1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2014,2015,2016,2017,2018 by the GROMACS development team.
5 * Copyright (c) 2019,2020, by the GROMACS development team, led by
6 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
7 * and including many others, as listed in the AUTHORS file in the
8 * top-level source directory and at http://www.gromacs.org.
9 *
10 * GROMACS is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public License
12 * as published by the Free Software Foundation; either version 2.1
13 * of the License, or (at your option) any later version.
14 *
15 * GROMACS is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
19 *
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with GROMACS; if not, see
22 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
23 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 *
25 * If you want to redistribute modifications to GROMACS, please
26 * consider that scientific software is very special. Version
27 * control is crucial - bugs must be traceable. We will be happy to
28 * consider code for inclusion in the official distribution, but
29 * derived work must not be called official GROMACS. Details are found
30 * in the README & COPYING files - if they are missing, get the
31 * official version at http://www.gromacs.org.
32 *
33 * To help us fund GROMACS development, we humbly ask that you cite
34 * the research papers on the package. Check out http://www.gromacs.org.
35 */
36 #include "gmxpre.h"
37
38 #include <cmath>
39
40 #include "gromacs/math/utilities.h"
41 #include "gromacs/simd/simd.h"
42 #include "gromacs/utility/basedefinitions.h"
43
44 #include "testutils/testasserts.h"
45
46 #include "data.h"
47 #include "simd4.h"
48
49 #if GMX_SIMD
50
51 namespace gmx
52 {
53 namespace test
54 {
55 namespace
56 {
57
58 /*! \cond internal */
59 /*! \addtogroup module_simd */
60 /*! \{ */
61
62 # if GMX_SIMD4_HAVE_REAL
63
64 /*! \brief Test fixture for SIMD4 floating-point operations (identical to the SIMD4 \ref Simd4Test) */
65 typedef Simd4Test Simd4FloatingpointTest;
66
TEST_F(Simd4FloatingpointTest,setZero)67 TEST_F(Simd4FloatingpointTest, setZero)
68 {
69 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(0.0), setZero());
70 }
71
TEST_F(Simd4FloatingpointTest,set)72 TEST_F(Simd4FloatingpointTest, set)
73 {
74 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(c1), Simd4Real(c1));
75 }
76
TEST_F(Simd4FloatingpointTest,add)77 TEST_F(Simd4FloatingpointTest, add)
78 {
79 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(c0 + c3, c1 + c4, c2 + c5), rSimd4_c0c1c2 + rSimd4_c3c4c5);
80 }
81
TEST_F(Simd4FloatingpointTest,sub)82 TEST_F(Simd4FloatingpointTest, sub)
83 {
84 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(c0 - c3, c1 - c4, c2 - c5), rSimd4_c0c1c2 - rSimd4_c3c4c5);
85 }
86
TEST_F(Simd4FloatingpointTest,mul)87 TEST_F(Simd4FloatingpointTest, mul)
88 {
89 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(c0 * c3, c1 * c4, c2 * c5), rSimd4_c0c1c2 * rSimd4_c3c4c5);
90 }
91
TEST_F(Simd4FloatingpointTest,fma)92 TEST_F(Simd4FloatingpointTest, fma)
93 {
94 // The last bit of FMA operations depends on hardware, so we don't require exact match
95 GMX_EXPECT_SIMD4_REAL_NEAR(setSimd4RealFrom3R(c0 * c3 + c6, c1 * c4 + c7, c2 * c5 + c8),
96 fma(rSimd4_c0c1c2, rSimd4_c3c4c5, rSimd4_c6c7c8));
97 }
98
TEST_F(Simd4FloatingpointTest,fms)99 TEST_F(Simd4FloatingpointTest, fms)
100 {
101 // The last bit of FMA operations depends on hardware, so we don't require exact match
102 GMX_EXPECT_SIMD4_REAL_NEAR(setSimd4RealFrom3R(c0 * c3 - c6, c1 * c4 - c7, c2 * c5 - c8),
103 fms(rSimd4_c0c1c2, rSimd4_c3c4c5, rSimd4_c6c7c8));
104 }
105
TEST_F(Simd4FloatingpointTest,fnma)106 TEST_F(Simd4FloatingpointTest, fnma)
107 {
108 // The last bit of FMA operations depends on hardware, so we don't require exact match
109 GMX_EXPECT_SIMD4_REAL_NEAR(setSimd4RealFrom3R(c6 - c0 * c3, c7 - c1 * c4, c8 - c2 * c5),
110 fnma(rSimd4_c0c1c2, rSimd4_c3c4c5, rSimd4_c6c7c8));
111 }
112
TEST_F(Simd4FloatingpointTest,fnms)113 TEST_F(Simd4FloatingpointTest, fnms)
114 {
115 // The last bit of FMA operations depends on hardware, so we don't require exact match
116 GMX_EXPECT_SIMD4_REAL_NEAR(setSimd4RealFrom3R(-c0 * c3 - c6, -c1 * c4 - c7, -c2 * c5 - c8),
117 fnms(rSimd4_c0c1c2, rSimd4_c3c4c5, rSimd4_c6c7c8));
118 }
119
TEST_F(Simd4FloatingpointTest,abs)120 TEST_F(Simd4FloatingpointTest, abs)
121 {
122 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_c0c1c2, abs(rSimd4_c0c1c2)); // fabs(x)=x
123 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_c0c1c2, abs(rSimd4_m0m1m2)); // fabs(-x)=x
124 }
125
TEST_F(Simd4FloatingpointTest,neg)126 TEST_F(Simd4FloatingpointTest, neg)
127 {
128 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_m0m1m2, -(rSimd4_c0c1c2)); // fneg(x)=-x
129 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_c0c1c2, -(rSimd4_m0m1m2)); // fneg(-x)=x
130 }
131
132 # if GMX_SIMD_HAVE_LOGICAL
TEST_F(Simd4FloatingpointTest,and)133 TEST_F(Simd4FloatingpointTest, and)
134 {
135 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_logicalResultAnd, (rSimd4_logicalA & rSimd4_logicalB));
136 }
137
TEST_F(Simd4FloatingpointTest,or)138 TEST_F(Simd4FloatingpointTest, or)
139 {
140 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_logicalResultOr, (rSimd4_logicalA | rSimd4_logicalB));
141 }
142
TEST_F(Simd4FloatingpointTest,xor)143 TEST_F(Simd4FloatingpointTest, xor)
144 {
145 /* Test xor by taking xor with a number and its negative. This should result
146 * in only the sign bit being set. We then use this bit change the sign of
147 * different numbers.
148 */
149 Simd4Real signbit = Simd4Real(c1) ^ Simd4Real(-c1);
150 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(-c2, c3, -c4), signbit ^ setSimd4RealFrom3R(c2, -c3, c4));
151 }
152
TEST_F(Simd4FloatingpointTest,andNot)153 TEST_F(Simd4FloatingpointTest, andNot)
154 {
155 /* Use xor (which we already tested, so fix that first if both tests fail)
156 * to extract the sign bit, and then use andnot to take absolute values.
157 */
158 Simd4Real signbit = Simd4Real(c1) ^ Simd4Real(-c1);
159 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(c2, c3, c4),
160 andNot(signbit, setSimd4RealFrom3R(-c2, c3, -c4)));
161 }
162
163 # endif
164
TEST_F(Simd4FloatingpointTest,max)165 TEST_F(Simd4FloatingpointTest, max)
166 {
167 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(c3, c1, c4), max(rSimd4_c0c1c2, rSimd4_c3c0c4));
168 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(c3, c1, c4), max(rSimd4_c3c0c4, rSimd4_c0c1c2));
169 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(-c0, -c0, -c2), max(rSimd4_m0m1m2, rSimd4_m3m0m4));
170 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(-c0, -c0, -c2), max(rSimd4_m3m0m4, rSimd4_m0m1m2));
171 }
172
TEST_F(Simd4FloatingpointTest,min)173 TEST_F(Simd4FloatingpointTest, min)
174 {
175 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(c0, c0, c2), min(rSimd4_c0c1c2, rSimd4_c3c0c4));
176 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(c0, c0, c2), min(rSimd4_c3c0c4, rSimd4_c0c1c2));
177 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(-c3, -c1, -c4), min(rSimd4_m0m1m2, rSimd4_m3m0m4));
178 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(-c3, -c1, -c4), min(rSimd4_m3m0m4, rSimd4_m0m1m2));
179 }
180
TEST_F(Simd4FloatingpointTest,round)181 TEST_F(Simd4FloatingpointTest, round)
182 {
183 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(2), round(Simd4Real(2.25)));
184 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(4), round(Simd4Real(3.75)));
185 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(-2), round(Simd4Real(-2.25)));
186 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(-4), round(Simd4Real(-3.75)));
187 }
188
TEST_F(Simd4FloatingpointTest,trunc)189 TEST_F(Simd4FloatingpointTest, trunc)
190 {
191 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(2), trunc(rSimd4_2p25));
192 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(3), trunc(rSimd4_3p75));
193 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(-2), trunc(rSimd4_m2p25));
194 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom1R(-3), trunc(rSimd4_m3p75));
195 }
196
197 /* We do extensive 1/sqrt(x) and 1/x accuracy testing in the tests for
198 * the SIMD math functions, so we just make sure the lookup instructions
199 * appear to work for a few values here.
200 */
TEST_F(Simd4FloatingpointTest,gmxSimd4RsqrtR)201 TEST_F(Simd4FloatingpointTest, gmxSimd4RsqrtR)
202 {
203 Simd4Real x = setSimd4RealFrom3R(4.0, M_PI, 1234567890.0);
204 Simd4Real ref = setSimd4RealFrom3R(0.5, 1.0 / std::sqrt(M_PI), 1.0 / std::sqrt(1234567890.0));
205 int shiftbits = std::numeric_limits<real>::digits - GMX_SIMD_RSQRT_BITS;
206
207 if (shiftbits < 0)
208 {
209 shiftbits = 0;
210 }
211
212 // The allowed Ulp deviation is 2 to the power of the number of mantissa
213 // digits, minus the number of bits provided by the table lookup
214 setUlpTol(1LL << shiftbits);
215 GMX_EXPECT_SIMD4_REAL_NEAR(ref, rsqrt(x));
216 }
217
TEST_F(Simd4FloatingpointTest,cmpEqAndSelectByMask)218 TEST_F(Simd4FloatingpointTest, cmpEqAndSelectByMask)
219 {
220 Simd4Bool eq = rSimd4_c4c6c8 == rSimd4_c6c7c8;
221 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(0, 0, c2), selectByMask(rSimd4_c0c1c2, eq));
222 }
223
TEST_F(Simd4FloatingpointTest,selectByNotMask)224 TEST_F(Simd4FloatingpointTest, selectByNotMask)
225 {
226 Simd4Bool eq = rSimd4_c4c6c8 == rSimd4_c6c7c8;
227 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(c0, c1, 0), selectByNotMask(rSimd4_c0c1c2, eq));
228 }
229
TEST_F(Simd4FloatingpointTest,cmpNe)230 TEST_F(Simd4FloatingpointTest, cmpNe)
231 {
232 Simd4Bool eq = rSimd4_c4c6c8 != rSimd4_c6c7c8;
233 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(c0, c1, 0), selectByMask(rSimd4_c0c1c2, eq));
234 }
235
TEST_F(Simd4FloatingpointTest,cmpLe)236 TEST_F(Simd4FloatingpointTest, cmpLe)
237 {
238 Simd4Bool le = rSimd4_c4c6c8 <= rSimd4_c6c7c8;
239 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_c0c1c2, selectByMask(rSimd4_c0c1c2, le));
240 }
241
TEST_F(Simd4FloatingpointTest,cmpLt)242 TEST_F(Simd4FloatingpointTest, cmpLt)
243 {
244 Simd4Bool lt = rSimd4_c4c6c8 < rSimd4_c6c7c8;
245 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(c0, c1, 0), selectByMask(rSimd4_c0c1c2, lt));
246 }
247
TEST_F(Simd4FloatingpointTest,andB)248 TEST_F(Simd4FloatingpointTest, andB)
249 {
250 Simd4Bool eq = rSimd4_c4c6c8 == rSimd4_c6c7c8;
251 Simd4Bool le = rSimd4_c4c6c8 <= rSimd4_c6c7c8;
252 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(0, 0, c2), selectByMask(rSimd4_c0c1c2, (eq && le)));
253 }
254
TEST_F(Simd4FloatingpointTest,orB)255 TEST_F(Simd4FloatingpointTest, orB)
256 {
257 Simd4Bool eq = rSimd4_c4c6c8 == rSimd4_c6c7c8;
258 Simd4Bool lt = rSimd4_c4c6c8 < rSimd4_c6c7c8;
259 GMX_EXPECT_SIMD4_REAL_EQ(rSimd4_c0c1c2, selectByMask(rSimd4_c0c1c2, (eq || lt)));
260 }
261
TEST_F(Simd4FloatingpointTest,anyTrue)262 TEST_F(Simd4FloatingpointTest, anyTrue)
263 {
264 Simd4Bool eq;
265
266 /* this test is a bit tricky since we don't know the simd width.
267 * We cannot check for truth values for "any" element beyond the first,
268 * since that part of the data will not be used if simd width is 1.
269 */
270 eq = (rSimd4_c4c6c8 == setSimd4RealFrom3R(c4, 0, 0));
271 EXPECT_TRUE(anyTrue(eq));
272
273 eq = (rSimd4_c0c1c2 == rSimd4_c3c4c5);
274 EXPECT_FALSE(anyTrue(eq));
275 }
276
TEST_F(Simd4FloatingpointTest,blend)277 TEST_F(Simd4FloatingpointTest, blend)
278 {
279 Simd4Bool lt = rSimd4_c4c6c8 < rSimd4_c6c7c8;
280 GMX_EXPECT_SIMD4_REAL_EQ(setSimd4RealFrom3R(c3, c4, c2), blend(rSimd4_c0c1c2, rSimd4_c3c4c5, lt));
281 }
282
TEST_F(Simd4FloatingpointTest,reduce)283 TEST_F(Simd4FloatingpointTest, reduce)
284 {
285 // The horizontal sum of the SIMD variable depends on the width, so
286 // simply store it an extra time and calculate what the sum should be
287 std::vector<real> v = simd4Real2Vector(rSimd4_c3c4c5);
288 real sum = 0.0;
289
290 for (int i = 0; i < GMX_SIMD4_WIDTH; i++)
291 {
292 sum += v[i];
293 }
294
295 EXPECT_REAL_EQ_TOL(sum, reduce(rSimd4_c3c4c5), defaultRealTolerance());
296 }
297
298
TEST_F(Simd4FloatingpointTest,dotProduct)299 TEST_F(Simd4FloatingpointTest, dotProduct)
300 {
301 real res = c0 * c3 + c1 * c4 + c2 * c5;
302
303 EXPECT_REAL_EQ_TOL(res, dotProduct(rSimd4_c0c1c2, rSimd4_c3c4c5), defaultRealTolerance());
304 }
305
TEST_F(Simd4FloatingpointTest,transpose)306 TEST_F(Simd4FloatingpointTest, transpose)
307 {
308 Simd4Real v0, v1, v2, v3;
309 int i;
310 // aligned pointers
311 alignas(GMX_SIMD_ALIGNMENT) real p0[4 * GMX_SIMD4_WIDTH];
312 real* p1 = p0 + GMX_SIMD4_WIDTH;
313 real* p2 = p0 + 2 * GMX_SIMD4_WIDTH;
314 real* p3 = p0 + 3 * GMX_SIMD4_WIDTH;
315
316 // Assign data with tens as row, single-digit as column
317 for (i = 0; i < 4; i++)
318 {
319 // Scale by 1+100*eps to use low bits tii
320 p0[i] = (0 * 10 + i * 1) * (1.0 + 100 * GMX_REAL_EPS);
321 p1[i] = (1 * 10 + i * 1) * (1.0 + 100 * GMX_REAL_EPS);
322 p2[i] = (2 * 10 + i * 1) * (1.0 + 100 * GMX_REAL_EPS);
323 p3[i] = (3 * 10 + i * 1) * (1.0 + 100 * GMX_REAL_EPS);
324 }
325
326 v0 = load4(p0);
327 v1 = load4(p1);
328 v2 = load4(p2);
329 v3 = load4(p3);
330
331 transpose(&v0, &v1, &v2, &v3);
332
333 store4(p0, v0);
334 store4(p1, v1);
335 store4(p2, v2);
336 store4(p3, v3);
337
338 for (i = 0; i < 4; i++)
339 {
340 EXPECT_REAL_EQ_TOL((i * 10 + 0) * (1.0 + 100 * GMX_REAL_EPS), p0[i], defaultRealTolerance());
341 EXPECT_REAL_EQ_TOL((i * 10 + 1) * (1.0 + 100 * GMX_REAL_EPS), p1[i], defaultRealTolerance());
342 EXPECT_REAL_EQ_TOL((i * 10 + 2) * (1.0 + 100 * GMX_REAL_EPS), p2[i], defaultRealTolerance());
343 EXPECT_REAL_EQ_TOL((i * 10 + 3) * (1.0 + 100 * GMX_REAL_EPS), p3[i], defaultRealTolerance());
344 }
345 }
346
347 # endif // GMX_SIMD4_HAVE_REAL
348
349 /*! \} */
350 /*! \endcond */
351
352 } // namespace
353 } // namespace test
354 } // namespace gmx
355
356 #endif // GMX_SIMD
357