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_ElecRFCut_VdwLJSw_GeomW4P1_VF_sse4_1_single
52 * Electrostatics interaction: ReactionField
53 * VdW interaction: LennardJones
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecRFCut_VdwLJSw_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 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
86 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
90 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
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 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
97 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
100 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
103 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
104 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
105 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
106 real rswitch_scalar,d_scalar;
107 __m128 dummy_mask,cutoff_mask;
108 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
109 __m128 one = _mm_set1_ps(1.0);
110 __m128 two = _mm_set1_ps(2.0);
116 jindex = nlist->jindex;
118 shiftidx = nlist->shift;
120 shiftvec = fr->shift_vec[0];
121 fshift = fr->fshift[0];
122 facel = _mm_set1_ps(fr->epsfac);
123 charge = mdatoms->chargeA;
124 krf = _mm_set1_ps(fr->ic->k_rf);
125 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
126 crf = _mm_set1_ps(fr->ic->c_rf);
127 nvdwtype = fr->ntype;
129 vdwtype = mdatoms->typeA;
131 /* Setup water-specific parameters */
132 inr = nlist->iinr[0];
133 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
134 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
135 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
136 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
138 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
139 rcutoff_scalar = fr->rcoulomb;
140 rcutoff = _mm_set1_ps(rcutoff_scalar);
141 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
143 rswitch_scalar = fr->rvdw_switch;
144 rswitch = _mm_set1_ps(rswitch_scalar);
145 /* Setup switch parameters */
146 d_scalar = rcutoff_scalar-rswitch_scalar;
147 d = _mm_set1_ps(d_scalar);
148 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
149 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
150 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
151 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
152 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
153 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
155 /* Avoid stupid compiler warnings */
156 jnrA = jnrB = jnrC = jnrD = 0;
165 for(iidx=0;iidx<4*DIM;iidx++)
170 /* Start outer loop over neighborlists */
171 for(iidx=0; iidx<nri; iidx++)
173 /* Load shift vector for this list */
174 i_shift_offset = DIM*shiftidx[iidx];
176 /* Load limits for loop over neighbors */
177 j_index_start = jindex[iidx];
178 j_index_end = jindex[iidx+1];
180 /* Get outer coordinate index */
182 i_coord_offset = DIM*inr;
184 /* Load i particle coords and add shift vector */
185 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
186 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
188 fix0 = _mm_setzero_ps();
189 fiy0 = _mm_setzero_ps();
190 fiz0 = _mm_setzero_ps();
191 fix1 = _mm_setzero_ps();
192 fiy1 = _mm_setzero_ps();
193 fiz1 = _mm_setzero_ps();
194 fix2 = _mm_setzero_ps();
195 fiy2 = _mm_setzero_ps();
196 fiz2 = _mm_setzero_ps();
197 fix3 = _mm_setzero_ps();
198 fiy3 = _mm_setzero_ps();
199 fiz3 = _mm_setzero_ps();
201 /* Reset potential sums */
202 velecsum = _mm_setzero_ps();
203 vvdwsum = _mm_setzero_ps();
205 /* Start inner kernel loop */
206 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
209 /* Get j neighbor index, and coordinate index */
214 j_coord_offsetA = DIM*jnrA;
215 j_coord_offsetB = DIM*jnrB;
216 j_coord_offsetC = DIM*jnrC;
217 j_coord_offsetD = DIM*jnrD;
219 /* load j atom coordinates */
220 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
221 x+j_coord_offsetC,x+j_coord_offsetD,
224 /* Calculate displacement vector */
225 dx00 = _mm_sub_ps(ix0,jx0);
226 dy00 = _mm_sub_ps(iy0,jy0);
227 dz00 = _mm_sub_ps(iz0,jz0);
228 dx10 = _mm_sub_ps(ix1,jx0);
229 dy10 = _mm_sub_ps(iy1,jy0);
230 dz10 = _mm_sub_ps(iz1,jz0);
231 dx20 = _mm_sub_ps(ix2,jx0);
232 dy20 = _mm_sub_ps(iy2,jy0);
233 dz20 = _mm_sub_ps(iz2,jz0);
234 dx30 = _mm_sub_ps(ix3,jx0);
235 dy30 = _mm_sub_ps(iy3,jy0);
236 dz30 = _mm_sub_ps(iz3,jz0);
238 /* Calculate squared distance and things based on it */
239 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
240 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
241 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
242 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
244 rinv00 = gmx_mm_invsqrt_ps(rsq00);
245 rinv10 = gmx_mm_invsqrt_ps(rsq10);
246 rinv20 = gmx_mm_invsqrt_ps(rsq20);
247 rinv30 = gmx_mm_invsqrt_ps(rsq30);
249 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
250 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
251 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
252 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
254 /* Load parameters for j particles */
255 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
256 charge+jnrC+0,charge+jnrD+0);
257 vdwjidx0A = 2*vdwtype[jnrA+0];
258 vdwjidx0B = 2*vdwtype[jnrB+0];
259 vdwjidx0C = 2*vdwtype[jnrC+0];
260 vdwjidx0D = 2*vdwtype[jnrD+0];
262 fjx0 = _mm_setzero_ps();
263 fjy0 = _mm_setzero_ps();
264 fjz0 = _mm_setzero_ps();
266 /**************************
267 * CALCULATE INTERACTIONS *
268 **************************/
270 if (gmx_mm_any_lt(rsq00,rcutoff2))
273 r00 = _mm_mul_ps(rsq00,rinv00);
275 /* Compute parameters for interactions between i and j atoms */
276 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
277 vdwparam+vdwioffset0+vdwjidx0B,
278 vdwparam+vdwioffset0+vdwjidx0C,
279 vdwparam+vdwioffset0+vdwjidx0D,
282 /* LENNARD-JONES DISPERSION/REPULSION */
284 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
285 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
286 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
287 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
288 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
290 d = _mm_sub_ps(r00,rswitch);
291 d = _mm_max_ps(d,_mm_setzero_ps());
292 d2 = _mm_mul_ps(d,d);
293 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
295 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
297 /* Evaluate switch function */
298 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
299 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
300 vvdw = _mm_mul_ps(vvdw,sw);
301 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
303 /* Update potential sum for this i atom from the interaction with this j atom. */
304 vvdw = _mm_and_ps(vvdw,cutoff_mask);
305 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
309 fscal = _mm_and_ps(fscal,cutoff_mask);
311 /* Calculate temporary vectorial force */
312 tx = _mm_mul_ps(fscal,dx00);
313 ty = _mm_mul_ps(fscal,dy00);
314 tz = _mm_mul_ps(fscal,dz00);
316 /* Update vectorial force */
317 fix0 = _mm_add_ps(fix0,tx);
318 fiy0 = _mm_add_ps(fiy0,ty);
319 fiz0 = _mm_add_ps(fiz0,tz);
321 fjx0 = _mm_add_ps(fjx0,tx);
322 fjy0 = _mm_add_ps(fjy0,ty);
323 fjz0 = _mm_add_ps(fjz0,tz);
327 /**************************
328 * CALCULATE INTERACTIONS *
329 **************************/
331 if (gmx_mm_any_lt(rsq10,rcutoff2))
334 /* Compute parameters for interactions between i and j atoms */
335 qq10 = _mm_mul_ps(iq1,jq0);
337 /* REACTION-FIELD ELECTROSTATICS */
338 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
339 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
341 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
343 /* Update potential sum for this i atom from the interaction with this j atom. */
344 velec = _mm_and_ps(velec,cutoff_mask);
345 velecsum = _mm_add_ps(velecsum,velec);
349 fscal = _mm_and_ps(fscal,cutoff_mask);
351 /* Calculate temporary vectorial force */
352 tx = _mm_mul_ps(fscal,dx10);
353 ty = _mm_mul_ps(fscal,dy10);
354 tz = _mm_mul_ps(fscal,dz10);
356 /* Update vectorial force */
357 fix1 = _mm_add_ps(fix1,tx);
358 fiy1 = _mm_add_ps(fiy1,ty);
359 fiz1 = _mm_add_ps(fiz1,tz);
361 fjx0 = _mm_add_ps(fjx0,tx);
362 fjy0 = _mm_add_ps(fjy0,ty);
363 fjz0 = _mm_add_ps(fjz0,tz);
367 /**************************
368 * CALCULATE INTERACTIONS *
369 **************************/
371 if (gmx_mm_any_lt(rsq20,rcutoff2))
374 /* Compute parameters for interactions between i and j atoms */
375 qq20 = _mm_mul_ps(iq2,jq0);
377 /* REACTION-FIELD ELECTROSTATICS */
378 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
379 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
381 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
383 /* Update potential sum for this i atom from the interaction with this j atom. */
384 velec = _mm_and_ps(velec,cutoff_mask);
385 velecsum = _mm_add_ps(velecsum,velec);
389 fscal = _mm_and_ps(fscal,cutoff_mask);
391 /* Calculate temporary vectorial force */
392 tx = _mm_mul_ps(fscal,dx20);
393 ty = _mm_mul_ps(fscal,dy20);
394 tz = _mm_mul_ps(fscal,dz20);
396 /* Update vectorial force */
397 fix2 = _mm_add_ps(fix2,tx);
398 fiy2 = _mm_add_ps(fiy2,ty);
399 fiz2 = _mm_add_ps(fiz2,tz);
401 fjx0 = _mm_add_ps(fjx0,tx);
402 fjy0 = _mm_add_ps(fjy0,ty);
403 fjz0 = _mm_add_ps(fjz0,tz);
407 /**************************
408 * CALCULATE INTERACTIONS *
409 **************************/
411 if (gmx_mm_any_lt(rsq30,rcutoff2))
414 /* Compute parameters for interactions between i and j atoms */
415 qq30 = _mm_mul_ps(iq3,jq0);
417 /* REACTION-FIELD ELECTROSTATICS */
418 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
419 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
421 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
423 /* Update potential sum for this i atom from the interaction with this j atom. */
424 velec = _mm_and_ps(velec,cutoff_mask);
425 velecsum = _mm_add_ps(velecsum,velec);
429 fscal = _mm_and_ps(fscal,cutoff_mask);
431 /* Calculate temporary vectorial force */
432 tx = _mm_mul_ps(fscal,dx30);
433 ty = _mm_mul_ps(fscal,dy30);
434 tz = _mm_mul_ps(fscal,dz30);
436 /* Update vectorial force */
437 fix3 = _mm_add_ps(fix3,tx);
438 fiy3 = _mm_add_ps(fiy3,ty);
439 fiz3 = _mm_add_ps(fiz3,tz);
441 fjx0 = _mm_add_ps(fjx0,tx);
442 fjy0 = _mm_add_ps(fjy0,ty);
443 fjz0 = _mm_add_ps(fjz0,tz);
447 fjptrA = f+j_coord_offsetA;
448 fjptrB = f+j_coord_offsetB;
449 fjptrC = f+j_coord_offsetC;
450 fjptrD = f+j_coord_offsetD;
452 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
454 /* Inner loop uses 167 flops */
460 /* Get j neighbor index, and coordinate index */
461 jnrlistA = jjnr[jidx];
462 jnrlistB = jjnr[jidx+1];
463 jnrlistC = jjnr[jidx+2];
464 jnrlistD = jjnr[jidx+3];
465 /* Sign of each element will be negative for non-real atoms.
466 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
467 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
469 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
470 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
471 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
472 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
473 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
474 j_coord_offsetA = DIM*jnrA;
475 j_coord_offsetB = DIM*jnrB;
476 j_coord_offsetC = DIM*jnrC;
477 j_coord_offsetD = DIM*jnrD;
479 /* load j atom coordinates */
480 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
481 x+j_coord_offsetC,x+j_coord_offsetD,
484 /* Calculate displacement vector */
485 dx00 = _mm_sub_ps(ix0,jx0);
486 dy00 = _mm_sub_ps(iy0,jy0);
487 dz00 = _mm_sub_ps(iz0,jz0);
488 dx10 = _mm_sub_ps(ix1,jx0);
489 dy10 = _mm_sub_ps(iy1,jy0);
490 dz10 = _mm_sub_ps(iz1,jz0);
491 dx20 = _mm_sub_ps(ix2,jx0);
492 dy20 = _mm_sub_ps(iy2,jy0);
493 dz20 = _mm_sub_ps(iz2,jz0);
494 dx30 = _mm_sub_ps(ix3,jx0);
495 dy30 = _mm_sub_ps(iy3,jy0);
496 dz30 = _mm_sub_ps(iz3,jz0);
498 /* Calculate squared distance and things based on it */
499 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
500 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
501 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
502 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
504 rinv00 = gmx_mm_invsqrt_ps(rsq00);
505 rinv10 = gmx_mm_invsqrt_ps(rsq10);
506 rinv20 = gmx_mm_invsqrt_ps(rsq20);
507 rinv30 = gmx_mm_invsqrt_ps(rsq30);
509 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
510 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
511 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
512 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
514 /* Load parameters for j particles */
515 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
516 charge+jnrC+0,charge+jnrD+0);
517 vdwjidx0A = 2*vdwtype[jnrA+0];
518 vdwjidx0B = 2*vdwtype[jnrB+0];
519 vdwjidx0C = 2*vdwtype[jnrC+0];
520 vdwjidx0D = 2*vdwtype[jnrD+0];
522 fjx0 = _mm_setzero_ps();
523 fjy0 = _mm_setzero_ps();
524 fjz0 = _mm_setzero_ps();
526 /**************************
527 * CALCULATE INTERACTIONS *
528 **************************/
530 if (gmx_mm_any_lt(rsq00,rcutoff2))
533 r00 = _mm_mul_ps(rsq00,rinv00);
534 r00 = _mm_andnot_ps(dummy_mask,r00);
536 /* Compute parameters for interactions between i and j atoms */
537 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
538 vdwparam+vdwioffset0+vdwjidx0B,
539 vdwparam+vdwioffset0+vdwjidx0C,
540 vdwparam+vdwioffset0+vdwjidx0D,
543 /* LENNARD-JONES DISPERSION/REPULSION */
545 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
546 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
547 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
548 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
549 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
551 d = _mm_sub_ps(r00,rswitch);
552 d = _mm_max_ps(d,_mm_setzero_ps());
553 d2 = _mm_mul_ps(d,d);
554 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
556 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
558 /* Evaluate switch function */
559 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
560 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
561 vvdw = _mm_mul_ps(vvdw,sw);
562 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
564 /* Update potential sum for this i atom from the interaction with this j atom. */
565 vvdw = _mm_and_ps(vvdw,cutoff_mask);
566 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
567 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
571 fscal = _mm_and_ps(fscal,cutoff_mask);
573 fscal = _mm_andnot_ps(dummy_mask,fscal);
575 /* Calculate temporary vectorial force */
576 tx = _mm_mul_ps(fscal,dx00);
577 ty = _mm_mul_ps(fscal,dy00);
578 tz = _mm_mul_ps(fscal,dz00);
580 /* Update vectorial force */
581 fix0 = _mm_add_ps(fix0,tx);
582 fiy0 = _mm_add_ps(fiy0,ty);
583 fiz0 = _mm_add_ps(fiz0,tz);
585 fjx0 = _mm_add_ps(fjx0,tx);
586 fjy0 = _mm_add_ps(fjy0,ty);
587 fjz0 = _mm_add_ps(fjz0,tz);
591 /**************************
592 * CALCULATE INTERACTIONS *
593 **************************/
595 if (gmx_mm_any_lt(rsq10,rcutoff2))
598 /* Compute parameters for interactions between i and j atoms */
599 qq10 = _mm_mul_ps(iq1,jq0);
601 /* REACTION-FIELD ELECTROSTATICS */
602 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
603 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
605 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
607 /* Update potential sum for this i atom from the interaction with this j atom. */
608 velec = _mm_and_ps(velec,cutoff_mask);
609 velec = _mm_andnot_ps(dummy_mask,velec);
610 velecsum = _mm_add_ps(velecsum,velec);
614 fscal = _mm_and_ps(fscal,cutoff_mask);
616 fscal = _mm_andnot_ps(dummy_mask,fscal);
618 /* Calculate temporary vectorial force */
619 tx = _mm_mul_ps(fscal,dx10);
620 ty = _mm_mul_ps(fscal,dy10);
621 tz = _mm_mul_ps(fscal,dz10);
623 /* Update vectorial force */
624 fix1 = _mm_add_ps(fix1,tx);
625 fiy1 = _mm_add_ps(fiy1,ty);
626 fiz1 = _mm_add_ps(fiz1,tz);
628 fjx0 = _mm_add_ps(fjx0,tx);
629 fjy0 = _mm_add_ps(fjy0,ty);
630 fjz0 = _mm_add_ps(fjz0,tz);
634 /**************************
635 * CALCULATE INTERACTIONS *
636 **************************/
638 if (gmx_mm_any_lt(rsq20,rcutoff2))
641 /* Compute parameters for interactions between i and j atoms */
642 qq20 = _mm_mul_ps(iq2,jq0);
644 /* REACTION-FIELD ELECTROSTATICS */
645 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
646 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
648 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
650 /* Update potential sum for this i atom from the interaction with this j atom. */
651 velec = _mm_and_ps(velec,cutoff_mask);
652 velec = _mm_andnot_ps(dummy_mask,velec);
653 velecsum = _mm_add_ps(velecsum,velec);
657 fscal = _mm_and_ps(fscal,cutoff_mask);
659 fscal = _mm_andnot_ps(dummy_mask,fscal);
661 /* Calculate temporary vectorial force */
662 tx = _mm_mul_ps(fscal,dx20);
663 ty = _mm_mul_ps(fscal,dy20);
664 tz = _mm_mul_ps(fscal,dz20);
666 /* Update vectorial force */
667 fix2 = _mm_add_ps(fix2,tx);
668 fiy2 = _mm_add_ps(fiy2,ty);
669 fiz2 = _mm_add_ps(fiz2,tz);
671 fjx0 = _mm_add_ps(fjx0,tx);
672 fjy0 = _mm_add_ps(fjy0,ty);
673 fjz0 = _mm_add_ps(fjz0,tz);
677 /**************************
678 * CALCULATE INTERACTIONS *
679 **************************/
681 if (gmx_mm_any_lt(rsq30,rcutoff2))
684 /* Compute parameters for interactions between i and j atoms */
685 qq30 = _mm_mul_ps(iq3,jq0);
687 /* REACTION-FIELD ELECTROSTATICS */
688 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
689 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
691 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
693 /* Update potential sum for this i atom from the interaction with this j atom. */
694 velec = _mm_and_ps(velec,cutoff_mask);
695 velec = _mm_andnot_ps(dummy_mask,velec);
696 velecsum = _mm_add_ps(velecsum,velec);
700 fscal = _mm_and_ps(fscal,cutoff_mask);
702 fscal = _mm_andnot_ps(dummy_mask,fscal);
704 /* Calculate temporary vectorial force */
705 tx = _mm_mul_ps(fscal,dx30);
706 ty = _mm_mul_ps(fscal,dy30);
707 tz = _mm_mul_ps(fscal,dz30);
709 /* Update vectorial force */
710 fix3 = _mm_add_ps(fix3,tx);
711 fiy3 = _mm_add_ps(fiy3,ty);
712 fiz3 = _mm_add_ps(fiz3,tz);
714 fjx0 = _mm_add_ps(fjx0,tx);
715 fjy0 = _mm_add_ps(fjy0,ty);
716 fjz0 = _mm_add_ps(fjz0,tz);
720 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
721 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
722 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
723 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
725 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
727 /* Inner loop uses 168 flops */
730 /* End of innermost loop */
732 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
733 f+i_coord_offset,fshift+i_shift_offset);
736 /* Update potential energies */
737 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
738 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
740 /* Increment number of inner iterations */
741 inneriter += j_index_end - j_index_start;
743 /* Outer loop uses 26 flops */
746 /* Increment number of outer iterations */
749 /* Update outer/inner flops */
751 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*168);
754 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse4_1_single
755 * Electrostatics interaction: ReactionField
756 * VdW interaction: LennardJones
757 * Geometry: Water4-Particle
758 * Calculate force/pot: Force
761 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse4_1_single
762 (t_nblist * gmx_restrict nlist,
763 rvec * gmx_restrict xx,
764 rvec * gmx_restrict ff,
765 t_forcerec * gmx_restrict fr,
766 t_mdatoms * gmx_restrict mdatoms,
767 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
768 t_nrnb * gmx_restrict nrnb)
770 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
771 * just 0 for non-waters.
772 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
773 * jnr indices corresponding to data put in the four positions in the SIMD register.
775 int i_shift_offset,i_coord_offset,outeriter,inneriter;
776 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
777 int jnrA,jnrB,jnrC,jnrD;
778 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
779 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
780 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
782 real *shiftvec,*fshift,*x,*f;
783 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
785 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
787 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
789 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
791 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
793 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
794 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
795 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
796 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
797 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
798 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
799 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
800 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
803 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
806 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
807 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
808 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
809 real rswitch_scalar,d_scalar;
810 __m128 dummy_mask,cutoff_mask;
811 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
812 __m128 one = _mm_set1_ps(1.0);
813 __m128 two = _mm_set1_ps(2.0);
819 jindex = nlist->jindex;
821 shiftidx = nlist->shift;
823 shiftvec = fr->shift_vec[0];
824 fshift = fr->fshift[0];
825 facel = _mm_set1_ps(fr->epsfac);
826 charge = mdatoms->chargeA;
827 krf = _mm_set1_ps(fr->ic->k_rf);
828 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
829 crf = _mm_set1_ps(fr->ic->c_rf);
830 nvdwtype = fr->ntype;
832 vdwtype = mdatoms->typeA;
834 /* Setup water-specific parameters */
835 inr = nlist->iinr[0];
836 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
837 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
838 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
839 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
841 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
842 rcutoff_scalar = fr->rcoulomb;
843 rcutoff = _mm_set1_ps(rcutoff_scalar);
844 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
846 rswitch_scalar = fr->rvdw_switch;
847 rswitch = _mm_set1_ps(rswitch_scalar);
848 /* Setup switch parameters */
849 d_scalar = rcutoff_scalar-rswitch_scalar;
850 d = _mm_set1_ps(d_scalar);
851 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
852 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
853 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
854 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
855 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
856 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
858 /* Avoid stupid compiler warnings */
859 jnrA = jnrB = jnrC = jnrD = 0;
868 for(iidx=0;iidx<4*DIM;iidx++)
873 /* Start outer loop over neighborlists */
874 for(iidx=0; iidx<nri; iidx++)
876 /* Load shift vector for this list */
877 i_shift_offset = DIM*shiftidx[iidx];
879 /* Load limits for loop over neighbors */
880 j_index_start = jindex[iidx];
881 j_index_end = jindex[iidx+1];
883 /* Get outer coordinate index */
885 i_coord_offset = DIM*inr;
887 /* Load i particle coords and add shift vector */
888 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
889 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
891 fix0 = _mm_setzero_ps();
892 fiy0 = _mm_setzero_ps();
893 fiz0 = _mm_setzero_ps();
894 fix1 = _mm_setzero_ps();
895 fiy1 = _mm_setzero_ps();
896 fiz1 = _mm_setzero_ps();
897 fix2 = _mm_setzero_ps();
898 fiy2 = _mm_setzero_ps();
899 fiz2 = _mm_setzero_ps();
900 fix3 = _mm_setzero_ps();
901 fiy3 = _mm_setzero_ps();
902 fiz3 = _mm_setzero_ps();
904 /* Start inner kernel loop */
905 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
908 /* Get j neighbor index, and coordinate index */
913 j_coord_offsetA = DIM*jnrA;
914 j_coord_offsetB = DIM*jnrB;
915 j_coord_offsetC = DIM*jnrC;
916 j_coord_offsetD = DIM*jnrD;
918 /* load j atom coordinates */
919 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
920 x+j_coord_offsetC,x+j_coord_offsetD,
923 /* Calculate displacement vector */
924 dx00 = _mm_sub_ps(ix0,jx0);
925 dy00 = _mm_sub_ps(iy0,jy0);
926 dz00 = _mm_sub_ps(iz0,jz0);
927 dx10 = _mm_sub_ps(ix1,jx0);
928 dy10 = _mm_sub_ps(iy1,jy0);
929 dz10 = _mm_sub_ps(iz1,jz0);
930 dx20 = _mm_sub_ps(ix2,jx0);
931 dy20 = _mm_sub_ps(iy2,jy0);
932 dz20 = _mm_sub_ps(iz2,jz0);
933 dx30 = _mm_sub_ps(ix3,jx0);
934 dy30 = _mm_sub_ps(iy3,jy0);
935 dz30 = _mm_sub_ps(iz3,jz0);
937 /* Calculate squared distance and things based on it */
938 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
939 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
940 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
941 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
943 rinv00 = gmx_mm_invsqrt_ps(rsq00);
944 rinv10 = gmx_mm_invsqrt_ps(rsq10);
945 rinv20 = gmx_mm_invsqrt_ps(rsq20);
946 rinv30 = gmx_mm_invsqrt_ps(rsq30);
948 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
949 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
950 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
951 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
953 /* Load parameters for j particles */
954 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
955 charge+jnrC+0,charge+jnrD+0);
956 vdwjidx0A = 2*vdwtype[jnrA+0];
957 vdwjidx0B = 2*vdwtype[jnrB+0];
958 vdwjidx0C = 2*vdwtype[jnrC+0];
959 vdwjidx0D = 2*vdwtype[jnrD+0];
961 fjx0 = _mm_setzero_ps();
962 fjy0 = _mm_setzero_ps();
963 fjz0 = _mm_setzero_ps();
965 /**************************
966 * CALCULATE INTERACTIONS *
967 **************************/
969 if (gmx_mm_any_lt(rsq00,rcutoff2))
972 r00 = _mm_mul_ps(rsq00,rinv00);
974 /* Compute parameters for interactions between i and j atoms */
975 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
976 vdwparam+vdwioffset0+vdwjidx0B,
977 vdwparam+vdwioffset0+vdwjidx0C,
978 vdwparam+vdwioffset0+vdwjidx0D,
981 /* LENNARD-JONES DISPERSION/REPULSION */
983 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
984 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
985 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
986 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
987 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
989 d = _mm_sub_ps(r00,rswitch);
990 d = _mm_max_ps(d,_mm_setzero_ps());
991 d2 = _mm_mul_ps(d,d);
992 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
994 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
996 /* Evaluate switch function */
997 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
998 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
999 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1003 fscal = _mm_and_ps(fscal,cutoff_mask);
1005 /* Calculate temporary vectorial force */
1006 tx = _mm_mul_ps(fscal,dx00);
1007 ty = _mm_mul_ps(fscal,dy00);
1008 tz = _mm_mul_ps(fscal,dz00);
1010 /* Update vectorial force */
1011 fix0 = _mm_add_ps(fix0,tx);
1012 fiy0 = _mm_add_ps(fiy0,ty);
1013 fiz0 = _mm_add_ps(fiz0,tz);
1015 fjx0 = _mm_add_ps(fjx0,tx);
1016 fjy0 = _mm_add_ps(fjy0,ty);
1017 fjz0 = _mm_add_ps(fjz0,tz);
1021 /**************************
1022 * CALCULATE INTERACTIONS *
1023 **************************/
1025 if (gmx_mm_any_lt(rsq10,rcutoff2))
1028 /* Compute parameters for interactions between i and j atoms */
1029 qq10 = _mm_mul_ps(iq1,jq0);
1031 /* REACTION-FIELD ELECTROSTATICS */
1032 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1034 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1038 fscal = _mm_and_ps(fscal,cutoff_mask);
1040 /* Calculate temporary vectorial force */
1041 tx = _mm_mul_ps(fscal,dx10);
1042 ty = _mm_mul_ps(fscal,dy10);
1043 tz = _mm_mul_ps(fscal,dz10);
1045 /* Update vectorial force */
1046 fix1 = _mm_add_ps(fix1,tx);
1047 fiy1 = _mm_add_ps(fiy1,ty);
1048 fiz1 = _mm_add_ps(fiz1,tz);
1050 fjx0 = _mm_add_ps(fjx0,tx);
1051 fjy0 = _mm_add_ps(fjy0,ty);
1052 fjz0 = _mm_add_ps(fjz0,tz);
1056 /**************************
1057 * CALCULATE INTERACTIONS *
1058 **************************/
1060 if (gmx_mm_any_lt(rsq20,rcutoff2))
1063 /* Compute parameters for interactions between i and j atoms */
1064 qq20 = _mm_mul_ps(iq2,jq0);
1066 /* REACTION-FIELD ELECTROSTATICS */
1067 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1069 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1073 fscal = _mm_and_ps(fscal,cutoff_mask);
1075 /* Calculate temporary vectorial force */
1076 tx = _mm_mul_ps(fscal,dx20);
1077 ty = _mm_mul_ps(fscal,dy20);
1078 tz = _mm_mul_ps(fscal,dz20);
1080 /* Update vectorial force */
1081 fix2 = _mm_add_ps(fix2,tx);
1082 fiy2 = _mm_add_ps(fiy2,ty);
1083 fiz2 = _mm_add_ps(fiz2,tz);
1085 fjx0 = _mm_add_ps(fjx0,tx);
1086 fjy0 = _mm_add_ps(fjy0,ty);
1087 fjz0 = _mm_add_ps(fjz0,tz);
1091 /**************************
1092 * CALCULATE INTERACTIONS *
1093 **************************/
1095 if (gmx_mm_any_lt(rsq30,rcutoff2))
1098 /* Compute parameters for interactions between i and j atoms */
1099 qq30 = _mm_mul_ps(iq3,jq0);
1101 /* REACTION-FIELD ELECTROSTATICS */
1102 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1104 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1108 fscal = _mm_and_ps(fscal,cutoff_mask);
1110 /* Calculate temporary vectorial force */
1111 tx = _mm_mul_ps(fscal,dx30);
1112 ty = _mm_mul_ps(fscal,dy30);
1113 tz = _mm_mul_ps(fscal,dz30);
1115 /* Update vectorial force */
1116 fix3 = _mm_add_ps(fix3,tx);
1117 fiy3 = _mm_add_ps(fiy3,ty);
1118 fiz3 = _mm_add_ps(fiz3,tz);
1120 fjx0 = _mm_add_ps(fjx0,tx);
1121 fjy0 = _mm_add_ps(fjy0,ty);
1122 fjz0 = _mm_add_ps(fjz0,tz);
1126 fjptrA = f+j_coord_offsetA;
1127 fjptrB = f+j_coord_offsetB;
1128 fjptrC = f+j_coord_offsetC;
1129 fjptrD = f+j_coord_offsetD;
1131 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1133 /* Inner loop uses 146 flops */
1136 if(jidx<j_index_end)
1139 /* Get j neighbor index, and coordinate index */
1140 jnrlistA = jjnr[jidx];
1141 jnrlistB = jjnr[jidx+1];
1142 jnrlistC = jjnr[jidx+2];
1143 jnrlistD = jjnr[jidx+3];
1144 /* Sign of each element will be negative for non-real atoms.
1145 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1146 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1148 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1149 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1150 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1151 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1152 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1153 j_coord_offsetA = DIM*jnrA;
1154 j_coord_offsetB = DIM*jnrB;
1155 j_coord_offsetC = DIM*jnrC;
1156 j_coord_offsetD = DIM*jnrD;
1158 /* load j atom coordinates */
1159 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1160 x+j_coord_offsetC,x+j_coord_offsetD,
1163 /* Calculate displacement vector */
1164 dx00 = _mm_sub_ps(ix0,jx0);
1165 dy00 = _mm_sub_ps(iy0,jy0);
1166 dz00 = _mm_sub_ps(iz0,jz0);
1167 dx10 = _mm_sub_ps(ix1,jx0);
1168 dy10 = _mm_sub_ps(iy1,jy0);
1169 dz10 = _mm_sub_ps(iz1,jz0);
1170 dx20 = _mm_sub_ps(ix2,jx0);
1171 dy20 = _mm_sub_ps(iy2,jy0);
1172 dz20 = _mm_sub_ps(iz2,jz0);
1173 dx30 = _mm_sub_ps(ix3,jx0);
1174 dy30 = _mm_sub_ps(iy3,jy0);
1175 dz30 = _mm_sub_ps(iz3,jz0);
1177 /* Calculate squared distance and things based on it */
1178 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1179 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1180 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1181 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1183 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1184 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1185 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1186 rinv30 = gmx_mm_invsqrt_ps(rsq30);
1188 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1189 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1190 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1191 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
1193 /* Load parameters for j particles */
1194 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1195 charge+jnrC+0,charge+jnrD+0);
1196 vdwjidx0A = 2*vdwtype[jnrA+0];
1197 vdwjidx0B = 2*vdwtype[jnrB+0];
1198 vdwjidx0C = 2*vdwtype[jnrC+0];
1199 vdwjidx0D = 2*vdwtype[jnrD+0];
1201 fjx0 = _mm_setzero_ps();
1202 fjy0 = _mm_setzero_ps();
1203 fjz0 = _mm_setzero_ps();
1205 /**************************
1206 * CALCULATE INTERACTIONS *
1207 **************************/
1209 if (gmx_mm_any_lt(rsq00,rcutoff2))
1212 r00 = _mm_mul_ps(rsq00,rinv00);
1213 r00 = _mm_andnot_ps(dummy_mask,r00);
1215 /* Compute parameters for interactions between i and j atoms */
1216 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1217 vdwparam+vdwioffset0+vdwjidx0B,
1218 vdwparam+vdwioffset0+vdwjidx0C,
1219 vdwparam+vdwioffset0+vdwjidx0D,
1222 /* LENNARD-JONES DISPERSION/REPULSION */
1224 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1225 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
1226 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1227 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1228 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1230 d = _mm_sub_ps(r00,rswitch);
1231 d = _mm_max_ps(d,_mm_setzero_ps());
1232 d2 = _mm_mul_ps(d,d);
1233 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1235 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1237 /* Evaluate switch function */
1238 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1239 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1240 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1244 fscal = _mm_and_ps(fscal,cutoff_mask);
1246 fscal = _mm_andnot_ps(dummy_mask,fscal);
1248 /* Calculate temporary vectorial force */
1249 tx = _mm_mul_ps(fscal,dx00);
1250 ty = _mm_mul_ps(fscal,dy00);
1251 tz = _mm_mul_ps(fscal,dz00);
1253 /* Update vectorial force */
1254 fix0 = _mm_add_ps(fix0,tx);
1255 fiy0 = _mm_add_ps(fiy0,ty);
1256 fiz0 = _mm_add_ps(fiz0,tz);
1258 fjx0 = _mm_add_ps(fjx0,tx);
1259 fjy0 = _mm_add_ps(fjy0,ty);
1260 fjz0 = _mm_add_ps(fjz0,tz);
1264 /**************************
1265 * CALCULATE INTERACTIONS *
1266 **************************/
1268 if (gmx_mm_any_lt(rsq10,rcutoff2))
1271 /* Compute parameters for interactions between i and j atoms */
1272 qq10 = _mm_mul_ps(iq1,jq0);
1274 /* REACTION-FIELD ELECTROSTATICS */
1275 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1277 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1281 fscal = _mm_and_ps(fscal,cutoff_mask);
1283 fscal = _mm_andnot_ps(dummy_mask,fscal);
1285 /* Calculate temporary vectorial force */
1286 tx = _mm_mul_ps(fscal,dx10);
1287 ty = _mm_mul_ps(fscal,dy10);
1288 tz = _mm_mul_ps(fscal,dz10);
1290 /* Update vectorial force */
1291 fix1 = _mm_add_ps(fix1,tx);
1292 fiy1 = _mm_add_ps(fiy1,ty);
1293 fiz1 = _mm_add_ps(fiz1,tz);
1295 fjx0 = _mm_add_ps(fjx0,tx);
1296 fjy0 = _mm_add_ps(fjy0,ty);
1297 fjz0 = _mm_add_ps(fjz0,tz);
1301 /**************************
1302 * CALCULATE INTERACTIONS *
1303 **************************/
1305 if (gmx_mm_any_lt(rsq20,rcutoff2))
1308 /* Compute parameters for interactions between i and j atoms */
1309 qq20 = _mm_mul_ps(iq2,jq0);
1311 /* REACTION-FIELD ELECTROSTATICS */
1312 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1314 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1318 fscal = _mm_and_ps(fscal,cutoff_mask);
1320 fscal = _mm_andnot_ps(dummy_mask,fscal);
1322 /* Calculate temporary vectorial force */
1323 tx = _mm_mul_ps(fscal,dx20);
1324 ty = _mm_mul_ps(fscal,dy20);
1325 tz = _mm_mul_ps(fscal,dz20);
1327 /* Update vectorial force */
1328 fix2 = _mm_add_ps(fix2,tx);
1329 fiy2 = _mm_add_ps(fiy2,ty);
1330 fiz2 = _mm_add_ps(fiz2,tz);
1332 fjx0 = _mm_add_ps(fjx0,tx);
1333 fjy0 = _mm_add_ps(fjy0,ty);
1334 fjz0 = _mm_add_ps(fjz0,tz);
1338 /**************************
1339 * CALCULATE INTERACTIONS *
1340 **************************/
1342 if (gmx_mm_any_lt(rsq30,rcutoff2))
1345 /* Compute parameters for interactions between i and j atoms */
1346 qq30 = _mm_mul_ps(iq3,jq0);
1348 /* REACTION-FIELD ELECTROSTATICS */
1349 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1351 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1355 fscal = _mm_and_ps(fscal,cutoff_mask);
1357 fscal = _mm_andnot_ps(dummy_mask,fscal);
1359 /* Calculate temporary vectorial force */
1360 tx = _mm_mul_ps(fscal,dx30);
1361 ty = _mm_mul_ps(fscal,dy30);
1362 tz = _mm_mul_ps(fscal,dz30);
1364 /* Update vectorial force */
1365 fix3 = _mm_add_ps(fix3,tx);
1366 fiy3 = _mm_add_ps(fiy3,ty);
1367 fiz3 = _mm_add_ps(fiz3,tz);
1369 fjx0 = _mm_add_ps(fjx0,tx);
1370 fjy0 = _mm_add_ps(fjy0,ty);
1371 fjz0 = _mm_add_ps(fjz0,tz);
1375 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1376 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1377 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1378 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1380 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1382 /* Inner loop uses 147 flops */
1385 /* End of innermost loop */
1387 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1388 f+i_coord_offset,fshift+i_shift_offset);
1390 /* Increment number of inner iterations */
1391 inneriter += j_index_end - j_index_start;
1393 /* Outer loop uses 24 flops */
1396 /* Increment number of outer iterations */
1399 /* Update outer/inner flops */
1401 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);