2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2016 by the GROMACS development team.
5 * Copyright (c) 2017,2018,2019,2020, by the GROMACS development team, led by
6 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
7 * and including many others, as listed in the AUTHORS file in the
8 * top-level source directory and at http://www.gromacs.org.
10 * GROMACS is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public License
12 * as published by the Free Software Foundation; either version 2.1
13 * of the License, or (at your option) any later version.
15 * GROMACS is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with GROMACS; if not, see
22 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
23 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
25 * If you want to redistribute modifications to GROMACS, please
26 * consider that scientific software is very special. Version
27 * control is crucial - bugs must be traceable. We will be happy to
28 * consider code for inclusion in the official distribution, but
29 * derived work must not be called official GROMACS. Details are found
30 * in the README & COPYING files - if they are missing, get the
31 * official version at http://www.gromacs.org.
33 * To help us fund GROMACS development, we humbly ask that you cite
34 * the research papers on the package. Check out http://www.gromacs.org.
39 * Data types used internally in the nbnxm_ocl module.
41 * \author Anca Hamuraru <anca@streamcomputing.eu>
42 * \author Szilárd Páll <pszilard@kth.se>
43 * \ingroup module_nbnxm
46 #ifndef GMX_NBNXM_NBNXM_OPENCL_TYPES_H
47 #define GMX_NBNXM_NBNXM_OPENCL_TYPES_H
49 #include "gromacs/gpu_utils/devicebuffer.h"
50 #include "gromacs/gpu_utils/gmxopencl.h"
51 #include "gromacs/gpu_utils/gputraits_ocl.h"
52 #include "gromacs/gpu_utils/oclutils.h"
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/utility/enumerationhelpers.h"
58 #include "gromacs/utility/fatalerror.h"
59 #include "gromacs/utility/real.h"
61 #include "nbnxm_ocl_consts.h"
63 struct gmx_wallclock_gpu_nbnxn_t;
65 /* kernel does #include "gromacs/math/utilities.h" */
66 /* Move the actual useful stuff here: */
69 #define M_FLOAT_1_SQRTPI 0.564189583547756f
72 /*! \brief Constants for platform-dependent defaults for the prune kernel's j4 processing concurrency.
74 * Initialized using macros that can be overridden at compile-time (using #GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY).
77 const int c_oclPruneKernelJ4ConcurrencyDEFAULT = GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY_DEFAULT;
80 /*! \brief Electrostatic OpenCL kernel flavors.
82 * Types of electrostatics implementations available in the OpenCL 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.
101 eelOclEWALD_ANA_TWIN,
105 /*! \brief VdW OpenCL kernel flavors.
107 * The enumerates values correspond to the LJ implementations in the OpenCL 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 /*! \brief Pruning kernel flavors.
128 * The values correspond to the first call of the pruning post-list generation
129 * and the rolling pruning, respectively.
139 * \brief Staging area for temporary data downloaded from the GPU.
141 * The energies/shift forces get downloaded here first, before getting added
142 * to the CPU-side aggregate values.
147 float* e_lj = nullptr;
148 //! electrostatic energy
149 float* e_el = nullptr;
150 //! float3 buffer with shift forces
151 float (*fshift)[3] = nullptr;
155 * \brief Nonbonded atom data - both inputs and outputs.
157 typedef struct cl_atomdata
161 //! number of local atoms
163 //! allocation size for the atom data (xq, f)
166 //! float4 buffer with atom coordinates + charges, size natoms
167 DeviceBuffer<float> xq;
169 //! float3 buffer with force output array, size natoms
170 DeviceBuffer<float> f;
172 //! LJ energy output, size 1
173 DeviceBuffer<float> e_lj;
174 //! Electrostatics energy input, size 1
175 DeviceBuffer<float> e_el;
177 //! float3 buffer with shift forces
178 DeviceBuffer<float> fshift;
180 //! number of atom types
182 //! int buffer with atom type indices, size natoms
183 DeviceBuffer<int> atom_types;
184 //! float2 buffer with sqrt(c6),sqrt(c12), size natoms
185 DeviceBuffer<float> lj_comb;
187 //! float3 buffer with shifts values
188 DeviceBuffer<float> shift_vec;
190 //! true if the shift vector has been uploaded
191 bool bShiftVecUploaded;
195 * \brief Parameters required for the OpenCL nonbonded calculations.
197 typedef struct cl_nbparam
200 //! type of electrostatics, takes values from #eelOcl
202 //! type of VdW impl., takes values from #evdwOcl
205 //! charge multiplication factor
207 //! Reaction-field/plain cutoff electrostatics const.
209 //! Reaction-field electrostatics constant
211 //! Ewald/PME parameter
213 //! Ewald/PME correction term substracted from the direct-space potential
215 //! LJ-Ewald/PME correction term added to the correction potential
217 //! LJ-Ewald/PME coefficient
220 //! Coulomb cut-off squared
223 //! VdW cut-off squared
225 //! VdW switched cut-off
227 //! Full, outer pair-list cut-off squared
229 //! Inner, dynamic pruned pair-list cut-off squared
231 //! True if we use dynamic pair-list pruning
232 bool useDynamicPruning;
234 //! VdW shift dispersion constants
235 shift_consts_t dispersion_shift;
236 //! VdW shift repulsion constants
237 shift_consts_t repulsion_shift;
238 //! VdW switch constants
239 switch_consts_t vdw_switch;
241 /* LJ non-bonded parameters - accessed through texture memory */
242 //! nonbonded parameter table with C6/C12 pairs per atom type-pair, 2*ntype^2 elements
244 //! nonbonded parameter table per atom type, 2*ntype elements
245 cl_mem nbfp_comb_climg2d;
247 /* Ewald Coulomb force table data - accessed through texture memory */
248 //! table scale/spacing
249 float coulomb_tab_scale;
250 //! pointer to the table in the device memory
251 cl_mem coulomb_tab_climg2d;
255 * \brief Data structure shared between the OpenCL device code and OpenCL host code
257 * Must not contain OpenCL objects (buffers)
258 * TODO: review, improve */
259 typedef struct cl_nbparam_params
262 //! type of electrostatics, takes values from #eelCu
264 //! type of VdW impl., takes values from #evdwCu
267 //! charge multiplication factor
269 //! Reaction-field/plain cutoff electrostatics const.
271 //! Reaction-field electrostatics constant
273 //! Ewald/PME parameter
275 //! Ewald/PME correction term substracted from the direct-space potential
277 //! LJ-Ewald/PME correction term added to the correction potential
279 //! LJ-Ewald/PME coefficient
282 //! Coulomb cut-off squared
285 //! VdW cut-off squared
287 //! VdW switched cut-off
289 //! Full, outer pair-list cut-off squared
291 //! Inner, dynamic pruned pair-list cut-off squared
294 //! VdW shift dispersion constants
295 shift_consts_t dispersion_shift;
296 //! VdW shift repulsion constants
297 shift_consts_t repulsion_shift;
298 //! VdW switch constants
299 switch_consts_t vdw_switch;
301 /* Ewald Coulomb force table data - accessed through texture memory */
302 //! table scale/spacing
303 float coulomb_tab_scale;
304 } cl_nbparam_params_t;
308 * \brief Pair list data.
310 using cl_plist_t = Nbnxm::gpu_plist;
313 * \brief Typedef of actual timer type.
315 typedef struct Nbnxm::gpu_timers_t cl_timers_t;
318 * \brief Main data structure for OpenCL nonbonded force calculations.
322 /* \brief OpenCL device context
324 * \todo Make it constant reference, once NbnxmGpu is a proper class.
326 const DeviceContext* deviceContext_;
327 //! OpenCL runtime data (context, kernels)
328 struct gmx_device_runtime_data_t* dev_rundata = nullptr;
330 /**< Pointers to non-bonded kernel functions
331 * organized similar with nb_kfunc_xxx arrays in nbnxn_ocl.cpp */
333 cl_kernel kernel_noener_noprune_ptr[eelOclNR][evdwOclNR] = { { nullptr } };
334 cl_kernel kernel_ener_noprune_ptr[eelOclNR][evdwOclNR] = { { nullptr } };
335 cl_kernel kernel_noener_prune_ptr[eelOclNR][evdwOclNR] = { { nullptr } };
336 cl_kernel kernel_ener_prune_ptr[eelOclNR][evdwOclNR] = { { nullptr } };
338 //! prune kernels, ePruneKind defined the kernel kinds
339 cl_kernel kernel_pruneonly[ePruneNR] = { nullptr };
341 //! true if prefetching fg i-atom LJ parameters should be used in the kernels
342 bool bPrefetchLjParam = false;
344 /**< auxiliary kernels implementing memset-like functions */
346 cl_kernel kernel_memset_f = nullptr;
347 cl_kernel kernel_memset_f2 = nullptr;
348 cl_kernel kernel_memset_f3 = nullptr;
349 cl_kernel kernel_zero_e_fshift = nullptr;
352 //! true if doing both local/non-local NB work on GPU
353 bool bUseTwoStreams = false;
354 //! true indicates that the nonlocal_done event was enqueued
355 bool bNonLocalStreamActive = false;
358 cl_atomdata_t* atdat = nullptr;
359 //! parameters required for the non-bonded calc.
360 cl_nbparam_t* nbparam = nullptr;
361 //! pair-list data structures (local and non-local)
362 gmx::EnumerationArray<Nbnxm::InteractionLocality, cl_plist_t*> plist = { nullptr };
363 //! staging area where fshift/energies get downloaded
366 //! local and non-local GPU queues
367 gmx::EnumerationArray<Nbnxm::InteractionLocality, const DeviceStream*> deviceStreams;
369 /*! \brief Events used for synchronization */
371 /*! \brief Event triggered when the non-local non-bonded
372 * kernel is done (and the local transfer can proceed) */
373 cl_event nonlocal_done = nullptr;
374 /*! \brief Event triggered when the tasks issued in the local
375 * stream that need to precede the non-local force or buffer
376 * operation calculations are done (e.g. f buffer 0-ing, local
377 * x/q H2D, buffer op initialization in local stream that is
378 * required also by nonlocal stream ) */
379 cl_event misc_ops_and_local_H2D_done = nullptr;
382 //! True if there has been local/nonlocal GPU work, either bonded or nonbonded, scheduled
383 // to be executed in the current domain. As long as bonded work is not split up into
384 // local/nonlocal, if there is bonded GPU work, both flags will be true.
385 gmx::EnumerationArray<Nbnxm::InteractionLocality, bool> haveWork;
388 //! True if event-based timing is enabled.
389 bool bDoTime = false;
390 //! OpenCL event-based timers.
391 cl_timers_t* timers = nullptr;
392 //! Timing data. TODO: deprecate this and query timers for accumulated data instead
393 gmx_wallclock_gpu_nbnxn_t* timings = nullptr;
396 #endif /* NBNXN_OPENCL_TYPES_H */