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 * check out http://www.gromacs.org for more information.
7 * Copyright (c) 2012, by the GROMACS development team, led by
8 * David van der Spoel, Berk Hess, Erik Lindahl, and including many
9 * others, as listed in the AUTHORS file in the top-level source
10 * directory and at http://www.gromacs.org.
12 * GROMACS is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public License
14 * as published by the Free Software Foundation; either version 2.1
15 * of the License, or (at your option) any later version.
17 * GROMACS is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with GROMACS; if not, see
24 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
25 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
27 * If you want to redistribute modifications to GROMACS, please
28 * consider that scientific software is very special. Version
29 * control is crucial - bugs must be traceable. We will be happy to
30 * consider code for inclusion in the official distribution, but
31 * derived work must not be called official GROMACS. Details are found
32 * in the README & COPYING files - if they are missing, get the
33 * official version at http://www.gromacs.org.
35 * To help us fund GROMACS development, we humbly ask that you cite
36 * the research papers on the package. Check out http://www.gromacs.org.
48 #include "types/simple.h"
49 #include "types/nbnxn_pairlist.h"
50 #include "types/nb_verlet.h"
51 #include "types/ishift.h"
52 #include "types/force_flags.h"
53 #include "../nbnxn_consts.h"
56 #include "thread_mpi/atomic.h"
59 #include "nbnxn_cuda_types.h"
60 #include "../../gmxlib/cuda_tools/cudautils.cuh"
61 #include "nbnxn_cuda.h"
62 #include "nbnxn_cuda_data_mgmt.h"
64 #if defined TEXOBJ_SUPPORTED && __CUDA_ARCH__ >= 300
68 /*! Texture reference for nonbonded parameters; bound to cu_nbparam_t.nbfp*/
69 texture<float, 1, cudaReadModeElementType> nbfp_texref;
71 /*! Texture reference for Ewald coulomb force table; bound to cu_nbparam_t.coulomb_tab */
72 texture<float, 1, cudaReadModeElementType> coulomb_tab_texref;
74 /* Convenience defines */
75 #define NCL_PER_SUPERCL (NBNXN_GPU_NCLUSTER_PER_SUPERCLUSTER)
76 #define CL_SIZE (NBNXN_GPU_CLUSTER_SIZE)
78 /***** The kernels come here *****/
79 #include "nbnxn_cuda_kernel_utils.cuh"
81 /* Top-level kernel generation: will generate through multiple inclusion the
82 * following flavors for all kernels:
83 * - force-only output;
84 * - force and energy output;
85 * - force-only with pair list pruning;
86 * - force and energy output with pair list pruning.
89 #include "nbnxn_cuda_kernels.cuh"
90 /** Force & energy **/
92 #include "nbnxn_cuda_kernels.cuh"
95 /*** Pair-list pruning kernels ***/
98 #include "nbnxn_cuda_kernels.cuh"
99 /** Force & energy **/
100 #define CALC_ENERGIES
101 #include "nbnxn_cuda_kernels.cuh"
105 /*! Nonbonded kernel function pointer type */
106 typedef void (*nbnxn_cu_kfunc_ptr_t)(const cu_atomdata_t,
111 /*********************************/
113 /* XXX always/never run the energy/pruning kernels -- only for benchmarking purposes */
114 static bool always_ener = (getenv("GMX_GPU_ALWAYS_ENER") != NULL);
115 static bool never_ener = (getenv("GMX_GPU_NEVER_ENER") != NULL);
116 static bool always_prune = (getenv("GMX_GPU_ALWAYS_PRUNE") != NULL);
119 /* Bit-pattern used for polling-based GPU synchronization. It is used as a float
120 * and corresponds to having the exponent set to the maximum (127 -- single
121 * precision) and the mantissa to 0.
123 static unsigned int poll_wait_pattern = (0x7FU << 23);
125 /*! Returns the number of blocks to be used for the nonbonded GPU kernel. */
126 static inline int calc_nb_kernel_nblock(int nwork_units, cuda_dev_info_t *dinfo)
132 max_grid_x_size = dinfo->prop.maxGridSize[0];
134 /* do we exceed the grid x dimension limit? */
135 if (nwork_units > max_grid_x_size)
137 gmx_fatal(FARGS, "Watch out, the input system is too large to simulate!\n"
138 "The number of nonbonded work units (=number of super-clusters) exceeds the"
139 "maximum grid size in x dimension (%d > %d)!", nwork_units, max_grid_x_size);
146 /* Constant arrays listing all kernel function pointers and enabling selection
147 of a kernel in an elegant manner. */
149 static const int nEnergyKernelTypes = 2; /* 0 - no energy, 1 - energy */
150 static const int nPruneKernelTypes = 2; /* 0 - no prune, 1 - prune */
152 /*! Pointers to the default kernels organized in a 3 dim array by:
153 * electrostatics type, energy calculation on/off, and pruning on/off.
155 * Note that the order of electrostatics (1st dimension) has to match the
156 * order of corresponding enumerated types defined in nbnxn_cuda_types.h.
158 static const nbnxn_cu_kfunc_ptr_t
159 nb_default_kfunc_ptr[eelCuNR][nEnergyKernelTypes][nPruneKernelTypes] =
161 { { k_nbnxn_cutoff, k_nbnxn_cutoff_prune },
162 { k_nbnxn_cutoff_ener, k_nbnxn_cutoff_ener_prune } },
163 { { k_nbnxn_rf, k_nbnxn_rf_prune },
164 { k_nbnxn_rf_ener, k_nbnxn_rf_ener_prune } },
165 { { k_nbnxn_ewald_tab, k_nbnxn_ewald_tab_prune },
166 { k_nbnxn_ewald_tab_ener, k_nbnxn_ewald_tab_ener_prune } },
167 { { k_nbnxn_ewald_tab_twin, k_nbnxn_ewald_tab_twin_prune },
168 { k_nbnxn_ewald_tab_twin_ener, k_nbnxn_ewald_twin_ener_prune } },
169 { { k_nbnxn_ewald, k_nbnxn_ewald_prune },
170 { k_nbnxn_ewald_ener, k_nbnxn_ewald_ener_prune } },
171 { { k_nbnxn_ewald_twin, k_nbnxn_ewald_twin_prune },
172 { k_nbnxn_ewald_twin_ener, k_nbnxn_ewald_twin_ener_prune } },
175 /*! Pointers to the legacy kernels organized in a 3 dim array by:
176 * electrostatics type, energy calculation on/off, and pruning on/off.
178 * Note that the order of electrostatics (1st dimension) has to match the
179 * order of corresponding enumerated types defined in nbnxn_cuda_types.h.
181 static const nbnxn_cu_kfunc_ptr_t
182 nb_legacy_kfunc_ptr[eelCuNR][nEnergyKernelTypes][nPruneKernelTypes] =
184 { { k_nbnxn_cutoff_legacy, k_nbnxn_cutoff_prune_legacy },
185 { k_nbnxn_cutoff_ener_legacy, k_nbnxn_cutoff_ener_prune_legacy } },
186 { { k_nbnxn_rf_legacy, k_nbnxn_rf_prune_legacy },
187 { k_nbnxn_rf_ener_legacy, k_nbnxn_rf_ener_prune_legacy } },
188 { { k_nbnxn_ewald_tab_legacy, k_nbnxn_ewald_tab_prune_legacy },
189 { k_nbnxn_ewald_tab_ener_legacy, k_nbnxn_ewald_tab_ener_prune_legacy } },
190 { { k_nbnxn_ewald_tab_twin_legacy, k_nbnxn_ewald_tab_twin_prune_legacy },
191 { k_nbnxn_ewald_tab_twin_ener_legacy, k_nbnxn_ewald_tab_twin_ener_prune_legacy } },
194 /*! Return a pointer to the kernel version to be executed at the current step. */
195 static inline nbnxn_cu_kfunc_ptr_t select_nbnxn_kernel(int kver, int eeltype,
196 bool bDoEne, bool bDoPrune)
198 assert(kver < eNbnxnCuKNR);
199 assert(eeltype < eelCuNR);
201 if (NBNXN_KVER_LEGACY(kver))
203 /* no analytical Ewald with legacy kernels */
204 assert(eeltype <= eelCuEWALD_TAB_TWIN);
206 return nb_legacy_kfunc_ptr[eeltype][bDoEne][bDoPrune];
210 return nb_default_kfunc_ptr[eeltype][bDoEne][bDoPrune];
214 /*! Calculates the amount of shared memory required for kernel version in use. */
215 static inline int calc_shmem_required(int kver)
219 /* size of shmem (force-buffers/xq/atom type preloading) */
220 if (NBNXN_KVER_LEGACY(kver))
222 /* i-atom x+q in shared memory */
223 shmem = NCL_PER_SUPERCL * CL_SIZE * sizeof(float4);
224 /* force reduction buffers in shared memory */
225 shmem += CL_SIZE * CL_SIZE * 3 * sizeof(float);
229 /* NOTE: with the default kernel on sm3.0 we need shmem only for pre-loading */
230 /* i-atom x+q in shared memory */
231 shmem = NCL_PER_SUPERCL * CL_SIZE * sizeof(float4);
232 /* cj in shared memory, for both warps separately */
233 shmem += 2 * NBNXN_GPU_JGROUP_SIZE * sizeof(int);
235 /* i-atom types in shared memory */
236 shmem += NCL_PER_SUPERCL * CL_SIZE * sizeof(int);
238 #if __CUDA_ARCH__ < 300
239 /* force reduction buffers in shared memory */
240 shmem += CL_SIZE * CL_SIZE * 3 * sizeof(float);
247 /*! As we execute nonbonded workload in separate streams, before launching
248 the kernel we need to make sure that he following operations have completed:
249 - atomdata allocation and related H2D transfers (every nstlist step);
250 - pair list H2D transfer (every nstlist step);
251 - shift vector H2D transfer (every nstlist step);
252 - force (+shift force and energy) output clearing (every step).
254 These operations are issued in the local stream at the beginning of the step
255 and therefore always complete before the local kernel launch. The non-local
256 kernel is launched after the local on the same device/context hence it is
257 inherently scheduled after the operations in the local stream (including the
258 above "misc_ops") on pre-GK110 devices with single hardware queue, but on later
259 devices with multiple hardware queues the dependency needs to be enforced.
260 We use the misc_ops_and_local_H2D_done event to record the point where
261 the local x+q H2D (and all preceding) tasks are complete and synchronize
262 with this event in the non-local stream before launching the non-bonded kernel.
264 void nbnxn_cuda_launch_kernel(nbnxn_cuda_ptr_t cu_nb,
265 const nbnxn_atomdata_t *nbatom,
270 int adat_begin, adat_len; /* local/nonlocal offset and length used for xq and f */
271 /* CUDA kernel launch-related stuff */
273 dim3 dim_block, dim_grid;
274 nbnxn_cu_kfunc_ptr_t nb_kernel = NULL; /* fn pointer to the nonbonded kernel */
276 cu_atomdata_t *adat = cu_nb->atdat;
277 cu_nbparam_t *nbp = cu_nb->nbparam;
278 cu_plist_t *plist = cu_nb->plist[iloc];
279 cu_timers_t *t = cu_nb->timers;
280 cudaStream_t stream = cu_nb->stream[iloc];
282 bool bCalcEner = flags & GMX_FORCE_VIRIAL;
283 bool bCalcFshift = flags & GMX_FORCE_VIRIAL;
284 bool bDoTime = cu_nb->bDoTime;
286 /* turn energy calculation always on/off (for debugging/testing only) */
287 bCalcEner = (bCalcEner || always_ener) && !never_ener;
289 /* Don't launch the non-local kernel if there is no work to do.
290 Doing the same for the local kernel is more complicated, since the
291 local part of the force array also depends on the non-local kernel.
292 So to avoid complicating the code and to reduce the risk of bugs,
293 we always call the local kernel, the local x+q copy and later (not in
294 this function) the stream wait, local f copyback and the f buffer
295 clearing. All these operations, except for the local interaction kernel,
296 are needed for the non-local interactions. The skip of the local kernel
297 call is taken care of later in this function. */
298 if (iloc == eintNonlocal && plist->nsci == 0)
303 /* calculate the atom data index range based on locality */
307 adat_len = adat->natoms_local;
311 adat_begin = adat->natoms_local;
312 adat_len = adat->natoms - adat->natoms_local;
315 /* beginning of timed HtoD section */
318 stat = cudaEventRecord(t->start_nb_h2d[iloc], stream);
319 CU_RET_ERR(stat, "cudaEventRecord failed");
323 cu_copy_H2D_async(adat->xq + adat_begin, nbatom->x + adat_begin * 4,
324 adat_len * sizeof(*adat->xq), stream);
326 /* When we get here all misc operations issues in the local stream as well as
327 the local xq H2D are done,
328 so we record that in the local stream and wait for it in the nonlocal one. */
329 if (cu_nb->bUseTwoStreams)
331 if (iloc == eintLocal)
333 stat = cudaEventRecord(cu_nb->misc_ops_and_local_H2D_done, stream);
334 CU_RET_ERR(stat, "cudaEventRecord on misc_ops_and_local_H2D_done failed");
338 stat = cudaStreamWaitEvent(stream, cu_nb->misc_ops_and_local_H2D_done, 0);
339 CU_RET_ERR(stat, "cudaStreamWaitEvent on misc_ops_and_local_H2D_done failed");
345 stat = cudaEventRecord(t->stop_nb_h2d[iloc], stream);
346 CU_RET_ERR(stat, "cudaEventRecord failed");
349 if (plist->nsci == 0)
351 /* Don't launch an empty local kernel (not allowed with CUDA) */
355 /* beginning of timed nonbonded calculation section */
358 stat = cudaEventRecord(t->start_nb_k[iloc], stream);
359 CU_RET_ERR(stat, "cudaEventRecord failed");
362 /* get the pointer to the kernel flavor we need to use */
363 nb_kernel = select_nbnxn_kernel(cu_nb->kernel_ver, nbp->eeltype, bCalcEner,
364 plist->bDoPrune || always_prune);
366 /* kernel launch config */
367 nblock = calc_nb_kernel_nblock(plist->nsci, cu_nb->dev_info);
368 dim_block = dim3(CL_SIZE, CL_SIZE, 1);
369 dim_grid = dim3(nblock, 1, 1);
370 shmem = calc_shmem_required(cu_nb->kernel_ver);
374 fprintf(debug, "GPU launch configuration:\n\tThread block: %dx%dx%d\n\t"
375 "Grid: %dx%d\n\t#Super-clusters/clusters: %d/%d (%d)\n",
376 dim_block.x, dim_block.y, dim_block.z,
377 dim_grid.x, dim_grid.y, plist->nsci*NCL_PER_SUPERCL,
378 NCL_PER_SUPERCL, plist->na_c);
381 nb_kernel<<<dim_grid, dim_block, shmem, stream>>>(*adat, *nbp, *plist, bCalcFshift);
382 CU_LAUNCH_ERR("k_calc_nb");
386 stat = cudaEventRecord(t->stop_nb_k[iloc], stream);
387 CU_RET_ERR(stat, "cudaEventRecord failed");
391 void nbnxn_cuda_launch_cpyback(nbnxn_cuda_ptr_t cu_nb,
392 const nbnxn_atomdata_t *nbatom,
397 int adat_begin, adat_len, adat_end; /* local/nonlocal offset and length used for xq and f */
400 /* determine interaction locality from atom locality */
405 else if (NONLOCAL_A(aloc))
412 sprintf(stmp, "Invalid atom locality passed (%d); valid here is only "
413 "local (%d) or nonlocal (%d)", aloc, eatLocal, eatNonlocal);
417 cu_atomdata_t *adat = cu_nb->atdat;
418 cu_timers_t *t = cu_nb->timers;
419 bool bDoTime = cu_nb->bDoTime;
420 cudaStream_t stream = cu_nb->stream[iloc];
422 bool bCalcEner = flags & GMX_FORCE_VIRIAL;
423 bool bCalcFshift = flags & GMX_FORCE_VIRIAL;
425 /* don't launch non-local copy-back if there was no non-local work to do */
426 if (iloc == eintNonlocal && cu_nb->plist[iloc]->nsci == 0)
431 /* calculate the atom data index range based on locality */
435 adat_len = adat->natoms_local;
436 adat_end = cu_nb->atdat->natoms_local;
440 adat_begin = adat->natoms_local;
441 adat_len = adat->natoms - adat->natoms_local;
442 adat_end = cu_nb->atdat->natoms;
445 /* beginning of timed D2H section */
448 stat = cudaEventRecord(t->start_nb_d2h[iloc], stream);
449 CU_RET_ERR(stat, "cudaEventRecord failed");
452 if (!cu_nb->bUseStreamSync)
454 /* For safety reasons set a few (5%) forces to NaN. This way even if the
455 polling "hack" fails with some future NVIDIA driver we'll get a crash. */
456 for (int i = adat_begin; i < 3*adat_end + 2; i += adat_len/20)
459 nbatom->out[0].f[i] = NAN;
462 if (numeric_limits<float>::has_quiet_NaN)
464 nbatom->out[0].f[i] = numeric_limits<float>::quiet_NaN();
469 nbatom->out[0].f[i] = GMX_REAL_MAX;
474 /* Set the last four bytes of the force array to a bit pattern
475 which can't be the result of the force calculation:
476 max exponent (127) and zero mantissa. */
477 *(unsigned int*)&nbatom->out[0].f[adat_end*3 - 1] = poll_wait_pattern;
480 /* With DD the local D2H transfer can only start after the non-local
481 kernel has finished. */
482 if (iloc == eintLocal && cu_nb->bUseTwoStreams)
484 stat = cudaStreamWaitEvent(stream, cu_nb->nonlocal_done, 0);
485 CU_RET_ERR(stat, "cudaStreamWaitEvent on nonlocal_done failed");
489 cu_copy_D2H_async(nbatom->out[0].f + adat_begin * 3, adat->f + adat_begin,
490 (adat_len)*sizeof(*adat->f), stream);
492 /* After the non-local D2H is launched the nonlocal_done event can be
493 recorded which signals that the local D2H can proceed. This event is not
494 placed after the non-local kernel because we want the non-local data
496 if (iloc == eintNonlocal)
498 stat = cudaEventRecord(cu_nb->nonlocal_done, stream);
499 CU_RET_ERR(stat, "cudaEventRecord on nonlocal_done failed");
502 /* only transfer energies in the local stream */
508 cu_copy_D2H_async(cu_nb->nbst.fshift, adat->fshift,
509 SHIFTS * sizeof(*cu_nb->nbst.fshift), stream);
515 cu_copy_D2H_async(cu_nb->nbst.e_lj, adat->e_lj,
516 sizeof(*cu_nb->nbst.e_lj), stream);
517 cu_copy_D2H_async(cu_nb->nbst.e_el, adat->e_el,
518 sizeof(*cu_nb->nbst.e_el), stream);
524 stat = cudaEventRecord(t->stop_nb_d2h[iloc], stream);
525 CU_RET_ERR(stat, "cudaEventRecord failed");
529 /* Atomic compare-exchange operation on unsigned values. It is used in
530 * polling wait for the GPU.
532 static inline bool atomic_cas(volatile unsigned int *ptr,
539 return tMPI_Atomic_cas((tMPI_Atomic_t *)ptr, oldval, newval);
541 gmx_incons("Atomic operations not available, atomic_cas() should not have been called!");
546 void nbnxn_cuda_wait_gpu(nbnxn_cuda_ptr_t cu_nb,
547 const nbnxn_atomdata_t *nbatom,
549 real *e_lj, real *e_el, rvec *fshift)
551 /* NOTE: only implemented for single-precision at this time */
553 int i, adat_end, iloc = -1;
554 volatile unsigned int *poll_word;
556 /* determine interaction locality from atom locality */
561 else if (NONLOCAL_A(aloc))
568 sprintf(stmp, "Invalid atom locality passed (%d); valid here is only "
569 "local (%d) or nonlocal (%d)", aloc, eatLocal, eatNonlocal);
573 cu_plist_t *plist = cu_nb->plist[iloc];
574 cu_timers_t *timers = cu_nb->timers;
575 wallclock_gpu_t *timings = cu_nb->timings;
576 nb_staging nbst = cu_nb->nbst;
578 bool bCalcEner = flags & GMX_FORCE_VIRIAL;
579 bool bCalcFshift = flags & GMX_FORCE_VIRIAL;
581 /* turn energy calculation always on/off (for debugging/testing only) */
582 bCalcEner = (bCalcEner || always_ener) && !never_ener;
584 /* Launch wait/update timers & counters, unless doing the non-local phase
585 when there is not actually work to do. This is consistent with
586 nbnxn_cuda_launch_kernel.
588 NOTE: if timing with multiple GPUs (streams) becomes possible, the
589 counters could end up being inconsistent due to not being incremented
590 on some of the nodes! */
591 if (iloc == eintNonlocal && cu_nb->plist[iloc]->nsci == 0)
596 /* calculate the atom data index range based on locality */
599 adat_end = cu_nb->atdat->natoms_local;
603 adat_end = cu_nb->atdat->natoms;
606 if (cu_nb->bUseStreamSync)
608 stat = cudaStreamSynchronize(cu_nb->stream[iloc]);
609 CU_RET_ERR(stat, "cudaStreamSynchronize failed in cu_blockwait_nb");
613 /* Busy-wait until we get the signal pattern set in last byte
614 * of the l/nl float vector. This pattern corresponds to a floating
615 * point number which can't be the result of the force calculation
616 * (maximum, 127 exponent and 0 mantissa).
617 * The polling uses atomic compare-exchange.
619 poll_word = (volatile unsigned int*)&nbatom->out[0].f[adat_end*3 - 1];
620 while (atomic_cas(poll_word, poll_wait_pattern, poll_wait_pattern)) {}
623 /* timing data accumulation */
626 /* only increase counter once (at local F wait) */
630 timings->ktime[plist->bDoPrune ? 1 : 0][bCalcEner ? 1 : 0].c += 1;
634 timings->ktime[plist->bDoPrune ? 1 : 0][bCalcEner ? 1 : 0].t +=
635 cu_event_elapsed(timers->start_nb_k[iloc], timers->stop_nb_k[iloc]);
637 /* X/q H2D and F D2H timings */
638 timings->nb_h2d_t += cu_event_elapsed(timers->start_nb_h2d[iloc],
639 timers->stop_nb_h2d[iloc]);
640 timings->nb_d2h_t += cu_event_elapsed(timers->start_nb_d2h[iloc],
641 timers->stop_nb_d2h[iloc]);
643 /* only count atdat and pair-list H2D at pair-search step */
646 /* atdat transfer timing (add only once, at local F wait) */
650 timings->pl_h2d_t += cu_event_elapsed(timers->start_atdat,
654 timings->pl_h2d_t += cu_event_elapsed(timers->start_pl_h2d[iloc],
655 timers->stop_pl_h2d[iloc]);
659 /* add up energies and shift forces (only once at local F wait) */
670 for (i = 0; i < SHIFTS; i++)
672 fshift[i][0] += nbst.fshift[i].x;
673 fshift[i][1] += nbst.fshift[i].y;
674 fshift[i][2] += nbst.fshift[i].z;
679 /* turn off pruning (doesn't matter if this is pair-search step or not) */
680 plist->bDoPrune = false;
683 /*! Return the reference to the nbfp texture. */
684 const struct texture<float, 1, cudaReadModeElementType>& nbnxn_cuda_get_nbfp_texref()
689 /*! Return the reference to the coulomb_tab. */
690 const struct texture<float, 1, cudaReadModeElementType>& nbnxn_cuda_get_coulomb_tab_texref()
692 return coulomb_tab_texref;
695 /*! Set up the cache configuration for the non-bonded kernels,
697 void nbnxn_cuda_set_cacheconfig(cuda_dev_info_t *devinfo)
701 for (int i = 0; i < eelCuNR; i++)
703 for (int j = 0; j < nEnergyKernelTypes; j++)
705 for (int k = 0; k < nPruneKernelTypes; k++)
707 /* Legacy kernel 16/48 kB Shared/L1
708 * No analytical Ewald!
710 if (i != eelCuEWALD_ANA && i != eelCuEWALD_ANA_TWIN)
712 stat = cudaFuncSetCacheConfig(nb_legacy_kfunc_ptr[i][j][k], cudaFuncCachePreferL1);
713 CU_RET_ERR(stat, "cudaFuncSetCacheConfig failed");
716 if (devinfo->prop.major >= 3)
718 /* Default kernel on sm 3.x 48/16 kB Shared/L1 */
719 stat = cudaFuncSetCacheConfig(nb_default_kfunc_ptr[i][j][k], cudaFuncCachePreferShared);
723 /* On Fermi prefer L1 gives 2% higher performance */
724 /* Default kernel on sm_2.x 16/48 kB Shared/L1 */
725 stat = cudaFuncSetCacheConfig(nb_default_kfunc_ptr[i][j][k], cudaFuncCachePreferL1);
727 CU_RET_ERR(stat, "cudaFuncSetCacheConfig failed");