2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2016,2017,2018,2019, 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.
37 * \brief Implements high-level PME GPU functions which do not require GPU framework-specific code.
39 * \author Aleksei Iupinov <a.yupinov@gmail.com>
40 * \ingroup module_ewald
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/utility/exceptions.h"
58 #include "gromacs/utility/fatalerror.h"
59 #include "gromacs/utility/gmxassert.h"
60 #include "gromacs/utility/stringutil.h"
62 #include "pme_gpu_internal.h"
64 #include "pme_internal.h"
65 #include "pme_solve.h"
67 void pme_gpu_reset_timings(const gmx_pme_t* pme)
69 if (pme_gpu_active(pme))
71 pme_gpu_reset_timings(pme->gpu);
75 void pme_gpu_get_timings(const gmx_pme_t* pme, gmx_wallclock_gpu_pme_t* timings)
77 if (pme_gpu_active(pme))
79 pme_gpu_get_timings(pme->gpu, timings);
83 int pme_gpu_get_padding_size(const gmx_pme_t* pme)
86 if (!pme || !pme_gpu_active(pme))
92 return pme_gpu_get_atom_data_alignment(pme->gpu);
97 * A convenience wrapper for launching either the GPU or CPU FFT.
99 * \param[in] pme The PME structure.
100 * \param[in] gridIndex The grid index - should currently always be 0.
101 * \param[in] dir The FFT direction enum.
102 * \param[in] wcycle The wallclock counter.
104 void inline parallel_3dfft_execute_gpu_wrapper(gmx_pme_t* pme,
106 enum gmx_fft_direction dir,
107 gmx_wallcycle_t wcycle)
109 GMX_ASSERT(gridIndex == 0, "Only single grid supported");
110 if (pme_gpu_performs_FFT(pme->gpu))
112 wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU);
113 wallcycle_sub_start_nocount(wcycle, ewcsLAUNCH_GPU_PME);
114 pme_gpu_3dfft(pme->gpu, dir, gridIndex);
115 wallcycle_sub_stop(wcycle, ewcsLAUNCH_GPU_PME);
116 wallcycle_stop(wcycle, ewcLAUNCH_GPU);
120 wallcycle_start(wcycle, ewcPME_FFT_MIXED_MODE);
121 #pragma omp parallel for num_threads(pme->nthread) schedule(static)
122 for (int thread = 0; thread < pme->nthread; thread++)
124 gmx_parallel_3dfft_execute(pme->pfft_setup[gridIndex], dir, thread, wcycle);
126 wallcycle_stop(wcycle, ewcPME_FFT_MIXED_MODE);
130 /* The PME computation code split into a few separate functions. */
132 void pme_gpu_prepare_computation(gmx_pme_t* pme,
133 bool needToUpdateBox,
135 gmx_wallcycle* wcycle,
137 bool useGpuForceReduction)
139 GMX_ASSERT(pme_gpu_active(pme), "This should be a GPU run of PME but it is not enabled.");
140 GMX_ASSERT(pme->nnodes > 0, "");
141 GMX_ASSERT(pme->nnodes == 1 || pme->ndecompdim > 0, "");
143 PmeGpu* pmeGpu = pme->gpu;
144 pmeGpu->settings.currentFlags = flags;
145 // TODO these flags are only here to honor the CPU PME code, and probably should be removed
146 pmeGpu->settings.useGpuForceReduction = useGpuForceReduction;
148 bool shouldUpdateBox = false;
149 for (int i = 0; i < DIM; ++i)
151 for (int j = 0; j <= i; ++j)
153 shouldUpdateBox |= (pmeGpu->common->previousBox[i][j] != box[i][j]);
154 pmeGpu->common->previousBox[i][j] = box[i][j];
158 if (needToUpdateBox || shouldUpdateBox) // || is to make the first computation always update
160 wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU);
161 wallcycle_sub_start_nocount(wcycle, ewcsLAUNCH_GPU_PME);
162 pme_gpu_update_input_box(pmeGpu, box);
163 wallcycle_sub_stop(wcycle, ewcsLAUNCH_GPU_PME);
164 wallcycle_stop(wcycle, ewcLAUNCH_GPU);
166 if (!pme_gpu_performs_solve(pmeGpu))
168 // TODO remove code duplication and add test coverage
170 pmeGpu->common->boxScaler->scaleBox(box, scaledBox);
171 gmx::invertBoxMatrix(scaledBox, pme->recipbox);
172 pme->boxVolume = scaledBox[XX][XX] * scaledBox[YY][YY] * scaledBox[ZZ][ZZ];
177 void pme_gpu_launch_spread(gmx_pme_t* pme, GpuEventSynchronizer* xReadyOnDevice, gmx_wallcycle* wcycle)
179 GMX_ASSERT(pme_gpu_active(pme), "This should be a GPU run of PME but it is not enabled.");
180 GMX_ASSERT(xReadyOnDevice || !pme->bPPnode || (GMX_GPU != GMX_GPU_CUDA),
181 "Need a valid xReadyOnDevice on PP+PME ranks with CUDA.");
183 PmeGpu* pmeGpu = pme->gpu;
185 const unsigned int gridIndex = 0;
186 real* fftgrid = pme->fftgrid[gridIndex];
187 if (pmeGpu->settings.currentFlags & GMX_PME_SPREAD)
189 /* Spread the coefficients on a grid */
190 const bool computeSplines = true;
191 const bool spreadCharges = true;
192 wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU);
193 wallcycle_sub_start_nocount(wcycle, ewcsLAUNCH_GPU_PME);
194 pme_gpu_spread(pmeGpu, xReadyOnDevice, gridIndex, fftgrid, computeSplines, spreadCharges);
195 wallcycle_sub_stop(wcycle, ewcsLAUNCH_GPU_PME);
196 wallcycle_stop(wcycle, ewcLAUNCH_GPU);
200 void pme_gpu_launch_complex_transforms(gmx_pme_t* pme, gmx_wallcycle* wcycle)
202 PmeGpu* pmeGpu = pme->gpu;
203 const bool computeEnergyAndVirial = (pmeGpu->settings.currentFlags & GMX_PME_CALC_ENER_VIR) != 0;
204 const bool performBackFFT = (pmeGpu->settings.currentFlags & (GMX_PME_CALC_F | GMX_PME_CALC_POT)) != 0;
205 const unsigned int gridIndex = 0;
206 t_complex* cfftgrid = pme->cfftgrid[gridIndex];
208 if (pmeGpu->settings.currentFlags & GMX_PME_SPREAD)
210 if (!pme_gpu_performs_FFT(pmeGpu))
212 wallcycle_start(wcycle, ewcWAIT_GPU_PME_SPREAD);
213 pme_gpu_sync_spread_grid(pme->gpu);
214 wallcycle_stop(wcycle, ewcWAIT_GPU_PME_SPREAD);
220 if (pmeGpu->settings.currentFlags & GMX_PME_SOLVE)
223 parallel_3dfft_execute_gpu_wrapper(pme, gridIndex, GMX_FFT_REAL_TO_COMPLEX, wcycle);
225 /* solve in k-space for our local cells */
226 if (pme_gpu_performs_solve(pmeGpu))
228 const auto gridOrdering = pme_gpu_uses_dd(pmeGpu) ? GridOrdering::YZX : GridOrdering::XYZ;
229 wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU);
230 wallcycle_sub_start_nocount(wcycle, ewcsLAUNCH_GPU_PME);
231 pme_gpu_solve(pmeGpu, cfftgrid, gridOrdering, computeEnergyAndVirial);
232 wallcycle_sub_stop(wcycle, ewcsLAUNCH_GPU_PME);
233 wallcycle_stop(wcycle, ewcLAUNCH_GPU);
237 wallcycle_start(wcycle, ewcPME_SOLVE_MIXED_MODE);
238 #pragma omp parallel for num_threads(pme->nthread) schedule(static)
239 for (int thread = 0; thread < pme->nthread; thread++)
241 solve_pme_yzx(pme, cfftgrid, pme->boxVolume, computeEnergyAndVirial,
242 pme->nthread, thread);
244 wallcycle_stop(wcycle, ewcPME_SOLVE_MIXED_MODE);
250 parallel_3dfft_execute_gpu_wrapper(pme, gridIndex, GMX_FFT_COMPLEX_TO_REAL, wcycle);
253 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
256 void pme_gpu_launch_gather(const gmx_pme_t* pme, gmx_wallcycle gmx_unused* wcycle, PmeForceOutputHandling forceTreatment)
258 GMX_ASSERT(pme_gpu_active(pme), "This should be a GPU run of PME but it is not enabled.");
260 if (!pme_gpu_performs_gather(pme->gpu))
265 wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU);
266 wallcycle_sub_start_nocount(wcycle, ewcsLAUNCH_GPU_PME);
267 const unsigned int gridIndex = 0;
268 real* fftgrid = pme->fftgrid[gridIndex];
269 pme_gpu_gather(pme->gpu, forceTreatment, reinterpret_cast<float*>(fftgrid));
270 wallcycle_sub_stop(wcycle, ewcsLAUNCH_GPU_PME);
271 wallcycle_stop(wcycle, ewcLAUNCH_GPU);
274 //! Accumulate the \c forcesToAdd to \c f, using the available threads.
275 static void sum_forces(gmx::ArrayRef<gmx::RVec> f, gmx::ArrayRef<const gmx::RVec> forceToAdd)
277 const int end = forceToAdd.size();
279 int gmx_unused nt = gmx_omp_nthreads_get(emntPME);
280 #pragma omp parallel for num_threads(nt) schedule(static)
281 for (int i = 0; i < end; i++)
283 f[i] += forceToAdd[i];
287 //! Reduce quantities from \c output to \c forceWithVirial and \c enerd.
288 static void pme_gpu_reduce_outputs(const int flags,
289 const PmeOutput& output,
290 gmx_wallcycle* wcycle,
291 gmx::ForceWithVirial* forceWithVirial,
292 gmx_enerdata_t* enerd)
294 wallcycle_start(wcycle, ewcPME_GPU_F_REDUCTION);
295 GMX_ASSERT(forceWithVirial, "Invalid force pointer");
297 const bool haveComputedEnergyAndVirial = (flags & GMX_PME_CALC_ENER_VIR) != 0;
298 if (haveComputedEnergyAndVirial)
300 GMX_ASSERT(enerd, "Invalid energy output manager");
301 forceWithVirial->addVirialContribution(output.coulombVirial_);
302 enerd->term[F_COUL_RECIP] += output.coulombEnergy_;
304 if (output.haveForceOutput_)
306 sum_forces(forceWithVirial->force_, output.forces_);
308 wallcycle_stop(wcycle, ewcPME_GPU_F_REDUCTION);
311 bool pme_gpu_try_finish_task(gmx_pme_t* pme,
313 gmx_wallcycle* wcycle,
314 gmx::ForceWithVirial* forceWithVirial,
315 gmx_enerdata_t* enerd,
316 GpuTaskCompletion completionKind)
318 GMX_ASSERT(pme_gpu_active(pme), "This should be a GPU run of PME but it is not enabled.");
319 GMX_ASSERT(!pme->gpu->settings.useGpuForceReduction,
320 "GPU force reduction should not be active on the pme_gpu_try_finish_task() path");
322 // First, if possible, check whether all tasks on the stream have
323 // completed, and return fast if not. Accumulate to wcycle the
324 // time needed for that checking, but do not yet record that the
325 // gather has occured.
326 bool needToSynchronize = true;
327 constexpr bool c_streamQuerySupported = (GMX_GPU == GMX_GPU_CUDA);
328 // TODO: implement c_streamQuerySupported with an additional GpuEventSynchronizer per stream (#2521)
329 if ((completionKind == GpuTaskCompletion::Check) && c_streamQuerySupported)
331 wallcycle_start_nocount(wcycle, ewcWAIT_GPU_PME_GATHER);
332 // Query the PME stream for completion of all tasks enqueued and
333 // if we're not done, stop the timer before early return.
334 const bool pmeGpuDone = pme_gpu_stream_query(pme->gpu);
335 wallcycle_stop(wcycle, ewcWAIT_GPU_PME_GATHER);
341 needToSynchronize = false;
344 wallcycle_start(wcycle, ewcWAIT_GPU_PME_GATHER);
345 // If the above check passed, then there is no need to make an
346 // explicit synchronization call.
347 if (needToSynchronize)
349 // Synchronize the whole PME stream at once, including D2H result transfers.
350 pme_gpu_synchronize(pme->gpu);
352 pme_gpu_update_timings(pme->gpu);
353 PmeOutput output = pme_gpu_getOutput(*pme, flags);
354 wallcycle_stop(wcycle, ewcWAIT_GPU_PME_GATHER);
356 GMX_ASSERT(pme->gpu->settings.useGpuForceReduction == !output.haveForceOutput_,
357 "When forces are reduced on the CPU, there needs to be force output");
358 pme_gpu_reduce_outputs(flags, output, wcycle, forceWithVirial, enerd);
363 // This is used by PME-only ranks
364 PmeOutput pme_gpu_wait_finish_task(gmx_pme_t* pme, const int flags, gmx_wallcycle* wcycle)
366 GMX_ASSERT(pme_gpu_active(pme), "This should be a GPU run of PME but it is not enabled.");
368 wallcycle_start(wcycle, ewcWAIT_GPU_PME_GATHER);
370 // Synchronize the whole PME stream at once, including D2H result transfers
371 // if there are outputs we need to wait for at this step; we still call getOutputs
372 // for uniformity and because it sets the PmeOutput.haveForceOutput_.
373 const bool haveComputedEnergyAndVirial = (flags & GMX_PME_CALC_ENER_VIR) != 0;
374 if (!pme->gpu->settings.useGpuForceReduction || haveComputedEnergyAndVirial)
376 pme_gpu_synchronize(pme->gpu);
379 PmeOutput output = pme_gpu_getOutput(*pme, flags);
380 wallcycle_stop(wcycle, ewcWAIT_GPU_PME_GATHER);
384 // This is used when not using the alternate-waiting reduction
385 void pme_gpu_wait_and_reduce(gmx_pme_t* pme,
387 gmx_wallcycle* wcycle,
388 gmx::ForceWithVirial* forceWithVirial,
389 gmx_enerdata_t* enerd)
391 PmeOutput output = pme_gpu_wait_finish_task(pme, flags, wcycle);
392 GMX_ASSERT(pme->gpu->settings.useGpuForceReduction == !output.haveForceOutput_,
393 "When forces are reduced on the CPU, there needs to be force output");
394 pme_gpu_reduce_outputs(flags, output, wcycle, forceWithVirial, enerd);
397 void pme_gpu_reinit_computation(const gmx_pme_t* pme, gmx_wallcycle* wcycle)
399 GMX_ASSERT(pme_gpu_active(pme), "This should be a GPU run of PME but it is not enabled.");
401 wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU);
402 wallcycle_sub_start_nocount(wcycle, ewcsLAUNCH_GPU_PME);
404 pme_gpu_update_timings(pme->gpu);
406 pme_gpu_clear_grids(pme->gpu);
407 pme_gpu_clear_energy_virial(pme->gpu);
409 wallcycle_sub_stop(wcycle, ewcsLAUNCH_GPU_PME);
410 wallcycle_stop(wcycle, ewcLAUNCH_GPU);
413 DeviceBuffer<float> pme_gpu_get_device_x(const gmx_pme_t* pme)
415 GMX_ASSERT((pme && pme_gpu_active(pme)),
416 "PME GPU coordinates buffer was requested from uninitialized PME module");
417 return pme_gpu_get_kernelparam_coordinates(pme->gpu);
420 void* pme_gpu_get_device_f(const gmx_pme_t* pme)
422 if (!pme || !pme_gpu_active(pme))
426 return pme_gpu_get_kernelparam_forces(pme->gpu);
429 void pme_gpu_set_device_x(const gmx_pme_t* pme, DeviceBuffer<float> d_x)
431 GMX_ASSERT(pme != nullptr, "Null pointer is passed as a PME to the set coordinates function.");
432 GMX_ASSERT(pme_gpu_active(pme), "This should be a GPU run of PME but it is not enabled.");
434 pme_gpu_set_kernelparam_coordinates(pme->gpu, d_x);
437 void* pme_gpu_get_device_stream(const gmx_pme_t* pme)
439 if (!pme || !pme_gpu_active(pme))
443 return pme_gpu_get_stream(pme->gpu);
446 void* pme_gpu_get_device_context(const gmx_pme_t* pme)
448 if (!pme || !pme_gpu_active(pme))
452 return pme_gpu_get_context(pme->gpu);
455 GpuEventSynchronizer* pme_gpu_get_f_ready_synchronizer(const gmx_pme_t* pme)
457 if (!pme || !pme_gpu_active(pme))
462 return pme_gpu_get_forces_ready_synchronizer(pme->gpu);