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 number of clusters per supercluster. */
76 static const int c_numClPerSupercl = c_nbnxnGpuNumClusterPerSupercluster;
77 /*! \brief cluster size = number of atoms per cluster. */
78 static const int c_clSize = c_nbnxnGpuClusterSize;
80 /*! \brief Electrostatic CUDA kernel flavors.
82 * Types of electrostatics implementations available in the CUDA non-bonded
83 * force kernels. These represent both the electrostatics types implemented
84 * by the kernels (cut-off, RF, and Ewald - a subset of what's defined in
85 * enums.h) as well as encode implementation details analytical/tabulated
86 * and single or twin cut-off (for Ewald kernels).
87 * Note that the cut-off and RF kernels have only analytical flavor and unlike
88 * in the CPU kernels, the tabulated kernels are ATM Ewald-only.
90 * The row-order of pointers to different electrostatic kernels defined in
91 * nbnxn_cuda.cu by the nb_*_kfunc_ptr function pointer table
92 * should match the order of enumerated types below.
105 /*! \brief VdW CUDA kernel flavors.
107 * The enumerates values correspond to the LJ implementations in the CUDA non-bonded
110 * The column-order of pointers to different electrostatic kernels defined in
111 * nbnxn_cuda.cu by the nb_*_kfunc_ptr function pointer table
112 * should match the order of enumerated types below.
126 /* All structs prefixed with "cu_" hold data used in GPU calculations and
127 * are passed to the kernels, except cu_timers_t. */
129 typedef struct cu_atomdata cu_atomdata_t;
130 typedef struct cu_nbparam cu_nbparam_t;
131 typedef struct nb_staging nb_staging_t;
136 * \brief Staging area for temporary data downloaded from the GPU.
138 * The energies/shift forces get downloaded here first, before getting added
139 * to the CPU-side aggregate values.
143 float* e_lj; /**< LJ energy */
144 float* e_el; /**< electrostatic energy */
145 float3* fshift; /**< shift forces */
149 * \brief Nonbonded atom data - both inputs and outputs.
153 int natoms; /**< number of atoms */
154 int natoms_local; /**< number of local atoms */
155 int nalloc; /**< allocation size for the atom data (xq, f) */
157 float4* xq; /**< atom coordinates + charges, size natoms */
158 float3* f; /**< force output array, size natoms */
160 float* e_lj; /**< LJ energy output, size 1 */
161 float* e_el; /**< Electrostatics energy input, size 1 */
163 float3* fshift; /**< shift forces */
165 int ntypes; /**< number of atom types */
166 int* atom_types; /**< atom type indices, size natoms */
167 float2* lj_comb; /**< sqrt(c6),sqrt(c12) size natoms */
169 float3* shift_vec; /**< shifts */
170 bool bShiftVecUploaded; /**< true if the shift vector has been uploaded */
174 * \brief Parameters required for the CUDA nonbonded calculations.
179 int eeltype; /**< type of electrostatics, takes values from #eelCu */
180 int vdwtype; /**< type of VdW impl., takes values from #evdwCu */
182 float epsfac; /**< charge multiplication factor */
183 float c_rf; /**< Reaction-field/plain cutoff electrostatics const. */
184 float two_k_rf; /**< Reaction-field electrostatics constant */
185 float ewald_beta; /**< Ewald/PME parameter */
186 float sh_ewald; /**< Ewald/PME correction term substracted from the direct-space potential */
187 float sh_lj_ewald; /**< LJ-Ewald/PME correction term added to the correction potential */
188 float ewaldcoeff_lj; /**< LJ-Ewald/PME coefficient */
190 float rcoulomb_sq; /**< Coulomb cut-off squared */
192 float rvdw_sq; /**< VdW cut-off squared */
193 float rvdw_switch; /**< VdW switched cut-off */
194 float rlistOuter_sq; /**< Full, outer pair-list cut-off squared */
195 float rlistInner_sq; /**< Inner, dynamic pruned pair-list cut-off squared */
196 bool useDynamicPruning; /**< True if we use dynamic pair-list pruning */
198 shift_consts_t dispersion_shift; /**< VdW shift dispersion constants */
199 shift_consts_t repulsion_shift; /**< VdW shift repulsion constants */
200 switch_consts_t vdw_switch; /**< VdW switch constants */
202 /* LJ non-bonded parameters - accessed through texture memory */
203 float* nbfp; /**< nonbonded parameter table with C6/C12 pairs per atom type-pair, 2*ntype^2 elements */
204 cudaTextureObject_t nbfp_texobj; /**< texture object bound to nbfp */
205 float* nbfp_comb; /**< nonbonded parameter table per atom type, 2*ntype elements */
206 cudaTextureObject_t nbfp_comb_texobj; /**< texture object bound to nbfp_texobj */
208 /* Ewald Coulomb force table data - accessed through texture memory */
209 float coulomb_tab_scale; /**< table scale/spacing */
210 float* coulomb_tab; /**< pointer to the table in the device memory */
211 cudaTextureObject_t coulomb_tab_texobj; /**< texture object bound to coulomb_tab */
215 * \brief Pair list data.
217 using cu_plist_t = Nbnxm::gpu_plist;
220 * \brief Typedef of actual timer type.
222 typedef struct Nbnxm::gpu_timers_t cu_timers_t;
224 class GpuEventSynchronizer;
227 * \brief Main data structure for CUDA nonbonded force calculations.
229 struct gmx_nbnxm_gpu_t
231 /*! \brief CUDA device information */
232 const gmx_device_info_t* dev_info;
233 /*! \brief true if doing both local/non-local NB work on GPU */
235 /*! \brief atom data */
236 cu_atomdata_t* atdat;
237 /*! \brief f buf ops cell index mapping */
239 /*! \brief number of indices in cell buffer */
241 /*! \brief number of indices allocated in cell buffer */
243 /*! \brief array of atom indices */
245 /*! \brief size of atom indices */
247 /*! \brief size of atom indices allocated in device buffer */
248 int atomIndicesSize_alloc;
249 /*! \brief x buf ops num of atoms */
251 /*! \brief number of elements in cxy_na */
253 /*! \brief number of elements allocated allocated in device buffer */
255 /*! \brief x buf ops cell index mapping */
257 /*! \brief number of elements in cxy_ind */
259 /*! \brief number of elements allocated allocated in device buffer */
261 /*! \brief parameters required for the non-bonded calc. */
262 cu_nbparam_t* nbparam;
263 /*! \brief pair-list data structures (local and non-local) */
264 gmx::EnumerationArray<Nbnxm::InteractionLocality, cu_plist_t*> plist;
265 /*! \brief staging area where fshift/energies get downloaded */
267 /*! \brief local and non-local GPU streams */
268 gmx::EnumerationArray<Nbnxm::InteractionLocality, cudaStream_t> stream;
270 /*! \brief Events used for synchronization */
272 /*! \brief Event triggered when the non-local non-bonded
273 * kernel is done (and the local transfer can proceed) */
274 cudaEvent_t nonlocal_done;
275 /*! \brief Event triggered when the tasks issued in the local
276 * stream that need to precede the non-local force or buffer
277 * operation calculations are done (e.g. f buffer 0-ing, local
278 * x/q H2D, buffer op initialization in local stream that is
279 * required also by nonlocal stream ) */
280 cudaEvent_t misc_ops_and_local_H2D_done;
283 /*! \brief True if there is work for the current domain in the
284 * respective locality.
286 * This includes local/nonlocal GPU work, either bonded or
287 * nonbonded, scheduled to be executed in the current
288 * domain. As long as bonded work is not split up into
289 * local/nonlocal, if there is bonded GPU work, both flags
291 gmx::EnumerationArray<Nbnxm::InteractionLocality, bool> haveWork;
293 /*! \brief Pointer to event synchronizer triggered when the local
294 * GPU buffer ops / reduction is complete
296 * \note That the synchronizer is managed outside of this module
297 * in StatePropagatorDataGpu.
299 GpuEventSynchronizer* localFReductionDone;
301 /*! \brief Event triggered when non-local coordinate buffer
302 * has been copied from device to host. */
303 GpuEventSynchronizer* xNonLocalCopyD2HDone;
305 /* NOTE: With current CUDA versions (<=5.0) timing doesn't work with multiple
306 * concurrent streams, so we won't time if both l/nl work is done on GPUs.
307 * Timer init/uninit is still done even with timing off so only the condition
308 * setting bDoTime needs to be change if this CUDA "feature" gets fixed. */
309 /*! \brief True if event-based timing is enabled. */
311 /*! \brief CUDA event-based timers. */
313 /*! \brief Timing data. TODO: deprecate this and query timers for accumulated data instead */
314 gmx_wallclock_gpu_nbnxn_t* timings;
317 #endif /* NBNXN_CUDA_TYPES_H */