2 * Note: this file was generated by the Gromacs sse2_double 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_sse2_double.h"
34 #include "kernelutil_x86_sse2_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_double
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: LennardJones
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_double
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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
61 int j_coord_offsetA,j_coord_offsetB;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
64 real *shiftvec,*fshift,*x,*f;
65 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
69 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
71 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
72 int vdwjidx0A,vdwjidx0B;
73 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
74 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
75 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
76 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
77 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
80 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
83 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
84 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
85 __m128d dummy_mask,cutoff_mask;
86 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
87 __m128d one = _mm_set1_pd(1.0);
88 __m128d two = _mm_set1_pd(2.0);
94 jindex = nlist->jindex;
96 shiftidx = nlist->shift;
98 shiftvec = fr->shift_vec[0];
99 fshift = fr->fshift[0];
100 facel = _mm_set1_pd(fr->epsfac);
101 charge = mdatoms->chargeA;
102 krf = _mm_set1_pd(fr->ic->k_rf);
103 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
104 crf = _mm_set1_pd(fr->ic->c_rf);
105 nvdwtype = fr->ntype;
107 vdwtype = mdatoms->typeA;
109 /* Setup water-specific parameters */
110 inr = nlist->iinr[0];
111 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
112 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
113 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
114 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
116 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
117 rcutoff_scalar = fr->rcoulomb;
118 rcutoff = _mm_set1_pd(rcutoff_scalar);
119 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
121 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
122 rvdw = _mm_set1_pd(fr->rvdw);
124 /* Avoid stupid compiler warnings */
132 /* Start outer loop over neighborlists */
133 for(iidx=0; iidx<nri; iidx++)
135 /* Load shift vector for this list */
136 i_shift_offset = DIM*shiftidx[iidx];
138 /* Load limits for loop over neighbors */
139 j_index_start = jindex[iidx];
140 j_index_end = jindex[iidx+1];
142 /* Get outer coordinate index */
144 i_coord_offset = DIM*inr;
146 /* Load i particle coords and add shift vector */
147 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
148 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
150 fix0 = _mm_setzero_pd();
151 fiy0 = _mm_setzero_pd();
152 fiz0 = _mm_setzero_pd();
153 fix1 = _mm_setzero_pd();
154 fiy1 = _mm_setzero_pd();
155 fiz1 = _mm_setzero_pd();
156 fix2 = _mm_setzero_pd();
157 fiy2 = _mm_setzero_pd();
158 fiz2 = _mm_setzero_pd();
160 /* Reset potential sums */
161 velecsum = _mm_setzero_pd();
162 vvdwsum = _mm_setzero_pd();
164 /* Start inner kernel loop */
165 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
168 /* Get j neighbor index, and coordinate index */
171 j_coord_offsetA = DIM*jnrA;
172 j_coord_offsetB = DIM*jnrB;
174 /* load j atom coordinates */
175 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
178 /* Calculate displacement vector */
179 dx00 = _mm_sub_pd(ix0,jx0);
180 dy00 = _mm_sub_pd(iy0,jy0);
181 dz00 = _mm_sub_pd(iz0,jz0);
182 dx10 = _mm_sub_pd(ix1,jx0);
183 dy10 = _mm_sub_pd(iy1,jy0);
184 dz10 = _mm_sub_pd(iz1,jz0);
185 dx20 = _mm_sub_pd(ix2,jx0);
186 dy20 = _mm_sub_pd(iy2,jy0);
187 dz20 = _mm_sub_pd(iz2,jz0);
189 /* Calculate squared distance and things based on it */
190 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
191 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
192 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
194 rinv00 = gmx_mm_invsqrt_pd(rsq00);
195 rinv10 = gmx_mm_invsqrt_pd(rsq10);
196 rinv20 = gmx_mm_invsqrt_pd(rsq20);
198 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
199 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
200 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
202 /* Load parameters for j particles */
203 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
204 vdwjidx0A = 2*vdwtype[jnrA+0];
205 vdwjidx0B = 2*vdwtype[jnrB+0];
207 fjx0 = _mm_setzero_pd();
208 fjy0 = _mm_setzero_pd();
209 fjz0 = _mm_setzero_pd();
211 /**************************
212 * CALCULATE INTERACTIONS *
213 **************************/
215 if (gmx_mm_any_lt(rsq00,rcutoff2))
218 /* Compute parameters for interactions between i and j atoms */
219 qq00 = _mm_mul_pd(iq0,jq0);
220 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
221 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
223 /* REACTION-FIELD ELECTROSTATICS */
224 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
225 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
227 /* LENNARD-JONES DISPERSION/REPULSION */
229 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
230 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
231 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
232 vvdw = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
233 _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
234 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
236 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
238 /* Update potential sum for this i atom from the interaction with this j atom. */
239 velec = _mm_and_pd(velec,cutoff_mask);
240 velecsum = _mm_add_pd(velecsum,velec);
241 vvdw = _mm_and_pd(vvdw,cutoff_mask);
242 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
244 fscal = _mm_add_pd(felec,fvdw);
246 fscal = _mm_and_pd(fscal,cutoff_mask);
248 /* Calculate temporary vectorial force */
249 tx = _mm_mul_pd(fscal,dx00);
250 ty = _mm_mul_pd(fscal,dy00);
251 tz = _mm_mul_pd(fscal,dz00);
253 /* Update vectorial force */
254 fix0 = _mm_add_pd(fix0,tx);
255 fiy0 = _mm_add_pd(fiy0,ty);
256 fiz0 = _mm_add_pd(fiz0,tz);
258 fjx0 = _mm_add_pd(fjx0,tx);
259 fjy0 = _mm_add_pd(fjy0,ty);
260 fjz0 = _mm_add_pd(fjz0,tz);
264 /**************************
265 * CALCULATE INTERACTIONS *
266 **************************/
268 if (gmx_mm_any_lt(rsq10,rcutoff2))
271 /* Compute parameters for interactions between i and j atoms */
272 qq10 = _mm_mul_pd(iq1,jq0);
274 /* REACTION-FIELD ELECTROSTATICS */
275 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
276 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
278 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
280 /* Update potential sum for this i atom from the interaction with this j atom. */
281 velec = _mm_and_pd(velec,cutoff_mask);
282 velecsum = _mm_add_pd(velecsum,velec);
286 fscal = _mm_and_pd(fscal,cutoff_mask);
288 /* Calculate temporary vectorial force */
289 tx = _mm_mul_pd(fscal,dx10);
290 ty = _mm_mul_pd(fscal,dy10);
291 tz = _mm_mul_pd(fscal,dz10);
293 /* Update vectorial force */
294 fix1 = _mm_add_pd(fix1,tx);
295 fiy1 = _mm_add_pd(fiy1,ty);
296 fiz1 = _mm_add_pd(fiz1,tz);
298 fjx0 = _mm_add_pd(fjx0,tx);
299 fjy0 = _mm_add_pd(fjy0,ty);
300 fjz0 = _mm_add_pd(fjz0,tz);
304 /**************************
305 * CALCULATE INTERACTIONS *
306 **************************/
308 if (gmx_mm_any_lt(rsq20,rcutoff2))
311 /* Compute parameters for interactions between i and j atoms */
312 qq20 = _mm_mul_pd(iq2,jq0);
314 /* REACTION-FIELD ELECTROSTATICS */
315 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
316 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
318 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
320 /* Update potential sum for this i atom from the interaction with this j atom. */
321 velec = _mm_and_pd(velec,cutoff_mask);
322 velecsum = _mm_add_pd(velecsum,velec);
326 fscal = _mm_and_pd(fscal,cutoff_mask);
328 /* Calculate temporary vectorial force */
329 tx = _mm_mul_pd(fscal,dx20);
330 ty = _mm_mul_pd(fscal,dy20);
331 tz = _mm_mul_pd(fscal,dz20);
333 /* Update vectorial force */
334 fix2 = _mm_add_pd(fix2,tx);
335 fiy2 = _mm_add_pd(fiy2,ty);
336 fiz2 = _mm_add_pd(fiz2,tz);
338 fjx0 = _mm_add_pd(fjx0,tx);
339 fjy0 = _mm_add_pd(fjy0,ty);
340 fjz0 = _mm_add_pd(fjz0,tz);
344 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
346 /* Inner loop uses 129 flops */
353 j_coord_offsetA = DIM*jnrA;
355 /* load j atom coordinates */
356 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
359 /* Calculate displacement vector */
360 dx00 = _mm_sub_pd(ix0,jx0);
361 dy00 = _mm_sub_pd(iy0,jy0);
362 dz00 = _mm_sub_pd(iz0,jz0);
363 dx10 = _mm_sub_pd(ix1,jx0);
364 dy10 = _mm_sub_pd(iy1,jy0);
365 dz10 = _mm_sub_pd(iz1,jz0);
366 dx20 = _mm_sub_pd(ix2,jx0);
367 dy20 = _mm_sub_pd(iy2,jy0);
368 dz20 = _mm_sub_pd(iz2,jz0);
370 /* Calculate squared distance and things based on it */
371 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
372 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
373 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
375 rinv00 = gmx_mm_invsqrt_pd(rsq00);
376 rinv10 = gmx_mm_invsqrt_pd(rsq10);
377 rinv20 = gmx_mm_invsqrt_pd(rsq20);
379 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
380 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
381 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
383 /* Load parameters for j particles */
384 jq0 = _mm_load_sd(charge+jnrA+0);
385 vdwjidx0A = 2*vdwtype[jnrA+0];
387 fjx0 = _mm_setzero_pd();
388 fjy0 = _mm_setzero_pd();
389 fjz0 = _mm_setzero_pd();
391 /**************************
392 * CALCULATE INTERACTIONS *
393 **************************/
395 if (gmx_mm_any_lt(rsq00,rcutoff2))
398 /* Compute parameters for interactions between i and j atoms */
399 qq00 = _mm_mul_pd(iq0,jq0);
400 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
402 /* REACTION-FIELD ELECTROSTATICS */
403 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
404 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
406 /* LENNARD-JONES DISPERSION/REPULSION */
408 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
409 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
410 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
411 vvdw = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
412 _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
413 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
415 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
417 /* Update potential sum for this i atom from the interaction with this j atom. */
418 velec = _mm_and_pd(velec,cutoff_mask);
419 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
420 velecsum = _mm_add_pd(velecsum,velec);
421 vvdw = _mm_and_pd(vvdw,cutoff_mask);
422 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
423 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
425 fscal = _mm_add_pd(felec,fvdw);
427 fscal = _mm_and_pd(fscal,cutoff_mask);
429 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
431 /* Calculate temporary vectorial force */
432 tx = _mm_mul_pd(fscal,dx00);
433 ty = _mm_mul_pd(fscal,dy00);
434 tz = _mm_mul_pd(fscal,dz00);
436 /* Update vectorial force */
437 fix0 = _mm_add_pd(fix0,tx);
438 fiy0 = _mm_add_pd(fiy0,ty);
439 fiz0 = _mm_add_pd(fiz0,tz);
441 fjx0 = _mm_add_pd(fjx0,tx);
442 fjy0 = _mm_add_pd(fjy0,ty);
443 fjz0 = _mm_add_pd(fjz0,tz);
447 /**************************
448 * CALCULATE INTERACTIONS *
449 **************************/
451 if (gmx_mm_any_lt(rsq10,rcutoff2))
454 /* Compute parameters for interactions between i and j atoms */
455 qq10 = _mm_mul_pd(iq1,jq0);
457 /* REACTION-FIELD ELECTROSTATICS */
458 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
459 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
461 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
463 /* Update potential sum for this i atom from the interaction with this j atom. */
464 velec = _mm_and_pd(velec,cutoff_mask);
465 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
466 velecsum = _mm_add_pd(velecsum,velec);
470 fscal = _mm_and_pd(fscal,cutoff_mask);
472 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
474 /* Calculate temporary vectorial force */
475 tx = _mm_mul_pd(fscal,dx10);
476 ty = _mm_mul_pd(fscal,dy10);
477 tz = _mm_mul_pd(fscal,dz10);
479 /* Update vectorial force */
480 fix1 = _mm_add_pd(fix1,tx);
481 fiy1 = _mm_add_pd(fiy1,ty);
482 fiz1 = _mm_add_pd(fiz1,tz);
484 fjx0 = _mm_add_pd(fjx0,tx);
485 fjy0 = _mm_add_pd(fjy0,ty);
486 fjz0 = _mm_add_pd(fjz0,tz);
490 /**************************
491 * CALCULATE INTERACTIONS *
492 **************************/
494 if (gmx_mm_any_lt(rsq20,rcutoff2))
497 /* Compute parameters for interactions between i and j atoms */
498 qq20 = _mm_mul_pd(iq2,jq0);
500 /* REACTION-FIELD ELECTROSTATICS */
501 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
502 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
504 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
506 /* Update potential sum for this i atom from the interaction with this j atom. */
507 velec = _mm_and_pd(velec,cutoff_mask);
508 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
509 velecsum = _mm_add_pd(velecsum,velec);
513 fscal = _mm_and_pd(fscal,cutoff_mask);
515 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
517 /* Calculate temporary vectorial force */
518 tx = _mm_mul_pd(fscal,dx20);
519 ty = _mm_mul_pd(fscal,dy20);
520 tz = _mm_mul_pd(fscal,dz20);
522 /* Update vectorial force */
523 fix2 = _mm_add_pd(fix2,tx);
524 fiy2 = _mm_add_pd(fiy2,ty);
525 fiz2 = _mm_add_pd(fiz2,tz);
527 fjx0 = _mm_add_pd(fjx0,tx);
528 fjy0 = _mm_add_pd(fjy0,ty);
529 fjz0 = _mm_add_pd(fjz0,tz);
533 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
535 /* Inner loop uses 129 flops */
538 /* End of innermost loop */
540 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
541 f+i_coord_offset,fshift+i_shift_offset);
544 /* Update potential energies */
545 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
546 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
548 /* Increment number of inner iterations */
549 inneriter += j_index_end - j_index_start;
551 /* Outer loop uses 20 flops */
554 /* Increment number of outer iterations */
557 /* Update outer/inner flops */
559 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
562 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_double
563 * Electrostatics interaction: ReactionField
564 * VdW interaction: LennardJones
565 * Geometry: Water3-Particle
566 * Calculate force/pot: Force
569 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_double
570 (t_nblist * gmx_restrict nlist,
571 rvec * gmx_restrict xx,
572 rvec * gmx_restrict ff,
573 t_forcerec * gmx_restrict fr,
574 t_mdatoms * gmx_restrict mdatoms,
575 nb_kernel_data_t * gmx_restrict kernel_data,
576 t_nrnb * gmx_restrict nrnb)
578 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
579 * just 0 for non-waters.
580 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
581 * jnr indices corresponding to data put in the four positions in the SIMD register.
583 int i_shift_offset,i_coord_offset,outeriter,inneriter;
584 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
586 int j_coord_offsetA,j_coord_offsetB;
587 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
589 real *shiftvec,*fshift,*x,*f;
590 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
592 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
594 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
596 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
597 int vdwjidx0A,vdwjidx0B;
598 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
599 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
600 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
601 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
602 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
605 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
608 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
609 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
610 __m128d dummy_mask,cutoff_mask;
611 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
612 __m128d one = _mm_set1_pd(1.0);
613 __m128d two = _mm_set1_pd(2.0);
619 jindex = nlist->jindex;
621 shiftidx = nlist->shift;
623 shiftvec = fr->shift_vec[0];
624 fshift = fr->fshift[0];
625 facel = _mm_set1_pd(fr->epsfac);
626 charge = mdatoms->chargeA;
627 krf = _mm_set1_pd(fr->ic->k_rf);
628 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
629 crf = _mm_set1_pd(fr->ic->c_rf);
630 nvdwtype = fr->ntype;
632 vdwtype = mdatoms->typeA;
634 /* Setup water-specific parameters */
635 inr = nlist->iinr[0];
636 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
637 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
638 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
639 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
641 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
642 rcutoff_scalar = fr->rcoulomb;
643 rcutoff = _mm_set1_pd(rcutoff_scalar);
644 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
646 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
647 rvdw = _mm_set1_pd(fr->rvdw);
649 /* Avoid stupid compiler warnings */
657 /* Start outer loop over neighborlists */
658 for(iidx=0; iidx<nri; iidx++)
660 /* Load shift vector for this list */
661 i_shift_offset = DIM*shiftidx[iidx];
663 /* Load limits for loop over neighbors */
664 j_index_start = jindex[iidx];
665 j_index_end = jindex[iidx+1];
667 /* Get outer coordinate index */
669 i_coord_offset = DIM*inr;
671 /* Load i particle coords and add shift vector */
672 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
673 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
675 fix0 = _mm_setzero_pd();
676 fiy0 = _mm_setzero_pd();
677 fiz0 = _mm_setzero_pd();
678 fix1 = _mm_setzero_pd();
679 fiy1 = _mm_setzero_pd();
680 fiz1 = _mm_setzero_pd();
681 fix2 = _mm_setzero_pd();
682 fiy2 = _mm_setzero_pd();
683 fiz2 = _mm_setzero_pd();
685 /* Start inner kernel loop */
686 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
689 /* Get j neighbor index, and coordinate index */
692 j_coord_offsetA = DIM*jnrA;
693 j_coord_offsetB = DIM*jnrB;
695 /* load j atom coordinates */
696 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
699 /* Calculate displacement vector */
700 dx00 = _mm_sub_pd(ix0,jx0);
701 dy00 = _mm_sub_pd(iy0,jy0);
702 dz00 = _mm_sub_pd(iz0,jz0);
703 dx10 = _mm_sub_pd(ix1,jx0);
704 dy10 = _mm_sub_pd(iy1,jy0);
705 dz10 = _mm_sub_pd(iz1,jz0);
706 dx20 = _mm_sub_pd(ix2,jx0);
707 dy20 = _mm_sub_pd(iy2,jy0);
708 dz20 = _mm_sub_pd(iz2,jz0);
710 /* Calculate squared distance and things based on it */
711 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
712 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
713 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
715 rinv00 = gmx_mm_invsqrt_pd(rsq00);
716 rinv10 = gmx_mm_invsqrt_pd(rsq10);
717 rinv20 = gmx_mm_invsqrt_pd(rsq20);
719 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
720 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
721 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
723 /* Load parameters for j particles */
724 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
725 vdwjidx0A = 2*vdwtype[jnrA+0];
726 vdwjidx0B = 2*vdwtype[jnrB+0];
728 fjx0 = _mm_setzero_pd();
729 fjy0 = _mm_setzero_pd();
730 fjz0 = _mm_setzero_pd();
732 /**************************
733 * CALCULATE INTERACTIONS *
734 **************************/
736 if (gmx_mm_any_lt(rsq00,rcutoff2))
739 /* Compute parameters for interactions between i and j atoms */
740 qq00 = _mm_mul_pd(iq0,jq0);
741 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
742 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
744 /* REACTION-FIELD ELECTROSTATICS */
745 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
747 /* LENNARD-JONES DISPERSION/REPULSION */
749 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
750 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
752 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
754 fscal = _mm_add_pd(felec,fvdw);
756 fscal = _mm_and_pd(fscal,cutoff_mask);
758 /* Calculate temporary vectorial force */
759 tx = _mm_mul_pd(fscal,dx00);
760 ty = _mm_mul_pd(fscal,dy00);
761 tz = _mm_mul_pd(fscal,dz00);
763 /* Update vectorial force */
764 fix0 = _mm_add_pd(fix0,tx);
765 fiy0 = _mm_add_pd(fiy0,ty);
766 fiz0 = _mm_add_pd(fiz0,tz);
768 fjx0 = _mm_add_pd(fjx0,tx);
769 fjy0 = _mm_add_pd(fjy0,ty);
770 fjz0 = _mm_add_pd(fjz0,tz);
774 /**************************
775 * CALCULATE INTERACTIONS *
776 **************************/
778 if (gmx_mm_any_lt(rsq10,rcutoff2))
781 /* Compute parameters for interactions between i and j atoms */
782 qq10 = _mm_mul_pd(iq1,jq0);
784 /* REACTION-FIELD ELECTROSTATICS */
785 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
787 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
791 fscal = _mm_and_pd(fscal,cutoff_mask);
793 /* Calculate temporary vectorial force */
794 tx = _mm_mul_pd(fscal,dx10);
795 ty = _mm_mul_pd(fscal,dy10);
796 tz = _mm_mul_pd(fscal,dz10);
798 /* Update vectorial force */
799 fix1 = _mm_add_pd(fix1,tx);
800 fiy1 = _mm_add_pd(fiy1,ty);
801 fiz1 = _mm_add_pd(fiz1,tz);
803 fjx0 = _mm_add_pd(fjx0,tx);
804 fjy0 = _mm_add_pd(fjy0,ty);
805 fjz0 = _mm_add_pd(fjz0,tz);
809 /**************************
810 * CALCULATE INTERACTIONS *
811 **************************/
813 if (gmx_mm_any_lt(rsq20,rcutoff2))
816 /* Compute parameters for interactions between i and j atoms */
817 qq20 = _mm_mul_pd(iq2,jq0);
819 /* REACTION-FIELD ELECTROSTATICS */
820 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
822 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
826 fscal = _mm_and_pd(fscal,cutoff_mask);
828 /* Calculate temporary vectorial force */
829 tx = _mm_mul_pd(fscal,dx20);
830 ty = _mm_mul_pd(fscal,dy20);
831 tz = _mm_mul_pd(fscal,dz20);
833 /* Update vectorial force */
834 fix2 = _mm_add_pd(fix2,tx);
835 fiy2 = _mm_add_pd(fiy2,ty);
836 fiz2 = _mm_add_pd(fiz2,tz);
838 fjx0 = _mm_add_pd(fjx0,tx);
839 fjy0 = _mm_add_pd(fjy0,ty);
840 fjz0 = _mm_add_pd(fjz0,tz);
844 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
846 /* Inner loop uses 100 flops */
853 j_coord_offsetA = DIM*jnrA;
855 /* load j atom coordinates */
856 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
859 /* Calculate displacement vector */
860 dx00 = _mm_sub_pd(ix0,jx0);
861 dy00 = _mm_sub_pd(iy0,jy0);
862 dz00 = _mm_sub_pd(iz0,jz0);
863 dx10 = _mm_sub_pd(ix1,jx0);
864 dy10 = _mm_sub_pd(iy1,jy0);
865 dz10 = _mm_sub_pd(iz1,jz0);
866 dx20 = _mm_sub_pd(ix2,jx0);
867 dy20 = _mm_sub_pd(iy2,jy0);
868 dz20 = _mm_sub_pd(iz2,jz0);
870 /* Calculate squared distance and things based on it */
871 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
872 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
873 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
875 rinv00 = gmx_mm_invsqrt_pd(rsq00);
876 rinv10 = gmx_mm_invsqrt_pd(rsq10);
877 rinv20 = gmx_mm_invsqrt_pd(rsq20);
879 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
880 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
881 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
883 /* Load parameters for j particles */
884 jq0 = _mm_load_sd(charge+jnrA+0);
885 vdwjidx0A = 2*vdwtype[jnrA+0];
887 fjx0 = _mm_setzero_pd();
888 fjy0 = _mm_setzero_pd();
889 fjz0 = _mm_setzero_pd();
891 /**************************
892 * CALCULATE INTERACTIONS *
893 **************************/
895 if (gmx_mm_any_lt(rsq00,rcutoff2))
898 /* Compute parameters for interactions between i and j atoms */
899 qq00 = _mm_mul_pd(iq0,jq0);
900 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
902 /* REACTION-FIELD ELECTROSTATICS */
903 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
905 /* LENNARD-JONES DISPERSION/REPULSION */
907 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
908 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
910 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
912 fscal = _mm_add_pd(felec,fvdw);
914 fscal = _mm_and_pd(fscal,cutoff_mask);
916 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
918 /* Calculate temporary vectorial force */
919 tx = _mm_mul_pd(fscal,dx00);
920 ty = _mm_mul_pd(fscal,dy00);
921 tz = _mm_mul_pd(fscal,dz00);
923 /* Update vectorial force */
924 fix0 = _mm_add_pd(fix0,tx);
925 fiy0 = _mm_add_pd(fiy0,ty);
926 fiz0 = _mm_add_pd(fiz0,tz);
928 fjx0 = _mm_add_pd(fjx0,tx);
929 fjy0 = _mm_add_pd(fjy0,ty);
930 fjz0 = _mm_add_pd(fjz0,tz);
934 /**************************
935 * CALCULATE INTERACTIONS *
936 **************************/
938 if (gmx_mm_any_lt(rsq10,rcutoff2))
941 /* Compute parameters for interactions between i and j atoms */
942 qq10 = _mm_mul_pd(iq1,jq0);
944 /* REACTION-FIELD ELECTROSTATICS */
945 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
947 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
951 fscal = _mm_and_pd(fscal,cutoff_mask);
953 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
955 /* Calculate temporary vectorial force */
956 tx = _mm_mul_pd(fscal,dx10);
957 ty = _mm_mul_pd(fscal,dy10);
958 tz = _mm_mul_pd(fscal,dz10);
960 /* Update vectorial force */
961 fix1 = _mm_add_pd(fix1,tx);
962 fiy1 = _mm_add_pd(fiy1,ty);
963 fiz1 = _mm_add_pd(fiz1,tz);
965 fjx0 = _mm_add_pd(fjx0,tx);
966 fjy0 = _mm_add_pd(fjy0,ty);
967 fjz0 = _mm_add_pd(fjz0,tz);
971 /**************************
972 * CALCULATE INTERACTIONS *
973 **************************/
975 if (gmx_mm_any_lt(rsq20,rcutoff2))
978 /* Compute parameters for interactions between i and j atoms */
979 qq20 = _mm_mul_pd(iq2,jq0);
981 /* REACTION-FIELD ELECTROSTATICS */
982 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
984 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
988 fscal = _mm_and_pd(fscal,cutoff_mask);
990 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
992 /* Calculate temporary vectorial force */
993 tx = _mm_mul_pd(fscal,dx20);
994 ty = _mm_mul_pd(fscal,dy20);
995 tz = _mm_mul_pd(fscal,dz20);
997 /* Update vectorial force */
998 fix2 = _mm_add_pd(fix2,tx);
999 fiy2 = _mm_add_pd(fiy2,ty);
1000 fiz2 = _mm_add_pd(fiz2,tz);
1002 fjx0 = _mm_add_pd(fjx0,tx);
1003 fjy0 = _mm_add_pd(fjy0,ty);
1004 fjz0 = _mm_add_pd(fjz0,tz);
1008 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1010 /* Inner loop uses 100 flops */
1013 /* End of innermost loop */
1015 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1016 f+i_coord_offset,fshift+i_shift_offset);
1018 /* Increment number of inner iterations */
1019 inneriter += j_index_end - j_index_start;
1021 /* Outer loop uses 18 flops */
1024 /* Increment number of outer iterations */
1027 /* Update outer/inner flops */
1029 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);