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.
42 #include "../nb_kernel.h"
43 #include "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.h"
47 #include "kernelutil_sparc64_hpc_ace_double.h"
50 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sparc64_hpc_ace_double
51 * Electrostatics interaction: Ewald
52 * VdW interaction: LJEwald
53 * Geometry: Water3-Particle
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sparc64_hpc_ace_double
58 (t_nblist * gmx_restrict nlist,
59 rvec * gmx_restrict xx,
60 rvec * gmx_restrict ff,
61 t_forcerec * gmx_restrict fr,
62 t_mdatoms * gmx_restrict mdatoms,
63 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
64 t_nrnb * gmx_restrict nrnb)
66 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
67 * just 0 for non-waters.
68 * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
69 * jnr indices corresponding to data put in the four positions in the SIMD register.
71 int i_shift_offset,i_coord_offset,outeriter,inneriter;
72 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int j_coord_offsetA,j_coord_offsetB;
75 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
77 real *shiftvec,*fshift,*x,*f;
78 _fjsp_v2r8 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
80 _fjsp_v2r8 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
82 _fjsp_v2r8 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
84 _fjsp_v2r8 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
85 int vdwjidx0A,vdwjidx0B;
86 _fjsp_v2r8 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
87 _fjsp_v2r8 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
88 _fjsp_v2r8 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
89 _fjsp_v2r8 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
90 _fjsp_v2r8 velec,felec,velecsum,facel,crf,krf,krf2;
93 _fjsp_v2r8 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
96 _fjsp_v2r8 one_sixth = gmx_fjsp_set1_v2r8(1.0/6.0);
97 _fjsp_v2r8 one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
100 _fjsp_v2r8 c6grid_20;
102 _fjsp_v2r8 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
103 _fjsp_v2r8 one_half = gmx_fjsp_set1_v2r8(0.5);
104 _fjsp_v2r8 minus_one = gmx_fjsp_set1_v2r8(-1.0);
105 _fjsp_v2r8 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
108 _fjsp_v2r8 dummy_mask,cutoff_mask;
109 _fjsp_v2r8 one = gmx_fjsp_set1_v2r8(1.0);
110 _fjsp_v2r8 two = gmx_fjsp_set1_v2r8(2.0);
111 union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
118 jindex = nlist->jindex;
120 shiftidx = nlist->shift;
122 shiftvec = fr->shift_vec[0];
123 fshift = fr->fshift[0];
124 facel = gmx_fjsp_set1_v2r8(fr->epsfac);
125 charge = mdatoms->chargeA;
126 nvdwtype = fr->ntype;
128 vdwtype = mdatoms->typeA;
129 vdwgridparam = fr->ljpme_c6grid;
130 sh_lj_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
131 ewclj = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
132 ewclj2 = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
134 sh_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
135 ewtab = fr->ic->tabq_coul_FDV0;
136 ewtabscale = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
137 ewtabhalfspace = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
139 /* Setup water-specific parameters */
140 inr = nlist->iinr[0];
141 iq0 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+0]));
142 iq1 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+1]));
143 iq2 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+2]));
144 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
146 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
147 rcutoff_scalar = fr->rcoulomb;
148 rcutoff = gmx_fjsp_set1_v2r8(rcutoff_scalar);
149 rcutoff2 = _fjsp_mul_v2r8(rcutoff,rcutoff);
151 sh_vdw_invrcut6 = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
152 rvdw = gmx_fjsp_set1_v2r8(fr->rvdw);
154 /* Avoid stupid compiler warnings */
162 /* Start outer loop over neighborlists */
163 for(iidx=0; iidx<nri; iidx++)
165 /* Load shift vector for this list */
166 i_shift_offset = DIM*shiftidx[iidx];
168 /* Load limits for loop over neighbors */
169 j_index_start = jindex[iidx];
170 j_index_end = jindex[iidx+1];
172 /* Get outer coordinate index */
174 i_coord_offset = DIM*inr;
176 /* Load i particle coords and add shift vector */
177 gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
178 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
180 fix0 = _fjsp_setzero_v2r8();
181 fiy0 = _fjsp_setzero_v2r8();
182 fiz0 = _fjsp_setzero_v2r8();
183 fix1 = _fjsp_setzero_v2r8();
184 fiy1 = _fjsp_setzero_v2r8();
185 fiz1 = _fjsp_setzero_v2r8();
186 fix2 = _fjsp_setzero_v2r8();
187 fiy2 = _fjsp_setzero_v2r8();
188 fiz2 = _fjsp_setzero_v2r8();
190 /* Reset potential sums */
191 velecsum = _fjsp_setzero_v2r8();
192 vvdwsum = _fjsp_setzero_v2r8();
194 /* Start inner kernel loop */
195 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
198 /* Get j neighbor index, and coordinate index */
201 j_coord_offsetA = DIM*jnrA;
202 j_coord_offsetB = DIM*jnrB;
204 /* load j atom coordinates */
205 gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
208 /* Calculate displacement vector */
209 dx00 = _fjsp_sub_v2r8(ix0,jx0);
210 dy00 = _fjsp_sub_v2r8(iy0,jy0);
211 dz00 = _fjsp_sub_v2r8(iz0,jz0);
212 dx10 = _fjsp_sub_v2r8(ix1,jx0);
213 dy10 = _fjsp_sub_v2r8(iy1,jy0);
214 dz10 = _fjsp_sub_v2r8(iz1,jz0);
215 dx20 = _fjsp_sub_v2r8(ix2,jx0);
216 dy20 = _fjsp_sub_v2r8(iy2,jy0);
217 dz20 = _fjsp_sub_v2r8(iz2,jz0);
219 /* Calculate squared distance and things based on it */
220 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
221 rsq10 = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
222 rsq20 = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
224 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
225 rinv10 = gmx_fjsp_invsqrt_v2r8(rsq10);
226 rinv20 = gmx_fjsp_invsqrt_v2r8(rsq20);
228 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
229 rinvsq10 = _fjsp_mul_v2r8(rinv10,rinv10);
230 rinvsq20 = _fjsp_mul_v2r8(rinv20,rinv20);
232 /* Load parameters for j particles */
233 jq0 = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
234 vdwjidx0A = 2*vdwtype[jnrA+0];
235 vdwjidx0B = 2*vdwtype[jnrB+0];
237 fjx0 = _fjsp_setzero_v2r8();
238 fjy0 = _fjsp_setzero_v2r8();
239 fjz0 = _fjsp_setzero_v2r8();
241 /**************************
242 * CALCULATE INTERACTIONS *
243 **************************/
245 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
248 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
250 /* Compute parameters for interactions between i and j atoms */
251 qq00 = _fjsp_mul_v2r8(iq0,jq0);
252 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
253 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
255 c6grid_00 = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
256 vdwgridparam+vdwioffset0+vdwjidx0B);
258 /* EWALD ELECTROSTATICS */
260 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
261 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
262 itab_tmp = _fjsp_dtox_v2r8(ewrt);
263 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
264 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
266 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
267 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
268 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
269 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
270 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
271 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
272 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
273 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
274 velec = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
275 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
277 /* Analytical LJ-PME */
278 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
279 ewcljrsq = _fjsp_mul_v2r8(ewclj2,rsq00);
280 ewclj6 = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
281 exponent = gmx_simd_exp_d(-ewcljrsq);
282 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
283 poly = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
284 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
285 vvdw6 = _fjsp_mul_v2r8(_fjsp_madd_v2r8(-c6grid_00,_fjsp_sub_v2r8(one,poly),c6_00),rinvsix);
286 vvdw12 = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
287 vvdw = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
288 _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));
289 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
290 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);
292 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
294 /* Update potential sum for this i atom from the interaction with this j atom. */
295 velec = _fjsp_and_v2r8(velec,cutoff_mask);
296 velecsum = _fjsp_add_v2r8(velecsum,velec);
297 vvdw = _fjsp_and_v2r8(vvdw,cutoff_mask);
298 vvdwsum = _fjsp_add_v2r8(vvdwsum,vvdw);
300 fscal = _fjsp_add_v2r8(felec,fvdw);
302 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
304 /* Update vectorial force */
305 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
306 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
307 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
309 fjx0 = _fjsp_madd_v2r8(dx00,fscal,fjx0);
310 fjy0 = _fjsp_madd_v2r8(dy00,fscal,fjy0);
311 fjz0 = _fjsp_madd_v2r8(dz00,fscal,fjz0);
315 /**************************
316 * CALCULATE INTERACTIONS *
317 **************************/
319 if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
322 r10 = _fjsp_mul_v2r8(rsq10,rinv10);
324 /* Compute parameters for interactions between i and j atoms */
325 qq10 = _fjsp_mul_v2r8(iq1,jq0);
327 /* EWALD ELECTROSTATICS */
329 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
330 ewrt = _fjsp_mul_v2r8(r10,ewtabscale);
331 itab_tmp = _fjsp_dtox_v2r8(ewrt);
332 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
333 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
335 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
336 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
337 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
338 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
339 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
340 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
341 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
342 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
343 velec = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv10,sh_ewald),velec));
344 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
346 cutoff_mask = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
348 /* Update potential sum for this i atom from the interaction with this j atom. */
349 velec = _fjsp_and_v2r8(velec,cutoff_mask);
350 velecsum = _fjsp_add_v2r8(velecsum,velec);
354 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
356 /* Update vectorial force */
357 fix1 = _fjsp_madd_v2r8(dx10,fscal,fix1);
358 fiy1 = _fjsp_madd_v2r8(dy10,fscal,fiy1);
359 fiz1 = _fjsp_madd_v2r8(dz10,fscal,fiz1);
361 fjx0 = _fjsp_madd_v2r8(dx10,fscal,fjx0);
362 fjy0 = _fjsp_madd_v2r8(dy10,fscal,fjy0);
363 fjz0 = _fjsp_madd_v2r8(dz10,fscal,fjz0);
367 /**************************
368 * CALCULATE INTERACTIONS *
369 **************************/
371 if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
374 r20 = _fjsp_mul_v2r8(rsq20,rinv20);
376 /* Compute parameters for interactions between i and j atoms */
377 qq20 = _fjsp_mul_v2r8(iq2,jq0);
379 /* EWALD ELECTROSTATICS */
381 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
382 ewrt = _fjsp_mul_v2r8(r20,ewtabscale);
383 itab_tmp = _fjsp_dtox_v2r8(ewrt);
384 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
385 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
387 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
388 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
389 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
390 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
391 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
392 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
393 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
394 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
395 velec = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv20,sh_ewald),velec));
396 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
398 cutoff_mask = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
400 /* Update potential sum for this i atom from the interaction with this j atom. */
401 velec = _fjsp_and_v2r8(velec,cutoff_mask);
402 velecsum = _fjsp_add_v2r8(velecsum,velec);
406 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
408 /* Update vectorial force */
409 fix2 = _fjsp_madd_v2r8(dx20,fscal,fix2);
410 fiy2 = _fjsp_madd_v2r8(dy20,fscal,fiy2);
411 fiz2 = _fjsp_madd_v2r8(dz20,fscal,fiz2);
413 fjx0 = _fjsp_madd_v2r8(dx20,fscal,fjx0);
414 fjy0 = _fjsp_madd_v2r8(dy20,fscal,fjy0);
415 fjz0 = _fjsp_madd_v2r8(dz20,fscal,fjz0);
419 gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
421 /* Inner loop uses 180 flops */
428 j_coord_offsetA = DIM*jnrA;
430 /* load j atom coordinates */
431 gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
434 /* Calculate displacement vector */
435 dx00 = _fjsp_sub_v2r8(ix0,jx0);
436 dy00 = _fjsp_sub_v2r8(iy0,jy0);
437 dz00 = _fjsp_sub_v2r8(iz0,jz0);
438 dx10 = _fjsp_sub_v2r8(ix1,jx0);
439 dy10 = _fjsp_sub_v2r8(iy1,jy0);
440 dz10 = _fjsp_sub_v2r8(iz1,jz0);
441 dx20 = _fjsp_sub_v2r8(ix2,jx0);
442 dy20 = _fjsp_sub_v2r8(iy2,jy0);
443 dz20 = _fjsp_sub_v2r8(iz2,jz0);
445 /* Calculate squared distance and things based on it */
446 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
447 rsq10 = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
448 rsq20 = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
450 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
451 rinv10 = gmx_fjsp_invsqrt_v2r8(rsq10);
452 rinv20 = gmx_fjsp_invsqrt_v2r8(rsq20);
454 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
455 rinvsq10 = _fjsp_mul_v2r8(rinv10,rinv10);
456 rinvsq20 = _fjsp_mul_v2r8(rinv20,rinv20);
458 /* Load parameters for j particles */
459 jq0 = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
460 vdwjidx0A = 2*vdwtype[jnrA+0];
462 fjx0 = _fjsp_setzero_v2r8();
463 fjy0 = _fjsp_setzero_v2r8();
464 fjz0 = _fjsp_setzero_v2r8();
466 /**************************
467 * CALCULATE INTERACTIONS *
468 **************************/
470 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
473 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
475 /* Compute parameters for interactions between i and j atoms */
476 qq00 = _fjsp_mul_v2r8(iq0,jq0);
477 gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
479 c6grid_00 = gmx_fjsp_load_1real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A);
481 /* EWALD ELECTROSTATICS */
483 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
484 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
485 itab_tmp = _fjsp_dtox_v2r8(ewrt);
486 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
487 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
489 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
490 ewtabD = _fjsp_setzero_v2r8();
491 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
492 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
493 ewtabFn = _fjsp_setzero_v2r8();
494 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
495 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
496 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
497 velec = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
498 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
500 /* Analytical LJ-PME */
501 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
502 ewcljrsq = _fjsp_mul_v2r8(ewclj2,rsq00);
503 ewclj6 = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
504 exponent = gmx_simd_exp_d(-ewcljrsq);
505 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
506 poly = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
507 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
508 vvdw6 = _fjsp_mul_v2r8(_fjsp_madd_v2r8(-c6grid_00,_fjsp_sub_v2r8(one,poly),c6_00),rinvsix);
509 vvdw12 = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
510 vvdw = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
511 _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));
512 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
513 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);
515 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
517 /* Update potential sum for this i atom from the interaction with this j atom. */
518 velec = _fjsp_and_v2r8(velec,cutoff_mask);
519 velec = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
520 velecsum = _fjsp_add_v2r8(velecsum,velec);
521 vvdw = _fjsp_and_v2r8(vvdw,cutoff_mask);
522 vvdw = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
523 vvdwsum = _fjsp_add_v2r8(vvdwsum,vvdw);
525 fscal = _fjsp_add_v2r8(felec,fvdw);
527 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
529 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
531 /* Update vectorial force */
532 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
533 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
534 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
536 fjx0 = _fjsp_madd_v2r8(dx00,fscal,fjx0);
537 fjy0 = _fjsp_madd_v2r8(dy00,fscal,fjy0);
538 fjz0 = _fjsp_madd_v2r8(dz00,fscal,fjz0);
542 /**************************
543 * CALCULATE INTERACTIONS *
544 **************************/
546 if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
549 r10 = _fjsp_mul_v2r8(rsq10,rinv10);
551 /* Compute parameters for interactions between i and j atoms */
552 qq10 = _fjsp_mul_v2r8(iq1,jq0);
554 /* EWALD ELECTROSTATICS */
556 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
557 ewrt = _fjsp_mul_v2r8(r10,ewtabscale);
558 itab_tmp = _fjsp_dtox_v2r8(ewrt);
559 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
560 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
562 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
563 ewtabD = _fjsp_setzero_v2r8();
564 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
565 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
566 ewtabFn = _fjsp_setzero_v2r8();
567 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
568 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
569 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
570 velec = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv10,sh_ewald),velec));
571 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
573 cutoff_mask = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
575 /* Update potential sum for this i atom from the interaction with this j atom. */
576 velec = _fjsp_and_v2r8(velec,cutoff_mask);
577 velec = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
578 velecsum = _fjsp_add_v2r8(velecsum,velec);
582 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
584 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
586 /* Update vectorial force */
587 fix1 = _fjsp_madd_v2r8(dx10,fscal,fix1);
588 fiy1 = _fjsp_madd_v2r8(dy10,fscal,fiy1);
589 fiz1 = _fjsp_madd_v2r8(dz10,fscal,fiz1);
591 fjx0 = _fjsp_madd_v2r8(dx10,fscal,fjx0);
592 fjy0 = _fjsp_madd_v2r8(dy10,fscal,fjy0);
593 fjz0 = _fjsp_madd_v2r8(dz10,fscal,fjz0);
597 /**************************
598 * CALCULATE INTERACTIONS *
599 **************************/
601 if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
604 r20 = _fjsp_mul_v2r8(rsq20,rinv20);
606 /* Compute parameters for interactions between i and j atoms */
607 qq20 = _fjsp_mul_v2r8(iq2,jq0);
609 /* EWALD ELECTROSTATICS */
611 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
612 ewrt = _fjsp_mul_v2r8(r20,ewtabscale);
613 itab_tmp = _fjsp_dtox_v2r8(ewrt);
614 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
615 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
617 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
618 ewtabD = _fjsp_setzero_v2r8();
619 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
620 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
621 ewtabFn = _fjsp_setzero_v2r8();
622 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
623 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
624 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
625 velec = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv20,sh_ewald),velec));
626 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
628 cutoff_mask = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
630 /* Update potential sum for this i atom from the interaction with this j atom. */
631 velec = _fjsp_and_v2r8(velec,cutoff_mask);
632 velec = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
633 velecsum = _fjsp_add_v2r8(velecsum,velec);
637 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
639 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
641 /* Update vectorial force */
642 fix2 = _fjsp_madd_v2r8(dx20,fscal,fix2);
643 fiy2 = _fjsp_madd_v2r8(dy20,fscal,fiy2);
644 fiz2 = _fjsp_madd_v2r8(dz20,fscal,fiz2);
646 fjx0 = _fjsp_madd_v2r8(dx20,fscal,fjx0);
647 fjy0 = _fjsp_madd_v2r8(dy20,fscal,fjy0);
648 fjz0 = _fjsp_madd_v2r8(dz20,fscal,fjz0);
652 gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
654 /* Inner loop uses 180 flops */
657 /* End of innermost loop */
659 gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
660 f+i_coord_offset,fshift+i_shift_offset);
663 /* Update potential energies */
664 gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
665 gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
667 /* Increment number of inner iterations */
668 inneriter += j_index_end - j_index_start;
670 /* Outer loop uses 20 flops */
673 /* Increment number of outer iterations */
676 /* Update outer/inner flops */
678 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*180);
681 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sparc64_hpc_ace_double
682 * Electrostatics interaction: Ewald
683 * VdW interaction: LJEwald
684 * Geometry: Water3-Particle
685 * Calculate force/pot: Force
688 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sparc64_hpc_ace_double
689 (t_nblist * gmx_restrict nlist,
690 rvec * gmx_restrict xx,
691 rvec * gmx_restrict ff,
692 t_forcerec * gmx_restrict fr,
693 t_mdatoms * gmx_restrict mdatoms,
694 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
695 t_nrnb * gmx_restrict nrnb)
697 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
698 * just 0 for non-waters.
699 * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
700 * jnr indices corresponding to data put in the four positions in the SIMD register.
702 int i_shift_offset,i_coord_offset,outeriter,inneriter;
703 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
705 int j_coord_offsetA,j_coord_offsetB;
706 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
708 real *shiftvec,*fshift,*x,*f;
709 _fjsp_v2r8 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
711 _fjsp_v2r8 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
713 _fjsp_v2r8 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
715 _fjsp_v2r8 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
716 int vdwjidx0A,vdwjidx0B;
717 _fjsp_v2r8 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
718 _fjsp_v2r8 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
719 _fjsp_v2r8 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
720 _fjsp_v2r8 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
721 _fjsp_v2r8 velec,felec,velecsum,facel,crf,krf,krf2;
724 _fjsp_v2r8 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
727 _fjsp_v2r8 one_sixth = gmx_fjsp_set1_v2r8(1.0/6.0);
728 _fjsp_v2r8 one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
729 _fjsp_v2r8 c6grid_00;
730 _fjsp_v2r8 c6grid_10;
731 _fjsp_v2r8 c6grid_20;
733 _fjsp_v2r8 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
734 _fjsp_v2r8 one_half = gmx_fjsp_set1_v2r8(0.5);
735 _fjsp_v2r8 minus_one = gmx_fjsp_set1_v2r8(-1.0);
736 _fjsp_v2r8 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
739 _fjsp_v2r8 dummy_mask,cutoff_mask;
740 _fjsp_v2r8 one = gmx_fjsp_set1_v2r8(1.0);
741 _fjsp_v2r8 two = gmx_fjsp_set1_v2r8(2.0);
742 union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
749 jindex = nlist->jindex;
751 shiftidx = nlist->shift;
753 shiftvec = fr->shift_vec[0];
754 fshift = fr->fshift[0];
755 facel = gmx_fjsp_set1_v2r8(fr->epsfac);
756 charge = mdatoms->chargeA;
757 nvdwtype = fr->ntype;
759 vdwtype = mdatoms->typeA;
760 vdwgridparam = fr->ljpme_c6grid;
761 sh_lj_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
762 ewclj = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
763 ewclj2 = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
765 sh_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
766 ewtab = fr->ic->tabq_coul_F;
767 ewtabscale = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
768 ewtabhalfspace = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
770 /* Setup water-specific parameters */
771 inr = nlist->iinr[0];
772 iq0 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+0]));
773 iq1 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+1]));
774 iq2 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+2]));
775 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
777 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
778 rcutoff_scalar = fr->rcoulomb;
779 rcutoff = gmx_fjsp_set1_v2r8(rcutoff_scalar);
780 rcutoff2 = _fjsp_mul_v2r8(rcutoff,rcutoff);
782 sh_vdw_invrcut6 = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
783 rvdw = gmx_fjsp_set1_v2r8(fr->rvdw);
785 /* Avoid stupid compiler warnings */
793 /* Start outer loop over neighborlists */
794 for(iidx=0; iidx<nri; iidx++)
796 /* Load shift vector for this list */
797 i_shift_offset = DIM*shiftidx[iidx];
799 /* Load limits for loop over neighbors */
800 j_index_start = jindex[iidx];
801 j_index_end = jindex[iidx+1];
803 /* Get outer coordinate index */
805 i_coord_offset = DIM*inr;
807 /* Load i particle coords and add shift vector */
808 gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
809 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
811 fix0 = _fjsp_setzero_v2r8();
812 fiy0 = _fjsp_setzero_v2r8();
813 fiz0 = _fjsp_setzero_v2r8();
814 fix1 = _fjsp_setzero_v2r8();
815 fiy1 = _fjsp_setzero_v2r8();
816 fiz1 = _fjsp_setzero_v2r8();
817 fix2 = _fjsp_setzero_v2r8();
818 fiy2 = _fjsp_setzero_v2r8();
819 fiz2 = _fjsp_setzero_v2r8();
821 /* Start inner kernel loop */
822 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
825 /* Get j neighbor index, and coordinate index */
828 j_coord_offsetA = DIM*jnrA;
829 j_coord_offsetB = DIM*jnrB;
831 /* load j atom coordinates */
832 gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
835 /* Calculate displacement vector */
836 dx00 = _fjsp_sub_v2r8(ix0,jx0);
837 dy00 = _fjsp_sub_v2r8(iy0,jy0);
838 dz00 = _fjsp_sub_v2r8(iz0,jz0);
839 dx10 = _fjsp_sub_v2r8(ix1,jx0);
840 dy10 = _fjsp_sub_v2r8(iy1,jy0);
841 dz10 = _fjsp_sub_v2r8(iz1,jz0);
842 dx20 = _fjsp_sub_v2r8(ix2,jx0);
843 dy20 = _fjsp_sub_v2r8(iy2,jy0);
844 dz20 = _fjsp_sub_v2r8(iz2,jz0);
846 /* Calculate squared distance and things based on it */
847 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
848 rsq10 = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
849 rsq20 = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
851 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
852 rinv10 = gmx_fjsp_invsqrt_v2r8(rsq10);
853 rinv20 = gmx_fjsp_invsqrt_v2r8(rsq20);
855 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
856 rinvsq10 = _fjsp_mul_v2r8(rinv10,rinv10);
857 rinvsq20 = _fjsp_mul_v2r8(rinv20,rinv20);
859 /* Load parameters for j particles */
860 jq0 = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
861 vdwjidx0A = 2*vdwtype[jnrA+0];
862 vdwjidx0B = 2*vdwtype[jnrB+0];
864 fjx0 = _fjsp_setzero_v2r8();
865 fjy0 = _fjsp_setzero_v2r8();
866 fjz0 = _fjsp_setzero_v2r8();
868 /**************************
869 * CALCULATE INTERACTIONS *
870 **************************/
872 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
875 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
877 /* Compute parameters for interactions between i and j atoms */
878 qq00 = _fjsp_mul_v2r8(iq0,jq0);
879 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
880 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
882 c6grid_00 = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
883 vdwgridparam+vdwioffset0+vdwjidx0B);
885 /* EWALD ELECTROSTATICS */
887 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
888 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
889 itab_tmp = _fjsp_dtox_v2r8(ewrt);
890 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
891 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
893 gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
895 felec = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
896 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
898 /* Analytical LJ-PME */
899 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
900 ewcljrsq = _fjsp_mul_v2r8(ewclj2,rsq00);
901 ewclj6 = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
902 exponent = gmx_simd_exp_d(-ewcljrsq);
903 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
904 poly = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
905 /* f6A = 6 * C6grid * (1 - poly) */
906 f6A = _fjsp_mul_v2r8(c6grid_00,_fjsp_msub_v2r8(one,poly));
907 /* f6B = C6grid * exponent * beta^6 */
908 f6B = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
909 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
910 fvdw = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
912 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
914 fscal = _fjsp_add_v2r8(felec,fvdw);
916 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
918 /* Update vectorial force */
919 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
920 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
921 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
923 fjx0 = _fjsp_madd_v2r8(dx00,fscal,fjx0);
924 fjy0 = _fjsp_madd_v2r8(dy00,fscal,fjy0);
925 fjz0 = _fjsp_madd_v2r8(dz00,fscal,fjz0);
929 /**************************
930 * CALCULATE INTERACTIONS *
931 **************************/
933 if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
936 r10 = _fjsp_mul_v2r8(rsq10,rinv10);
938 /* Compute parameters for interactions between i and j atoms */
939 qq10 = _fjsp_mul_v2r8(iq1,jq0);
941 /* EWALD ELECTROSTATICS */
943 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
944 ewrt = _fjsp_mul_v2r8(r10,ewtabscale);
945 itab_tmp = _fjsp_dtox_v2r8(ewrt);
946 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
947 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
949 gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
951 felec = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
952 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
954 cutoff_mask = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
958 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
960 /* Update vectorial force */
961 fix1 = _fjsp_madd_v2r8(dx10,fscal,fix1);
962 fiy1 = _fjsp_madd_v2r8(dy10,fscal,fiy1);
963 fiz1 = _fjsp_madd_v2r8(dz10,fscal,fiz1);
965 fjx0 = _fjsp_madd_v2r8(dx10,fscal,fjx0);
966 fjy0 = _fjsp_madd_v2r8(dy10,fscal,fjy0);
967 fjz0 = _fjsp_madd_v2r8(dz10,fscal,fjz0);
971 /**************************
972 * CALCULATE INTERACTIONS *
973 **************************/
975 if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
978 r20 = _fjsp_mul_v2r8(rsq20,rinv20);
980 /* Compute parameters for interactions between i and j atoms */
981 qq20 = _fjsp_mul_v2r8(iq2,jq0);
983 /* EWALD ELECTROSTATICS */
985 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
986 ewrt = _fjsp_mul_v2r8(r20,ewtabscale);
987 itab_tmp = _fjsp_dtox_v2r8(ewrt);
988 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
989 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
991 gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
993 felec = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
994 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
996 cutoff_mask = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
1000 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1002 /* Update vectorial force */
1003 fix2 = _fjsp_madd_v2r8(dx20,fscal,fix2);
1004 fiy2 = _fjsp_madd_v2r8(dy20,fscal,fiy2);
1005 fiz2 = _fjsp_madd_v2r8(dz20,fscal,fiz2);
1007 fjx0 = _fjsp_madd_v2r8(dx20,fscal,fjx0);
1008 fjy0 = _fjsp_madd_v2r8(dy20,fscal,fjy0);
1009 fjz0 = _fjsp_madd_v2r8(dz20,fscal,fjz0);
1013 gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1015 /* Inner loop uses 151 flops */
1018 if(jidx<j_index_end)
1022 j_coord_offsetA = DIM*jnrA;
1024 /* load j atom coordinates */
1025 gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
1028 /* Calculate displacement vector */
1029 dx00 = _fjsp_sub_v2r8(ix0,jx0);
1030 dy00 = _fjsp_sub_v2r8(iy0,jy0);
1031 dz00 = _fjsp_sub_v2r8(iz0,jz0);
1032 dx10 = _fjsp_sub_v2r8(ix1,jx0);
1033 dy10 = _fjsp_sub_v2r8(iy1,jy0);
1034 dz10 = _fjsp_sub_v2r8(iz1,jz0);
1035 dx20 = _fjsp_sub_v2r8(ix2,jx0);
1036 dy20 = _fjsp_sub_v2r8(iy2,jy0);
1037 dz20 = _fjsp_sub_v2r8(iz2,jz0);
1039 /* Calculate squared distance and things based on it */
1040 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
1041 rsq10 = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
1042 rsq20 = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
1044 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
1045 rinv10 = gmx_fjsp_invsqrt_v2r8(rsq10);
1046 rinv20 = gmx_fjsp_invsqrt_v2r8(rsq20);
1048 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
1049 rinvsq10 = _fjsp_mul_v2r8(rinv10,rinv10);
1050 rinvsq20 = _fjsp_mul_v2r8(rinv20,rinv20);
1052 /* Load parameters for j particles */
1053 jq0 = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
1054 vdwjidx0A = 2*vdwtype[jnrA+0];
1056 fjx0 = _fjsp_setzero_v2r8();
1057 fjy0 = _fjsp_setzero_v2r8();
1058 fjz0 = _fjsp_setzero_v2r8();
1060 /**************************
1061 * CALCULATE INTERACTIONS *
1062 **************************/
1064 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
1067 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
1069 /* Compute parameters for interactions between i and j atoms */
1070 qq00 = _fjsp_mul_v2r8(iq0,jq0);
1071 gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1073 c6grid_00 = gmx_fjsp_load_1real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A);
1075 /* EWALD ELECTROSTATICS */
1077 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1078 ewrt = _fjsp_mul_v2r8(r00,ewtabscale);
1079 itab_tmp = _fjsp_dtox_v2r8(ewrt);
1080 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1081 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1083 gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1084 felec = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1085 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
1087 /* Analytical LJ-PME */
1088 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
1089 ewcljrsq = _fjsp_mul_v2r8(ewclj2,rsq00);
1090 ewclj6 = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
1091 exponent = gmx_simd_exp_d(-ewcljrsq);
1092 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1093 poly = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
1094 /* f6A = 6 * C6grid * (1 - poly) */
1095 f6A = _fjsp_mul_v2r8(c6grid_00,_fjsp_msub_v2r8(one,poly));
1096 /* f6B = C6grid * exponent * beta^6 */
1097 f6B = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
1098 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1099 fvdw = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1101 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
1103 fscal = _fjsp_add_v2r8(felec,fvdw);
1105 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1107 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1109 /* Update vectorial force */
1110 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
1111 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
1112 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
1114 fjx0 = _fjsp_madd_v2r8(dx00,fscal,fjx0);
1115 fjy0 = _fjsp_madd_v2r8(dy00,fscal,fjy0);
1116 fjz0 = _fjsp_madd_v2r8(dz00,fscal,fjz0);
1120 /**************************
1121 * CALCULATE INTERACTIONS *
1122 **************************/
1124 if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
1127 r10 = _fjsp_mul_v2r8(rsq10,rinv10);
1129 /* Compute parameters for interactions between i and j atoms */
1130 qq10 = _fjsp_mul_v2r8(iq1,jq0);
1132 /* EWALD ELECTROSTATICS */
1134 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1135 ewrt = _fjsp_mul_v2r8(r10,ewtabscale);
1136 itab_tmp = _fjsp_dtox_v2r8(ewrt);
1137 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1138 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1140 gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1141 felec = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1142 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
1144 cutoff_mask = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
1148 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1150 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1152 /* Update vectorial force */
1153 fix1 = _fjsp_madd_v2r8(dx10,fscal,fix1);
1154 fiy1 = _fjsp_madd_v2r8(dy10,fscal,fiy1);
1155 fiz1 = _fjsp_madd_v2r8(dz10,fscal,fiz1);
1157 fjx0 = _fjsp_madd_v2r8(dx10,fscal,fjx0);
1158 fjy0 = _fjsp_madd_v2r8(dy10,fscal,fjy0);
1159 fjz0 = _fjsp_madd_v2r8(dz10,fscal,fjz0);
1163 /**************************
1164 * CALCULATE INTERACTIONS *
1165 **************************/
1167 if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
1170 r20 = _fjsp_mul_v2r8(rsq20,rinv20);
1172 /* Compute parameters for interactions between i and j atoms */
1173 qq20 = _fjsp_mul_v2r8(iq2,jq0);
1175 /* EWALD ELECTROSTATICS */
1177 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1178 ewrt = _fjsp_mul_v2r8(r20,ewtabscale);
1179 itab_tmp = _fjsp_dtox_v2r8(ewrt);
1180 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1181 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1183 gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1184 felec = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1185 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
1187 cutoff_mask = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
1191 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1193 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1195 /* Update vectorial force */
1196 fix2 = _fjsp_madd_v2r8(dx20,fscal,fix2);
1197 fiy2 = _fjsp_madd_v2r8(dy20,fscal,fiy2);
1198 fiz2 = _fjsp_madd_v2r8(dz20,fscal,fiz2);
1200 fjx0 = _fjsp_madd_v2r8(dx20,fscal,fjx0);
1201 fjy0 = _fjsp_madd_v2r8(dy20,fscal,fjy0);
1202 fjz0 = _fjsp_madd_v2r8(dz20,fscal,fjz0);
1206 gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1208 /* Inner loop uses 151 flops */
1211 /* End of innermost loop */
1213 gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1214 f+i_coord_offset,fshift+i_shift_offset);
1216 /* Increment number of inner iterations */
1217 inneriter += j_index_end - j_index_start;
1219 /* Outer loop uses 18 flops */
1222 /* Increment number of outer iterations */
1225 /* Update outer/inner flops */
1227 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*151);