2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 sse4_1_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/gmxlib/nrnb.h"
47 #include "kernelutil_x86_sse4_1_single.h"
50 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_sse4_1_single
51 * Electrostatics interaction: ReactionField
52 * VdW interaction: LennardJones
53 * Geometry: Water4-Particle
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_sse4_1_single
58 (t_nblist * gmx_restrict nlist,
59 rvec * gmx_restrict xx,
60 rvec * gmx_restrict ff,
61 struct 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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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;
73 int jnrA,jnrB,jnrC,jnrD;
74 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
75 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
81 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
87 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
89 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
90 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
91 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
93 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
94 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
95 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
96 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
99 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
102 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
103 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
104 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
105 real rswitch_scalar,d_scalar;
106 __m128 dummy_mask,cutoff_mask;
107 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
108 __m128 one = _mm_set1_ps(1.0);
109 __m128 two = _mm_set1_ps(2.0);
115 jindex = nlist->jindex;
117 shiftidx = nlist->shift;
119 shiftvec = fr->shift_vec[0];
120 fshift = fr->fshift[0];
121 facel = _mm_set1_ps(fr->ic->epsfac);
122 charge = mdatoms->chargeA;
123 krf = _mm_set1_ps(fr->ic->k_rf);
124 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
125 crf = _mm_set1_ps(fr->ic->c_rf);
126 nvdwtype = fr->ntype;
128 vdwtype = mdatoms->typeA;
130 /* Setup water-specific parameters */
131 inr = nlist->iinr[0];
132 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
133 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
134 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
135 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
137 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
138 rcutoff_scalar = fr->ic->rcoulomb;
139 rcutoff = _mm_set1_ps(rcutoff_scalar);
140 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
142 rswitch_scalar = fr->ic->rvdw_switch;
143 rswitch = _mm_set1_ps(rswitch_scalar);
144 /* Setup switch parameters */
145 d_scalar = rcutoff_scalar-rswitch_scalar;
146 d = _mm_set1_ps(d_scalar);
147 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
148 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
149 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
150 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
151 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
152 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
154 /* Avoid stupid compiler warnings */
155 jnrA = jnrB = jnrC = jnrD = 0;
164 for(iidx=0;iidx<4*DIM;iidx++)
169 /* Start outer loop over neighborlists */
170 for(iidx=0; iidx<nri; iidx++)
172 /* Load shift vector for this list */
173 i_shift_offset = DIM*shiftidx[iidx];
175 /* Load limits for loop over neighbors */
176 j_index_start = jindex[iidx];
177 j_index_end = jindex[iidx+1];
179 /* Get outer coordinate index */
181 i_coord_offset = DIM*inr;
183 /* Load i particle coords and add shift vector */
184 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
185 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
187 fix0 = _mm_setzero_ps();
188 fiy0 = _mm_setzero_ps();
189 fiz0 = _mm_setzero_ps();
190 fix1 = _mm_setzero_ps();
191 fiy1 = _mm_setzero_ps();
192 fiz1 = _mm_setzero_ps();
193 fix2 = _mm_setzero_ps();
194 fiy2 = _mm_setzero_ps();
195 fiz2 = _mm_setzero_ps();
196 fix3 = _mm_setzero_ps();
197 fiy3 = _mm_setzero_ps();
198 fiz3 = _mm_setzero_ps();
200 /* Reset potential sums */
201 velecsum = _mm_setzero_ps();
202 vvdwsum = _mm_setzero_ps();
204 /* Start inner kernel loop */
205 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
208 /* Get j neighbor index, and coordinate index */
213 j_coord_offsetA = DIM*jnrA;
214 j_coord_offsetB = DIM*jnrB;
215 j_coord_offsetC = DIM*jnrC;
216 j_coord_offsetD = DIM*jnrD;
218 /* load j atom coordinates */
219 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
220 x+j_coord_offsetC,x+j_coord_offsetD,
223 /* Calculate displacement vector */
224 dx00 = _mm_sub_ps(ix0,jx0);
225 dy00 = _mm_sub_ps(iy0,jy0);
226 dz00 = _mm_sub_ps(iz0,jz0);
227 dx10 = _mm_sub_ps(ix1,jx0);
228 dy10 = _mm_sub_ps(iy1,jy0);
229 dz10 = _mm_sub_ps(iz1,jz0);
230 dx20 = _mm_sub_ps(ix2,jx0);
231 dy20 = _mm_sub_ps(iy2,jy0);
232 dz20 = _mm_sub_ps(iz2,jz0);
233 dx30 = _mm_sub_ps(ix3,jx0);
234 dy30 = _mm_sub_ps(iy3,jy0);
235 dz30 = _mm_sub_ps(iz3,jz0);
237 /* Calculate squared distance and things based on it */
238 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
239 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
240 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
241 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
243 rinv00 = sse41_invsqrt_f(rsq00);
244 rinv10 = sse41_invsqrt_f(rsq10);
245 rinv20 = sse41_invsqrt_f(rsq20);
246 rinv30 = sse41_invsqrt_f(rsq30);
248 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
249 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
250 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
251 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
253 /* Load parameters for j particles */
254 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
255 charge+jnrC+0,charge+jnrD+0);
256 vdwjidx0A = 2*vdwtype[jnrA+0];
257 vdwjidx0B = 2*vdwtype[jnrB+0];
258 vdwjidx0C = 2*vdwtype[jnrC+0];
259 vdwjidx0D = 2*vdwtype[jnrD+0];
261 fjx0 = _mm_setzero_ps();
262 fjy0 = _mm_setzero_ps();
263 fjz0 = _mm_setzero_ps();
265 /**************************
266 * CALCULATE INTERACTIONS *
267 **************************/
269 if (gmx_mm_any_lt(rsq00,rcutoff2))
272 r00 = _mm_mul_ps(rsq00,rinv00);
274 /* Compute parameters for interactions between i and j atoms */
275 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
276 vdwparam+vdwioffset0+vdwjidx0B,
277 vdwparam+vdwioffset0+vdwjidx0C,
278 vdwparam+vdwioffset0+vdwjidx0D,
281 /* LENNARD-JONES DISPERSION/REPULSION */
283 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
284 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
285 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
286 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
287 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
289 d = _mm_sub_ps(r00,rswitch);
290 d = _mm_max_ps(d,_mm_setzero_ps());
291 d2 = _mm_mul_ps(d,d);
292 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
294 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
296 /* Evaluate switch function */
297 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
298 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
299 vvdw = _mm_mul_ps(vvdw,sw);
300 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
302 /* Update potential sum for this i atom from the interaction with this j atom. */
303 vvdw = _mm_and_ps(vvdw,cutoff_mask);
304 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
308 fscal = _mm_and_ps(fscal,cutoff_mask);
310 /* Calculate temporary vectorial force */
311 tx = _mm_mul_ps(fscal,dx00);
312 ty = _mm_mul_ps(fscal,dy00);
313 tz = _mm_mul_ps(fscal,dz00);
315 /* Update vectorial force */
316 fix0 = _mm_add_ps(fix0,tx);
317 fiy0 = _mm_add_ps(fiy0,ty);
318 fiz0 = _mm_add_ps(fiz0,tz);
320 fjx0 = _mm_add_ps(fjx0,tx);
321 fjy0 = _mm_add_ps(fjy0,ty);
322 fjz0 = _mm_add_ps(fjz0,tz);
326 /**************************
327 * CALCULATE INTERACTIONS *
328 **************************/
330 if (gmx_mm_any_lt(rsq10,rcutoff2))
333 /* Compute parameters for interactions between i and j atoms */
334 qq10 = _mm_mul_ps(iq1,jq0);
336 /* REACTION-FIELD ELECTROSTATICS */
337 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
338 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
340 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
342 /* Update potential sum for this i atom from the interaction with this j atom. */
343 velec = _mm_and_ps(velec,cutoff_mask);
344 velecsum = _mm_add_ps(velecsum,velec);
348 fscal = _mm_and_ps(fscal,cutoff_mask);
350 /* Calculate temporary vectorial force */
351 tx = _mm_mul_ps(fscal,dx10);
352 ty = _mm_mul_ps(fscal,dy10);
353 tz = _mm_mul_ps(fscal,dz10);
355 /* Update vectorial force */
356 fix1 = _mm_add_ps(fix1,tx);
357 fiy1 = _mm_add_ps(fiy1,ty);
358 fiz1 = _mm_add_ps(fiz1,tz);
360 fjx0 = _mm_add_ps(fjx0,tx);
361 fjy0 = _mm_add_ps(fjy0,ty);
362 fjz0 = _mm_add_ps(fjz0,tz);
366 /**************************
367 * CALCULATE INTERACTIONS *
368 **************************/
370 if (gmx_mm_any_lt(rsq20,rcutoff2))
373 /* Compute parameters for interactions between i and j atoms */
374 qq20 = _mm_mul_ps(iq2,jq0);
376 /* REACTION-FIELD ELECTROSTATICS */
377 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
378 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
380 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
382 /* Update potential sum for this i atom from the interaction with this j atom. */
383 velec = _mm_and_ps(velec,cutoff_mask);
384 velecsum = _mm_add_ps(velecsum,velec);
388 fscal = _mm_and_ps(fscal,cutoff_mask);
390 /* Calculate temporary vectorial force */
391 tx = _mm_mul_ps(fscal,dx20);
392 ty = _mm_mul_ps(fscal,dy20);
393 tz = _mm_mul_ps(fscal,dz20);
395 /* Update vectorial force */
396 fix2 = _mm_add_ps(fix2,tx);
397 fiy2 = _mm_add_ps(fiy2,ty);
398 fiz2 = _mm_add_ps(fiz2,tz);
400 fjx0 = _mm_add_ps(fjx0,tx);
401 fjy0 = _mm_add_ps(fjy0,ty);
402 fjz0 = _mm_add_ps(fjz0,tz);
406 /**************************
407 * CALCULATE INTERACTIONS *
408 **************************/
410 if (gmx_mm_any_lt(rsq30,rcutoff2))
413 /* Compute parameters for interactions between i and j atoms */
414 qq30 = _mm_mul_ps(iq3,jq0);
416 /* REACTION-FIELD ELECTROSTATICS */
417 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
418 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
420 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
422 /* Update potential sum for this i atom from the interaction with this j atom. */
423 velec = _mm_and_ps(velec,cutoff_mask);
424 velecsum = _mm_add_ps(velecsum,velec);
428 fscal = _mm_and_ps(fscal,cutoff_mask);
430 /* Calculate temporary vectorial force */
431 tx = _mm_mul_ps(fscal,dx30);
432 ty = _mm_mul_ps(fscal,dy30);
433 tz = _mm_mul_ps(fscal,dz30);
435 /* Update vectorial force */
436 fix3 = _mm_add_ps(fix3,tx);
437 fiy3 = _mm_add_ps(fiy3,ty);
438 fiz3 = _mm_add_ps(fiz3,tz);
440 fjx0 = _mm_add_ps(fjx0,tx);
441 fjy0 = _mm_add_ps(fjy0,ty);
442 fjz0 = _mm_add_ps(fjz0,tz);
446 fjptrA = f+j_coord_offsetA;
447 fjptrB = f+j_coord_offsetB;
448 fjptrC = f+j_coord_offsetC;
449 fjptrD = f+j_coord_offsetD;
451 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
453 /* Inner loop uses 167 flops */
459 /* Get j neighbor index, and coordinate index */
460 jnrlistA = jjnr[jidx];
461 jnrlistB = jjnr[jidx+1];
462 jnrlistC = jjnr[jidx+2];
463 jnrlistD = jjnr[jidx+3];
464 /* Sign of each element will be negative for non-real atoms.
465 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
466 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
468 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
469 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
470 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
471 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
472 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
473 j_coord_offsetA = DIM*jnrA;
474 j_coord_offsetB = DIM*jnrB;
475 j_coord_offsetC = DIM*jnrC;
476 j_coord_offsetD = DIM*jnrD;
478 /* load j atom coordinates */
479 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
480 x+j_coord_offsetC,x+j_coord_offsetD,
483 /* Calculate displacement vector */
484 dx00 = _mm_sub_ps(ix0,jx0);
485 dy00 = _mm_sub_ps(iy0,jy0);
486 dz00 = _mm_sub_ps(iz0,jz0);
487 dx10 = _mm_sub_ps(ix1,jx0);
488 dy10 = _mm_sub_ps(iy1,jy0);
489 dz10 = _mm_sub_ps(iz1,jz0);
490 dx20 = _mm_sub_ps(ix2,jx0);
491 dy20 = _mm_sub_ps(iy2,jy0);
492 dz20 = _mm_sub_ps(iz2,jz0);
493 dx30 = _mm_sub_ps(ix3,jx0);
494 dy30 = _mm_sub_ps(iy3,jy0);
495 dz30 = _mm_sub_ps(iz3,jz0);
497 /* Calculate squared distance and things based on it */
498 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
499 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
500 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
501 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
503 rinv00 = sse41_invsqrt_f(rsq00);
504 rinv10 = sse41_invsqrt_f(rsq10);
505 rinv20 = sse41_invsqrt_f(rsq20);
506 rinv30 = sse41_invsqrt_f(rsq30);
508 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
509 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
510 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
511 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
513 /* Load parameters for j particles */
514 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
515 charge+jnrC+0,charge+jnrD+0);
516 vdwjidx0A = 2*vdwtype[jnrA+0];
517 vdwjidx0B = 2*vdwtype[jnrB+0];
518 vdwjidx0C = 2*vdwtype[jnrC+0];
519 vdwjidx0D = 2*vdwtype[jnrD+0];
521 fjx0 = _mm_setzero_ps();
522 fjy0 = _mm_setzero_ps();
523 fjz0 = _mm_setzero_ps();
525 /**************************
526 * CALCULATE INTERACTIONS *
527 **************************/
529 if (gmx_mm_any_lt(rsq00,rcutoff2))
532 r00 = _mm_mul_ps(rsq00,rinv00);
533 r00 = _mm_andnot_ps(dummy_mask,r00);
535 /* Compute parameters for interactions between i and j atoms */
536 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
537 vdwparam+vdwioffset0+vdwjidx0B,
538 vdwparam+vdwioffset0+vdwjidx0C,
539 vdwparam+vdwioffset0+vdwjidx0D,
542 /* LENNARD-JONES DISPERSION/REPULSION */
544 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
545 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
546 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
547 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
548 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
550 d = _mm_sub_ps(r00,rswitch);
551 d = _mm_max_ps(d,_mm_setzero_ps());
552 d2 = _mm_mul_ps(d,d);
553 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
555 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
557 /* Evaluate switch function */
558 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
559 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
560 vvdw = _mm_mul_ps(vvdw,sw);
561 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
563 /* Update potential sum for this i atom from the interaction with this j atom. */
564 vvdw = _mm_and_ps(vvdw,cutoff_mask);
565 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
566 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
570 fscal = _mm_and_ps(fscal,cutoff_mask);
572 fscal = _mm_andnot_ps(dummy_mask,fscal);
574 /* Calculate temporary vectorial force */
575 tx = _mm_mul_ps(fscal,dx00);
576 ty = _mm_mul_ps(fscal,dy00);
577 tz = _mm_mul_ps(fscal,dz00);
579 /* Update vectorial force */
580 fix0 = _mm_add_ps(fix0,tx);
581 fiy0 = _mm_add_ps(fiy0,ty);
582 fiz0 = _mm_add_ps(fiz0,tz);
584 fjx0 = _mm_add_ps(fjx0,tx);
585 fjy0 = _mm_add_ps(fjy0,ty);
586 fjz0 = _mm_add_ps(fjz0,tz);
590 /**************************
591 * CALCULATE INTERACTIONS *
592 **************************/
594 if (gmx_mm_any_lt(rsq10,rcutoff2))
597 /* Compute parameters for interactions between i and j atoms */
598 qq10 = _mm_mul_ps(iq1,jq0);
600 /* REACTION-FIELD ELECTROSTATICS */
601 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
602 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
604 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
606 /* Update potential sum for this i atom from the interaction with this j atom. */
607 velec = _mm_and_ps(velec,cutoff_mask);
608 velec = _mm_andnot_ps(dummy_mask,velec);
609 velecsum = _mm_add_ps(velecsum,velec);
613 fscal = _mm_and_ps(fscal,cutoff_mask);
615 fscal = _mm_andnot_ps(dummy_mask,fscal);
617 /* Calculate temporary vectorial force */
618 tx = _mm_mul_ps(fscal,dx10);
619 ty = _mm_mul_ps(fscal,dy10);
620 tz = _mm_mul_ps(fscal,dz10);
622 /* Update vectorial force */
623 fix1 = _mm_add_ps(fix1,tx);
624 fiy1 = _mm_add_ps(fiy1,ty);
625 fiz1 = _mm_add_ps(fiz1,tz);
627 fjx0 = _mm_add_ps(fjx0,tx);
628 fjy0 = _mm_add_ps(fjy0,ty);
629 fjz0 = _mm_add_ps(fjz0,tz);
633 /**************************
634 * CALCULATE INTERACTIONS *
635 **************************/
637 if (gmx_mm_any_lt(rsq20,rcutoff2))
640 /* Compute parameters for interactions between i and j atoms */
641 qq20 = _mm_mul_ps(iq2,jq0);
643 /* REACTION-FIELD ELECTROSTATICS */
644 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
645 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
647 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
649 /* Update potential sum for this i atom from the interaction with this j atom. */
650 velec = _mm_and_ps(velec,cutoff_mask);
651 velec = _mm_andnot_ps(dummy_mask,velec);
652 velecsum = _mm_add_ps(velecsum,velec);
656 fscal = _mm_and_ps(fscal,cutoff_mask);
658 fscal = _mm_andnot_ps(dummy_mask,fscal);
660 /* Calculate temporary vectorial force */
661 tx = _mm_mul_ps(fscal,dx20);
662 ty = _mm_mul_ps(fscal,dy20);
663 tz = _mm_mul_ps(fscal,dz20);
665 /* Update vectorial force */
666 fix2 = _mm_add_ps(fix2,tx);
667 fiy2 = _mm_add_ps(fiy2,ty);
668 fiz2 = _mm_add_ps(fiz2,tz);
670 fjx0 = _mm_add_ps(fjx0,tx);
671 fjy0 = _mm_add_ps(fjy0,ty);
672 fjz0 = _mm_add_ps(fjz0,tz);
676 /**************************
677 * CALCULATE INTERACTIONS *
678 **************************/
680 if (gmx_mm_any_lt(rsq30,rcutoff2))
683 /* Compute parameters for interactions between i and j atoms */
684 qq30 = _mm_mul_ps(iq3,jq0);
686 /* REACTION-FIELD ELECTROSTATICS */
687 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
688 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
690 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
692 /* Update potential sum for this i atom from the interaction with this j atom. */
693 velec = _mm_and_ps(velec,cutoff_mask);
694 velec = _mm_andnot_ps(dummy_mask,velec);
695 velecsum = _mm_add_ps(velecsum,velec);
699 fscal = _mm_and_ps(fscal,cutoff_mask);
701 fscal = _mm_andnot_ps(dummy_mask,fscal);
703 /* Calculate temporary vectorial force */
704 tx = _mm_mul_ps(fscal,dx30);
705 ty = _mm_mul_ps(fscal,dy30);
706 tz = _mm_mul_ps(fscal,dz30);
708 /* Update vectorial force */
709 fix3 = _mm_add_ps(fix3,tx);
710 fiy3 = _mm_add_ps(fiy3,ty);
711 fiz3 = _mm_add_ps(fiz3,tz);
713 fjx0 = _mm_add_ps(fjx0,tx);
714 fjy0 = _mm_add_ps(fjy0,ty);
715 fjz0 = _mm_add_ps(fjz0,tz);
719 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
720 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
721 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
722 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
724 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
726 /* Inner loop uses 168 flops */
729 /* End of innermost loop */
731 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
732 f+i_coord_offset,fshift+i_shift_offset);
735 /* Update potential energies */
736 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
737 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
739 /* Increment number of inner iterations */
740 inneriter += j_index_end - j_index_start;
742 /* Outer loop uses 26 flops */
745 /* Increment number of outer iterations */
748 /* Update outer/inner flops */
750 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*168);
753 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse4_1_single
754 * Electrostatics interaction: ReactionField
755 * VdW interaction: LennardJones
756 * Geometry: Water4-Particle
757 * Calculate force/pot: Force
760 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse4_1_single
761 (t_nblist * gmx_restrict nlist,
762 rvec * gmx_restrict xx,
763 rvec * gmx_restrict ff,
764 struct t_forcerec * gmx_restrict fr,
765 t_mdatoms * gmx_restrict mdatoms,
766 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
767 t_nrnb * gmx_restrict nrnb)
769 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
770 * just 0 for non-waters.
771 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
772 * jnr indices corresponding to data put in the four positions in the SIMD register.
774 int i_shift_offset,i_coord_offset,outeriter,inneriter;
775 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
776 int jnrA,jnrB,jnrC,jnrD;
777 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
778 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
779 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
781 real *shiftvec,*fshift,*x,*f;
782 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
784 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
786 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
788 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
790 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
792 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
793 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
794 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
795 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
796 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
797 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
798 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
799 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
802 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
805 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
806 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
807 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
808 real rswitch_scalar,d_scalar;
809 __m128 dummy_mask,cutoff_mask;
810 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
811 __m128 one = _mm_set1_ps(1.0);
812 __m128 two = _mm_set1_ps(2.0);
818 jindex = nlist->jindex;
820 shiftidx = nlist->shift;
822 shiftvec = fr->shift_vec[0];
823 fshift = fr->fshift[0];
824 facel = _mm_set1_ps(fr->ic->epsfac);
825 charge = mdatoms->chargeA;
826 krf = _mm_set1_ps(fr->ic->k_rf);
827 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
828 crf = _mm_set1_ps(fr->ic->c_rf);
829 nvdwtype = fr->ntype;
831 vdwtype = mdatoms->typeA;
833 /* Setup water-specific parameters */
834 inr = nlist->iinr[0];
835 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
836 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
837 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
838 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
840 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
841 rcutoff_scalar = fr->ic->rcoulomb;
842 rcutoff = _mm_set1_ps(rcutoff_scalar);
843 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
845 rswitch_scalar = fr->ic->rvdw_switch;
846 rswitch = _mm_set1_ps(rswitch_scalar);
847 /* Setup switch parameters */
848 d_scalar = rcutoff_scalar-rswitch_scalar;
849 d = _mm_set1_ps(d_scalar);
850 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
851 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
852 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
853 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
854 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
855 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
857 /* Avoid stupid compiler warnings */
858 jnrA = jnrB = jnrC = jnrD = 0;
867 for(iidx=0;iidx<4*DIM;iidx++)
872 /* Start outer loop over neighborlists */
873 for(iidx=0; iidx<nri; iidx++)
875 /* Load shift vector for this list */
876 i_shift_offset = DIM*shiftidx[iidx];
878 /* Load limits for loop over neighbors */
879 j_index_start = jindex[iidx];
880 j_index_end = jindex[iidx+1];
882 /* Get outer coordinate index */
884 i_coord_offset = DIM*inr;
886 /* Load i particle coords and add shift vector */
887 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
888 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
890 fix0 = _mm_setzero_ps();
891 fiy0 = _mm_setzero_ps();
892 fiz0 = _mm_setzero_ps();
893 fix1 = _mm_setzero_ps();
894 fiy1 = _mm_setzero_ps();
895 fiz1 = _mm_setzero_ps();
896 fix2 = _mm_setzero_ps();
897 fiy2 = _mm_setzero_ps();
898 fiz2 = _mm_setzero_ps();
899 fix3 = _mm_setzero_ps();
900 fiy3 = _mm_setzero_ps();
901 fiz3 = _mm_setzero_ps();
903 /* Start inner kernel loop */
904 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
907 /* Get j neighbor index, and coordinate index */
912 j_coord_offsetA = DIM*jnrA;
913 j_coord_offsetB = DIM*jnrB;
914 j_coord_offsetC = DIM*jnrC;
915 j_coord_offsetD = DIM*jnrD;
917 /* load j atom coordinates */
918 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
919 x+j_coord_offsetC,x+j_coord_offsetD,
922 /* Calculate displacement vector */
923 dx00 = _mm_sub_ps(ix0,jx0);
924 dy00 = _mm_sub_ps(iy0,jy0);
925 dz00 = _mm_sub_ps(iz0,jz0);
926 dx10 = _mm_sub_ps(ix1,jx0);
927 dy10 = _mm_sub_ps(iy1,jy0);
928 dz10 = _mm_sub_ps(iz1,jz0);
929 dx20 = _mm_sub_ps(ix2,jx0);
930 dy20 = _mm_sub_ps(iy2,jy0);
931 dz20 = _mm_sub_ps(iz2,jz0);
932 dx30 = _mm_sub_ps(ix3,jx0);
933 dy30 = _mm_sub_ps(iy3,jy0);
934 dz30 = _mm_sub_ps(iz3,jz0);
936 /* Calculate squared distance and things based on it */
937 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
938 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
939 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
940 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
942 rinv00 = sse41_invsqrt_f(rsq00);
943 rinv10 = sse41_invsqrt_f(rsq10);
944 rinv20 = sse41_invsqrt_f(rsq20);
945 rinv30 = sse41_invsqrt_f(rsq30);
947 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
948 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
949 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
950 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
952 /* Load parameters for j particles */
953 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
954 charge+jnrC+0,charge+jnrD+0);
955 vdwjidx0A = 2*vdwtype[jnrA+0];
956 vdwjidx0B = 2*vdwtype[jnrB+0];
957 vdwjidx0C = 2*vdwtype[jnrC+0];
958 vdwjidx0D = 2*vdwtype[jnrD+0];
960 fjx0 = _mm_setzero_ps();
961 fjy0 = _mm_setzero_ps();
962 fjz0 = _mm_setzero_ps();
964 /**************************
965 * CALCULATE INTERACTIONS *
966 **************************/
968 if (gmx_mm_any_lt(rsq00,rcutoff2))
971 r00 = _mm_mul_ps(rsq00,rinv00);
973 /* Compute parameters for interactions between i and j atoms */
974 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
975 vdwparam+vdwioffset0+vdwjidx0B,
976 vdwparam+vdwioffset0+vdwjidx0C,
977 vdwparam+vdwioffset0+vdwjidx0D,
980 /* LENNARD-JONES DISPERSION/REPULSION */
982 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
983 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
984 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
985 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
986 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
988 d = _mm_sub_ps(r00,rswitch);
989 d = _mm_max_ps(d,_mm_setzero_ps());
990 d2 = _mm_mul_ps(d,d);
991 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
993 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
995 /* Evaluate switch function */
996 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
997 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
998 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1002 fscal = _mm_and_ps(fscal,cutoff_mask);
1004 /* Calculate temporary vectorial force */
1005 tx = _mm_mul_ps(fscal,dx00);
1006 ty = _mm_mul_ps(fscal,dy00);
1007 tz = _mm_mul_ps(fscal,dz00);
1009 /* Update vectorial force */
1010 fix0 = _mm_add_ps(fix0,tx);
1011 fiy0 = _mm_add_ps(fiy0,ty);
1012 fiz0 = _mm_add_ps(fiz0,tz);
1014 fjx0 = _mm_add_ps(fjx0,tx);
1015 fjy0 = _mm_add_ps(fjy0,ty);
1016 fjz0 = _mm_add_ps(fjz0,tz);
1020 /**************************
1021 * CALCULATE INTERACTIONS *
1022 **************************/
1024 if (gmx_mm_any_lt(rsq10,rcutoff2))
1027 /* Compute parameters for interactions between i and j atoms */
1028 qq10 = _mm_mul_ps(iq1,jq0);
1030 /* REACTION-FIELD ELECTROSTATICS */
1031 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1033 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1037 fscal = _mm_and_ps(fscal,cutoff_mask);
1039 /* Calculate temporary vectorial force */
1040 tx = _mm_mul_ps(fscal,dx10);
1041 ty = _mm_mul_ps(fscal,dy10);
1042 tz = _mm_mul_ps(fscal,dz10);
1044 /* Update vectorial force */
1045 fix1 = _mm_add_ps(fix1,tx);
1046 fiy1 = _mm_add_ps(fiy1,ty);
1047 fiz1 = _mm_add_ps(fiz1,tz);
1049 fjx0 = _mm_add_ps(fjx0,tx);
1050 fjy0 = _mm_add_ps(fjy0,ty);
1051 fjz0 = _mm_add_ps(fjz0,tz);
1055 /**************************
1056 * CALCULATE INTERACTIONS *
1057 **************************/
1059 if (gmx_mm_any_lt(rsq20,rcutoff2))
1062 /* Compute parameters for interactions between i and j atoms */
1063 qq20 = _mm_mul_ps(iq2,jq0);
1065 /* REACTION-FIELD ELECTROSTATICS */
1066 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1068 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1072 fscal = _mm_and_ps(fscal,cutoff_mask);
1074 /* Calculate temporary vectorial force */
1075 tx = _mm_mul_ps(fscal,dx20);
1076 ty = _mm_mul_ps(fscal,dy20);
1077 tz = _mm_mul_ps(fscal,dz20);
1079 /* Update vectorial force */
1080 fix2 = _mm_add_ps(fix2,tx);
1081 fiy2 = _mm_add_ps(fiy2,ty);
1082 fiz2 = _mm_add_ps(fiz2,tz);
1084 fjx0 = _mm_add_ps(fjx0,tx);
1085 fjy0 = _mm_add_ps(fjy0,ty);
1086 fjz0 = _mm_add_ps(fjz0,tz);
1090 /**************************
1091 * CALCULATE INTERACTIONS *
1092 **************************/
1094 if (gmx_mm_any_lt(rsq30,rcutoff2))
1097 /* Compute parameters for interactions between i and j atoms */
1098 qq30 = _mm_mul_ps(iq3,jq0);
1100 /* REACTION-FIELD ELECTROSTATICS */
1101 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1103 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1107 fscal = _mm_and_ps(fscal,cutoff_mask);
1109 /* Calculate temporary vectorial force */
1110 tx = _mm_mul_ps(fscal,dx30);
1111 ty = _mm_mul_ps(fscal,dy30);
1112 tz = _mm_mul_ps(fscal,dz30);
1114 /* Update vectorial force */
1115 fix3 = _mm_add_ps(fix3,tx);
1116 fiy3 = _mm_add_ps(fiy3,ty);
1117 fiz3 = _mm_add_ps(fiz3,tz);
1119 fjx0 = _mm_add_ps(fjx0,tx);
1120 fjy0 = _mm_add_ps(fjy0,ty);
1121 fjz0 = _mm_add_ps(fjz0,tz);
1125 fjptrA = f+j_coord_offsetA;
1126 fjptrB = f+j_coord_offsetB;
1127 fjptrC = f+j_coord_offsetC;
1128 fjptrD = f+j_coord_offsetD;
1130 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1132 /* Inner loop uses 146 flops */
1135 if(jidx<j_index_end)
1138 /* Get j neighbor index, and coordinate index */
1139 jnrlistA = jjnr[jidx];
1140 jnrlistB = jjnr[jidx+1];
1141 jnrlistC = jjnr[jidx+2];
1142 jnrlistD = jjnr[jidx+3];
1143 /* Sign of each element will be negative for non-real atoms.
1144 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1145 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1147 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1148 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1149 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1150 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1151 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1152 j_coord_offsetA = DIM*jnrA;
1153 j_coord_offsetB = DIM*jnrB;
1154 j_coord_offsetC = DIM*jnrC;
1155 j_coord_offsetD = DIM*jnrD;
1157 /* load j atom coordinates */
1158 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1159 x+j_coord_offsetC,x+j_coord_offsetD,
1162 /* Calculate displacement vector */
1163 dx00 = _mm_sub_ps(ix0,jx0);
1164 dy00 = _mm_sub_ps(iy0,jy0);
1165 dz00 = _mm_sub_ps(iz0,jz0);
1166 dx10 = _mm_sub_ps(ix1,jx0);
1167 dy10 = _mm_sub_ps(iy1,jy0);
1168 dz10 = _mm_sub_ps(iz1,jz0);
1169 dx20 = _mm_sub_ps(ix2,jx0);
1170 dy20 = _mm_sub_ps(iy2,jy0);
1171 dz20 = _mm_sub_ps(iz2,jz0);
1172 dx30 = _mm_sub_ps(ix3,jx0);
1173 dy30 = _mm_sub_ps(iy3,jy0);
1174 dz30 = _mm_sub_ps(iz3,jz0);
1176 /* Calculate squared distance and things based on it */
1177 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1178 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1179 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1180 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1182 rinv00 = sse41_invsqrt_f(rsq00);
1183 rinv10 = sse41_invsqrt_f(rsq10);
1184 rinv20 = sse41_invsqrt_f(rsq20);
1185 rinv30 = sse41_invsqrt_f(rsq30);
1187 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1188 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1189 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1190 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
1192 /* Load parameters for j particles */
1193 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1194 charge+jnrC+0,charge+jnrD+0);
1195 vdwjidx0A = 2*vdwtype[jnrA+0];
1196 vdwjidx0B = 2*vdwtype[jnrB+0];
1197 vdwjidx0C = 2*vdwtype[jnrC+0];
1198 vdwjidx0D = 2*vdwtype[jnrD+0];
1200 fjx0 = _mm_setzero_ps();
1201 fjy0 = _mm_setzero_ps();
1202 fjz0 = _mm_setzero_ps();
1204 /**************************
1205 * CALCULATE INTERACTIONS *
1206 **************************/
1208 if (gmx_mm_any_lt(rsq00,rcutoff2))
1211 r00 = _mm_mul_ps(rsq00,rinv00);
1212 r00 = _mm_andnot_ps(dummy_mask,r00);
1214 /* Compute parameters for interactions between i and j atoms */
1215 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1216 vdwparam+vdwioffset0+vdwjidx0B,
1217 vdwparam+vdwioffset0+vdwjidx0C,
1218 vdwparam+vdwioffset0+vdwjidx0D,
1221 /* LENNARD-JONES DISPERSION/REPULSION */
1223 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1224 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
1225 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1226 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1227 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1229 d = _mm_sub_ps(r00,rswitch);
1230 d = _mm_max_ps(d,_mm_setzero_ps());
1231 d2 = _mm_mul_ps(d,d);
1232 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1234 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1236 /* Evaluate switch function */
1237 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1238 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1239 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1243 fscal = _mm_and_ps(fscal,cutoff_mask);
1245 fscal = _mm_andnot_ps(dummy_mask,fscal);
1247 /* Calculate temporary vectorial force */
1248 tx = _mm_mul_ps(fscal,dx00);
1249 ty = _mm_mul_ps(fscal,dy00);
1250 tz = _mm_mul_ps(fscal,dz00);
1252 /* Update vectorial force */
1253 fix0 = _mm_add_ps(fix0,tx);
1254 fiy0 = _mm_add_ps(fiy0,ty);
1255 fiz0 = _mm_add_ps(fiz0,tz);
1257 fjx0 = _mm_add_ps(fjx0,tx);
1258 fjy0 = _mm_add_ps(fjy0,ty);
1259 fjz0 = _mm_add_ps(fjz0,tz);
1263 /**************************
1264 * CALCULATE INTERACTIONS *
1265 **************************/
1267 if (gmx_mm_any_lt(rsq10,rcutoff2))
1270 /* Compute parameters for interactions between i and j atoms */
1271 qq10 = _mm_mul_ps(iq1,jq0);
1273 /* REACTION-FIELD ELECTROSTATICS */
1274 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1276 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1280 fscal = _mm_and_ps(fscal,cutoff_mask);
1282 fscal = _mm_andnot_ps(dummy_mask,fscal);
1284 /* Calculate temporary vectorial force */
1285 tx = _mm_mul_ps(fscal,dx10);
1286 ty = _mm_mul_ps(fscal,dy10);
1287 tz = _mm_mul_ps(fscal,dz10);
1289 /* Update vectorial force */
1290 fix1 = _mm_add_ps(fix1,tx);
1291 fiy1 = _mm_add_ps(fiy1,ty);
1292 fiz1 = _mm_add_ps(fiz1,tz);
1294 fjx0 = _mm_add_ps(fjx0,tx);
1295 fjy0 = _mm_add_ps(fjy0,ty);
1296 fjz0 = _mm_add_ps(fjz0,tz);
1300 /**************************
1301 * CALCULATE INTERACTIONS *
1302 **************************/
1304 if (gmx_mm_any_lt(rsq20,rcutoff2))
1307 /* Compute parameters for interactions between i and j atoms */
1308 qq20 = _mm_mul_ps(iq2,jq0);
1310 /* REACTION-FIELD ELECTROSTATICS */
1311 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1313 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1317 fscal = _mm_and_ps(fscal,cutoff_mask);
1319 fscal = _mm_andnot_ps(dummy_mask,fscal);
1321 /* Calculate temporary vectorial force */
1322 tx = _mm_mul_ps(fscal,dx20);
1323 ty = _mm_mul_ps(fscal,dy20);
1324 tz = _mm_mul_ps(fscal,dz20);
1326 /* Update vectorial force */
1327 fix2 = _mm_add_ps(fix2,tx);
1328 fiy2 = _mm_add_ps(fiy2,ty);
1329 fiz2 = _mm_add_ps(fiz2,tz);
1331 fjx0 = _mm_add_ps(fjx0,tx);
1332 fjy0 = _mm_add_ps(fjy0,ty);
1333 fjz0 = _mm_add_ps(fjz0,tz);
1337 /**************************
1338 * CALCULATE INTERACTIONS *
1339 **************************/
1341 if (gmx_mm_any_lt(rsq30,rcutoff2))
1344 /* Compute parameters for interactions between i and j atoms */
1345 qq30 = _mm_mul_ps(iq3,jq0);
1347 /* REACTION-FIELD ELECTROSTATICS */
1348 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1350 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1354 fscal = _mm_and_ps(fscal,cutoff_mask);
1356 fscal = _mm_andnot_ps(dummy_mask,fscal);
1358 /* Calculate temporary vectorial force */
1359 tx = _mm_mul_ps(fscal,dx30);
1360 ty = _mm_mul_ps(fscal,dy30);
1361 tz = _mm_mul_ps(fscal,dz30);
1363 /* Update vectorial force */
1364 fix3 = _mm_add_ps(fix3,tx);
1365 fiy3 = _mm_add_ps(fiy3,ty);
1366 fiz3 = _mm_add_ps(fiz3,tz);
1368 fjx0 = _mm_add_ps(fjx0,tx);
1369 fjy0 = _mm_add_ps(fjy0,ty);
1370 fjz0 = _mm_add_ps(fjz0,tz);
1374 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1375 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1376 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1377 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1379 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1381 /* Inner loop uses 147 flops */
1384 /* End of innermost loop */
1386 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1387 f+i_coord_offset,fshift+i_shift_offset);
1389 /* Increment number of inner iterations */
1390 inneriter += j_index_end - j_index_start;
1392 /* Outer loop uses 24 flops */
1395 /* Increment number of outer iterations */
1398 /* Update outer/inner flops */
1400 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);