2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, 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.
45 #include "types/simple.h"
46 #include "types/nbnxn_pairlist.h"
47 #include "types/nb_verlet.h"
48 #include "types/ishift.h"
49 #include "types/force_flags.h"
50 #include "../nbnxn_consts.h"
53 #include "thread_mpi/atomic.h"
56 #include "nbnxn_cuda_types.h"
57 #include "../../gmxlib/cuda_tools/cudautils.cuh"
58 #include "nbnxn_cuda.h"
59 #include "nbnxn_cuda_data_mgmt.h"
61 #if defined TEXOBJ_SUPPORTED && __CUDA_ARCH__ >= 300
65 /*! Texture reference for nonbonded parameters; bound to cu_nbparam_t.nbfp*/
66 texture<float, 1, cudaReadModeElementType> nbfp_texref;
68 /*! Texture reference for Ewald coulomb force table; bound to cu_nbparam_t.coulomb_tab */
69 texture<float, 1, cudaReadModeElementType> coulomb_tab_texref;
71 /* Convenience defines */
72 #define NCL_PER_SUPERCL (NBNXN_GPU_NCLUSTER_PER_SUPERCLUSTER)
73 #define CL_SIZE (NBNXN_GPU_CLUSTER_SIZE)
75 /***** The kernels come here *****/
76 #include "nbnxn_cuda_kernel_utils.cuh"
78 /* Top-level kernel generation: will generate through multiple inclusion the
79 * following flavors for all kernels:
80 * - force-only output;
81 * - force and energy output;
82 * - force-only with pair list pruning;
83 * - force and energy output with pair list pruning.
86 #include "nbnxn_cuda_kernels.cuh"
87 /** Force & energy **/
89 #include "nbnxn_cuda_kernels.cuh"
92 /*** Pair-list pruning kernels ***/
95 #include "nbnxn_cuda_kernels.cuh"
96 /** Force & energy **/
98 #include "nbnxn_cuda_kernels.cuh"
102 /*! Nonbonded kernel function pointer type */
103 typedef void (*nbnxn_cu_kfunc_ptr_t)(const cu_atomdata_t,
108 /*********************************/
110 /* XXX always/never run the energy/pruning kernels -- only for benchmarking purposes */
111 static bool always_ener = (getenv("GMX_GPU_ALWAYS_ENER") != NULL);
112 static bool never_ener = (getenv("GMX_GPU_NEVER_ENER") != NULL);
113 static bool always_prune = (getenv("GMX_GPU_ALWAYS_PRUNE") != NULL);
116 /* Bit-pattern used for polling-based GPU synchronization. It is used as a float
117 * and corresponds to having the exponent set to the maximum (127 -- single
118 * precision) and the mantissa to 0.
120 static unsigned int poll_wait_pattern = (0x7FU << 23);
122 /*! Returns the number of blocks to be used for the nonbonded GPU kernel. */
123 static inline int calc_nb_kernel_nblock(int nwork_units, cuda_dev_info_t *dinfo)
129 max_grid_x_size = dinfo->prop.maxGridSize[0];
131 /* do we exceed the grid x dimension limit? */
132 if (nwork_units > max_grid_x_size)
134 gmx_fatal(FARGS, "Watch out, the input system is too large to simulate!\n"
135 "The number of nonbonded work units (=number of super-clusters) exceeds the"
136 "maximum grid size in x dimension (%d > %d)!", nwork_units, max_grid_x_size);
143 /* Constant arrays listing all kernel function pointers and enabling selection
144 of a kernel in an elegant manner. */
146 /*! Pointers to the non-bonded kernels organized in 2-dim arrays by:
147 * electrostatics and VDW type.
149 * Note that the row- and column-order of function pointers has to match the
150 * order of corresponding enumerated electrostatics and vdw types, resp.,
151 * defined in nbnxn_cuda_types.h.
154 /*! Force-only kernel function pointers. */
155 static const nbnxn_cu_kfunc_ptr_t nb_kfunc_noener_noprune_ptr[eelCuNR][evdwCuNR] =
157 { nbnxn_kernel_ElecCut_VdwLJ_F_cuda, nbnxn_kernel_ElecCut_VdwLJFsw_F_cuda, nbnxn_kernel_ElecCut_VdwLJPsw_F_cuda },
158 { nbnxn_kernel_ElecRF_VdwLJ_F_cuda, nbnxn_kernel_ElecRF_VdwLJFsw_F_cuda, nbnxn_kernel_ElecRF_VdwLJPsw_F_cuda },
159 { nbnxn_kernel_ElecEwQSTab_VdwLJ_F_cuda, nbnxn_kernel_ElecEwQSTab_VdwLJFsw_F_cuda, nbnxn_kernel_ElecEwQSTab_VdwLJPsw_F_cuda },
160 { nbnxn_kernel_ElecEwQSTabTwinCut_VdwLJ_F_cuda, nbnxn_kernel_ElecEwQSTabTwinCut_VdwLJFsw_F_cuda, nbnxn_kernel_ElecEwQSTabTwinCut_VdwLJPsw_F_cuda },
161 { nbnxn_kernel_ElecEw_VdwLJ_F_cuda, nbnxn_kernel_ElecEw_VdwLJFsw_F_cuda, nbnxn_kernel_ElecEw_VdwLJPsw_F_cuda },
162 { nbnxn_kernel_ElecEwTwinCut_VdwLJ_F_cuda, nbnxn_kernel_ElecEwTwinCut_VdwLJFsw_F_cuda, nbnxn_kernel_ElecEwTwinCut_VdwLJPsw_F_cuda }
165 /*! Force + energy kernel function pointers. */
166 static const nbnxn_cu_kfunc_ptr_t nb_kfunc_ener_noprune_ptr[eelCuNR][evdwCuNR] =
168 { nbnxn_kernel_ElecCut_VdwLJ_VF_cuda, nbnxn_kernel_ElecCut_VdwLJFsw_VF_cuda, nbnxn_kernel_ElecCut_VdwLJPsw_VF_cuda },
169 { nbnxn_kernel_ElecRF_VdwLJ_VF_cuda, nbnxn_kernel_ElecRF_VdwLJFsw_VF_cuda, nbnxn_kernel_ElecRF_VdwLJPsw_VF_cuda },
170 { nbnxn_kernel_ElecEwQSTab_VdwLJ_VF_cuda, nbnxn_kernel_ElecEwQSTab_VdwLJFsw_VF_cuda, nbnxn_kernel_ElecEwQSTab_VdwLJPsw_VF_cuda },
171 { nbnxn_kernel_ElecEwQSTabTwinCut_VdwLJ_VF_cuda, nbnxn_kernel_ElecEwQSTabTwinCut_VdwLJFsw_VF_cuda, nbnxn_kernel_ElecEwQSTabTwinCut_VdwLJPsw_VF_cuda },
172 { nbnxn_kernel_ElecEw_VdwLJ_VF_cuda, nbnxn_kernel_ElecEw_VdwLJFsw_VF_cuda, nbnxn_kernel_ElecEw_VdwLJPsw_VF_cuda },
173 { nbnxn_kernel_ElecEwTwinCut_VdwLJ_VF_cuda, nbnxn_kernel_ElecEwTwinCut_VdwLJFsw_VF_cuda, nbnxn_kernel_ElecEwTwinCut_VdwLJPsw_VF_cuda }
176 /*! Force + pruning kernel function pointers. */
177 static const nbnxn_cu_kfunc_ptr_t nb_kfunc_noener_prune_ptr[eelCuNR][evdwCuNR] =
179 { nbnxn_kernel_ElecCut_VdwLJ_F_prune_cuda, nbnxn_kernel_ElecCut_VdwLJFsw_F_prune_cuda, nbnxn_kernel_ElecCut_VdwLJPsw_F_prune_cuda },
180 { nbnxn_kernel_ElecRF_VdwLJ_F_prune_cuda, nbnxn_kernel_ElecRF_VdwLJFsw_F_prune_cuda, nbnxn_kernel_ElecRF_VdwLJPsw_F_prune_cuda },
181 { nbnxn_kernel_ElecEwQSTab_VdwLJ_F_prune_cuda, nbnxn_kernel_ElecEwQSTab_VdwLJFsw_F_prune_cuda, nbnxn_kernel_ElecEwQSTab_VdwLJPsw_F_prune_cuda },
182 { nbnxn_kernel_ElecEwQSTabTwinCut_VdwLJ_F_prune_cuda, nbnxn_kernel_ElecEwQSTabTwinCut_VdwLJFsw_F_prune_cuda, nbnxn_kernel_ElecEwQSTabTwinCut_VdwLJPsw_F_prune_cuda },
183 { nbnxn_kernel_ElecEw_VdwLJ_F_prune_cuda, nbnxn_kernel_ElecEw_VdwLJFsw_F_prune_cuda, nbnxn_kernel_ElecEw_VdwLJPsw_F_prune_cuda },
184 { nbnxn_kernel_ElecEwTwinCut_VdwLJ_F_prune_cuda, nbnxn_kernel_ElecEwTwinCut_VdwLJFsw_F_prune_cuda, nbnxn_kernel_ElecEwTwinCut_VdwLJPsw_F_prune_cuda }
187 /*! Force + energy + pruning kernel function pointers. */
188 static const nbnxn_cu_kfunc_ptr_t nb_kfunc_ener_prune_ptr[eelCuNR][evdwCuNR] =
190 { nbnxn_kernel_ElecCut_VdwLJ_VF_prune_cuda, nbnxn_kernel_ElecCut_VdwLJFsw_VF_prune_cuda, nbnxn_kernel_ElecCut_VdwLJPsw_VF_prune_cuda },
191 { nbnxn_kernel_ElecRF_VdwLJ_VF_prune_cuda, nbnxn_kernel_ElecRF_VdwLJFsw_VF_prune_cuda, nbnxn_kernel_ElecRF_VdwLJPsw_VF_prune_cuda },
192 { nbnxn_kernel_ElecEwQSTab_VdwLJ_VF_prune_cuda, nbnxn_kernel_ElecEwQSTab_VdwLJFsw_VF_prune_cuda, nbnxn_kernel_ElecEwQSTab_VdwLJPsw_VF_prune_cuda },
193 { nbnxn_kernel_ElecEwQSTabTwinCut_VdwLJ_VF_prune_cuda, nbnxn_kernel_ElecEwQSTabTwinCut_VdwLJFsw_VF_prune_cuda, nbnxn_kernel_ElecEwQSTabTwinCut_VdwLJPsw_VF_prune_cuda },
194 { nbnxn_kernel_ElecEw_VdwLJ_VF_prune_cuda, nbnxn_kernel_ElecEw_VdwLJFsw_VF_prune_cuda, nbnxn_kernel_ElecEw_VdwLJPsw_VF_prune_cuda },
195 { nbnxn_kernel_ElecEwTwinCut_VdwLJ_VF_prune_cuda, nbnxn_kernel_ElecEwTwinCut_VdwLJFsw_VF_prune_cuda, nbnxn_kernel_ElecEwTwinCut_VdwLJPsw_VF_prune_cuda }
198 /*! Return a pointer to the kernel version to be executed at the current step. */
199 static inline nbnxn_cu_kfunc_ptr_t select_nbnxn_kernel(int eeltype,
204 nbnxn_cu_kfunc_ptr_t res;
206 assert(eeltype < eelCuNR);
207 assert(evdwtype < eelCuNR);
213 res = nb_kfunc_ener_prune_ptr[eeltype][evdwtype];
217 res = nb_kfunc_ener_noprune_ptr[eeltype][evdwtype];
224 res = nb_kfunc_noener_prune_ptr[eeltype][evdwtype];
228 res = nb_kfunc_noener_noprune_ptr[eeltype][evdwtype];
235 /*! Calculates the amount of shared memory required by the CUDA kernel in use. */
236 static inline int calc_shmem_required()
240 /* size of shmem (force-buffers/xq/atom type preloading) */
241 /* NOTE: with the default kernel on sm3.0 we need shmem only for pre-loading */
242 /* i-atom x+q in shared memory */
243 shmem = NCL_PER_SUPERCL * CL_SIZE * sizeof(float4);
244 /* cj in shared memory, for both warps separately */
245 shmem += 2 * NBNXN_GPU_JGROUP_SIZE * sizeof(int);
247 /* i-atom types in shared memory */
248 shmem += NCL_PER_SUPERCL * CL_SIZE * sizeof(int);
250 #if __CUDA_ARCH__ < 300
251 /* force reduction buffers in shared memory */
252 shmem += CL_SIZE * CL_SIZE * 3 * sizeof(float);
258 /*! As we execute nonbonded workload in separate streams, before launching
259 the kernel we need to make sure that he following operations have completed:
260 - atomdata allocation and related H2D transfers (every nstlist step);
261 - pair list H2D transfer (every nstlist step);
262 - shift vector H2D transfer (every nstlist step);
263 - force (+shift force and energy) output clearing (every step).
265 These operations are issued in the local stream at the beginning of the step
266 and therefore always complete before the local kernel launch. The non-local
267 kernel is launched after the local on the same device/context, so this is
268 inherently scheduled after the operations in the local stream (including the
270 However, for the sake of having a future-proof implementation, we use the
271 misc_ops_done event to record the point in time when the above operations
272 are finished and synchronize with this event in the non-local stream.
274 void nbnxn_cuda_launch_kernel(nbnxn_cuda_ptr_t cu_nb,
275 const nbnxn_atomdata_t *nbatom,
280 int adat_begin, adat_len; /* local/nonlocal offset and length used for xq and f */
281 /* CUDA kernel launch-related stuff */
283 dim3 dim_block, dim_grid;
284 nbnxn_cu_kfunc_ptr_t nb_kernel = NULL; /* fn pointer to the nonbonded kernel */
286 cu_atomdata_t *adat = cu_nb->atdat;
287 cu_nbparam_t *nbp = cu_nb->nbparam;
288 cu_plist_t *plist = cu_nb->plist[iloc];
289 cu_timers_t *t = cu_nb->timers;
290 cudaStream_t stream = cu_nb->stream[iloc];
292 bool bCalcEner = flags & GMX_FORCE_VIRIAL;
293 bool bCalcFshift = flags & GMX_FORCE_VIRIAL;
294 bool bDoTime = cu_nb->bDoTime;
296 /* turn energy calculation always on/off (for debugging/testing only) */
297 bCalcEner = (bCalcEner || always_ener) && !never_ener;
299 /* don't launch the kernel if there is no work to do */
300 if (plist->nsci == 0)
305 /* calculate the atom data index range based on locality */
309 adat_len = adat->natoms_local;
313 adat_begin = adat->natoms_local;
314 adat_len = adat->natoms - adat->natoms_local;
317 /* When we get here all misc operations issues in the local stream are done,
318 so we record that in the local stream and wait for it in the nonlocal one. */
319 if (cu_nb->bUseTwoStreams)
321 if (iloc == eintLocal)
323 stat = cudaEventRecord(cu_nb->misc_ops_done, stream);
324 CU_RET_ERR(stat, "cudaEventRecord on misc_ops_done failed");
328 stat = cudaStreamWaitEvent(stream, cu_nb->misc_ops_done, 0);
329 CU_RET_ERR(stat, "cudaStreamWaitEvent on misc_ops_done failed");
333 /* beginning of timed HtoD section */
336 stat = cudaEventRecord(t->start_nb_h2d[iloc], stream);
337 CU_RET_ERR(stat, "cudaEventRecord failed");
341 cu_copy_H2D_async(adat->xq + adat_begin, nbatom->x + adat_begin * 4,
342 adat_len * sizeof(*adat->xq), stream);
346 stat = cudaEventRecord(t->stop_nb_h2d[iloc], stream);
347 CU_RET_ERR(stat, "cudaEventRecord failed");
350 /* beginning of timed nonbonded calculation section */
353 stat = cudaEventRecord(t->start_nb_k[iloc], stream);
354 CU_RET_ERR(stat, "cudaEventRecord failed");
357 /* get the pointer to the kernel flavor we need to use */
358 nb_kernel = select_nbnxn_kernel(nbp->eeltype,
361 plist->bDoPrune || always_prune);
363 /* kernel launch config */
364 nblock = calc_nb_kernel_nblock(plist->nsci, cu_nb->dev_info);
365 dim_block = dim3(CL_SIZE, CL_SIZE, 1);
366 dim_grid = dim3(nblock, 1, 1);
367 shmem = calc_shmem_required();
371 fprintf(debug, "GPU launch configuration:\n\tThread block: %dx%dx%d\n\t"
372 "Grid: %dx%d\n\t#Super-clusters/clusters: %d/%d (%d)\n",
373 dim_block.x, dim_block.y, dim_block.z,
374 dim_grid.x, dim_grid.y, plist->nsci*NCL_PER_SUPERCL,
375 NCL_PER_SUPERCL, plist->na_c);
378 nb_kernel<<< dim_grid, dim_block, shmem, stream>>> (*adat, *nbp, *plist, bCalcFshift);
379 CU_LAUNCH_ERR("k_calc_nb");
383 stat = cudaEventRecord(t->stop_nb_k[iloc], stream);
384 CU_RET_ERR(stat, "cudaEventRecord failed");
388 void nbnxn_cuda_launch_cpyback(nbnxn_cuda_ptr_t cu_nb,
389 const nbnxn_atomdata_t *nbatom,
394 int adat_begin, adat_len, adat_end; /* local/nonlocal offset and length used for xq and f */
397 /* determine interaction locality from atom locality */
402 else if (NONLOCAL_A(aloc))
409 sprintf(stmp, "Invalid atom locality passed (%d); valid here is only "
410 "local (%d) or nonlocal (%d)", aloc, eatLocal, eatNonlocal);
414 cu_atomdata_t *adat = cu_nb->atdat;
415 cu_timers_t *t = cu_nb->timers;
416 bool bDoTime = cu_nb->bDoTime;
417 cudaStream_t stream = cu_nb->stream[iloc];
419 bool bCalcEner = flags & GMX_FORCE_VIRIAL;
420 bool bCalcFshift = flags & GMX_FORCE_VIRIAL;
422 /* don't launch copy-back if there was no work to do */
423 if (cu_nb->plist[iloc]->nsci == 0)
428 /* calculate the atom data index range based on locality */
432 adat_len = adat->natoms_local;
433 adat_end = cu_nb->atdat->natoms_local;
437 adat_begin = adat->natoms_local;
438 adat_len = adat->natoms - adat->natoms_local;
439 adat_end = cu_nb->atdat->natoms;
442 /* beginning of timed D2H section */
445 stat = cudaEventRecord(t->start_nb_d2h[iloc], stream);
446 CU_RET_ERR(stat, "cudaEventRecord failed");
449 if (!cu_nb->bUseStreamSync)
451 /* For safety reasons set a few (5%) forces to NaN. This way even if the
452 polling "hack" fails with some future NVIDIA driver we'll get a crash. */
453 for (int i = adat_begin; i < 3*adat_end + 2; i += adat_len/20)
456 nbatom->out[0].f[i] = NAN;
459 if (numeric_limits<float>::has_quiet_NaN)
461 nbatom->out[0].f[i] = numeric_limits<float>::quiet_NaN();
466 nbatom->out[0].f[i] = GMX_REAL_MAX;
471 /* Set the last four bytes of the force array to a bit pattern
472 which can't be the result of the force calculation:
473 max exponent (127) and zero mantissa. */
474 *(unsigned int*)&nbatom->out[0].f[adat_end*3 - 1] = poll_wait_pattern;
477 /* With DD the local D2H transfer can only start after the non-local
478 has been launched. */
479 if (iloc == eintLocal && cu_nb->bUseTwoStreams)
481 stat = cudaStreamWaitEvent(stream, cu_nb->nonlocal_done, 0);
482 CU_RET_ERR(stat, "cudaStreamWaitEvent on nonlocal_done failed");
486 cu_copy_D2H_async(nbatom->out[0].f + adat_begin * 3, adat->f + adat_begin,
487 (adat_len)*sizeof(*adat->f), stream);
489 /* After the non-local D2H is launched the nonlocal_done event can be
490 recorded which signals that the local D2H can proceed. This event is not
491 placed after the non-local kernel because we first need the non-local
493 if (iloc == eintNonlocal)
495 stat = cudaEventRecord(cu_nb->nonlocal_done, stream);
496 CU_RET_ERR(stat, "cudaEventRecord on nonlocal_done failed");
499 /* only transfer energies in the local stream */
505 cu_copy_D2H_async(cu_nb->nbst.fshift, adat->fshift,
506 SHIFTS * sizeof(*cu_nb->nbst.fshift), stream);
512 cu_copy_D2H_async(cu_nb->nbst.e_lj, adat->e_lj,
513 sizeof(*cu_nb->nbst.e_lj), stream);
514 cu_copy_D2H_async(cu_nb->nbst.e_el, adat->e_el,
515 sizeof(*cu_nb->nbst.e_el), stream);
521 stat = cudaEventRecord(t->stop_nb_d2h[iloc], stream);
522 CU_RET_ERR(stat, "cudaEventRecord failed");
526 /* Atomic compare-exchange operation on unsigned values. It is used in
527 * polling wait for the GPU.
529 static inline bool atomic_cas(volatile unsigned int *ptr,
536 return tMPI_Atomic_cas((tMPI_Atomic_t *)ptr, oldval, newval);
538 gmx_incons("Atomic operations not available, atomic_cas() should not have been called!");
543 void nbnxn_cuda_wait_gpu(nbnxn_cuda_ptr_t cu_nb,
544 const nbnxn_atomdata_t *nbatom,
546 real *e_lj, real *e_el, rvec *fshift)
548 /* NOTE: only implemented for single-precision at this time */
550 int i, adat_end, iloc = -1;
551 volatile unsigned int *poll_word;
553 /* determine interaction locality from atom locality */
558 else if (NONLOCAL_A(aloc))
565 sprintf(stmp, "Invalid atom locality passed (%d); valid here is only "
566 "local (%d) or nonlocal (%d)", aloc, eatLocal, eatNonlocal);
570 cu_plist_t *plist = cu_nb->plist[iloc];
571 cu_timers_t *timers = cu_nb->timers;
572 wallclock_gpu_t *timings = cu_nb->timings;
573 nb_staging nbst = cu_nb->nbst;
575 bool bCalcEner = flags & GMX_FORCE_VIRIAL;
576 bool bCalcFshift = flags & GMX_FORCE_VIRIAL;
578 /* turn energy calculation always on/off (for debugging/testing only) */
579 bCalcEner = (bCalcEner || always_ener) && !never_ener;
581 /* don't launch wait/update timers & counters if there was no work to do
583 NOTE: if timing with multiple GPUs (streams) becomes possible, the
584 counters could end up being inconsistent due to not being incremented
585 on some of the nodes! */
586 if (cu_nb->plist[iloc]->nsci == 0)
591 /* calculate the atom data index range based on locality */
594 adat_end = cu_nb->atdat->natoms_local;
598 adat_end = cu_nb->atdat->natoms;
601 if (cu_nb->bUseStreamSync)
603 stat = cudaStreamSynchronize(cu_nb->stream[iloc]);
604 CU_RET_ERR(stat, "cudaStreamSynchronize failed in cu_blockwait_nb");
608 /* Busy-wait until we get the signal pattern set in last byte
609 * of the l/nl float vector. This pattern corresponds to a floating
610 * point number which can't be the result of the force calculation
611 * (maximum, 127 exponent and 0 mantissa).
612 * The polling uses atomic compare-exchange.
614 poll_word = (volatile unsigned int*)&nbatom->out[0].f[adat_end*3 - 1];
615 while (atomic_cas(poll_word, poll_wait_pattern, poll_wait_pattern))
620 /* timing data accumulation */
623 /* only increase counter once (at local F wait) */
627 timings->ktime[plist->bDoPrune ? 1 : 0][bCalcEner ? 1 : 0].c += 1;
631 timings->ktime[plist->bDoPrune ? 1 : 0][bCalcEner ? 1 : 0].t +=
632 cu_event_elapsed(timers->start_nb_k[iloc], timers->stop_nb_k[iloc]);
634 /* X/q H2D and F D2H timings */
635 timings->nb_h2d_t += cu_event_elapsed(timers->start_nb_h2d[iloc],
636 timers->stop_nb_h2d[iloc]);
637 timings->nb_d2h_t += cu_event_elapsed(timers->start_nb_d2h[iloc],
638 timers->stop_nb_d2h[iloc]);
640 /* only count atdat and pair-list H2D at pair-search step */
643 /* atdat transfer timing (add only once, at local F wait) */
647 timings->pl_h2d_t += cu_event_elapsed(timers->start_atdat,
651 timings->pl_h2d_t += cu_event_elapsed(timers->start_pl_h2d[iloc],
652 timers->stop_pl_h2d[iloc]);
656 /* add up energies and shift forces (only once at local F wait) */
667 for (i = 0; i < SHIFTS; i++)
669 fshift[i][0] += nbst.fshift[i].x;
670 fshift[i][1] += nbst.fshift[i].y;
671 fshift[i][2] += nbst.fshift[i].z;
676 /* turn off pruning (doesn't matter if this is pair-search step or not) */
677 plist->bDoPrune = false;
680 /*! Return the reference to the nbfp texture. */
681 const struct texture<float, 1, cudaReadModeElementType> &nbnxn_cuda_get_nbfp_texref()
686 /*! Return the reference to the coulomb_tab. */
687 const struct texture<float, 1, cudaReadModeElementType> &nbnxn_cuda_get_coulomb_tab_texref()
689 return coulomb_tab_texref;
692 /*! Set up the cache configuration for the non-bonded kernels,
694 void nbnxn_cuda_set_cacheconfig(cuda_dev_info_t *devinfo)
698 for (int i = 0; i < eelCuNR; i++)
700 for (int j = 0; j < evdwCuNR; j++)
702 if (devinfo->prop.major >= 3)
704 /* Default kernel on sm 3.x 48/16 kB Shared/L1 */
705 stat = cudaFuncSetCacheConfig(nb_kfunc_ener_prune_ptr[i][j], cudaFuncCachePreferShared);
706 stat = cudaFuncSetCacheConfig(nb_kfunc_ener_noprune_ptr[i][j], cudaFuncCachePreferShared);
707 stat = cudaFuncSetCacheConfig(nb_kfunc_noener_prune_ptr[i][j], cudaFuncCachePreferShared);
708 stat = cudaFuncSetCacheConfig(nb_kfunc_noener_noprune_ptr[i][j], cudaFuncCachePreferShared);
712 /* On Fermi prefer L1 gives 2% higher performance */
713 /* Default kernel on sm_2.x 16/48 kB Shared/L1 */
714 stat = cudaFuncSetCacheConfig(nb_kfunc_ener_prune_ptr[i][j], cudaFuncCachePreferL1);
715 stat = cudaFuncSetCacheConfig(nb_kfunc_ener_noprune_ptr[i][j], cudaFuncCachePreferL1);
716 stat = cudaFuncSetCacheConfig(nb_kfunc_noener_prune_ptr[i][j], cudaFuncCachePreferL1);
717 stat = cudaFuncSetCacheConfig(nb_kfunc_noener_noprune_ptr[i][j], cudaFuncCachePreferL1);
719 CU_RET_ERR(stat, "cudaFuncSetCacheConfig failed");