2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2017,2018,2019, 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
36 * \brief Declares functionality for deciding whether tasks will run on GPUs.
38 * \author Mark Abraham <mark.j.abraham@gmail.com>
39 * \ingroup module_taskassignment
43 #ifndef GMX_TASKASSIGNMENT_DECIDEGPUUSAGE_H
44 #define GMX_TASKASSIGNMENT_DECIDEGPUUSAGE_H
55 //! Record where a compute task is targetted.
56 enum class TaskTarget : int
63 //! Help pass GPU-emulation parameters with type safety.
64 enum class EmulateGpuNonbonded : bool
66 //! Do not emulate GPUs.
74 /*! \brief Decide whether this thread-MPI simulation will run
75 * nonbonded tasks on GPUs.
77 * The number of GPU tasks and devices influences both the choice of
78 * the number of ranks, and checks upon any such choice made by the
79 * user. So we need to consider this before any automated choice of
80 * the number of thread-MPI ranks.
82 * \param[in] nonbondedTarget The user's choice for mdrun -nb for where to assign short-ranged nonbonded interaction tasks.
83 * \param[in] gpuIdsToUse The compatible GPUs that the user permitted us to use.
84 * \param[in] userGpuTaskAssignment The user-specified assignment of GPU tasks to device IDs.
85 * \param[in] emulateGpuNonbonded Whether we will emulate GPU calculation of nonbonded interactions.
86 * \param[in] buildSupportsNonbondedOnGpu Whether GROMACS was built with GPU support.
87 * \param[in] nonbondedOnGpuIsUseful Whether computing nonbonded interactions on a GPU is useful for this calculation.
88 * \param[in] numRanksPerSimulation The number of ranks in each simulation.
90 * \returns Whether the simulation will run nonbonded tasks on GPUs.
92 * \throws std::bad_alloc If out of memory
93 * InconsistentInputError If the user requirements are inconsistent. */
94 bool decideWhetherToUseGpusForNonbondedWithThreadMpi(TaskTarget nonbondedTarget,
95 const std::vector<int>& gpuIdsToUse,
96 const std::vector<int>& userGpuTaskAssignment,
97 EmulateGpuNonbonded emulateGpuNonbonded,
98 bool buildSupportsNonbondedOnGpu,
99 bool nonbondedOnGpuIsUseful,
100 int numRanksPerSimulation);
102 /*! \brief Decide whether this thread-MPI simulation will run
105 * The number of GPU tasks and devices influences both the choice of
106 * the number of ranks, and checks upon any such choice made by the
107 * user. So we need to consider this before any automated choice of
108 * the number of thread-MPI ranks.
110 * \param[in] useGpuForNonbonded Whether GPUs will be used for nonbonded interactions.
111 * \param[in] pmeTarget The user's choice for mdrun -pme for where to assign
112 * long-ranged PME nonbonded interaction tasks.
113 * \param[in] gpuIdsToUse The compatible GPUs that the user permitted us to use.
114 * \param[in] userGpuTaskAssignment The user-specified assignment of GPU tasks to device IDs.
115 * \param[in] hardwareInfo Hardware information
116 * \param[in] inputrec The user input
117 * \param[in] mtop Global system topology
118 * \param[in] numRanksPerSimulation The number of ranks in each simulation.
119 * \param[in] numPmeRanksPerSimulation The number of PME ranks in each simulation.
121 * \returns Whether the simulation will run PME tasks on GPUs.
123 * \throws std::bad_alloc If out of memory
124 * InconsistentInputError If the user requirements are inconsistent. */
125 bool decideWhetherToUseGpusForPmeWithThreadMpi(bool useGpuForNonbonded,
126 TaskTarget pmeTarget,
127 const std::vector<int>& gpuIdsToUse,
128 const std::vector<int>& userGpuTaskAssignment,
129 const gmx_hw_info_t& hardwareInfo,
130 const t_inputrec& inputrec,
131 const gmx_mtop_t& mtop,
132 int numRanksPerSimulation,
133 int numPmeRanksPerSimulation);
135 /*! \brief Decide whether the simulation will try to run nonbonded
138 * The final decision cannot be made until after the duty of the rank
139 * is known. But we need to know if nonbonded will run on GPUs for
140 * setting up DD (particularly rlist) and determining duty. If the
141 * user requires GPUs for the tasks of that duty, then it will be an
142 * error when none are found.
144 * With thread-MPI, calls have been made to
145 * decideWhetherToUseGpusForNonbondedWithThreadMpi() and
146 * decideWhetherToUseGpusForPmeWithThreadMpi() to help determine
147 * the number of ranks and run some checks, but the final
148 * decision is made in this routine, along with many more
149 * consistency checks.
151 * \param[in] nonbondedTarget The user's choice for mdrun -nb for where to assign short-ranged nonbonded interaction tasks.
152 * \param[in] userGpuTaskAssignment The user-specified assignment of GPU tasks to device IDs.
153 * \param[in] emulateGpuNonbonded Whether we will emulate GPU calculation of nonbonded interactions.
154 * \param[in] buildSupportsNonbondedOnGpu Whether GROMACS was build with GPU support.
155 * \param[in] nonbondedOnGpuIsUseful Whether computing nonbonded interactions on a GPU is useful for this calculation.
156 * \param[in] gpusWereDetected Whether compatible GPUs were detected on any node.
158 * \returns Whether the simulation will run nonbonded and PME tasks, respectively, on GPUs.
160 * \throws std::bad_alloc If out of memory
161 * InconsistentInputError If the user requirements are inconsistent. */
162 bool decideWhetherToUseGpusForNonbonded(TaskTarget nonbondedTarget,
163 const std::vector<int>& userGpuTaskAssignment,
164 EmulateGpuNonbonded emulateGpuNonbonded,
165 bool buildSupportsNonbondedOnGpu,
166 bool nonbondedOnGpuIsUseful,
167 bool gpusWereDetected);
169 /*! \brief Decide whether the simulation will try to run tasks of
170 * different types on GPUs.
172 * The final decision cannot be made until after the duty of the rank
173 * is known. But we need to know if nonbonded will run on GPUs for
174 * setting up DD (particularly rlist) and determining duty. If the
175 * user requires GPUs for the tasks of that duty, then it will be an
176 * error when none are found.
178 * With thread-MPI, calls have been made to
179 * decideWhetherToUseGpusForNonbondedWithThreadMpi() and
180 * decideWhetherToUseGpusForPmeWithThreadMpi() to help determine
181 * the number of ranks and run some checks, but the final
182 * decision is made in this routine, along with many more
183 * consistency checks.
185 * \param[in] useGpuForNonbonded Whether GPUs will be used for nonbonded interactions.
186 * \param[in] pmeTarget The user's choice for mdrun -pme for where to assign long-ranged PME nonbonded interaction tasks.
187 * \param[in] userGpuTaskAssignment The user-specified assignment of GPU tasks to device IDs.
188 * \param[in] hardwareInfo Hardware information
189 * \param[in] inputrec The user input
190 * \param[in] mtop Global system topology
191 * \param[in] numRanksPerSimulation The number of ranks in each simulation.
192 * \param[in] numPmeRanksPerSimulation The number of PME ranks in each simulation.
193 * \param[in] gpusWereDetected Whether compatible GPUs were detected on any node.
195 * \returns Whether the simulation will run nonbonded and PME tasks, respectively, on GPUs.
197 * \throws std::bad_alloc If out of memory
198 * InconsistentInputError If the user requirements are inconsistent. */
199 bool decideWhetherToUseGpusForPme(bool useGpuForNonbonded,
200 TaskTarget pmeTarget,
201 const std::vector<int>& userGpuTaskAssignment,
202 const gmx_hw_info_t& hardwareInfo,
203 const t_inputrec& inputrec,
204 const gmx_mtop_t& mtop,
205 int numRanksPerSimulation,
206 int numPmeRanksPerSimulation,
207 bool gpusWereDetected);
209 /*! \brief Decide whether the simulation will try to run bonded tasks on GPUs.
211 * \param[in] useGpuForNonbonded Whether GPUs will be used for nonbonded interactions.
212 * \param[in] useGpuForPme Whether GPUs will be used for PME interactions.
213 * \param[in] bondedTarget The user's choice for mdrun -bonded for where to assign tasks.
214 * \param[in] canUseGpuForBonded Whether the bonded interactions can run on a GPU
215 * \param[in] usingLJPme Whether Vdw interactions use LJ-PME.
216 * \param[in] usingElecPmeOrEwald Whether a PME or Ewald type method is used for electrostatics.
217 * \param[in] numPmeRanksPerSimulation The number of PME ranks in each simulation, can be -1 for auto.
218 * \param[in] gpusWereDetected Whether compatible GPUs were detected on any node.
220 * \returns Whether the simulation will run bondeded tasks on GPUs.
222 * \throws std::bad_alloc If out of memory
223 * InconsistentInputError If the user requirements are inconsistent. */
224 bool decideWhetherToUseGpusForBonded(bool useGpuForNonbonded,
226 TaskTarget bondedTarget,
227 bool canUseGpuForBonded,
229 bool usingElecPmeOrEwald,
230 int numPmeRanksPerSimulation,
231 bool gpusWereDetected);
233 /*! \brief Decide whether to use GPU for update.
235 * \param[in] forceGpuUpdateDefaultWithDD If update should run on GPU with DD by default.
236 * \param[in] isDomainDecomposition Whether there more than one domain.
237 * \param[in] useUpdateGroups If the constraints can be split across domains.
238 * \param[in] useGpuForPme Whether GPUs will be used for PME interactions.
239 * \param[in] useGpuForNonbonded Whether GPUs will be used for nonbonded interactions.
240 * \param[in] updateTarget User choice for running simulation on GPU.
241 * \param[in] gpusWereDetected Whether compatible GPUs were detected on any node.
242 * \param[in] inputrec The user input.
243 * \param[in] mtop The global topology.
244 * \param[in] useEssentialDynamics If essential dynamics is active.
245 * \param[in] doOrientationRestraints If orientation restraints are enabled.
246 * \param[in] useReplicaExchange If this is a REMD simulation.
247 * \param[in] doRerun It this is a rerun.
249 * \returns Whether complete simulation can be run on GPU.
250 * \throws std::bad_alloc If out of memory
251 * InconsistentInputError If the user requirements are inconsistent.
253 bool decideWhetherToUseGpuForUpdate(bool forceGpuUpdateDefaultWithDD,
254 bool isDomainDecomposition,
255 bool useUpdateGroups,
257 bool useGpuForNonbonded,
258 TaskTarget updateTarget,
259 bool gpusWereDetected,
260 const t_inputrec& inputrec,
261 const gmx_mtop_t& mtop,
262 bool useEssentialDynamics,
263 bool doOrientationRestraints,
264 bool useReplicaExchange,