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_GeomW3P1_VF_sparc64_hpc_ace_double
53 * Electrostatics interaction: Ewald
54 * VdW interaction: LJEwald
55 * Geometry: Water3-Particle
56 * Calculate force/pot: PotentialAndForce
59 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_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;
84 _fjsp_v2r8 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
86 _fjsp_v2r8 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
87 int vdwjidx0A,vdwjidx0B;
88 _fjsp_v2r8 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
89 _fjsp_v2r8 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
90 _fjsp_v2r8 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
91 _fjsp_v2r8 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
92 _fjsp_v2r8 velec,felec,velecsum,facel,crf,krf,krf2;
95 _fjsp_v2r8 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
98 _fjsp_v2r8 one_sixth = gmx_fjsp_set1_v2r8(1.0/6.0);
99 _fjsp_v2r8 one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
100 _fjsp_v2r8 c6grid_00;
101 _fjsp_v2r8 c6grid_10;
102 _fjsp_v2r8 c6grid_20;
104 _fjsp_v2r8 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
105 _fjsp_v2r8 one_half = gmx_fjsp_set1_v2r8(0.5);
106 _fjsp_v2r8 minus_one = gmx_fjsp_set1_v2r8(-1.0);
107 _fjsp_v2r8 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
110 _fjsp_v2r8 dummy_mask,cutoff_mask;
111 _fjsp_v2r8 one = gmx_fjsp_set1_v2r8(1.0);
112 _fjsp_v2r8 two = gmx_fjsp_set1_v2r8(2.0);
113 union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
120 jindex = nlist->jindex;
122 shiftidx = nlist->shift;
124 shiftvec = fr->shift_vec[0];
125 fshift = fr->fshift[0];
126 facel = gmx_fjsp_set1_v2r8(fr->epsfac);
127 charge = mdatoms->chargeA;
128 nvdwtype = fr->ntype;
130 vdwtype = mdatoms->typeA;
131 vdwgridparam = fr->ljpme_c6grid;
132 sh_lj_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
133 ewclj = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
134 ewclj2 = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
136 sh_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
137 ewtab = fr->ic->tabq_coul_FDV0;
138 ewtabscale = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
139 ewtabhalfspace = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
141 /* Setup water-specific parameters */
142 inr = nlist->iinr[0];
143 iq0 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+0]));
144 iq1 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+1]));
145 iq2 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+2]));
146 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
148 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
149 rcutoff_scalar = fr->rcoulomb;
150 rcutoff = gmx_fjsp_set1_v2r8(rcutoff_scalar);
151 rcutoff2 = _fjsp_mul_v2r8(rcutoff,rcutoff);
153 sh_vdw_invrcut6 = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
154 rvdw = gmx_fjsp_set1_v2r8(fr->rvdw);
156 /* Avoid stupid compiler warnings */
164 /* Start outer loop over neighborlists */
165 for(iidx=0; iidx<nri; iidx++)
167 /* Load shift vector for this list */
168 i_shift_offset = DIM*shiftidx[iidx];
170 /* Load limits for loop over neighbors */
171 j_index_start = jindex[iidx];
172 j_index_end = jindex[iidx+1];
174 /* Get outer coordinate index */
176 i_coord_offset = DIM*inr;
178 /* Load i particle coords and add shift vector */
179 gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
180 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
182 fix0 = _fjsp_setzero_v2r8();
183 fiy0 = _fjsp_setzero_v2r8();
184 fiz0 = _fjsp_setzero_v2r8();
185 fix1 = _fjsp_setzero_v2r8();
186 fiy1 = _fjsp_setzero_v2r8();
187 fiz1 = _fjsp_setzero_v2r8();
188 fix2 = _fjsp_setzero_v2r8();
189 fiy2 = _fjsp_setzero_v2r8();
190 fiz2 = _fjsp_setzero_v2r8();
192 /* Reset potential sums */
193 velecsum = _fjsp_setzero_v2r8();
194 vvdwsum = _fjsp_setzero_v2r8();
196 /* Start inner kernel loop */
197 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
200 /* Get j neighbor index, and coordinate index */
203 j_coord_offsetA = DIM*jnrA;
204 j_coord_offsetB = DIM*jnrB;
206 /* load j atom coordinates */
207 gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
210 /* Calculate displacement vector */
211 dx00 = _fjsp_sub_v2r8(ix0,jx0);
212 dy00 = _fjsp_sub_v2r8(iy0,jy0);
213 dz00 = _fjsp_sub_v2r8(iz0,jz0);
214 dx10 = _fjsp_sub_v2r8(ix1,jx0);
215 dy10 = _fjsp_sub_v2r8(iy1,jy0);
216 dz10 = _fjsp_sub_v2r8(iz1,jz0);
217 dx20 = _fjsp_sub_v2r8(ix2,jx0);
218 dy20 = _fjsp_sub_v2r8(iy2,jy0);
219 dz20 = _fjsp_sub_v2r8(iz2,jz0);
221 /* Calculate squared distance and things based on it */
222 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
223 rsq10 = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
224 rsq20 = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
226 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
227 rinv10 = gmx_fjsp_invsqrt_v2r8(rsq10);
228 rinv20 = gmx_fjsp_invsqrt_v2r8(rsq20);
230 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
231 rinvsq10 = _fjsp_mul_v2r8(rinv10,rinv10);
232 rinvsq20 = _fjsp_mul_v2r8(rinv20,rinv20);
234 /* Load parameters for j particles */
235 jq0 = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
236 vdwjidx0A = 2*vdwtype[jnrA+0];
237 vdwjidx0B = 2*vdwtype[jnrB+0];
239 fjx0 = _fjsp_setzero_v2r8();
240 fjy0 = _fjsp_setzero_v2r8();
241 fjz0 = _fjsp_setzero_v2r8();
243 /**************************
244 * CALCULATE INTERACTIONS *
245 **************************/
247 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
250 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
252 /* Compute parameters for interactions between i and j atoms */
253 qq00 = _fjsp_mul_v2r8(iq0,jq0);
254 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
255 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
257 c6grid_00 = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
258 vdwgridparam+vdwioffset0+vdwjidx0B);
260 /* EWALD ELECTROSTATICS */
262 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
263 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
264 itab_tmp = _fjsp_dtox_v2r8(ewrt);
265 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
266 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
268 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
269 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
270 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
271 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
272 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
273 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
274 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
275 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
276 velec = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
277 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
279 /* Analytical LJ-PME */
280 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
281 ewcljrsq = _fjsp_mul_v2r8(ewclj2,rsq00);
282 ewclj6 = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
283 exponent = gmx_simd_exp_d(ewcljrsq);
284 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
285 poly = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
286 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
287 vvdw6 = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
288 vvdw12 = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
289 vvdw = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
290 _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));
291 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
292 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);
294 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
296 /* Update potential sum for this i atom from the interaction with this j atom. */
297 velec = _fjsp_and_v2r8(velec,cutoff_mask);
298 velecsum = _fjsp_add_v2r8(velecsum,velec);
299 vvdw = _fjsp_and_v2r8(vvdw,cutoff_mask);
300 vvdwsum = _fjsp_add_v2r8(vvdwsum,vvdw);
302 fscal = _fjsp_add_v2r8(felec,fvdw);
304 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
306 /* Update vectorial force */
307 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
308 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
309 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
311 fjx0 = _fjsp_madd_v2r8(dx00,fscal,fjx0);
312 fjy0 = _fjsp_madd_v2r8(dy00,fscal,fjy0);
313 fjz0 = _fjsp_madd_v2r8(dz00,fscal,fjz0);
317 /**************************
318 * CALCULATE INTERACTIONS *
319 **************************/
321 if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
324 r10 = _fjsp_mul_v2r8(rsq10,rinv10);
326 /* Compute parameters for interactions between i and j atoms */
327 qq10 = _fjsp_mul_v2r8(iq1,jq0);
329 /* EWALD ELECTROSTATICS */
331 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
332 ewrt = _fjsp_mul_v2r8(r10,ewtabscale);
333 itab_tmp = _fjsp_dtox_v2r8(ewrt);
334 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
335 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
337 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
338 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
339 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
340 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
341 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
342 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
343 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
344 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
345 velec = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv10,sh_ewald),velec));
346 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
348 cutoff_mask = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
350 /* Update potential sum for this i atom from the interaction with this j atom. */
351 velec = _fjsp_and_v2r8(velec,cutoff_mask);
352 velecsum = _fjsp_add_v2r8(velecsum,velec);
356 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
358 /* Update vectorial force */
359 fix1 = _fjsp_madd_v2r8(dx10,fscal,fix1);
360 fiy1 = _fjsp_madd_v2r8(dy10,fscal,fiy1);
361 fiz1 = _fjsp_madd_v2r8(dz10,fscal,fiz1);
363 fjx0 = _fjsp_madd_v2r8(dx10,fscal,fjx0);
364 fjy0 = _fjsp_madd_v2r8(dy10,fscal,fjy0);
365 fjz0 = _fjsp_madd_v2r8(dz10,fscal,fjz0);
369 /**************************
370 * CALCULATE INTERACTIONS *
371 **************************/
373 if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
376 r20 = _fjsp_mul_v2r8(rsq20,rinv20);
378 /* Compute parameters for interactions between i and j atoms */
379 qq20 = _fjsp_mul_v2r8(iq2,jq0);
381 /* EWALD ELECTROSTATICS */
383 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
384 ewrt = _fjsp_mul_v2r8(r20,ewtabscale);
385 itab_tmp = _fjsp_dtox_v2r8(ewrt);
386 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
387 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
389 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
390 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
391 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
392 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
393 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
394 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
395 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
396 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
397 velec = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv20,sh_ewald),velec));
398 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
400 cutoff_mask = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
402 /* Update potential sum for this i atom from the interaction with this j atom. */
403 velec = _fjsp_and_v2r8(velec,cutoff_mask);
404 velecsum = _fjsp_add_v2r8(velecsum,velec);
408 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
410 /* Update vectorial force */
411 fix2 = _fjsp_madd_v2r8(dx20,fscal,fix2);
412 fiy2 = _fjsp_madd_v2r8(dy20,fscal,fiy2);
413 fiz2 = _fjsp_madd_v2r8(dz20,fscal,fiz2);
415 fjx0 = _fjsp_madd_v2r8(dx20,fscal,fjx0);
416 fjy0 = _fjsp_madd_v2r8(dy20,fscal,fjy0);
417 fjz0 = _fjsp_madd_v2r8(dz20,fscal,fjz0);
421 gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
423 /* Inner loop uses 180 flops */
430 j_coord_offsetA = DIM*jnrA;
432 /* load j atom coordinates */
433 gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
436 /* Calculate displacement vector */
437 dx00 = _fjsp_sub_v2r8(ix0,jx0);
438 dy00 = _fjsp_sub_v2r8(iy0,jy0);
439 dz00 = _fjsp_sub_v2r8(iz0,jz0);
440 dx10 = _fjsp_sub_v2r8(ix1,jx0);
441 dy10 = _fjsp_sub_v2r8(iy1,jy0);
442 dz10 = _fjsp_sub_v2r8(iz1,jz0);
443 dx20 = _fjsp_sub_v2r8(ix2,jx0);
444 dy20 = _fjsp_sub_v2r8(iy2,jy0);
445 dz20 = _fjsp_sub_v2r8(iz2,jz0);
447 /* Calculate squared distance and things based on it */
448 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
449 rsq10 = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
450 rsq20 = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
452 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
453 rinv10 = gmx_fjsp_invsqrt_v2r8(rsq10);
454 rinv20 = gmx_fjsp_invsqrt_v2r8(rsq20);
456 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
457 rinvsq10 = _fjsp_mul_v2r8(rinv10,rinv10);
458 rinvsq20 = _fjsp_mul_v2r8(rinv20,rinv20);
460 /* Load parameters for j particles */
461 jq0 = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
462 vdwjidx0A = 2*vdwtype[jnrA+0];
464 fjx0 = _fjsp_setzero_v2r8();
465 fjy0 = _fjsp_setzero_v2r8();
466 fjz0 = _fjsp_setzero_v2r8();
468 /**************************
469 * CALCULATE INTERACTIONS *
470 **************************/
472 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
475 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
477 /* Compute parameters for interactions between i and j atoms */
478 qq00 = _fjsp_mul_v2r8(iq0,jq0);
479 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
480 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
482 c6grid_00 = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
483 vdwgridparam+vdwioffset0+vdwjidx0B);
485 /* EWALD ELECTROSTATICS */
487 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
488 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
489 itab_tmp = _fjsp_dtox_v2r8(ewrt);
490 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
491 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
493 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
494 ewtabD = _fjsp_setzero_v2r8();
495 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
496 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
497 ewtabFn = _fjsp_setzero_v2r8();
498 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
499 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
500 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
501 velec = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
502 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
504 /* Analytical LJ-PME */
505 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
506 ewcljrsq = _fjsp_mul_v2r8(ewclj2,rsq00);
507 ewclj6 = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
508 exponent = gmx_simd_exp_d(ewcljrsq);
509 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
510 poly = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
511 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
512 vvdw6 = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
513 vvdw12 = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
514 vvdw = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
515 _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));
516 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
517 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);
519 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
521 /* Update potential sum for this i atom from the interaction with this j atom. */
522 velec = _fjsp_and_v2r8(velec,cutoff_mask);
523 velec = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
524 velecsum = _fjsp_add_v2r8(velecsum,velec);
525 vvdw = _fjsp_and_v2r8(vvdw,cutoff_mask);
526 vvdw = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
527 vvdwsum = _fjsp_add_v2r8(vvdwsum,vvdw);
529 fscal = _fjsp_add_v2r8(felec,fvdw);
531 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
533 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
535 /* Update vectorial force */
536 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
537 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
538 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
540 fjx0 = _fjsp_madd_v2r8(dx00,fscal,fjx0);
541 fjy0 = _fjsp_madd_v2r8(dy00,fscal,fjy0);
542 fjz0 = _fjsp_madd_v2r8(dz00,fscal,fjz0);
546 /**************************
547 * CALCULATE INTERACTIONS *
548 **************************/
550 if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
553 r10 = _fjsp_mul_v2r8(rsq10,rinv10);
555 /* Compute parameters for interactions between i and j atoms */
556 qq10 = _fjsp_mul_v2r8(iq1,jq0);
558 /* EWALD ELECTROSTATICS */
560 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
561 ewrt = _fjsp_mul_v2r8(r10,ewtabscale);
562 itab_tmp = _fjsp_dtox_v2r8(ewrt);
563 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
564 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
566 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
567 ewtabD = _fjsp_setzero_v2r8();
568 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
569 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
570 ewtabFn = _fjsp_setzero_v2r8();
571 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
572 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
573 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
574 velec = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv10,sh_ewald),velec));
575 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
577 cutoff_mask = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
579 /* Update potential sum for this i atom from the interaction with this j atom. */
580 velec = _fjsp_and_v2r8(velec,cutoff_mask);
581 velec = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
582 velecsum = _fjsp_add_v2r8(velecsum,velec);
586 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
588 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
590 /* Update vectorial force */
591 fix1 = _fjsp_madd_v2r8(dx10,fscal,fix1);
592 fiy1 = _fjsp_madd_v2r8(dy10,fscal,fiy1);
593 fiz1 = _fjsp_madd_v2r8(dz10,fscal,fiz1);
595 fjx0 = _fjsp_madd_v2r8(dx10,fscal,fjx0);
596 fjy0 = _fjsp_madd_v2r8(dy10,fscal,fjy0);
597 fjz0 = _fjsp_madd_v2r8(dz10,fscal,fjz0);
601 /**************************
602 * CALCULATE INTERACTIONS *
603 **************************/
605 if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
608 r20 = _fjsp_mul_v2r8(rsq20,rinv20);
610 /* Compute parameters for interactions between i and j atoms */
611 qq20 = _fjsp_mul_v2r8(iq2,jq0);
613 /* EWALD ELECTROSTATICS */
615 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
616 ewrt = _fjsp_mul_v2r8(r20,ewtabscale);
617 itab_tmp = _fjsp_dtox_v2r8(ewrt);
618 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
619 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
621 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
622 ewtabD = _fjsp_setzero_v2r8();
623 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
624 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
625 ewtabFn = _fjsp_setzero_v2r8();
626 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
627 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
628 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
629 velec = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv20,sh_ewald),velec));
630 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
632 cutoff_mask = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
634 /* Update potential sum for this i atom from the interaction with this j atom. */
635 velec = _fjsp_and_v2r8(velec,cutoff_mask);
636 velec = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
637 velecsum = _fjsp_add_v2r8(velecsum,velec);
641 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
643 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
645 /* Update vectorial force */
646 fix2 = _fjsp_madd_v2r8(dx20,fscal,fix2);
647 fiy2 = _fjsp_madd_v2r8(dy20,fscal,fiy2);
648 fiz2 = _fjsp_madd_v2r8(dz20,fscal,fiz2);
650 fjx0 = _fjsp_madd_v2r8(dx20,fscal,fjx0);
651 fjy0 = _fjsp_madd_v2r8(dy20,fscal,fjy0);
652 fjz0 = _fjsp_madd_v2r8(dz20,fscal,fjz0);
656 gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
658 /* Inner loop uses 180 flops */
661 /* End of innermost loop */
663 gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
664 f+i_coord_offset,fshift+i_shift_offset);
667 /* Update potential energies */
668 gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
669 gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
671 /* Increment number of inner iterations */
672 inneriter += j_index_end - j_index_start;
674 /* Outer loop uses 20 flops */
677 /* Increment number of outer iterations */
680 /* Update outer/inner flops */
682 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*180);
685 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sparc64_hpc_ace_double
686 * Electrostatics interaction: Ewald
687 * VdW interaction: LJEwald
688 * Geometry: Water3-Particle
689 * Calculate force/pot: Force
692 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sparc64_hpc_ace_double
693 (t_nblist * gmx_restrict nlist,
694 rvec * gmx_restrict xx,
695 rvec * gmx_restrict ff,
696 t_forcerec * gmx_restrict fr,
697 t_mdatoms * gmx_restrict mdatoms,
698 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
699 t_nrnb * gmx_restrict nrnb)
701 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
702 * just 0 for non-waters.
703 * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
704 * jnr indices corresponding to data put in the four positions in the SIMD register.
706 int i_shift_offset,i_coord_offset,outeriter,inneriter;
707 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
709 int j_coord_offsetA,j_coord_offsetB;
710 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
712 real *shiftvec,*fshift,*x,*f;
713 _fjsp_v2r8 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
715 _fjsp_v2r8 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
717 _fjsp_v2r8 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
719 _fjsp_v2r8 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
720 int vdwjidx0A,vdwjidx0B;
721 _fjsp_v2r8 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
722 _fjsp_v2r8 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
723 _fjsp_v2r8 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
724 _fjsp_v2r8 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
725 _fjsp_v2r8 velec,felec,velecsum,facel,crf,krf,krf2;
728 _fjsp_v2r8 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
731 _fjsp_v2r8 one_sixth = gmx_fjsp_set1_v2r8(1.0/6.0);
732 _fjsp_v2r8 one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
733 _fjsp_v2r8 c6grid_00;
734 _fjsp_v2r8 c6grid_10;
735 _fjsp_v2r8 c6grid_20;
737 _fjsp_v2r8 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
738 _fjsp_v2r8 one_half = gmx_fjsp_set1_v2r8(0.5);
739 _fjsp_v2r8 minus_one = gmx_fjsp_set1_v2r8(-1.0);
740 _fjsp_v2r8 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
743 _fjsp_v2r8 dummy_mask,cutoff_mask;
744 _fjsp_v2r8 one = gmx_fjsp_set1_v2r8(1.0);
745 _fjsp_v2r8 two = gmx_fjsp_set1_v2r8(2.0);
746 union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
753 jindex = nlist->jindex;
755 shiftidx = nlist->shift;
757 shiftvec = fr->shift_vec[0];
758 fshift = fr->fshift[0];
759 facel = gmx_fjsp_set1_v2r8(fr->epsfac);
760 charge = mdatoms->chargeA;
761 nvdwtype = fr->ntype;
763 vdwtype = mdatoms->typeA;
764 vdwgridparam = fr->ljpme_c6grid;
765 sh_lj_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
766 ewclj = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
767 ewclj2 = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
769 sh_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
770 ewtab = fr->ic->tabq_coul_F;
771 ewtabscale = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
772 ewtabhalfspace = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
774 /* Setup water-specific parameters */
775 inr = nlist->iinr[0];
776 iq0 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+0]));
777 iq1 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+1]));
778 iq2 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+2]));
779 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
781 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
782 rcutoff_scalar = fr->rcoulomb;
783 rcutoff = gmx_fjsp_set1_v2r8(rcutoff_scalar);
784 rcutoff2 = _fjsp_mul_v2r8(rcutoff,rcutoff);
786 sh_vdw_invrcut6 = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
787 rvdw = gmx_fjsp_set1_v2r8(fr->rvdw);
789 /* Avoid stupid compiler warnings */
797 /* Start outer loop over neighborlists */
798 for(iidx=0; iidx<nri; iidx++)
800 /* Load shift vector for this list */
801 i_shift_offset = DIM*shiftidx[iidx];
803 /* Load limits for loop over neighbors */
804 j_index_start = jindex[iidx];
805 j_index_end = jindex[iidx+1];
807 /* Get outer coordinate index */
809 i_coord_offset = DIM*inr;
811 /* Load i particle coords and add shift vector */
812 gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
813 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
815 fix0 = _fjsp_setzero_v2r8();
816 fiy0 = _fjsp_setzero_v2r8();
817 fiz0 = _fjsp_setzero_v2r8();
818 fix1 = _fjsp_setzero_v2r8();
819 fiy1 = _fjsp_setzero_v2r8();
820 fiz1 = _fjsp_setzero_v2r8();
821 fix2 = _fjsp_setzero_v2r8();
822 fiy2 = _fjsp_setzero_v2r8();
823 fiz2 = _fjsp_setzero_v2r8();
825 /* Start inner kernel loop */
826 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
829 /* Get j neighbor index, and coordinate index */
832 j_coord_offsetA = DIM*jnrA;
833 j_coord_offsetB = DIM*jnrB;
835 /* load j atom coordinates */
836 gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
839 /* Calculate displacement vector */
840 dx00 = _fjsp_sub_v2r8(ix0,jx0);
841 dy00 = _fjsp_sub_v2r8(iy0,jy0);
842 dz00 = _fjsp_sub_v2r8(iz0,jz0);
843 dx10 = _fjsp_sub_v2r8(ix1,jx0);
844 dy10 = _fjsp_sub_v2r8(iy1,jy0);
845 dz10 = _fjsp_sub_v2r8(iz1,jz0);
846 dx20 = _fjsp_sub_v2r8(ix2,jx0);
847 dy20 = _fjsp_sub_v2r8(iy2,jy0);
848 dz20 = _fjsp_sub_v2r8(iz2,jz0);
850 /* Calculate squared distance and things based on it */
851 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
852 rsq10 = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
853 rsq20 = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
855 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
856 rinv10 = gmx_fjsp_invsqrt_v2r8(rsq10);
857 rinv20 = gmx_fjsp_invsqrt_v2r8(rsq20);
859 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
860 rinvsq10 = _fjsp_mul_v2r8(rinv10,rinv10);
861 rinvsq20 = _fjsp_mul_v2r8(rinv20,rinv20);
863 /* Load parameters for j particles */
864 jq0 = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
865 vdwjidx0A = 2*vdwtype[jnrA+0];
866 vdwjidx0B = 2*vdwtype[jnrB+0];
868 fjx0 = _fjsp_setzero_v2r8();
869 fjy0 = _fjsp_setzero_v2r8();
870 fjz0 = _fjsp_setzero_v2r8();
872 /**************************
873 * CALCULATE INTERACTIONS *
874 **************************/
876 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
879 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
881 /* Compute parameters for interactions between i and j atoms */
882 qq00 = _fjsp_mul_v2r8(iq0,jq0);
883 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
884 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
886 c6grid_00 = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
887 vdwgridparam+vdwioffset0+vdwjidx0B);
889 /* EWALD ELECTROSTATICS */
891 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
892 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
893 itab_tmp = _fjsp_dtox_v2r8(ewrt);
894 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
895 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
897 gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
899 felec = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
900 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
902 /* Analytical LJ-PME */
903 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
904 ewcljrsq = _fjsp_mul_v2r8(ewclj2,rsq00);
905 ewclj6 = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
906 exponent = gmx_simd_exp_d(ewcljrsq);
907 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
908 poly = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
909 /* f6A = 6 * C6grid * (1 - poly) */
910 f6A = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
911 /* f6B = C6grid * exponent * beta^6 */
912 f6B = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
913 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
914 fvdw = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
916 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
918 fscal = _fjsp_add_v2r8(felec,fvdw);
920 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
922 /* Update vectorial force */
923 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
924 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
925 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
927 fjx0 = _fjsp_madd_v2r8(dx00,fscal,fjx0);
928 fjy0 = _fjsp_madd_v2r8(dy00,fscal,fjy0);
929 fjz0 = _fjsp_madd_v2r8(dz00,fscal,fjz0);
933 /**************************
934 * CALCULATE INTERACTIONS *
935 **************************/
937 if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
940 r10 = _fjsp_mul_v2r8(rsq10,rinv10);
942 /* Compute parameters for interactions between i and j atoms */
943 qq10 = _fjsp_mul_v2r8(iq1,jq0);
945 /* EWALD ELECTROSTATICS */
947 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
948 ewrt = _fjsp_mul_v2r8(r10,ewtabscale);
949 itab_tmp = _fjsp_dtox_v2r8(ewrt);
950 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
951 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
953 gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
955 felec = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
956 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
958 cutoff_mask = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
962 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
964 /* Update vectorial force */
965 fix1 = _fjsp_madd_v2r8(dx10,fscal,fix1);
966 fiy1 = _fjsp_madd_v2r8(dy10,fscal,fiy1);
967 fiz1 = _fjsp_madd_v2r8(dz10,fscal,fiz1);
969 fjx0 = _fjsp_madd_v2r8(dx10,fscal,fjx0);
970 fjy0 = _fjsp_madd_v2r8(dy10,fscal,fjy0);
971 fjz0 = _fjsp_madd_v2r8(dz10,fscal,fjz0);
975 /**************************
976 * CALCULATE INTERACTIONS *
977 **************************/
979 if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
982 r20 = _fjsp_mul_v2r8(rsq20,rinv20);
984 /* Compute parameters for interactions between i and j atoms */
985 qq20 = _fjsp_mul_v2r8(iq2,jq0);
987 /* EWALD ELECTROSTATICS */
989 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
990 ewrt = _fjsp_mul_v2r8(r20,ewtabscale);
991 itab_tmp = _fjsp_dtox_v2r8(ewrt);
992 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
993 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
995 gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
997 felec = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
998 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
1000 cutoff_mask = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
1004 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1006 /* Update vectorial force */
1007 fix2 = _fjsp_madd_v2r8(dx20,fscal,fix2);
1008 fiy2 = _fjsp_madd_v2r8(dy20,fscal,fiy2);
1009 fiz2 = _fjsp_madd_v2r8(dz20,fscal,fiz2);
1011 fjx0 = _fjsp_madd_v2r8(dx20,fscal,fjx0);
1012 fjy0 = _fjsp_madd_v2r8(dy20,fscal,fjy0);
1013 fjz0 = _fjsp_madd_v2r8(dz20,fscal,fjz0);
1017 gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1019 /* Inner loop uses 151 flops */
1022 if(jidx<j_index_end)
1026 j_coord_offsetA = DIM*jnrA;
1028 /* load j atom coordinates */
1029 gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
1032 /* Calculate displacement vector */
1033 dx00 = _fjsp_sub_v2r8(ix0,jx0);
1034 dy00 = _fjsp_sub_v2r8(iy0,jy0);
1035 dz00 = _fjsp_sub_v2r8(iz0,jz0);
1036 dx10 = _fjsp_sub_v2r8(ix1,jx0);
1037 dy10 = _fjsp_sub_v2r8(iy1,jy0);
1038 dz10 = _fjsp_sub_v2r8(iz1,jz0);
1039 dx20 = _fjsp_sub_v2r8(ix2,jx0);
1040 dy20 = _fjsp_sub_v2r8(iy2,jy0);
1041 dz20 = _fjsp_sub_v2r8(iz2,jz0);
1043 /* Calculate squared distance and things based on it */
1044 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
1045 rsq10 = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
1046 rsq20 = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
1048 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
1049 rinv10 = gmx_fjsp_invsqrt_v2r8(rsq10);
1050 rinv20 = gmx_fjsp_invsqrt_v2r8(rsq20);
1052 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
1053 rinvsq10 = _fjsp_mul_v2r8(rinv10,rinv10);
1054 rinvsq20 = _fjsp_mul_v2r8(rinv20,rinv20);
1056 /* Load parameters for j particles */
1057 jq0 = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
1058 vdwjidx0A = 2*vdwtype[jnrA+0];
1060 fjx0 = _fjsp_setzero_v2r8();
1061 fjy0 = _fjsp_setzero_v2r8();
1062 fjz0 = _fjsp_setzero_v2r8();
1064 /**************************
1065 * CALCULATE INTERACTIONS *
1066 **************************/
1068 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
1071 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
1073 /* Compute parameters for interactions between i and j atoms */
1074 qq00 = _fjsp_mul_v2r8(iq0,jq0);
1075 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
1076 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
1078 c6grid_00 = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
1079 vdwgridparam+vdwioffset0+vdwjidx0B);
1081 /* EWALD ELECTROSTATICS */
1083 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1084 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
1085 itab_tmp = _fjsp_dtox_v2r8(ewrt);
1086 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1087 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1089 gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1090 felec = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1091 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
1093 /* Analytical LJ-PME */
1094 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
1095 ewcljrsq = _fjsp_mul_v2r8(ewclj2,rsq00);
1096 ewclj6 = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
1097 exponent = gmx_simd_exp_d(ewcljrsq);
1098 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1099 poly = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
1100 /* f6A = 6 * C6grid * (1 - poly) */
1101 f6A = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
1102 /* f6B = C6grid * exponent * beta^6 */
1103 f6B = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
1104 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1105 fvdw = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1107 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
1109 fscal = _fjsp_add_v2r8(felec,fvdw);
1111 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1113 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1115 /* Update vectorial force */
1116 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
1117 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
1118 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
1120 fjx0 = _fjsp_madd_v2r8(dx00,fscal,fjx0);
1121 fjy0 = _fjsp_madd_v2r8(dy00,fscal,fjy0);
1122 fjz0 = _fjsp_madd_v2r8(dz00,fscal,fjz0);
1126 /**************************
1127 * CALCULATE INTERACTIONS *
1128 **************************/
1130 if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
1133 r10 = _fjsp_mul_v2r8(rsq10,rinv10);
1135 /* Compute parameters for interactions between i and j atoms */
1136 qq10 = _fjsp_mul_v2r8(iq1,jq0);
1138 /* EWALD ELECTROSTATICS */
1140 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1141 ewrt = _fjsp_mul_v2r8(r10,ewtabscale);
1142 itab_tmp = _fjsp_dtox_v2r8(ewrt);
1143 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1144 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1146 gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1147 felec = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1148 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
1150 cutoff_mask = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
1154 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1156 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1158 /* Update vectorial force */
1159 fix1 = _fjsp_madd_v2r8(dx10,fscal,fix1);
1160 fiy1 = _fjsp_madd_v2r8(dy10,fscal,fiy1);
1161 fiz1 = _fjsp_madd_v2r8(dz10,fscal,fiz1);
1163 fjx0 = _fjsp_madd_v2r8(dx10,fscal,fjx0);
1164 fjy0 = _fjsp_madd_v2r8(dy10,fscal,fjy0);
1165 fjz0 = _fjsp_madd_v2r8(dz10,fscal,fjz0);
1169 /**************************
1170 * CALCULATE INTERACTIONS *
1171 **************************/
1173 if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
1176 r20 = _fjsp_mul_v2r8(rsq20,rinv20);
1178 /* Compute parameters for interactions between i and j atoms */
1179 qq20 = _fjsp_mul_v2r8(iq2,jq0);
1181 /* EWALD ELECTROSTATICS */
1183 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1184 ewrt = _fjsp_mul_v2r8(r20,ewtabscale);
1185 itab_tmp = _fjsp_dtox_v2r8(ewrt);
1186 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1187 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1189 gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1190 felec = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1191 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
1193 cutoff_mask = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
1197 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1199 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1201 /* Update vectorial force */
1202 fix2 = _fjsp_madd_v2r8(dx20,fscal,fix2);
1203 fiy2 = _fjsp_madd_v2r8(dy20,fscal,fiy2);
1204 fiz2 = _fjsp_madd_v2r8(dz20,fscal,fiz2);
1206 fjx0 = _fjsp_madd_v2r8(dx20,fscal,fjx0);
1207 fjy0 = _fjsp_madd_v2r8(dy20,fscal,fjy0);
1208 fjz0 = _fjsp_madd_v2r8(dz20,fscal,fjz0);
1212 gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1214 /* Inner loop uses 151 flops */
1217 /* End of innermost loop */
1219 gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1220 f+i_coord_offset,fshift+i_shift_offset);
1222 /* Increment number of inner iterations */
1223 inneriter += j_index_end - j_index_start;
1225 /* Outer loop uses 18 flops */
1228 /* Increment number of outer iterations */
1231 /* Update outer/inner flops */
1233 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*151);