2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, 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 sse4_1_single kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_sse4_1_single.h"
48 #include "kernelutil_x86_sse4_1_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_sse4_1_single
52 * Electrostatics interaction: ReactionField
53 * VdW interaction: None
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_sse4_1_single
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real *shiftvec,*fshift,*x,*f;
80 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
86 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
88 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
89 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
90 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
91 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
92 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
93 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
94 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
96 __m128 dummy_mask,cutoff_mask;
97 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
98 __m128 one = _mm_set1_ps(1.0);
99 __m128 two = _mm_set1_ps(2.0);
105 jindex = nlist->jindex;
107 shiftidx = nlist->shift;
109 shiftvec = fr->shift_vec[0];
110 fshift = fr->fshift[0];
111 facel = _mm_set1_ps(fr->epsfac);
112 charge = mdatoms->chargeA;
113 krf = _mm_set1_ps(fr->ic->k_rf);
114 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
115 crf = _mm_set1_ps(fr->ic->c_rf);
117 /* Setup water-specific parameters */
118 inr = nlist->iinr[0];
119 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
120 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
121 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
123 /* Avoid stupid compiler warnings */
124 jnrA = jnrB = jnrC = jnrD = 0;
133 for(iidx=0;iidx<4*DIM;iidx++)
138 /* Start outer loop over neighborlists */
139 for(iidx=0; iidx<nri; iidx++)
141 /* Load shift vector for this list */
142 i_shift_offset = DIM*shiftidx[iidx];
144 /* Load limits for loop over neighbors */
145 j_index_start = jindex[iidx];
146 j_index_end = jindex[iidx+1];
148 /* Get outer coordinate index */
150 i_coord_offset = DIM*inr;
152 /* Load i particle coords and add shift vector */
153 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
154 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
156 fix1 = _mm_setzero_ps();
157 fiy1 = _mm_setzero_ps();
158 fiz1 = _mm_setzero_ps();
159 fix2 = _mm_setzero_ps();
160 fiy2 = _mm_setzero_ps();
161 fiz2 = _mm_setzero_ps();
162 fix3 = _mm_setzero_ps();
163 fiy3 = _mm_setzero_ps();
164 fiz3 = _mm_setzero_ps();
166 /* Reset potential sums */
167 velecsum = _mm_setzero_ps();
169 /* Start inner kernel loop */
170 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
173 /* Get j neighbor index, and coordinate index */
178 j_coord_offsetA = DIM*jnrA;
179 j_coord_offsetB = DIM*jnrB;
180 j_coord_offsetC = DIM*jnrC;
181 j_coord_offsetD = DIM*jnrD;
183 /* load j atom coordinates */
184 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
185 x+j_coord_offsetC,x+j_coord_offsetD,
188 /* Calculate displacement vector */
189 dx10 = _mm_sub_ps(ix1,jx0);
190 dy10 = _mm_sub_ps(iy1,jy0);
191 dz10 = _mm_sub_ps(iz1,jz0);
192 dx20 = _mm_sub_ps(ix2,jx0);
193 dy20 = _mm_sub_ps(iy2,jy0);
194 dz20 = _mm_sub_ps(iz2,jz0);
195 dx30 = _mm_sub_ps(ix3,jx0);
196 dy30 = _mm_sub_ps(iy3,jy0);
197 dz30 = _mm_sub_ps(iz3,jz0);
199 /* Calculate squared distance and things based on it */
200 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
201 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
202 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
204 rinv10 = gmx_mm_invsqrt_ps(rsq10);
205 rinv20 = gmx_mm_invsqrt_ps(rsq20);
206 rinv30 = gmx_mm_invsqrt_ps(rsq30);
208 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
209 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
210 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
212 /* Load parameters for j particles */
213 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
214 charge+jnrC+0,charge+jnrD+0);
216 fjx0 = _mm_setzero_ps();
217 fjy0 = _mm_setzero_ps();
218 fjz0 = _mm_setzero_ps();
220 /**************************
221 * CALCULATE INTERACTIONS *
222 **************************/
224 /* Compute parameters for interactions between i and j atoms */
225 qq10 = _mm_mul_ps(iq1,jq0);
227 /* REACTION-FIELD ELECTROSTATICS */
228 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
229 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
231 /* Update potential sum for this i atom from the interaction with this j atom. */
232 velecsum = _mm_add_ps(velecsum,velec);
236 /* Calculate temporary vectorial force */
237 tx = _mm_mul_ps(fscal,dx10);
238 ty = _mm_mul_ps(fscal,dy10);
239 tz = _mm_mul_ps(fscal,dz10);
241 /* Update vectorial force */
242 fix1 = _mm_add_ps(fix1,tx);
243 fiy1 = _mm_add_ps(fiy1,ty);
244 fiz1 = _mm_add_ps(fiz1,tz);
246 fjx0 = _mm_add_ps(fjx0,tx);
247 fjy0 = _mm_add_ps(fjy0,ty);
248 fjz0 = _mm_add_ps(fjz0,tz);
250 /**************************
251 * CALCULATE INTERACTIONS *
252 **************************/
254 /* Compute parameters for interactions between i and j atoms */
255 qq20 = _mm_mul_ps(iq2,jq0);
257 /* REACTION-FIELD ELECTROSTATICS */
258 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
259 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
261 /* Update potential sum for this i atom from the interaction with this j atom. */
262 velecsum = _mm_add_ps(velecsum,velec);
266 /* Calculate temporary vectorial force */
267 tx = _mm_mul_ps(fscal,dx20);
268 ty = _mm_mul_ps(fscal,dy20);
269 tz = _mm_mul_ps(fscal,dz20);
271 /* Update vectorial force */
272 fix2 = _mm_add_ps(fix2,tx);
273 fiy2 = _mm_add_ps(fiy2,ty);
274 fiz2 = _mm_add_ps(fiz2,tz);
276 fjx0 = _mm_add_ps(fjx0,tx);
277 fjy0 = _mm_add_ps(fjy0,ty);
278 fjz0 = _mm_add_ps(fjz0,tz);
280 /**************************
281 * CALCULATE INTERACTIONS *
282 **************************/
284 /* Compute parameters for interactions between i and j atoms */
285 qq30 = _mm_mul_ps(iq3,jq0);
287 /* REACTION-FIELD ELECTROSTATICS */
288 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
289 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
291 /* Update potential sum for this i atom from the interaction with this j atom. */
292 velecsum = _mm_add_ps(velecsum,velec);
296 /* Calculate temporary vectorial force */
297 tx = _mm_mul_ps(fscal,dx30);
298 ty = _mm_mul_ps(fscal,dy30);
299 tz = _mm_mul_ps(fscal,dz30);
301 /* Update vectorial force */
302 fix3 = _mm_add_ps(fix3,tx);
303 fiy3 = _mm_add_ps(fiy3,ty);
304 fiz3 = _mm_add_ps(fiz3,tz);
306 fjx0 = _mm_add_ps(fjx0,tx);
307 fjy0 = _mm_add_ps(fjy0,ty);
308 fjz0 = _mm_add_ps(fjz0,tz);
310 fjptrA = f+j_coord_offsetA;
311 fjptrB = f+j_coord_offsetB;
312 fjptrC = f+j_coord_offsetC;
313 fjptrD = f+j_coord_offsetD;
315 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
317 /* Inner loop uses 96 flops */
323 /* Get j neighbor index, and coordinate index */
324 jnrlistA = jjnr[jidx];
325 jnrlistB = jjnr[jidx+1];
326 jnrlistC = jjnr[jidx+2];
327 jnrlistD = jjnr[jidx+3];
328 /* Sign of each element will be negative for non-real atoms.
329 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
330 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
332 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
333 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
334 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
335 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
336 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
337 j_coord_offsetA = DIM*jnrA;
338 j_coord_offsetB = DIM*jnrB;
339 j_coord_offsetC = DIM*jnrC;
340 j_coord_offsetD = DIM*jnrD;
342 /* load j atom coordinates */
343 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
344 x+j_coord_offsetC,x+j_coord_offsetD,
347 /* Calculate displacement vector */
348 dx10 = _mm_sub_ps(ix1,jx0);
349 dy10 = _mm_sub_ps(iy1,jy0);
350 dz10 = _mm_sub_ps(iz1,jz0);
351 dx20 = _mm_sub_ps(ix2,jx0);
352 dy20 = _mm_sub_ps(iy2,jy0);
353 dz20 = _mm_sub_ps(iz2,jz0);
354 dx30 = _mm_sub_ps(ix3,jx0);
355 dy30 = _mm_sub_ps(iy3,jy0);
356 dz30 = _mm_sub_ps(iz3,jz0);
358 /* Calculate squared distance and things based on it */
359 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
360 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
361 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
363 rinv10 = gmx_mm_invsqrt_ps(rsq10);
364 rinv20 = gmx_mm_invsqrt_ps(rsq20);
365 rinv30 = gmx_mm_invsqrt_ps(rsq30);
367 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
368 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
369 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
371 /* Load parameters for j particles */
372 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
373 charge+jnrC+0,charge+jnrD+0);
375 fjx0 = _mm_setzero_ps();
376 fjy0 = _mm_setzero_ps();
377 fjz0 = _mm_setzero_ps();
379 /**************************
380 * CALCULATE INTERACTIONS *
381 **************************/
383 /* Compute parameters for interactions between i and j atoms */
384 qq10 = _mm_mul_ps(iq1,jq0);
386 /* REACTION-FIELD ELECTROSTATICS */
387 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
388 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
390 /* Update potential sum for this i atom from the interaction with this j atom. */
391 velec = _mm_andnot_ps(dummy_mask,velec);
392 velecsum = _mm_add_ps(velecsum,velec);
396 fscal = _mm_andnot_ps(dummy_mask,fscal);
398 /* Calculate temporary vectorial force */
399 tx = _mm_mul_ps(fscal,dx10);
400 ty = _mm_mul_ps(fscal,dy10);
401 tz = _mm_mul_ps(fscal,dz10);
403 /* Update vectorial force */
404 fix1 = _mm_add_ps(fix1,tx);
405 fiy1 = _mm_add_ps(fiy1,ty);
406 fiz1 = _mm_add_ps(fiz1,tz);
408 fjx0 = _mm_add_ps(fjx0,tx);
409 fjy0 = _mm_add_ps(fjy0,ty);
410 fjz0 = _mm_add_ps(fjz0,tz);
412 /**************************
413 * CALCULATE INTERACTIONS *
414 **************************/
416 /* Compute parameters for interactions between i and j atoms */
417 qq20 = _mm_mul_ps(iq2,jq0);
419 /* REACTION-FIELD ELECTROSTATICS */
420 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
421 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
423 /* Update potential sum for this i atom from the interaction with this j atom. */
424 velec = _mm_andnot_ps(dummy_mask,velec);
425 velecsum = _mm_add_ps(velecsum,velec);
429 fscal = _mm_andnot_ps(dummy_mask,fscal);
431 /* Calculate temporary vectorial force */
432 tx = _mm_mul_ps(fscal,dx20);
433 ty = _mm_mul_ps(fscal,dy20);
434 tz = _mm_mul_ps(fscal,dz20);
436 /* Update vectorial force */
437 fix2 = _mm_add_ps(fix2,tx);
438 fiy2 = _mm_add_ps(fiy2,ty);
439 fiz2 = _mm_add_ps(fiz2,tz);
441 fjx0 = _mm_add_ps(fjx0,tx);
442 fjy0 = _mm_add_ps(fjy0,ty);
443 fjz0 = _mm_add_ps(fjz0,tz);
445 /**************************
446 * CALCULATE INTERACTIONS *
447 **************************/
449 /* Compute parameters for interactions between i and j atoms */
450 qq30 = _mm_mul_ps(iq3,jq0);
452 /* REACTION-FIELD ELECTROSTATICS */
453 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
454 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
456 /* Update potential sum for this i atom from the interaction with this j atom. */
457 velec = _mm_andnot_ps(dummy_mask,velec);
458 velecsum = _mm_add_ps(velecsum,velec);
462 fscal = _mm_andnot_ps(dummy_mask,fscal);
464 /* Calculate temporary vectorial force */
465 tx = _mm_mul_ps(fscal,dx30);
466 ty = _mm_mul_ps(fscal,dy30);
467 tz = _mm_mul_ps(fscal,dz30);
469 /* Update vectorial force */
470 fix3 = _mm_add_ps(fix3,tx);
471 fiy3 = _mm_add_ps(fiy3,ty);
472 fiz3 = _mm_add_ps(fiz3,tz);
474 fjx0 = _mm_add_ps(fjx0,tx);
475 fjy0 = _mm_add_ps(fjy0,ty);
476 fjz0 = _mm_add_ps(fjz0,tz);
478 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
479 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
480 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
481 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
483 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
485 /* Inner loop uses 96 flops */
488 /* End of innermost loop */
490 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
491 f+i_coord_offset+DIM,fshift+i_shift_offset);
494 /* Update potential energies */
495 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
497 /* Increment number of inner iterations */
498 inneriter += j_index_end - j_index_start;
500 /* Outer loop uses 19 flops */
503 /* Increment number of outer iterations */
506 /* Update outer/inner flops */
508 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*96);
511 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW4P1_F_sse4_1_single
512 * Electrostatics interaction: ReactionField
513 * VdW interaction: None
514 * Geometry: Water4-Particle
515 * Calculate force/pot: Force
518 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_sse4_1_single
519 (t_nblist * gmx_restrict nlist,
520 rvec * gmx_restrict xx,
521 rvec * gmx_restrict ff,
522 t_forcerec * gmx_restrict fr,
523 t_mdatoms * gmx_restrict mdatoms,
524 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
525 t_nrnb * gmx_restrict nrnb)
527 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
528 * just 0 for non-waters.
529 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
530 * jnr indices corresponding to data put in the four positions in the SIMD register.
532 int i_shift_offset,i_coord_offset,outeriter,inneriter;
533 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
534 int jnrA,jnrB,jnrC,jnrD;
535 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
536 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
537 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
539 real *shiftvec,*fshift,*x,*f;
540 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
542 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
544 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
546 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
548 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
549 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
550 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
551 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
552 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
553 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
554 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
556 __m128 dummy_mask,cutoff_mask;
557 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
558 __m128 one = _mm_set1_ps(1.0);
559 __m128 two = _mm_set1_ps(2.0);
565 jindex = nlist->jindex;
567 shiftidx = nlist->shift;
569 shiftvec = fr->shift_vec[0];
570 fshift = fr->fshift[0];
571 facel = _mm_set1_ps(fr->epsfac);
572 charge = mdatoms->chargeA;
573 krf = _mm_set1_ps(fr->ic->k_rf);
574 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
575 crf = _mm_set1_ps(fr->ic->c_rf);
577 /* Setup water-specific parameters */
578 inr = nlist->iinr[0];
579 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
580 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
581 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
583 /* Avoid stupid compiler warnings */
584 jnrA = jnrB = jnrC = jnrD = 0;
593 for(iidx=0;iidx<4*DIM;iidx++)
598 /* Start outer loop over neighborlists */
599 for(iidx=0; iidx<nri; iidx++)
601 /* Load shift vector for this list */
602 i_shift_offset = DIM*shiftidx[iidx];
604 /* Load limits for loop over neighbors */
605 j_index_start = jindex[iidx];
606 j_index_end = jindex[iidx+1];
608 /* Get outer coordinate index */
610 i_coord_offset = DIM*inr;
612 /* Load i particle coords and add shift vector */
613 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
614 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
616 fix1 = _mm_setzero_ps();
617 fiy1 = _mm_setzero_ps();
618 fiz1 = _mm_setzero_ps();
619 fix2 = _mm_setzero_ps();
620 fiy2 = _mm_setzero_ps();
621 fiz2 = _mm_setzero_ps();
622 fix3 = _mm_setzero_ps();
623 fiy3 = _mm_setzero_ps();
624 fiz3 = _mm_setzero_ps();
626 /* Start inner kernel loop */
627 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
630 /* Get j neighbor index, and coordinate index */
635 j_coord_offsetA = DIM*jnrA;
636 j_coord_offsetB = DIM*jnrB;
637 j_coord_offsetC = DIM*jnrC;
638 j_coord_offsetD = DIM*jnrD;
640 /* load j atom coordinates */
641 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
642 x+j_coord_offsetC,x+j_coord_offsetD,
645 /* Calculate displacement vector */
646 dx10 = _mm_sub_ps(ix1,jx0);
647 dy10 = _mm_sub_ps(iy1,jy0);
648 dz10 = _mm_sub_ps(iz1,jz0);
649 dx20 = _mm_sub_ps(ix2,jx0);
650 dy20 = _mm_sub_ps(iy2,jy0);
651 dz20 = _mm_sub_ps(iz2,jz0);
652 dx30 = _mm_sub_ps(ix3,jx0);
653 dy30 = _mm_sub_ps(iy3,jy0);
654 dz30 = _mm_sub_ps(iz3,jz0);
656 /* Calculate squared distance and things based on it */
657 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
658 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
659 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
661 rinv10 = gmx_mm_invsqrt_ps(rsq10);
662 rinv20 = gmx_mm_invsqrt_ps(rsq20);
663 rinv30 = gmx_mm_invsqrt_ps(rsq30);
665 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
666 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
667 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
669 /* Load parameters for j particles */
670 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
671 charge+jnrC+0,charge+jnrD+0);
673 fjx0 = _mm_setzero_ps();
674 fjy0 = _mm_setzero_ps();
675 fjz0 = _mm_setzero_ps();
677 /**************************
678 * CALCULATE INTERACTIONS *
679 **************************/
681 /* Compute parameters for interactions between i and j atoms */
682 qq10 = _mm_mul_ps(iq1,jq0);
684 /* REACTION-FIELD ELECTROSTATICS */
685 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
689 /* Calculate temporary vectorial force */
690 tx = _mm_mul_ps(fscal,dx10);
691 ty = _mm_mul_ps(fscal,dy10);
692 tz = _mm_mul_ps(fscal,dz10);
694 /* Update vectorial force */
695 fix1 = _mm_add_ps(fix1,tx);
696 fiy1 = _mm_add_ps(fiy1,ty);
697 fiz1 = _mm_add_ps(fiz1,tz);
699 fjx0 = _mm_add_ps(fjx0,tx);
700 fjy0 = _mm_add_ps(fjy0,ty);
701 fjz0 = _mm_add_ps(fjz0,tz);
703 /**************************
704 * CALCULATE INTERACTIONS *
705 **************************/
707 /* Compute parameters for interactions between i and j atoms */
708 qq20 = _mm_mul_ps(iq2,jq0);
710 /* REACTION-FIELD ELECTROSTATICS */
711 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
715 /* Calculate temporary vectorial force */
716 tx = _mm_mul_ps(fscal,dx20);
717 ty = _mm_mul_ps(fscal,dy20);
718 tz = _mm_mul_ps(fscal,dz20);
720 /* Update vectorial force */
721 fix2 = _mm_add_ps(fix2,tx);
722 fiy2 = _mm_add_ps(fiy2,ty);
723 fiz2 = _mm_add_ps(fiz2,tz);
725 fjx0 = _mm_add_ps(fjx0,tx);
726 fjy0 = _mm_add_ps(fjy0,ty);
727 fjz0 = _mm_add_ps(fjz0,tz);
729 /**************************
730 * CALCULATE INTERACTIONS *
731 **************************/
733 /* Compute parameters for interactions between i and j atoms */
734 qq30 = _mm_mul_ps(iq3,jq0);
736 /* REACTION-FIELD ELECTROSTATICS */
737 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
741 /* Calculate temporary vectorial force */
742 tx = _mm_mul_ps(fscal,dx30);
743 ty = _mm_mul_ps(fscal,dy30);
744 tz = _mm_mul_ps(fscal,dz30);
746 /* Update vectorial force */
747 fix3 = _mm_add_ps(fix3,tx);
748 fiy3 = _mm_add_ps(fiy3,ty);
749 fiz3 = _mm_add_ps(fiz3,tz);
751 fjx0 = _mm_add_ps(fjx0,tx);
752 fjy0 = _mm_add_ps(fjy0,ty);
753 fjz0 = _mm_add_ps(fjz0,tz);
755 fjptrA = f+j_coord_offsetA;
756 fjptrB = f+j_coord_offsetB;
757 fjptrC = f+j_coord_offsetC;
758 fjptrD = f+j_coord_offsetD;
760 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
762 /* Inner loop uses 81 flops */
768 /* Get j neighbor index, and coordinate index */
769 jnrlistA = jjnr[jidx];
770 jnrlistB = jjnr[jidx+1];
771 jnrlistC = jjnr[jidx+2];
772 jnrlistD = jjnr[jidx+3];
773 /* Sign of each element will be negative for non-real atoms.
774 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
775 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
777 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
778 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
779 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
780 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
781 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
782 j_coord_offsetA = DIM*jnrA;
783 j_coord_offsetB = DIM*jnrB;
784 j_coord_offsetC = DIM*jnrC;
785 j_coord_offsetD = DIM*jnrD;
787 /* load j atom coordinates */
788 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
789 x+j_coord_offsetC,x+j_coord_offsetD,
792 /* Calculate displacement vector */
793 dx10 = _mm_sub_ps(ix1,jx0);
794 dy10 = _mm_sub_ps(iy1,jy0);
795 dz10 = _mm_sub_ps(iz1,jz0);
796 dx20 = _mm_sub_ps(ix2,jx0);
797 dy20 = _mm_sub_ps(iy2,jy0);
798 dz20 = _mm_sub_ps(iz2,jz0);
799 dx30 = _mm_sub_ps(ix3,jx0);
800 dy30 = _mm_sub_ps(iy3,jy0);
801 dz30 = _mm_sub_ps(iz3,jz0);
803 /* Calculate squared distance and things based on it */
804 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
805 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
806 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
808 rinv10 = gmx_mm_invsqrt_ps(rsq10);
809 rinv20 = gmx_mm_invsqrt_ps(rsq20);
810 rinv30 = gmx_mm_invsqrt_ps(rsq30);
812 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
813 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
814 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
816 /* Load parameters for j particles */
817 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
818 charge+jnrC+0,charge+jnrD+0);
820 fjx0 = _mm_setzero_ps();
821 fjy0 = _mm_setzero_ps();
822 fjz0 = _mm_setzero_ps();
824 /**************************
825 * CALCULATE INTERACTIONS *
826 **************************/
828 /* Compute parameters for interactions between i and j atoms */
829 qq10 = _mm_mul_ps(iq1,jq0);
831 /* REACTION-FIELD ELECTROSTATICS */
832 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
836 fscal = _mm_andnot_ps(dummy_mask,fscal);
838 /* Calculate temporary vectorial force */
839 tx = _mm_mul_ps(fscal,dx10);
840 ty = _mm_mul_ps(fscal,dy10);
841 tz = _mm_mul_ps(fscal,dz10);
843 /* Update vectorial force */
844 fix1 = _mm_add_ps(fix1,tx);
845 fiy1 = _mm_add_ps(fiy1,ty);
846 fiz1 = _mm_add_ps(fiz1,tz);
848 fjx0 = _mm_add_ps(fjx0,tx);
849 fjy0 = _mm_add_ps(fjy0,ty);
850 fjz0 = _mm_add_ps(fjz0,tz);
852 /**************************
853 * CALCULATE INTERACTIONS *
854 **************************/
856 /* Compute parameters for interactions between i and j atoms */
857 qq20 = _mm_mul_ps(iq2,jq0);
859 /* REACTION-FIELD ELECTROSTATICS */
860 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
864 fscal = _mm_andnot_ps(dummy_mask,fscal);
866 /* Calculate temporary vectorial force */
867 tx = _mm_mul_ps(fscal,dx20);
868 ty = _mm_mul_ps(fscal,dy20);
869 tz = _mm_mul_ps(fscal,dz20);
871 /* Update vectorial force */
872 fix2 = _mm_add_ps(fix2,tx);
873 fiy2 = _mm_add_ps(fiy2,ty);
874 fiz2 = _mm_add_ps(fiz2,tz);
876 fjx0 = _mm_add_ps(fjx0,tx);
877 fjy0 = _mm_add_ps(fjy0,ty);
878 fjz0 = _mm_add_ps(fjz0,tz);
880 /**************************
881 * CALCULATE INTERACTIONS *
882 **************************/
884 /* Compute parameters for interactions between i and j atoms */
885 qq30 = _mm_mul_ps(iq3,jq0);
887 /* REACTION-FIELD ELECTROSTATICS */
888 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
892 fscal = _mm_andnot_ps(dummy_mask,fscal);
894 /* Calculate temporary vectorial force */
895 tx = _mm_mul_ps(fscal,dx30);
896 ty = _mm_mul_ps(fscal,dy30);
897 tz = _mm_mul_ps(fscal,dz30);
899 /* Update vectorial force */
900 fix3 = _mm_add_ps(fix3,tx);
901 fiy3 = _mm_add_ps(fiy3,ty);
902 fiz3 = _mm_add_ps(fiz3,tz);
904 fjx0 = _mm_add_ps(fjx0,tx);
905 fjy0 = _mm_add_ps(fjy0,ty);
906 fjz0 = _mm_add_ps(fjz0,tz);
908 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
909 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
910 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
911 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
913 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
915 /* Inner loop uses 81 flops */
918 /* End of innermost loop */
920 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
921 f+i_coord_offset+DIM,fshift+i_shift_offset);
923 /* Increment number of inner iterations */
924 inneriter += j_index_end - j_index_start;
926 /* Outer loop uses 18 flops */
929 /* Increment number of outer iterations */
932 /* Update outer/inner flops */
934 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*81);