2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
5 * Copyright (c) 2001-2012, The GROMACS development team.
6 * Copyright (c) 2013-2019,2020, 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.
40 * Data types used internally in the nbnxn_cuda module.
42 * \author Szilárd Páll <pall.szilard@gmail.com>
43 * \ingroup module_nbnxm
46 #ifndef NBNXM_CUDA_TYPES_H
47 #define NBNXM_CUDA_TYPES_H
49 #include "gromacs/gpu_utils/cuda_arch_utils.cuh"
50 #include "gromacs/gpu_utils/cudautils.cuh"
51 #include "gromacs/gpu_utils/devicebuffer.h"
52 #include "gromacs/gpu_utils/gputraits.cuh"
53 #include "gromacs/mdtypes/interaction_const.h"
54 #include "gromacs/nbnxm/gpu_types_common.h"
55 #include "gromacs/nbnxm/nbnxm.h"
56 #include "gromacs/nbnxm/pairlist.h"
57 #include "gromacs/timing/gpu_timing.h"
58 #include "gromacs/utility/enumerationhelpers.h"
60 /*! \brief Macro definining default for the prune kernel's j4 processing concurrency.
62 * The GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY macro allows compile-time override.
64 #ifndef GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY
65 # define GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY 4
67 /*! \brief Default for the prune kernel's j4 processing concurrency.
69 * Initialized using the #GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY macro which allows compile-time override.
71 const int c_cudaPruneKernelJ4Concurrency = GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY;
73 /* TODO: consider moving this to kernel_utils */
74 /* Convenience defines */
75 /*! \brief cluster size = number of atoms per cluster. */
76 static constexpr int c_clSize = c_nbnxnGpuClusterSize;
78 /*! \brief Electrostatic CUDA kernel flavors.
80 * Types of electrostatics implementations available in the CUDA non-bonded
81 * force kernels. These represent both the electrostatics types implemented
82 * by the kernels (cut-off, RF, and Ewald - a subset of what's defined in
83 * enums.h) as well as encode implementation details analytical/tabulated
84 * and single or twin cut-off (for Ewald kernels).
85 * Note that the cut-off and RF kernels have only analytical flavor and unlike
86 * in the CPU kernels, the tabulated kernels are ATM Ewald-only.
88 * The row-order of pointers to different electrostatic kernels defined in
89 * nbnxn_cuda.cu by the nb_*_kfunc_ptr function pointer table
90 * should match the order of enumerated types below.
103 /*! \brief VdW CUDA kernel flavors.
105 * The enumerates values correspond to the LJ implementations in the CUDA non-bonded
108 * The column-order of pointers to different electrostatic kernels defined in
109 * nbnxn_cuda.cu by the nb_*_kfunc_ptr function pointer table
110 * should match the order of enumerated types below.
124 /* All structs prefixed with "cu_" hold data used in GPU calculations and
125 * are passed to the kernels, except cu_timers_t. */
127 typedef struct cu_atomdata cu_atomdata_t;
128 typedef struct cu_nbparam cu_nbparam_t;
133 * \brief Staging area for temporary data downloaded from the GPU.
135 * The energies/shift forces get downloaded here first, before getting added
136 * to the CPU-side aggregate values.
141 float* e_lj = nullptr;
142 //! electrostatic energy
143 float* e_el = nullptr;
145 float3* fshift = nullptr;
149 * \brief Nonbonded atom data - both inputs and outputs.
155 //! number of local atoms
157 //! allocation size for the atom data (xq, f)
160 //! atom coordinates + charges, size natoms
162 //! force output array, size natoms
165 //! LJ energy output, size 1
167 //! Electrostatics energy input, size 1
173 //! number of atom types
175 //! atom type indices, size natoms
177 //! sqrt(c6),sqrt(c12) size natoms
182 //! true if the shift vector has been uploaded
183 bool bShiftVecUploaded;
187 * \brief Parameters required for the CUDA nonbonded calculations.
192 //! type of electrostatics, takes values from #eelCu
194 //! type of VdW impl., takes values from #evdwCu
197 //! charge multiplication factor
199 //! Reaction-field/plain cutoff electrostatics const.
201 //! Reaction-field electrostatics constant
203 //! Ewald/PME parameter
205 //! Ewald/PME correction term substracted from the direct-space potential
207 //! LJ-Ewald/PME correction term added to the correction potential
209 //! LJ-Ewald/PME coefficient
212 //! Coulomb cut-off squared
215 //! VdW cut-off squared
217 //! VdW switched cut-off
219 //! Full, outer pair-list cut-off squared
221 //! Inner, dynamic pruned pair-list cut-off squared
223 //! True if we use dynamic pair-list pruning
224 bool useDynamicPruning;
226 //! VdW shift dispersion constants
227 shift_consts_t dispersion_shift;
228 //! VdW shift repulsion constants
229 shift_consts_t repulsion_shift;
230 //! VdW switch constants
231 switch_consts_t vdw_switch;
233 /* LJ non-bonded parameters - accessed through texture memory */
234 //! nonbonded parameter table with C6/C12 pairs per atom type-pair, 2*ntype^2 elements
236 //! texture object bound to nbfp
237 cudaTextureObject_t nbfp_texobj;
238 //! nonbonded parameter table per atom type, 2*ntype elements
240 //! texture object bound to nbfp_texobj
241 cudaTextureObject_t nbfp_comb_texobj;
243 /* Ewald Coulomb force table data - accessed through texture memory */
244 //! table scale/spacing
245 float coulomb_tab_scale;
246 //! pointer to the table in the device memory
248 //! texture object bound to coulomb_tab
249 cudaTextureObject_t coulomb_tab_texobj;
253 * \brief Pair list data.
255 using cu_plist_t = Nbnxm::gpu_plist;
258 * \brief Typedef of actual timer type.
260 typedef struct Nbnxm::gpu_timers_t cu_timers_t;
262 class GpuEventSynchronizer;
265 * \brief Main data structure for CUDA nonbonded force calculations.
269 /*! \brief GPU device context.
271 * \todo Make it constant reference, once NbnxmGpu is a proper class.
273 const DeviceContext* deviceContext_;
274 /*! \brief true if doing both local/non-local NB work on GPU */
275 bool bUseTwoStreams = false;
276 /*! \brief atom data */
277 cu_atomdata_t* atdat = nullptr;
278 /*! \brief f buf ops cell index mapping */
280 /*! \brief number of indices in cell buffer */
282 /*! \brief number of indices allocated in cell buffer */
284 /*! \brief array of atom indices */
285 int* atomIndices = nullptr;
286 /*! \brief size of atom indices */
287 int atomIndicesSize = 0;
288 /*! \brief size of atom indices allocated in device buffer */
289 int atomIndicesSize_alloc = 0;
290 /*! \brief x buf ops num of atoms */
291 int* cxy_na = nullptr;
292 /*! \brief number of elements in cxy_na */
294 /*! \brief number of elements allocated allocated in device buffer */
295 int ncxy_na_alloc = 0;
296 /*! \brief x buf ops cell index mapping */
297 int* cxy_ind = nullptr;
298 /*! \brief number of elements in cxy_ind */
300 /*! \brief number of elements allocated allocated in device buffer */
301 int ncxy_ind_alloc = 0;
302 /*! \brief parameters required for the non-bonded calc. */
303 cu_nbparam_t* nbparam = nullptr;
304 /*! \brief pair-list data structures (local and non-local) */
305 gmx::EnumerationArray<Nbnxm::InteractionLocality, cu_plist_t*> plist = { { nullptr } };
306 /*! \brief staging area where fshift/energies get downloaded */
308 /*! \brief local and non-local GPU streams */
309 gmx::EnumerationArray<Nbnxm::InteractionLocality, const DeviceStream*> deviceStreams;
311 /*! \brief Events used for synchronization */
313 /*! \brief Event triggered when the non-local non-bonded
314 * kernel is done (and the local transfer can proceed) */
315 cudaEvent_t nonlocal_done = nullptr;
316 /*! \brief Event triggered when the tasks issued in the local
317 * stream that need to precede the non-local force or buffer
318 * operation calculations are done (e.g. f buffer 0-ing, local
319 * x/q H2D, buffer op initialization in local stream that is
320 * required also by nonlocal stream ) */
321 cudaEvent_t misc_ops_and_local_H2D_done = nullptr;
324 /*! \brief True if there is work for the current domain in the
325 * respective locality.
327 * This includes local/nonlocal GPU work, either bonded or
328 * nonbonded, scheduled to be executed in the current
329 * domain. As long as bonded work is not split up into
330 * local/nonlocal, if there is bonded GPU work, both flags
332 gmx::EnumerationArray<Nbnxm::InteractionLocality, bool> haveWork = { { false } };
334 /*! \brief Pointer to event synchronizer triggered when the local
335 * GPU buffer ops / reduction is complete
337 * \note That the synchronizer is managed outside of this module
338 * in StatePropagatorDataGpu.
340 GpuEventSynchronizer* localFReductionDone = nullptr;
342 /*! \brief Event triggered when non-local coordinate buffer
343 * has been copied from device to host. */
344 GpuEventSynchronizer* xNonLocalCopyD2HDone = nullptr;
346 /* NOTE: With current CUDA versions (<=5.0) timing doesn't work with multiple
347 * concurrent streams, so we won't time if both l/nl work is done on GPUs.
348 * Timer init/uninit is still done even with timing off so only the condition
349 * setting bDoTime needs to be change if this CUDA "feature" gets fixed. */
350 /*! \brief True if event-based timing is enabled. */
351 bool bDoTime = false;
352 /*! \brief CUDA event-based timers. */
353 cu_timers_t* timers = nullptr;
354 /*! \brief Timing data. TODO: deprecate this and query timers for accumulated data instead */
355 gmx_wallclock_gpu_nbnxn_t* timings = nullptr;
358 #endif /* NBNXN_CUDA_TYPES_H */