2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, 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 sse2_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_sse2_single.h"
50 #include "kernelutil_x86_sse2_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_single
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: LennardJones
56 * Geometry: Water3-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_single
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
79 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
81 real *shiftvec,*fshift,*x,*f;
82 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
84 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
91 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
92 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
93 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
94 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
95 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
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 dummy_mask,cutoff_mask;
105 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
106 __m128 one = _mm_set1_ps(1.0);
107 __m128 two = _mm_set1_ps(2.0);
113 jindex = nlist->jindex;
115 shiftidx = nlist->shift;
117 shiftvec = fr->shift_vec[0];
118 fshift = fr->fshift[0];
119 facel = _mm_set1_ps(fr->epsfac);
120 charge = mdatoms->chargeA;
121 krf = _mm_set1_ps(fr->ic->k_rf);
122 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
123 crf = _mm_set1_ps(fr->ic->c_rf);
124 nvdwtype = fr->ntype;
126 vdwtype = mdatoms->typeA;
128 /* Setup water-specific parameters */
129 inr = nlist->iinr[0];
130 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
131 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
132 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
133 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
135 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
136 rcutoff_scalar = fr->rcoulomb;
137 rcutoff = _mm_set1_ps(rcutoff_scalar);
138 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
140 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
141 rvdw = _mm_set1_ps(fr->rvdw);
143 /* Avoid stupid compiler warnings */
144 jnrA = jnrB = jnrC = jnrD = 0;
153 for(iidx=0;iidx<4*DIM;iidx++)
158 /* Start outer loop over neighborlists */
159 for(iidx=0; iidx<nri; iidx++)
161 /* Load shift vector for this list */
162 i_shift_offset = DIM*shiftidx[iidx];
164 /* Load limits for loop over neighbors */
165 j_index_start = jindex[iidx];
166 j_index_end = jindex[iidx+1];
168 /* Get outer coordinate index */
170 i_coord_offset = DIM*inr;
172 /* Load i particle coords and add shift vector */
173 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
174 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
176 fix0 = _mm_setzero_ps();
177 fiy0 = _mm_setzero_ps();
178 fiz0 = _mm_setzero_ps();
179 fix1 = _mm_setzero_ps();
180 fiy1 = _mm_setzero_ps();
181 fiz1 = _mm_setzero_ps();
182 fix2 = _mm_setzero_ps();
183 fiy2 = _mm_setzero_ps();
184 fiz2 = _mm_setzero_ps();
186 /* Reset potential sums */
187 velecsum = _mm_setzero_ps();
188 vvdwsum = _mm_setzero_ps();
190 /* Start inner kernel loop */
191 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
194 /* Get j neighbor index, and coordinate index */
199 j_coord_offsetA = DIM*jnrA;
200 j_coord_offsetB = DIM*jnrB;
201 j_coord_offsetC = DIM*jnrC;
202 j_coord_offsetD = DIM*jnrD;
204 /* load j atom coordinates */
205 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
206 x+j_coord_offsetC,x+j_coord_offsetD,
209 /* Calculate displacement vector */
210 dx00 = _mm_sub_ps(ix0,jx0);
211 dy00 = _mm_sub_ps(iy0,jy0);
212 dz00 = _mm_sub_ps(iz0,jz0);
213 dx10 = _mm_sub_ps(ix1,jx0);
214 dy10 = _mm_sub_ps(iy1,jy0);
215 dz10 = _mm_sub_ps(iz1,jz0);
216 dx20 = _mm_sub_ps(ix2,jx0);
217 dy20 = _mm_sub_ps(iy2,jy0);
218 dz20 = _mm_sub_ps(iz2,jz0);
220 /* Calculate squared distance and things based on it */
221 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
222 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
223 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
225 rinv00 = gmx_mm_invsqrt_ps(rsq00);
226 rinv10 = gmx_mm_invsqrt_ps(rsq10);
227 rinv20 = gmx_mm_invsqrt_ps(rsq20);
229 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
230 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
231 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
233 /* Load parameters for j particles */
234 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
235 charge+jnrC+0,charge+jnrD+0);
236 vdwjidx0A = 2*vdwtype[jnrA+0];
237 vdwjidx0B = 2*vdwtype[jnrB+0];
238 vdwjidx0C = 2*vdwtype[jnrC+0];
239 vdwjidx0D = 2*vdwtype[jnrD+0];
241 fjx0 = _mm_setzero_ps();
242 fjy0 = _mm_setzero_ps();
243 fjz0 = _mm_setzero_ps();
245 /**************************
246 * CALCULATE INTERACTIONS *
247 **************************/
249 if (gmx_mm_any_lt(rsq00,rcutoff2))
252 /* Compute parameters for interactions between i and j atoms */
253 qq00 = _mm_mul_ps(iq0,jq0);
254 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
255 vdwparam+vdwioffset0+vdwjidx0B,
256 vdwparam+vdwioffset0+vdwjidx0C,
257 vdwparam+vdwioffset0+vdwjidx0D,
260 /* REACTION-FIELD ELECTROSTATICS */
261 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
262 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
264 /* LENNARD-JONES DISPERSION/REPULSION */
266 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
267 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
268 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
269 vvdw = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
270 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
271 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
273 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
275 /* Update potential sum for this i atom from the interaction with this j atom. */
276 velec = _mm_and_ps(velec,cutoff_mask);
277 velecsum = _mm_add_ps(velecsum,velec);
278 vvdw = _mm_and_ps(vvdw,cutoff_mask);
279 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
281 fscal = _mm_add_ps(felec,fvdw);
283 fscal = _mm_and_ps(fscal,cutoff_mask);
285 /* Calculate temporary vectorial force */
286 tx = _mm_mul_ps(fscal,dx00);
287 ty = _mm_mul_ps(fscal,dy00);
288 tz = _mm_mul_ps(fscal,dz00);
290 /* Update vectorial force */
291 fix0 = _mm_add_ps(fix0,tx);
292 fiy0 = _mm_add_ps(fiy0,ty);
293 fiz0 = _mm_add_ps(fiz0,tz);
295 fjx0 = _mm_add_ps(fjx0,tx);
296 fjy0 = _mm_add_ps(fjy0,ty);
297 fjz0 = _mm_add_ps(fjz0,tz);
301 /**************************
302 * CALCULATE INTERACTIONS *
303 **************************/
305 if (gmx_mm_any_lt(rsq10,rcutoff2))
308 /* Compute parameters for interactions between i and j atoms */
309 qq10 = _mm_mul_ps(iq1,jq0);
311 /* REACTION-FIELD ELECTROSTATICS */
312 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
313 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
315 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
317 /* Update potential sum for this i atom from the interaction with this j atom. */
318 velec = _mm_and_ps(velec,cutoff_mask);
319 velecsum = _mm_add_ps(velecsum,velec);
323 fscal = _mm_and_ps(fscal,cutoff_mask);
325 /* Calculate temporary vectorial force */
326 tx = _mm_mul_ps(fscal,dx10);
327 ty = _mm_mul_ps(fscal,dy10);
328 tz = _mm_mul_ps(fscal,dz10);
330 /* Update vectorial force */
331 fix1 = _mm_add_ps(fix1,tx);
332 fiy1 = _mm_add_ps(fiy1,ty);
333 fiz1 = _mm_add_ps(fiz1,tz);
335 fjx0 = _mm_add_ps(fjx0,tx);
336 fjy0 = _mm_add_ps(fjy0,ty);
337 fjz0 = _mm_add_ps(fjz0,tz);
341 /**************************
342 * CALCULATE INTERACTIONS *
343 **************************/
345 if (gmx_mm_any_lt(rsq20,rcutoff2))
348 /* Compute parameters for interactions between i and j atoms */
349 qq20 = _mm_mul_ps(iq2,jq0);
351 /* REACTION-FIELD ELECTROSTATICS */
352 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
353 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
355 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
357 /* Update potential sum for this i atom from the interaction with this j atom. */
358 velec = _mm_and_ps(velec,cutoff_mask);
359 velecsum = _mm_add_ps(velecsum,velec);
363 fscal = _mm_and_ps(fscal,cutoff_mask);
365 /* Calculate temporary vectorial force */
366 tx = _mm_mul_ps(fscal,dx20);
367 ty = _mm_mul_ps(fscal,dy20);
368 tz = _mm_mul_ps(fscal,dz20);
370 /* Update vectorial force */
371 fix2 = _mm_add_ps(fix2,tx);
372 fiy2 = _mm_add_ps(fiy2,ty);
373 fiz2 = _mm_add_ps(fiz2,tz);
375 fjx0 = _mm_add_ps(fjx0,tx);
376 fjy0 = _mm_add_ps(fjy0,ty);
377 fjz0 = _mm_add_ps(fjz0,tz);
381 fjptrA = f+j_coord_offsetA;
382 fjptrB = f+j_coord_offsetB;
383 fjptrC = f+j_coord_offsetC;
384 fjptrD = f+j_coord_offsetD;
386 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
388 /* Inner loop uses 126 flops */
394 /* Get j neighbor index, and coordinate index */
395 jnrlistA = jjnr[jidx];
396 jnrlistB = jjnr[jidx+1];
397 jnrlistC = jjnr[jidx+2];
398 jnrlistD = jjnr[jidx+3];
399 /* Sign of each element will be negative for non-real atoms.
400 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
401 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
403 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
404 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
405 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
406 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
407 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
408 j_coord_offsetA = DIM*jnrA;
409 j_coord_offsetB = DIM*jnrB;
410 j_coord_offsetC = DIM*jnrC;
411 j_coord_offsetD = DIM*jnrD;
413 /* load j atom coordinates */
414 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
415 x+j_coord_offsetC,x+j_coord_offsetD,
418 /* Calculate displacement vector */
419 dx00 = _mm_sub_ps(ix0,jx0);
420 dy00 = _mm_sub_ps(iy0,jy0);
421 dz00 = _mm_sub_ps(iz0,jz0);
422 dx10 = _mm_sub_ps(ix1,jx0);
423 dy10 = _mm_sub_ps(iy1,jy0);
424 dz10 = _mm_sub_ps(iz1,jz0);
425 dx20 = _mm_sub_ps(ix2,jx0);
426 dy20 = _mm_sub_ps(iy2,jy0);
427 dz20 = _mm_sub_ps(iz2,jz0);
429 /* Calculate squared distance and things based on it */
430 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
431 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
432 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
434 rinv00 = gmx_mm_invsqrt_ps(rsq00);
435 rinv10 = gmx_mm_invsqrt_ps(rsq10);
436 rinv20 = gmx_mm_invsqrt_ps(rsq20);
438 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
439 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
440 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
442 /* Load parameters for j particles */
443 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
444 charge+jnrC+0,charge+jnrD+0);
445 vdwjidx0A = 2*vdwtype[jnrA+0];
446 vdwjidx0B = 2*vdwtype[jnrB+0];
447 vdwjidx0C = 2*vdwtype[jnrC+0];
448 vdwjidx0D = 2*vdwtype[jnrD+0];
450 fjx0 = _mm_setzero_ps();
451 fjy0 = _mm_setzero_ps();
452 fjz0 = _mm_setzero_ps();
454 /**************************
455 * CALCULATE INTERACTIONS *
456 **************************/
458 if (gmx_mm_any_lt(rsq00,rcutoff2))
461 /* Compute parameters for interactions between i and j atoms */
462 qq00 = _mm_mul_ps(iq0,jq0);
463 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
464 vdwparam+vdwioffset0+vdwjidx0B,
465 vdwparam+vdwioffset0+vdwjidx0C,
466 vdwparam+vdwioffset0+vdwjidx0D,
469 /* REACTION-FIELD ELECTROSTATICS */
470 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
471 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
473 /* LENNARD-JONES DISPERSION/REPULSION */
475 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
476 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
477 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
478 vvdw = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
479 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
480 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
482 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
484 /* Update potential sum for this i atom from the interaction with this j atom. */
485 velec = _mm_and_ps(velec,cutoff_mask);
486 velec = _mm_andnot_ps(dummy_mask,velec);
487 velecsum = _mm_add_ps(velecsum,velec);
488 vvdw = _mm_and_ps(vvdw,cutoff_mask);
489 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
490 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
492 fscal = _mm_add_ps(felec,fvdw);
494 fscal = _mm_and_ps(fscal,cutoff_mask);
496 fscal = _mm_andnot_ps(dummy_mask,fscal);
498 /* Calculate temporary vectorial force */
499 tx = _mm_mul_ps(fscal,dx00);
500 ty = _mm_mul_ps(fscal,dy00);
501 tz = _mm_mul_ps(fscal,dz00);
503 /* Update vectorial force */
504 fix0 = _mm_add_ps(fix0,tx);
505 fiy0 = _mm_add_ps(fiy0,ty);
506 fiz0 = _mm_add_ps(fiz0,tz);
508 fjx0 = _mm_add_ps(fjx0,tx);
509 fjy0 = _mm_add_ps(fjy0,ty);
510 fjz0 = _mm_add_ps(fjz0,tz);
514 /**************************
515 * CALCULATE INTERACTIONS *
516 **************************/
518 if (gmx_mm_any_lt(rsq10,rcutoff2))
521 /* Compute parameters for interactions between i and j atoms */
522 qq10 = _mm_mul_ps(iq1,jq0);
524 /* REACTION-FIELD ELECTROSTATICS */
525 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
526 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
528 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
530 /* Update potential sum for this i atom from the interaction with this j atom. */
531 velec = _mm_and_ps(velec,cutoff_mask);
532 velec = _mm_andnot_ps(dummy_mask,velec);
533 velecsum = _mm_add_ps(velecsum,velec);
537 fscal = _mm_and_ps(fscal,cutoff_mask);
539 fscal = _mm_andnot_ps(dummy_mask,fscal);
541 /* Calculate temporary vectorial force */
542 tx = _mm_mul_ps(fscal,dx10);
543 ty = _mm_mul_ps(fscal,dy10);
544 tz = _mm_mul_ps(fscal,dz10);
546 /* Update vectorial force */
547 fix1 = _mm_add_ps(fix1,tx);
548 fiy1 = _mm_add_ps(fiy1,ty);
549 fiz1 = _mm_add_ps(fiz1,tz);
551 fjx0 = _mm_add_ps(fjx0,tx);
552 fjy0 = _mm_add_ps(fjy0,ty);
553 fjz0 = _mm_add_ps(fjz0,tz);
557 /**************************
558 * CALCULATE INTERACTIONS *
559 **************************/
561 if (gmx_mm_any_lt(rsq20,rcutoff2))
564 /* Compute parameters for interactions between i and j atoms */
565 qq20 = _mm_mul_ps(iq2,jq0);
567 /* REACTION-FIELD ELECTROSTATICS */
568 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
569 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
571 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
573 /* Update potential sum for this i atom from the interaction with this j atom. */
574 velec = _mm_and_ps(velec,cutoff_mask);
575 velec = _mm_andnot_ps(dummy_mask,velec);
576 velecsum = _mm_add_ps(velecsum,velec);
580 fscal = _mm_and_ps(fscal,cutoff_mask);
582 fscal = _mm_andnot_ps(dummy_mask,fscal);
584 /* Calculate temporary vectorial force */
585 tx = _mm_mul_ps(fscal,dx20);
586 ty = _mm_mul_ps(fscal,dy20);
587 tz = _mm_mul_ps(fscal,dz20);
589 /* Update vectorial force */
590 fix2 = _mm_add_ps(fix2,tx);
591 fiy2 = _mm_add_ps(fiy2,ty);
592 fiz2 = _mm_add_ps(fiz2,tz);
594 fjx0 = _mm_add_ps(fjx0,tx);
595 fjy0 = _mm_add_ps(fjy0,ty);
596 fjz0 = _mm_add_ps(fjz0,tz);
600 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
601 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
602 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
603 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
605 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
607 /* Inner loop uses 126 flops */
610 /* End of innermost loop */
612 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
613 f+i_coord_offset,fshift+i_shift_offset);
616 /* Update potential energies */
617 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
618 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
620 /* Increment number of inner iterations */
621 inneriter += j_index_end - j_index_start;
623 /* Outer loop uses 20 flops */
626 /* Increment number of outer iterations */
629 /* Update outer/inner flops */
631 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*126);
634 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_single
635 * Electrostatics interaction: ReactionField
636 * VdW interaction: LennardJones
637 * Geometry: Water3-Particle
638 * Calculate force/pot: Force
641 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_single
642 (t_nblist * gmx_restrict nlist,
643 rvec * gmx_restrict xx,
644 rvec * gmx_restrict ff,
645 t_forcerec * gmx_restrict fr,
646 t_mdatoms * gmx_restrict mdatoms,
647 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
648 t_nrnb * gmx_restrict nrnb)
650 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
651 * just 0 for non-waters.
652 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
653 * jnr indices corresponding to data put in the four positions in the SIMD register.
655 int i_shift_offset,i_coord_offset,outeriter,inneriter;
656 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
657 int jnrA,jnrB,jnrC,jnrD;
658 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
659 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
660 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
662 real *shiftvec,*fshift,*x,*f;
663 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
665 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
667 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
669 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
671 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
672 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
673 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
674 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
675 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
676 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
677 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
680 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
683 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
684 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
685 __m128 dummy_mask,cutoff_mask;
686 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
687 __m128 one = _mm_set1_ps(1.0);
688 __m128 two = _mm_set1_ps(2.0);
694 jindex = nlist->jindex;
696 shiftidx = nlist->shift;
698 shiftvec = fr->shift_vec[0];
699 fshift = fr->fshift[0];
700 facel = _mm_set1_ps(fr->epsfac);
701 charge = mdatoms->chargeA;
702 krf = _mm_set1_ps(fr->ic->k_rf);
703 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
704 crf = _mm_set1_ps(fr->ic->c_rf);
705 nvdwtype = fr->ntype;
707 vdwtype = mdatoms->typeA;
709 /* Setup water-specific parameters */
710 inr = nlist->iinr[0];
711 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
712 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
713 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
714 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
716 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
717 rcutoff_scalar = fr->rcoulomb;
718 rcutoff = _mm_set1_ps(rcutoff_scalar);
719 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
721 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
722 rvdw = _mm_set1_ps(fr->rvdw);
724 /* Avoid stupid compiler warnings */
725 jnrA = jnrB = jnrC = jnrD = 0;
734 for(iidx=0;iidx<4*DIM;iidx++)
739 /* Start outer loop over neighborlists */
740 for(iidx=0; iidx<nri; iidx++)
742 /* Load shift vector for this list */
743 i_shift_offset = DIM*shiftidx[iidx];
745 /* Load limits for loop over neighbors */
746 j_index_start = jindex[iidx];
747 j_index_end = jindex[iidx+1];
749 /* Get outer coordinate index */
751 i_coord_offset = DIM*inr;
753 /* Load i particle coords and add shift vector */
754 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
755 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
757 fix0 = _mm_setzero_ps();
758 fiy0 = _mm_setzero_ps();
759 fiz0 = _mm_setzero_ps();
760 fix1 = _mm_setzero_ps();
761 fiy1 = _mm_setzero_ps();
762 fiz1 = _mm_setzero_ps();
763 fix2 = _mm_setzero_ps();
764 fiy2 = _mm_setzero_ps();
765 fiz2 = _mm_setzero_ps();
767 /* Start inner kernel loop */
768 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
771 /* Get j neighbor index, and coordinate index */
776 j_coord_offsetA = DIM*jnrA;
777 j_coord_offsetB = DIM*jnrB;
778 j_coord_offsetC = DIM*jnrC;
779 j_coord_offsetD = DIM*jnrD;
781 /* load j atom coordinates */
782 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
783 x+j_coord_offsetC,x+j_coord_offsetD,
786 /* Calculate displacement vector */
787 dx00 = _mm_sub_ps(ix0,jx0);
788 dy00 = _mm_sub_ps(iy0,jy0);
789 dz00 = _mm_sub_ps(iz0,jz0);
790 dx10 = _mm_sub_ps(ix1,jx0);
791 dy10 = _mm_sub_ps(iy1,jy0);
792 dz10 = _mm_sub_ps(iz1,jz0);
793 dx20 = _mm_sub_ps(ix2,jx0);
794 dy20 = _mm_sub_ps(iy2,jy0);
795 dz20 = _mm_sub_ps(iz2,jz0);
797 /* Calculate squared distance and things based on it */
798 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
799 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
800 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
802 rinv00 = gmx_mm_invsqrt_ps(rsq00);
803 rinv10 = gmx_mm_invsqrt_ps(rsq10);
804 rinv20 = gmx_mm_invsqrt_ps(rsq20);
806 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
807 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
808 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
810 /* Load parameters for j particles */
811 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
812 charge+jnrC+0,charge+jnrD+0);
813 vdwjidx0A = 2*vdwtype[jnrA+0];
814 vdwjidx0B = 2*vdwtype[jnrB+0];
815 vdwjidx0C = 2*vdwtype[jnrC+0];
816 vdwjidx0D = 2*vdwtype[jnrD+0];
818 fjx0 = _mm_setzero_ps();
819 fjy0 = _mm_setzero_ps();
820 fjz0 = _mm_setzero_ps();
822 /**************************
823 * CALCULATE INTERACTIONS *
824 **************************/
826 if (gmx_mm_any_lt(rsq00,rcutoff2))
829 /* Compute parameters for interactions between i and j atoms */
830 qq00 = _mm_mul_ps(iq0,jq0);
831 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
832 vdwparam+vdwioffset0+vdwjidx0B,
833 vdwparam+vdwioffset0+vdwjidx0C,
834 vdwparam+vdwioffset0+vdwjidx0D,
837 /* REACTION-FIELD ELECTROSTATICS */
838 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
840 /* LENNARD-JONES DISPERSION/REPULSION */
842 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
843 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
845 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
847 fscal = _mm_add_ps(felec,fvdw);
849 fscal = _mm_and_ps(fscal,cutoff_mask);
851 /* Calculate temporary vectorial force */
852 tx = _mm_mul_ps(fscal,dx00);
853 ty = _mm_mul_ps(fscal,dy00);
854 tz = _mm_mul_ps(fscal,dz00);
856 /* Update vectorial force */
857 fix0 = _mm_add_ps(fix0,tx);
858 fiy0 = _mm_add_ps(fiy0,ty);
859 fiz0 = _mm_add_ps(fiz0,tz);
861 fjx0 = _mm_add_ps(fjx0,tx);
862 fjy0 = _mm_add_ps(fjy0,ty);
863 fjz0 = _mm_add_ps(fjz0,tz);
867 /**************************
868 * CALCULATE INTERACTIONS *
869 **************************/
871 if (gmx_mm_any_lt(rsq10,rcutoff2))
874 /* Compute parameters for interactions between i and j atoms */
875 qq10 = _mm_mul_ps(iq1,jq0);
877 /* REACTION-FIELD ELECTROSTATICS */
878 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
880 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
884 fscal = _mm_and_ps(fscal,cutoff_mask);
886 /* Calculate temporary vectorial force */
887 tx = _mm_mul_ps(fscal,dx10);
888 ty = _mm_mul_ps(fscal,dy10);
889 tz = _mm_mul_ps(fscal,dz10);
891 /* Update vectorial force */
892 fix1 = _mm_add_ps(fix1,tx);
893 fiy1 = _mm_add_ps(fiy1,ty);
894 fiz1 = _mm_add_ps(fiz1,tz);
896 fjx0 = _mm_add_ps(fjx0,tx);
897 fjy0 = _mm_add_ps(fjy0,ty);
898 fjz0 = _mm_add_ps(fjz0,tz);
902 /**************************
903 * CALCULATE INTERACTIONS *
904 **************************/
906 if (gmx_mm_any_lt(rsq20,rcutoff2))
909 /* Compute parameters for interactions between i and j atoms */
910 qq20 = _mm_mul_ps(iq2,jq0);
912 /* REACTION-FIELD ELECTROSTATICS */
913 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
915 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
919 fscal = _mm_and_ps(fscal,cutoff_mask);
921 /* Calculate temporary vectorial force */
922 tx = _mm_mul_ps(fscal,dx20);
923 ty = _mm_mul_ps(fscal,dy20);
924 tz = _mm_mul_ps(fscal,dz20);
926 /* Update vectorial force */
927 fix2 = _mm_add_ps(fix2,tx);
928 fiy2 = _mm_add_ps(fiy2,ty);
929 fiz2 = _mm_add_ps(fiz2,tz);
931 fjx0 = _mm_add_ps(fjx0,tx);
932 fjy0 = _mm_add_ps(fjy0,ty);
933 fjz0 = _mm_add_ps(fjz0,tz);
937 fjptrA = f+j_coord_offsetA;
938 fjptrB = f+j_coord_offsetB;
939 fjptrC = f+j_coord_offsetC;
940 fjptrD = f+j_coord_offsetD;
942 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
944 /* Inner loop uses 97 flops */
950 /* Get j neighbor index, and coordinate index */
951 jnrlistA = jjnr[jidx];
952 jnrlistB = jjnr[jidx+1];
953 jnrlistC = jjnr[jidx+2];
954 jnrlistD = jjnr[jidx+3];
955 /* Sign of each element will be negative for non-real atoms.
956 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
957 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
959 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
960 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
961 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
962 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
963 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
964 j_coord_offsetA = DIM*jnrA;
965 j_coord_offsetB = DIM*jnrB;
966 j_coord_offsetC = DIM*jnrC;
967 j_coord_offsetD = DIM*jnrD;
969 /* load j atom coordinates */
970 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
971 x+j_coord_offsetC,x+j_coord_offsetD,
974 /* Calculate displacement vector */
975 dx00 = _mm_sub_ps(ix0,jx0);
976 dy00 = _mm_sub_ps(iy0,jy0);
977 dz00 = _mm_sub_ps(iz0,jz0);
978 dx10 = _mm_sub_ps(ix1,jx0);
979 dy10 = _mm_sub_ps(iy1,jy0);
980 dz10 = _mm_sub_ps(iz1,jz0);
981 dx20 = _mm_sub_ps(ix2,jx0);
982 dy20 = _mm_sub_ps(iy2,jy0);
983 dz20 = _mm_sub_ps(iz2,jz0);
985 /* Calculate squared distance and things based on it */
986 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
987 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
988 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
990 rinv00 = gmx_mm_invsqrt_ps(rsq00);
991 rinv10 = gmx_mm_invsqrt_ps(rsq10);
992 rinv20 = gmx_mm_invsqrt_ps(rsq20);
994 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
995 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
996 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
998 /* Load parameters for j particles */
999 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1000 charge+jnrC+0,charge+jnrD+0);
1001 vdwjidx0A = 2*vdwtype[jnrA+0];
1002 vdwjidx0B = 2*vdwtype[jnrB+0];
1003 vdwjidx0C = 2*vdwtype[jnrC+0];
1004 vdwjidx0D = 2*vdwtype[jnrD+0];
1006 fjx0 = _mm_setzero_ps();
1007 fjy0 = _mm_setzero_ps();
1008 fjz0 = _mm_setzero_ps();
1010 /**************************
1011 * CALCULATE INTERACTIONS *
1012 **************************/
1014 if (gmx_mm_any_lt(rsq00,rcutoff2))
1017 /* Compute parameters for interactions between i and j atoms */
1018 qq00 = _mm_mul_ps(iq0,jq0);
1019 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1020 vdwparam+vdwioffset0+vdwjidx0B,
1021 vdwparam+vdwioffset0+vdwjidx0C,
1022 vdwparam+vdwioffset0+vdwjidx0D,
1025 /* REACTION-FIELD ELECTROSTATICS */
1026 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1028 /* LENNARD-JONES DISPERSION/REPULSION */
1030 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1031 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1033 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1035 fscal = _mm_add_ps(felec,fvdw);
1037 fscal = _mm_and_ps(fscal,cutoff_mask);
1039 fscal = _mm_andnot_ps(dummy_mask,fscal);
1041 /* Calculate temporary vectorial force */
1042 tx = _mm_mul_ps(fscal,dx00);
1043 ty = _mm_mul_ps(fscal,dy00);
1044 tz = _mm_mul_ps(fscal,dz00);
1046 /* Update vectorial force */
1047 fix0 = _mm_add_ps(fix0,tx);
1048 fiy0 = _mm_add_ps(fiy0,ty);
1049 fiz0 = _mm_add_ps(fiz0,tz);
1051 fjx0 = _mm_add_ps(fjx0,tx);
1052 fjy0 = _mm_add_ps(fjy0,ty);
1053 fjz0 = _mm_add_ps(fjz0,tz);
1057 /**************************
1058 * CALCULATE INTERACTIONS *
1059 **************************/
1061 if (gmx_mm_any_lt(rsq10,rcutoff2))
1064 /* Compute parameters for interactions between i and j atoms */
1065 qq10 = _mm_mul_ps(iq1,jq0);
1067 /* REACTION-FIELD ELECTROSTATICS */
1068 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1070 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1074 fscal = _mm_and_ps(fscal,cutoff_mask);
1076 fscal = _mm_andnot_ps(dummy_mask,fscal);
1078 /* Calculate temporary vectorial force */
1079 tx = _mm_mul_ps(fscal,dx10);
1080 ty = _mm_mul_ps(fscal,dy10);
1081 tz = _mm_mul_ps(fscal,dz10);
1083 /* Update vectorial force */
1084 fix1 = _mm_add_ps(fix1,tx);
1085 fiy1 = _mm_add_ps(fiy1,ty);
1086 fiz1 = _mm_add_ps(fiz1,tz);
1088 fjx0 = _mm_add_ps(fjx0,tx);
1089 fjy0 = _mm_add_ps(fjy0,ty);
1090 fjz0 = _mm_add_ps(fjz0,tz);
1094 /**************************
1095 * CALCULATE INTERACTIONS *
1096 **************************/
1098 if (gmx_mm_any_lt(rsq20,rcutoff2))
1101 /* Compute parameters for interactions between i and j atoms */
1102 qq20 = _mm_mul_ps(iq2,jq0);
1104 /* REACTION-FIELD ELECTROSTATICS */
1105 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1107 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1111 fscal = _mm_and_ps(fscal,cutoff_mask);
1113 fscal = _mm_andnot_ps(dummy_mask,fscal);
1115 /* Calculate temporary vectorial force */
1116 tx = _mm_mul_ps(fscal,dx20);
1117 ty = _mm_mul_ps(fscal,dy20);
1118 tz = _mm_mul_ps(fscal,dz20);
1120 /* Update vectorial force */
1121 fix2 = _mm_add_ps(fix2,tx);
1122 fiy2 = _mm_add_ps(fiy2,ty);
1123 fiz2 = _mm_add_ps(fiz2,tz);
1125 fjx0 = _mm_add_ps(fjx0,tx);
1126 fjy0 = _mm_add_ps(fjy0,ty);
1127 fjz0 = _mm_add_ps(fjz0,tz);
1131 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1132 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1133 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1134 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1136 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1138 /* Inner loop uses 97 flops */
1141 /* End of innermost loop */
1143 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1144 f+i_coord_offset,fshift+i_shift_offset);
1146 /* Increment number of inner iterations */
1147 inneriter += j_index_end - j_index_start;
1149 /* Outer loop uses 18 flops */
1152 /* Increment number of outer iterations */
1155 /* Update outer/inner flops */
1157 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*97);