2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2016,2017,2018,2019, 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.
38 * Data types used internally in the nbnxm_ocl module.
40 * \author Anca Hamuraru <anca@streamcomputing.eu>
41 * \author Szilárd Páll <pszilard@kth.se>
42 * \ingroup module_nbnxm
45 #ifndef GMX_NBNXM_NBNXM_OPENCL_TYPES_H
46 #define GMX_NBNXM_NBNXM_OPENCL_TYPES_H
48 #include "gromacs/gpu_utils/devicebuffer.h"
49 #include "gromacs/gpu_utils/gmxopencl.h"
50 #include "gromacs/gpu_utils/gputraits_ocl.h"
51 #include "gromacs/gpu_utils/oclutils.h"
52 #include "gromacs/mdtypes/interaction_const.h"
53 #include "gromacs/nbnxm/gpu_types_common.h"
54 #include "gromacs/nbnxm/nbnxm.h"
55 #include "gromacs/nbnxm/pairlist.h"
56 #include "gromacs/utility/enumerationhelpers.h"
57 #include "gromacs/utility/fatalerror.h"
58 #include "gromacs/utility/real.h"
60 #include "nbnxm_ocl_consts.h"
62 /* kernel does #include "gromacs/math/utilities.h" */
63 /* Move the actual useful stuff here: */
66 #define M_FLOAT_1_SQRTPI 0.564189583547756f
69 /*! \brief Constants for platform-dependent defaults for the prune kernel's j4 processing concurrency.
71 * Initialized using macros that can be overridden at compile-time (using #GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY).
74 const int c_oclPruneKernelJ4ConcurrencyDEFAULT = GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY_DEFAULT;
77 /*! \brief Returns the j4 processing concurrency parameter for the vendor \p vendorId
78 * \param vendorId takes values from #ocl_vendor_id_t.
80 static inline int getOclPruneKernelJ4Concurrency(int vendorId)
84 default: return c_oclPruneKernelJ4ConcurrencyDEFAULT;
89 /*! \brief Electrostatic OpenCL kernel flavors.
91 * Types of electrostatics implementations available in the OpenCL non-bonded
92 * force kernels. These represent both the electrostatics types implemented
93 * by the kernels (cut-off, RF, and Ewald - a subset of what's defined in
94 * enums.h) as well as encode implementation details analytical/tabulated
95 * and single or twin cut-off (for Ewald kernels).
96 * Note that the cut-off and RF kernels have only analytical flavor and unlike
97 * in the CPU kernels, the tabulated kernels are ATM Ewald-only.
99 * The row-order of pointers to different electrostatic kernels defined in
100 * nbnxn_cuda.cu by the nb_*_kfunc_ptr function pointer table
101 * should match the order of enumerated types below.
104 eelOclCUT, eelOclRF, eelOclEWALD_TAB, eelOclEWALD_TAB_TWIN, eelOclEWALD_ANA, eelOclEWALD_ANA_TWIN, eelOclNR
107 /*! \brief VdW OpenCL kernel flavors.
109 * The enumerates values correspond to the LJ implementations in the OpenCL non-bonded
112 * The column-order of pointers to different electrostatic kernels defined in
113 * nbnxn_cuda.cu by the nb_*_kfunc_ptr function pointer table
114 * should match the order of enumerated types below.
117 evdwOclCUT, evdwOclCUTCOMBGEOM, evdwOclCUTCOMBLB, evdwOclFSWITCH, evdwOclPSWITCH, evdwOclEWALDGEOM, evdwOclEWALDLB, evdwOclNR
120 /*! \brief Pruning kernel flavors.
122 * The values correspond to the first call of the pruning post-list generation
123 * and the rolling pruning, respectively.
126 epruneFirst, epruneRolling, ePruneNR
130 * \brief Staging area for temporary data downloaded from the GPU.
132 * The energies/shift forces get downloaded here first, before getting added
133 * to the CPU-side aggregate values.
135 typedef struct cl_nb_staging
137 float *e_lj; /**< LJ energy */
138 float *e_el; /**< electrostatic energy */
139 float (*fshift)[3]; /**< float3 buffer with shift forces */
143 * \brief Nonbonded atom data - both inputs and outputs.
145 typedef struct cl_atomdata
147 int natoms; /**< number of atoms */
148 int natoms_local; /**< number of local atoms */
149 int nalloc; /**< allocation size for the atom data (xq, f) */
151 cl_mem xq; /**< float4 buffer with atom coordinates + charges, size natoms */
153 cl_mem f; /**< float3 buffer with force output array, size natoms */
154 size_t f_elem_size; /**< Size in bytes for one element of f buffer */
156 cl_mem e_lj; /**< LJ energy output, size 1 */
157 cl_mem e_el; /**< Electrostatics energy input, size 1 */
159 cl_mem fshift; /**< float3 buffer with shift forces */
160 size_t fshift_elem_size; /**< Size in bytes for one element of fshift buffer */
162 int ntypes; /**< number of atom types */
163 cl_mem atom_types; /**< int buffer with atom type indices, size natoms */
164 cl_mem lj_comb; /**< float2 buffer with sqrt(c6),sqrt(c12), size natoms */
166 cl_mem shift_vec; /**< float3 buffer with shifts values */
167 size_t shift_vec_elem_size; /**< Size in bytes for one element of shift_vec buffer */
169 cl_bool bShiftVecUploaded; /**< true if the shift vector has been uploaded */
173 * \brief Parameters required for the OpenCL nonbonded calculations.
175 typedef struct cl_nbparam
178 int eeltype; /**< type of electrostatics, takes values from #eelOcl */
179 int vdwtype; /**< type of VdW impl., takes values from #evdwOcl */
181 float epsfac; /**< charge multiplication factor */
182 float c_rf; /**< Reaction-field/plain cutoff electrostatics const. */
183 float two_k_rf; /**< Reaction-field electrostatics constant */
184 float ewald_beta; /**< Ewald/PME parameter */
185 float sh_ewald; /**< Ewald/PME correction term substracted from the direct-space potential */
186 float sh_lj_ewald; /**< LJ-Ewald/PME correction term added to the correction potential */
187 float ewaldcoeff_lj; /**< LJ-Ewald/PME coefficient */
189 float rcoulomb_sq; /**< Coulomb cut-off squared */
191 float rvdw_sq; /**< VdW cut-off squared */
192 float rvdw_switch; /**< VdW switched cut-off */
193 float rlistOuter_sq; /**< Full, outer pair-list cut-off squared */
194 float rlistInner_sq; /**< Inner, dynamic pruned pair-list cut-off squared */
195 bool useDynamicPruning; /**< True if we use dynamic pair-list pruning */
197 shift_consts_t dispersion_shift; /**< VdW shift dispersion constants */
198 shift_consts_t repulsion_shift; /**< VdW shift repulsion constants */
199 switch_consts_t vdw_switch; /**< VdW switch constants */
201 /* LJ non-bonded parameters - accessed through texture memory */
202 cl_mem nbfp_climg2d; /**< nonbonded parameter table with C6/C12 pairs per atom type-pair, 2*ntype^2 elements */
203 cl_mem nbfp_comb_climg2d; /**< nonbonded parameter table per atom type, 2*ntype elements */
205 /* Ewald Coulomb force table data - accessed through texture memory */
206 float coulomb_tab_scale; /**< table scale/spacing */
207 cl_mem coulomb_tab_climg2d; /**< pointer to the table in the device memory */
211 * \brief Data structure shared between the OpenCL device code and OpenCL host code
213 * Must not contain OpenCL objects (buffers)
214 * TODO: review, improve */
215 typedef struct cl_nbparam_params
218 int eeltype; /**< type of electrostatics, takes values from #eelCu */
219 int vdwtype; /**< type of VdW impl., takes values from #evdwCu */
221 float epsfac; /**< charge multiplication factor */
222 float c_rf; /**< Reaction-field/plain cutoff electrostatics const. */
223 float two_k_rf; /**< Reaction-field electrostatics constant */
224 float ewald_beta; /**< Ewald/PME parameter */
225 float sh_ewald; /**< Ewald/PME correction term substracted from the direct-space potential */
226 float sh_lj_ewald; /**< LJ-Ewald/PME correction term added to the correction potential */
227 float ewaldcoeff_lj; /**< LJ-Ewald/PME coefficient */
229 float rcoulomb_sq; /**< Coulomb cut-off squared */
231 float rvdw_sq; /**< VdW cut-off squared */
232 float rvdw_switch; /**< VdW switched cut-off */
233 float rlistOuter_sq; /**< Full, outer pair-list cut-off squared */
234 float rlistInner_sq; /**< Inner, dynamic pruned pair-list cut-off squared */
236 shift_consts_t dispersion_shift; /**< VdW shift dispersion constants */
237 shift_consts_t repulsion_shift; /**< VdW shift repulsion constants */
238 switch_consts_t vdw_switch; /**< VdW switch constants */
240 /* Ewald Coulomb force table data - accessed through texture memory */
241 float coulomb_tab_scale; /**< table scale/spacing */
242 } cl_nbparam_params_t;
246 * \brief Pair list data.
248 using cl_plist_t = Nbnxm::gpu_plist;
251 * \brief Typedef of actual timer type.
253 typedef struct Nbnxm::gpu_timers_t cl_timers_t;
256 * \brief Main data structure for OpenCL nonbonded force calculations.
258 struct gmx_nbnxn_ocl_t
260 const gmx_device_info_t *dev_info; /**< OpenCL device information */
261 struct gmx_device_runtime_data_t *dev_rundata; /**< OpenCL runtime data (context, kernels) */
263 /**< Pointers to non-bonded kernel functions
264 * organized similar with nb_kfunc_xxx arrays in nbnxn_ocl.cpp */
266 cl_kernel kernel_noener_noprune_ptr[eelOclNR][evdwOclNR];
267 cl_kernel kernel_ener_noprune_ptr[eelOclNR][evdwOclNR];
268 cl_kernel kernel_noener_prune_ptr[eelOclNR][evdwOclNR];
269 cl_kernel kernel_ener_prune_ptr[eelOclNR][evdwOclNR];
271 cl_kernel kernel_pruneonly[ePruneNR]; /**< prune kernels, ePruneKind defined the kernel kinds */
273 bool bPrefetchLjParam; /**< true if prefetching fg i-atom LJ parameters should be used in the kernels */
275 /**< auxiliary kernels implementing memset-like functions */
277 cl_kernel kernel_memset_f;
278 cl_kernel kernel_memset_f2;
279 cl_kernel kernel_memset_f3;
280 cl_kernel kernel_zero_e_fshift;
283 cl_bool bUseTwoStreams; /**< true if doing both local/non-local NB work on GPU */
284 cl_bool bNonLocalStreamActive; /**< true indicates that the nonlocal_done event was enqueued */
286 cl_atomdata_t *atdat; /**< atom data */
287 cl_nbparam_t *nbparam; /**< parameters required for the non-bonded calc. */
288 gmx::EnumerationArray<Nbnxm::InteractionLocality, cl_plist_t *> plist; /**< pair-list data structures (local and non-local) */
289 cl_nb_staging_t nbst; /**< staging area where fshift/energies get downloaded */
291 gmx::EnumerationArray<Nbnxm::InteractionLocality, cl_command_queue> stream; /**< local and non-local GPU queues */
293 /** events used for synchronization */
294 cl_event nonlocal_done; /**< event triggered when the non-local non-bonded kernel
295 is done (and the local transfer can proceed) */
296 cl_event misc_ops_and_local_H2D_done; /**< event triggered when the tasks issued in
297 the local stream that need to precede the
298 non-local force calculations are done
299 (e.g. f buffer 0-ing, local x/q H2D) */
301 cl_bool bDoTime; /**< True if event-based timing is enabled. */
302 cl_timers_t *timers; /**< OpenCL event-based timers. */
303 struct gmx_wallclock_gpu_nbnxn_t *timings; /**< Timing data. TODO: deprecate this and query timers for accumulated data instead */
306 #endif /* NBNXN_OPENCL_TYPES_H */