2 * Note: this file was generated by the Gromacs sse4_1_single kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_sse4_1_single.h"
34 #include "kernelutil_x86_sse4_1_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse4_1_single
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: LennardJones
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse4_1_single
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
60 int jnrA,jnrB,jnrC,jnrD;
61 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
62 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
63 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
65 real *shiftvec,*fshift,*x,*f;
66 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
68 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
70 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
72 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
74 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
75 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
76 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
77 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
78 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
79 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
80 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
83 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
86 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
87 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
88 __m128 dummy_mask,cutoff_mask;
89 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
90 __m128 one = _mm_set1_ps(1.0);
91 __m128 two = _mm_set1_ps(2.0);
97 jindex = nlist->jindex;
99 shiftidx = nlist->shift;
101 shiftvec = fr->shift_vec[0];
102 fshift = fr->fshift[0];
103 facel = _mm_set1_ps(fr->epsfac);
104 charge = mdatoms->chargeA;
105 krf = _mm_set1_ps(fr->ic->k_rf);
106 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
107 crf = _mm_set1_ps(fr->ic->c_rf);
108 nvdwtype = fr->ntype;
110 vdwtype = mdatoms->typeA;
112 /* Setup water-specific parameters */
113 inr = nlist->iinr[0];
114 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
115 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
116 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
117 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
119 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
120 rcutoff_scalar = fr->rcoulomb;
121 rcutoff = _mm_set1_ps(rcutoff_scalar);
122 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
124 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
125 rvdw = _mm_set1_ps(fr->rvdw);
127 /* Avoid stupid compiler warnings */
128 jnrA = jnrB = jnrC = jnrD = 0;
137 for(iidx=0;iidx<4*DIM;iidx++)
142 /* Start outer loop over neighborlists */
143 for(iidx=0; iidx<nri; iidx++)
145 /* Load shift vector for this list */
146 i_shift_offset = DIM*shiftidx[iidx];
148 /* Load limits for loop over neighbors */
149 j_index_start = jindex[iidx];
150 j_index_end = jindex[iidx+1];
152 /* Get outer coordinate index */
154 i_coord_offset = DIM*inr;
156 /* Load i particle coords and add shift vector */
157 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
158 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
160 fix0 = _mm_setzero_ps();
161 fiy0 = _mm_setzero_ps();
162 fiz0 = _mm_setzero_ps();
163 fix1 = _mm_setzero_ps();
164 fiy1 = _mm_setzero_ps();
165 fiz1 = _mm_setzero_ps();
166 fix2 = _mm_setzero_ps();
167 fiy2 = _mm_setzero_ps();
168 fiz2 = _mm_setzero_ps();
170 /* Reset potential sums */
171 velecsum = _mm_setzero_ps();
172 vvdwsum = _mm_setzero_ps();
174 /* Start inner kernel loop */
175 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
178 /* Get j neighbor index, and coordinate index */
183 j_coord_offsetA = DIM*jnrA;
184 j_coord_offsetB = DIM*jnrB;
185 j_coord_offsetC = DIM*jnrC;
186 j_coord_offsetD = DIM*jnrD;
188 /* load j atom coordinates */
189 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
190 x+j_coord_offsetC,x+j_coord_offsetD,
193 /* Calculate displacement vector */
194 dx00 = _mm_sub_ps(ix0,jx0);
195 dy00 = _mm_sub_ps(iy0,jy0);
196 dz00 = _mm_sub_ps(iz0,jz0);
197 dx10 = _mm_sub_ps(ix1,jx0);
198 dy10 = _mm_sub_ps(iy1,jy0);
199 dz10 = _mm_sub_ps(iz1,jz0);
200 dx20 = _mm_sub_ps(ix2,jx0);
201 dy20 = _mm_sub_ps(iy2,jy0);
202 dz20 = _mm_sub_ps(iz2,jz0);
204 /* Calculate squared distance and things based on it */
205 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
206 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
207 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
209 rinv00 = gmx_mm_invsqrt_ps(rsq00);
210 rinv10 = gmx_mm_invsqrt_ps(rsq10);
211 rinv20 = gmx_mm_invsqrt_ps(rsq20);
213 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
214 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
215 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
217 /* Load parameters for j particles */
218 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
219 charge+jnrC+0,charge+jnrD+0);
220 vdwjidx0A = 2*vdwtype[jnrA+0];
221 vdwjidx0B = 2*vdwtype[jnrB+0];
222 vdwjidx0C = 2*vdwtype[jnrC+0];
223 vdwjidx0D = 2*vdwtype[jnrD+0];
225 fjx0 = _mm_setzero_ps();
226 fjy0 = _mm_setzero_ps();
227 fjz0 = _mm_setzero_ps();
229 /**************************
230 * CALCULATE INTERACTIONS *
231 **************************/
233 if (gmx_mm_any_lt(rsq00,rcutoff2))
236 /* Compute parameters for interactions between i and j atoms */
237 qq00 = _mm_mul_ps(iq0,jq0);
238 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
239 vdwparam+vdwioffset0+vdwjidx0B,
240 vdwparam+vdwioffset0+vdwjidx0C,
241 vdwparam+vdwioffset0+vdwjidx0D,
244 /* REACTION-FIELD ELECTROSTATICS */
245 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
246 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
248 /* LENNARD-JONES DISPERSION/REPULSION */
250 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
251 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
252 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
253 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) ,
254 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
255 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
257 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
259 /* Update potential sum for this i atom from the interaction with this j atom. */
260 velec = _mm_and_ps(velec,cutoff_mask);
261 velecsum = _mm_add_ps(velecsum,velec);
262 vvdw = _mm_and_ps(vvdw,cutoff_mask);
263 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
265 fscal = _mm_add_ps(felec,fvdw);
267 fscal = _mm_and_ps(fscal,cutoff_mask);
269 /* Calculate temporary vectorial force */
270 tx = _mm_mul_ps(fscal,dx00);
271 ty = _mm_mul_ps(fscal,dy00);
272 tz = _mm_mul_ps(fscal,dz00);
274 /* Update vectorial force */
275 fix0 = _mm_add_ps(fix0,tx);
276 fiy0 = _mm_add_ps(fiy0,ty);
277 fiz0 = _mm_add_ps(fiz0,tz);
279 fjx0 = _mm_add_ps(fjx0,tx);
280 fjy0 = _mm_add_ps(fjy0,ty);
281 fjz0 = _mm_add_ps(fjz0,tz);
285 /**************************
286 * CALCULATE INTERACTIONS *
287 **************************/
289 if (gmx_mm_any_lt(rsq10,rcutoff2))
292 /* Compute parameters for interactions between i and j atoms */
293 qq10 = _mm_mul_ps(iq1,jq0);
295 /* REACTION-FIELD ELECTROSTATICS */
296 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
297 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
299 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
301 /* Update potential sum for this i atom from the interaction with this j atom. */
302 velec = _mm_and_ps(velec,cutoff_mask);
303 velecsum = _mm_add_ps(velecsum,velec);
307 fscal = _mm_and_ps(fscal,cutoff_mask);
309 /* Calculate temporary vectorial force */
310 tx = _mm_mul_ps(fscal,dx10);
311 ty = _mm_mul_ps(fscal,dy10);
312 tz = _mm_mul_ps(fscal,dz10);
314 /* Update vectorial force */
315 fix1 = _mm_add_ps(fix1,tx);
316 fiy1 = _mm_add_ps(fiy1,ty);
317 fiz1 = _mm_add_ps(fiz1,tz);
319 fjx0 = _mm_add_ps(fjx0,tx);
320 fjy0 = _mm_add_ps(fjy0,ty);
321 fjz0 = _mm_add_ps(fjz0,tz);
325 /**************************
326 * CALCULATE INTERACTIONS *
327 **************************/
329 if (gmx_mm_any_lt(rsq20,rcutoff2))
332 /* Compute parameters for interactions between i and j atoms */
333 qq20 = _mm_mul_ps(iq2,jq0);
335 /* REACTION-FIELD ELECTROSTATICS */
336 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
337 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
339 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
341 /* Update potential sum for this i atom from the interaction with this j atom. */
342 velec = _mm_and_ps(velec,cutoff_mask);
343 velecsum = _mm_add_ps(velecsum,velec);
347 fscal = _mm_and_ps(fscal,cutoff_mask);
349 /* Calculate temporary vectorial force */
350 tx = _mm_mul_ps(fscal,dx20);
351 ty = _mm_mul_ps(fscal,dy20);
352 tz = _mm_mul_ps(fscal,dz20);
354 /* Update vectorial force */
355 fix2 = _mm_add_ps(fix2,tx);
356 fiy2 = _mm_add_ps(fiy2,ty);
357 fiz2 = _mm_add_ps(fiz2,tz);
359 fjx0 = _mm_add_ps(fjx0,tx);
360 fjy0 = _mm_add_ps(fjy0,ty);
361 fjz0 = _mm_add_ps(fjz0,tz);
365 fjptrA = f+j_coord_offsetA;
366 fjptrB = f+j_coord_offsetB;
367 fjptrC = f+j_coord_offsetC;
368 fjptrD = f+j_coord_offsetD;
370 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
372 /* Inner loop uses 126 flops */
378 /* Get j neighbor index, and coordinate index */
379 jnrlistA = jjnr[jidx];
380 jnrlistB = jjnr[jidx+1];
381 jnrlistC = jjnr[jidx+2];
382 jnrlistD = jjnr[jidx+3];
383 /* Sign of each element will be negative for non-real atoms.
384 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
385 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
387 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
388 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
389 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
390 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
391 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
392 j_coord_offsetA = DIM*jnrA;
393 j_coord_offsetB = DIM*jnrB;
394 j_coord_offsetC = DIM*jnrC;
395 j_coord_offsetD = DIM*jnrD;
397 /* load j atom coordinates */
398 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
399 x+j_coord_offsetC,x+j_coord_offsetD,
402 /* Calculate displacement vector */
403 dx00 = _mm_sub_ps(ix0,jx0);
404 dy00 = _mm_sub_ps(iy0,jy0);
405 dz00 = _mm_sub_ps(iz0,jz0);
406 dx10 = _mm_sub_ps(ix1,jx0);
407 dy10 = _mm_sub_ps(iy1,jy0);
408 dz10 = _mm_sub_ps(iz1,jz0);
409 dx20 = _mm_sub_ps(ix2,jx0);
410 dy20 = _mm_sub_ps(iy2,jy0);
411 dz20 = _mm_sub_ps(iz2,jz0);
413 /* Calculate squared distance and things based on it */
414 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
415 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
416 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
418 rinv00 = gmx_mm_invsqrt_ps(rsq00);
419 rinv10 = gmx_mm_invsqrt_ps(rsq10);
420 rinv20 = gmx_mm_invsqrt_ps(rsq20);
422 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
423 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
424 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
426 /* Load parameters for j particles */
427 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
428 charge+jnrC+0,charge+jnrD+0);
429 vdwjidx0A = 2*vdwtype[jnrA+0];
430 vdwjidx0B = 2*vdwtype[jnrB+0];
431 vdwjidx0C = 2*vdwtype[jnrC+0];
432 vdwjidx0D = 2*vdwtype[jnrD+0];
434 fjx0 = _mm_setzero_ps();
435 fjy0 = _mm_setzero_ps();
436 fjz0 = _mm_setzero_ps();
438 /**************************
439 * CALCULATE INTERACTIONS *
440 **************************/
442 if (gmx_mm_any_lt(rsq00,rcutoff2))
445 /* Compute parameters for interactions between i and j atoms */
446 qq00 = _mm_mul_ps(iq0,jq0);
447 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
448 vdwparam+vdwioffset0+vdwjidx0B,
449 vdwparam+vdwioffset0+vdwjidx0C,
450 vdwparam+vdwioffset0+vdwjidx0D,
453 /* REACTION-FIELD ELECTROSTATICS */
454 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
455 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
457 /* LENNARD-JONES DISPERSION/REPULSION */
459 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
460 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
461 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
462 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) ,
463 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
464 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
466 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
468 /* Update potential sum for this i atom from the interaction with this j atom. */
469 velec = _mm_and_ps(velec,cutoff_mask);
470 velec = _mm_andnot_ps(dummy_mask,velec);
471 velecsum = _mm_add_ps(velecsum,velec);
472 vvdw = _mm_and_ps(vvdw,cutoff_mask);
473 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
474 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
476 fscal = _mm_add_ps(felec,fvdw);
478 fscal = _mm_and_ps(fscal,cutoff_mask);
480 fscal = _mm_andnot_ps(dummy_mask,fscal);
482 /* Calculate temporary vectorial force */
483 tx = _mm_mul_ps(fscal,dx00);
484 ty = _mm_mul_ps(fscal,dy00);
485 tz = _mm_mul_ps(fscal,dz00);
487 /* Update vectorial force */
488 fix0 = _mm_add_ps(fix0,tx);
489 fiy0 = _mm_add_ps(fiy0,ty);
490 fiz0 = _mm_add_ps(fiz0,tz);
492 fjx0 = _mm_add_ps(fjx0,tx);
493 fjy0 = _mm_add_ps(fjy0,ty);
494 fjz0 = _mm_add_ps(fjz0,tz);
498 /**************************
499 * CALCULATE INTERACTIONS *
500 **************************/
502 if (gmx_mm_any_lt(rsq10,rcutoff2))
505 /* Compute parameters for interactions between i and j atoms */
506 qq10 = _mm_mul_ps(iq1,jq0);
508 /* REACTION-FIELD ELECTROSTATICS */
509 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
510 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
512 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
514 /* Update potential sum for this i atom from the interaction with this j atom. */
515 velec = _mm_and_ps(velec,cutoff_mask);
516 velec = _mm_andnot_ps(dummy_mask,velec);
517 velecsum = _mm_add_ps(velecsum,velec);
521 fscal = _mm_and_ps(fscal,cutoff_mask);
523 fscal = _mm_andnot_ps(dummy_mask,fscal);
525 /* Calculate temporary vectorial force */
526 tx = _mm_mul_ps(fscal,dx10);
527 ty = _mm_mul_ps(fscal,dy10);
528 tz = _mm_mul_ps(fscal,dz10);
530 /* Update vectorial force */
531 fix1 = _mm_add_ps(fix1,tx);
532 fiy1 = _mm_add_ps(fiy1,ty);
533 fiz1 = _mm_add_ps(fiz1,tz);
535 fjx0 = _mm_add_ps(fjx0,tx);
536 fjy0 = _mm_add_ps(fjy0,ty);
537 fjz0 = _mm_add_ps(fjz0,tz);
541 /**************************
542 * CALCULATE INTERACTIONS *
543 **************************/
545 if (gmx_mm_any_lt(rsq20,rcutoff2))
548 /* Compute parameters for interactions between i and j atoms */
549 qq20 = _mm_mul_ps(iq2,jq0);
551 /* REACTION-FIELD ELECTROSTATICS */
552 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
553 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
555 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
557 /* Update potential sum for this i atom from the interaction with this j atom. */
558 velec = _mm_and_ps(velec,cutoff_mask);
559 velec = _mm_andnot_ps(dummy_mask,velec);
560 velecsum = _mm_add_ps(velecsum,velec);
564 fscal = _mm_and_ps(fscal,cutoff_mask);
566 fscal = _mm_andnot_ps(dummy_mask,fscal);
568 /* Calculate temporary vectorial force */
569 tx = _mm_mul_ps(fscal,dx20);
570 ty = _mm_mul_ps(fscal,dy20);
571 tz = _mm_mul_ps(fscal,dz20);
573 /* Update vectorial force */
574 fix2 = _mm_add_ps(fix2,tx);
575 fiy2 = _mm_add_ps(fiy2,ty);
576 fiz2 = _mm_add_ps(fiz2,tz);
578 fjx0 = _mm_add_ps(fjx0,tx);
579 fjy0 = _mm_add_ps(fjy0,ty);
580 fjz0 = _mm_add_ps(fjz0,tz);
584 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
585 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
586 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
587 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
589 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
591 /* Inner loop uses 126 flops */
594 /* End of innermost loop */
596 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
597 f+i_coord_offset,fshift+i_shift_offset);
600 /* Update potential energies */
601 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
602 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
604 /* Increment number of inner iterations */
605 inneriter += j_index_end - j_index_start;
607 /* Outer loop uses 20 flops */
610 /* Increment number of outer iterations */
613 /* Update outer/inner flops */
615 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*126);
618 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse4_1_single
619 * Electrostatics interaction: ReactionField
620 * VdW interaction: LennardJones
621 * Geometry: Water3-Particle
622 * Calculate force/pot: Force
625 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse4_1_single
626 (t_nblist * gmx_restrict nlist,
627 rvec * gmx_restrict xx,
628 rvec * gmx_restrict ff,
629 t_forcerec * gmx_restrict fr,
630 t_mdatoms * gmx_restrict mdatoms,
631 nb_kernel_data_t * gmx_restrict kernel_data,
632 t_nrnb * gmx_restrict nrnb)
634 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
635 * just 0 for non-waters.
636 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
637 * jnr indices corresponding to data put in the four positions in the SIMD register.
639 int i_shift_offset,i_coord_offset,outeriter,inneriter;
640 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
641 int jnrA,jnrB,jnrC,jnrD;
642 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
643 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
644 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
646 real *shiftvec,*fshift,*x,*f;
647 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
649 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
651 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
653 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
655 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
656 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
657 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
658 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
659 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
660 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
661 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
664 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
667 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
668 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
669 __m128 dummy_mask,cutoff_mask;
670 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
671 __m128 one = _mm_set1_ps(1.0);
672 __m128 two = _mm_set1_ps(2.0);
678 jindex = nlist->jindex;
680 shiftidx = nlist->shift;
682 shiftvec = fr->shift_vec[0];
683 fshift = fr->fshift[0];
684 facel = _mm_set1_ps(fr->epsfac);
685 charge = mdatoms->chargeA;
686 krf = _mm_set1_ps(fr->ic->k_rf);
687 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
688 crf = _mm_set1_ps(fr->ic->c_rf);
689 nvdwtype = fr->ntype;
691 vdwtype = mdatoms->typeA;
693 /* Setup water-specific parameters */
694 inr = nlist->iinr[0];
695 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
696 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
697 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
698 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
700 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
701 rcutoff_scalar = fr->rcoulomb;
702 rcutoff = _mm_set1_ps(rcutoff_scalar);
703 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
705 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
706 rvdw = _mm_set1_ps(fr->rvdw);
708 /* Avoid stupid compiler warnings */
709 jnrA = jnrB = jnrC = jnrD = 0;
718 for(iidx=0;iidx<4*DIM;iidx++)
723 /* Start outer loop over neighborlists */
724 for(iidx=0; iidx<nri; iidx++)
726 /* Load shift vector for this list */
727 i_shift_offset = DIM*shiftidx[iidx];
729 /* Load limits for loop over neighbors */
730 j_index_start = jindex[iidx];
731 j_index_end = jindex[iidx+1];
733 /* Get outer coordinate index */
735 i_coord_offset = DIM*inr;
737 /* Load i particle coords and add shift vector */
738 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
739 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
741 fix0 = _mm_setzero_ps();
742 fiy0 = _mm_setzero_ps();
743 fiz0 = _mm_setzero_ps();
744 fix1 = _mm_setzero_ps();
745 fiy1 = _mm_setzero_ps();
746 fiz1 = _mm_setzero_ps();
747 fix2 = _mm_setzero_ps();
748 fiy2 = _mm_setzero_ps();
749 fiz2 = _mm_setzero_ps();
751 /* Start inner kernel loop */
752 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
755 /* Get j neighbor index, and coordinate index */
760 j_coord_offsetA = DIM*jnrA;
761 j_coord_offsetB = DIM*jnrB;
762 j_coord_offsetC = DIM*jnrC;
763 j_coord_offsetD = DIM*jnrD;
765 /* load j atom coordinates */
766 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
767 x+j_coord_offsetC,x+j_coord_offsetD,
770 /* Calculate displacement vector */
771 dx00 = _mm_sub_ps(ix0,jx0);
772 dy00 = _mm_sub_ps(iy0,jy0);
773 dz00 = _mm_sub_ps(iz0,jz0);
774 dx10 = _mm_sub_ps(ix1,jx0);
775 dy10 = _mm_sub_ps(iy1,jy0);
776 dz10 = _mm_sub_ps(iz1,jz0);
777 dx20 = _mm_sub_ps(ix2,jx0);
778 dy20 = _mm_sub_ps(iy2,jy0);
779 dz20 = _mm_sub_ps(iz2,jz0);
781 /* Calculate squared distance and things based on it */
782 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
783 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
784 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
786 rinv00 = gmx_mm_invsqrt_ps(rsq00);
787 rinv10 = gmx_mm_invsqrt_ps(rsq10);
788 rinv20 = gmx_mm_invsqrt_ps(rsq20);
790 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
791 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
792 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
794 /* Load parameters for j particles */
795 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
796 charge+jnrC+0,charge+jnrD+0);
797 vdwjidx0A = 2*vdwtype[jnrA+0];
798 vdwjidx0B = 2*vdwtype[jnrB+0];
799 vdwjidx0C = 2*vdwtype[jnrC+0];
800 vdwjidx0D = 2*vdwtype[jnrD+0];
802 fjx0 = _mm_setzero_ps();
803 fjy0 = _mm_setzero_ps();
804 fjz0 = _mm_setzero_ps();
806 /**************************
807 * CALCULATE INTERACTIONS *
808 **************************/
810 if (gmx_mm_any_lt(rsq00,rcutoff2))
813 /* Compute parameters for interactions between i and j atoms */
814 qq00 = _mm_mul_ps(iq0,jq0);
815 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
816 vdwparam+vdwioffset0+vdwjidx0B,
817 vdwparam+vdwioffset0+vdwjidx0C,
818 vdwparam+vdwioffset0+vdwjidx0D,
821 /* REACTION-FIELD ELECTROSTATICS */
822 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
824 /* LENNARD-JONES DISPERSION/REPULSION */
826 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
827 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
829 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
831 fscal = _mm_add_ps(felec,fvdw);
833 fscal = _mm_and_ps(fscal,cutoff_mask);
835 /* Calculate temporary vectorial force */
836 tx = _mm_mul_ps(fscal,dx00);
837 ty = _mm_mul_ps(fscal,dy00);
838 tz = _mm_mul_ps(fscal,dz00);
840 /* Update vectorial force */
841 fix0 = _mm_add_ps(fix0,tx);
842 fiy0 = _mm_add_ps(fiy0,ty);
843 fiz0 = _mm_add_ps(fiz0,tz);
845 fjx0 = _mm_add_ps(fjx0,tx);
846 fjy0 = _mm_add_ps(fjy0,ty);
847 fjz0 = _mm_add_ps(fjz0,tz);
851 /**************************
852 * CALCULATE INTERACTIONS *
853 **************************/
855 if (gmx_mm_any_lt(rsq10,rcutoff2))
858 /* Compute parameters for interactions between i and j atoms */
859 qq10 = _mm_mul_ps(iq1,jq0);
861 /* REACTION-FIELD ELECTROSTATICS */
862 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
864 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
868 fscal = _mm_and_ps(fscal,cutoff_mask);
870 /* Calculate temporary vectorial force */
871 tx = _mm_mul_ps(fscal,dx10);
872 ty = _mm_mul_ps(fscal,dy10);
873 tz = _mm_mul_ps(fscal,dz10);
875 /* Update vectorial force */
876 fix1 = _mm_add_ps(fix1,tx);
877 fiy1 = _mm_add_ps(fiy1,ty);
878 fiz1 = _mm_add_ps(fiz1,tz);
880 fjx0 = _mm_add_ps(fjx0,tx);
881 fjy0 = _mm_add_ps(fjy0,ty);
882 fjz0 = _mm_add_ps(fjz0,tz);
886 /**************************
887 * CALCULATE INTERACTIONS *
888 **************************/
890 if (gmx_mm_any_lt(rsq20,rcutoff2))
893 /* Compute parameters for interactions between i and j atoms */
894 qq20 = _mm_mul_ps(iq2,jq0);
896 /* REACTION-FIELD ELECTROSTATICS */
897 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
899 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
903 fscal = _mm_and_ps(fscal,cutoff_mask);
905 /* Calculate temporary vectorial force */
906 tx = _mm_mul_ps(fscal,dx20);
907 ty = _mm_mul_ps(fscal,dy20);
908 tz = _mm_mul_ps(fscal,dz20);
910 /* Update vectorial force */
911 fix2 = _mm_add_ps(fix2,tx);
912 fiy2 = _mm_add_ps(fiy2,ty);
913 fiz2 = _mm_add_ps(fiz2,tz);
915 fjx0 = _mm_add_ps(fjx0,tx);
916 fjy0 = _mm_add_ps(fjy0,ty);
917 fjz0 = _mm_add_ps(fjz0,tz);
921 fjptrA = f+j_coord_offsetA;
922 fjptrB = f+j_coord_offsetB;
923 fjptrC = f+j_coord_offsetC;
924 fjptrD = f+j_coord_offsetD;
926 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
928 /* Inner loop uses 97 flops */
934 /* Get j neighbor index, and coordinate index */
935 jnrlistA = jjnr[jidx];
936 jnrlistB = jjnr[jidx+1];
937 jnrlistC = jjnr[jidx+2];
938 jnrlistD = jjnr[jidx+3];
939 /* Sign of each element will be negative for non-real atoms.
940 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
941 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
943 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
944 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
945 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
946 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
947 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
948 j_coord_offsetA = DIM*jnrA;
949 j_coord_offsetB = DIM*jnrB;
950 j_coord_offsetC = DIM*jnrC;
951 j_coord_offsetD = DIM*jnrD;
953 /* load j atom coordinates */
954 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
955 x+j_coord_offsetC,x+j_coord_offsetD,
958 /* Calculate displacement vector */
959 dx00 = _mm_sub_ps(ix0,jx0);
960 dy00 = _mm_sub_ps(iy0,jy0);
961 dz00 = _mm_sub_ps(iz0,jz0);
962 dx10 = _mm_sub_ps(ix1,jx0);
963 dy10 = _mm_sub_ps(iy1,jy0);
964 dz10 = _mm_sub_ps(iz1,jz0);
965 dx20 = _mm_sub_ps(ix2,jx0);
966 dy20 = _mm_sub_ps(iy2,jy0);
967 dz20 = _mm_sub_ps(iz2,jz0);
969 /* Calculate squared distance and things based on it */
970 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
971 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
972 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
974 rinv00 = gmx_mm_invsqrt_ps(rsq00);
975 rinv10 = gmx_mm_invsqrt_ps(rsq10);
976 rinv20 = gmx_mm_invsqrt_ps(rsq20);
978 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
979 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
980 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
982 /* Load parameters for j particles */
983 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
984 charge+jnrC+0,charge+jnrD+0);
985 vdwjidx0A = 2*vdwtype[jnrA+0];
986 vdwjidx0B = 2*vdwtype[jnrB+0];
987 vdwjidx0C = 2*vdwtype[jnrC+0];
988 vdwjidx0D = 2*vdwtype[jnrD+0];
990 fjx0 = _mm_setzero_ps();
991 fjy0 = _mm_setzero_ps();
992 fjz0 = _mm_setzero_ps();
994 /**************************
995 * CALCULATE INTERACTIONS *
996 **************************/
998 if (gmx_mm_any_lt(rsq00,rcutoff2))
1001 /* Compute parameters for interactions between i and j atoms */
1002 qq00 = _mm_mul_ps(iq0,jq0);
1003 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1004 vdwparam+vdwioffset0+vdwjidx0B,
1005 vdwparam+vdwioffset0+vdwjidx0C,
1006 vdwparam+vdwioffset0+vdwjidx0D,
1009 /* REACTION-FIELD ELECTROSTATICS */
1010 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1012 /* LENNARD-JONES DISPERSION/REPULSION */
1014 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1015 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1017 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1019 fscal = _mm_add_ps(felec,fvdw);
1021 fscal = _mm_and_ps(fscal,cutoff_mask);
1023 fscal = _mm_andnot_ps(dummy_mask,fscal);
1025 /* Calculate temporary vectorial force */
1026 tx = _mm_mul_ps(fscal,dx00);
1027 ty = _mm_mul_ps(fscal,dy00);
1028 tz = _mm_mul_ps(fscal,dz00);
1030 /* Update vectorial force */
1031 fix0 = _mm_add_ps(fix0,tx);
1032 fiy0 = _mm_add_ps(fiy0,ty);
1033 fiz0 = _mm_add_ps(fiz0,tz);
1035 fjx0 = _mm_add_ps(fjx0,tx);
1036 fjy0 = _mm_add_ps(fjy0,ty);
1037 fjz0 = _mm_add_ps(fjz0,tz);
1041 /**************************
1042 * CALCULATE INTERACTIONS *
1043 **************************/
1045 if (gmx_mm_any_lt(rsq10,rcutoff2))
1048 /* Compute parameters for interactions between i and j atoms */
1049 qq10 = _mm_mul_ps(iq1,jq0);
1051 /* REACTION-FIELD ELECTROSTATICS */
1052 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1054 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1058 fscal = _mm_and_ps(fscal,cutoff_mask);
1060 fscal = _mm_andnot_ps(dummy_mask,fscal);
1062 /* Calculate temporary vectorial force */
1063 tx = _mm_mul_ps(fscal,dx10);
1064 ty = _mm_mul_ps(fscal,dy10);
1065 tz = _mm_mul_ps(fscal,dz10);
1067 /* Update vectorial force */
1068 fix1 = _mm_add_ps(fix1,tx);
1069 fiy1 = _mm_add_ps(fiy1,ty);
1070 fiz1 = _mm_add_ps(fiz1,tz);
1072 fjx0 = _mm_add_ps(fjx0,tx);
1073 fjy0 = _mm_add_ps(fjy0,ty);
1074 fjz0 = _mm_add_ps(fjz0,tz);
1078 /**************************
1079 * CALCULATE INTERACTIONS *
1080 **************************/
1082 if (gmx_mm_any_lt(rsq20,rcutoff2))
1085 /* Compute parameters for interactions between i and j atoms */
1086 qq20 = _mm_mul_ps(iq2,jq0);
1088 /* REACTION-FIELD ELECTROSTATICS */
1089 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1091 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1095 fscal = _mm_and_ps(fscal,cutoff_mask);
1097 fscal = _mm_andnot_ps(dummy_mask,fscal);
1099 /* Calculate temporary vectorial force */
1100 tx = _mm_mul_ps(fscal,dx20);
1101 ty = _mm_mul_ps(fscal,dy20);
1102 tz = _mm_mul_ps(fscal,dz20);
1104 /* Update vectorial force */
1105 fix2 = _mm_add_ps(fix2,tx);
1106 fiy2 = _mm_add_ps(fiy2,ty);
1107 fiz2 = _mm_add_ps(fiz2,tz);
1109 fjx0 = _mm_add_ps(fjx0,tx);
1110 fjy0 = _mm_add_ps(fjy0,ty);
1111 fjz0 = _mm_add_ps(fjz0,tz);
1115 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1116 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1117 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1118 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1120 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1122 /* Inner loop uses 97 flops */
1125 /* End of innermost loop */
1127 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1128 f+i_coord_offset,fshift+i_shift_offset);
1130 /* Increment number of inner iterations */
1131 inneriter += j_index_end - j_index_start;
1133 /* Outer loop uses 18 flops */
1136 /* Increment number of outer iterations */
1139 /* Update outer/inner flops */
1141 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*97);