2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2016,2017, 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.
36 #include "vectype_ops.clh"
38 #define CL_SIZE (NBNXN_GPU_CLUSTER_SIZE)
39 #define NCL_PER_SUPERCL (NBNXN_GPU_NCLUSTER_PER_SUPERCLUSTER)
43 #undef KERNEL_UTILS_INLINE
44 #ifdef KERNEL_UTILS_INLINE
45 #define __INLINE__ inline
50 /* 1.0 / sqrt(M_PI) */
51 #define M_FLOAT_1_SQRTPI 0.564189583547756f
55 #ifndef NBNXN_OPENCL_KERNEL_UTILS_CLH
56 #define NBNXN_OPENCL_KERNEL_UTILS_CLH
58 __constant sampler_t generic_sampler = CLK_NORMALIZED_COORDS_FALSE /* Natural coords */
59 | CLK_ADDRESS_NONE /* No clamp/repeat*/
60 | CLK_FILTER_NEAREST ; /* No interpolation */
64 #define WARP_SIZE_LOG2 (5)
65 #define CL_SIZE_LOG2 (3) /* change this together with CL_SIZE !*/
66 #define CL_SIZE_SQ (CL_SIZE * CL_SIZE)
67 #define FBUF_STRIDE (CL_SIZE_SQ)
69 #define ONE_SIXTH_F 0.16666667f
70 #define ONE_TWELVETH_F 0.08333333f
73 // Data structures shared between OpenCL device code and OpenCL host code
74 // TODO: review, improve
75 // Replaced real by float for now, to avoid including any other header
82 /* Used with potential switching:
83 * rsw = max(r - r_switch, 0)
84 * sw = 1 + c3*rsw^3 + c4*rsw^4 + c5*rsw^5
85 * dsw = 3*c3*rsw^2 + 4*c4*rsw^3 + 5*c5*rsw^4
86 * force = force*dsw - potential*sw
95 // Data structure shared between the OpenCL device code and OpenCL host code
96 // Must not contain OpenCL objects (buffers)
97 typedef struct cl_nbparam_params
100 int eeltype; /**< type of electrostatics, takes values from #eelCu */
101 int vdwtype; /**< type of VdW impl., takes values from #evdwCu */
103 float epsfac; /**< charge multiplication factor */
104 float c_rf; /**< Reaction-field/plain cutoff electrostatics const. */
105 float two_k_rf; /**< Reaction-field electrostatics constant */
106 float ewald_beta; /**< Ewald/PME parameter */
107 float sh_ewald; /**< Ewald/PME correction term substracted from the direct-space potential */
108 float sh_lj_ewald; /**< LJ-Ewald/PME correction term added to the correction potential */
109 float ewaldcoeff_lj; /**< LJ-Ewald/PME coefficient */
111 float rcoulomb_sq; /**< Coulomb cut-off squared */
113 float rvdw_sq; /**< VdW cut-off squared */
114 float rvdw_switch; /**< VdW switched cut-off */
115 float rlistOuter_sq; /**< Full, outer pair-list cut-off squared */
116 float rlistInner_sq; /**< Inner, dynamic pruned pair-list cut-off squared XXX: this is only needed in the pruning kernels, but for now we also pass it to the nonbondeds */
118 shift_consts_t dispersion_shift; /**< VdW shift dispersion constants */
119 shift_consts_t repulsion_shift; /**< VdW shift repulsion constants */
120 switch_consts_t vdw_switch; /**< VdW switch constants */
122 /* Ewald Coulomb force table data - accessed through texture memory */
123 float coulomb_tab_scale; /**< table scale/spacing */
124 }cl_nbparam_params_t;
127 int sci; /* i-super-cluster */
128 int shift; /* Shift vector index plus possible flags */
129 int cj4_ind_start; /* Start index into cj4 */
130 int cj4_ind_end; /* End index into cj4 */
134 unsigned int imask; /* The i-cluster interactions mask for 1 warp */
135 int excl_ind; /* Index into the exclusion array for 1 warp */
139 int cj[4]; /* The 4 j-clusters */
140 nbnxn_im_ei_t imei[2]; /* The i-cluster mask data for 2 warps */
145 unsigned int pair[32]; /* Topology exclusion interaction bits for one warp,
146 * each unsigned has bitS for 4*8 i clusters
150 /*! i-cluster interaction mask for a super-cluster with all NCL_PER_SUPERCL bits set */
151 __constant unsigned supercl_interaction_mask = ((1U << NCL_PER_SUPERCL) - 1U);
153 /*! Convert LJ sigma,epsilon parameters to C6,C12. */
154 __INLINE__ __device__
155 void convert_sigma_epsilon_to_c6_c12(const float sigma,
160 float sigma2, sigma6;
162 sigma2 = sigma * sigma;
163 sigma6 = sigma2 *sigma2 * sigma2;
164 *c6 = epsilon * sigma6;
169 /*! Apply force switch, force + energy version. */
170 __INLINE__ __device__
171 void calculate_force_switch_F(cl_nbparam_params_t *nbparam,
180 /* force switch constants */
181 float disp_shift_V2 = nbparam->dispersion_shift.c2;
182 float disp_shift_V3 = nbparam->dispersion_shift.c3;
183 float repu_shift_V2 = nbparam->repulsion_shift.c2;
184 float repu_shift_V3 = nbparam->repulsion_shift.c3;
187 r_switch = r - nbparam->rvdw_switch;
188 r_switch = r_switch >= 0.0f ? r_switch : 0.0f;
191 -c6*(disp_shift_V2 + disp_shift_V3*r_switch)*r_switch*r_switch*inv_r +
192 c12*(-repu_shift_V2 + repu_shift_V3*r_switch)*r_switch*r_switch*inv_r;
195 /*! Apply force switch, force-only version. */
196 __INLINE__ __device__
197 void calculate_force_switch_F_E(cl_nbparam_params_t *nbparam,
207 /* force switch constants */
208 float disp_shift_V2 = nbparam->dispersion_shift.c2;
209 float disp_shift_V3 = nbparam->dispersion_shift.c3;
210 float repu_shift_V2 = nbparam->repulsion_shift.c2;
211 float repu_shift_V3 = nbparam->repulsion_shift.c3;
213 float disp_shift_F2 = nbparam->dispersion_shift.c2/3;
214 float disp_shift_F3 = nbparam->dispersion_shift.c3/4;
215 float repu_shift_F2 = nbparam->repulsion_shift.c2/3;
216 float repu_shift_F3 = nbparam->repulsion_shift.c3/4;
219 r_switch = r - nbparam->rvdw_switch;
220 r_switch = r_switch >= 0.0f ? r_switch : 0.0f;
223 -c6*(disp_shift_V2 + disp_shift_V3*r_switch)*r_switch*r_switch*inv_r +
224 c12*(-repu_shift_V2 + repu_shift_V3*r_switch)*r_switch*r_switch*inv_r;
226 c6*(disp_shift_F2 + disp_shift_F3*r_switch)*r_switch*r_switch*r_switch -
227 c12*(repu_shift_F2 + repu_shift_F3*r_switch)*r_switch*r_switch*r_switch;
230 /*! Apply potential switch, force-only version. */
231 __INLINE__ __device__
232 void calculate_potential_switch_F(cl_nbparam_params_t *nbparam,
241 /* potential switch constants */
242 float switch_V3 = nbparam->vdw_switch.c3;
243 float switch_V4 = nbparam->vdw_switch.c4;
244 float switch_V5 = nbparam->vdw_switch.c5;
245 float switch_F2 = nbparam->vdw_switch.c3;
246 float switch_F3 = nbparam->vdw_switch.c4;
247 float switch_F4 = nbparam->vdw_switch.c5;
250 r_switch = r - nbparam->rvdw_switch;
252 /* Unlike in the F+E kernel, conditional is faster here */
255 sw = 1.0f + (switch_V3 + (switch_V4 + switch_V5*r_switch)*r_switch)*r_switch*r_switch*r_switch;
256 dsw = (switch_F2 + (switch_F3 + switch_F4*r_switch)*r_switch)*r_switch*r_switch;
258 *F_invr = (*F_invr)*sw - inv_r*(*E_lj)*dsw;
262 /*! Apply potential switch, force + energy version. */
263 __INLINE__ __device__
264 void calculate_potential_switch_F_E(cl_nbparam_params_t *nbparam,
273 /* potential switch constants */
274 float switch_V3 = nbparam->vdw_switch.c3;
275 float switch_V4 = nbparam->vdw_switch.c4;
276 float switch_V5 = nbparam->vdw_switch.c5;
277 float switch_F2 = nbparam->vdw_switch.c3;
278 float switch_F3 = nbparam->vdw_switch.c4;
279 float switch_F4 = nbparam->vdw_switch.c5;
282 r_switch = r - nbparam->rvdw_switch;
283 r_switch = r_switch >= 0.0f ? r_switch : 0.0f;
285 /* Unlike in the F-only kernel, masking is faster here */
286 sw = 1.0f + (switch_V3 + (switch_V4 + switch_V5*r_switch)*r_switch)*r_switch*r_switch*r_switch;
287 dsw = (switch_F2 + (switch_F3 + switch_F4*r_switch)*r_switch)*r_switch*r_switch;
289 *F_invr = (*F_invr)*sw - inv_r*(*E_lj)*dsw;
293 /*! Calculate LJ-PME grid force contribution with
294 * geometric combination rule.
296 __INLINE__ __device__
297 void calculate_lj_ewald_comb_geom_F(__constant float * nbfp_comb_climg2d,
306 float c6grid, inv_r6_nm, cr2, expmcr2, poly;
308 c6grid = nbfp_comb_climg2d[2*typei]*nbfp_comb_climg2d[2*typej];
310 /* Recalculate inv_r6 without exclusion mask */
311 inv_r6_nm = inv_r2*inv_r2*inv_r2;
314 poly = 1.0f + cr2 + 0.5f*cr2*cr2;
316 /* Subtract the grid force from the total LJ force */
317 *F_invr += c6grid*(inv_r6_nm - expmcr2*(inv_r6_nm*poly + lje_coeff6_6))*inv_r2;
320 /*! Calculate LJ-PME grid force + energy contribution with
321 * geometric combination rule.
323 __INLINE__ __device__
324 void calculate_lj_ewald_comb_geom_F_E(__constant float *nbfp_comb_climg2d,
325 cl_nbparam_params_t *nbparam,
336 float c6grid, inv_r6_nm, cr2, expmcr2, poly, sh_mask;
338 c6grid = nbfp_comb_climg2d[2*typei]*nbfp_comb_climg2d[2*typej];
340 /* Recalculate inv_r6 without exclusion mask */
341 inv_r6_nm = inv_r2*inv_r2*inv_r2;
344 poly = 1.0f + cr2 + 0.5f*cr2*cr2;
346 /* Subtract the grid force from the total LJ force */
347 *F_invr += c6grid*(inv_r6_nm - expmcr2*(inv_r6_nm*poly + lje_coeff6_6))*inv_r2;
349 /* Shift should be applied only to real LJ pairs */
350 sh_mask = nbparam->sh_lj_ewald*int_bit;
351 *E_lj += ONE_SIXTH_F*c6grid*(inv_r6_nm*(1.0f - expmcr2*poly) + sh_mask);
354 /*! Calculate LJ-PME grid force + energy contribution (if E_lj != NULL) with
355 * Lorentz-Berthelot combination rule.
356 * We use a single F+E kernel with conditional because the performance impact
357 * of this is pretty small and LB on the CPU is anyway very slow.
359 __INLINE__ __device__
360 void calculate_lj_ewald_comb_LB_F_E(__constant float *nbfp_comb_climg2d,
361 cl_nbparam_params_t *nbparam,
373 float c6grid, inv_r6_nm, cr2, expmcr2, poly;
374 float sigma, sigma2, epsilon;
376 /* sigma and epsilon are scaled to give 6*C6 */
377 sigma = nbfp_comb_climg2d[2*typei] + nbfp_comb_climg2d[2*typej];
379 epsilon = nbfp_comb_climg2d[2*typei+1]*nbfp_comb_climg2d[2*typej+1];
381 sigma2 = sigma*sigma;
382 c6grid = epsilon*sigma2*sigma2*sigma2;
384 /* Recalculate inv_r6 without exclusion mask */
385 inv_r6_nm = inv_r2*inv_r2*inv_r2;
388 poly = 1.0f + cr2 + 0.5f*cr2*cr2;
390 /* Subtract the grid force from the total LJ force */
391 *F_invr += c6grid*(inv_r6_nm - expmcr2*(inv_r6_nm*poly + lje_coeff6_6))*inv_r2;
397 /* Shift should be applied only to real LJ pairs */
398 sh_mask = nbparam->sh_lj_ewald*int_bit;
399 *E_lj += ONE_SIXTH_F*c6grid*(inv_r6_nm*(1.0f - expmcr2*poly) + sh_mask);
403 /*! Interpolate Ewald coulomb force using the table through the tex_nbfp texture.
404 * Original idea: from the OpenMM project
406 __INLINE__ __device__ float
407 interpolate_coulomb_force_r(__constant float *coulomb_tab_climg2d,
411 float normalized = scale * r;
412 int index = (int) normalized;
413 float fract2 = normalized - index;
414 float fract1 = 1.0f - fract2;
416 return fract1*coulomb_tab_climg2d[index] +
417 fract2*coulomb_tab_climg2d[index + 1];
420 /*! Calculate analytical Ewald correction term. */
421 __INLINE__ __device__
422 float pmecorrF(float z2)
424 const float FN6 = -1.7357322914161492954e-8f;
425 const float FN5 = 1.4703624142580877519e-6f;
426 const float FN4 = -0.000053401640219807709149f;
427 const float FN3 = 0.0010054721316683106153f;
428 const float FN2 = -0.019278317264888380590f;
429 const float FN1 = 0.069670166153766424023f;
430 const float FN0 = -0.75225204789749321333f;
432 const float FD4 = 0.0011193462567257629232f;
433 const float FD3 = 0.014866955030185295499f;
434 const float FD2 = 0.11583842382862377919f;
435 const float FD1 = 0.50736591960530292870f;
436 const float FD0 = 1.0f;
439 float polyFN0, polyFN1, polyFD0, polyFD1;
443 polyFD0 = FD4*z4 + FD2;
444 polyFD1 = FD3*z4 + FD1;
445 polyFD0 = polyFD0*z4 + FD0;
446 polyFD0 = polyFD1*z2 + polyFD0;
448 polyFD0 = 1.0f/polyFD0;
450 polyFN0 = FN6*z4 + FN4;
451 polyFN1 = FN5*z4 + FN3;
452 polyFN0 = polyFN0*z4 + FN2;
453 polyFN1 = polyFN1*z4 + FN1;
454 polyFN0 = polyFN0*z4 + FN0;
455 polyFN0 = polyFN1*z2 + polyFN0;
457 return polyFN0*polyFD0;
460 /*! Final j-force reduction; this generic implementation works with
461 * arbitrary array sizes.
463 /* AMD OpenCL compiler error "Undeclared function index 1024" if __INLINE__d */
464 __INLINE__ __device__
465 void reduce_force_j_generic(__local float *f_buf, __global float *fout,
466 int tidxi, int tidxj, int aidx)
468 /* Split the reduction between the first 3 column threads
469 Threads with column id 0 will do the reduction for (float3).x components
470 Threads with column id 1 will do the reduction for (float3).y components
471 Threads with column id 2 will do the reduction for (float3).z components.
472 The reduction is performed for each line tidxj of f_buf. */
476 for (int j = tidxj * CL_SIZE; j < (tidxj + 1) * CL_SIZE; j++)
478 f += f_buf[FBUF_STRIDE * tidxi + j];
481 atomicAdd_g_f(&fout[3 * aidx + tidxi], f);
485 /*! Final i-force reduction; this generic implementation works with
486 * arbitrary array sizes.
488 __INLINE__ __device__
489 void reduce_force_i_generic(__local float *f_buf, __global float *fout,
490 float *fshift_buf, bool bCalcFshift,
491 int tidxi, int tidxj, int aidx)
493 /* Split the reduction between the first 3 line threads
494 Threads with line id 0 will do the reduction for (float3).x components
495 Threads with line id 1 will do the reduction for (float3).y components
496 Threads with line id 2 will do the reduction for (float3).z components. */
500 for (int j = tidxi; j < CL_SIZE_SQ; j += CL_SIZE)
502 f += f_buf[tidxj * FBUF_STRIDE + j];
505 atomicAdd_g_f(&fout[3 * aidx + tidxj], f);
514 /*! Final i-force reduction; this implementation works only with power of two
517 __INLINE__ __device__
518 void reduce_force_i_pow2(volatile __local float *f_buf, __global float *fout,
519 float *fshift_buf, bool bCalcFshift,
520 int tidxi, int tidxj, int aidx)
523 /* Reduce the initial CL_SIZE values for each i atom to half
524 * every step by using CL_SIZE * i threads.
525 * Can't just use i as loop variable because than nvcc refuses to unroll.
528 for (j = CL_SIZE_LOG2 - 1; j > 0; j--)
533 f_buf[ tidxj * CL_SIZE + tidxi] += f_buf[ (tidxj + i) * CL_SIZE + tidxi];
534 f_buf[ FBUF_STRIDE + tidxj * CL_SIZE + tidxi] += f_buf[ FBUF_STRIDE + (tidxj + i) * CL_SIZE + tidxi];
535 f_buf[2 * FBUF_STRIDE + tidxj * CL_SIZE + tidxi] += f_buf[2 * FBUF_STRIDE + (tidxj + i) * CL_SIZE + tidxi];
540 /* i == 1, last reduction step, writing to global mem */
541 /* Split the reduction between the first 3 line threads
542 Threads with line id 0 will do the reduction for (float3).x components
543 Threads with line id 1 will do the reduction for (float3).y components
544 Threads with line id 2 will do the reduction for (float3).z components. */
547 float f = f_buf[tidxj * FBUF_STRIDE + tidxi] + f_buf[tidxj * FBUF_STRIDE + i * CL_SIZE + tidxi];
549 atomicAdd_g_f(&fout[3 * aidx + tidxj], f);
558 /*! Final i-force reduction wrapper; calls the generic or pow2 reduction depending
559 * on whether the size of the array to be reduced is power of two or not.
561 __INLINE__ __device__
562 void reduce_force_i(__local float *f_buf, __global float *f,
563 float *fshift_buf, bool bCalcFshift,
564 int tidxi, int tidxj, int ai)
566 if ((CL_SIZE & (CL_SIZE - 1)))
568 reduce_force_i_generic(f_buf, f, fshift_buf, bCalcFshift, tidxi, tidxj, ai);
572 reduce_force_i_pow2(f_buf, f, fshift_buf, bCalcFshift, tidxi, tidxj, ai);
576 /*! Energy reduction; this implementation works only with power of two
579 __INLINE__ __device__
580 void reduce_energy_pow2(volatile __local float *buf,
581 volatile __global float *e_lj,
582 volatile __global float *e_el,
590 /* Can't just use i as loop variable because than nvcc refuses to unroll. */
591 for (j = WARP_SIZE_LOG2 - 1; j > 0; j--)
595 buf[ tidx] += buf[ tidx + i];
596 buf[FBUF_STRIDE + tidx] += buf[FBUF_STRIDE + tidx + i];
601 /* last reduction step, writing to global mem */
604 e1 = buf[ tidx] + buf[ tidx + i];
605 e2 = buf[FBUF_STRIDE + tidx] + buf[FBUF_STRIDE + tidx + i];
607 atomicAdd_g_f(e_lj, e1);
608 atomicAdd_g_f(e_el, e2);
612 /*! Writes in debug_buffer the input value.
613 * Each thread has its own unique location in debug_buffer.
614 * Works for 2D global configurations.
616 void print_to_debug_buffer_f(__global float* debug_buffer, float value)
619 debug_buffer[get_global_id(1) * get_global_size(0) + get_global_id(0)] = value;
622 #endif /* NBNXN_OPENCL_KERNEL_UTILS_CLH */