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 sse4_1_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/gmxlib/nrnb.h"
47 #include "kernelutil_x86_sse4_1_single.h"
50 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW3P1_VF_sse4_1_single
51 * Electrostatics interaction: ReactionField
52 * VdW interaction: None
53 * Geometry: Water3-Particle
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecRF_VdwNone_GeomW3P1_VF_sse4_1_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 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
87 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
88 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
89 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
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 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;
112 krf = _mm_set1_ps(fr->ic->k_rf);
113 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
114 crf = _mm_set1_ps(fr->ic->c_rf);
116 /* Setup water-specific parameters */
117 inr = nlist->iinr[0];
118 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+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]));
122 /* Avoid stupid compiler warnings */
123 jnrA = jnrB = jnrC = jnrD = 0;
132 for(iidx=0;iidx<4*DIM;iidx++)
137 /* Start outer loop over neighborlists */
138 for(iidx=0; iidx<nri; iidx++)
140 /* Load shift vector for this list */
141 i_shift_offset = DIM*shiftidx[iidx];
143 /* Load limits for loop over neighbors */
144 j_index_start = jindex[iidx];
145 j_index_end = jindex[iidx+1];
147 /* Get outer coordinate index */
149 i_coord_offset = DIM*inr;
151 /* Load i particle coords and add shift vector */
152 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
153 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
155 fix0 = _mm_setzero_ps();
156 fiy0 = _mm_setzero_ps();
157 fiz0 = _mm_setzero_ps();
158 fix1 = _mm_setzero_ps();
159 fiy1 = _mm_setzero_ps();
160 fiz1 = _mm_setzero_ps();
161 fix2 = _mm_setzero_ps();
162 fiy2 = _mm_setzero_ps();
163 fiz2 = _mm_setzero_ps();
165 /* Reset potential sums */
166 velecsum = _mm_setzero_ps();
168 /* Start inner kernel loop */
169 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
172 /* Get j neighbor index, and coordinate index */
177 j_coord_offsetA = DIM*jnrA;
178 j_coord_offsetB = DIM*jnrB;
179 j_coord_offsetC = DIM*jnrC;
180 j_coord_offsetD = DIM*jnrD;
182 /* load j atom coordinates */
183 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
184 x+j_coord_offsetC,x+j_coord_offsetD,
187 /* Calculate displacement vector */
188 dx00 = _mm_sub_ps(ix0,jx0);
189 dy00 = _mm_sub_ps(iy0,jy0);
190 dz00 = _mm_sub_ps(iz0,jz0);
191 dx10 = _mm_sub_ps(ix1,jx0);
192 dy10 = _mm_sub_ps(iy1,jy0);
193 dz10 = _mm_sub_ps(iz1,jz0);
194 dx20 = _mm_sub_ps(ix2,jx0);
195 dy20 = _mm_sub_ps(iy2,jy0);
196 dz20 = _mm_sub_ps(iz2,jz0);
198 /* Calculate squared distance and things based on it */
199 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
200 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
201 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
203 rinv00 = sse41_invsqrt_f(rsq00);
204 rinv10 = sse41_invsqrt_f(rsq10);
205 rinv20 = sse41_invsqrt_f(rsq20);
207 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
208 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
209 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
211 /* Load parameters for j particles */
212 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
213 charge+jnrC+0,charge+jnrD+0);
215 fjx0 = _mm_setzero_ps();
216 fjy0 = _mm_setzero_ps();
217 fjz0 = _mm_setzero_ps();
219 /**************************
220 * CALCULATE INTERACTIONS *
221 **************************/
223 /* Compute parameters for interactions between i and j atoms */
224 qq00 = _mm_mul_ps(iq0,jq0);
226 /* REACTION-FIELD ELECTROSTATICS */
227 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
228 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
230 /* Update potential sum for this i atom from the interaction with this j atom. */
231 velecsum = _mm_add_ps(velecsum,velec);
235 /* Calculate temporary vectorial force */
236 tx = _mm_mul_ps(fscal,dx00);
237 ty = _mm_mul_ps(fscal,dy00);
238 tz = _mm_mul_ps(fscal,dz00);
240 /* Update vectorial force */
241 fix0 = _mm_add_ps(fix0,tx);
242 fiy0 = _mm_add_ps(fiy0,ty);
243 fiz0 = _mm_add_ps(fiz0,tz);
245 fjx0 = _mm_add_ps(fjx0,tx);
246 fjy0 = _mm_add_ps(fjy0,ty);
247 fjz0 = _mm_add_ps(fjz0,tz);
249 /**************************
250 * CALCULATE INTERACTIONS *
251 **************************/
253 /* Compute parameters for interactions between i and j atoms */
254 qq10 = _mm_mul_ps(iq1,jq0);
256 /* REACTION-FIELD ELECTROSTATICS */
257 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
258 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
260 /* Update potential sum for this i atom from the interaction with this j atom. */
261 velecsum = _mm_add_ps(velecsum,velec);
265 /* Calculate temporary vectorial force */
266 tx = _mm_mul_ps(fscal,dx10);
267 ty = _mm_mul_ps(fscal,dy10);
268 tz = _mm_mul_ps(fscal,dz10);
270 /* Update vectorial force */
271 fix1 = _mm_add_ps(fix1,tx);
272 fiy1 = _mm_add_ps(fiy1,ty);
273 fiz1 = _mm_add_ps(fiz1,tz);
275 fjx0 = _mm_add_ps(fjx0,tx);
276 fjy0 = _mm_add_ps(fjy0,ty);
277 fjz0 = _mm_add_ps(fjz0,tz);
279 /**************************
280 * CALCULATE INTERACTIONS *
281 **************************/
283 /* Compute parameters for interactions between i and j atoms */
284 qq20 = _mm_mul_ps(iq2,jq0);
286 /* REACTION-FIELD ELECTROSTATICS */
287 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
288 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
290 /* Update potential sum for this i atom from the interaction with this j atom. */
291 velecsum = _mm_add_ps(velecsum,velec);
295 /* Calculate temporary vectorial force */
296 tx = _mm_mul_ps(fscal,dx20);
297 ty = _mm_mul_ps(fscal,dy20);
298 tz = _mm_mul_ps(fscal,dz20);
300 /* Update vectorial force */
301 fix2 = _mm_add_ps(fix2,tx);
302 fiy2 = _mm_add_ps(fiy2,ty);
303 fiz2 = _mm_add_ps(fiz2,tz);
305 fjx0 = _mm_add_ps(fjx0,tx);
306 fjy0 = _mm_add_ps(fjy0,ty);
307 fjz0 = _mm_add_ps(fjz0,tz);
309 fjptrA = f+j_coord_offsetA;
310 fjptrB = f+j_coord_offsetB;
311 fjptrC = f+j_coord_offsetC;
312 fjptrD = f+j_coord_offsetD;
314 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
316 /* Inner loop uses 96 flops */
322 /* Get j neighbor index, and coordinate index */
323 jnrlistA = jjnr[jidx];
324 jnrlistB = jjnr[jidx+1];
325 jnrlistC = jjnr[jidx+2];
326 jnrlistD = jjnr[jidx+3];
327 /* Sign of each element will be negative for non-real atoms.
328 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
329 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
331 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
332 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
333 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
334 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
335 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
336 j_coord_offsetA = DIM*jnrA;
337 j_coord_offsetB = DIM*jnrB;
338 j_coord_offsetC = DIM*jnrC;
339 j_coord_offsetD = DIM*jnrD;
341 /* load j atom coordinates */
342 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
343 x+j_coord_offsetC,x+j_coord_offsetD,
346 /* Calculate displacement vector */
347 dx00 = _mm_sub_ps(ix0,jx0);
348 dy00 = _mm_sub_ps(iy0,jy0);
349 dz00 = _mm_sub_ps(iz0,jz0);
350 dx10 = _mm_sub_ps(ix1,jx0);
351 dy10 = _mm_sub_ps(iy1,jy0);
352 dz10 = _mm_sub_ps(iz1,jz0);
353 dx20 = _mm_sub_ps(ix2,jx0);
354 dy20 = _mm_sub_ps(iy2,jy0);
355 dz20 = _mm_sub_ps(iz2,jz0);
357 /* Calculate squared distance and things based on it */
358 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
359 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
360 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
362 rinv00 = sse41_invsqrt_f(rsq00);
363 rinv10 = sse41_invsqrt_f(rsq10);
364 rinv20 = sse41_invsqrt_f(rsq20);
366 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
367 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
368 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
370 /* Load parameters for j particles */
371 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
372 charge+jnrC+0,charge+jnrD+0);
374 fjx0 = _mm_setzero_ps();
375 fjy0 = _mm_setzero_ps();
376 fjz0 = _mm_setzero_ps();
378 /**************************
379 * CALCULATE INTERACTIONS *
380 **************************/
382 /* Compute parameters for interactions between i and j atoms */
383 qq00 = _mm_mul_ps(iq0,jq0);
385 /* REACTION-FIELD ELECTROSTATICS */
386 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
387 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
389 /* Update potential sum for this i atom from the interaction with this j atom. */
390 velec = _mm_andnot_ps(dummy_mask,velec);
391 velecsum = _mm_add_ps(velecsum,velec);
395 fscal = _mm_andnot_ps(dummy_mask,fscal);
397 /* Calculate temporary vectorial force */
398 tx = _mm_mul_ps(fscal,dx00);
399 ty = _mm_mul_ps(fscal,dy00);
400 tz = _mm_mul_ps(fscal,dz00);
402 /* Update vectorial force */
403 fix0 = _mm_add_ps(fix0,tx);
404 fiy0 = _mm_add_ps(fiy0,ty);
405 fiz0 = _mm_add_ps(fiz0,tz);
407 fjx0 = _mm_add_ps(fjx0,tx);
408 fjy0 = _mm_add_ps(fjy0,ty);
409 fjz0 = _mm_add_ps(fjz0,tz);
411 /**************************
412 * CALCULATE INTERACTIONS *
413 **************************/
415 /* Compute parameters for interactions between i and j atoms */
416 qq10 = _mm_mul_ps(iq1,jq0);
418 /* REACTION-FIELD ELECTROSTATICS */
419 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
420 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
422 /* Update potential sum for this i atom from the interaction with this j atom. */
423 velec = _mm_andnot_ps(dummy_mask,velec);
424 velecsum = _mm_add_ps(velecsum,velec);
428 fscal = _mm_andnot_ps(dummy_mask,fscal);
430 /* Calculate temporary vectorial force */
431 tx = _mm_mul_ps(fscal,dx10);
432 ty = _mm_mul_ps(fscal,dy10);
433 tz = _mm_mul_ps(fscal,dz10);
435 /* Update vectorial force */
436 fix1 = _mm_add_ps(fix1,tx);
437 fiy1 = _mm_add_ps(fiy1,ty);
438 fiz1 = _mm_add_ps(fiz1,tz);
440 fjx0 = _mm_add_ps(fjx0,tx);
441 fjy0 = _mm_add_ps(fjy0,ty);
442 fjz0 = _mm_add_ps(fjz0,tz);
444 /**************************
445 * CALCULATE INTERACTIONS *
446 **************************/
448 /* Compute parameters for interactions between i and j atoms */
449 qq20 = _mm_mul_ps(iq2,jq0);
451 /* REACTION-FIELD ELECTROSTATICS */
452 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
453 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
455 /* Update potential sum for this i atom from the interaction with this j atom. */
456 velec = _mm_andnot_ps(dummy_mask,velec);
457 velecsum = _mm_add_ps(velecsum,velec);
461 fscal = _mm_andnot_ps(dummy_mask,fscal);
463 /* Calculate temporary vectorial force */
464 tx = _mm_mul_ps(fscal,dx20);
465 ty = _mm_mul_ps(fscal,dy20);
466 tz = _mm_mul_ps(fscal,dz20);
468 /* Update vectorial force */
469 fix2 = _mm_add_ps(fix2,tx);
470 fiy2 = _mm_add_ps(fiy2,ty);
471 fiz2 = _mm_add_ps(fiz2,tz);
473 fjx0 = _mm_add_ps(fjx0,tx);
474 fjy0 = _mm_add_ps(fjy0,ty);
475 fjz0 = _mm_add_ps(fjz0,tz);
477 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
478 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
479 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
480 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
482 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
484 /* Inner loop uses 96 flops */
487 /* End of innermost loop */
489 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
490 f+i_coord_offset,fshift+i_shift_offset);
493 /* Update potential energies */
494 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
496 /* Increment number of inner iterations */
497 inneriter += j_index_end - j_index_start;
499 /* Outer loop uses 19 flops */
502 /* Increment number of outer iterations */
505 /* Update outer/inner flops */
507 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*96);
510 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW3P1_F_sse4_1_single
511 * Electrostatics interaction: ReactionField
512 * VdW interaction: None
513 * Geometry: Water3-Particle
514 * Calculate force/pot: Force
517 nb_kernel_ElecRF_VdwNone_GeomW3P1_F_sse4_1_single
518 (t_nblist * gmx_restrict nlist,
519 rvec * gmx_restrict xx,
520 rvec * gmx_restrict ff,
521 struct t_forcerec * gmx_restrict fr,
522 t_mdatoms * gmx_restrict mdatoms,
523 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
524 t_nrnb * gmx_restrict nrnb)
526 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
527 * just 0 for non-waters.
528 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
529 * jnr indices corresponding to data put in the four positions in the SIMD register.
531 int i_shift_offset,i_coord_offset,outeriter,inneriter;
532 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
533 int jnrA,jnrB,jnrC,jnrD;
534 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
535 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
536 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
538 real *shiftvec,*fshift,*x,*f;
539 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
541 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
543 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
545 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
547 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
548 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
549 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
550 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
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 velec,felec,velecsum,facel,crf,krf,krf2;
555 __m128 dummy_mask,cutoff_mask;
556 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
557 __m128 one = _mm_set1_ps(1.0);
558 __m128 two = _mm_set1_ps(2.0);
564 jindex = nlist->jindex;
566 shiftidx = nlist->shift;
568 shiftvec = fr->shift_vec[0];
569 fshift = fr->fshift[0];
570 facel = _mm_set1_ps(fr->ic->epsfac);
571 charge = mdatoms->chargeA;
572 krf = _mm_set1_ps(fr->ic->k_rf);
573 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
574 crf = _mm_set1_ps(fr->ic->c_rf);
576 /* Setup water-specific parameters */
577 inr = nlist->iinr[0];
578 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+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]));
582 /* Avoid stupid compiler warnings */
583 jnrA = jnrB = jnrC = jnrD = 0;
592 for(iidx=0;iidx<4*DIM;iidx++)
597 /* Start outer loop over neighborlists */
598 for(iidx=0; iidx<nri; iidx++)
600 /* Load shift vector for this list */
601 i_shift_offset = DIM*shiftidx[iidx];
603 /* Load limits for loop over neighbors */
604 j_index_start = jindex[iidx];
605 j_index_end = jindex[iidx+1];
607 /* Get outer coordinate index */
609 i_coord_offset = DIM*inr;
611 /* Load i particle coords and add shift vector */
612 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
613 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
615 fix0 = _mm_setzero_ps();
616 fiy0 = _mm_setzero_ps();
617 fiz0 = _mm_setzero_ps();
618 fix1 = _mm_setzero_ps();
619 fiy1 = _mm_setzero_ps();
620 fiz1 = _mm_setzero_ps();
621 fix2 = _mm_setzero_ps();
622 fiy2 = _mm_setzero_ps();
623 fiz2 = _mm_setzero_ps();
625 /* Start inner kernel loop */
626 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
629 /* Get j neighbor index, and coordinate index */
634 j_coord_offsetA = DIM*jnrA;
635 j_coord_offsetB = DIM*jnrB;
636 j_coord_offsetC = DIM*jnrC;
637 j_coord_offsetD = DIM*jnrD;
639 /* load j atom coordinates */
640 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
641 x+j_coord_offsetC,x+j_coord_offsetD,
644 /* Calculate displacement vector */
645 dx00 = _mm_sub_ps(ix0,jx0);
646 dy00 = _mm_sub_ps(iy0,jy0);
647 dz00 = _mm_sub_ps(iz0,jz0);
648 dx10 = _mm_sub_ps(ix1,jx0);
649 dy10 = _mm_sub_ps(iy1,jy0);
650 dz10 = _mm_sub_ps(iz1,jz0);
651 dx20 = _mm_sub_ps(ix2,jx0);
652 dy20 = _mm_sub_ps(iy2,jy0);
653 dz20 = _mm_sub_ps(iz2,jz0);
655 /* Calculate squared distance and things based on it */
656 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
657 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
658 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
660 rinv00 = sse41_invsqrt_f(rsq00);
661 rinv10 = sse41_invsqrt_f(rsq10);
662 rinv20 = sse41_invsqrt_f(rsq20);
664 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
665 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
666 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
668 /* Load parameters for j particles */
669 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
670 charge+jnrC+0,charge+jnrD+0);
672 fjx0 = _mm_setzero_ps();
673 fjy0 = _mm_setzero_ps();
674 fjz0 = _mm_setzero_ps();
676 /**************************
677 * CALCULATE INTERACTIONS *
678 **************************/
680 /* Compute parameters for interactions between i and j atoms */
681 qq00 = _mm_mul_ps(iq0,jq0);
683 /* REACTION-FIELD ELECTROSTATICS */
684 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
688 /* Calculate temporary vectorial force */
689 tx = _mm_mul_ps(fscal,dx00);
690 ty = _mm_mul_ps(fscal,dy00);
691 tz = _mm_mul_ps(fscal,dz00);
693 /* Update vectorial force */
694 fix0 = _mm_add_ps(fix0,tx);
695 fiy0 = _mm_add_ps(fiy0,ty);
696 fiz0 = _mm_add_ps(fiz0,tz);
698 fjx0 = _mm_add_ps(fjx0,tx);
699 fjy0 = _mm_add_ps(fjy0,ty);
700 fjz0 = _mm_add_ps(fjz0,tz);
702 /**************************
703 * CALCULATE INTERACTIONS *
704 **************************/
706 /* Compute parameters for interactions between i and j atoms */
707 qq10 = _mm_mul_ps(iq1,jq0);
709 /* REACTION-FIELD ELECTROSTATICS */
710 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
714 /* Calculate temporary vectorial force */
715 tx = _mm_mul_ps(fscal,dx10);
716 ty = _mm_mul_ps(fscal,dy10);
717 tz = _mm_mul_ps(fscal,dz10);
719 /* Update vectorial force */
720 fix1 = _mm_add_ps(fix1,tx);
721 fiy1 = _mm_add_ps(fiy1,ty);
722 fiz1 = _mm_add_ps(fiz1,tz);
724 fjx0 = _mm_add_ps(fjx0,tx);
725 fjy0 = _mm_add_ps(fjy0,ty);
726 fjz0 = _mm_add_ps(fjz0,tz);
728 /**************************
729 * CALCULATE INTERACTIONS *
730 **************************/
732 /* Compute parameters for interactions between i and j atoms */
733 qq20 = _mm_mul_ps(iq2,jq0);
735 /* REACTION-FIELD ELECTROSTATICS */
736 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
740 /* Calculate temporary vectorial force */
741 tx = _mm_mul_ps(fscal,dx20);
742 ty = _mm_mul_ps(fscal,dy20);
743 tz = _mm_mul_ps(fscal,dz20);
745 /* Update vectorial force */
746 fix2 = _mm_add_ps(fix2,tx);
747 fiy2 = _mm_add_ps(fiy2,ty);
748 fiz2 = _mm_add_ps(fiz2,tz);
750 fjx0 = _mm_add_ps(fjx0,tx);
751 fjy0 = _mm_add_ps(fjy0,ty);
752 fjz0 = _mm_add_ps(fjz0,tz);
754 fjptrA = f+j_coord_offsetA;
755 fjptrB = f+j_coord_offsetB;
756 fjptrC = f+j_coord_offsetC;
757 fjptrD = f+j_coord_offsetD;
759 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
761 /* Inner loop uses 81 flops */
767 /* Get j neighbor index, and coordinate index */
768 jnrlistA = jjnr[jidx];
769 jnrlistB = jjnr[jidx+1];
770 jnrlistC = jjnr[jidx+2];
771 jnrlistD = jjnr[jidx+3];
772 /* Sign of each element will be negative for non-real atoms.
773 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
774 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
776 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
777 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
778 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
779 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
780 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
781 j_coord_offsetA = DIM*jnrA;
782 j_coord_offsetB = DIM*jnrB;
783 j_coord_offsetC = DIM*jnrC;
784 j_coord_offsetD = DIM*jnrD;
786 /* load j atom coordinates */
787 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
788 x+j_coord_offsetC,x+j_coord_offsetD,
791 /* Calculate displacement vector */
792 dx00 = _mm_sub_ps(ix0,jx0);
793 dy00 = _mm_sub_ps(iy0,jy0);
794 dz00 = _mm_sub_ps(iz0,jz0);
795 dx10 = _mm_sub_ps(ix1,jx0);
796 dy10 = _mm_sub_ps(iy1,jy0);
797 dz10 = _mm_sub_ps(iz1,jz0);
798 dx20 = _mm_sub_ps(ix2,jx0);
799 dy20 = _mm_sub_ps(iy2,jy0);
800 dz20 = _mm_sub_ps(iz2,jz0);
802 /* Calculate squared distance and things based on it */
803 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
804 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
805 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
807 rinv00 = sse41_invsqrt_f(rsq00);
808 rinv10 = sse41_invsqrt_f(rsq10);
809 rinv20 = sse41_invsqrt_f(rsq20);
811 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
812 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
813 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
815 /* Load parameters for j particles */
816 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
817 charge+jnrC+0,charge+jnrD+0);
819 fjx0 = _mm_setzero_ps();
820 fjy0 = _mm_setzero_ps();
821 fjz0 = _mm_setzero_ps();
823 /**************************
824 * CALCULATE INTERACTIONS *
825 **************************/
827 /* Compute parameters for interactions between i and j atoms */
828 qq00 = _mm_mul_ps(iq0,jq0);
830 /* REACTION-FIELD ELECTROSTATICS */
831 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
835 fscal = _mm_andnot_ps(dummy_mask,fscal);
837 /* Calculate temporary vectorial force */
838 tx = _mm_mul_ps(fscal,dx00);
839 ty = _mm_mul_ps(fscal,dy00);
840 tz = _mm_mul_ps(fscal,dz00);
842 /* Update vectorial force */
843 fix0 = _mm_add_ps(fix0,tx);
844 fiy0 = _mm_add_ps(fiy0,ty);
845 fiz0 = _mm_add_ps(fiz0,tz);
847 fjx0 = _mm_add_ps(fjx0,tx);
848 fjy0 = _mm_add_ps(fjy0,ty);
849 fjz0 = _mm_add_ps(fjz0,tz);
851 /**************************
852 * CALCULATE INTERACTIONS *
853 **************************/
855 /* Compute parameters for interactions between i and j atoms */
856 qq10 = _mm_mul_ps(iq1,jq0);
858 /* REACTION-FIELD ELECTROSTATICS */
859 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
863 fscal = _mm_andnot_ps(dummy_mask,fscal);
865 /* Calculate temporary vectorial force */
866 tx = _mm_mul_ps(fscal,dx10);
867 ty = _mm_mul_ps(fscal,dy10);
868 tz = _mm_mul_ps(fscal,dz10);
870 /* Update vectorial force */
871 fix1 = _mm_add_ps(fix1,tx);
872 fiy1 = _mm_add_ps(fiy1,ty);
873 fiz1 = _mm_add_ps(fiz1,tz);
875 fjx0 = _mm_add_ps(fjx0,tx);
876 fjy0 = _mm_add_ps(fjy0,ty);
877 fjz0 = _mm_add_ps(fjz0,tz);
879 /**************************
880 * CALCULATE INTERACTIONS *
881 **************************/
883 /* Compute parameters for interactions between i and j atoms */
884 qq20 = _mm_mul_ps(iq2,jq0);
886 /* REACTION-FIELD ELECTROSTATICS */
887 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
891 fscal = _mm_andnot_ps(dummy_mask,fscal);
893 /* Calculate temporary vectorial force */
894 tx = _mm_mul_ps(fscal,dx20);
895 ty = _mm_mul_ps(fscal,dy20);
896 tz = _mm_mul_ps(fscal,dz20);
898 /* Update vectorial force */
899 fix2 = _mm_add_ps(fix2,tx);
900 fiy2 = _mm_add_ps(fiy2,ty);
901 fiz2 = _mm_add_ps(fiz2,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(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
920 f+i_coord_offset,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_W3_F,outeriter*18 + inneriter*81);