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 "types/simple.h"
46 #include "gromacs/math/vec.h"
49 #include "kernelutil_sparc64_hpc_ace_double.h"
52 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_VF_sparc64_hpc_ace_double
53 * Electrostatics interaction: Ewald
54 * VdW interaction: LennardJones
55 * Geometry: Water4-Particle
56 * Calculate force/pot: PotentialAndForce
59 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_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;
88 _fjsp_v2r8 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
89 int vdwjidx0A,vdwjidx0B;
90 _fjsp_v2r8 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
91 _fjsp_v2r8 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
92 _fjsp_v2r8 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
93 _fjsp_v2r8 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
94 _fjsp_v2r8 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
95 _fjsp_v2r8 velec,felec,velecsum,facel,crf,krf,krf2;
98 _fjsp_v2r8 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
101 _fjsp_v2r8 one_sixth = gmx_fjsp_set1_v2r8(1.0/6.0);
102 _fjsp_v2r8 one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
103 _fjsp_v2r8 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
105 _fjsp_v2r8 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
106 real rswitch_scalar,d_scalar;
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;
130 sh_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
131 ewtab = fr->ic->tabq_coul_FDV0;
132 ewtabscale = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
133 ewtabhalfspace = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
135 /* Setup water-specific parameters */
136 inr = nlist->iinr[0];
137 iq1 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+1]));
138 iq2 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+2]));
139 iq3 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+3]));
140 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
142 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
143 rcutoff_scalar = fr->rcoulomb;
144 rcutoff = gmx_fjsp_set1_v2r8(rcutoff_scalar);
145 rcutoff2 = _fjsp_mul_v2r8(rcutoff,rcutoff);
147 rswitch_scalar = fr->rcoulomb_switch;
148 rswitch = gmx_fjsp_set1_v2r8(rswitch_scalar);
149 /* Setup switch parameters */
150 d_scalar = rcutoff_scalar-rswitch_scalar;
151 d = gmx_fjsp_set1_v2r8(d_scalar);
152 swV3 = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
153 swV4 = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
154 swV5 = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
155 swF2 = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
156 swF3 = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
157 swF4 = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
159 /* Avoid stupid compiler warnings */
167 /* Start outer loop over neighborlists */
168 for(iidx=0; iidx<nri; iidx++)
170 /* Load shift vector for this list */
171 i_shift_offset = DIM*shiftidx[iidx];
173 /* Load limits for loop over neighbors */
174 j_index_start = jindex[iidx];
175 j_index_end = jindex[iidx+1];
177 /* Get outer coordinate index */
179 i_coord_offset = DIM*inr;
181 /* Load i particle coords and add shift vector */
182 gmx_fjsp_load_shift_and_4rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
183 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
185 fix0 = _fjsp_setzero_v2r8();
186 fiy0 = _fjsp_setzero_v2r8();
187 fiz0 = _fjsp_setzero_v2r8();
188 fix1 = _fjsp_setzero_v2r8();
189 fiy1 = _fjsp_setzero_v2r8();
190 fiz1 = _fjsp_setzero_v2r8();
191 fix2 = _fjsp_setzero_v2r8();
192 fiy2 = _fjsp_setzero_v2r8();
193 fiz2 = _fjsp_setzero_v2r8();
194 fix3 = _fjsp_setzero_v2r8();
195 fiy3 = _fjsp_setzero_v2r8();
196 fiz3 = _fjsp_setzero_v2r8();
198 /* Reset potential sums */
199 velecsum = _fjsp_setzero_v2r8();
200 vvdwsum = _fjsp_setzero_v2r8();
202 /* Start inner kernel loop */
203 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
206 /* Get j neighbor index, and coordinate index */
209 j_coord_offsetA = DIM*jnrA;
210 j_coord_offsetB = DIM*jnrB;
212 /* load j atom coordinates */
213 gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
216 /* Calculate displacement vector */
217 dx00 = _fjsp_sub_v2r8(ix0,jx0);
218 dy00 = _fjsp_sub_v2r8(iy0,jy0);
219 dz00 = _fjsp_sub_v2r8(iz0,jz0);
220 dx10 = _fjsp_sub_v2r8(ix1,jx0);
221 dy10 = _fjsp_sub_v2r8(iy1,jy0);
222 dz10 = _fjsp_sub_v2r8(iz1,jz0);
223 dx20 = _fjsp_sub_v2r8(ix2,jx0);
224 dy20 = _fjsp_sub_v2r8(iy2,jy0);
225 dz20 = _fjsp_sub_v2r8(iz2,jz0);
226 dx30 = _fjsp_sub_v2r8(ix3,jx0);
227 dy30 = _fjsp_sub_v2r8(iy3,jy0);
228 dz30 = _fjsp_sub_v2r8(iz3,jz0);
230 /* Calculate squared distance and things based on it */
231 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
232 rsq10 = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
233 rsq20 = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
234 rsq30 = gmx_fjsp_calc_rsq_v2r8(dx30,dy30,dz30);
236 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
237 rinv10 = gmx_fjsp_invsqrt_v2r8(rsq10);
238 rinv20 = gmx_fjsp_invsqrt_v2r8(rsq20);
239 rinv30 = gmx_fjsp_invsqrt_v2r8(rsq30);
241 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
242 rinvsq10 = _fjsp_mul_v2r8(rinv10,rinv10);
243 rinvsq20 = _fjsp_mul_v2r8(rinv20,rinv20);
244 rinvsq30 = _fjsp_mul_v2r8(rinv30,rinv30);
246 /* Load parameters for j particles */
247 jq0 = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
248 vdwjidx0A = 2*vdwtype[jnrA+0];
249 vdwjidx0B = 2*vdwtype[jnrB+0];
251 fjx0 = _fjsp_setzero_v2r8();
252 fjy0 = _fjsp_setzero_v2r8();
253 fjz0 = _fjsp_setzero_v2r8();
255 /**************************
256 * CALCULATE INTERACTIONS *
257 **************************/
259 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
262 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
264 /* Compute parameters for interactions between i and j atoms */
265 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
266 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
268 /* LENNARD-JONES DISPERSION/REPULSION */
270 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
271 vvdw6 = _fjsp_mul_v2r8(c6_00,rinvsix);
272 vvdw12 = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
273 vvdw = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
274 fvdw = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
276 d = _fjsp_sub_v2r8(r00,rswitch);
277 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
278 d2 = _fjsp_mul_v2r8(d,d);
279 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
281 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
283 /* Evaluate switch function */
284 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
285 fvdw = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
286 vvdw = _fjsp_mul_v2r8(vvdw,sw);
287 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
289 /* Update potential sum for this i atom from the interaction with this j atom. */
290 vvdw = _fjsp_and_v2r8(vvdw,cutoff_mask);
291 vvdwsum = _fjsp_add_v2r8(vvdwsum,vvdw);
295 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
297 /* Update vectorial force */
298 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
299 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
300 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
302 fjx0 = _fjsp_madd_v2r8(dx00,fscal,fjx0);
303 fjy0 = _fjsp_madd_v2r8(dy00,fscal,fjy0);
304 fjz0 = _fjsp_madd_v2r8(dz00,fscal,fjz0);
308 /**************************
309 * CALCULATE INTERACTIONS *
310 **************************/
312 if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
315 r10 = _fjsp_mul_v2r8(rsq10,rinv10);
317 /* Compute parameters for interactions between i and j atoms */
318 qq10 = _fjsp_mul_v2r8(iq1,jq0);
320 /* EWALD ELECTROSTATICS */
322 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
323 ewrt = _fjsp_mul_v2r8(r10,ewtabscale);
324 itab_tmp = _fjsp_dtox_v2r8(ewrt);
325 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
326 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
328 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
329 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
330 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
331 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
332 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
333 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
334 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
335 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
336 velec = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(rinv10,velec));
337 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
339 d = _fjsp_sub_v2r8(r10,rswitch);
340 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
341 d2 = _fjsp_mul_v2r8(d,d);
342 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
344 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
346 /* Evaluate switch function */
347 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
348 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv10,_fjsp_mul_v2r8(velec,dsw)) );
349 velec = _fjsp_mul_v2r8(velec,sw);
350 cutoff_mask = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
352 /* Update potential sum for this i atom from the interaction with this j atom. */
353 velec = _fjsp_and_v2r8(velec,cutoff_mask);
354 velecsum = _fjsp_add_v2r8(velecsum,velec);
358 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
360 /* Update vectorial force */
361 fix1 = _fjsp_madd_v2r8(dx10,fscal,fix1);
362 fiy1 = _fjsp_madd_v2r8(dy10,fscal,fiy1);
363 fiz1 = _fjsp_madd_v2r8(dz10,fscal,fiz1);
365 fjx0 = _fjsp_madd_v2r8(dx10,fscal,fjx0);
366 fjy0 = _fjsp_madd_v2r8(dy10,fscal,fjy0);
367 fjz0 = _fjsp_madd_v2r8(dz10,fscal,fjz0);
371 /**************************
372 * CALCULATE INTERACTIONS *
373 **************************/
375 if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
378 r20 = _fjsp_mul_v2r8(rsq20,rinv20);
380 /* Compute parameters for interactions between i and j atoms */
381 qq20 = _fjsp_mul_v2r8(iq2,jq0);
383 /* EWALD ELECTROSTATICS */
385 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
386 ewrt = _fjsp_mul_v2r8(r20,ewtabscale);
387 itab_tmp = _fjsp_dtox_v2r8(ewrt);
388 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
389 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
391 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
392 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
393 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
394 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
395 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
396 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
397 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
398 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
399 velec = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(rinv20,velec));
400 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
402 d = _fjsp_sub_v2r8(r20,rswitch);
403 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
404 d2 = _fjsp_mul_v2r8(d,d);
405 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
407 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
409 /* Evaluate switch function */
410 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
411 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv20,_fjsp_mul_v2r8(velec,dsw)) );
412 velec = _fjsp_mul_v2r8(velec,sw);
413 cutoff_mask = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
415 /* Update potential sum for this i atom from the interaction with this j atom. */
416 velec = _fjsp_and_v2r8(velec,cutoff_mask);
417 velecsum = _fjsp_add_v2r8(velecsum,velec);
421 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
423 /* Update vectorial force */
424 fix2 = _fjsp_madd_v2r8(dx20,fscal,fix2);
425 fiy2 = _fjsp_madd_v2r8(dy20,fscal,fiy2);
426 fiz2 = _fjsp_madd_v2r8(dz20,fscal,fiz2);
428 fjx0 = _fjsp_madd_v2r8(dx20,fscal,fjx0);
429 fjy0 = _fjsp_madd_v2r8(dy20,fscal,fjy0);
430 fjz0 = _fjsp_madd_v2r8(dz20,fscal,fjz0);
434 /**************************
435 * CALCULATE INTERACTIONS *
436 **************************/
438 if (gmx_fjsp_any_lt_v2r8(rsq30,rcutoff2))
441 r30 = _fjsp_mul_v2r8(rsq30,rinv30);
443 /* Compute parameters for interactions between i and j atoms */
444 qq30 = _fjsp_mul_v2r8(iq3,jq0);
446 /* EWALD ELECTROSTATICS */
448 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
449 ewrt = _fjsp_mul_v2r8(r30,ewtabscale);
450 itab_tmp = _fjsp_dtox_v2r8(ewrt);
451 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
452 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
454 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
455 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
456 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
457 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
458 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
459 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
460 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
461 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
462 velec = _fjsp_mul_v2r8(qq30,_fjsp_sub_v2r8(rinv30,velec));
463 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq30,rinv30),_fjsp_sub_v2r8(rinvsq30,felec));
465 d = _fjsp_sub_v2r8(r30,rswitch);
466 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
467 d2 = _fjsp_mul_v2r8(d,d);
468 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
470 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
472 /* Evaluate switch function */
473 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
474 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv30,_fjsp_mul_v2r8(velec,dsw)) );
475 velec = _fjsp_mul_v2r8(velec,sw);
476 cutoff_mask = _fjsp_cmplt_v2r8(rsq30,rcutoff2);
478 /* Update potential sum for this i atom from the interaction with this j atom. */
479 velec = _fjsp_and_v2r8(velec,cutoff_mask);
480 velecsum = _fjsp_add_v2r8(velecsum,velec);
484 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
486 /* Update vectorial force */
487 fix3 = _fjsp_madd_v2r8(dx30,fscal,fix3);
488 fiy3 = _fjsp_madd_v2r8(dy30,fscal,fiy3);
489 fiz3 = _fjsp_madd_v2r8(dz30,fscal,fiz3);
491 fjx0 = _fjsp_madd_v2r8(dx30,fscal,fjx0);
492 fjy0 = _fjsp_madd_v2r8(dy30,fscal,fjy0);
493 fjz0 = _fjsp_madd_v2r8(dz30,fscal,fjz0);
497 gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
499 /* Inner loop uses 269 flops */
506 j_coord_offsetA = DIM*jnrA;
508 /* load j atom coordinates */
509 gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
512 /* Calculate displacement vector */
513 dx00 = _fjsp_sub_v2r8(ix0,jx0);
514 dy00 = _fjsp_sub_v2r8(iy0,jy0);
515 dz00 = _fjsp_sub_v2r8(iz0,jz0);
516 dx10 = _fjsp_sub_v2r8(ix1,jx0);
517 dy10 = _fjsp_sub_v2r8(iy1,jy0);
518 dz10 = _fjsp_sub_v2r8(iz1,jz0);
519 dx20 = _fjsp_sub_v2r8(ix2,jx0);
520 dy20 = _fjsp_sub_v2r8(iy2,jy0);
521 dz20 = _fjsp_sub_v2r8(iz2,jz0);
522 dx30 = _fjsp_sub_v2r8(ix3,jx0);
523 dy30 = _fjsp_sub_v2r8(iy3,jy0);
524 dz30 = _fjsp_sub_v2r8(iz3,jz0);
526 /* Calculate squared distance and things based on it */
527 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
528 rsq10 = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
529 rsq20 = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
530 rsq30 = gmx_fjsp_calc_rsq_v2r8(dx30,dy30,dz30);
532 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
533 rinv10 = gmx_fjsp_invsqrt_v2r8(rsq10);
534 rinv20 = gmx_fjsp_invsqrt_v2r8(rsq20);
535 rinv30 = gmx_fjsp_invsqrt_v2r8(rsq30);
537 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
538 rinvsq10 = _fjsp_mul_v2r8(rinv10,rinv10);
539 rinvsq20 = _fjsp_mul_v2r8(rinv20,rinv20);
540 rinvsq30 = _fjsp_mul_v2r8(rinv30,rinv30);
542 /* Load parameters for j particles */
543 jq0 = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
544 vdwjidx0A = 2*vdwtype[jnrA+0];
546 fjx0 = _fjsp_setzero_v2r8();
547 fjy0 = _fjsp_setzero_v2r8();
548 fjz0 = _fjsp_setzero_v2r8();
550 /**************************
551 * CALCULATE INTERACTIONS *
552 **************************/
554 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
557 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
559 /* Compute parameters for interactions between i and j atoms */
560 gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
562 /* LENNARD-JONES DISPERSION/REPULSION */
564 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
565 vvdw6 = _fjsp_mul_v2r8(c6_00,rinvsix);
566 vvdw12 = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
567 vvdw = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
568 fvdw = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
570 d = _fjsp_sub_v2r8(r00,rswitch);
571 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
572 d2 = _fjsp_mul_v2r8(d,d);
573 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
575 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
577 /* Evaluate switch function */
578 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
579 fvdw = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
580 vvdw = _fjsp_mul_v2r8(vvdw,sw);
581 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
583 /* Update potential sum for this i atom from the interaction with this j atom. */
584 vvdw = _fjsp_and_v2r8(vvdw,cutoff_mask);
585 vvdw = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
586 vvdwsum = _fjsp_add_v2r8(vvdwsum,vvdw);
590 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
592 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
594 /* Update vectorial force */
595 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
596 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
597 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
599 fjx0 = _fjsp_madd_v2r8(dx00,fscal,fjx0);
600 fjy0 = _fjsp_madd_v2r8(dy00,fscal,fjy0);
601 fjz0 = _fjsp_madd_v2r8(dz00,fscal,fjz0);
605 /**************************
606 * CALCULATE INTERACTIONS *
607 **************************/
609 if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
612 r10 = _fjsp_mul_v2r8(rsq10,rinv10);
614 /* Compute parameters for interactions between i and j atoms */
615 qq10 = _fjsp_mul_v2r8(iq1,jq0);
617 /* EWALD ELECTROSTATICS */
619 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
620 ewrt = _fjsp_mul_v2r8(r10,ewtabscale);
621 itab_tmp = _fjsp_dtox_v2r8(ewrt);
622 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
623 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
625 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
626 ewtabD = _fjsp_setzero_v2r8();
627 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
628 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
629 ewtabFn = _fjsp_setzero_v2r8();
630 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
631 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
632 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
633 velec = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(rinv10,velec));
634 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
636 d = _fjsp_sub_v2r8(r10,rswitch);
637 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
638 d2 = _fjsp_mul_v2r8(d,d);
639 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
641 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
643 /* Evaluate switch function */
644 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
645 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv10,_fjsp_mul_v2r8(velec,dsw)) );
646 velec = _fjsp_mul_v2r8(velec,sw);
647 cutoff_mask = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
649 /* Update potential sum for this i atom from the interaction with this j atom. */
650 velec = _fjsp_and_v2r8(velec,cutoff_mask);
651 velec = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
652 velecsum = _fjsp_add_v2r8(velecsum,velec);
656 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
658 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
660 /* Update vectorial force */
661 fix1 = _fjsp_madd_v2r8(dx10,fscal,fix1);
662 fiy1 = _fjsp_madd_v2r8(dy10,fscal,fiy1);
663 fiz1 = _fjsp_madd_v2r8(dz10,fscal,fiz1);
665 fjx0 = _fjsp_madd_v2r8(dx10,fscal,fjx0);
666 fjy0 = _fjsp_madd_v2r8(dy10,fscal,fjy0);
667 fjz0 = _fjsp_madd_v2r8(dz10,fscal,fjz0);
671 /**************************
672 * CALCULATE INTERACTIONS *
673 **************************/
675 if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
678 r20 = _fjsp_mul_v2r8(rsq20,rinv20);
680 /* Compute parameters for interactions between i and j atoms */
681 qq20 = _fjsp_mul_v2r8(iq2,jq0);
683 /* EWALD ELECTROSTATICS */
685 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
686 ewrt = _fjsp_mul_v2r8(r20,ewtabscale);
687 itab_tmp = _fjsp_dtox_v2r8(ewrt);
688 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
689 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
691 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
692 ewtabD = _fjsp_setzero_v2r8();
693 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
694 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
695 ewtabFn = _fjsp_setzero_v2r8();
696 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
697 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
698 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
699 velec = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(rinv20,velec));
700 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
702 d = _fjsp_sub_v2r8(r20,rswitch);
703 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
704 d2 = _fjsp_mul_v2r8(d,d);
705 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
707 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
709 /* Evaluate switch function */
710 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
711 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv20,_fjsp_mul_v2r8(velec,dsw)) );
712 velec = _fjsp_mul_v2r8(velec,sw);
713 cutoff_mask = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
715 /* Update potential sum for this i atom from the interaction with this j atom. */
716 velec = _fjsp_and_v2r8(velec,cutoff_mask);
717 velec = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
718 velecsum = _fjsp_add_v2r8(velecsum,velec);
722 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
724 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
726 /* Update vectorial force */
727 fix2 = _fjsp_madd_v2r8(dx20,fscal,fix2);
728 fiy2 = _fjsp_madd_v2r8(dy20,fscal,fiy2);
729 fiz2 = _fjsp_madd_v2r8(dz20,fscal,fiz2);
731 fjx0 = _fjsp_madd_v2r8(dx20,fscal,fjx0);
732 fjy0 = _fjsp_madd_v2r8(dy20,fscal,fjy0);
733 fjz0 = _fjsp_madd_v2r8(dz20,fscal,fjz0);
737 /**************************
738 * CALCULATE INTERACTIONS *
739 **************************/
741 if (gmx_fjsp_any_lt_v2r8(rsq30,rcutoff2))
744 r30 = _fjsp_mul_v2r8(rsq30,rinv30);
746 /* Compute parameters for interactions between i and j atoms */
747 qq30 = _fjsp_mul_v2r8(iq3,jq0);
749 /* EWALD ELECTROSTATICS */
751 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
752 ewrt = _fjsp_mul_v2r8(r30,ewtabscale);
753 itab_tmp = _fjsp_dtox_v2r8(ewrt);
754 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
755 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
757 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
758 ewtabD = _fjsp_setzero_v2r8();
759 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
760 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
761 ewtabFn = _fjsp_setzero_v2r8();
762 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
763 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
764 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
765 velec = _fjsp_mul_v2r8(qq30,_fjsp_sub_v2r8(rinv30,velec));
766 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq30,rinv30),_fjsp_sub_v2r8(rinvsq30,felec));
768 d = _fjsp_sub_v2r8(r30,rswitch);
769 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
770 d2 = _fjsp_mul_v2r8(d,d);
771 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
773 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
775 /* Evaluate switch function */
776 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
777 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv30,_fjsp_mul_v2r8(velec,dsw)) );
778 velec = _fjsp_mul_v2r8(velec,sw);
779 cutoff_mask = _fjsp_cmplt_v2r8(rsq30,rcutoff2);
781 /* Update potential sum for this i atom from the interaction with this j atom. */
782 velec = _fjsp_and_v2r8(velec,cutoff_mask);
783 velec = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
784 velecsum = _fjsp_add_v2r8(velecsum,velec);
788 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
790 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
792 /* Update vectorial force */
793 fix3 = _fjsp_madd_v2r8(dx30,fscal,fix3);
794 fiy3 = _fjsp_madd_v2r8(dy30,fscal,fiy3);
795 fiz3 = _fjsp_madd_v2r8(dz30,fscal,fiz3);
797 fjx0 = _fjsp_madd_v2r8(dx30,fscal,fjx0);
798 fjy0 = _fjsp_madd_v2r8(dy30,fscal,fjy0);
799 fjz0 = _fjsp_madd_v2r8(dz30,fscal,fjz0);
803 gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
805 /* Inner loop uses 269 flops */
808 /* End of innermost loop */
810 gmx_fjsp_update_iforce_4atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
811 f+i_coord_offset,fshift+i_shift_offset);
814 /* Update potential energies */
815 gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
816 gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
818 /* Increment number of inner iterations */
819 inneriter += j_index_end - j_index_start;
821 /* Outer loop uses 26 flops */
824 /* Increment number of outer iterations */
827 /* Update outer/inner flops */
829 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*269);
832 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_sparc64_hpc_ace_double
833 * Electrostatics interaction: Ewald
834 * VdW interaction: LennardJones
835 * Geometry: Water4-Particle
836 * Calculate force/pot: Force
839 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_sparc64_hpc_ace_double
840 (t_nblist * gmx_restrict nlist,
841 rvec * gmx_restrict xx,
842 rvec * gmx_restrict ff,
843 t_forcerec * gmx_restrict fr,
844 t_mdatoms * gmx_restrict mdatoms,
845 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
846 t_nrnb * gmx_restrict nrnb)
848 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
849 * just 0 for non-waters.
850 * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
851 * jnr indices corresponding to data put in the four positions in the SIMD register.
853 int i_shift_offset,i_coord_offset,outeriter,inneriter;
854 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
856 int j_coord_offsetA,j_coord_offsetB;
857 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
859 real *shiftvec,*fshift,*x,*f;
860 _fjsp_v2r8 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
862 _fjsp_v2r8 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
864 _fjsp_v2r8 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
866 _fjsp_v2r8 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
868 _fjsp_v2r8 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
869 int vdwjidx0A,vdwjidx0B;
870 _fjsp_v2r8 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
871 _fjsp_v2r8 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
872 _fjsp_v2r8 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
873 _fjsp_v2r8 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
874 _fjsp_v2r8 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
875 _fjsp_v2r8 velec,felec,velecsum,facel,crf,krf,krf2;
878 _fjsp_v2r8 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
881 _fjsp_v2r8 one_sixth = gmx_fjsp_set1_v2r8(1.0/6.0);
882 _fjsp_v2r8 one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
883 _fjsp_v2r8 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
885 _fjsp_v2r8 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
886 real rswitch_scalar,d_scalar;
888 _fjsp_v2r8 dummy_mask,cutoff_mask;
889 _fjsp_v2r8 one = gmx_fjsp_set1_v2r8(1.0);
890 _fjsp_v2r8 two = gmx_fjsp_set1_v2r8(2.0);
891 union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
898 jindex = nlist->jindex;
900 shiftidx = nlist->shift;
902 shiftvec = fr->shift_vec[0];
903 fshift = fr->fshift[0];
904 facel = gmx_fjsp_set1_v2r8(fr->epsfac);
905 charge = mdatoms->chargeA;
906 nvdwtype = fr->ntype;
908 vdwtype = mdatoms->typeA;
910 sh_ewald = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
911 ewtab = fr->ic->tabq_coul_FDV0;
912 ewtabscale = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
913 ewtabhalfspace = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
915 /* Setup water-specific parameters */
916 inr = nlist->iinr[0];
917 iq1 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+1]));
918 iq2 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+2]));
919 iq3 = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+3]));
920 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
922 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
923 rcutoff_scalar = fr->rcoulomb;
924 rcutoff = gmx_fjsp_set1_v2r8(rcutoff_scalar);
925 rcutoff2 = _fjsp_mul_v2r8(rcutoff,rcutoff);
927 rswitch_scalar = fr->rcoulomb_switch;
928 rswitch = gmx_fjsp_set1_v2r8(rswitch_scalar);
929 /* Setup switch parameters */
930 d_scalar = rcutoff_scalar-rswitch_scalar;
931 d = gmx_fjsp_set1_v2r8(d_scalar);
932 swV3 = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
933 swV4 = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
934 swV5 = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
935 swF2 = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
936 swF3 = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
937 swF4 = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
939 /* Avoid stupid compiler warnings */
947 /* Start outer loop over neighborlists */
948 for(iidx=0; iidx<nri; iidx++)
950 /* Load shift vector for this list */
951 i_shift_offset = DIM*shiftidx[iidx];
953 /* Load limits for loop over neighbors */
954 j_index_start = jindex[iidx];
955 j_index_end = jindex[iidx+1];
957 /* Get outer coordinate index */
959 i_coord_offset = DIM*inr;
961 /* Load i particle coords and add shift vector */
962 gmx_fjsp_load_shift_and_4rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
963 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
965 fix0 = _fjsp_setzero_v2r8();
966 fiy0 = _fjsp_setzero_v2r8();
967 fiz0 = _fjsp_setzero_v2r8();
968 fix1 = _fjsp_setzero_v2r8();
969 fiy1 = _fjsp_setzero_v2r8();
970 fiz1 = _fjsp_setzero_v2r8();
971 fix2 = _fjsp_setzero_v2r8();
972 fiy2 = _fjsp_setzero_v2r8();
973 fiz2 = _fjsp_setzero_v2r8();
974 fix3 = _fjsp_setzero_v2r8();
975 fiy3 = _fjsp_setzero_v2r8();
976 fiz3 = _fjsp_setzero_v2r8();
978 /* Start inner kernel loop */
979 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
982 /* Get j neighbor index, and coordinate index */
985 j_coord_offsetA = DIM*jnrA;
986 j_coord_offsetB = DIM*jnrB;
988 /* load j atom coordinates */
989 gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
992 /* Calculate displacement vector */
993 dx00 = _fjsp_sub_v2r8(ix0,jx0);
994 dy00 = _fjsp_sub_v2r8(iy0,jy0);
995 dz00 = _fjsp_sub_v2r8(iz0,jz0);
996 dx10 = _fjsp_sub_v2r8(ix1,jx0);
997 dy10 = _fjsp_sub_v2r8(iy1,jy0);
998 dz10 = _fjsp_sub_v2r8(iz1,jz0);
999 dx20 = _fjsp_sub_v2r8(ix2,jx0);
1000 dy20 = _fjsp_sub_v2r8(iy2,jy0);
1001 dz20 = _fjsp_sub_v2r8(iz2,jz0);
1002 dx30 = _fjsp_sub_v2r8(ix3,jx0);
1003 dy30 = _fjsp_sub_v2r8(iy3,jy0);
1004 dz30 = _fjsp_sub_v2r8(iz3,jz0);
1006 /* Calculate squared distance and things based on it */
1007 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
1008 rsq10 = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
1009 rsq20 = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
1010 rsq30 = gmx_fjsp_calc_rsq_v2r8(dx30,dy30,dz30);
1012 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
1013 rinv10 = gmx_fjsp_invsqrt_v2r8(rsq10);
1014 rinv20 = gmx_fjsp_invsqrt_v2r8(rsq20);
1015 rinv30 = gmx_fjsp_invsqrt_v2r8(rsq30);
1017 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
1018 rinvsq10 = _fjsp_mul_v2r8(rinv10,rinv10);
1019 rinvsq20 = _fjsp_mul_v2r8(rinv20,rinv20);
1020 rinvsq30 = _fjsp_mul_v2r8(rinv30,rinv30);
1022 /* Load parameters for j particles */
1023 jq0 = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
1024 vdwjidx0A = 2*vdwtype[jnrA+0];
1025 vdwjidx0B = 2*vdwtype[jnrB+0];
1027 fjx0 = _fjsp_setzero_v2r8();
1028 fjy0 = _fjsp_setzero_v2r8();
1029 fjz0 = _fjsp_setzero_v2r8();
1031 /**************************
1032 * CALCULATE INTERACTIONS *
1033 **************************/
1035 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
1038 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
1040 /* Compute parameters for interactions between i and j atoms */
1041 gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
1042 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
1044 /* LENNARD-JONES DISPERSION/REPULSION */
1046 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
1047 vvdw6 = _fjsp_mul_v2r8(c6_00,rinvsix);
1048 vvdw12 = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
1049 vvdw = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
1050 fvdw = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
1052 d = _fjsp_sub_v2r8(r00,rswitch);
1053 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
1054 d2 = _fjsp_mul_v2r8(d,d);
1055 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
1057 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
1059 /* Evaluate switch function */
1060 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1061 fvdw = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
1062 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
1066 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1068 /* Update vectorial force */
1069 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
1070 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
1071 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
1073 fjx0 = _fjsp_madd_v2r8(dx00,fscal,fjx0);
1074 fjy0 = _fjsp_madd_v2r8(dy00,fscal,fjy0);
1075 fjz0 = _fjsp_madd_v2r8(dz00,fscal,fjz0);
1079 /**************************
1080 * CALCULATE INTERACTIONS *
1081 **************************/
1083 if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
1086 r10 = _fjsp_mul_v2r8(rsq10,rinv10);
1088 /* Compute parameters for interactions between i and j atoms */
1089 qq10 = _fjsp_mul_v2r8(iq1,jq0);
1091 /* EWALD ELECTROSTATICS */
1093 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1094 ewrt = _fjsp_mul_v2r8(r10,ewtabscale);
1095 itab_tmp = _fjsp_dtox_v2r8(ewrt);
1096 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1097 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1099 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
1100 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
1101 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
1102 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
1103 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
1104 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
1105 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
1106 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
1107 velec = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(rinv10,velec));
1108 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
1110 d = _fjsp_sub_v2r8(r10,rswitch);
1111 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
1112 d2 = _fjsp_mul_v2r8(d,d);
1113 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
1115 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
1117 /* Evaluate switch function */
1118 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1119 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv10,_fjsp_mul_v2r8(velec,dsw)) );
1120 cutoff_mask = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
1124 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1126 /* Update vectorial force */
1127 fix1 = _fjsp_madd_v2r8(dx10,fscal,fix1);
1128 fiy1 = _fjsp_madd_v2r8(dy10,fscal,fiy1);
1129 fiz1 = _fjsp_madd_v2r8(dz10,fscal,fiz1);
1131 fjx0 = _fjsp_madd_v2r8(dx10,fscal,fjx0);
1132 fjy0 = _fjsp_madd_v2r8(dy10,fscal,fjy0);
1133 fjz0 = _fjsp_madd_v2r8(dz10,fscal,fjz0);
1137 /**************************
1138 * CALCULATE INTERACTIONS *
1139 **************************/
1141 if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
1144 r20 = _fjsp_mul_v2r8(rsq20,rinv20);
1146 /* Compute parameters for interactions between i and j atoms */
1147 qq20 = _fjsp_mul_v2r8(iq2,jq0);
1149 /* EWALD ELECTROSTATICS */
1151 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1152 ewrt = _fjsp_mul_v2r8(r20,ewtabscale);
1153 itab_tmp = _fjsp_dtox_v2r8(ewrt);
1154 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1155 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1157 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
1158 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
1159 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
1160 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
1161 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
1162 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
1163 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
1164 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
1165 velec = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(rinv20,velec));
1166 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
1168 d = _fjsp_sub_v2r8(r20,rswitch);
1169 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
1170 d2 = _fjsp_mul_v2r8(d,d);
1171 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
1173 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
1175 /* Evaluate switch function */
1176 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1177 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv20,_fjsp_mul_v2r8(velec,dsw)) );
1178 cutoff_mask = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
1182 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1184 /* Update vectorial force */
1185 fix2 = _fjsp_madd_v2r8(dx20,fscal,fix2);
1186 fiy2 = _fjsp_madd_v2r8(dy20,fscal,fiy2);
1187 fiz2 = _fjsp_madd_v2r8(dz20,fscal,fiz2);
1189 fjx0 = _fjsp_madd_v2r8(dx20,fscal,fjx0);
1190 fjy0 = _fjsp_madd_v2r8(dy20,fscal,fjy0);
1191 fjz0 = _fjsp_madd_v2r8(dz20,fscal,fjz0);
1195 /**************************
1196 * CALCULATE INTERACTIONS *
1197 **************************/
1199 if (gmx_fjsp_any_lt_v2r8(rsq30,rcutoff2))
1202 r30 = _fjsp_mul_v2r8(rsq30,rinv30);
1204 /* Compute parameters for interactions between i and j atoms */
1205 qq30 = _fjsp_mul_v2r8(iq3,jq0);
1207 /* EWALD ELECTROSTATICS */
1209 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1210 ewrt = _fjsp_mul_v2r8(r30,ewtabscale);
1211 itab_tmp = _fjsp_dtox_v2r8(ewrt);
1212 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1213 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1215 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
1216 ewtabD = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
1217 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
1218 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
1219 ewtabFn = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
1220 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
1221 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
1222 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
1223 velec = _fjsp_mul_v2r8(qq30,_fjsp_sub_v2r8(rinv30,velec));
1224 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq30,rinv30),_fjsp_sub_v2r8(rinvsq30,felec));
1226 d = _fjsp_sub_v2r8(r30,rswitch);
1227 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
1228 d2 = _fjsp_mul_v2r8(d,d);
1229 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
1231 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
1233 /* Evaluate switch function */
1234 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1235 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv30,_fjsp_mul_v2r8(velec,dsw)) );
1236 cutoff_mask = _fjsp_cmplt_v2r8(rsq30,rcutoff2);
1240 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1242 /* Update vectorial force */
1243 fix3 = _fjsp_madd_v2r8(dx30,fscal,fix3);
1244 fiy3 = _fjsp_madd_v2r8(dy30,fscal,fiy3);
1245 fiz3 = _fjsp_madd_v2r8(dz30,fscal,fiz3);
1247 fjx0 = _fjsp_madd_v2r8(dx30,fscal,fjx0);
1248 fjy0 = _fjsp_madd_v2r8(dy30,fscal,fjy0);
1249 fjz0 = _fjsp_madd_v2r8(dz30,fscal,fjz0);
1253 gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1255 /* Inner loop uses 257 flops */
1258 if(jidx<j_index_end)
1262 j_coord_offsetA = DIM*jnrA;
1264 /* load j atom coordinates */
1265 gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
1268 /* Calculate displacement vector */
1269 dx00 = _fjsp_sub_v2r8(ix0,jx0);
1270 dy00 = _fjsp_sub_v2r8(iy0,jy0);
1271 dz00 = _fjsp_sub_v2r8(iz0,jz0);
1272 dx10 = _fjsp_sub_v2r8(ix1,jx0);
1273 dy10 = _fjsp_sub_v2r8(iy1,jy0);
1274 dz10 = _fjsp_sub_v2r8(iz1,jz0);
1275 dx20 = _fjsp_sub_v2r8(ix2,jx0);
1276 dy20 = _fjsp_sub_v2r8(iy2,jy0);
1277 dz20 = _fjsp_sub_v2r8(iz2,jz0);
1278 dx30 = _fjsp_sub_v2r8(ix3,jx0);
1279 dy30 = _fjsp_sub_v2r8(iy3,jy0);
1280 dz30 = _fjsp_sub_v2r8(iz3,jz0);
1282 /* Calculate squared distance and things based on it */
1283 rsq00 = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
1284 rsq10 = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
1285 rsq20 = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
1286 rsq30 = gmx_fjsp_calc_rsq_v2r8(dx30,dy30,dz30);
1288 rinv00 = gmx_fjsp_invsqrt_v2r8(rsq00);
1289 rinv10 = gmx_fjsp_invsqrt_v2r8(rsq10);
1290 rinv20 = gmx_fjsp_invsqrt_v2r8(rsq20);
1291 rinv30 = gmx_fjsp_invsqrt_v2r8(rsq30);
1293 rinvsq00 = _fjsp_mul_v2r8(rinv00,rinv00);
1294 rinvsq10 = _fjsp_mul_v2r8(rinv10,rinv10);
1295 rinvsq20 = _fjsp_mul_v2r8(rinv20,rinv20);
1296 rinvsq30 = _fjsp_mul_v2r8(rinv30,rinv30);
1298 /* Load parameters for j particles */
1299 jq0 = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
1300 vdwjidx0A = 2*vdwtype[jnrA+0];
1302 fjx0 = _fjsp_setzero_v2r8();
1303 fjy0 = _fjsp_setzero_v2r8();
1304 fjz0 = _fjsp_setzero_v2r8();
1306 /**************************
1307 * CALCULATE INTERACTIONS *
1308 **************************/
1310 if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
1313 r00 = _fjsp_mul_v2r8(rsq00,rinv00);
1315 /* Compute parameters for interactions between i and j atoms */
1316 gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1318 /* LENNARD-JONES DISPERSION/REPULSION */
1320 rinvsix = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
1321 vvdw6 = _fjsp_mul_v2r8(c6_00,rinvsix);
1322 vvdw12 = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
1323 vvdw = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
1324 fvdw = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
1326 d = _fjsp_sub_v2r8(r00,rswitch);
1327 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
1328 d2 = _fjsp_mul_v2r8(d,d);
1329 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
1331 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
1333 /* Evaluate switch function */
1334 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1335 fvdw = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
1336 cutoff_mask = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
1340 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1342 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1344 /* Update vectorial force */
1345 fix0 = _fjsp_madd_v2r8(dx00,fscal,fix0);
1346 fiy0 = _fjsp_madd_v2r8(dy00,fscal,fiy0);
1347 fiz0 = _fjsp_madd_v2r8(dz00,fscal,fiz0);
1349 fjx0 = _fjsp_madd_v2r8(dx00,fscal,fjx0);
1350 fjy0 = _fjsp_madd_v2r8(dy00,fscal,fjy0);
1351 fjz0 = _fjsp_madd_v2r8(dz00,fscal,fjz0);
1355 /**************************
1356 * CALCULATE INTERACTIONS *
1357 **************************/
1359 if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
1362 r10 = _fjsp_mul_v2r8(rsq10,rinv10);
1364 /* Compute parameters for interactions between i and j atoms */
1365 qq10 = _fjsp_mul_v2r8(iq1,jq0);
1367 /* EWALD ELECTROSTATICS */
1369 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1370 ewrt = _fjsp_mul_v2r8(r10,ewtabscale);
1371 itab_tmp = _fjsp_dtox_v2r8(ewrt);
1372 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1373 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1375 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
1376 ewtabD = _fjsp_setzero_v2r8();
1377 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
1378 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
1379 ewtabFn = _fjsp_setzero_v2r8();
1380 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
1381 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
1382 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
1383 velec = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(rinv10,velec));
1384 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
1386 d = _fjsp_sub_v2r8(r10,rswitch);
1387 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
1388 d2 = _fjsp_mul_v2r8(d,d);
1389 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
1391 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
1393 /* Evaluate switch function */
1394 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1395 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv10,_fjsp_mul_v2r8(velec,dsw)) );
1396 cutoff_mask = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
1400 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1402 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1404 /* Update vectorial force */
1405 fix1 = _fjsp_madd_v2r8(dx10,fscal,fix1);
1406 fiy1 = _fjsp_madd_v2r8(dy10,fscal,fiy1);
1407 fiz1 = _fjsp_madd_v2r8(dz10,fscal,fiz1);
1409 fjx0 = _fjsp_madd_v2r8(dx10,fscal,fjx0);
1410 fjy0 = _fjsp_madd_v2r8(dy10,fscal,fjy0);
1411 fjz0 = _fjsp_madd_v2r8(dz10,fscal,fjz0);
1415 /**************************
1416 * CALCULATE INTERACTIONS *
1417 **************************/
1419 if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
1422 r20 = _fjsp_mul_v2r8(rsq20,rinv20);
1424 /* Compute parameters for interactions between i and j atoms */
1425 qq20 = _fjsp_mul_v2r8(iq2,jq0);
1427 /* EWALD ELECTROSTATICS */
1429 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1430 ewrt = _fjsp_mul_v2r8(r20,ewtabscale);
1431 itab_tmp = _fjsp_dtox_v2r8(ewrt);
1432 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1433 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1435 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
1436 ewtabD = _fjsp_setzero_v2r8();
1437 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
1438 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
1439 ewtabFn = _fjsp_setzero_v2r8();
1440 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
1441 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
1442 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
1443 velec = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(rinv20,velec));
1444 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
1446 d = _fjsp_sub_v2r8(r20,rswitch);
1447 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
1448 d2 = _fjsp_mul_v2r8(d,d);
1449 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
1451 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
1453 /* Evaluate switch function */
1454 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1455 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv20,_fjsp_mul_v2r8(velec,dsw)) );
1456 cutoff_mask = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
1460 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1462 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1464 /* Update vectorial force */
1465 fix2 = _fjsp_madd_v2r8(dx20,fscal,fix2);
1466 fiy2 = _fjsp_madd_v2r8(dy20,fscal,fiy2);
1467 fiz2 = _fjsp_madd_v2r8(dz20,fscal,fiz2);
1469 fjx0 = _fjsp_madd_v2r8(dx20,fscal,fjx0);
1470 fjy0 = _fjsp_madd_v2r8(dy20,fscal,fjy0);
1471 fjz0 = _fjsp_madd_v2r8(dz20,fscal,fjz0);
1475 /**************************
1476 * CALCULATE INTERACTIONS *
1477 **************************/
1479 if (gmx_fjsp_any_lt_v2r8(rsq30,rcutoff2))
1482 r30 = _fjsp_mul_v2r8(rsq30,rinv30);
1484 /* Compute parameters for interactions between i and j atoms */
1485 qq30 = _fjsp_mul_v2r8(iq3,jq0);
1487 /* EWALD ELECTROSTATICS */
1489 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1490 ewrt = _fjsp_mul_v2r8(r30,ewtabscale);
1491 itab_tmp = _fjsp_dtox_v2r8(ewrt);
1492 eweps = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1493 _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1495 ewtabF = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
1496 ewtabD = _fjsp_setzero_v2r8();
1497 GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
1498 ewtabV = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
1499 ewtabFn = _fjsp_setzero_v2r8();
1500 GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
1501 felec = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
1502 velec = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
1503 velec = _fjsp_mul_v2r8(qq30,_fjsp_sub_v2r8(rinv30,velec));
1504 felec = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq30,rinv30),_fjsp_sub_v2r8(rinvsq30,felec));
1506 d = _fjsp_sub_v2r8(r30,rswitch);
1507 d = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
1508 d2 = _fjsp_mul_v2r8(d,d);
1509 sw = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
1511 dsw = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
1513 /* Evaluate switch function */
1514 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1515 felec = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv30,_fjsp_mul_v2r8(velec,dsw)) );
1516 cutoff_mask = _fjsp_cmplt_v2r8(rsq30,rcutoff2);
1520 fscal = _fjsp_and_v2r8(fscal,cutoff_mask);
1522 fscal = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1524 /* Update vectorial force */
1525 fix3 = _fjsp_madd_v2r8(dx30,fscal,fix3);
1526 fiy3 = _fjsp_madd_v2r8(dy30,fscal,fiy3);
1527 fiz3 = _fjsp_madd_v2r8(dz30,fscal,fiz3);
1529 fjx0 = _fjsp_madd_v2r8(dx30,fscal,fjx0);
1530 fjy0 = _fjsp_madd_v2r8(dy30,fscal,fjy0);
1531 fjz0 = _fjsp_madd_v2r8(dz30,fscal,fjz0);
1535 gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1537 /* Inner loop uses 257 flops */
1540 /* End of innermost loop */
1542 gmx_fjsp_update_iforce_4atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1543 f+i_coord_offset,fshift+i_shift_offset);
1545 /* Increment number of inner iterations */
1546 inneriter += j_index_end - j_index_start;
1548 /* Outer loop uses 24 flops */
1551 /* Increment number of outer iterations */
1554 /* Update outer/inner flops */
1556 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*257);