src/gromacs/ewald/pme_gpu.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2016,2017,2018,2019,2020, 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.
   8  *
   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.
  13  *
  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.
  18  *
  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.
  23  *
  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.
  31  *
  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.
  34  */
  35
  36 /*! \internal \file
  37  * \brief Implements high-level PME GPU functions which do not require GPU framework-specific code.
  38  *
  39  * \author Aleksei Iupinov <a.yupinov@gmail.com>
  40  * \ingroup module_ewald
  41  */
  42
  43 #include "gmxpre.h"
  44
  45 #include "config.h"
  46
  47 #include <list>
  48
  49 #include "gromacs/ewald/ewald_utils.h"
  50 #include "gromacs/ewald/pme.h"
  51 #include "gromacs/fft/parallel_3dfft.h"
  52 #include "gromacs/math/invertmatrix.h"
  53 #include "gromacs/mdlib/gmx_omp_nthreads.h"
  54 #include "gromacs/mdtypes/enerdata.h"
  55 #include "gromacs/mdtypes/forceoutput.h"
  56 #include "gromacs/mdtypes/inputrec.h"
  57 #include "gromacs/mdtypes/simulation_workload.h"
  58 #include "gromacs/utility/exceptions.h"
  59 #include "gromacs/utility/fatalerror.h"
  60 #include "gromacs/utility/gmxassert.h"
  61 #include "gromacs/utility/stringutil.h"
  62
  63 #include "pme_gpu_internal.h"
  64 #include "pme_gpu_settings.h"
  65 #include "pme_gpu_timings.h"
  66 #include "pme_gpu_types_host.h"
  67 #include "pme_grid.h"
  68 #include "pme_internal.h"
  69 #include "pme_solve.h"
  70
  71 /*! \brief
  72  * Finds out if PME is currently running on GPU.
  73  *
  74  * \todo The GPU module should not be constructed (or at least called)
  75  * when it is not active, so there should be no need to check whether
  76  * it is active. An assertion that this is true makes sense.
  77  *
  78  * \param[in] pme  The PME structure.
  79  * \returns        True if PME runs on GPU currently, false otherwise.
  80  */
  81 static inline bool pme_gpu_active(const gmx_pme_t* pme)
  82 {
  83     return (pme != nullptr) && (pme->runMode != PmeRunMode::CPU);
  84 }
  85
  86 void pme_gpu_reset_timings(const gmx_pme_t* pme)
  87 {
  88     if (pme_gpu_active(pme))
  89     {
  90         pme_gpu_reset_timings(pme->gpu);
  91     }
  92 }
  93
  94 void pme_gpu_get_timings(const gmx_pme_t* pme, gmx_wallclock_gpu_pme_t* timings)
  95 {
  96     if (pme_gpu_active(pme))
  97     {
  98         pme_gpu_get_timings(pme->gpu, timings);
  99     }
 100 }
 101
 102 int pme_gpu_get_block_size(const gmx_pme_t* pme)
 103 {
 104
 105     if (!pme || !pme_gpu_active(pme))
 106     {
 107         return 0;
 108     }
 109     else
 110     {
 111         return pme_gpu_get_atom_data_block_size();
 112     }
 113 }
 114
 115 /*! \brief
 116  * A convenience wrapper for launching either the GPU or CPU FFT.
 117  *
 118  * \param[in] pme            The PME structure.
 119  * \param[in] gridIndex      The grid index - should currently always be 0.
 120  * \param[in] dir            The FFT direction enum.
 121  * \param[in] wcycle         The wallclock counter.
 122  */
 123 void inline parallel_3dfft_execute_gpu_wrapper(gmx_pme_t*             pme,
 124                                                const int              gridIndex,
 125                                                enum gmx_fft_direction dir,
 126                                                gmx_wallcycle_t        wcycle)
 127 {
 128     GMX_ASSERT(gridIndex == 0, "Only single grid supported");
 129     if (pme_gpu_settings(pme->gpu).performGPUFFT)
 130     {
 131         wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU);
 132         wallcycle_sub_start_nocount(wcycle, ewcsLAUNCH_GPU_PME);
 133         pme_gpu_3dfft(pme->gpu, dir, gridIndex);
 134         wallcycle_sub_stop(wcycle, ewcsLAUNCH_GPU_PME);
 135         wallcycle_stop(wcycle, ewcLAUNCH_GPU);
 136     }
 137     else
 138     {
 139         wallcycle_start(wcycle, ewcPME_FFT_MIXED_MODE);
 140 #pragma omp parallel for num_threads(pme->nthread) schedule(static)
 141         for (int thread = 0; thread < pme->nthread; thread++)
 142         {
 143             gmx_parallel_3dfft_execute(pme->pfft_setup[gridIndex], dir, thread, wcycle);
 144         }
 145         wallcycle_stop(wcycle, ewcPME_FFT_MIXED_MODE);
 146     }
 147 }
 148
 149 /* The PME computation code split into a few separate functions. */
 150
 151 void pme_gpu_prepare_computation(gmx_pme_t*               pme,
 152                                  const matrix             box,
 153                                  gmx_wallcycle*           wcycle,
 154                                  const gmx::StepWorkload& stepWork)
 155 {
 156     GMX_ASSERT(pme_gpu_active(pme), "This should be a GPU run of PME but it is not enabled.");
 157     GMX_ASSERT(pme->nnodes > 0, "");
 158     GMX_ASSERT(pme->nnodes == 1 || pme->ndecompdim > 0, "");
 159
 160     PmeGpu* pmeGpu = pme->gpu;
 161     // TODO these flags are only here to honor the CPU PME code, and probably should be removed
 162     pmeGpu->settings.useGpuForceReduction = stepWork.useGpuPmeFReduction;
 163
 164     bool shouldUpdateBox = false;
 165     for (int i = 0; i < DIM; ++i)
 166     {
 167         for (int j = 0; j <= i; ++j)
 168         {
 169             shouldUpdateBox |= (pmeGpu->common->previousBox[i][j] != box[i][j]);
 170             pmeGpu->common->previousBox[i][j] = box[i][j];
 171         }
 172     }
 173
 174     if (stepWork.haveDynamicBox || shouldUpdateBox) // || is to make the first computation always update
 175     {
 176         wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU);
 177         wallcycle_sub_start_nocount(wcycle, ewcsLAUNCH_GPU_PME);
 178         pme_gpu_update_input_box(pmeGpu, box);
 179         wallcycle_sub_stop(wcycle, ewcsLAUNCH_GPU_PME);
 180         wallcycle_stop(wcycle, ewcLAUNCH_GPU);
 181
 182         if (!pme_gpu_settings(pmeGpu).performGPUSolve)
 183         {
 184             // TODO remove code duplication and add test coverage
 185             matrix scaledBox;
 186             pmeGpu->common->boxScaler->scaleBox(box, scaledBox);
 187             gmx::invertBoxMatrix(scaledBox, pme->recipbox);
 188             pme->boxVolume = scaledBox[XX][XX] * scaledBox[YY][YY] * scaledBox[ZZ][ZZ];
 189         }
 190     }
 191 }
 192
 193 void pme_gpu_launch_spread(gmx_pme_t* pme, GpuEventSynchronizer* xReadyOnDevice, gmx_wallcycle* wcycle)
 194 {
 195     GMX_ASSERT(pme_gpu_active(pme), "This should be a GPU run of PME but it is not enabled.");
 196     GMX_ASSERT(xReadyOnDevice || !pme->bPPnode || (GMX_GPU != GMX_GPU_CUDA),
 197                "Need a valid xReadyOnDevice on PP+PME ranks with CUDA.");
 198
 199     PmeGpu* pmeGpu = pme->gpu;
 200
 201     const unsigned int gridIndex = 0;
 202     real*              fftgrid   = pme->fftgrid[gridIndex];
 203     /* Spread the coefficients on a grid */
 204     const bool computeSplines = true;
 205     const bool spreadCharges  = true;
 206     wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU);
 207     wallcycle_sub_start_nocount(wcycle, ewcsLAUNCH_GPU_PME);
 208     pme_gpu_spread(pmeGpu, xReadyOnDevice, gridIndex, fftgrid, computeSplines, spreadCharges);
 209     wallcycle_sub_stop(wcycle, ewcsLAUNCH_GPU_PME);
 210     wallcycle_stop(wcycle, ewcLAUNCH_GPU);
 211 }
 212
 213 void pme_gpu_launch_complex_transforms(gmx_pme_t* pme, gmx_wallcycle* wcycle, const gmx::StepWorkload& stepWork)
 214 {
 215     PmeGpu*            pmeGpu    = pme->gpu;
 216     const auto&        settings  = pmeGpu->settings;
 217     const unsigned int gridIndex = 0;
 218     t_complex*         cfftgrid  = pme->cfftgrid[gridIndex];
 219     // There's no support for computing energy without virial, or vice versa
 220     const bool computeEnergyAndVirial = stepWork.computeEnergy || stepWork.computeVirial;
 221     if (!settings.performGPUFFT)
 222     {
 223         wallcycle_start(wcycle, ewcWAIT_GPU_PME_SPREAD);
 224         pme_gpu_sync_spread_grid(pme->gpu);
 225         wallcycle_stop(wcycle, ewcWAIT_GPU_PME_SPREAD);
 226     }
 227
 228     try
 229     {
 230         /* do R2C 3D-FFT */
 231         parallel_3dfft_execute_gpu_wrapper(pme, gridIndex, GMX_FFT_REAL_TO_COMPLEX, wcycle);
 232
 233         /* solve in k-space for our local cells */
 234         if (settings.performGPUSolve)
 235         {
 236             // TODO grid ordering should be set up at pme init time.
 237             const auto gridOrdering = settings.useDecomposition ? GridOrdering::YZX : GridOrdering::XYZ;
 238             wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU);
 239             wallcycle_sub_start_nocount(wcycle, ewcsLAUNCH_GPU_PME);
 240             pme_gpu_solve(pmeGpu, cfftgrid, gridOrdering, computeEnergyAndVirial);
 241             wallcycle_sub_stop(wcycle, ewcsLAUNCH_GPU_PME);
 242             wallcycle_stop(wcycle, ewcLAUNCH_GPU);
 243         }
 244         else
 245         {
 246             wallcycle_start(wcycle, ewcPME_SOLVE_MIXED_MODE);
 247 #pragma omp parallel for num_threads(pme->nthread) schedule(static)
 248             for (int thread = 0; thread < pme->nthread; thread++)
 249             {
 250                 solve_pme_yzx(pme, cfftgrid, pme->boxVolume, computeEnergyAndVirial, pme->nthread, thread);
 251             }
 252             wallcycle_stop(wcycle, ewcPME_SOLVE_MIXED_MODE);
 253         }
 254
 255         parallel_3dfft_execute_gpu_wrapper(pme, gridIndex, GMX_FFT_COMPLEX_TO_REAL, wcycle);
 256     }
 257     GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
 258 }
 259
 260 void pme_gpu_launch_gather(const gmx_pme_t* pme, gmx_wallcycle gmx_unused* wcycle)
 261 {
 262     GMX_ASSERT(pme_gpu_active(pme), "This should be a GPU run of PME but it is not enabled.");
 263
 264     if (!pme_gpu_settings(pme->gpu).performGPUGather)
 265     {
 266         return;
 267     }
 268
 269     wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU);
 270     wallcycle_sub_start_nocount(wcycle, ewcsLAUNCH_GPU_PME);
 271     const unsigned int gridIndex = 0;
 272     real*              fftgrid   = pme->fftgrid[gridIndex];
 273     pme_gpu_gather(pme->gpu, reinterpret_cast<float*>(fftgrid));
 274     wallcycle_sub_stop(wcycle, ewcsLAUNCH_GPU_PME);
 275     wallcycle_stop(wcycle, ewcLAUNCH_GPU);
 276 }
 277
 278 //! Accumulate the \c forcesToAdd to \c f, using the available threads.
 279 static void sum_forces(gmx::ArrayRef<gmx::RVec> f, gmx::ArrayRef<const gmx::RVec> forceToAdd)
 280 {
 281     const int end = forceToAdd.size();
 282
 283     int gmx_unused nt = gmx_omp_nthreads_get(emntPME);
 284 #pragma omp parallel for num_threads(nt) schedule(static)
 285     for (int i = 0; i < end; i++)
 286     {
 287         f[i] += forceToAdd[i];
 288     }
 289 }
 290
 291 //! Reduce quantities from \c output to \c forceWithVirial and \c enerd.
 292 static void pme_gpu_reduce_outputs(const bool            computeEnergyAndVirial,
 293                                    const PmeOutput&      output,
 294                                    gmx_wallcycle*        wcycle,
 295                                    gmx::ForceWithVirial* forceWithVirial,
 296                                    gmx_enerdata_t*       enerd)
 297 {
 298     wallcycle_start(wcycle, ewcPME_GPU_F_REDUCTION);
 299     GMX_ASSERT(forceWithVirial, "Invalid force pointer");
 300
 301     if (computeEnergyAndVirial)
 302     {
 303         GMX_ASSERT(enerd, "Invalid energy output manager");
 304         forceWithVirial->addVirialContribution(output.coulombVirial_);
 305         enerd->term[F_COUL_RECIP] += output.coulombEnergy_;
 306     }
 307     if (output.haveForceOutput_)
 308     {
 309         sum_forces(forceWithVirial->force_, output.forces_);
 310     }
 311     wallcycle_stop(wcycle, ewcPME_GPU_F_REDUCTION);
 312 }
 313
 314 bool pme_gpu_try_finish_task(gmx_pme_t*               pme,
 315                              const gmx::StepWorkload& stepWork,
 316                              gmx_wallcycle*           wcycle,
 317                              gmx::ForceWithVirial*    forceWithVirial,
 318                              gmx_enerdata_t*          enerd,
 319                              GpuTaskCompletion        completionKind)
 320 {
 321     GMX_ASSERT(pme_gpu_active(pme), "This should be a GPU run of PME but it is not enabled.");
 322     GMX_ASSERT(!pme->gpu->settings.useGpuForceReduction,
 323                "GPU force reduction should not be active on the pme_gpu_try_finish_task() path");
 324
 325     // First, if possible, check whether all tasks on the stream have
 326     // completed, and return fast if not. Accumulate to wcycle the
 327     // time needed for that checking, but do not yet record that the
 328     // gather has occured.
 329     bool           needToSynchronize      = true;
 330     constexpr bool c_streamQuerySupported = (GMX_GPU == GMX_GPU_CUDA);
 331     // TODO: implement c_streamQuerySupported with an additional GpuEventSynchronizer per stream (#2521)
 332     if ((completionKind == GpuTaskCompletion::Check) && c_streamQuerySupported)
 333     {
 334         wallcycle_start_nocount(wcycle, ewcWAIT_GPU_PME_GATHER);
 335         // Query the PME stream for completion of all tasks enqueued and
 336         // if we're not done, stop the timer before early return.
 337         const bool pmeGpuDone = pme_gpu_stream_query(pme->gpu);
 338         wallcycle_stop(wcycle, ewcWAIT_GPU_PME_GATHER);
 339
 340         if (!pmeGpuDone)
 341         {
 342             return false;
 343         }
 344         needToSynchronize = false;
 345     }
 346
 347     wallcycle_start(wcycle, ewcWAIT_GPU_PME_GATHER);
 348     // If the above check passed, then there is no need to make an
 349     // explicit synchronization call.
 350     if (needToSynchronize)
 351     {
 352         // Synchronize the whole PME stream at once, including D2H result transfers.
 353         pme_gpu_synchronize(pme->gpu);
 354     }
 355     pme_gpu_update_timings(pme->gpu);
 356     // There's no support for computing energy without virial, or vice versa
 357     const bool computeEnergyAndVirial = stepWork.computeEnergy || stepWork.computeVirial;
 358     PmeOutput  output                 = pme_gpu_getOutput(*pme, computeEnergyAndVirial);
 359     wallcycle_stop(wcycle, ewcWAIT_GPU_PME_GATHER);
 360
 361     GMX_ASSERT(pme->gpu->settings.useGpuForceReduction == !output.haveForceOutput_,
 362                "When forces are reduced on the CPU, there needs to be force output");
 363     pme_gpu_reduce_outputs(computeEnergyAndVirial, output, wcycle, forceWithVirial, enerd);
 364
 365     return true;
 366 }
 367
 368 // This is used by PME-only ranks
 369 PmeOutput pme_gpu_wait_finish_task(gmx_pme_t* pme, const bool computeEnergyAndVirial, gmx_wallcycle* wcycle)
 370 {
 371     GMX_ASSERT(pme_gpu_active(pme), "This should be a GPU run of PME but it is not enabled.");
 372
 373     wallcycle_start(wcycle, ewcWAIT_GPU_PME_GATHER);
 374
 375     // Synchronize the whole PME stream at once, including D2H result transfers
 376     // if there are outputs we need to wait for at this step; we still call getOutputs
 377     // for uniformity and because it sets the PmeOutput.haveForceOutput_.
 378     if (!pme->gpu->settings.useGpuForceReduction || computeEnergyAndVirial)
 379     {
 380         pme_gpu_synchronize(pme->gpu);
 381     }
 382
 383     PmeOutput output = pme_gpu_getOutput(*pme, computeEnergyAndVirial);
 384     wallcycle_stop(wcycle, ewcWAIT_GPU_PME_GATHER);
 385     return output;
 386 }
 387
 388 // This is used when not using the alternate-waiting reduction
 389 void pme_gpu_wait_and_reduce(gmx_pme_t*               pme,
 390                              const gmx::StepWorkload& stepWork,
 391                              gmx_wallcycle*           wcycle,
 392                              gmx::ForceWithVirial*    forceWithVirial,
 393                              gmx_enerdata_t*          enerd)
 394 {
 395     // There's no support for computing energy without virial, or vice versa
 396     const bool computeEnergyAndVirial = stepWork.computeEnergy || stepWork.computeVirial;
 397     PmeOutput  output = pme_gpu_wait_finish_task(pme, computeEnergyAndVirial, wcycle);
 398     GMX_ASSERT(pme->gpu->settings.useGpuForceReduction == !output.haveForceOutput_,
 399                "When forces are reduced on the CPU, there needs to be force output");
 400     pme_gpu_reduce_outputs(computeEnergyAndVirial, output, wcycle, forceWithVirial, enerd);
 401 }
 402
 403 void pme_gpu_reinit_computation(const gmx_pme_t* pme, gmx_wallcycle* wcycle)
 404 {
 405     GMX_ASSERT(pme_gpu_active(pme), "This should be a GPU run of PME but it is not enabled.");
 406
 407     wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU);
 408     wallcycle_sub_start_nocount(wcycle, ewcsLAUNCH_GPU_PME);
 409
 410     pme_gpu_update_timings(pme->gpu);
 411
 412     pme_gpu_clear_grids(pme->gpu);
 413     pme_gpu_clear_energy_virial(pme->gpu);
 414
 415     wallcycle_sub_stop(wcycle, ewcsLAUNCH_GPU_PME);
 416     wallcycle_stop(wcycle, ewcLAUNCH_GPU);
 417 }
 418
 419 void* pme_gpu_get_device_f(const gmx_pme_t* pme)
 420 {
 421     if (!pme || !pme_gpu_active(pme))
 422     {
 423         return nullptr;
 424     }
 425     return pme_gpu_get_kernelparam_forces(pme->gpu);
 426 }
 427
 428 void pme_gpu_set_device_x(const gmx_pme_t* pme, DeviceBuffer<gmx::RVec> d_x)
 429 {
 430     GMX_ASSERT(pme != nullptr, "Null pointer is passed as a PME to the set coordinates function.");
 431     GMX_ASSERT(pme_gpu_active(pme), "This should be a GPU run of PME but it is not enabled.");
 432
 433     pme_gpu_set_kernelparam_coordinates(pme->gpu, d_x);
 434 }
 435
 436 GpuEventSynchronizer* pme_gpu_get_f_ready_synchronizer(const gmx_pme_t* pme)
 437 {
 438     if (!pme || !pme_gpu_active(pme))
 439     {
 440         return nullptr;
 441     }
 442
 443     return pme_gpu_get_forces_ready_synchronizer(pme->gpu);
 444 }