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.
45 /*! \addtogroup module_simd */
48 /* Unfortunately we cannot keep static SIMD constants in the test fixture class.
49 * The problem is that SIMD memory need to be aligned, and in particular
50 * this applies to automatic storage of variables in classes. For SSE registers
51 * this means 16-byte alignment (which seems to work), but AVX requires 32-bit
52 * alignment. At least both gcc-4.7.3 and Apple clang-5.0 (OS X 10.9) fail to
53 * align these variables when they are stored as data in a class.
55 * In theory we could set some of these on-the-fly e.g. with setSimdRealFrom3R()
56 * instead (although that would mean repeating code between tests), but many of
57 * the constants depend on the current precision not to mention they
58 * occasionally have many digits that need to be exactly right, and keeping
59 * them in a single place makes sure they are consistent.
61 #ifdef GMX_SIMD_HAVE_REAL
62 const gmx_simd_real_t rSimd_1_2_3 = setSimdRealFrom3R(1, 2, 3);
63 const gmx_simd_real_t rSimd_4_5_6 = setSimdRealFrom3R(4, 5, 6);
64 const gmx_simd_real_t rSimd_7_8_9 = setSimdRealFrom3R(7, 8, 9);
65 const gmx_simd_real_t rSimd_5_7_9 = setSimdRealFrom3R(5, 7, 9);
66 const gmx_simd_real_t rSimd_m1_m2_m3 = setSimdRealFrom3R(-1, -2, -3);
67 const gmx_simd_real_t rSimd_3_1_4 = setSimdRealFrom3R(3, 1, 4);
68 const gmx_simd_real_t rSimd_m3_m1_m4 = setSimdRealFrom3R(-3, -1, -4);
69 const gmx_simd_real_t rSimd_2p25 = setSimdRealFrom1R(2.25);
70 const gmx_simd_real_t rSimd_3p75 = setSimdRealFrom1R(3.75);
71 const gmx_simd_real_t rSimd_m2p25 = setSimdRealFrom1R(-2.25);
72 const gmx_simd_real_t rSimd_m3p75 = setSimdRealFrom1R(-3.75);
73 const gmx_simd_real_t rSimd_Exp = setSimdRealFrom3R( 1.4055235171027452623914516e+18,
74 5.3057102734253445623914516e-13,
75 -2.1057102745623934534514516e+16);
76 # if (defined GMX_SIMD_HAVE_DOUBLE) && (defined GMX_DOUBLE)
77 // Make sure we also test exponents outside single precision when we use double
78 const gmx_simd_real_t rSimd_ExpDouble = setSimdRealFrom3R( 6.287393598732017379054414e+176,
79 8.794495252903116023030553e-140,
80 -3.637060701570496477655022e+202);
81 // Magic FP numbers corresponding to specific bit patterns
82 const gmx_simd_real_t rSimd_Bits1 = setSimdRealFrom1R(-1.07730874267432137e+236);
83 const gmx_simd_real_t rSimd_Bits2 = setSimdRealFrom1R(-9.25596313493178307e+061);
84 const gmx_simd_real_t rSimd_Bits3 = setSimdRealFrom1R(-8.57750588235293981e+003);
85 const gmx_simd_real_t rSimd_Bits4 = setSimdRealFrom1R( 1.22416778341839096e-250);
86 const gmx_simd_real_t rSimd_Bits5 = setSimdRealFrom1R(-1.15711777004554095e+294);
87 const gmx_simd_real_t rSimd_Bits6 = setSimdRealFrom1R( 1.53063836115600621e-018);
89 // Magic FP numbers corresponding to specific bit patterns
90 const gmx_simd_real_t rSimd_Bits1 = setSimdRealFrom1R(-5.9654142337e+29);
91 const gmx_simd_real_t rSimd_Bits2 = setSimdRealFrom1R(-1.0737417600e+08);
92 const gmx_simd_real_t rSimd_Bits3 = setSimdRealFrom1R(-6.0235290527e+00);
93 const gmx_simd_real_t rSimd_Bits4 = setSimdRealFrom1R( 1.0788832913e-31);
94 const gmx_simd_real_t rSimd_Bits5 = setSimdRealFrom1R(-1.0508719529e+37);
95 const gmx_simd_real_t rSimd_Bits6 = setSimdRealFrom1R( 1.1488970369e-02);
97 #endif // GMX_SIMD_HAVE_REAL
98 #ifdef GMX_SIMD_HAVE_INT32
99 const gmx_simd_int32_t iSimd_1_2_3 = setSimdIntFrom3I(1, 2, 3);
100 const gmx_simd_int32_t iSimd_4_5_6 = setSimdIntFrom3I(4, 5, 6);
101 const gmx_simd_int32_t iSimd_7_8_9 = setSimdIntFrom3I(7, 8, 9);
102 const gmx_simd_int32_t iSimd_5_7_9 = setSimdIntFrom3I(5, 7, 9);
103 const gmx_simd_int32_t iSimd_1M_2M_3M = setSimdIntFrom3I(1000000, 2000000, 3000000);
104 const gmx_simd_int32_t iSimd_4M_5M_6M = setSimdIntFrom3I(4000000, 5000000, 6000000);
105 const gmx_simd_int32_t iSimd_5M_7M_9M = setSimdIntFrom3I(5000000, 7000000, 9000000);
106 const gmx_simd_int32_t iSimd_0xF0F0F0F0 = setSimdIntFrom1I(0xF0F0F0F0);
107 const gmx_simd_int32_t iSimd_0xCCCCCCCC = setSimdIntFrom1I(0xCCCCCCCC);
108 #endif // GMX_SIMD_HAVE_INT32
110 #ifdef GMX_SIMD_HAVE_REAL
112 simdReal2Vector(const gmx_simd_real_t simd)
114 real mem[GMX_SIMD_REAL_WIDTH*2];
115 real * p = gmx_simd_align_r(mem);
117 gmx_simd_store_r(p, simd);
118 std::vector<real> v(p, p+GMX_SIMD_REAL_WIDTH);
124 vector2SimdReal(const std::vector<real> &v)
126 real mem[GMX_SIMD_REAL_WIDTH*2];
127 real * p = gmx_simd_align_r(mem);
129 for (int i = 0; i < GMX_SIMD_REAL_WIDTH; i++)
131 p[i] = v[i % v.size()]; // repeat vector contents to fill simd width
133 return gmx_simd_load_r(p);
137 setSimdRealFrom3R(real r0, real r1, real r2)
139 std::vector<real> v(3);
143 return vector2SimdReal(v);
147 setSimdRealFrom1R(real value)
149 std::vector<real> v(GMX_SIMD_REAL_WIDTH);
150 for (int i = 0; i < GMX_SIMD_REAL_WIDTH; i++)
154 return vector2SimdReal(v);
157 testing::AssertionResult
158 SimdTest::compareSimdRealUlp(const char * refExpr, const char * tstExpr,
159 const gmx_simd_real_t ref, const gmx_simd_real_t tst)
161 return compareVectorRealUlp(refExpr, tstExpr, simdReal2Vector(ref), simdReal2Vector(tst));
164 testing::AssertionResult
165 SimdTest::compareSimdRealEq(const char * refExpr, const char * tstExpr,
166 const gmx_simd_real_t ref, const gmx_simd_real_t tst)
168 return compareVectorEq(refExpr, tstExpr, simdReal2Vector(ref), simdReal2Vector(tst));
171 #endif // GMX_SIMD_HAVE_REAL
173 #ifdef GMX_SIMD_HAVE_INT32
175 simdInt2Vector(const gmx_simd_int32_t simd)
177 int mem[GMX_SIMD_INT32_WIDTH*2];
178 int * p = gmx_simd_align_i(mem);
180 gmx_simd_store_i(p, simd);
181 std::vector<int> v(p, p+GMX_SIMD_INT32_WIDTH);
187 vector2SimdInt(const std::vector<int> &v)
189 int mem[GMX_SIMD_INT32_WIDTH*2];
190 int * p = gmx_simd_align_i(mem);
192 for (int i = 0; i < GMX_SIMD_INT32_WIDTH; i++)
194 p[i] = v[i % v.size()]; // repeat vector contents to fill simd width
196 return gmx_simd_load_i(p);
200 setSimdIntFrom3I(int i0, int i1, int i2)
202 std::vector<int> v(3);
206 return vector2SimdInt(v);
210 setSimdIntFrom1I(int value)
212 std::vector<int> v(GMX_SIMD_INT32_WIDTH);
213 for (int i = 0; i < GMX_SIMD_INT32_WIDTH; i++)
217 return vector2SimdInt(v);
220 ::testing::AssertionResult
221 SimdTest::compareSimdInt32(const char * refExpr, const char * tstExpr,
222 const gmx_simd_int32_t ref, const gmx_simd_int32_t tst)
224 return compareVectorEq(refExpr, tstExpr, simdInt2Vector(ref), simdInt2Vector(tst));
227 #endif // GMX_SIMD_HAVE_INT32