src/gromacs/ewald/pme.h

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   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, 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.
  10  *
  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.
  15  *
  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.
  20  *
  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.
  25  *
  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.
  33  *
  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.
  36  */
  37 /*! \libinternal \file
  38  *
  39  * \brief This file contains function declarations necessary for
  40  * computing energies and forces for the PME long-ranged part (Coulomb
  41  * and LJ).
  42  *
  43  * \author Berk Hess <hess@kth.se>
  44  * \inlibraryapi
  45  * \ingroup module_ewald
  46  */
  47
  48 #ifndef GMX_EWALD_PME_H
  49 #define GMX_EWALD_PME_H
  50
  51 #include "gromacs/timing/wallcycle.h"
  52 #include "gromacs/timing/walltime_accounting.h"
  53 #include "gromacs/utility/real.h"
  54
  55 #include "pme-gpu-types.h"
  56
  57 struct interaction_const_t;
  58 struct t_commrec;
  59 struct t_inputrec;
  60 struct t_nrnb;
  61 struct pme_gpu_t;
  62 struct gmx_wallclock_gpu_pme_t;
  63 struct gmx_device_info_t;
  64
  65 namespace gmx
  66 {
  67 class ForceWithVirial;
  68 class MDLogger;
  69 }
  70
  71 enum {
  72     GMX_SUM_GRID_FORWARD, GMX_SUM_GRID_BACKWARD
  73 };
  74
  75 /*! \brief Return the smallest allowed PME grid size for \p pmeOrder */
  76 int minimalPmeGridSize(int pmeOrder);
  77
  78 /*! \brief Check restrictions on pme_order and the PME grid nkx,nky,nkz.
  79  *
  80  * With errorsAreFatal=true, an exception or fatal error is generated
  81  * on violation of restrictions.
  82  * With errorsAreFatal=false, false is returned on violation of restrictions.
  83  * When all restrictions are obeyed, true is returned.
  84  * Argument useThreads tells if any MPI rank doing PME uses more than 1 threads.
  85  * If at calling useThreads is unknown, pass true for conservative checking.
  86  *
  87  * The PME GPU restrictions are checked separately during pme_gpu_init().
  88  */
  89 bool gmx_pme_check_restrictions(int pme_order,
  90                                 int nkx, int nky, int nkz,
  91                                 int nnodes_major,
  92                                 bool useThreads,
  93                                 bool errorsAreFatal);
  94
  95 /*! \brief Initialize \p pmedata
  96  *
  97  * \returns  0 indicates all well, non zero is an error code.
  98  * \throws   gmx::InconsistentInputError if input grid sizes/PME order are inconsistent.
  99  */
 100 int gmx_pme_init(struct gmx_pme_t **pmedata, struct t_commrec *cr,
 101                  int nnodes_major, int nnodes_minor,
 102                  const t_inputrec *ir, int homenr,
 103                  gmx_bool bFreeEnergy_q, gmx_bool bFreeEnergy_lj,
 104                  gmx_bool bReproducible,
 105                  real ewaldcoeff_q, real ewaldcoeff_lj,
 106                  int nthread,
 107                  PmeRunMode runMode,
 108                  pme_gpu_t *pmeGPU,
 109                  gmx_device_info_t *gpuInfo,
 110                  const gmx::MDLogger &mdlog);
 111
 112 /*! \brief Destroys the PME data structure.*/
 113 void gmx_pme_destroy(gmx_pme_t *pme);
 114
 115 //@{
 116 /*! \brief Flag values that control what gmx_pme_do() will calculate
 117  *
 118  * These can be combined with bitwise-OR if more than one thing is required.
 119  */
 120 #define GMX_PME_SPREAD        (1<<0)
 121 #define GMX_PME_SOLVE         (1<<1)
 122 #define GMX_PME_CALC_F        (1<<2)
 123 #define GMX_PME_CALC_ENER_VIR (1<<3)
 124 /* This forces the grid to be backtransformed even without GMX_PME_CALC_F */
 125 #define GMX_PME_CALC_POT      (1<<4)
 126
 127 #define GMX_PME_DO_ALL_F  (GMX_PME_SPREAD | GMX_PME_SOLVE | GMX_PME_CALC_F)
 128 //@}
 129
 130 /*! \brief Do a PME calculation on a CPU for the long range electrostatics and/or LJ.
 131  *
 132  * The meaning of \p flags is defined above, and determines which
 133  * parts of the calculation are performed.
 134  *
 135  * \return 0 indicates all well, non zero is an error code.
 136  */
 137 int gmx_pme_do(struct gmx_pme_t *pme,
 138                int start,       int homenr,
 139                rvec x[],        rvec f[],
 140                real chargeA[],  real chargeB[],
 141                real c6A[],      real c6B[],
 142                real sigmaA[],   real sigmaB[],
 143                matrix box,      t_commrec *cr,
 144                int  maxshift_x, int maxshift_y,
 145                t_nrnb *nrnb,    gmx_wallcycle_t wcycle,
 146                matrix vir_q,    matrix vir_lj,
 147                real *energy_q,  real *energy_lj,
 148                real lambda_q,   real lambda_lj,
 149                real *dvdlambda_q, real *dvdlambda_lj,
 150                int flags);
 151
 152 /*! \brief Called on the nodes that do PME exclusively (as slaves) */
 153 int gmx_pmeonly(struct gmx_pme_t *pme,
 154                 struct t_commrec *cr,     t_nrnb *mynrnb,
 155                 gmx_wallcycle_t wcycle,
 156                 gmx_walltime_accounting_t walltime_accounting,
 157                 t_inputrec *ir, PmeRunMode runMode);
 158
 159 /*! \brief Calculate the PME grid energy V for n charges.
 160  *
 161  * The potential (found in \p pme) must have been found already with a
 162  * call to gmx_pme_do() with at least GMX_PME_SPREAD and GMX_PME_SOLVE
 163  * specified. Note that the charges are not spread on the grid in the
 164  * pme struct. Currently does not work in parallel or with free
 165  * energy.
 166  */
 167 void gmx_pme_calc_energy(struct gmx_pme_t *pme, int n, rvec *x, real *q, real *V);
 168
 169 /*! \brief Send the charges and maxshift to out PME-only node. */
 170 void gmx_pme_send_parameters(struct t_commrec *cr,
 171                              const interaction_const_t *ic,
 172                              gmx_bool bFreeEnergy_q, gmx_bool bFreeEnergy_lj,
 173                              real *chargeA, real *chargeB,
 174                              real *sqrt_c6A, real *sqrt_c6B,
 175                              real *sigmaA, real *sigmaB,
 176                              int maxshift_x, int maxshift_y);
 177
 178 /*! \brief Send the coordinates to our PME-only node and request a PME calculation */
 179 void gmx_pme_send_coordinates(struct t_commrec *cr, matrix box, rvec *x,
 180                               real lambda_q, real lambda_lj,
 181                               gmx_bool bEnerVir,
 182                               gmx_int64_t step);
 183
 184 /*! \brief Tell our PME-only node to finish */
 185 void gmx_pme_send_finish(struct t_commrec *cr);
 186
 187 /*! \brief Tell our PME-only node to reset all cycle and flop counters */
 188 void gmx_pme_send_resetcounters(struct t_commrec *cr, gmx_int64_t step);
 189
 190 /*! \brief PP nodes receive the long range forces from the PME nodes */
 191 void gmx_pme_receive_f(struct t_commrec *cr,
 192                        gmx::ForceWithVirial *forceWithVirial,
 193                        real *energy_q, real *energy_lj,
 194                        real *dvdlambda_q, real *dvdlambda_lj,
 195                        float *pme_cycles);
 196
 197 /*! \brief
 198  * This function updates the local atom data on GPU after DD (charges, coordinates, etc.).
 199  * TODO: it should update the PME CPU atom data as well.
 200  * (currently PME CPU call gmx_pme_do() gets passed the input pointers each step).
 201  *
 202  * \param[in] pme            The PME structure.
 203  * \param[in] nAtoms         The number of particles.
 204  * \param[in] charges        The pointer to the array of particle charges.
 205  */
 206 void gmx_pme_reinit_atoms(const gmx_pme_t *pme, const int nAtoms, const real *charges);
 207
 208 /* A block of PME GPU functions */
 209
 210 /*! \brief Checks whether the input system allows to run PME on GPU.
 211  * TODO: this mostly duplicates an internal PME assert function
 212  * pme_gpu_check_restrictions(), except that works with a
 213  * formed gmx_pme_t structure. Should that one go away/work with inputrec?
 214  *
 215  * \param[in]  ir     Input system.
 216  * \param[out] error  The error message if the input is not supported on GPU.
 217  *
 218  * \returns true if PME can run on GPU with this input, false otherwise.
 219  */
 220 bool pme_gpu_supports_input(const t_inputrec *ir, std::string *error);
 221
 222 /*! \brief
 223  * Returns the active PME codepath (CPU, GPU, mixed).
 224  * \todo This is a rather static data that should be managed by the higher level task scheduler.
 225  *
 226  * \param[in]  pme            The PME data structure.
 227  * \returns active PME codepath.
 228  */
 229 PmeRunMode pme_run_mode(const gmx_pme_t *pme);
 230
 231 /*! \brief
 232  * Tells if PME is enabled to run on GPU (not necessarily active at the moment).
 233  * \todo This is a rather static data that should be managed by the hardware assignment manager.
 234  * For now, it is synonymous with the active PME codepath (in the absence of dynamic switching).
 235  *
 236  * \param[in]  pme            The PME data structure.
 237  * \returns true if PME can run on GPU, false otherwise.
 238  */
 239 inline bool pme_gpu_task_enabled(const gmx_pme_t *pme)
 240 {
 241     return (pme != nullptr) && (pme_run_mode(pme) != PmeRunMode::CPU);
 242 }
 243
 244 /*! \brief
 245  * Resets the PME GPU timings. To be called at the reset step.
 246  *
 247  * \param[in] pme            The PME structure.
 248  */
 249 void pme_gpu_reset_timings(const gmx_pme_t *pme);
 250
 251 /*! \brief
 252  * Copies the PME GPU timings to the gmx_wallclock_gpu_pme_t structure (for log output). To be called at the run end.
 253  *
 254  * \param[in] pme               The PME structure.
 255  * \param[in] timings           The gmx_wallclock_gpu_pme_t structure.
 256  */
 257 void pme_gpu_get_timings(const gmx_pme_t         *pme,
 258                          gmx_wallclock_gpu_pme_t *timings);
 259
 260 /* The main PME GPU functions */
 261
 262 /*! \brief
 263  * Prepares PME on GPU step (updating the box if needed)
 264  * \param[in] pme               The PME data structure.
 265  * \param[in] needToUpdateBox   Tells if the stored unit cell parameters should be updated from \p box.
 266  * \param[in] box               The unit cell box.
 267  * \param[in] wcycle            The wallclock counter.
 268  * \param[in] flags             The combination of flags to affect the PME computation on this step.
 269  *                              The flags are the GMX_PME_ flags from pme.h.
 270  */
 271 void pme_gpu_prepare_step(gmx_pme_t      *pme,
 272                           bool            needToUpdateBox,
 273                           const matrix    box,
 274                           gmx_wallcycle_t wcycle,
 275                           int             flags);
 276
 277 /*! \brief
 278  * Launches first stage of PME on GPU - H2D input transfers, spreading kernel, and D2H grid transfer if needed.
 279  *
 280  * \param[in] pme               The PME data structure.
 281  * \param[in] x                 The array of local atoms' coordinates.
 282  * \param[in] wcycle            The wallclock counter.
 283  */
 284 void pme_gpu_launch_spread(gmx_pme_t      *pme,
 285                            const rvec     *x,
 286                            gmx_wallcycle_t wcycle);
 287
 288 /*! \brief
 289  * Launches middle stages of PME (FFT R2C, solving, FFT C2R) either on GPU or on CPU, depending on the run mode.
 290  *
 291  * \param[in] pme               The PME data structure.
 292  * \param[in] wcycle            The wallclock counter.
 293  */
 294 void pme_gpu_launch_complex_transforms(gmx_pme_t       *pme,
 295                                        gmx_wallcycle_t  wcycle);
 296
 297 /*! \brief
 298  * Launches last stage of PME on GPU - force gathering and D2H force transfer.
 299  *
 300  * \param[in]  pme               The PME data structure.
 301  * \param[in]  wcycle            The wallclock counter.
 302  * \param[in,out] forces         The array of local atoms' resulting forces.
 303  * \param[in]  forceTreatment    Tells how data in h_forces should be treated. The gathering kernel either stores
 304  *                               the output reciprocal forces into the host array, or copies its contents to the GPU first
 305  *                               and accumulates. The reduction is non-atomic.
 306  */
 307 void pme_gpu_launch_gather(const gmx_pme_t        *pme,
 308                            gmx_wallcycle_t         wcycle,
 309                            rvec                   *forces,
 310                            PmeForceOutputHandling  forceTreatment);
 311
 312 /*! \brief
 313  * Blocks until PME GPU tasks are completed, and gets the output forces and virial/energy
 314  * (if they were to be computed).
 315  *
 316  * \param[in]  pme            The PME data structure.
 317  * \param[in]  wcycle         The wallclock counter.
 318  * \param[out] vir_q          The output virial matrix.
 319  * \param[out] energy_q       The output energy.
 320  */
 321 void pme_gpu_wait_for_gpu(const gmx_pme_t *pme,
 322                           gmx_wallcycle_t  wcycle,
 323                           matrix           vir_q,
 324                           real            *energy_q);
 325
 326 #endif