2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2016,2017,2018, 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 This file contains internal CUDA function implementations
38 * for performing the PME calculations on GPU.
40 * \author Aleksei Iupinov <a.yupinov@gmail.com>
50 #include "gromacs/gpu_utils/cudautils.cuh"
51 #include "gromacs/gpu_utils/pmalloc_cuda.h"
52 #include "gromacs/utility/gmxassert.h"
53 #include "gromacs/utility/smalloc.h"
56 #include "pme-3dfft.cuh"
59 int pme_gpu_get_atom_data_alignment(const PmeGpu *pmeGpu)
61 const int order = pmeGpu->common->pme_order;
62 GMX_ASSERT(order > 0, "Invalid PME order");
63 return PME_ATOM_DATA_ALIGNMENT;
66 int pme_gpu_get_atoms_per_warp(const PmeGpu *pmeGpu)
68 const int order = pmeGpu->common->pme_order;
69 GMX_ASSERT(order > 0, "Invalid PME order");
70 return PME_SPREADGATHER_ATOMS_PER_WARP;
73 void pme_gpu_synchronize(const PmeGpu *pmeGpu)
75 cudaError_t stat = cudaStreamSynchronize(pmeGpu->archSpecific->pmeStream);
76 CU_RET_ERR(stat, "Failed to synchronize the PME GPU stream!");
79 void pme_gpu_alloc_energy_virial(const PmeGpu *pmeGpu)
81 const size_t energyAndVirialSize = c_virialAndEnergyCount * sizeof(float);
82 cudaError_t stat = cudaMalloc((void **)&pmeGpu->kernelParams->constants.d_virialAndEnergy, energyAndVirialSize);
83 CU_RET_ERR(stat, "cudaMalloc failed on PME energy and virial");
84 pmalloc((void **)&pmeGpu->staging.h_virialAndEnergy, energyAndVirialSize);
87 void pme_gpu_free_energy_virial(PmeGpu *pmeGpu)
89 cudaError_t stat = cudaFree(pmeGpu->kernelParams->constants.d_virialAndEnergy);
90 CU_RET_ERR(stat, "cudaFree failed on PME energy and virial");
91 pmeGpu->kernelParams->constants.d_virialAndEnergy = nullptr;
92 pfree(pmeGpu->staging.h_virialAndEnergy);
93 pmeGpu->staging.h_virialAndEnergy = nullptr;
96 void pme_gpu_clear_energy_virial(const PmeGpu *pmeGpu)
98 cudaError_t stat = cudaMemsetAsync(pmeGpu->kernelParams->constants.d_virialAndEnergy, 0,
99 c_virialAndEnergyCount * sizeof(float), pmeGpu->archSpecific->pmeStream);
100 CU_RET_ERR(stat, "PME energy/virial cudaMemsetAsync error");
103 void pme_gpu_realloc_and_copy_bspline_values(const PmeGpu *pmeGpu)
105 const int splineValuesOffset[DIM] = {
107 pmeGpu->kernelParams->grid.realGridSize[XX],
108 pmeGpu->kernelParams->grid.realGridSize[XX] + pmeGpu->kernelParams->grid.realGridSize[YY]
110 memcpy((void *)&pmeGpu->kernelParams->grid.splineValuesOffset, &splineValuesOffset, sizeof(splineValuesOffset));
112 const int newSplineValuesSize = pmeGpu->kernelParams->grid.realGridSize[XX] +
113 pmeGpu->kernelParams->grid.realGridSize[YY] +
114 pmeGpu->kernelParams->grid.realGridSize[ZZ];
115 const bool shouldRealloc = (newSplineValuesSize > pmeGpu->archSpecific->splineValuesSize);
116 cu_realloc_buffered((void **)&pmeGpu->kernelParams->grid.d_splineModuli, nullptr, sizeof(float),
117 &pmeGpu->archSpecific->splineValuesSize, &pmeGpu->archSpecific->splineValuesSizeAlloc, newSplineValuesSize, pmeGpu->archSpecific->pmeStream, true);
120 /* Reallocate the host buffer */
121 pfree(pmeGpu->staging.h_splineModuli);
122 pmalloc((void **)&pmeGpu->staging.h_splineModuli, newSplineValuesSize * sizeof(float));
124 for (int i = 0; i < DIM; i++)
126 memcpy(pmeGpu->staging.h_splineModuli + splineValuesOffset[i], pmeGpu->common->bsp_mod[i].data(), pmeGpu->common->bsp_mod[i].size() * sizeof(float));
128 /* TODO: pin original buffer instead! */
129 cu_copy_H2D(pmeGpu->kernelParams->grid.d_splineModuli, pmeGpu->staging.h_splineModuli,
130 newSplineValuesSize * sizeof(float), pmeGpu->settings.transferKind, pmeGpu->archSpecific->pmeStream);
133 void pme_gpu_free_bspline_values(const PmeGpu *pmeGpu)
135 pfree(pmeGpu->staging.h_splineModuli);
136 freeDeviceBuffer(&pmeGpu->kernelParams->grid.d_splineModuli);
139 void pme_gpu_realloc_forces(PmeGpu *pmeGpu)
141 const size_t newForcesSize = pmeGpu->nAtomsAlloc * DIM;
142 GMX_ASSERT(newForcesSize > 0, "Bad number of atoms in PME GPU");
143 cu_realloc_buffered((void **)&pmeGpu->kernelParams->atoms.d_forces, nullptr, sizeof(float),
144 &pmeGpu->archSpecific->forcesSize, &pmeGpu->archSpecific->forcesSizeAlloc, newForcesSize, pmeGpu->archSpecific->pmeStream, true);
145 pmeGpu->staging.h_forces.reserve(pmeGpu->nAtomsAlloc);
146 pmeGpu->staging.h_forces.resize(pmeGpu->kernelParams->atoms.nAtoms);
149 void pme_gpu_free_forces(const PmeGpu *pmeGpu)
151 freeDeviceBuffer(&pmeGpu->kernelParams->atoms.d_forces);
154 void pme_gpu_copy_input_forces(PmeGpu *pmeGpu)
156 const size_t forcesSize = DIM * pmeGpu->kernelParams->atoms.nAtoms * sizeof(float);
157 GMX_ASSERT(forcesSize > 0, "Bad number of atoms in PME GPU");
158 cu_copy_H2D(pmeGpu->kernelParams->atoms.d_forces, pmeGpu->staging.h_forces.data(), forcesSize, pmeGpu->settings.transferKind, pmeGpu->archSpecific->pmeStream);
161 void pme_gpu_copy_output_forces(PmeGpu *pmeGpu)
163 const size_t forcesSize = DIM * pmeGpu->kernelParams->atoms.nAtoms * sizeof(float);
164 GMX_ASSERT(forcesSize > 0, "Bad number of atoms in PME GPU");
165 cu_copy_D2H(pmeGpu->staging.h_forces.data(), pmeGpu->kernelParams->atoms.d_forces, forcesSize, pmeGpu->settings.transferKind, pmeGpu->archSpecific->pmeStream);
168 void pme_gpu_realloc_coordinates(const PmeGpu *pmeGpu)
170 const size_t newCoordinatesSize = pmeGpu->nAtomsAlloc * DIM;
171 GMX_ASSERT(newCoordinatesSize > 0, "Bad number of atoms in PME GPU");
172 cu_realloc_buffered((void **)&pmeGpu->kernelParams->atoms.d_coordinates, nullptr, sizeof(float),
173 &pmeGpu->archSpecific->coordinatesSize, &pmeGpu->archSpecific->coordinatesSizeAlloc, newCoordinatesSize, pmeGpu->archSpecific->pmeStream, true);
176 const size_t paddingIndex = DIM * pmeGpu->kernelParams->atoms.nAtoms;
177 const size_t paddingCount = DIM * pmeGpu->nAtomsAlloc - paddingIndex;
178 if (paddingCount > 0)
180 cudaError_t stat = cudaMemsetAsync(pmeGpu->kernelParams->atoms.d_coordinates + paddingIndex, 0, paddingCount * sizeof(float), pmeGpu->archSpecific->pmeStream);
181 CU_RET_ERR(stat, "PME failed to clear the padded coordinates");
186 void pme_gpu_copy_input_coordinates(const PmeGpu *pmeGpu, const rvec *h_coordinates)
188 GMX_ASSERT(h_coordinates, "Bad host-side coordinate buffer in PME GPU");
190 GMX_RELEASE_ASSERT(false, "Only single precision is supported");
191 GMX_UNUSED_VALUE(h_coordinates);
193 cu_copy_H2D(pmeGpu->kernelParams->atoms.d_coordinates, const_cast<rvec *>(h_coordinates),
194 pmeGpu->kernelParams->atoms.nAtoms * sizeof(rvec), pmeGpu->settings.transferKind, pmeGpu->archSpecific->pmeStream);
198 void pme_gpu_free_coordinates(const PmeGpu *pmeGpu)
200 freeDeviceBuffer(&pmeGpu->kernelParams->atoms.d_coordinates);
203 void pme_gpu_realloc_and_copy_input_coefficients(const PmeGpu *pmeGpu, const float *h_coefficients)
205 GMX_ASSERT(h_coefficients, "Bad host-side charge buffer in PME GPU");
206 const size_t newCoefficientsSize = pmeGpu->nAtomsAlloc;
207 GMX_ASSERT(newCoefficientsSize > 0, "Bad number of atoms in PME GPU");
208 cu_realloc_buffered((void **)&pmeGpu->kernelParams->atoms.d_coefficients, nullptr, sizeof(float),
209 &pmeGpu->archSpecific->coefficientsSize, &pmeGpu->archSpecific->coefficientsSizeAlloc,
210 newCoefficientsSize, pmeGpu->archSpecific->pmeStream, true);
211 cu_copy_H2D(pmeGpu->kernelParams->atoms.d_coefficients, const_cast<float *>(h_coefficients),
212 pmeGpu->kernelParams->atoms.nAtoms * sizeof(float), pmeGpu->settings.transferKind, pmeGpu->archSpecific->pmeStream);
215 const size_t paddingIndex = pmeGpu->kernelParams->atoms.nAtoms;
216 const size_t paddingCount = pmeGpu->nAtomsAlloc - paddingIndex;
217 if (paddingCount > 0)
219 cudaError_t stat = cudaMemsetAsync(pmeGpu->kernelParams->atoms.d_coefficients + paddingIndex, 0, paddingCount * sizeof(float), pmeGpu->archSpecific->pmeStream);
220 CU_RET_ERR(stat, "PME failed to clear the padded charges");
225 void pme_gpu_free_coefficients(const PmeGpu *pmeGpu)
227 freeDeviceBuffer(&pmeGpu->kernelParams->atoms.d_coefficients);
230 void pme_gpu_realloc_spline_data(const PmeGpu *pmeGpu)
232 const int order = pmeGpu->common->pme_order;
233 const int alignment = pme_gpu_get_atoms_per_warp(pmeGpu);
234 const size_t nAtomsPadded = ((pmeGpu->nAtomsAlloc + alignment - 1) / alignment) * alignment;
235 const int newSplineDataSize = DIM * order * nAtomsPadded;
236 GMX_ASSERT(newSplineDataSize > 0, "Bad number of atoms in PME GPU");
237 /* Two arrays of the same size */
238 const bool shouldRealloc = (newSplineDataSize > pmeGpu->archSpecific->splineDataSize);
239 int currentSizeTemp = pmeGpu->archSpecific->splineDataSize;
240 int currentSizeTempAlloc = pmeGpu->archSpecific->splineDataSizeAlloc;
241 cu_realloc_buffered((void **)&pmeGpu->kernelParams->atoms.d_theta, nullptr, sizeof(float),
242 ¤tSizeTemp, ¤tSizeTempAlloc, newSplineDataSize, pmeGpu->archSpecific->pmeStream, true);
243 cu_realloc_buffered((void **)&pmeGpu->kernelParams->atoms.d_dtheta, nullptr, sizeof(float),
244 &pmeGpu->archSpecific->splineDataSize, &pmeGpu->archSpecific->splineDataSizeAlloc, newSplineDataSize, pmeGpu->archSpecific->pmeStream, true);
245 // the host side reallocation
248 pfree(pmeGpu->staging.h_theta);
249 pmalloc((void **)&pmeGpu->staging.h_theta, newSplineDataSize * sizeof(float));
250 pfree(pmeGpu->staging.h_dtheta);
251 pmalloc((void **)&pmeGpu->staging.h_dtheta, newSplineDataSize * sizeof(float));
255 void pme_gpu_free_spline_data(const PmeGpu *pmeGpu)
257 /* Two arrays of the same size */
258 freeDeviceBuffer(&pmeGpu->kernelParams->atoms.d_theta);
259 freeDeviceBuffer(&pmeGpu->kernelParams->atoms.d_dtheta);
260 pfree(pmeGpu->staging.h_theta);
261 pfree(pmeGpu->staging.h_dtheta);
264 void pme_gpu_realloc_grid_indices(const PmeGpu *pmeGpu)
266 const size_t newIndicesSize = DIM * pmeGpu->nAtomsAlloc;
267 GMX_ASSERT(newIndicesSize > 0, "Bad number of atoms in PME GPU");
268 cu_realloc_buffered((void **)&pmeGpu->kernelParams->atoms.d_gridlineIndices, nullptr, sizeof(int),
269 &pmeGpu->archSpecific->gridlineIndicesSize, &pmeGpu->archSpecific->gridlineIndicesSizeAlloc, newIndicesSize, pmeGpu->archSpecific->pmeStream, true);
270 pfree(pmeGpu->staging.h_gridlineIndices);
271 pmalloc((void **)&pmeGpu->staging.h_gridlineIndices, newIndicesSize * sizeof(int));
274 void pme_gpu_free_grid_indices(const PmeGpu *pmeGpu)
276 freeDeviceBuffer(&pmeGpu->kernelParams->atoms.d_gridlineIndices);
277 pfree(pmeGpu->staging.h_gridlineIndices);
280 void pme_gpu_realloc_grids(PmeGpu *pmeGpu)
282 auto *kernelParamsPtr = pmeGpu->kernelParams.get();
283 const int newRealGridSize = kernelParamsPtr->grid.realGridSizePadded[XX] *
284 kernelParamsPtr->grid.realGridSizePadded[YY] *
285 kernelParamsPtr->grid.realGridSizePadded[ZZ];
286 const int newComplexGridSize = kernelParamsPtr->grid.complexGridSizePadded[XX] *
287 kernelParamsPtr->grid.complexGridSizePadded[YY] *
288 kernelParamsPtr->grid.complexGridSizePadded[ZZ] * 2;
289 // Multiplied by 2 because we count complex grid size for complex numbers, but all allocations/pointers are float
290 if (pmeGpu->archSpecific->performOutOfPlaceFFT)
292 /* 2 separate grids */
293 cu_realloc_buffered((void **)&kernelParamsPtr->grid.d_fourierGrid, nullptr, sizeof(float),
294 &pmeGpu->archSpecific->complexGridSize, &pmeGpu->archSpecific->complexGridSizeAlloc,
295 newComplexGridSize, pmeGpu->archSpecific->pmeStream, true);
296 cu_realloc_buffered((void **)&kernelParamsPtr->grid.d_realGrid, nullptr, sizeof(float),
297 &pmeGpu->archSpecific->realGridSize, &pmeGpu->archSpecific->realGridSizeAlloc,
298 newRealGridSize, pmeGpu->archSpecific->pmeStream, true);
302 /* A single buffer so that any grid will fit */
303 const int newGridsSize = std::max(newRealGridSize, newComplexGridSize);
304 cu_realloc_buffered((void **)&kernelParamsPtr->grid.d_realGrid, nullptr, sizeof(float),
305 &pmeGpu->archSpecific->realGridSize, &pmeGpu->archSpecific->realGridSizeAlloc,
306 newGridsSize, pmeGpu->archSpecific->pmeStream, true);
307 kernelParamsPtr->grid.d_fourierGrid = kernelParamsPtr->grid.d_realGrid;
308 pmeGpu->archSpecific->complexGridSize = pmeGpu->archSpecific->realGridSize;
309 // the size might get used later for copying the grid
313 void pme_gpu_free_grids(const PmeGpu *pmeGpu)
315 if (pmeGpu->archSpecific->performOutOfPlaceFFT)
317 freeDeviceBuffer(&pmeGpu->kernelParams->grid.d_fourierGrid);
319 freeDeviceBuffer(&pmeGpu->kernelParams->grid.d_realGrid);
322 void pme_gpu_clear_grids(const PmeGpu *pmeGpu)
324 cudaError_t stat = cudaMemsetAsync(pmeGpu->kernelParams->grid.d_realGrid, 0,
325 pmeGpu->archSpecific->realGridSize * sizeof(float), pmeGpu->archSpecific->pmeStream);
326 /* Should the complex grid be cleared in some weird case? */
327 CU_RET_ERR(stat, "cudaMemsetAsync on the PME grid error");
330 void pme_gpu_realloc_and_copy_fract_shifts(PmeGpu *pmeGpu)
332 pme_gpu_free_fract_shifts(pmeGpu);
334 auto *kernelParamsPtr = pmeGpu->kernelParams.get();
336 const int nx = kernelParamsPtr->grid.realGridSize[XX];
337 const int ny = kernelParamsPtr->grid.realGridSize[YY];
338 const int nz = kernelParamsPtr->grid.realGridSize[ZZ];
339 const int cellCount = c_pmeNeighborUnitcellCount;
340 const int gridDataOffset[DIM] = {0, cellCount * nx, cellCount * (nx + ny)};
342 memcpy(kernelParamsPtr->grid.tablesOffsets, &gridDataOffset, sizeof(gridDataOffset));
344 const int newFractShiftsSize = cellCount * (nx + ny + nz);
346 initParamLookupTable(kernelParamsPtr->grid.d_fractShiftsTable,
347 kernelParamsPtr->fractShiftsTableTexture,
348 pmeGpu->common->fsh.data(),
352 initParamLookupTable(kernelParamsPtr->grid.d_gridlineIndicesTable,
353 kernelParamsPtr->gridlineIndicesTableTexture,
354 pmeGpu->common->nn.data(),
359 void pme_gpu_free_fract_shifts(const PmeGpu *pmeGpu)
361 auto *kernelParamsPtr = pmeGpu->kernelParams.get();
362 destroyParamLookupTable(kernelParamsPtr->grid.d_fractShiftsTable,
363 kernelParamsPtr->fractShiftsTableTexture,
365 destroyParamLookupTable(kernelParamsPtr->grid.d_gridlineIndicesTable,
366 kernelParamsPtr->gridlineIndicesTableTexture,
370 bool pme_gpu_stream_query(const PmeGpu *pmeGpu)
372 return haveStreamTasksCompleted(pmeGpu->archSpecific->pmeStream);
375 void pme_gpu_copy_input_gather_grid(const PmeGpu *pmeGpu, float *h_grid)
377 const size_t gridSize = pmeGpu->archSpecific->realGridSize * sizeof(float);
378 cu_copy_H2D(pmeGpu->kernelParams->grid.d_realGrid, h_grid, gridSize, pmeGpu->settings.transferKind, pmeGpu->archSpecific->pmeStream);
381 void pme_gpu_copy_output_spread_grid(const PmeGpu *pmeGpu, float *h_grid)
383 const size_t gridSize = pmeGpu->archSpecific->realGridSize * sizeof(float);
384 cu_copy_D2H(h_grid, pmeGpu->kernelParams->grid.d_realGrid, gridSize, pmeGpu->settings.transferKind, pmeGpu->archSpecific->pmeStream);
385 cudaError_t stat = cudaEventRecord(pmeGpu->archSpecific->syncSpreadGridD2H, pmeGpu->archSpecific->pmeStream);
386 CU_RET_ERR(stat, "PME spread grid sync event record failure");
389 void pme_gpu_copy_output_spread_atom_data(const PmeGpu *pmeGpu)
391 const int alignment = pme_gpu_get_atoms_per_warp(pmeGpu);
392 const size_t nAtomsPadded = ((pmeGpu->nAtomsAlloc + alignment - 1) / alignment) * alignment;
393 const size_t splinesSize = DIM * nAtomsPadded * pmeGpu->common->pme_order * sizeof(float);
394 auto *kernelParamsPtr = pmeGpu->kernelParams.get();
395 cu_copy_D2H(pmeGpu->staging.h_dtheta, kernelParamsPtr->atoms.d_dtheta, splinesSize, pmeGpu->settings.transferKind, pmeGpu->archSpecific->pmeStream);
396 cu_copy_D2H(pmeGpu->staging.h_theta, kernelParamsPtr->atoms.d_theta, splinesSize, pmeGpu->settings.transferKind, pmeGpu->archSpecific->pmeStream);
397 cu_copy_D2H(pmeGpu->staging.h_gridlineIndices, kernelParamsPtr->atoms.d_gridlineIndices,
398 kernelParamsPtr->atoms.nAtoms * DIM * sizeof(int), pmeGpu->settings.transferKind, pmeGpu->archSpecific->pmeStream);
401 void pme_gpu_copy_input_gather_atom_data(const PmeGpu *pmeGpu)
403 const int alignment = pme_gpu_get_atoms_per_warp(pmeGpu);
404 const size_t nAtomsPadded = ((pmeGpu->nAtomsAlloc + alignment - 1) / alignment) * alignment;
405 const size_t splinesSize = DIM * nAtomsPadded * pmeGpu->common->pme_order * sizeof(float);
406 auto *kernelParamsPtr = pmeGpu->kernelParams.get();
409 const size_t gridlineIndicesSizePerAtom = DIM * sizeof(int);
410 const size_t splineDataSizePerAtom = pmeGpu->common->pme_order * DIM * sizeof(float);
411 // TODO: could clear only the padding and not the whole thing, but this is a test-exclusive code anyway
412 CU_RET_ERR(cudaMemsetAsync(kernelParamsPtr->atoms.d_gridlineIndices, 0, pmeGpu->nAtomsAlloc * gridlineIndicesSizePerAtom, pmeGpu->archSpecific->pmeStream),
413 "PME failed to clear the gridline indices");
414 CU_RET_ERR(cudaMemsetAsync(kernelParamsPtr->atoms.d_dtheta, 0, pmeGpu->nAtomsAlloc * splineDataSizePerAtom, pmeGpu->archSpecific->pmeStream),
415 "PME failed to clear the spline derivatives");
416 CU_RET_ERR(cudaMemsetAsync(kernelParamsPtr->atoms.d_theta, 0, pmeGpu->nAtomsAlloc * splineDataSizePerAtom, pmeGpu->archSpecific->pmeStream),
417 "PME failed to clear the spline values");
419 cu_copy_H2D(kernelParamsPtr->atoms.d_dtheta, pmeGpu->staging.h_dtheta, splinesSize, pmeGpu->settings.transferKind, pmeGpu->archSpecific->pmeStream);
420 cu_copy_H2D(kernelParamsPtr->atoms.d_theta, pmeGpu->staging.h_theta, splinesSize, pmeGpu->settings.transferKind, pmeGpu->archSpecific->pmeStream);
421 cu_copy_H2D(kernelParamsPtr->atoms.d_gridlineIndices, pmeGpu->staging.h_gridlineIndices,
422 kernelParamsPtr->atoms.nAtoms * DIM * sizeof(int), pmeGpu->settings.transferKind, pmeGpu->archSpecific->pmeStream);
425 void pme_gpu_sync_spread_grid(const PmeGpu *pmeGpu)
427 cudaError_t stat = cudaEventSynchronize(pmeGpu->archSpecific->syncSpreadGridD2H);
428 CU_RET_ERR(stat, "Error while waiting for the PME GPU spread grid to be copied to the host");
431 void pme_gpu_init_internal(PmeGpu *pmeGpu)
433 /* Allocate the target-specific structures */
434 pmeGpu->archSpecific.reset(new PmeGpuSpecific());
435 pmeGpu->kernelParams.reset(new PmeGpuKernelParams());
437 pmeGpu->archSpecific->performOutOfPlaceFFT = true;
438 /* This should give better performance, according to the cuFFT documentation.
439 * The performance seems to be the same though.
440 * TODO: PME could also try to pick up nice grid sizes (with factors of 2, 3, 5, 7).
443 /* WARNING: CUDA timings are incorrect with multiple streams.
444 * This is the main reason why they are disabled by default.
446 // TODO: Consider turning on by default when we can detect nr of streams.
447 pmeGpu->archSpecific->useTiming = (getenv("GMX_ENABLE_GPU_TIMING") != nullptr);
449 pmeGpu->maxGridWidthX = pmeGpu->deviceInfo->prop.maxGridSize[0];
451 /* Creating a PME CUDA stream */
453 int highest_priority, lowest_priority;
454 stat = cudaDeviceGetStreamPriorityRange(&lowest_priority, &highest_priority);
455 CU_RET_ERR(stat, "PME cudaDeviceGetStreamPriorityRange failed");
456 stat = cudaStreamCreateWithPriority(&pmeGpu->archSpecific->pmeStream,
457 cudaStreamDefault, //cudaStreamNonBlocking,
459 CU_RET_ERR(stat, "cudaStreamCreateWithPriority on the PME stream failed");
462 void pme_gpu_destroy_specific(const PmeGpu *pmeGpu)
464 /* Destroy the CUDA stream */
465 cudaError_t stat = cudaStreamDestroy(pmeGpu->archSpecific->pmeStream);
466 CU_RET_ERR(stat, "PME cudaStreamDestroy error");
469 void pme_gpu_init_sync_events(const PmeGpu *pmeGpu)
471 const auto eventFlags = cudaEventDisableTiming;
472 CU_RET_ERR(cudaEventCreateWithFlags(&pmeGpu->archSpecific->syncSpreadGridD2H, eventFlags), "cudaEventCreate on syncSpreadGridD2H failed");
475 void pme_gpu_destroy_sync_events(const PmeGpu *pmeGpu)
477 CU_RET_ERR(cudaEventDestroy(pmeGpu->archSpecific->syncSpreadGridD2H), "cudaEventDestroy failed on syncSpreadGridD2H");
480 void pme_gpu_reinit_3dfft(const PmeGpu *pmeGpu)
482 if (pme_gpu_performs_FFT(pmeGpu))
484 pmeGpu->archSpecific->fftSetup.resize(0);
485 for (int i = 0; i < pmeGpu->common->ngrids; i++)
487 pmeGpu->archSpecific->fftSetup.push_back(std::unique_ptr<GpuParallel3dFft>(new GpuParallel3dFft(pmeGpu)));
492 void pme_gpu_destroy_3dfft(const PmeGpu *pmeGpu)
494 pmeGpu->archSpecific->fftSetup.resize(0);
497 int getSplineParamFullIndex(int order, int splineIndex, int dimIndex, int warpIndex, int atomWarpIndex)
503 constexpr int fixedOrder = 4;
504 GMX_UNUSED_VALUE(fixedOrder);
505 const int indexBase = getSplineParamIndexBase<fixedOrder>(warpIndex, atomWarpIndex);
506 return getSplineParamIndex<fixedOrder>(indexBase, dimIndex, splineIndex);