2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, 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.
35 /*! \libinternal \file
37 * Declares OpenMP wrappers to avoid conditional compilation.
39 * This module defines wrappers for OpenMP API functions and enables compiling
40 * code without conditional compilation even when OpenMP is turned off in the
42 * Therefore, OpenMP API functions should always be used through these wrappers
43 * and omp.h should never be directly included. Instead, this header should be
44 * used whenever OpenMP API functions are needed.
47 * \ingroup module_utility
49 #ifndef GMX_UTILITY_OMP_H
50 #define GMX_UTILITY_OMP_H
58 #ifndef GMX_NATIVE_WINDOWS
59 /* Ugly hack because the openmp implementation below hacks into the SIMD
60 * settings to decide when to use _mm_pause(). This should eventually be
61 * changed into proper detection of the intrinsics uses, not SIMD.
63 #if (defined GMX_SIMD_X86_SSE2) || (defined GMX_SIMD_X86_SSE4_1) || \
64 (defined GMX_SIMD_X86_AVX_128_FMA) || (defined GMX_SIMD_X86_AVX_256) || \
65 (defined GMX_SIMD_X86_AVX2_256)
66 # include <xmmintrin.h>
72 #include "basedefinitions.h"
79 /*! \addtogroup module_utility
84 * Returns an integer equal to or greater than the number of threads
85 * that would be available if a parallel region without num_threads were
86 * defined at that point in the code.
88 * Acts as a wrapper for omp_get_max_threads().
90 int gmx_omp_get_max_threads(void);
93 * Returns the number of processors available when the function is called.
95 * Acts as a wrapper around omp_get_num_procs().
97 int gmx_omp_get_num_procs(void);
100 * Returns the thread number of the thread executing within its thread team.
102 * Acts as a wrapper for omp_get_thread_num().
104 int gmx_omp_get_thread_num(void);
107 * Sets the number of threads in subsequent parallel regions, unless overridden
108 * by a num_threads clause.
110 * Acts as a wrapper for omp_set_num_threads().
112 void gmx_omp_set_num_threads(int num_threads);
115 * Check for externally set thread affinity to avoid conflicts with \Gromacs
118 * \param[out] message Receives the message to be shown to the user.
119 * \returns `true` if we can set thread affinity ourselves.
121 * While GNU OpenMP does not set affinity by default, the Intel OpenMP library
122 * does. This conflicts with the internal affinity (especially thread-MPI)
123 * setting, results in incorrectly locked threads, and causes dreadful performance.
125 * The KMP_AFFINITY environment variable is used by Intel, GOMP_CPU_AFFINITY
126 * by the GNU compilers (Intel also honors it well). If any of the variables
127 * is set, we should honor it and disable the internal pinning.
128 * When using Intel OpenMP, we will disable affinity if the user did not set it
129 * manually through one of the aforementioned environment variables.
131 * Note that the Intel OpenMP affinity disabling will only take effect if this
132 * function is called before the OpenMP library gets initialized, which happens
133 * when the first call is made into a compilation unit that contains OpenMP
136 * If this function returns `false`, the caller is responsible to disable the
137 * pinning, show the message from \p *message to the user, and free the memory
138 * allocated for \p *message.
139 * If the return value is `true`, \p *message is NULL.
141 gmx_bool gmx_omp_check_thread_affinity(char **message);
144 * Pause for use in a spin-wait loop.
146 static gmx_inline void gmx_pause()
148 #ifndef GMX_NATIVE_WINDOWS
149 /* Ugly hack because the openmp implementation below hacks into the SIMD
150 * settings to decide when to use _mm_pause(). This should eventually be
151 * changed into proper detection of the intrinsics uses, not SIMD.
153 #if (defined GMX_SIMD_X86_SSE2) || (defined GMX_SIMD_X86_SSE4_1) || \
154 (defined GMX_SIMD_X86_AVX_128_FMA) || (defined GMX_SIMD_X86_AVX_256) || \
155 (defined GMX_SIMD_X86_AVX2_256)
156 /* Replace with tbb::internal::atomic_backoff when/if we use TBB */
158 #elif defined __MIC__
161 /* No wait for unknown architecture */