4FloatingpointTest, gmxSimd4SubR)
82 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_4_5_6, gmx_simd4_sub_r(rSimd4_5_7_9, rSimd4_1_2_3)); // 5-1=4, 7-2=5, 9-3=6
85 TEST_F(Simd4FloatingpointTest, gmxSimd4MulR)
87 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(4, 10, 18), gmx_simd4_mul_r(rSimd4_1_2_3, rSimd4_4_5_6));
90 TEST_F(Simd4FloatingpointTest, gmxSimd4FmaddR)
92 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(11, 18, 27), gmx_simd4_fmadd_r(rSimd4_1_2_3, rSimd4_4_5_6, rSimd4_7_8_9)); // 1*4+7, etc.
95 TEST_F(Simd4FloatingpointTest, gmxSimd4FmsubR)
97 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(-3, 2, 9), gmx_simd4_fmsub_r(rSimd4_1_2_3, rSimd4_4_5_6, rSimd4_7_8_9)); // 1*4-7, etc.
100 TEST_F(Simd4FloatingpointTest, gmxSimd4FnmaddR)
102 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(3, -2, -9), gmx_simd4_fnmadd_r(rSimd4_1_2_3, rSimd4_4_5_6, rSimd4_7_8_9)); // -1*4+7, etc.
105 TEST_F(Simd4FloatingpointTest, gmxSimd4FnmsubR)
107 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(-11, -18, -27), gmx_simd4_fnmsub_r(rSimd4_1_2_3, rSimd4_4_5_6, rSimd4_7_8_9)); // -1*4-7, etc.
110 TEST_F(Simd4FloatingpointTest, gmxSimd4FabsR)
112 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_1_2_3, gmx_simd4_fabs_r(rSimd4_1_2_3)); // fabs(x)=x
113 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_1_2_3, gmx_simd4_fabs_r(rSimd4_m1_m2_m3)); // fabs(-x)=x
116 TEST_F(Simd4FloatingpointTest, gmxSimd4FnegR)
118 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_m1_m2_m3, gmx_simd4_fneg_r(rSimd4_1_2_3)); // fneg(x)=-x
119 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_1_2_3, gmx_simd4_fneg_r(rSimd4_m1_m2_m3)); // fneg(-x)=x
122 #ifdef GMX_SIMD4_HAVE_LOGICAL
123 /* 1.3333282470703125 has mantissa 0101010101010101 (followed by zeros)
124 * 1.79998779296875 has mantissa 1100110011001100 (followed by zeros)
125 * 1.26666259765625 has mantissa 0100010001000100 (followed by zeros)
126 * 1.8666534423828125 has mantissa 1101110111011101 (followed by zeros)
128 * Since all of them have the same exponent (2^0), the exponent will
129 * not change with AND or OR operations.
131 TEST_F(Simd4FloatingpointTest, gmxSimd4AndR)
133 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(1.26666259765625),
134 gmx_simd4_and_r(gmx_simd4_set1_r(1.3333282470703125),
135 gmx_simd4_set1_r(1.79998779296875)));
138 TEST_F(Simd4FloatingpointTest, gmxSimd4OrR)
140 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(1.8666534423828125),
141 gmx_simd4_or_r(gmx_simd4_set1_r(1.3333282470703125),
142 gmx_simd4_set1_r(1.79998779296875)));
145 TEST_F(Simd4FloatingpointTest, gmxSimd4XorR)
147 /* Test xor by taking xor with a number and its negative. This should result
148 * in only the sign bit being set. We then use this bit change the sign of
151 gmx_simd4_real_t signbit = gmx_simd4_xor_r(gmx_simd4_set1_r(1.5), gmx_simd4_set1_r(-1.5));
152 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(-1, 2, -3), gmx_simd4_xor_r(signbit, setSimd4RealFrom3R(1, -2, 3)));
155 TEST_F(Simd4FloatingpointTest, gmxSimd4AndnotR)
157 /* Use xor (which we already tested, so fix that first if both tests fail)
158 * to extract the sign bit, and then use andnot to take absolute values.
160 gmx_simd4_real_t signbit = gmx_simd4_xor_r(gmx_simd4_set1_r(1.5), gmx_simd4_set1_r(-1.5));
161 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(1, 2, 3), gmx_simd4_andnot_r(signbit, setSimd4RealFrom3R(-1, 2, -3)));
166 TEST_F(Simd4FloatingpointTest, gmxSimd4MaxR)
168 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(3, 2, 4), gmx_simd4_max_r(rSimd4_1_2_3, rSimd4_3_1_4));
169 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(3, 2, 4), gmx_simd4_max_r(rSimd4_3_1_4, rSimd4_1_2_3));
170 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(-1, -1, -3), gmx_simd4_max_r(rSimd4_m1_m2_m3, rSimd4_m3_m1_m4));
171 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(-1, -1, -3), gmx_simd4_max_r(rSimd4_m3_m1_m4, rSimd4_m1_m2_m3));
174 TEST_F(Simd4FloatingpointTest, gmxSimd4MinR)
176 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(1, 1, 3), gmx_simd4_min_r(rSimd4_1_2_3, rSimd4_3_1_4));
177 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(1, 1, 3), gmx_simd4_min_r(rSimd4_3_1_4, rSimd4_1_2_3));
178 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(-3, -2, -4), gmx_simd4_min_r(rSimd4_m1_m2_m3, rSimd4_m3_m1_m4));
179 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(-3, -2, -4), gmx_simd4_min_r(rSimd4_m3_m1_m4, rSimd4_m1_m2_m3));
182 TEST_F(Simd4FloatingpointTest, gmxSimd4RoundR)
184 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(2), gmx_simd4_round_r(gmx_simd4_set1_r(2.25)));
185 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(4), gmx_simd4_round_r(gmx_simd4_set1_r(3.75)));
186 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(-2), gmx_simd4_round_r(gmx_simd4_set1_r(-2.25)));
187 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(-4), gmx_simd4_round_r(gmx_simd4_set1_r(-3.75)));
190 TEST_F(Simd4FloatingpointTest, gmxSimd4TruncR)
192 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(2), gmx_simd4_trunc_r(rSimd4_2p25));
193 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(3), gmx_simd4_trunc_r(rSimd4_3p75));
194 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(-2), gmx_simd4_trunc_r(rSimd4_m2p25));
195 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(-3), gmx_simd4_trunc_r(rSimd4_m3p75));
198 /* We do extensive 1/sqrt(x) and 1/x accuracy testing in the tests for
199 * the SIMD math functions, so we just make sure the lookup instructions
200 * appear to work for a few values here.
202 TEST_F(Simd4FloatingpointTest, gmxSimd4RsqrtR)
204 gmx_simd4_real_t x = setSimd4RealFrom3R(4.0, M_PI, 1234567890.0);
205 gmx_simd4_real_t ref = setSimd4RealFrom3R(0.5, 1.0/sqrt(M_PI), 1.0/sqrt(1234567890.0));
207 // The allowed Ulp deviation is 2 to the power of the number of mantissa
208 // digits, minus the number of bits provided by the table lookup
209 setUlpTol(1LL << (std::numeric_limits<real>::digits-GMX_SIMD_RSQRT_BITS));
210 GMX_EXPECT_SIMD4_REAL_NEAR(ref, gmx_simd4_rsqrt_r(x));
213 TEST_F(Simd4FloatingpointTest, gmxSimd4BoolCmpEqAndBlendZeroR)
215 gmx_simd4_bool_t eq = gmx_simd4_cmpeq_r(rSimd4_5_7_9, rSimd4_7_8_9);
216 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(0, 0, 3), gmx_simd4_blendzero_r(rSimd4_1_2_3, eq));
219 TEST_F(Simd4FloatingpointTest, gmxSimd4BlendNotZeroR)
221 gmx_simd4_bool_t eq = gmx_simd4_cmpeq_r(rSimd4_5_7_9, rSimd4_7_8_9);
222 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(1, 2, 0), gmx_simd4_blendnotzero_r(rSimd4_1_2_3, eq));
225 TEST_F(Simd4FloatingpointTest, gmxSimd4BoolCmpLER)
227 gmx_simd4_bool_t le = gmx_simd4_cmple_r(rSimd4_5_7_9, rSimd4_7_8_9);
228 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_1_2_3, gmx_simd4_blendzero_r(rSimd4_1_2_3, le));
231 TEST_F(Simd4FloatingpointTest, gmxSimd4BoolCmpLTR)
233 gmx_simd4_bool_t lt = gmx_simd4_cmplt_r(rSimd4_5_7_9, rSimd4_7_8_9);
234 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(1, 2, 0), gmx_simd4_blendzero_r(rSimd4_1_2_3, lt));
237 TEST_F(Simd4FloatingpointTest, gmxSimd4BoolAndB)
239 gmx_simd4_bool_t eq = gmx_simd4_cmpeq_r(rSimd4_5_7_9, rSimd4_7_8_9);
240 gmx_simd4_bool_t le = gmx_simd4_cmple_r(rSimd4_5_7_9, rSimd4_7_8_9);
241 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(0, 0, 3), gmx_simd4_blendzero_r(rSimd4_1_2_3, gmx_simd4_and_b(eq, le)));
244 TEST_F(Simd4FloatingpointTest, gmxSimd4BoolOrB)
246 gmx_simd4_bool_t eq = gmx_simd4_cmpeq_r(rSimd4_5_7_9, rSimd4_7_8_9);
247 gmx_simd4_bool_t lt = gmx_simd4_cmplt_r(rSimd4_5_7_9, rSimd4_7_8_9);
248 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(1, 2, 3), gmx_simd4_blendzero_r(rSimd4_1_2_3, gmx_simd4_or_b(eq, lt)));
251 TEST_F(Simd4FloatingpointTest, gmxSimd4AnytrueB)
255 /* this test is a bit tricky since we don't know the simd width.
256 * We cannot check for truth values for "any" element beyond the first,
257 * since that part of the data will not be used if simd width is 1.
259 eq = gmx_simd4_cmpeq_r(rSimd4_5_7_9, setSimd4RealFrom3R(5, 0, 0));
260 EXPECT_NE(0, gmx_simd4_anytrue_b(eq));
262 eq = gmx_simd4_cmpeq_r(rSimd4_1_2_3, rSimd4_4_5_6);
263 EXPECT_EQ(0, gmx_simd4_anytrue_b(eq));
266 TEST_F(Simd4FloatingpointTest, gmxSimd4BlendvR)
268 gmx_simd4_bool_t lt = gmx_simd4_cmplt_r(rSimd4_5_7_9, rSimd4_7_8_9);
269 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(4, 5, 3), gmx_simd4_blendv_r(rSimd4_1_2_3, rSimd4_4_5_6, lt));
272 TEST_F(Simd4FloatingpointTest, gmxSimd4ReduceR)
274 // The horizontal sum of the SIMD variable depends on the width, so
275 // simply store it an extra time and calculate what the sum should be
276 std::vector<real> v = simd4Real2Vector(rSimd4_1_2_3);
279 for (int i = 0; i < GMX_SIMD4_WIDTH; i++)
284 EXPECT_EQ(sum, gmx_simd4_reduce_r(rSimd4_1_2_3));
288 TEST_F(Simd4FloatingpointTest, gmxSimd4Dotproduct3R)
290 gmx_simd4_real_t v1 = setSimd4RealFrom3R(1, 4, 5);
291 gmx_simd4_real_t v2 = setSimd4RealFrom3R(3, 8, 2);
293 EXPECT_DOUBLE_EQ(45.0, gmx_simd4_dotproduct3_r(v1, v2));
295 EXPECT_FLOAT_EQ(45.0, gmx_simd4_dotproduct3_r(v1, v2));
299 #endif // GMX_SIMD4_HAVE_REAL