2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2017, 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 "gromacs/gmxlib/nrnb.h"
47 #include "kernelutil_x86_sse2_single.h"
50 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_sse2_single
51 * Electrostatics interaction: Coulomb
52 * VdW interaction: None
53 * Geometry: Water4-Particle
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_sse2_single
58 (t_nblist * gmx_restrict nlist,
59 rvec * gmx_restrict xx,
60 rvec * gmx_restrict ff,
61 struct t_forcerec * gmx_restrict fr,
62 t_mdatoms * gmx_restrict mdatoms,
63 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
64 t_nrnb * gmx_restrict nrnb)
66 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
67 * just 0 for non-waters.
68 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
69 * jnr indices corresponding to data put in the four positions in the SIMD register.
71 int i_shift_offset,i_coord_offset,outeriter,inneriter;
72 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
73 int jnrA,jnrB,jnrC,jnrD;
74 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
75 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
81 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
87 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
88 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
89 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
91 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
92 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
93 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
95 __m128 dummy_mask,cutoff_mask;
96 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
97 __m128 one = _mm_set1_ps(1.0);
98 __m128 two = _mm_set1_ps(2.0);
104 jindex = nlist->jindex;
106 shiftidx = nlist->shift;
108 shiftvec = fr->shift_vec[0];
109 fshift = fr->fshift[0];
110 facel = _mm_set1_ps(fr->ic->epsfac);
111 charge = mdatoms->chargeA;
113 /* Setup water-specific parameters */
114 inr = nlist->iinr[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 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
119 /* Avoid stupid compiler warnings */
120 jnrA = jnrB = jnrC = jnrD = 0;
129 for(iidx=0;iidx<4*DIM;iidx++)
134 /* Start outer loop over neighborlists */
135 for(iidx=0; iidx<nri; iidx++)
137 /* Load shift vector for this list */
138 i_shift_offset = DIM*shiftidx[iidx];
140 /* Load limits for loop over neighbors */
141 j_index_start = jindex[iidx];
142 j_index_end = jindex[iidx+1];
144 /* Get outer coordinate index */
146 i_coord_offset = DIM*inr;
148 /* Load i particle coords and add shift vector */
149 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
150 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
152 fix1 = _mm_setzero_ps();
153 fiy1 = _mm_setzero_ps();
154 fiz1 = _mm_setzero_ps();
155 fix2 = _mm_setzero_ps();
156 fiy2 = _mm_setzero_ps();
157 fiz2 = _mm_setzero_ps();
158 fix3 = _mm_setzero_ps();
159 fiy3 = _mm_setzero_ps();
160 fiz3 = _mm_setzero_ps();
162 /* Reset potential sums */
163 velecsum = _mm_setzero_ps();
165 /* Start inner kernel loop */
166 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
169 /* Get j neighbor index, and coordinate index */
174 j_coord_offsetA = DIM*jnrA;
175 j_coord_offsetB = DIM*jnrB;
176 j_coord_offsetC = DIM*jnrC;
177 j_coord_offsetD = DIM*jnrD;
179 /* load j atom coordinates */
180 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
181 x+j_coord_offsetC,x+j_coord_offsetD,
184 /* Calculate displacement vector */
185 dx10 = _mm_sub_ps(ix1,jx0);
186 dy10 = _mm_sub_ps(iy1,jy0);
187 dz10 = _mm_sub_ps(iz1,jz0);
188 dx20 = _mm_sub_ps(ix2,jx0);
189 dy20 = _mm_sub_ps(iy2,jy0);
190 dz20 = _mm_sub_ps(iz2,jz0);
191 dx30 = _mm_sub_ps(ix3,jx0);
192 dy30 = _mm_sub_ps(iy3,jy0);
193 dz30 = _mm_sub_ps(iz3,jz0);
195 /* Calculate squared distance and things based on it */
196 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
197 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
198 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
200 rinv10 = sse2_invsqrt_f(rsq10);
201 rinv20 = sse2_invsqrt_f(rsq20);
202 rinv30 = sse2_invsqrt_f(rsq30);
204 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
205 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
206 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
208 /* Load parameters for j particles */
209 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
210 charge+jnrC+0,charge+jnrD+0);
212 fjx0 = _mm_setzero_ps();
213 fjy0 = _mm_setzero_ps();
214 fjz0 = _mm_setzero_ps();
216 /**************************
217 * CALCULATE INTERACTIONS *
218 **************************/
220 /* Compute parameters for interactions between i and j atoms */
221 qq10 = _mm_mul_ps(iq1,jq0);
223 /* COULOMB ELECTROSTATICS */
224 velec = _mm_mul_ps(qq10,rinv10);
225 felec = _mm_mul_ps(velec,rinvsq10);
227 /* Update potential sum for this i atom from the interaction with this j atom. */
228 velecsum = _mm_add_ps(velecsum,velec);
232 /* Calculate temporary vectorial force */
233 tx = _mm_mul_ps(fscal,dx10);
234 ty = _mm_mul_ps(fscal,dy10);
235 tz = _mm_mul_ps(fscal,dz10);
237 /* Update vectorial force */
238 fix1 = _mm_add_ps(fix1,tx);
239 fiy1 = _mm_add_ps(fiy1,ty);
240 fiz1 = _mm_add_ps(fiz1,tz);
242 fjx0 = _mm_add_ps(fjx0,tx);
243 fjy0 = _mm_add_ps(fjy0,ty);
244 fjz0 = _mm_add_ps(fjz0,tz);
246 /**************************
247 * CALCULATE INTERACTIONS *
248 **************************/
250 /* Compute parameters for interactions between i and j atoms */
251 qq20 = _mm_mul_ps(iq2,jq0);
253 /* COULOMB ELECTROSTATICS */
254 velec = _mm_mul_ps(qq20,rinv20);
255 felec = _mm_mul_ps(velec,rinvsq20);
257 /* Update potential sum for this i atom from the interaction with this j atom. */
258 velecsum = _mm_add_ps(velecsum,velec);
262 /* Calculate temporary vectorial force */
263 tx = _mm_mul_ps(fscal,dx20);
264 ty = _mm_mul_ps(fscal,dy20);
265 tz = _mm_mul_ps(fscal,dz20);
267 /* Update vectorial force */
268 fix2 = _mm_add_ps(fix2,tx);
269 fiy2 = _mm_add_ps(fiy2,ty);
270 fiz2 = _mm_add_ps(fiz2,tz);
272 fjx0 = _mm_add_ps(fjx0,tx);
273 fjy0 = _mm_add_ps(fjy0,ty);
274 fjz0 = _mm_add_ps(fjz0,tz);
276 /**************************
277 * CALCULATE INTERACTIONS *
278 **************************/
280 /* Compute parameters for interactions between i and j atoms */
281 qq30 = _mm_mul_ps(iq3,jq0);
283 /* COULOMB ELECTROSTATICS */
284 velec = _mm_mul_ps(qq30,rinv30);
285 felec = _mm_mul_ps(velec,rinvsq30);
287 /* Update potential sum for this i atom from the interaction with this j atom. */
288 velecsum = _mm_add_ps(velecsum,velec);
292 /* Calculate temporary vectorial force */
293 tx = _mm_mul_ps(fscal,dx30);
294 ty = _mm_mul_ps(fscal,dy30);
295 tz = _mm_mul_ps(fscal,dz30);
297 /* Update vectorial force */
298 fix3 = _mm_add_ps(fix3,tx);
299 fiy3 = _mm_add_ps(fiy3,ty);
300 fiz3 = _mm_add_ps(fiz3,tz);
302 fjx0 = _mm_add_ps(fjx0,tx);
303 fjy0 = _mm_add_ps(fjy0,ty);
304 fjz0 = _mm_add_ps(fjz0,tz);
306 fjptrA = f+j_coord_offsetA;
307 fjptrB = f+j_coord_offsetB;
308 fjptrC = f+j_coord_offsetC;
309 fjptrD = f+j_coord_offsetD;
311 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
313 /* Inner loop uses 84 flops */
319 /* Get j neighbor index, and coordinate index */
320 jnrlistA = jjnr[jidx];
321 jnrlistB = jjnr[jidx+1];
322 jnrlistC = jjnr[jidx+2];
323 jnrlistD = jjnr[jidx+3];
324 /* Sign of each element will be negative for non-real atoms.
325 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
326 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
328 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
329 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
330 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
331 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
332 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
333 j_coord_offsetA = DIM*jnrA;
334 j_coord_offsetB = DIM*jnrB;
335 j_coord_offsetC = DIM*jnrC;
336 j_coord_offsetD = DIM*jnrD;
338 /* load j atom coordinates */
339 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
340 x+j_coord_offsetC,x+j_coord_offsetD,
343 /* Calculate displacement vector */
344 dx10 = _mm_sub_ps(ix1,jx0);
345 dy10 = _mm_sub_ps(iy1,jy0);
346 dz10 = _mm_sub_ps(iz1,jz0);
347 dx20 = _mm_sub_ps(ix2,jx0);
348 dy20 = _mm_sub_ps(iy2,jy0);
349 dz20 = _mm_sub_ps(iz2,jz0);
350 dx30 = _mm_sub_ps(ix3,jx0);
351 dy30 = _mm_sub_ps(iy3,jy0);
352 dz30 = _mm_sub_ps(iz3,jz0);
354 /* Calculate squared distance and things based on it */
355 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
356 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
357 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
359 rinv10 = sse2_invsqrt_f(rsq10);
360 rinv20 = sse2_invsqrt_f(rsq20);
361 rinv30 = sse2_invsqrt_f(rsq30);
363 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
364 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
365 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
367 /* Load parameters for j particles */
368 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
369 charge+jnrC+0,charge+jnrD+0);
371 fjx0 = _mm_setzero_ps();
372 fjy0 = _mm_setzero_ps();
373 fjz0 = _mm_setzero_ps();
375 /**************************
376 * CALCULATE INTERACTIONS *
377 **************************/
379 /* Compute parameters for interactions between i and j atoms */
380 qq10 = _mm_mul_ps(iq1,jq0);
382 /* COULOMB ELECTROSTATICS */
383 velec = _mm_mul_ps(qq10,rinv10);
384 felec = _mm_mul_ps(velec,rinvsq10);
386 /* Update potential sum for this i atom from the interaction with this j atom. */
387 velec = _mm_andnot_ps(dummy_mask,velec);
388 velecsum = _mm_add_ps(velecsum,velec);
392 fscal = _mm_andnot_ps(dummy_mask,fscal);
394 /* Calculate temporary vectorial force */
395 tx = _mm_mul_ps(fscal,dx10);
396 ty = _mm_mul_ps(fscal,dy10);
397 tz = _mm_mul_ps(fscal,dz10);
399 /* Update vectorial force */
400 fix1 = _mm_add_ps(fix1,tx);
401 fiy1 = _mm_add_ps(fiy1,ty);
402 fiz1 = _mm_add_ps(fiz1,tz);
404 fjx0 = _mm_add_ps(fjx0,tx);
405 fjy0 = _mm_add_ps(fjy0,ty);
406 fjz0 = _mm_add_ps(fjz0,tz);
408 /**************************
409 * CALCULATE INTERACTIONS *
410 **************************/
412 /* Compute parameters for interactions between i and j atoms */
413 qq20 = _mm_mul_ps(iq2,jq0);
415 /* COULOMB ELECTROSTATICS */
416 velec = _mm_mul_ps(qq20,rinv20);
417 felec = _mm_mul_ps(velec,rinvsq20);
419 /* Update potential sum for this i atom from the interaction with this j atom. */
420 velec = _mm_andnot_ps(dummy_mask,velec);
421 velecsum = _mm_add_ps(velecsum,velec);
425 fscal = _mm_andnot_ps(dummy_mask,fscal);
427 /* Calculate temporary vectorial force */
428 tx = _mm_mul_ps(fscal,dx20);
429 ty = _mm_mul_ps(fscal,dy20);
430 tz = _mm_mul_ps(fscal,dz20);
432 /* Update vectorial force */
433 fix2 = _mm_add_ps(fix2,tx);
434 fiy2 = _mm_add_ps(fiy2,ty);
435 fiz2 = _mm_add_ps(fiz2,tz);
437 fjx0 = _mm_add_ps(fjx0,tx);
438 fjy0 = _mm_add_ps(fjy0,ty);
439 fjz0 = _mm_add_ps(fjz0,tz);
441 /**************************
442 * CALCULATE INTERACTIONS *
443 **************************/
445 /* Compute parameters for interactions between i and j atoms */
446 qq30 = _mm_mul_ps(iq3,jq0);
448 /* COULOMB ELECTROSTATICS */
449 velec = _mm_mul_ps(qq30,rinv30);
450 felec = _mm_mul_ps(velec,rinvsq30);
452 /* Update potential sum for this i atom from the interaction with this j atom. */
453 velec = _mm_andnot_ps(dummy_mask,velec);
454 velecsum = _mm_add_ps(velecsum,velec);
458 fscal = _mm_andnot_ps(dummy_mask,fscal);
460 /* Calculate temporary vectorial force */
461 tx = _mm_mul_ps(fscal,dx30);
462 ty = _mm_mul_ps(fscal,dy30);
463 tz = _mm_mul_ps(fscal,dz30);
465 /* Update vectorial force */
466 fix3 = _mm_add_ps(fix3,tx);
467 fiy3 = _mm_add_ps(fiy3,ty);
468 fiz3 = _mm_add_ps(fiz3,tz);
470 fjx0 = _mm_add_ps(fjx0,tx);
471 fjy0 = _mm_add_ps(fjy0,ty);
472 fjz0 = _mm_add_ps(fjz0,tz);
474 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
475 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
476 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
477 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
479 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
481 /* Inner loop uses 84 flops */
484 /* End of innermost loop */
486 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
487 f+i_coord_offset+DIM,fshift+i_shift_offset);
490 /* Update potential energies */
491 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
493 /* Increment number of inner iterations */
494 inneriter += j_index_end - j_index_start;
496 /* Outer loop uses 19 flops */
499 /* Increment number of outer iterations */
502 /* Update outer/inner flops */
504 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*84);
507 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_sse2_single
508 * Electrostatics interaction: Coulomb
509 * VdW interaction: None
510 * Geometry: Water4-Particle
511 * Calculate force/pot: Force
514 nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_sse2_single
515 (t_nblist * gmx_restrict nlist,
516 rvec * gmx_restrict xx,
517 rvec * gmx_restrict ff,
518 struct t_forcerec * gmx_restrict fr,
519 t_mdatoms * gmx_restrict mdatoms,
520 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
521 t_nrnb * gmx_restrict nrnb)
523 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
524 * just 0 for non-waters.
525 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
526 * jnr indices corresponding to data put in the four positions in the SIMD register.
528 int i_shift_offset,i_coord_offset,outeriter,inneriter;
529 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
530 int jnrA,jnrB,jnrC,jnrD;
531 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
532 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
533 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
535 real *shiftvec,*fshift,*x,*f;
536 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
538 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
540 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
542 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
544 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
545 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
546 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
547 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
548 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
549 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
550 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
552 __m128 dummy_mask,cutoff_mask;
553 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
554 __m128 one = _mm_set1_ps(1.0);
555 __m128 two = _mm_set1_ps(2.0);
561 jindex = nlist->jindex;
563 shiftidx = nlist->shift;
565 shiftvec = fr->shift_vec[0];
566 fshift = fr->fshift[0];
567 facel = _mm_set1_ps(fr->ic->epsfac);
568 charge = mdatoms->chargeA;
570 /* Setup water-specific parameters */
571 inr = nlist->iinr[0];
572 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
573 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
574 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
576 /* Avoid stupid compiler warnings */
577 jnrA = jnrB = jnrC = jnrD = 0;
586 for(iidx=0;iidx<4*DIM;iidx++)
591 /* Start outer loop over neighborlists */
592 for(iidx=0; iidx<nri; iidx++)
594 /* Load shift vector for this list */
595 i_shift_offset = DIM*shiftidx[iidx];
597 /* Load limits for loop over neighbors */
598 j_index_start = jindex[iidx];
599 j_index_end = jindex[iidx+1];
601 /* Get outer coordinate index */
603 i_coord_offset = DIM*inr;
605 /* Load i particle coords and add shift vector */
606 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
607 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
609 fix1 = _mm_setzero_ps();
610 fiy1 = _mm_setzero_ps();
611 fiz1 = _mm_setzero_ps();
612 fix2 = _mm_setzero_ps();
613 fiy2 = _mm_setzero_ps();
614 fiz2 = _mm_setzero_ps();
615 fix3 = _mm_setzero_ps();
616 fiy3 = _mm_setzero_ps();
617 fiz3 = _mm_setzero_ps();
619 /* Start inner kernel loop */
620 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
623 /* Get j neighbor index, and coordinate index */
628 j_coord_offsetA = DIM*jnrA;
629 j_coord_offsetB = DIM*jnrB;
630 j_coord_offsetC = DIM*jnrC;
631 j_coord_offsetD = DIM*jnrD;
633 /* load j atom coordinates */
634 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
635 x+j_coord_offsetC,x+j_coord_offsetD,
638 /* Calculate displacement vector */
639 dx10 = _mm_sub_ps(ix1,jx0);
640 dy10 = _mm_sub_ps(iy1,jy0);
641 dz10 = _mm_sub_ps(iz1,jz0);
642 dx20 = _mm_sub_ps(ix2,jx0);
643 dy20 = _mm_sub_ps(iy2,jy0);
644 dz20 = _mm_sub_ps(iz2,jz0);
645 dx30 = _mm_sub_ps(ix3,jx0);
646 dy30 = _mm_sub_ps(iy3,jy0);
647 dz30 = _mm_sub_ps(iz3,jz0);
649 /* Calculate squared distance and things based on it */
650 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
651 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
652 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
654 rinv10 = sse2_invsqrt_f(rsq10);
655 rinv20 = sse2_invsqrt_f(rsq20);
656 rinv30 = sse2_invsqrt_f(rsq30);
658 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
659 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
660 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
662 /* Load parameters for j particles */
663 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
664 charge+jnrC+0,charge+jnrD+0);
666 fjx0 = _mm_setzero_ps();
667 fjy0 = _mm_setzero_ps();
668 fjz0 = _mm_setzero_ps();
670 /**************************
671 * CALCULATE INTERACTIONS *
672 **************************/
674 /* Compute parameters for interactions between i and j atoms */
675 qq10 = _mm_mul_ps(iq1,jq0);
677 /* COULOMB ELECTROSTATICS */
678 velec = _mm_mul_ps(qq10,rinv10);
679 felec = _mm_mul_ps(velec,rinvsq10);
683 /* Calculate temporary vectorial force */
684 tx = _mm_mul_ps(fscal,dx10);
685 ty = _mm_mul_ps(fscal,dy10);
686 tz = _mm_mul_ps(fscal,dz10);
688 /* Update vectorial force */
689 fix1 = _mm_add_ps(fix1,tx);
690 fiy1 = _mm_add_ps(fiy1,ty);
691 fiz1 = _mm_add_ps(fiz1,tz);
693 fjx0 = _mm_add_ps(fjx0,tx);
694 fjy0 = _mm_add_ps(fjy0,ty);
695 fjz0 = _mm_add_ps(fjz0,tz);
697 /**************************
698 * CALCULATE INTERACTIONS *
699 **************************/
701 /* Compute parameters for interactions between i and j atoms */
702 qq20 = _mm_mul_ps(iq2,jq0);
704 /* COULOMB ELECTROSTATICS */
705 velec = _mm_mul_ps(qq20,rinv20);
706 felec = _mm_mul_ps(velec,rinvsq20);
710 /* Calculate temporary vectorial force */
711 tx = _mm_mul_ps(fscal,dx20);
712 ty = _mm_mul_ps(fscal,dy20);
713 tz = _mm_mul_ps(fscal,dz20);
715 /* Update vectorial force */
716 fix2 = _mm_add_ps(fix2,tx);
717 fiy2 = _mm_add_ps(fiy2,ty);
718 fiz2 = _mm_add_ps(fiz2,tz);
720 fjx0 = _mm_add_ps(fjx0,tx);
721 fjy0 = _mm_add_ps(fjy0,ty);
722 fjz0 = _mm_add_ps(fjz0,tz);
724 /**************************
725 * CALCULATE INTERACTIONS *
726 **************************/
728 /* Compute parameters for interactions between i and j atoms */
729 qq30 = _mm_mul_ps(iq3,jq0);
731 /* COULOMB ELECTROSTATICS */
732 velec = _mm_mul_ps(qq30,rinv30);
733 felec = _mm_mul_ps(velec,rinvsq30);
737 /* Calculate temporary vectorial force */
738 tx = _mm_mul_ps(fscal,dx30);
739 ty = _mm_mul_ps(fscal,dy30);
740 tz = _mm_mul_ps(fscal,dz30);
742 /* Update vectorial force */
743 fix3 = _mm_add_ps(fix3,tx);
744 fiy3 = _mm_add_ps(fiy3,ty);
745 fiz3 = _mm_add_ps(fiz3,tz);
747 fjx0 = _mm_add_ps(fjx0,tx);
748 fjy0 = _mm_add_ps(fjy0,ty);
749 fjz0 = _mm_add_ps(fjz0,tz);
751 fjptrA = f+j_coord_offsetA;
752 fjptrB = f+j_coord_offsetB;
753 fjptrC = f+j_coord_offsetC;
754 fjptrD = f+j_coord_offsetD;
756 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
758 /* Inner loop uses 81 flops */
764 /* Get j neighbor index, and coordinate index */
765 jnrlistA = jjnr[jidx];
766 jnrlistB = jjnr[jidx+1];
767 jnrlistC = jjnr[jidx+2];
768 jnrlistD = jjnr[jidx+3];
769 /* Sign of each element will be negative for non-real atoms.
770 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
771 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
773 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
774 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
775 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
776 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
777 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
778 j_coord_offsetA = DIM*jnrA;
779 j_coord_offsetB = DIM*jnrB;
780 j_coord_offsetC = DIM*jnrC;
781 j_coord_offsetD = DIM*jnrD;
783 /* load j atom coordinates */
784 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
785 x+j_coord_offsetC,x+j_coord_offsetD,
788 /* Calculate displacement vector */
789 dx10 = _mm_sub_ps(ix1,jx0);
790 dy10 = _mm_sub_ps(iy1,jy0);
791 dz10 = _mm_sub_ps(iz1,jz0);
792 dx20 = _mm_sub_ps(ix2,jx0);
793 dy20 = _mm_sub_ps(iy2,jy0);
794 dz20 = _mm_sub_ps(iz2,jz0);
795 dx30 = _mm_sub_ps(ix3,jx0);
796 dy30 = _mm_sub_ps(iy3,jy0);
797 dz30 = _mm_sub_ps(iz3,jz0);
799 /* Calculate squared distance and things based on it */
800 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
801 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
802 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
804 rinv10 = sse2_invsqrt_f(rsq10);
805 rinv20 = sse2_invsqrt_f(rsq20);
806 rinv30 = sse2_invsqrt_f(rsq30);
808 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
809 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
810 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
812 /* Load parameters for j particles */
813 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
814 charge+jnrC+0,charge+jnrD+0);
816 fjx0 = _mm_setzero_ps();
817 fjy0 = _mm_setzero_ps();
818 fjz0 = _mm_setzero_ps();
820 /**************************
821 * CALCULATE INTERACTIONS *
822 **************************/
824 /* Compute parameters for interactions between i and j atoms */
825 qq10 = _mm_mul_ps(iq1,jq0);
827 /* COULOMB ELECTROSTATICS */
828 velec = _mm_mul_ps(qq10,rinv10);
829 felec = _mm_mul_ps(velec,rinvsq10);
833 fscal = _mm_andnot_ps(dummy_mask,fscal);
835 /* Calculate temporary vectorial force */
836 tx = _mm_mul_ps(fscal,dx10);
837 ty = _mm_mul_ps(fscal,dy10);
838 tz = _mm_mul_ps(fscal,dz10);
840 /* Update vectorial force */
841 fix1 = _mm_add_ps(fix1,tx);
842 fiy1 = _mm_add_ps(fiy1,ty);
843 fiz1 = _mm_add_ps(fiz1,tz);
845 fjx0 = _mm_add_ps(fjx0,tx);
846 fjy0 = _mm_add_ps(fjy0,ty);
847 fjz0 = _mm_add_ps(fjz0,tz);
849 /**************************
850 * CALCULATE INTERACTIONS *
851 **************************/
853 /* Compute parameters for interactions between i and j atoms */
854 qq20 = _mm_mul_ps(iq2,jq0);
856 /* COULOMB ELECTROSTATICS */
857 velec = _mm_mul_ps(qq20,rinv20);
858 felec = _mm_mul_ps(velec,rinvsq20);
862 fscal = _mm_andnot_ps(dummy_mask,fscal);
864 /* Calculate temporary vectorial force */
865 tx = _mm_mul_ps(fscal,dx20);
866 ty = _mm_mul_ps(fscal,dy20);
867 tz = _mm_mul_ps(fscal,dz20);
869 /* Update vectorial force */
870 fix2 = _mm_add_ps(fix2,tx);
871 fiy2 = _mm_add_ps(fiy2,ty);
872 fiz2 = _mm_add_ps(fiz2,tz);
874 fjx0 = _mm_add_ps(fjx0,tx);
875 fjy0 = _mm_add_ps(fjy0,ty);
876 fjz0 = _mm_add_ps(fjz0,tz);
878 /**************************
879 * CALCULATE INTERACTIONS *
880 **************************/
882 /* Compute parameters for interactions between i and j atoms */
883 qq30 = _mm_mul_ps(iq3,jq0);
885 /* COULOMB ELECTROSTATICS */
886 velec = _mm_mul_ps(qq30,rinv30);
887 felec = _mm_mul_ps(velec,rinvsq30);
891 fscal = _mm_andnot_ps(dummy_mask,fscal);
893 /* Calculate temporary vectorial force */
894 tx = _mm_mul_ps(fscal,dx30);
895 ty = _mm_mul_ps(fscal,dy30);
896 tz = _mm_mul_ps(fscal,dz30);
898 /* Update vectorial force */
899 fix3 = _mm_add_ps(fix3,tx);
900 fiy3 = _mm_add_ps(fiy3,ty);
901 fiz3 = _mm_add_ps(fiz3,tz);
903 fjx0 = _mm_add_ps(fjx0,tx);
904 fjy0 = _mm_add_ps(fjy0,ty);
905 fjz0 = _mm_add_ps(fjz0,tz);
907 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
908 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
909 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
910 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
912 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
914 /* Inner loop uses 81 flops */
917 /* End of innermost loop */
919 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
920 f+i_coord_offset+DIM,fshift+i_shift_offset);
922 /* Increment number of inner iterations */
923 inneriter += j_index_end - j_index_start;
925 /* Outer loop uses 18 flops */
928 /* Increment number of outer iterations */
931 /* Update outer/inner flops */
933 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*81);