2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2016,2017, 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 nbnxn_ocl module.
40 * \author Anca Hamuraru <anca@streamcomputing.eu>
41 * \author Szilárd Páll <pszilard@kth.se>
42 * \ingroup module_mdlib
45 #ifndef NBNXN_OPENCL_TYPES_H
46 #define NBNXN_OPENCL_TYPES_H
48 /*! \brief Declare to OpenCL SDKs that we intend to use OpenCL API
49 features that were deprecated in 2.0, so that they don't warn about
51 #define CL_USE_DEPRECATED_OPENCL_2_0_APIS
53 # include <OpenCL/opencl.h>
55 # include <CL/opencl.h>
58 #include "gromacs/gpu_utils/gpuregiontimer_ocl.h"
59 #include "gromacs/gpu_utils/oclutils.h"
60 #include "gromacs/mdlib/nbnxn_pairlist.h"
61 #include "gromacs/mdtypes/interaction_const.h"
62 #include "gromacs/utility/real.h"
64 /* kernel does #include "gromacs/math/utilities.h" */
65 /* Move the actual useful stuff here: */
68 #define M_FLOAT_1_SQRTPI 0.564189583547756f
70 /*! \brief Macros defining platform-dependent defaults for the prune kernel's j4 processing concurrency.
72 * The GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY macro allows compile-time override.
75 #ifndef GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY
76 #define GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY_AMD 4
77 #define GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY_NVIDIA 4
78 #define GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY_DEFAULT 4
80 #define GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY_AMD GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY
81 #define GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY_NVIDIA GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY
82 #define GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY_DEFAULT GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY
85 /*! \brief Constants for platform-dependent defaults for the prune kernel's j4 processing concurrency.
87 * Initialized using macros that can be overridden at compile-time (using #GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY).
90 const int c_oclPruneKernelJ4ConcurrencyAMD = GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY_AMD;
91 const int c_oclPruneKernelJ4ConcurrencyNVIDIA = GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY_NVIDIA;
92 const int c_oclPruneKernelJ4ConcurrencyDefault = GMX_NBNXN_PRUNE_KERNEL_J4_CONCURRENCY_DEFAULT;
95 /*! \brief Returns the j4 processing concurrency parameter for the vendor \p vendorId
96 * \param vendorId takes values from #ocl_vendor_id_t.
98 static inline int getOclPruneKernelJ4Concurrency(int vendorId)
100 assert(vendorId < OCL_VENDOR_UNKNOWN);
103 case OCL_VENDOR_AMD: return c_oclPruneKernelJ4ConcurrencyAMD; break;
104 case OCL_VENDOR_NVIDIA: return c_oclPruneKernelJ4ConcurrencyNVIDIA; break;
105 default: return c_oclPruneKernelJ4ConcurrencyDefault; break;
114 /*! \brief Electrostatic OpenCL kernel flavors.
116 * Types of electrostatics implementations available in the OpenCL non-bonded
117 * force kernels. These represent both the electrostatics types implemented
118 * by the kernels (cut-off, RF, and Ewald - a subset of what's defined in
119 * enums.h) as well as encode implementation details analytical/tabulated
120 * and single or twin cut-off (for Ewald kernels).
121 * Note that the cut-off and RF kernels have only analytical flavor and unlike
122 * in the CPU kernels, the tabulated kernels are ATM Ewald-only.
124 * The row-order of pointers to different electrostatic kernels defined in
125 * nbnxn_cuda.cu by the nb_*_kfunc_ptr function pointer table
126 * should match the order of enumerated types below.
129 eelOclCUT, eelOclRF, eelOclEWALD_TAB, eelOclEWALD_TAB_TWIN, eelOclEWALD_ANA, eelOclEWALD_ANA_TWIN, eelOclNR
132 /*! \brief VdW OpenCL kernel flavors.
134 * The enumerates values correspond to the LJ implementations in the OpenCL non-bonded
137 * The column-order of pointers to different electrostatic kernels defined in
138 * nbnxn_cuda.cu by the nb_*_kfunc_ptr function pointer table
139 * should match the order of enumerated types below.
142 evdwOclCUT, evdwOclCUTCOMBGEOM, evdwOclCUTCOMBLB, evdwOclFSWITCH, evdwOclPSWITCH, evdwOclEWALDGEOM, evdwOclEWALDLB, evdwOclNR
145 /*! \brief Pruning kernel flavors.
147 * The values correspond to the first call of the pruning post-list generation
148 * and the rolling pruning, respectively.
151 epruneFirst, epruneRolling, ePruneNR
155 * \brief Staging area for temporary data downloaded from the GPU.
157 * The energies/shift forces get downloaded here first, before getting added
158 * to the CPU-side aggregate values.
160 typedef struct cl_nb_staging
162 float *e_lj; /**< LJ energy */
163 float *e_el; /**< electrostatic energy */
164 float (*fshift)[3]; /**< float3 buffer with shift forces */
168 * \brief Nonbonded atom data - both inputs and outputs.
170 typedef struct cl_atomdata
172 int natoms; /**< number of atoms */
173 int natoms_local; /**< number of local atoms */
174 int nalloc; /**< allocation size for the atom data (xq, f) */
176 cl_mem xq; /**< float4 buffer with atom coordinates + charges, size natoms */
178 cl_mem f; /**< float3 buffer with force output array, size natoms */
179 size_t f_elem_size; /**< Size in bytes for one element of f buffer */
181 cl_mem e_lj; /**< LJ energy output, size 1 */
182 cl_mem e_el; /**< Electrostatics energy input, size 1 */
184 cl_mem fshift; /**< float3 buffer with shift forces */
185 size_t fshift_elem_size; /**< Size in bytes for one element of fshift buffer */
187 int ntypes; /**< number of atom types */
188 cl_mem atom_types; /**< int buffer with atom type indices, size natoms */
189 cl_mem lj_comb; /**< float2 buffer with sqrt(c6),sqrt(c12), size natoms */
191 cl_mem shift_vec; /**< float3 buffer with shifts values */
192 size_t shift_vec_elem_size; /**< Size in bytes for one element of shift_vec buffer */
194 cl_bool bShiftVecUploaded; /**< true if the shift vector has been uploaded */
198 * \brief Parameters required for the OpenCL nonbonded calculations.
200 typedef struct cl_nbparam
203 int eeltype; /**< type of electrostatics, takes values from #eelOcl */
204 int vdwtype; /**< type of VdW impl., takes values from #evdwOcl */
206 float epsfac; /**< charge multiplication factor */
207 float c_rf; /**< Reaction-field/plain cutoff electrostatics const. */
208 float two_k_rf; /**< Reaction-field electrostatics constant */
209 float ewald_beta; /**< Ewald/PME parameter */
210 float sh_ewald; /**< Ewald/PME correction term substracted from the direct-space potential */
211 float sh_lj_ewald; /**< LJ-Ewald/PME correction term added to the correction potential */
212 float ewaldcoeff_lj; /**< LJ-Ewald/PME coefficient */
214 float rcoulomb_sq; /**< Coulomb cut-off squared */
216 float rvdw_sq; /**< VdW cut-off squared */
217 float rvdw_switch; /**< VdW switched cut-off */
218 float rlistOuter_sq; /**< Full, outer pair-list cut-off squared */
219 float rlistInner_sq; /**< Inner, dynamic pruned pair-list cut-off squared */
220 bool useDynamicPruning; /**< True if we use dynamic pair-list pruning */
222 shift_consts_t dispersion_shift; /**< VdW shift dispersion constants */
223 shift_consts_t repulsion_shift; /**< VdW shift repulsion constants */
224 switch_consts_t vdw_switch; /**< VdW switch constants */
226 /* LJ non-bonded parameters - accessed through texture memory */
227 cl_mem nbfp_climg2d; /**< nonbonded parameter table with C6/C12 pairs per atom type-pair, 2*ntype^2 elements */
228 cl_mem nbfp_comb_climg2d; /**< nonbonded parameter table per atom type, 2*ntype elements */
230 /* Ewald Coulomb force table data - accessed through texture memory */
231 float coulomb_tab_scale; /**< table scale/spacing */
232 cl_mem coulomb_tab_climg2d; /**< pointer to the table in the device memory */
236 * \brief Data structure shared between the OpenCL device code and OpenCL host code
238 * Must not contain OpenCL objects (buffers)
239 * TODO: review, improve */
240 typedef struct cl_nbparam_params
243 int eeltype; /**< type of electrostatics, takes values from #eelCu */
244 int vdwtype; /**< type of VdW impl., takes values from #evdwCu */
246 float epsfac; /**< charge multiplication factor */
247 float c_rf; /**< Reaction-field/plain cutoff electrostatics const. */
248 float two_k_rf; /**< Reaction-field electrostatics constant */
249 float ewald_beta; /**< Ewald/PME parameter */
250 float sh_ewald; /**< Ewald/PME correction term substracted from the direct-space potential */
251 float sh_lj_ewald; /**< LJ-Ewald/PME correction term added to the correction potential */
252 float ewaldcoeff_lj; /**< LJ-Ewald/PME coefficient */
254 float rcoulomb_sq; /**< Coulomb cut-off squared */
256 float rvdw_sq; /**< VdW cut-off squared */
257 float rvdw_switch; /**< VdW switched cut-off */
258 float rlistOuter_sq; /**< Full, outer pair-list cut-off squared */
259 float rlistInner_sq; /**< Inner, dynamic pruned pair-list cut-off squared */
261 shift_consts_t dispersion_shift; /**< VdW shift dispersion constants */
262 shift_consts_t repulsion_shift; /**< VdW shift repulsion constants */
263 switch_consts_t vdw_switch; /**< VdW switch constants */
265 /* Ewald Coulomb force table data - accessed through texture memory */
266 float coulomb_tab_scale; /**< table scale/spacing */
267 } cl_nbparam_params_t;
271 * \brief Pair list data.
273 typedef struct cl_plist
275 int na_c; /**< number of atoms per cluster */
277 int nsci; /**< size of sci, # of i clusters in the list */
278 int sci_nalloc; /**< allocation size of sci */
279 cl_mem sci; /**< list of i-cluster ("super-clusters").
280 It contains elements of type nbnxn_sci_t */
282 int ncj4; /**< total # of 4*j clusters */
283 int cj4_nalloc; /**< allocation size of cj4 */
284 cl_mem cj4; /**< 4*j cluster list, contains j cluster number and
285 index into the i cluster list.
286 It contains elements of type nbnxn_cj4_t */
287 int nimask; /**< # of 4*j clusters * # of warps */
288 int imask_nalloc; /**< allocation size of imask */
289 cl_mem imask; /**< imask for 2 warps for each 4*j cluster group */
290 cl_mem excl; /**< atom interaction bits
291 It contains elements of type nbnxn_excl_t */
292 int nexcl; /**< count for excl */
293 int excl_nalloc; /**< allocation size of excl */
295 /* parameter+variables for normal and rolling pruning */
296 bool haveFreshList; /**< true after search, indictes that initial pruning with outer prunning is needed */
297 int rollingPruningNumParts; /**< the number of parts/steps over which one cyle of roling pruning takes places */
298 int rollingPruningPart; /**< the next part to which the roling pruning needs to be applied */
302 * \brief OpenCL events used for timing GPU kernels and H2D/D2H transfers.
304 * The two-sized arrays hold the local and non-local values and should always
305 * be indexed with eintLocal/eintNonlocal.
307 typedef struct cl_timers
309 GpuRegionTimer atdat; /**< timer for atom data transfer (every PS step) */
310 GpuRegionTimer nb_h2d[2]; /**< timer for x/q H2D transfers (l/nl, every step) */
311 GpuRegionTimer nb_d2h[2]; /**< timer for f D2H transfer (l/nl, every step) */
312 GpuRegionTimer pl_h2d[2]; /**< timer for pair-list H2D transfers (l/nl, every PS step) */
313 bool didPairlistH2D[2]; /**< true when a pair-list transfer has been done at this step */
314 GpuRegionTimer nb_k[2]; /**< timer for non-bonded kernels (l/nl, every step) */
315 GpuRegionTimer prune_k[2]; /**< timer for the 1st pass list pruning kernel (l/nl, every PS step) */
316 bool didPrune[2]; /**< true when we timed pruning and the timings need to be accounted for */
317 GpuRegionTimer rollingPrune_k[2]; /**< timer for rolling pruning kernels (l/nl, frequency depends on chunk size) */
318 bool didRollingPrune[2]; /**< true when we timed rolling pruning (at the previous step) and the timings need to be accounted for */
322 * \brief Main data structure for OpenCL nonbonded force calculations.
324 struct gmx_nbnxn_ocl_t
326 const gmx_device_info_t *dev_info; /**< OpenCL device information */
327 struct gmx_device_runtime_data_t *dev_rundata; /**< OpenCL runtime data (context, kernels) */
329 /**< Pointers to non-bonded kernel functions
330 * organized similar with nb_kfunc_xxx arrays in nbnxn_ocl.cpp */
332 cl_kernel kernel_noener_noprune_ptr[eelOclNR][evdwOclNR];
333 cl_kernel kernel_ener_noprune_ptr[eelOclNR][evdwOclNR];
334 cl_kernel kernel_noener_prune_ptr[eelOclNR][evdwOclNR];
335 cl_kernel kernel_ener_prune_ptr[eelOclNR][evdwOclNR];
337 cl_kernel kernel_pruneonly[ePruneNR]; /**< prune kernels, ePruneKind defined the kernel kinds */
339 bool bPrefetchLjParam; /**< true if prefetching fg i-atom LJ parameters should be used in the kernels */
341 /**< auxiliary kernels implementing memset-like functions */
343 cl_kernel kernel_memset_f;
344 cl_kernel kernel_memset_f2;
345 cl_kernel kernel_memset_f3;
346 cl_kernel kernel_zero_e_fshift;
349 cl_bool bUseTwoStreams; /**< true if doing both local/non-local NB work on GPU */
350 cl_bool bNonLocalStreamActive; /**< true indicates that the nonlocal_done event was enqueued */
352 cl_atomdata_t *atdat; /**< atom data */
353 cl_nbparam_t *nbparam; /**< parameters required for the non-bonded calc. */
354 cl_plist_t *plist[2]; /**< pair-list data structures (local and non-local) */
355 cl_nb_staging_t nbst; /**< staging area where fshift/energies get downloaded */
357 cl_mem debug_buffer; /**< debug buffer */
359 cl_command_queue stream[2]; /**< local and non-local GPU queues */
361 /** events used for synchronization */
362 cl_event nonlocal_done; /**< event triggered when the non-local non-bonded kernel
363 is done (and the local transfer can proceed) */
364 cl_event misc_ops_and_local_H2D_done; /**< event triggered when the tasks issued in
365 the local stream that need to precede the
366 non-local force calculations are done
367 (e.g. f buffer 0-ing, local x/q H2D) */
369 cl_bool bDoTime; /**< True if event-based timing is enabled. */
370 cl_timers_t *timers; /**< OpenCL event-based timers. */
371 struct gmx_wallclock_gpu_nbnxn_t *timings; /**< Timing data. TODO: deprecate this and query timers for accumulated data instead */
378 #endif /* NBNXN_OPENCL_TYPES_H */