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_ElecCoul_VdwLJ_GeomW3P1_VF_sse2_double
38 * Electrostatics interaction: Coulomb
39 * VdW interaction: LennardJones
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCoul_VdwLJ_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 nvdwtype = fr->ntype;
104 vdwtype = mdatoms->typeA;
106 /* Setup water-specific parameters */
107 inr = nlist->iinr[0];
108 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
109 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
110 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
111 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
113 /* Avoid stupid compiler warnings */
121 /* Start outer loop over neighborlists */
122 for(iidx=0; iidx<nri; iidx++)
124 /* Load shift vector for this list */
125 i_shift_offset = DIM*shiftidx[iidx];
127 /* Load limits for loop over neighbors */
128 j_index_start = jindex[iidx];
129 j_index_end = jindex[iidx+1];
131 /* Get outer coordinate index */
133 i_coord_offset = DIM*inr;
135 /* Load i particle coords and add shift vector */
136 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
137 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
139 fix0 = _mm_setzero_pd();
140 fiy0 = _mm_setzero_pd();
141 fiz0 = _mm_setzero_pd();
142 fix1 = _mm_setzero_pd();
143 fiy1 = _mm_setzero_pd();
144 fiz1 = _mm_setzero_pd();
145 fix2 = _mm_setzero_pd();
146 fiy2 = _mm_setzero_pd();
147 fiz2 = _mm_setzero_pd();
149 /* Reset potential sums */
150 velecsum = _mm_setzero_pd();
151 vvdwsum = _mm_setzero_pd();
153 /* Start inner kernel loop */
154 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
157 /* Get j neighbor index, and coordinate index */
160 j_coord_offsetA = DIM*jnrA;
161 j_coord_offsetB = DIM*jnrB;
163 /* load j atom coordinates */
164 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
167 /* Calculate displacement vector */
168 dx00 = _mm_sub_pd(ix0,jx0);
169 dy00 = _mm_sub_pd(iy0,jy0);
170 dz00 = _mm_sub_pd(iz0,jz0);
171 dx10 = _mm_sub_pd(ix1,jx0);
172 dy10 = _mm_sub_pd(iy1,jy0);
173 dz10 = _mm_sub_pd(iz1,jz0);
174 dx20 = _mm_sub_pd(ix2,jx0);
175 dy20 = _mm_sub_pd(iy2,jy0);
176 dz20 = _mm_sub_pd(iz2,jz0);
178 /* Calculate squared distance and things based on it */
179 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
180 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
181 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
183 rinv00 = gmx_mm_invsqrt_pd(rsq00);
184 rinv10 = gmx_mm_invsqrt_pd(rsq10);
185 rinv20 = gmx_mm_invsqrt_pd(rsq20);
187 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
188 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
189 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
191 /* Load parameters for j particles */
192 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
193 vdwjidx0A = 2*vdwtype[jnrA+0];
194 vdwjidx0B = 2*vdwtype[jnrB+0];
196 fjx0 = _mm_setzero_pd();
197 fjy0 = _mm_setzero_pd();
198 fjz0 = _mm_setzero_pd();
200 /**************************
201 * CALCULATE INTERACTIONS *
202 **************************/
204 /* Compute parameters for interactions between i and j atoms */
205 qq00 = _mm_mul_pd(iq0,jq0);
206 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
207 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
209 /* COULOMB ELECTROSTATICS */
210 velec = _mm_mul_pd(qq00,rinv00);
211 felec = _mm_mul_pd(velec,rinvsq00);
213 /* LENNARD-JONES DISPERSION/REPULSION */
215 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
216 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
217 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
218 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
219 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
221 /* Update potential sum for this i atom from the interaction with this j atom. */
222 velecsum = _mm_add_pd(velecsum,velec);
223 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
225 fscal = _mm_add_pd(felec,fvdw);
227 /* Calculate temporary vectorial force */
228 tx = _mm_mul_pd(fscal,dx00);
229 ty = _mm_mul_pd(fscal,dy00);
230 tz = _mm_mul_pd(fscal,dz00);
232 /* Update vectorial force */
233 fix0 = _mm_add_pd(fix0,tx);
234 fiy0 = _mm_add_pd(fiy0,ty);
235 fiz0 = _mm_add_pd(fiz0,tz);
237 fjx0 = _mm_add_pd(fjx0,tx);
238 fjy0 = _mm_add_pd(fjy0,ty);
239 fjz0 = _mm_add_pd(fjz0,tz);
241 /**************************
242 * CALCULATE INTERACTIONS *
243 **************************/
245 /* Compute parameters for interactions between i and j atoms */
246 qq10 = _mm_mul_pd(iq1,jq0);
248 /* COULOMB ELECTROSTATICS */
249 velec = _mm_mul_pd(qq10,rinv10);
250 felec = _mm_mul_pd(velec,rinvsq10);
252 /* Update potential sum for this i atom from the interaction with this j atom. */
253 velecsum = _mm_add_pd(velecsum,velec);
257 /* Calculate temporary vectorial force */
258 tx = _mm_mul_pd(fscal,dx10);
259 ty = _mm_mul_pd(fscal,dy10);
260 tz = _mm_mul_pd(fscal,dz10);
262 /* Update vectorial force */
263 fix1 = _mm_add_pd(fix1,tx);
264 fiy1 = _mm_add_pd(fiy1,ty);
265 fiz1 = _mm_add_pd(fiz1,tz);
267 fjx0 = _mm_add_pd(fjx0,tx);
268 fjy0 = _mm_add_pd(fjy0,ty);
269 fjz0 = _mm_add_pd(fjz0,tz);
271 /**************************
272 * CALCULATE INTERACTIONS *
273 **************************/
275 /* Compute parameters for interactions between i and j atoms */
276 qq20 = _mm_mul_pd(iq2,jq0);
278 /* COULOMB ELECTROSTATICS */
279 velec = _mm_mul_pd(qq20,rinv20);
280 felec = _mm_mul_pd(velec,rinvsq20);
282 /* Update potential sum for this i atom from the interaction with this j atom. */
283 velecsum = _mm_add_pd(velecsum,velec);
287 /* Calculate temporary vectorial force */
288 tx = _mm_mul_pd(fscal,dx20);
289 ty = _mm_mul_pd(fscal,dy20);
290 tz = _mm_mul_pd(fscal,dz20);
292 /* Update vectorial force */
293 fix2 = _mm_add_pd(fix2,tx);
294 fiy2 = _mm_add_pd(fiy2,ty);
295 fiz2 = _mm_add_pd(fiz2,tz);
297 fjx0 = _mm_add_pd(fjx0,tx);
298 fjy0 = _mm_add_pd(fjy0,ty);
299 fjz0 = _mm_add_pd(fjz0,tz);
301 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
303 /* Inner loop uses 99 flops */
310 j_coord_offsetA = DIM*jnrA;
312 /* load j atom coordinates */
313 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
316 /* Calculate displacement vector */
317 dx00 = _mm_sub_pd(ix0,jx0);
318 dy00 = _mm_sub_pd(iy0,jy0);
319 dz00 = _mm_sub_pd(iz0,jz0);
320 dx10 = _mm_sub_pd(ix1,jx0);
321 dy10 = _mm_sub_pd(iy1,jy0);
322 dz10 = _mm_sub_pd(iz1,jz0);
323 dx20 = _mm_sub_pd(ix2,jx0);
324 dy20 = _mm_sub_pd(iy2,jy0);
325 dz20 = _mm_sub_pd(iz2,jz0);
327 /* Calculate squared distance and things based on it */
328 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
329 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
330 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
332 rinv00 = gmx_mm_invsqrt_pd(rsq00);
333 rinv10 = gmx_mm_invsqrt_pd(rsq10);
334 rinv20 = gmx_mm_invsqrt_pd(rsq20);
336 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
337 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
338 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
340 /* Load parameters for j particles */
341 jq0 = _mm_load_sd(charge+jnrA+0);
342 vdwjidx0A = 2*vdwtype[jnrA+0];
344 fjx0 = _mm_setzero_pd();
345 fjy0 = _mm_setzero_pd();
346 fjz0 = _mm_setzero_pd();
348 /**************************
349 * CALCULATE INTERACTIONS *
350 **************************/
352 /* Compute parameters for interactions between i and j atoms */
353 qq00 = _mm_mul_pd(iq0,jq0);
354 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
356 /* COULOMB ELECTROSTATICS */
357 velec = _mm_mul_pd(qq00,rinv00);
358 felec = _mm_mul_pd(velec,rinvsq00);
360 /* LENNARD-JONES DISPERSION/REPULSION */
362 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
363 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
364 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
365 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
366 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
368 /* Update potential sum for this i atom from the interaction with this j atom. */
369 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
370 velecsum = _mm_add_pd(velecsum,velec);
371 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
372 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
374 fscal = _mm_add_pd(felec,fvdw);
376 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
378 /* Calculate temporary vectorial force */
379 tx = _mm_mul_pd(fscal,dx00);
380 ty = _mm_mul_pd(fscal,dy00);
381 tz = _mm_mul_pd(fscal,dz00);
383 /* Update vectorial force */
384 fix0 = _mm_add_pd(fix0,tx);
385 fiy0 = _mm_add_pd(fiy0,ty);
386 fiz0 = _mm_add_pd(fiz0,tz);
388 fjx0 = _mm_add_pd(fjx0,tx);
389 fjy0 = _mm_add_pd(fjy0,ty);
390 fjz0 = _mm_add_pd(fjz0,tz);
392 /**************************
393 * CALCULATE INTERACTIONS *
394 **************************/
396 /* Compute parameters for interactions between i and j atoms */
397 qq10 = _mm_mul_pd(iq1,jq0);
399 /* COULOMB ELECTROSTATICS */
400 velec = _mm_mul_pd(qq10,rinv10);
401 felec = _mm_mul_pd(velec,rinvsq10);
403 /* Update potential sum for this i atom from the interaction with this j atom. */
404 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
405 velecsum = _mm_add_pd(velecsum,velec);
409 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
411 /* Calculate temporary vectorial force */
412 tx = _mm_mul_pd(fscal,dx10);
413 ty = _mm_mul_pd(fscal,dy10);
414 tz = _mm_mul_pd(fscal,dz10);
416 /* Update vectorial force */
417 fix1 = _mm_add_pd(fix1,tx);
418 fiy1 = _mm_add_pd(fiy1,ty);
419 fiz1 = _mm_add_pd(fiz1,tz);
421 fjx0 = _mm_add_pd(fjx0,tx);
422 fjy0 = _mm_add_pd(fjy0,ty);
423 fjz0 = _mm_add_pd(fjz0,tz);
425 /**************************
426 * CALCULATE INTERACTIONS *
427 **************************/
429 /* Compute parameters for interactions between i and j atoms */
430 qq20 = _mm_mul_pd(iq2,jq0);
432 /* COULOMB ELECTROSTATICS */
433 velec = _mm_mul_pd(qq20,rinv20);
434 felec = _mm_mul_pd(velec,rinvsq20);
436 /* Update potential sum for this i atom from the interaction with this j atom. */
437 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
438 velecsum = _mm_add_pd(velecsum,velec);
442 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
444 /* Calculate temporary vectorial force */
445 tx = _mm_mul_pd(fscal,dx20);
446 ty = _mm_mul_pd(fscal,dy20);
447 tz = _mm_mul_pd(fscal,dz20);
449 /* Update vectorial force */
450 fix2 = _mm_add_pd(fix2,tx);
451 fiy2 = _mm_add_pd(fiy2,ty);
452 fiz2 = _mm_add_pd(fiz2,tz);
454 fjx0 = _mm_add_pd(fjx0,tx);
455 fjy0 = _mm_add_pd(fjy0,ty);
456 fjz0 = _mm_add_pd(fjz0,tz);
458 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
460 /* Inner loop uses 99 flops */
463 /* End of innermost loop */
465 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
466 f+i_coord_offset,fshift+i_shift_offset);
469 /* Update potential energies */
470 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
471 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
473 /* Increment number of inner iterations */
474 inneriter += j_index_end - j_index_start;
476 /* Outer loop uses 20 flops */
479 /* Increment number of outer iterations */
482 /* Update outer/inner flops */
484 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*99);
487 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse2_double
488 * Electrostatics interaction: Coulomb
489 * VdW interaction: LennardJones
490 * Geometry: Water3-Particle
491 * Calculate force/pot: Force
494 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse2_double
495 (t_nblist * gmx_restrict nlist,
496 rvec * gmx_restrict xx,
497 rvec * gmx_restrict ff,
498 t_forcerec * gmx_restrict fr,
499 t_mdatoms * gmx_restrict mdatoms,
500 nb_kernel_data_t * gmx_restrict kernel_data,
501 t_nrnb * gmx_restrict nrnb)
503 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
504 * just 0 for non-waters.
505 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
506 * jnr indices corresponding to data put in the four positions in the SIMD register.
508 int i_shift_offset,i_coord_offset,outeriter,inneriter;
509 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
511 int j_coord_offsetA,j_coord_offsetB;
512 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
514 real *shiftvec,*fshift,*x,*f;
515 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
517 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
519 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
521 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
522 int vdwjidx0A,vdwjidx0B;
523 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
524 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
525 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
526 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
527 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
530 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
533 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
534 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
535 __m128d dummy_mask,cutoff_mask;
536 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
537 __m128d one = _mm_set1_pd(1.0);
538 __m128d two = _mm_set1_pd(2.0);
544 jindex = nlist->jindex;
546 shiftidx = nlist->shift;
548 shiftvec = fr->shift_vec[0];
549 fshift = fr->fshift[0];
550 facel = _mm_set1_pd(fr->epsfac);
551 charge = mdatoms->chargeA;
552 nvdwtype = fr->ntype;
554 vdwtype = mdatoms->typeA;
556 /* Setup water-specific parameters */
557 inr = nlist->iinr[0];
558 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
559 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
560 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
561 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
563 /* Avoid stupid compiler warnings */
571 /* Start outer loop over neighborlists */
572 for(iidx=0; iidx<nri; iidx++)
574 /* Load shift vector for this list */
575 i_shift_offset = DIM*shiftidx[iidx];
577 /* Load limits for loop over neighbors */
578 j_index_start = jindex[iidx];
579 j_index_end = jindex[iidx+1];
581 /* Get outer coordinate index */
583 i_coord_offset = DIM*inr;
585 /* Load i particle coords and add shift vector */
586 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
587 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
589 fix0 = _mm_setzero_pd();
590 fiy0 = _mm_setzero_pd();
591 fiz0 = _mm_setzero_pd();
592 fix1 = _mm_setzero_pd();
593 fiy1 = _mm_setzero_pd();
594 fiz1 = _mm_setzero_pd();
595 fix2 = _mm_setzero_pd();
596 fiy2 = _mm_setzero_pd();
597 fiz2 = _mm_setzero_pd();
599 /* Start inner kernel loop */
600 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
603 /* Get j neighbor index, and coordinate index */
606 j_coord_offsetA = DIM*jnrA;
607 j_coord_offsetB = DIM*jnrB;
609 /* load j atom coordinates */
610 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
613 /* Calculate displacement vector */
614 dx00 = _mm_sub_pd(ix0,jx0);
615 dy00 = _mm_sub_pd(iy0,jy0);
616 dz00 = _mm_sub_pd(iz0,jz0);
617 dx10 = _mm_sub_pd(ix1,jx0);
618 dy10 = _mm_sub_pd(iy1,jy0);
619 dz10 = _mm_sub_pd(iz1,jz0);
620 dx20 = _mm_sub_pd(ix2,jx0);
621 dy20 = _mm_sub_pd(iy2,jy0);
622 dz20 = _mm_sub_pd(iz2,jz0);
624 /* Calculate squared distance and things based on it */
625 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
626 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
627 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
629 rinv00 = gmx_mm_invsqrt_pd(rsq00);
630 rinv10 = gmx_mm_invsqrt_pd(rsq10);
631 rinv20 = gmx_mm_invsqrt_pd(rsq20);
633 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
634 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
635 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
637 /* Load parameters for j particles */
638 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
639 vdwjidx0A = 2*vdwtype[jnrA+0];
640 vdwjidx0B = 2*vdwtype[jnrB+0];
642 fjx0 = _mm_setzero_pd();
643 fjy0 = _mm_setzero_pd();
644 fjz0 = _mm_setzero_pd();
646 /**************************
647 * CALCULATE INTERACTIONS *
648 **************************/
650 /* Compute parameters for interactions between i and j atoms */
651 qq00 = _mm_mul_pd(iq0,jq0);
652 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
653 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
655 /* COULOMB ELECTROSTATICS */
656 velec = _mm_mul_pd(qq00,rinv00);
657 felec = _mm_mul_pd(velec,rinvsq00);
659 /* LENNARD-JONES DISPERSION/REPULSION */
661 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
662 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
664 fscal = _mm_add_pd(felec,fvdw);
666 /* Calculate temporary vectorial force */
667 tx = _mm_mul_pd(fscal,dx00);
668 ty = _mm_mul_pd(fscal,dy00);
669 tz = _mm_mul_pd(fscal,dz00);
671 /* Update vectorial force */
672 fix0 = _mm_add_pd(fix0,tx);
673 fiy0 = _mm_add_pd(fiy0,ty);
674 fiz0 = _mm_add_pd(fiz0,tz);
676 fjx0 = _mm_add_pd(fjx0,tx);
677 fjy0 = _mm_add_pd(fjy0,ty);
678 fjz0 = _mm_add_pd(fjz0,tz);
680 /**************************
681 * CALCULATE INTERACTIONS *
682 **************************/
684 /* Compute parameters for interactions between i and j atoms */
685 qq10 = _mm_mul_pd(iq1,jq0);
687 /* COULOMB ELECTROSTATICS */
688 velec = _mm_mul_pd(qq10,rinv10);
689 felec = _mm_mul_pd(velec,rinvsq10);
693 /* Calculate temporary vectorial force */
694 tx = _mm_mul_pd(fscal,dx10);
695 ty = _mm_mul_pd(fscal,dy10);
696 tz = _mm_mul_pd(fscal,dz10);
698 /* Update vectorial force */
699 fix1 = _mm_add_pd(fix1,tx);
700 fiy1 = _mm_add_pd(fiy1,ty);
701 fiz1 = _mm_add_pd(fiz1,tz);
703 fjx0 = _mm_add_pd(fjx0,tx);
704 fjy0 = _mm_add_pd(fjy0,ty);
705 fjz0 = _mm_add_pd(fjz0,tz);
707 /**************************
708 * CALCULATE INTERACTIONS *
709 **************************/
711 /* Compute parameters for interactions between i and j atoms */
712 qq20 = _mm_mul_pd(iq2,jq0);
714 /* COULOMB ELECTROSTATICS */
715 velec = _mm_mul_pd(qq20,rinv20);
716 felec = _mm_mul_pd(velec,rinvsq20);
720 /* Calculate temporary vectorial force */
721 tx = _mm_mul_pd(fscal,dx20);
722 ty = _mm_mul_pd(fscal,dy20);
723 tz = _mm_mul_pd(fscal,dz20);
725 /* Update vectorial force */
726 fix2 = _mm_add_pd(fix2,tx);
727 fiy2 = _mm_add_pd(fiy2,ty);
728 fiz2 = _mm_add_pd(fiz2,tz);
730 fjx0 = _mm_add_pd(fjx0,tx);
731 fjy0 = _mm_add_pd(fjy0,ty);
732 fjz0 = _mm_add_pd(fjz0,tz);
734 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
736 /* Inner loop uses 91 flops */
743 j_coord_offsetA = DIM*jnrA;
745 /* load j atom coordinates */
746 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
749 /* Calculate displacement vector */
750 dx00 = _mm_sub_pd(ix0,jx0);
751 dy00 = _mm_sub_pd(iy0,jy0);
752 dz00 = _mm_sub_pd(iz0,jz0);
753 dx10 = _mm_sub_pd(ix1,jx0);
754 dy10 = _mm_sub_pd(iy1,jy0);
755 dz10 = _mm_sub_pd(iz1,jz0);
756 dx20 = _mm_sub_pd(ix2,jx0);
757 dy20 = _mm_sub_pd(iy2,jy0);
758 dz20 = _mm_sub_pd(iz2,jz0);
760 /* Calculate squared distance and things based on it */
761 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
762 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
763 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
765 rinv00 = gmx_mm_invsqrt_pd(rsq00);
766 rinv10 = gmx_mm_invsqrt_pd(rsq10);
767 rinv20 = gmx_mm_invsqrt_pd(rsq20);
769 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
770 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
771 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
773 /* Load parameters for j particles */
774 jq0 = _mm_load_sd(charge+jnrA+0);
775 vdwjidx0A = 2*vdwtype[jnrA+0];
777 fjx0 = _mm_setzero_pd();
778 fjy0 = _mm_setzero_pd();
779 fjz0 = _mm_setzero_pd();
781 /**************************
782 * CALCULATE INTERACTIONS *
783 **************************/
785 /* Compute parameters for interactions between i and j atoms */
786 qq00 = _mm_mul_pd(iq0,jq0);
787 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
789 /* COULOMB ELECTROSTATICS */
790 velec = _mm_mul_pd(qq00,rinv00);
791 felec = _mm_mul_pd(velec,rinvsq00);
793 /* LENNARD-JONES DISPERSION/REPULSION */
795 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
796 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
798 fscal = _mm_add_pd(felec,fvdw);
800 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
802 /* Calculate temporary vectorial force */
803 tx = _mm_mul_pd(fscal,dx00);
804 ty = _mm_mul_pd(fscal,dy00);
805 tz = _mm_mul_pd(fscal,dz00);
807 /* Update vectorial force */
808 fix0 = _mm_add_pd(fix0,tx);
809 fiy0 = _mm_add_pd(fiy0,ty);
810 fiz0 = _mm_add_pd(fiz0,tz);
812 fjx0 = _mm_add_pd(fjx0,tx);
813 fjy0 = _mm_add_pd(fjy0,ty);
814 fjz0 = _mm_add_pd(fjz0,tz);
816 /**************************
817 * CALCULATE INTERACTIONS *
818 **************************/
820 /* Compute parameters for interactions between i and j atoms */
821 qq10 = _mm_mul_pd(iq1,jq0);
823 /* COULOMB ELECTROSTATICS */
824 velec = _mm_mul_pd(qq10,rinv10);
825 felec = _mm_mul_pd(velec,rinvsq10);
829 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
831 /* Calculate temporary vectorial force */
832 tx = _mm_mul_pd(fscal,dx10);
833 ty = _mm_mul_pd(fscal,dy10);
834 tz = _mm_mul_pd(fscal,dz10);
836 /* Update vectorial force */
837 fix1 = _mm_add_pd(fix1,tx);
838 fiy1 = _mm_add_pd(fiy1,ty);
839 fiz1 = _mm_add_pd(fiz1,tz);
841 fjx0 = _mm_add_pd(fjx0,tx);
842 fjy0 = _mm_add_pd(fjy0,ty);
843 fjz0 = _mm_add_pd(fjz0,tz);
845 /**************************
846 * CALCULATE INTERACTIONS *
847 **************************/
849 /* Compute parameters for interactions between i and j atoms */
850 qq20 = _mm_mul_pd(iq2,jq0);
852 /* COULOMB ELECTROSTATICS */
853 velec = _mm_mul_pd(qq20,rinv20);
854 felec = _mm_mul_pd(velec,rinvsq20);
858 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
860 /* Calculate temporary vectorial force */
861 tx = _mm_mul_pd(fscal,dx20);
862 ty = _mm_mul_pd(fscal,dy20);
863 tz = _mm_mul_pd(fscal,dz20);
865 /* Update vectorial force */
866 fix2 = _mm_add_pd(fix2,tx);
867 fiy2 = _mm_add_pd(fiy2,ty);
868 fiz2 = _mm_add_pd(fiz2,tz);
870 fjx0 = _mm_add_pd(fjx0,tx);
871 fjy0 = _mm_add_pd(fjy0,ty);
872 fjz0 = _mm_add_pd(fjz0,tz);
874 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
876 /* Inner loop uses 91 flops */
879 /* End of innermost loop */
881 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
882 f+i_coord_offset,fshift+i_shift_offset);
884 /* Increment number of inner iterations */
885 inneriter += j_index_end - j_index_start;
887 /* Outer loop uses 18 flops */
890 /* Increment number of outer iterations */
893 /* Update outer/inner flops */
895 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*91);