2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
5 * Copyright (c) 2001-2004, The GROMACS development team.
6 * Copyright (c) 2013,2014,2015,2016,2017,2018, by the GROMACS development team, led by
7 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
8 * and including many others, as listed in the AUTHORS file in the
9 * top-level source directory and at http://www.gromacs.org.
11 * GROMACS is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public License
13 * as published by the Free Software Foundation; either version 2.1
14 * of the License, or (at your option) any later version.
16 * GROMACS is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * Lesser General Public License for more details.
21 * You should have received a copy of the GNU Lesser General Public
22 * License along with GROMACS; if not, see
23 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
24 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
26 * If you want to redistribute modifications to GROMACS, please
27 * consider that scientific software is very special. Version
28 * control is crucial - bugs must be traceable. We will be happy to
29 * consider code for inclusion in the official distribution, but
30 * derived work must not be called official GROMACS. Details are found
31 * in the README & COPYING files - if they are missing, get the
32 * official version at http://www.gromacs.org.
34 * To help us fund GROMACS development, we humbly ask that you cite
35 * the research papers on the package. Check out http://www.gromacs.org.
37 /*! \libinternal \file
39 * \brief This file contains function declarations necessary for
40 * computing energies and forces for the PME long-ranged part (Coulomb
43 * \author Berk Hess <hess@kth.se>
45 * \ingroup module_ewald
48 #ifndef GMX_EWALD_PME_H
49 #define GMX_EWALD_PME_H
53 #include "gromacs/gpu_utils/gpu_macros.h"
54 #include "gromacs/math/vectypes.h"
55 #include "gromacs/timing/walltime_accounting.h"
56 #include "gromacs/utility/arrayref.h"
57 #include "gromacs/utility/basedefinitions.h"
58 #include "gromacs/utility/real.h"
60 struct interaction_const_t;
65 struct gmx_wallclock_gpu_pme_t;
66 struct gmx_device_info_t;
71 enum class GpuTaskCompletion;
75 class ForceWithVirial;
80 GMX_SUM_GRID_FORWARD, GMX_SUM_GRID_BACKWARD
83 /*! \brief Possible PME codepaths on a rank.
84 * \todo: make this enum class with gmx_pme_t C++ refactoring
88 None, //!< No PME task is done
89 CPU, //!< Whole PME computation is done on CPU
90 GPU, //!< Whole PME computation is done on GPU
91 Mixed, //!< Mixed mode: only spread and gather run on GPU; FFT and solving are done on CPU.
94 //! PME gathering output forces treatment
95 enum class PmeForceOutputHandling
97 Set, /**< Gather simply writes into provided force buffer */
98 ReduceWithInput, /**< Gather adds its output to the buffer.
99 On GPU, that means additional H2D copy before the kernel launch. */
102 /*! \brief Return the smallest allowed PME grid size for \p pmeOrder */
103 int minimalPmeGridSize(int pmeOrder);
105 /*! \brief Check restrictions on pme_order and the PME grid nkx,nky,nkz.
107 * With errorsAreFatal=true, an exception or fatal error is generated
108 * on violation of restrictions.
109 * With errorsAreFatal=false, false is returned on violation of restrictions.
110 * When all restrictions are obeyed, true is returned.
111 * Argument useThreads tells if any MPI rank doing PME uses more than 1 threads.
112 * If at calling useThreads is unknown, pass true for conservative checking.
114 * The PME GPU restrictions are checked separately during pme_gpu_init().
116 bool gmx_pme_check_restrictions(int pme_order,
117 int nkx, int nky, int nkz,
118 int numPmeDomainsAlongX,
120 bool errorsAreFatal);
122 /*! \brief Construct PME data
124 * \throws gmx::InconsistentInputError if input grid sizes/PME order are inconsistent.
125 * \returns Pointer to newly allocated and initialized PME data.
127 gmx_pme_t *gmx_pme_init(const t_commrec *cr,
128 const NumPmeDomains &numPmeDomains,
129 const t_inputrec *ir, int homenr,
130 gmx_bool bFreeEnergy_q, gmx_bool bFreeEnergy_lj,
131 gmx_bool bReproducible,
132 real ewaldcoeff_q, real ewaldcoeff_lj,
136 gmx_device_info_t *gpuInfo,
137 const gmx::MDLogger &mdlog);
139 /*! \brief Destroys the PME data structure.*/
140 void gmx_pme_destroy(gmx_pme_t *pme);
143 /*! \brief Flag values that control what gmx_pme_do() will calculate
145 * These can be combined with bitwise-OR if more than one thing is required.
147 #define GMX_PME_SPREAD (1<<0)
148 #define GMX_PME_SOLVE (1<<1)
149 #define GMX_PME_CALC_F (1<<2)
150 #define GMX_PME_CALC_ENER_VIR (1<<3)
151 /* This forces the grid to be backtransformed even without GMX_PME_CALC_F */
152 #define GMX_PME_CALC_POT (1<<4)
154 #define GMX_PME_DO_ALL_F (GMX_PME_SPREAD | GMX_PME_SOLVE | GMX_PME_CALC_F)
157 /*! \brief Do a PME calculation on a CPU for the long range electrostatics and/or LJ.
159 * The meaning of \p flags is defined above, and determines which
160 * parts of the calculation are performed.
162 * \return 0 indicates all well, non zero is an error code.
164 int gmx_pme_do(struct gmx_pme_t *pme,
165 int start, int homenr,
167 real chargeA[], real chargeB[],
168 real c6A[], real c6B[],
169 real sigmaA[], real sigmaB[],
170 matrix box, const t_commrec *cr,
171 int maxshift_x, int maxshift_y,
172 t_nrnb *nrnb, gmx_wallcycle *wcycle,
173 matrix vir_q, matrix vir_lj,
174 real *energy_q, real *energy_lj,
175 real lambda_q, real lambda_lj,
176 real *dvdlambda_q, real *dvdlambda_lj,
179 /*! \brief Called on the nodes that do PME exclusively (as slaves) */
180 int gmx_pmeonly(struct gmx_pme_t *pme,
181 const t_commrec *cr, t_nrnb *mynrnb,
182 gmx_wallcycle *wcycle,
183 gmx_walltime_accounting_t walltime_accounting,
184 t_inputrec *ir, PmeRunMode runMode);
186 /*! \brief Calculate the PME grid energy V for n charges.
188 * The potential (found in \p pme) must have been found already with a
189 * call to gmx_pme_do() with at least GMX_PME_SPREAD and GMX_PME_SOLVE
190 * specified. Note that the charges are not spread on the grid in the
191 * pme struct. Currently does not work in parallel or with free
194 void gmx_pme_calc_energy(struct gmx_pme_t *pme, int n, rvec *x, real *q, real *V);
196 /*! \brief Send the charges and maxshift to out PME-only node. */
197 void gmx_pme_send_parameters(const t_commrec *cr,
198 const interaction_const_t *ic,
199 gmx_bool bFreeEnergy_q, gmx_bool bFreeEnergy_lj,
200 real *chargeA, real *chargeB,
201 real *sqrt_c6A, real *sqrt_c6B,
202 real *sigmaA, real *sigmaB,
203 int maxshift_x, int maxshift_y);
205 /*! \brief Send the coordinates to our PME-only node and request a PME calculation */
206 void gmx_pme_send_coordinates(const t_commrec *cr, matrix box, rvec *x,
207 real lambda_q, real lambda_lj,
209 gmx_int64_t step, gmx_wallcycle *wcycle);
211 /*! \brief Tell our PME-only node to finish */
212 void gmx_pme_send_finish(const t_commrec *cr);
214 /*! \brief Tell our PME-only node to reset all cycle and flop counters */
215 void gmx_pme_send_resetcounters(const t_commrec *cr, gmx_int64_t step);
217 /*! \brief PP nodes receive the long range forces from the PME nodes */
218 void gmx_pme_receive_f(const t_commrec *cr,
219 gmx::ForceWithVirial *forceWithVirial,
220 real *energy_q, real *energy_lj,
221 real *dvdlambda_q, real *dvdlambda_lj,
225 * This function updates the local atom data on GPU after DD (charges, coordinates, etc.).
226 * TODO: it should update the PME CPU atom data as well.
227 * (currently PME CPU call gmx_pme_do() gets passed the input pointers for each computation).
229 * \param[in] pme The PME structure.
230 * \param[in] nAtoms The number of particles.
231 * \param[in] charges The pointer to the array of particle charges.
233 void gmx_pme_reinit_atoms(const gmx_pme_t *pme, const int nAtoms, const real *charges);
235 /* A block of PME GPU functions */
237 /*! \brief Checks whether the GROMACS build allows to run PME on GPU.
238 * TODO: this partly duplicates an internal PME assert function
239 * pme_gpu_check_restrictions(), except that works with a
240 * formed gmx_pme_t structure. Should that one go away/work with inputrec?
242 * \param[out] error If non-null, the error message when PME is not supported on GPU.
244 * \returns true if PME can run on GPU on this build, false otherwise.
246 bool pme_gpu_supports_build(std::string *error);
248 /*! \brief Checks whether the input system allows to run PME on GPU.
249 * TODO: this partly duplicates an internal PME assert function
250 * pme_gpu_check_restrictions(), except that works with a
251 * formed gmx_pme_t structure. Should that one go away/work with inputrec?
253 * \param[in] ir Input system.
254 * \param[out] error If non-null, the error message if the input is not supported on GPU.
256 * \returns true if PME can run on GPU with this input, false otherwise.
258 bool pme_gpu_supports_input(const t_inputrec *ir, std::string *error);
261 * Returns the active PME codepath (CPU, GPU, mixed).
262 * \todo This is a rather static data that should be managed by the higher level task scheduler.
264 * \param[in] pme The PME data structure.
265 * \returns active PME codepath.
267 PmeRunMode pme_run_mode(const gmx_pme_t *pme);
270 * Tells if PME is enabled to run on GPU (not necessarily active at the moment).
271 * \todo This is a rather static data that should be managed by the hardware assignment manager.
272 * For now, it is synonymous with the active PME codepath (in the absence of dynamic switching).
274 * \param[in] pme The PME data structure.
275 * \returns true if PME can run on GPU, false otherwise.
277 inline bool pme_gpu_task_enabled(const gmx_pme_t *pme)
279 return (pme != nullptr) && (pme_run_mode(pme) != PmeRunMode::CPU);
282 // The following functions are all the PME GPU entry points,
283 // currently inlining to nothing on non-CUDA builds.
286 * Resets the PME GPU timings. To be called at the reset step.
288 * \param[in] pme The PME structure.
290 CUDA_FUNC_QUALIFIER void pme_gpu_reset_timings(const gmx_pme_t *CUDA_FUNC_ARGUMENT(pme)) CUDA_FUNC_TERM
293 * Copies the PME GPU timings to the gmx_wallclock_gpu_pme_t structure (for log output). To be called at the run end.
295 * \param[in] pme The PME structure.
296 * \param[in] timings The gmx_wallclock_gpu_pme_t structure.
298 CUDA_FUNC_QUALIFIER void pme_gpu_get_timings(const gmx_pme_t *CUDA_FUNC_ARGUMENT(pme),
299 gmx_wallclock_gpu_pme_t *CUDA_FUNC_ARGUMENT(timings)) CUDA_FUNC_TERM
301 /* The main PME GPU functions */
304 * Prepares PME on GPU computation (updating the box if needed)
305 * \param[in] pme The PME data structure.
306 * \param[in] needToUpdateBox Tells if the stored unit cell parameters should be updated from \p box.
307 * \param[in] box The unit cell box.
308 * \param[in] wcycle The wallclock counter.
309 * \param[in] flags The combination of flags to affect this PME computation.
310 * The flags are the GMX_PME_ flags from pme.h.
312 CUDA_FUNC_QUALIFIER void pme_gpu_prepare_computation(gmx_pme_t *CUDA_FUNC_ARGUMENT(pme),
313 bool CUDA_FUNC_ARGUMENT(needToUpdateBox),
314 const matrix CUDA_FUNC_ARGUMENT(box),
315 gmx_wallcycle *CUDA_FUNC_ARGUMENT(wcycle),
316 int CUDA_FUNC_ARGUMENT(flags)) CUDA_FUNC_TERM
319 * Launches first stage of PME on GPU - H2D input transfers, spreading kernel, and D2H grid transfer if needed.
321 * \param[in] pme The PME data structure.
322 * \param[in] x The array of local atoms' coordinates.
323 * \param[in] wcycle The wallclock counter.
325 CUDA_FUNC_QUALIFIER void pme_gpu_launch_spread(gmx_pme_t *CUDA_FUNC_ARGUMENT(pme),
326 const rvec *CUDA_FUNC_ARGUMENT(x),
327 gmx_wallcycle *CUDA_FUNC_ARGUMENT(wcycle)) CUDA_FUNC_TERM
330 * Launches middle stages of PME (FFT R2C, solving, FFT C2R) either on GPU or on CPU, depending on the run mode.
332 * \param[in] pme The PME data structure.
333 * \param[in] wcycle The wallclock counter.
335 CUDA_FUNC_QUALIFIER void pme_gpu_launch_complex_transforms(gmx_pme_t *CUDA_FUNC_ARGUMENT(pme),
336 gmx_wallcycle *CUDA_FUNC_ARGUMENT(wcycle)) CUDA_FUNC_TERM
339 * Launches last stage of PME on GPU - force gathering and D2H force transfer.
341 * \param[in] pme The PME data structure.
342 * \param[in] wcycle The wallclock counter.
343 * \param[in] forceTreatment Tells how data should be treated. The gathering kernel either stores
344 * the output reciprocal forces into the host array, or copies its contents to the GPU first
345 * and accumulates. The reduction is non-atomic.
347 CUDA_FUNC_QUALIFIER void pme_gpu_launch_gather(const gmx_pme_t *CUDA_FUNC_ARGUMENT(pme),
348 gmx_wallcycle *CUDA_FUNC_ARGUMENT(wcycle),
349 PmeForceOutputHandling CUDA_FUNC_ARGUMENT(forceTreatment)) CUDA_FUNC_TERM
352 * Blocks until PME GPU tasks are completed, and gets the output forces and virial/energy
353 * (if they were to be computed).
355 * \param[in] pme The PME data structure.
356 * \param[out] wcycle The wallclock counter.
357 * \param[out] forces The output forces.
358 * \param[out] virial The output virial matrix.
359 * \param[out] energy The output energy.
361 CUDA_FUNC_QUALIFIER void pme_gpu_wait_finish_task(const gmx_pme_t *CUDA_FUNC_ARGUMENT(pme),
362 gmx_wallcycle *CUDA_FUNC_ARGUMENT(wcycle),
363 gmx::ArrayRef<const gmx::RVec> *CUDA_FUNC_ARGUMENT(forces),
364 matrix CUDA_FUNC_ARGUMENT(virial),
365 real *CUDA_FUNC_ARGUMENT(energy)) CUDA_FUNC_TERM
367 * Attempts to complete PME GPU tasks.
369 * The \p completionKind argument controls whether the function blocks until all
370 * PME GPU tasks enqueued completed (as pme_gpu_wait_finish_task() does) or only
371 * checks and returns immediately if they did not.
372 * When blocking or the tasks have completed it also gets the output forces
373 * by assigning the ArrayRef to the \p forces pointer passed in.
374 * Virial/energy are also outputs if they were to be computed.
376 * Note: also launches the reinitalization of the PME output buffers.
377 * TODO: this should be moved out to avoid miscounting its wall-time (as wait iso launch).
379 * \param[in] pme The PME data structure.
380 * \param[in] wcycle The wallclock counter.
381 * \param[out] forces The output forces.
382 * \param[out] virial The output virial matrix.
383 * \param[out] energy The output energy.
384 * \param[in] completionKind Indicates whether PME task completion should only be checked rather than waited for
385 * \returns True if the PME GPU tasks have completed
387 CUDA_FUNC_QUALIFIER bool pme_gpu_try_finish_task(const gmx_pme_t *CUDA_FUNC_ARGUMENT(pme),
388 gmx_wallcycle *CUDA_FUNC_ARGUMENT(wcycle),
389 gmx::ArrayRef<const gmx::RVec> *CUDA_FUNC_ARGUMENT(forces),
390 matrix CUDA_FUNC_ARGUMENT(virial),
391 real *CUDA_FUNC_ARGUMENT(energy),
392 GpuTaskCompletion CUDA_FUNC_ARGUMENT(completionKind)) CUDA_FUNC_TERM_WITH_RETURN(false)
395 * The PME GPU reinitialization function that is called both at the end of any PME computation and on any load balancing.
397 * Clears the internal grid and energy/virial buffers; it is not safe to start
398 * the PME computation without calling this.
399 * Note that unlike in the nbnxn module, the force buffer does not need clearing.
401 * \todo Rename this function to *clear* -- it clearly only does output resetting
402 * and we should be clear about what the function does..
404 * \param[in] pme The PME data structure.
405 * \param[in] wcycle The wallclock counter.
407 CUDA_FUNC_QUALIFIER void pme_gpu_reinit_computation(const gmx_pme_t *CUDA_FUNC_ARGUMENT(pme),
408 gmx_wallcycle *CUDA_FUNC_ARGUMENT(wcycle)) CUDA_FUNC_TERM