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.
36 * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "kernelutil_sparc64_hpc_ace_double.h"
52 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_sparc64_hpc_ace_double
53 * Electrostatics interaction: Ewald
54 * VdW interaction: LJEwald
55 * Geometry: Particle-Particle
56 * Calculate force/pot: PotentialAndForce
59 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_sparc64_hpc_ace_double
60 (t_nblist * gmx_restrict nlist,
61 rvec * gmx_restrict xx,
62 rvec * gmx_restrict ff,
63 t_forcerec * gmx_restrict fr,
64 t_mdatoms * gmx_restrict mdatoms,
65 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
66 t_nrnb * gmx_restrict nrnb)
68 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
69 * just 0 for non-waters.
70 * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
71 * jnr indices corresponding to data put in the four positions in the SIMD register.
73 int i_shift_offset,i_coord_offset,outeriter,inneriter;
74 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int j_coord_offsetA,j_coord_offsetB;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real *shiftvec,*fshift,*x,*f;
80 _fjsp_v2r8 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
82 _fjsp_v2r8 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
83 int vdwjidx0A,vdwjidx0B;
84 _fjsp_v2r8 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
85 _fjsp_v2r8 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
86 _fjsp_v2r8 velec,felec,velecsum,facel,crf,krf,krf2;
89 _fjsp_v2r8 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
92 _fjsp_v2r8 one_sixth = gmx_fjsp_set1_v2r8(1.0/6.0);
93 _fjsp_v2r8 one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
96 _fjsp_v2r8 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
97 _fjsp_v2r8 one_half = gmx_fjsp_set1_v2r8(0.5);
98 _fjsp_v2r8 minus_one = gmx_fjsp_set1_v2r8(-1.0);
99 _fjsp_v2r8 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
102 _fjsp_v2r8 dummy_mask,cutoff_mask;
103 _fjsp_v2r8 one = gmx_fjsp_set1_v2r8(1.0);
104 _fjsp_v2r8 two = gmx_fjsp_set1_v2r8(2.0);
105 union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
112 jindex = nlist->jindex;
114 shiftidx = nlist->shift;
116 shiftvec = fr->shift_vec[0];
117 fshift = fr->fshift[0];
118 facel = gmx_fjsp_set1_v2r8(fr->epsfac);
119 charge = mdatoms->chargeA;
120 nvdwtype = fr->ntype;
122 vdwtype = mdatoms->typeA;
123 vdwgridparam = fr->ljpme_c6grid;
124 sh_lj_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
125 ewclj = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
126 ewclj2 = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
128 sh_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
129 ewtab = fr->ic->tabq_coul_FDV0;
130 ewtabscale = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
131 ewtabhalfspace = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
133 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
134 rcutoff_scalar = fr->rcoulomb;
135 rcutoff = gmx_fjsp_set1_v2r8(rcutoff_scalar);
136 rcutoff2 = _fjsp_mul_v2r8(rcutoff,rcutoff);
138 sh_vdw_invrcut6 = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
139 rvdw = gmx_fjsp_set1_v2r8(fr->rvdw);
141 /* Avoid stupid compiler warnings */
149 /* Start outer loop over neighborlists */
150 for(iidx=0; iidx<nri; iidx++)
152 /* Load shift vector for this list */
153 i_shift_offset = DIM*shiftidx[iidx];
155 /* Load limits for loop over neighbors */
156 j_index_start = jindex[iidx];
157 j_index_end = jindex[iidx+1];
159 /* Get outer coordinate index */
161 i_coord_offset = DIM*inr;
163 /* Load i particle coords and add shift vector */
164 gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
166 fix0 = _fjsp_setzero_v2r8();
167 fiy0 = _fjsp_setzero_v2r8();
168 fiz0 = _fjsp_setzero_v2r8();
170 /* Load parameters for i particles */
171 iq0 = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
172 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
174 /* Reset potential sums */
175 velecsum = _fjsp_setzero_v2r8();
176 vvdwsum = _fjsp_setzero_v2r8();
178 /* Start inner kernel loop */
179 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
182 /* Get j neighbor index, and coordinate index */
185 j_coord_offsetA = DIM*jnrA;
186 j_coord_offsetB = DIM*jnrB;
188 /* load j atom coordinates */
189 gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
192 /* Calculate displacement vector */
193 dx00 = _fjsp_sub_v2r8(ix0,jx0);
194 dy00 = _fjsp_sub_v2r8(iy0,jy0);
195 dz00 = _fjsp_sub_v2r8(iz0,jz0);
197 /* Calculate squared distance and things based on it */
198 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
200 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
202 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
204 /* Load parameters for j particles */
205 jq0 = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
206 vdwjidx0A = 2*vdwtype[jnrA+0];
207 vdwjidx0B = 2*vdwtype[jnrB+0];
209 /**************************
210 * CALCULATE INTERACTIONS *
211 **************************/
213 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
216 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
218 /* Compute parameters for interactions between i and j atoms */
219 qq00 = _fjsp_mul_v2r8(iq0,jq0);
220 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
221 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
223 c6grid_00 = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
224 vdwgridparam+vdwioffset0+vdwjidx0B);
226 /* EWALD ELECTROSTATICS */
228 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
229 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
230 itab_tmp = _fjsp_dtox_v2r8(ewrt);
231 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
232 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
234 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
235 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
236 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
237 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
238 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
239 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
240 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
241 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
242 velec = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
243 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
245 /* Analytical LJ-PME */
246 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
247 ewcljrsq = _fjsp_mul_v2r8(ewclj2,rsq00);
248 ewclj6 = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
249 exponent = gmx_simd_exp_d(-ewcljrsq);
250 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
251 poly = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
252 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
253 vvdw6 = _fjsp_mul_v2r8(_fjsp_madd_v2r8(-c6grid_00,_fjsp_sub_v2r8(one,poly),c6_00),rinvsix);
254 vvdw12 = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
255 vvdw = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
256 _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw6,_fjsp_madd_v2r8(c6grid_00,sh_lj_ewald,_fjsp_mul_v2r8(c6_00,sh_vdw_invrcut6))),one_sixth));
257 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
258 fvdw = _fjsp_mul_v2r8(_fjsp_add_v2r8(vvdw12,_fjsp_msub_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6),vvdw6)),rinvsq00);
260 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
262 /* Update potential sum for this i atom from the interaction with this j atom. */
263 velec = _fjsp_and_v2r8(velec,cutoff_mask);
264 velecsum = _fjsp_add_v2r8(velecsum,velec);
265 vvdw = _fjsp_and_v2r8(vvdw,cutoff_mask);
266 vvdwsum = _fjsp_add_v2r8(vvdwsum,vvdw);
268 fscal = _fjsp_add_v2r8(felec,fvdw);
270 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
272 /* Update vectorial force */
273 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
274 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
275 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
277 gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
281 /* Inner loop uses 79 flops */
288 j_coord_offsetA = DIM*jnrA;
290 /* load j atom coordinates */
291 gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
294 /* Calculate displacement vector */
295 dx00 = _fjsp_sub_v2r8(ix0,jx0);
296 dy00 = _fjsp_sub_v2r8(iy0,jy0);
297 dz00 = _fjsp_sub_v2r8(iz0,jz0);
299 /* Calculate squared distance and things based on it */
300 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
302 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
304 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
306 /* Load parameters for j particles */
307 jq0 = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
308 vdwjidx0A = 2*vdwtype[jnrA+0];
310 /**************************
311 * CALCULATE INTERACTIONS *
312 **************************/
314 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
317 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
319 /* Compute parameters for interactions between i and j atoms */
320 qq00 = _fjsp_mul_v2r8(iq0,jq0);
321 gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
323 c6grid_00 = gmx_fjsp_load_1real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A);
325 /* EWALD ELECTROSTATICS */
327 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
328 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
329 itab_tmp = _fjsp_dtox_v2r8(ewrt);
330 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
331 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
333 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
334 ewtabD = _fjsp_setzero_v2r8();
335 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
336 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
337 ewtabFn = _fjsp_setzero_v2r8();
338 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
339 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
340 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
341 velec = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
342 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
344 /* Analytical LJ-PME */
345 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
346 ewcljrsq = _fjsp_mul_v2r8(ewclj2,rsq00);
347 ewclj6 = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
348 exponent = gmx_simd_exp_d(-ewcljrsq);
349 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
350 poly = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
351 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
352 vvdw6 = _fjsp_mul_v2r8(_fjsp_madd_v2r8(-c6grid_00,_fjsp_sub_v2r8(one,poly),c6_00),rinvsix);
353 vvdw12 = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
354 vvdw = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
355 _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw6,_fjsp_madd_v2r8(c6grid_00,sh_lj_ewald,_fjsp_mul_v2r8(c6_00,sh_vdw_invrcut6))),one_sixth));
356 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
357 fvdw = _fjsp_mul_v2r8(_fjsp_add_v2r8(vvdw12,_fjsp_msub_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6),vvdw6)),rinvsq00);
359 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
361 /* Update potential sum for this i atom from the interaction with this j atom. */
362 velec = _fjsp_and_v2r8(velec,cutoff_mask);
363 velec = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
364 velecsum = _fjsp_add_v2r8(velecsum,velec);
365 vvdw = _fjsp_and_v2r8(vvdw,cutoff_mask);
366 vvdw = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
367 vvdwsum = _fjsp_add_v2r8(vvdwsum,vvdw);
369 fscal = _fjsp_add_v2r8(felec,fvdw);
371 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
373 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
375 /* Update vectorial force */
376 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
377 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
378 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
380 gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
384 /* Inner loop uses 79 flops */
387 /* End of innermost loop */
389 gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
390 f+i_coord_offset,fshift+i_shift_offset);
393 /* Update potential energies */
394 gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
395 gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
397 /* Increment number of inner iterations */
398 inneriter += j_index_end - j_index_start;
400 /* Outer loop uses 9 flops */
403 /* Increment number of outer iterations */
406 /* Update outer/inner flops */
408 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*79);
411 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sparc64_hpc_ace_double
412 * Electrostatics interaction: Ewald
413 * VdW interaction: LJEwald
414 * Geometry: Particle-Particle
415 * Calculate force/pot: Force
418 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sparc64_hpc_ace_double
419 (t_nblist * gmx_restrict nlist,
420 rvec * gmx_restrict xx,
421 rvec * gmx_restrict ff,
422 t_forcerec * gmx_restrict fr,
423 t_mdatoms * gmx_restrict mdatoms,
424 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
425 t_nrnb * gmx_restrict nrnb)
427 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
428 * just 0 for non-waters.
429 * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
430 * jnr indices corresponding to data put in the four positions in the SIMD register.
432 int i_shift_offset,i_coord_offset,outeriter,inneriter;
433 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
435 int j_coord_offsetA,j_coord_offsetB;
436 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
438 real *shiftvec,*fshift,*x,*f;
439 _fjsp_v2r8 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
441 _fjsp_v2r8 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
442 int vdwjidx0A,vdwjidx0B;
443 _fjsp_v2r8 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
444 _fjsp_v2r8 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
445 _fjsp_v2r8 velec,felec,velecsum,facel,crf,krf,krf2;
448 _fjsp_v2r8 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
451 _fjsp_v2r8 one_sixth = gmx_fjsp_set1_v2r8(1.0/6.0);
452 _fjsp_v2r8 one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
453 _fjsp_v2r8 c6grid_00;
455 _fjsp_v2r8 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
456 _fjsp_v2r8 one_half = gmx_fjsp_set1_v2r8(0.5);
457 _fjsp_v2r8 minus_one = gmx_fjsp_set1_v2r8(-1.0);
458 _fjsp_v2r8 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
461 _fjsp_v2r8 dummy_mask,cutoff_mask;
462 _fjsp_v2r8 one = gmx_fjsp_set1_v2r8(1.0);
463 _fjsp_v2r8 two = gmx_fjsp_set1_v2r8(2.0);
464 union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
471 jindex = nlist->jindex;
473 shiftidx = nlist->shift;
475 shiftvec = fr->shift_vec[0];
476 fshift = fr->fshift[0];
477 facel = gmx_fjsp_set1_v2r8(fr->epsfac);
478 charge = mdatoms->chargeA;
479 nvdwtype = fr->ntype;
481 vdwtype = mdatoms->typeA;
482 vdwgridparam = fr->ljpme_c6grid;
483 sh_lj_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
484 ewclj = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
485 ewclj2 = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
487 sh_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
488 ewtab = fr->ic->tabq_coul_F;
489 ewtabscale = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
490 ewtabhalfspace = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
492 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
493 rcutoff_scalar = fr->rcoulomb;
494 rcutoff = gmx_fjsp_set1_v2r8(rcutoff_scalar);
495 rcutoff2 = _fjsp_mul_v2r8(rcutoff,rcutoff);
497 sh_vdw_invrcut6 = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
498 rvdw = gmx_fjsp_set1_v2r8(fr->rvdw);
500 /* Avoid stupid compiler warnings */
508 /* Start outer loop over neighborlists */
509 for(iidx=0; iidx<nri; iidx++)
511 /* Load shift vector for this list */
512 i_shift_offset = DIM*shiftidx[iidx];
514 /* Load limits for loop over neighbors */
515 j_index_start = jindex[iidx];
516 j_index_end = jindex[iidx+1];
518 /* Get outer coordinate index */
520 i_coord_offset = DIM*inr;
522 /* Load i particle coords and add shift vector */
523 gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
525 fix0 = _fjsp_setzero_v2r8();
526 fiy0 = _fjsp_setzero_v2r8();
527 fiz0 = _fjsp_setzero_v2r8();
529 /* Load parameters for i particles */
530 iq0 = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
531 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
533 /* Start inner kernel loop */
534 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
537 /* Get j neighbor index, and coordinate index */
540 j_coord_offsetA = DIM*jnrA;
541 j_coord_offsetB = DIM*jnrB;
543 /* load j atom coordinates */
544 gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
547 /* Calculate displacement vector */
548 dx00 = _fjsp_sub_v2r8(ix0,jx0);
549 dy00 = _fjsp_sub_v2r8(iy0,jy0);
550 dz00 = _fjsp_sub_v2r8(iz0,jz0);
552 /* Calculate squared distance and things based on it */
553 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
555 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
557 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
559 /* Load parameters for j particles */
560 jq0 = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
561 vdwjidx0A = 2*vdwtype[jnrA+0];
562 vdwjidx0B = 2*vdwtype[jnrB+0];
564 /**************************
565 * CALCULATE INTERACTIONS *
566 **************************/
568 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
571 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
573 /* Compute parameters for interactions between i and j atoms */
574 qq00 = _fjsp_mul_v2r8(iq0,jq0);
575 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
576 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
578 c6grid_00 = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
579 vdwgridparam+vdwioffset0+vdwjidx0B);
581 /* EWALD ELECTROSTATICS */
583 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
584 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
585 itab_tmp = _fjsp_dtox_v2r8(ewrt);
586 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
587 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
589 gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
591 felec = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
592 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
594 /* Analytical LJ-PME */
595 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
596 ewcljrsq = _fjsp_mul_v2r8(ewclj2,rsq00);
597 ewclj6 = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
598 exponent = gmx_simd_exp_d(-ewcljrsq);
599 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
600 poly = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
601 /* f6A = 6 * C6grid * (1 - poly) */
602 f6A = _fjsp_mul_v2r8(c6grid_00,_fjsp_msub_v2r8(one,poly));
603 /* f6B = C6grid * exponent * beta^6 */
604 f6B = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
605 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
606 fvdw = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
608 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
610 fscal = _fjsp_add_v2r8(felec,fvdw);
612 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
614 /* Update vectorial force */
615 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
616 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
617 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
619 gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
623 /* Inner loop uses 64 flops */
630 j_coord_offsetA = DIM*jnrA;
632 /* load j atom coordinates */
633 gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
636 /* Calculate displacement vector */
637 dx00 = _fjsp_sub_v2r8(ix0,jx0);
638 dy00 = _fjsp_sub_v2r8(iy0,jy0);
639 dz00 = _fjsp_sub_v2r8(iz0,jz0);
641 /* Calculate squared distance and things based on it */
642 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
644 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
646 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
648 /* Load parameters for j particles */
649 jq0 = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
650 vdwjidx0A = 2*vdwtype[jnrA+0];
652 /**************************
653 * CALCULATE INTERACTIONS *
654 **************************/
656 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
659 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
661 /* Compute parameters for interactions between i and j atoms */
662 qq00 = _fjsp_mul_v2r8(iq0,jq0);
663 gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
665 c6grid_00 = gmx_fjsp_load_1real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A);
667 /* EWALD ELECTROSTATICS */
669 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
670 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
671 itab_tmp = _fjsp_dtox_v2r8(ewrt);
672 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
673 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
675 gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
676 felec = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
677 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
679 /* Analytical LJ-PME */
680 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
681 ewcljrsq = _fjsp_mul_v2r8(ewclj2,rsq00);
682 ewclj6 = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
683 exponent = gmx_simd_exp_d(-ewcljrsq);
684 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
685 poly = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
686 /* f6A = 6 * C6grid * (1 - poly) */
687 f6A = _fjsp_mul_v2r8(c6grid_00,_fjsp_msub_v2r8(one,poly));
688 /* f6B = C6grid * exponent * beta^6 */
689 f6B = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
690 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
691 fvdw = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
693 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
695 fscal = _fjsp_add_v2r8(felec,fvdw);
697 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
699 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
701 /* Update vectorial force */
702 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
703 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
704 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
706 gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
710 /* Inner loop uses 64 flops */
713 /* End of innermost loop */
715 gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
716 f+i_coord_offset,fshift+i_shift_offset);
718 /* Increment number of inner iterations */
719 inneriter += j_index_end - j_index_start;
721 /* Outer loop uses 7 flops */
724 /* Increment number of outer iterations */
727 /* Update outer/inner flops */
729 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*64);