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 sse2_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_sse2_single.h"
50 #include "kernelutil_x86_sse2_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW3P1_VF_sse2_single
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: None
56 * Geometry: Water3-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRF_VdwNone_GeomW3P1_VF_sse2_single
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
79 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
81 real *shiftvec,*fshift,*x,*f;
82 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
84 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
91 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
92 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
93 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
94 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
95 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
96 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
98 __m128 dummy_mask,cutoff_mask;
99 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
100 __m128 one = _mm_set1_ps(1.0);
101 __m128 two = _mm_set1_ps(2.0);
107 jindex = nlist->jindex;
109 shiftidx = nlist->shift;
111 shiftvec = fr->shift_vec[0];
112 fshift = fr->fshift[0];
113 facel = _mm_set1_ps(fr->epsfac);
114 charge = mdatoms->chargeA;
115 krf = _mm_set1_ps(fr->ic->k_rf);
116 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
117 crf = _mm_set1_ps(fr->ic->c_rf);
119 /* Setup water-specific parameters */
120 inr = nlist->iinr[0];
121 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
122 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
123 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
125 /* Avoid stupid compiler warnings */
126 jnrA = jnrB = jnrC = jnrD = 0;
135 for(iidx=0;iidx<4*DIM;iidx++)
140 /* Start outer loop over neighborlists */
141 for(iidx=0; iidx<nri; iidx++)
143 /* Load shift vector for this list */
144 i_shift_offset = DIM*shiftidx[iidx];
146 /* Load limits for loop over neighbors */
147 j_index_start = jindex[iidx];
148 j_index_end = jindex[iidx+1];
150 /* Get outer coordinate index */
152 i_coord_offset = DIM*inr;
154 /* Load i particle coords and add shift vector */
155 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
156 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
158 fix0 = _mm_setzero_ps();
159 fiy0 = _mm_setzero_ps();
160 fiz0 = _mm_setzero_ps();
161 fix1 = _mm_setzero_ps();
162 fiy1 = _mm_setzero_ps();
163 fiz1 = _mm_setzero_ps();
164 fix2 = _mm_setzero_ps();
165 fiy2 = _mm_setzero_ps();
166 fiz2 = _mm_setzero_ps();
168 /* Reset potential sums */
169 velecsum = _mm_setzero_ps();
171 /* Start inner kernel loop */
172 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
175 /* Get j neighbor index, and coordinate index */
180 j_coord_offsetA = DIM*jnrA;
181 j_coord_offsetB = DIM*jnrB;
182 j_coord_offsetC = DIM*jnrC;
183 j_coord_offsetD = DIM*jnrD;
185 /* load j atom coordinates */
186 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
187 x+j_coord_offsetC,x+j_coord_offsetD,
190 /* Calculate displacement vector */
191 dx00 = _mm_sub_ps(ix0,jx0);
192 dy00 = _mm_sub_ps(iy0,jy0);
193 dz00 = _mm_sub_ps(iz0,jz0);
194 dx10 = _mm_sub_ps(ix1,jx0);
195 dy10 = _mm_sub_ps(iy1,jy0);
196 dz10 = _mm_sub_ps(iz1,jz0);
197 dx20 = _mm_sub_ps(ix2,jx0);
198 dy20 = _mm_sub_ps(iy2,jy0);
199 dz20 = _mm_sub_ps(iz2,jz0);
201 /* Calculate squared distance and things based on it */
202 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
203 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
204 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
206 rinv00 = gmx_mm_invsqrt_ps(rsq00);
207 rinv10 = gmx_mm_invsqrt_ps(rsq10);
208 rinv20 = gmx_mm_invsqrt_ps(rsq20);
210 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
211 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
212 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
214 /* Load parameters for j particles */
215 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
216 charge+jnrC+0,charge+jnrD+0);
218 fjx0 = _mm_setzero_ps();
219 fjy0 = _mm_setzero_ps();
220 fjz0 = _mm_setzero_ps();
222 /**************************
223 * CALCULATE INTERACTIONS *
224 **************************/
226 /* Compute parameters for interactions between i and j atoms */
227 qq00 = _mm_mul_ps(iq0,jq0);
229 /* REACTION-FIELD ELECTROSTATICS */
230 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
231 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
233 /* Update potential sum for this i atom from the interaction with this j atom. */
234 velecsum = _mm_add_ps(velecsum,velec);
238 /* Calculate temporary vectorial force */
239 tx = _mm_mul_ps(fscal,dx00);
240 ty = _mm_mul_ps(fscal,dy00);
241 tz = _mm_mul_ps(fscal,dz00);
243 /* Update vectorial force */
244 fix0 = _mm_add_ps(fix0,tx);
245 fiy0 = _mm_add_ps(fiy0,ty);
246 fiz0 = _mm_add_ps(fiz0,tz);
248 fjx0 = _mm_add_ps(fjx0,tx);
249 fjy0 = _mm_add_ps(fjy0,ty);
250 fjz0 = _mm_add_ps(fjz0,tz);
252 /**************************
253 * CALCULATE INTERACTIONS *
254 **************************/
256 /* Compute parameters for interactions between i and j atoms */
257 qq10 = _mm_mul_ps(iq1,jq0);
259 /* REACTION-FIELD ELECTROSTATICS */
260 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
261 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
263 /* Update potential sum for this i atom from the interaction with this j atom. */
264 velecsum = _mm_add_ps(velecsum,velec);
268 /* Calculate temporary vectorial force */
269 tx = _mm_mul_ps(fscal,dx10);
270 ty = _mm_mul_ps(fscal,dy10);
271 tz = _mm_mul_ps(fscal,dz10);
273 /* Update vectorial force */
274 fix1 = _mm_add_ps(fix1,tx);
275 fiy1 = _mm_add_ps(fiy1,ty);
276 fiz1 = _mm_add_ps(fiz1,tz);
278 fjx0 = _mm_add_ps(fjx0,tx);
279 fjy0 = _mm_add_ps(fjy0,ty);
280 fjz0 = _mm_add_ps(fjz0,tz);
282 /**************************
283 * CALCULATE INTERACTIONS *
284 **************************/
286 /* Compute parameters for interactions between i and j atoms */
287 qq20 = _mm_mul_ps(iq2,jq0);
289 /* REACTION-FIELD ELECTROSTATICS */
290 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
291 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
293 /* Update potential sum for this i atom from the interaction with this j atom. */
294 velecsum = _mm_add_ps(velecsum,velec);
298 /* Calculate temporary vectorial force */
299 tx = _mm_mul_ps(fscal,dx20);
300 ty = _mm_mul_ps(fscal,dy20);
301 tz = _mm_mul_ps(fscal,dz20);
303 /* Update vectorial force */
304 fix2 = _mm_add_ps(fix2,tx);
305 fiy2 = _mm_add_ps(fiy2,ty);
306 fiz2 = _mm_add_ps(fiz2,tz);
308 fjx0 = _mm_add_ps(fjx0,tx);
309 fjy0 = _mm_add_ps(fjy0,ty);
310 fjz0 = _mm_add_ps(fjz0,tz);
312 fjptrA = f+j_coord_offsetA;
313 fjptrB = f+j_coord_offsetB;
314 fjptrC = f+j_coord_offsetC;
315 fjptrD = f+j_coord_offsetD;
317 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
319 /* Inner loop uses 96 flops */
325 /* Get j neighbor index, and coordinate index */
326 jnrlistA = jjnr[jidx];
327 jnrlistB = jjnr[jidx+1];
328 jnrlistC = jjnr[jidx+2];
329 jnrlistD = jjnr[jidx+3];
330 /* Sign of each element will be negative for non-real atoms.
331 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
332 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
334 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
335 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
336 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
337 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
338 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
339 j_coord_offsetA = DIM*jnrA;
340 j_coord_offsetB = DIM*jnrB;
341 j_coord_offsetC = DIM*jnrC;
342 j_coord_offsetD = DIM*jnrD;
344 /* load j atom coordinates */
345 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
346 x+j_coord_offsetC,x+j_coord_offsetD,
349 /* Calculate displacement vector */
350 dx00 = _mm_sub_ps(ix0,jx0);
351 dy00 = _mm_sub_ps(iy0,jy0);
352 dz00 = _mm_sub_ps(iz0,jz0);
353 dx10 = _mm_sub_ps(ix1,jx0);
354 dy10 = _mm_sub_ps(iy1,jy0);
355 dz10 = _mm_sub_ps(iz1,jz0);
356 dx20 = _mm_sub_ps(ix2,jx0);
357 dy20 = _mm_sub_ps(iy2,jy0);
358 dz20 = _mm_sub_ps(iz2,jz0);
360 /* Calculate squared distance and things based on it */
361 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
362 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
363 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
365 rinv00 = gmx_mm_invsqrt_ps(rsq00);
366 rinv10 = gmx_mm_invsqrt_ps(rsq10);
367 rinv20 = gmx_mm_invsqrt_ps(rsq20);
369 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
370 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
371 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
373 /* Load parameters for j particles */
374 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
375 charge+jnrC+0,charge+jnrD+0);
377 fjx0 = _mm_setzero_ps();
378 fjy0 = _mm_setzero_ps();
379 fjz0 = _mm_setzero_ps();
381 /**************************
382 * CALCULATE INTERACTIONS *
383 **************************/
385 /* Compute parameters for interactions between i and j atoms */
386 qq00 = _mm_mul_ps(iq0,jq0);
388 /* REACTION-FIELD ELECTROSTATICS */
389 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
390 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
392 /* Update potential sum for this i atom from the interaction with this j atom. */
393 velec = _mm_andnot_ps(dummy_mask,velec);
394 velecsum = _mm_add_ps(velecsum,velec);
398 fscal = _mm_andnot_ps(dummy_mask,fscal);
400 /* Calculate temporary vectorial force */
401 tx = _mm_mul_ps(fscal,dx00);
402 ty = _mm_mul_ps(fscal,dy00);
403 tz = _mm_mul_ps(fscal,dz00);
405 /* Update vectorial force */
406 fix0 = _mm_add_ps(fix0,tx);
407 fiy0 = _mm_add_ps(fiy0,ty);
408 fiz0 = _mm_add_ps(fiz0,tz);
410 fjx0 = _mm_add_ps(fjx0,tx);
411 fjy0 = _mm_add_ps(fjy0,ty);
412 fjz0 = _mm_add_ps(fjz0,tz);
414 /**************************
415 * CALCULATE INTERACTIONS *
416 **************************/
418 /* Compute parameters for interactions between i and j atoms */
419 qq10 = _mm_mul_ps(iq1,jq0);
421 /* REACTION-FIELD ELECTROSTATICS */
422 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
423 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
425 /* Update potential sum for this i atom from the interaction with this j atom. */
426 velec = _mm_andnot_ps(dummy_mask,velec);
427 velecsum = _mm_add_ps(velecsum,velec);
431 fscal = _mm_andnot_ps(dummy_mask,fscal);
433 /* Calculate temporary vectorial force */
434 tx = _mm_mul_ps(fscal,dx10);
435 ty = _mm_mul_ps(fscal,dy10);
436 tz = _mm_mul_ps(fscal,dz10);
438 /* Update vectorial force */
439 fix1 = _mm_add_ps(fix1,tx);
440 fiy1 = _mm_add_ps(fiy1,ty);
441 fiz1 = _mm_add_ps(fiz1,tz);
443 fjx0 = _mm_add_ps(fjx0,tx);
444 fjy0 = _mm_add_ps(fjy0,ty);
445 fjz0 = _mm_add_ps(fjz0,tz);
447 /**************************
448 * CALCULATE INTERACTIONS *
449 **************************/
451 /* Compute parameters for interactions between i and j atoms */
452 qq20 = _mm_mul_ps(iq2,jq0);
454 /* REACTION-FIELD ELECTROSTATICS */
455 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
456 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
458 /* Update potential sum for this i atom from the interaction with this j atom. */
459 velec = _mm_andnot_ps(dummy_mask,velec);
460 velecsum = _mm_add_ps(velecsum,velec);
464 fscal = _mm_andnot_ps(dummy_mask,fscal);
466 /* Calculate temporary vectorial force */
467 tx = _mm_mul_ps(fscal,dx20);
468 ty = _mm_mul_ps(fscal,dy20);
469 tz = _mm_mul_ps(fscal,dz20);
471 /* Update vectorial force */
472 fix2 = _mm_add_ps(fix2,tx);
473 fiy2 = _mm_add_ps(fiy2,ty);
474 fiz2 = _mm_add_ps(fiz2,tz);
476 fjx0 = _mm_add_ps(fjx0,tx);
477 fjy0 = _mm_add_ps(fjy0,ty);
478 fjz0 = _mm_add_ps(fjz0,tz);
480 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
481 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
482 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
483 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
485 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
487 /* Inner loop uses 96 flops */
490 /* End of innermost loop */
492 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
493 f+i_coord_offset,fshift+i_shift_offset);
496 /* Update potential energies */
497 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
499 /* Increment number of inner iterations */
500 inneriter += j_index_end - j_index_start;
502 /* Outer loop uses 19 flops */
505 /* Increment number of outer iterations */
508 /* Update outer/inner flops */
510 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*96);
513 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW3P1_F_sse2_single
514 * Electrostatics interaction: ReactionField
515 * VdW interaction: None
516 * Geometry: Water3-Particle
517 * Calculate force/pot: Force
520 nb_kernel_ElecRF_VdwNone_GeomW3P1_F_sse2_single
521 (t_nblist * gmx_restrict nlist,
522 rvec * gmx_restrict xx,
523 rvec * gmx_restrict ff,
524 t_forcerec * gmx_restrict fr,
525 t_mdatoms * gmx_restrict mdatoms,
526 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
527 t_nrnb * gmx_restrict nrnb)
529 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
530 * just 0 for non-waters.
531 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
532 * jnr indices corresponding to data put in the four positions in the SIMD register.
534 int i_shift_offset,i_coord_offset,outeriter,inneriter;
535 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
536 int jnrA,jnrB,jnrC,jnrD;
537 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
538 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
539 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
541 real *shiftvec,*fshift,*x,*f;
542 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
544 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
546 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
548 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
550 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
551 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
552 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
553 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
554 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
555 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
556 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
558 __m128 dummy_mask,cutoff_mask;
559 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
560 __m128 one = _mm_set1_ps(1.0);
561 __m128 two = _mm_set1_ps(2.0);
567 jindex = nlist->jindex;
569 shiftidx = nlist->shift;
571 shiftvec = fr->shift_vec[0];
572 fshift = fr->fshift[0];
573 facel = _mm_set1_ps(fr->epsfac);
574 charge = mdatoms->chargeA;
575 krf = _mm_set1_ps(fr->ic->k_rf);
576 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
577 crf = _mm_set1_ps(fr->ic->c_rf);
579 /* Setup water-specific parameters */
580 inr = nlist->iinr[0];
581 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
582 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
583 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
585 /* Avoid stupid compiler warnings */
586 jnrA = jnrB = jnrC = jnrD = 0;
595 for(iidx=0;iidx<4*DIM;iidx++)
600 /* Start outer loop over neighborlists */
601 for(iidx=0; iidx<nri; iidx++)
603 /* Load shift vector for this list */
604 i_shift_offset = DIM*shiftidx[iidx];
606 /* Load limits for loop over neighbors */
607 j_index_start = jindex[iidx];
608 j_index_end = jindex[iidx+1];
610 /* Get outer coordinate index */
612 i_coord_offset = DIM*inr;
614 /* Load i particle coords and add shift vector */
615 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
616 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
618 fix0 = _mm_setzero_ps();
619 fiy0 = _mm_setzero_ps();
620 fiz0 = _mm_setzero_ps();
621 fix1 = _mm_setzero_ps();
622 fiy1 = _mm_setzero_ps();
623 fiz1 = _mm_setzero_ps();
624 fix2 = _mm_setzero_ps();
625 fiy2 = _mm_setzero_ps();
626 fiz2 = _mm_setzero_ps();
628 /* Start inner kernel loop */
629 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
632 /* Get j neighbor index, and coordinate index */
637 j_coord_offsetA = DIM*jnrA;
638 j_coord_offsetB = DIM*jnrB;
639 j_coord_offsetC = DIM*jnrC;
640 j_coord_offsetD = DIM*jnrD;
642 /* load j atom coordinates */
643 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
644 x+j_coord_offsetC,x+j_coord_offsetD,
647 /* Calculate displacement vector */
648 dx00 = _mm_sub_ps(ix0,jx0);
649 dy00 = _mm_sub_ps(iy0,jy0);
650 dz00 = _mm_sub_ps(iz0,jz0);
651 dx10 = _mm_sub_ps(ix1,jx0);
652 dy10 = _mm_sub_ps(iy1,jy0);
653 dz10 = _mm_sub_ps(iz1,jz0);
654 dx20 = _mm_sub_ps(ix2,jx0);
655 dy20 = _mm_sub_ps(iy2,jy0);
656 dz20 = _mm_sub_ps(iz2,jz0);
658 /* Calculate squared distance and things based on it */
659 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
660 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
661 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
663 rinv00 = gmx_mm_invsqrt_ps(rsq00);
664 rinv10 = gmx_mm_invsqrt_ps(rsq10);
665 rinv20 = gmx_mm_invsqrt_ps(rsq20);
667 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
668 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
669 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
671 /* Load parameters for j particles */
672 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
673 charge+jnrC+0,charge+jnrD+0);
675 fjx0 = _mm_setzero_ps();
676 fjy0 = _mm_setzero_ps();
677 fjz0 = _mm_setzero_ps();
679 /**************************
680 * CALCULATE INTERACTIONS *
681 **************************/
683 /* Compute parameters for interactions between i and j atoms */
684 qq00 = _mm_mul_ps(iq0,jq0);
686 /* REACTION-FIELD ELECTROSTATICS */
687 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
691 /* Calculate temporary vectorial force */
692 tx = _mm_mul_ps(fscal,dx00);
693 ty = _mm_mul_ps(fscal,dy00);
694 tz = _mm_mul_ps(fscal,dz00);
696 /* Update vectorial force */
697 fix0 = _mm_add_ps(fix0,tx);
698 fiy0 = _mm_add_ps(fiy0,ty);
699 fiz0 = _mm_add_ps(fiz0,tz);
701 fjx0 = _mm_add_ps(fjx0,tx);
702 fjy0 = _mm_add_ps(fjy0,ty);
703 fjz0 = _mm_add_ps(fjz0,tz);
705 /**************************
706 * CALCULATE INTERACTIONS *
707 **************************/
709 /* Compute parameters for interactions between i and j atoms */
710 qq10 = _mm_mul_ps(iq1,jq0);
712 /* REACTION-FIELD ELECTROSTATICS */
713 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
717 /* Calculate temporary vectorial force */
718 tx = _mm_mul_ps(fscal,dx10);
719 ty = _mm_mul_ps(fscal,dy10);
720 tz = _mm_mul_ps(fscal,dz10);
722 /* Update vectorial force */
723 fix1 = _mm_add_ps(fix1,tx);
724 fiy1 = _mm_add_ps(fiy1,ty);
725 fiz1 = _mm_add_ps(fiz1,tz);
727 fjx0 = _mm_add_ps(fjx0,tx);
728 fjy0 = _mm_add_ps(fjy0,ty);
729 fjz0 = _mm_add_ps(fjz0,tz);
731 /**************************
732 * CALCULATE INTERACTIONS *
733 **************************/
735 /* Compute parameters for interactions between i and j atoms */
736 qq20 = _mm_mul_ps(iq2,jq0);
738 /* REACTION-FIELD ELECTROSTATICS */
739 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
743 /* Calculate temporary vectorial force */
744 tx = _mm_mul_ps(fscal,dx20);
745 ty = _mm_mul_ps(fscal,dy20);
746 tz = _mm_mul_ps(fscal,dz20);
748 /* Update vectorial force */
749 fix2 = _mm_add_ps(fix2,tx);
750 fiy2 = _mm_add_ps(fiy2,ty);
751 fiz2 = _mm_add_ps(fiz2,tz);
753 fjx0 = _mm_add_ps(fjx0,tx);
754 fjy0 = _mm_add_ps(fjy0,ty);
755 fjz0 = _mm_add_ps(fjz0,tz);
757 fjptrA = f+j_coord_offsetA;
758 fjptrB = f+j_coord_offsetB;
759 fjptrC = f+j_coord_offsetC;
760 fjptrD = f+j_coord_offsetD;
762 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
764 /* Inner loop uses 81 flops */
770 /* Get j neighbor index, and coordinate index */
771 jnrlistA = jjnr[jidx];
772 jnrlistB = jjnr[jidx+1];
773 jnrlistC = jjnr[jidx+2];
774 jnrlistD = jjnr[jidx+3];
775 /* Sign of each element will be negative for non-real atoms.
776 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
777 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
779 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
780 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
781 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
782 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
783 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
784 j_coord_offsetA = DIM*jnrA;
785 j_coord_offsetB = DIM*jnrB;
786 j_coord_offsetC = DIM*jnrC;
787 j_coord_offsetD = DIM*jnrD;
789 /* load j atom coordinates */
790 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
791 x+j_coord_offsetC,x+j_coord_offsetD,
794 /* Calculate displacement vector */
795 dx00 = _mm_sub_ps(ix0,jx0);
796 dy00 = _mm_sub_ps(iy0,jy0);
797 dz00 = _mm_sub_ps(iz0,jz0);
798 dx10 = _mm_sub_ps(ix1,jx0);
799 dy10 = _mm_sub_ps(iy1,jy0);
800 dz10 = _mm_sub_ps(iz1,jz0);
801 dx20 = _mm_sub_ps(ix2,jx0);
802 dy20 = _mm_sub_ps(iy2,jy0);
803 dz20 = _mm_sub_ps(iz2,jz0);
805 /* Calculate squared distance and things based on it */
806 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
807 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
808 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
810 rinv00 = gmx_mm_invsqrt_ps(rsq00);
811 rinv10 = gmx_mm_invsqrt_ps(rsq10);
812 rinv20 = gmx_mm_invsqrt_ps(rsq20);
814 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
815 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
816 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
818 /* Load parameters for j particles */
819 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
820 charge+jnrC+0,charge+jnrD+0);
822 fjx0 = _mm_setzero_ps();
823 fjy0 = _mm_setzero_ps();
824 fjz0 = _mm_setzero_ps();
826 /**************************
827 * CALCULATE INTERACTIONS *
828 **************************/
830 /* Compute parameters for interactions between i and j atoms */
831 qq00 = _mm_mul_ps(iq0,jq0);
833 /* REACTION-FIELD ELECTROSTATICS */
834 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
838 fscal = _mm_andnot_ps(dummy_mask,fscal);
840 /* Calculate temporary vectorial force */
841 tx = _mm_mul_ps(fscal,dx00);
842 ty = _mm_mul_ps(fscal,dy00);
843 tz = _mm_mul_ps(fscal,dz00);
845 /* Update vectorial force */
846 fix0 = _mm_add_ps(fix0,tx);
847 fiy0 = _mm_add_ps(fiy0,ty);
848 fiz0 = _mm_add_ps(fiz0,tz);
850 fjx0 = _mm_add_ps(fjx0,tx);
851 fjy0 = _mm_add_ps(fjy0,ty);
852 fjz0 = _mm_add_ps(fjz0,tz);
854 /**************************
855 * CALCULATE INTERACTIONS *
856 **************************/
858 /* Compute parameters for interactions between i and j atoms */
859 qq10 = _mm_mul_ps(iq1,jq0);
861 /* REACTION-FIELD ELECTROSTATICS */
862 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
866 fscal = _mm_andnot_ps(dummy_mask,fscal);
868 /* Calculate temporary vectorial force */
869 tx = _mm_mul_ps(fscal,dx10);
870 ty = _mm_mul_ps(fscal,dy10);
871 tz = _mm_mul_ps(fscal,dz10);
873 /* Update vectorial force */
874 fix1 = _mm_add_ps(fix1,tx);
875 fiy1 = _mm_add_ps(fiy1,ty);
876 fiz1 = _mm_add_ps(fiz1,tz);
878 fjx0 = _mm_add_ps(fjx0,tx);
879 fjy0 = _mm_add_ps(fjy0,ty);
880 fjz0 = _mm_add_ps(fjz0,tz);
882 /**************************
883 * CALCULATE INTERACTIONS *
884 **************************/
886 /* Compute parameters for interactions between i and j atoms */
887 qq20 = _mm_mul_ps(iq2,jq0);
889 /* REACTION-FIELD ELECTROSTATICS */
890 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
894 fscal = _mm_andnot_ps(dummy_mask,fscal);
896 /* Calculate temporary vectorial force */
897 tx = _mm_mul_ps(fscal,dx20);
898 ty = _mm_mul_ps(fscal,dy20);
899 tz = _mm_mul_ps(fscal,dz20);
901 /* Update vectorial force */
902 fix2 = _mm_add_ps(fix2,tx);
903 fiy2 = _mm_add_ps(fiy2,ty);
904 fiz2 = _mm_add_ps(fiz2,tz);
906 fjx0 = _mm_add_ps(fjx0,tx);
907 fjy0 = _mm_add_ps(fjy0,ty);
908 fjz0 = _mm_add_ps(fjz0,tz);
910 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
911 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
912 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
913 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
915 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
917 /* Inner loop uses 81 flops */
920 /* End of innermost loop */
922 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
923 f+i_coord_offset,fshift+i_shift_offset);
925 /* Increment number of inner iterations */
926 inneriter += j_index_end - j_index_start;
928 /* Outer loop uses 18 flops */
931 /* Increment number of outer iterations */
934 /* Update outer/inner flops */
936 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*81);