2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/gmxlib/nrnb.h"
47 #include "kernelutil_x86_sse4_1_double.h"
50 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse4_1_double
51 * Electrostatics interaction: Ewald
52 * VdW interaction: LennardJones
53 * Geometry: Water4-Particle
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse4_1_double
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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
74 int j_coord_offsetA,j_coord_offsetB;
75 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
77 real *shiftvec,*fshift,*x,*f;
78 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
80 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
82 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
84 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
86 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
87 int vdwjidx0A,vdwjidx0B;
88 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
89 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
90 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
91 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
92 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
93 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
96 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
99 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
100 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
102 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
104 __m128d dummy_mask,cutoff_mask;
105 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
106 __m128d one = _mm_set1_pd(1.0);
107 __m128d two = _mm_set1_pd(2.0);
113 jindex = nlist->jindex;
115 shiftidx = nlist->shift;
117 shiftvec = fr->shift_vec[0];
118 fshift = fr->fshift[0];
119 facel = _mm_set1_pd(fr->ic->epsfac);
120 charge = mdatoms->chargeA;
121 nvdwtype = fr->ntype;
123 vdwtype = mdatoms->typeA;
125 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
126 ewtab = fr->ic->tabq_coul_FDV0;
127 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
128 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
130 /* Setup water-specific parameters */
131 inr = nlist->iinr[0];
132 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
133 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
134 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
135 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
137 /* Avoid stupid compiler warnings */
145 /* Start outer loop over neighborlists */
146 for(iidx=0; iidx<nri; iidx++)
148 /* Load shift vector for this list */
149 i_shift_offset = DIM*shiftidx[iidx];
151 /* Load limits for loop over neighbors */
152 j_index_start = jindex[iidx];
153 j_index_end = jindex[iidx+1];
155 /* Get outer coordinate index */
157 i_coord_offset = DIM*inr;
159 /* Load i particle coords and add shift vector */
160 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
161 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
163 fix0 = _mm_setzero_pd();
164 fiy0 = _mm_setzero_pd();
165 fiz0 = _mm_setzero_pd();
166 fix1 = _mm_setzero_pd();
167 fiy1 = _mm_setzero_pd();
168 fiz1 = _mm_setzero_pd();
169 fix2 = _mm_setzero_pd();
170 fiy2 = _mm_setzero_pd();
171 fiz2 = _mm_setzero_pd();
172 fix3 = _mm_setzero_pd();
173 fiy3 = _mm_setzero_pd();
174 fiz3 = _mm_setzero_pd();
176 /* Reset potential sums */
177 velecsum = _mm_setzero_pd();
178 vvdwsum = _mm_setzero_pd();
180 /* Start inner kernel loop */
181 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
184 /* Get j neighbor index, and coordinate index */
187 j_coord_offsetA = DIM*jnrA;
188 j_coord_offsetB = DIM*jnrB;
190 /* load j atom coordinates */
191 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
194 /* Calculate displacement vector */
195 dx00 = _mm_sub_pd(ix0,jx0);
196 dy00 = _mm_sub_pd(iy0,jy0);
197 dz00 = _mm_sub_pd(iz0,jz0);
198 dx10 = _mm_sub_pd(ix1,jx0);
199 dy10 = _mm_sub_pd(iy1,jy0);
200 dz10 = _mm_sub_pd(iz1,jz0);
201 dx20 = _mm_sub_pd(ix2,jx0);
202 dy20 = _mm_sub_pd(iy2,jy0);
203 dz20 = _mm_sub_pd(iz2,jz0);
204 dx30 = _mm_sub_pd(ix3,jx0);
205 dy30 = _mm_sub_pd(iy3,jy0);
206 dz30 = _mm_sub_pd(iz3,jz0);
208 /* Calculate squared distance and things based on it */
209 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
210 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
211 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
212 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
214 rinv10 = sse41_invsqrt_d(rsq10);
215 rinv20 = sse41_invsqrt_d(rsq20);
216 rinv30 = sse41_invsqrt_d(rsq30);
218 rinvsq00 = sse41_inv_d(rsq00);
219 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
220 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
221 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
223 /* Load parameters for j particles */
224 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
225 vdwjidx0A = 2*vdwtype[jnrA+0];
226 vdwjidx0B = 2*vdwtype[jnrB+0];
228 fjx0 = _mm_setzero_pd();
229 fjy0 = _mm_setzero_pd();
230 fjz0 = _mm_setzero_pd();
232 /**************************
233 * CALCULATE INTERACTIONS *
234 **************************/
236 /* Compute parameters for interactions between i and j atoms */
237 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
238 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
240 /* LENNARD-JONES DISPERSION/REPULSION */
242 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
243 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
244 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
245 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
246 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
248 /* Update potential sum for this i atom from the interaction with this j atom. */
249 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
253 /* Calculate temporary vectorial force */
254 tx = _mm_mul_pd(fscal,dx00);
255 ty = _mm_mul_pd(fscal,dy00);
256 tz = _mm_mul_pd(fscal,dz00);
258 /* Update vectorial force */
259 fix0 = _mm_add_pd(fix0,tx);
260 fiy0 = _mm_add_pd(fiy0,ty);
261 fiz0 = _mm_add_pd(fiz0,tz);
263 fjx0 = _mm_add_pd(fjx0,tx);
264 fjy0 = _mm_add_pd(fjy0,ty);
265 fjz0 = _mm_add_pd(fjz0,tz);
267 /**************************
268 * CALCULATE INTERACTIONS *
269 **************************/
271 r10 = _mm_mul_pd(rsq10,rinv10);
273 /* Compute parameters for interactions between i and j atoms */
274 qq10 = _mm_mul_pd(iq1,jq0);
276 /* EWALD ELECTROSTATICS */
278 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
279 ewrt = _mm_mul_pd(r10,ewtabscale);
280 ewitab = _mm_cvttpd_epi32(ewrt);
281 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
282 ewitab = _mm_slli_epi32(ewitab,2);
283 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
284 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
285 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
286 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
287 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
288 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
289 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
290 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
291 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
292 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
294 /* Update potential sum for this i atom from the interaction with this j atom. */
295 velecsum = _mm_add_pd(velecsum,velec);
299 /* Calculate temporary vectorial force */
300 tx = _mm_mul_pd(fscal,dx10);
301 ty = _mm_mul_pd(fscal,dy10);
302 tz = _mm_mul_pd(fscal,dz10);
304 /* Update vectorial force */
305 fix1 = _mm_add_pd(fix1,tx);
306 fiy1 = _mm_add_pd(fiy1,ty);
307 fiz1 = _mm_add_pd(fiz1,tz);
309 fjx0 = _mm_add_pd(fjx0,tx);
310 fjy0 = _mm_add_pd(fjy0,ty);
311 fjz0 = _mm_add_pd(fjz0,tz);
313 /**************************
314 * CALCULATE INTERACTIONS *
315 **************************/
317 r20 = _mm_mul_pd(rsq20,rinv20);
319 /* Compute parameters for interactions between i and j atoms */
320 qq20 = _mm_mul_pd(iq2,jq0);
322 /* EWALD ELECTROSTATICS */
324 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
325 ewrt = _mm_mul_pd(r20,ewtabscale);
326 ewitab = _mm_cvttpd_epi32(ewrt);
327 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
328 ewitab = _mm_slli_epi32(ewitab,2);
329 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
330 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
331 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
332 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
333 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
334 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
335 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
336 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
337 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
338 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
340 /* Update potential sum for this i atom from the interaction with this j atom. */
341 velecsum = _mm_add_pd(velecsum,velec);
345 /* Calculate temporary vectorial force */
346 tx = _mm_mul_pd(fscal,dx20);
347 ty = _mm_mul_pd(fscal,dy20);
348 tz = _mm_mul_pd(fscal,dz20);
350 /* Update vectorial force */
351 fix2 = _mm_add_pd(fix2,tx);
352 fiy2 = _mm_add_pd(fiy2,ty);
353 fiz2 = _mm_add_pd(fiz2,tz);
355 fjx0 = _mm_add_pd(fjx0,tx);
356 fjy0 = _mm_add_pd(fjy0,ty);
357 fjz0 = _mm_add_pd(fjz0,tz);
359 /**************************
360 * CALCULATE INTERACTIONS *
361 **************************/
363 r30 = _mm_mul_pd(rsq30,rinv30);
365 /* Compute parameters for interactions between i and j atoms */
366 qq30 = _mm_mul_pd(iq3,jq0);
368 /* EWALD ELECTROSTATICS */
370 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
371 ewrt = _mm_mul_pd(r30,ewtabscale);
372 ewitab = _mm_cvttpd_epi32(ewrt);
373 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
374 ewitab = _mm_slli_epi32(ewitab,2);
375 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
376 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
377 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
378 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
379 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
380 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
381 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
382 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
383 velec = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
384 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
386 /* Update potential sum for this i atom from the interaction with this j atom. */
387 velecsum = _mm_add_pd(velecsum,velec);
391 /* Calculate temporary vectorial force */
392 tx = _mm_mul_pd(fscal,dx30);
393 ty = _mm_mul_pd(fscal,dy30);
394 tz = _mm_mul_pd(fscal,dz30);
396 /* Update vectorial force */
397 fix3 = _mm_add_pd(fix3,tx);
398 fiy3 = _mm_add_pd(fiy3,ty);
399 fiz3 = _mm_add_pd(fiz3,tz);
401 fjx0 = _mm_add_pd(fjx0,tx);
402 fjy0 = _mm_add_pd(fjy0,ty);
403 fjz0 = _mm_add_pd(fjz0,tz);
405 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
407 /* Inner loop uses 158 flops */
414 j_coord_offsetA = DIM*jnrA;
416 /* load j atom coordinates */
417 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
420 /* Calculate displacement vector */
421 dx00 = _mm_sub_pd(ix0,jx0);
422 dy00 = _mm_sub_pd(iy0,jy0);
423 dz00 = _mm_sub_pd(iz0,jz0);
424 dx10 = _mm_sub_pd(ix1,jx0);
425 dy10 = _mm_sub_pd(iy1,jy0);
426 dz10 = _mm_sub_pd(iz1,jz0);
427 dx20 = _mm_sub_pd(ix2,jx0);
428 dy20 = _mm_sub_pd(iy2,jy0);
429 dz20 = _mm_sub_pd(iz2,jz0);
430 dx30 = _mm_sub_pd(ix3,jx0);
431 dy30 = _mm_sub_pd(iy3,jy0);
432 dz30 = _mm_sub_pd(iz3,jz0);
434 /* Calculate squared distance and things based on it */
435 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
436 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
437 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
438 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
440 rinv10 = sse41_invsqrt_d(rsq10);
441 rinv20 = sse41_invsqrt_d(rsq20);
442 rinv30 = sse41_invsqrt_d(rsq30);
444 rinvsq00 = sse41_inv_d(rsq00);
445 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
446 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
447 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
449 /* Load parameters for j particles */
450 jq0 = _mm_load_sd(charge+jnrA+0);
451 vdwjidx0A = 2*vdwtype[jnrA+0];
453 fjx0 = _mm_setzero_pd();
454 fjy0 = _mm_setzero_pd();
455 fjz0 = _mm_setzero_pd();
457 /**************************
458 * CALCULATE INTERACTIONS *
459 **************************/
461 /* Compute parameters for interactions between i and j atoms */
462 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
464 /* LENNARD-JONES DISPERSION/REPULSION */
466 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
467 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
468 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
469 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
470 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
472 /* Update potential sum for this i atom from the interaction with this j atom. */
473 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
474 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
478 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
480 /* Calculate temporary vectorial force */
481 tx = _mm_mul_pd(fscal,dx00);
482 ty = _mm_mul_pd(fscal,dy00);
483 tz = _mm_mul_pd(fscal,dz00);
485 /* Update vectorial force */
486 fix0 = _mm_add_pd(fix0,tx);
487 fiy0 = _mm_add_pd(fiy0,ty);
488 fiz0 = _mm_add_pd(fiz0,tz);
490 fjx0 = _mm_add_pd(fjx0,tx);
491 fjy0 = _mm_add_pd(fjy0,ty);
492 fjz0 = _mm_add_pd(fjz0,tz);
494 /**************************
495 * CALCULATE INTERACTIONS *
496 **************************/
498 r10 = _mm_mul_pd(rsq10,rinv10);
500 /* Compute parameters for interactions between i and j atoms */
501 qq10 = _mm_mul_pd(iq1,jq0);
503 /* EWALD ELECTROSTATICS */
505 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
506 ewrt = _mm_mul_pd(r10,ewtabscale);
507 ewitab = _mm_cvttpd_epi32(ewrt);
508 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
509 ewitab = _mm_slli_epi32(ewitab,2);
510 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
511 ewtabD = _mm_setzero_pd();
512 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
513 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
514 ewtabFn = _mm_setzero_pd();
515 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
516 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
517 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
518 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
519 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
521 /* Update potential sum for this i atom from the interaction with this j atom. */
522 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
523 velecsum = _mm_add_pd(velecsum,velec);
527 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
529 /* Calculate temporary vectorial force */
530 tx = _mm_mul_pd(fscal,dx10);
531 ty = _mm_mul_pd(fscal,dy10);
532 tz = _mm_mul_pd(fscal,dz10);
534 /* Update vectorial force */
535 fix1 = _mm_add_pd(fix1,tx);
536 fiy1 = _mm_add_pd(fiy1,ty);
537 fiz1 = _mm_add_pd(fiz1,tz);
539 fjx0 = _mm_add_pd(fjx0,tx);
540 fjy0 = _mm_add_pd(fjy0,ty);
541 fjz0 = _mm_add_pd(fjz0,tz);
543 /**************************
544 * CALCULATE INTERACTIONS *
545 **************************/
547 r20 = _mm_mul_pd(rsq20,rinv20);
549 /* Compute parameters for interactions between i and j atoms */
550 qq20 = _mm_mul_pd(iq2,jq0);
552 /* EWALD ELECTROSTATICS */
554 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
555 ewrt = _mm_mul_pd(r20,ewtabscale);
556 ewitab = _mm_cvttpd_epi32(ewrt);
557 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
558 ewitab = _mm_slli_epi32(ewitab,2);
559 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
560 ewtabD = _mm_setzero_pd();
561 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
562 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
563 ewtabFn = _mm_setzero_pd();
564 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
565 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
566 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
567 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
568 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
570 /* Update potential sum for this i atom from the interaction with this j atom. */
571 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
572 velecsum = _mm_add_pd(velecsum,velec);
576 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
578 /* Calculate temporary vectorial force */
579 tx = _mm_mul_pd(fscal,dx20);
580 ty = _mm_mul_pd(fscal,dy20);
581 tz = _mm_mul_pd(fscal,dz20);
583 /* Update vectorial force */
584 fix2 = _mm_add_pd(fix2,tx);
585 fiy2 = _mm_add_pd(fiy2,ty);
586 fiz2 = _mm_add_pd(fiz2,tz);
588 fjx0 = _mm_add_pd(fjx0,tx);
589 fjy0 = _mm_add_pd(fjy0,ty);
590 fjz0 = _mm_add_pd(fjz0,tz);
592 /**************************
593 * CALCULATE INTERACTIONS *
594 **************************/
596 r30 = _mm_mul_pd(rsq30,rinv30);
598 /* Compute parameters for interactions between i and j atoms */
599 qq30 = _mm_mul_pd(iq3,jq0);
601 /* EWALD ELECTROSTATICS */
603 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
604 ewrt = _mm_mul_pd(r30,ewtabscale);
605 ewitab = _mm_cvttpd_epi32(ewrt);
606 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
607 ewitab = _mm_slli_epi32(ewitab,2);
608 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
609 ewtabD = _mm_setzero_pd();
610 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
611 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
612 ewtabFn = _mm_setzero_pd();
613 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
614 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
615 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
616 velec = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
617 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
619 /* Update potential sum for this i atom from the interaction with this j atom. */
620 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
621 velecsum = _mm_add_pd(velecsum,velec);
625 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
627 /* Calculate temporary vectorial force */
628 tx = _mm_mul_pd(fscal,dx30);
629 ty = _mm_mul_pd(fscal,dy30);
630 tz = _mm_mul_pd(fscal,dz30);
632 /* Update vectorial force */
633 fix3 = _mm_add_pd(fix3,tx);
634 fiy3 = _mm_add_pd(fiy3,ty);
635 fiz3 = _mm_add_pd(fiz3,tz);
637 fjx0 = _mm_add_pd(fjx0,tx);
638 fjy0 = _mm_add_pd(fjy0,ty);
639 fjz0 = _mm_add_pd(fjz0,tz);
641 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
643 /* Inner loop uses 158 flops */
646 /* End of innermost loop */
648 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
649 f+i_coord_offset,fshift+i_shift_offset);
652 /* Update potential energies */
653 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
654 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
656 /* Increment number of inner iterations */
657 inneriter += j_index_end - j_index_start;
659 /* Outer loop uses 26 flops */
662 /* Increment number of outer iterations */
665 /* Update outer/inner flops */
667 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*158);
670 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse4_1_double
671 * Electrostatics interaction: Ewald
672 * VdW interaction: LennardJones
673 * Geometry: Water4-Particle
674 * Calculate force/pot: Force
677 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse4_1_double
678 (t_nblist * gmx_restrict nlist,
679 rvec * gmx_restrict xx,
680 rvec * gmx_restrict ff,
681 struct t_forcerec * gmx_restrict fr,
682 t_mdatoms * gmx_restrict mdatoms,
683 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
684 t_nrnb * gmx_restrict nrnb)
686 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
687 * just 0 for non-waters.
688 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
689 * jnr indices corresponding to data put in the four positions in the SIMD register.
691 int i_shift_offset,i_coord_offset,outeriter,inneriter;
692 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
694 int j_coord_offsetA,j_coord_offsetB;
695 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
697 real *shiftvec,*fshift,*x,*f;
698 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
700 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
702 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
704 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
706 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
707 int vdwjidx0A,vdwjidx0B;
708 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
709 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
710 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
711 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
712 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
713 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
716 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
719 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
720 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
722 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
724 __m128d dummy_mask,cutoff_mask;
725 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
726 __m128d one = _mm_set1_pd(1.0);
727 __m128d two = _mm_set1_pd(2.0);
733 jindex = nlist->jindex;
735 shiftidx = nlist->shift;
737 shiftvec = fr->shift_vec[0];
738 fshift = fr->fshift[0];
739 facel = _mm_set1_pd(fr->ic->epsfac);
740 charge = mdatoms->chargeA;
741 nvdwtype = fr->ntype;
743 vdwtype = mdatoms->typeA;
745 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
746 ewtab = fr->ic->tabq_coul_F;
747 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
748 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
750 /* Setup water-specific parameters */
751 inr = nlist->iinr[0];
752 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
753 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
754 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
755 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
757 /* Avoid stupid compiler warnings */
765 /* Start outer loop over neighborlists */
766 for(iidx=0; iidx<nri; iidx++)
768 /* Load shift vector for this list */
769 i_shift_offset = DIM*shiftidx[iidx];
771 /* Load limits for loop over neighbors */
772 j_index_start = jindex[iidx];
773 j_index_end = jindex[iidx+1];
775 /* Get outer coordinate index */
777 i_coord_offset = DIM*inr;
779 /* Load i particle coords and add shift vector */
780 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
781 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
783 fix0 = _mm_setzero_pd();
784 fiy0 = _mm_setzero_pd();
785 fiz0 = _mm_setzero_pd();
786 fix1 = _mm_setzero_pd();
787 fiy1 = _mm_setzero_pd();
788 fiz1 = _mm_setzero_pd();
789 fix2 = _mm_setzero_pd();
790 fiy2 = _mm_setzero_pd();
791 fiz2 = _mm_setzero_pd();
792 fix3 = _mm_setzero_pd();
793 fiy3 = _mm_setzero_pd();
794 fiz3 = _mm_setzero_pd();
796 /* Start inner kernel loop */
797 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
800 /* Get j neighbor index, and coordinate index */
803 j_coord_offsetA = DIM*jnrA;
804 j_coord_offsetB = DIM*jnrB;
806 /* load j atom coordinates */
807 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
810 /* Calculate displacement vector */
811 dx00 = _mm_sub_pd(ix0,jx0);
812 dy00 = _mm_sub_pd(iy0,jy0);
813 dz00 = _mm_sub_pd(iz0,jz0);
814 dx10 = _mm_sub_pd(ix1,jx0);
815 dy10 = _mm_sub_pd(iy1,jy0);
816 dz10 = _mm_sub_pd(iz1,jz0);
817 dx20 = _mm_sub_pd(ix2,jx0);
818 dy20 = _mm_sub_pd(iy2,jy0);
819 dz20 = _mm_sub_pd(iz2,jz0);
820 dx30 = _mm_sub_pd(ix3,jx0);
821 dy30 = _mm_sub_pd(iy3,jy0);
822 dz30 = _mm_sub_pd(iz3,jz0);
824 /* Calculate squared distance and things based on it */
825 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
826 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
827 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
828 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
830 rinv10 = sse41_invsqrt_d(rsq10);
831 rinv20 = sse41_invsqrt_d(rsq20);
832 rinv30 = sse41_invsqrt_d(rsq30);
834 rinvsq00 = sse41_inv_d(rsq00);
835 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
836 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
837 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
839 /* Load parameters for j particles */
840 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
841 vdwjidx0A = 2*vdwtype[jnrA+0];
842 vdwjidx0B = 2*vdwtype[jnrB+0];
844 fjx0 = _mm_setzero_pd();
845 fjy0 = _mm_setzero_pd();
846 fjz0 = _mm_setzero_pd();
848 /**************************
849 * CALCULATE INTERACTIONS *
850 **************************/
852 /* Compute parameters for interactions between i and j atoms */
853 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
854 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
856 /* LENNARD-JONES DISPERSION/REPULSION */
858 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
859 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
863 /* Calculate temporary vectorial force */
864 tx = _mm_mul_pd(fscal,dx00);
865 ty = _mm_mul_pd(fscal,dy00);
866 tz = _mm_mul_pd(fscal,dz00);
868 /* Update vectorial force */
869 fix0 = _mm_add_pd(fix0,tx);
870 fiy0 = _mm_add_pd(fiy0,ty);
871 fiz0 = _mm_add_pd(fiz0,tz);
873 fjx0 = _mm_add_pd(fjx0,tx);
874 fjy0 = _mm_add_pd(fjy0,ty);
875 fjz0 = _mm_add_pd(fjz0,tz);
877 /**************************
878 * CALCULATE INTERACTIONS *
879 **************************/
881 r10 = _mm_mul_pd(rsq10,rinv10);
883 /* Compute parameters for interactions between i and j atoms */
884 qq10 = _mm_mul_pd(iq1,jq0);
886 /* EWALD ELECTROSTATICS */
888 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
889 ewrt = _mm_mul_pd(r10,ewtabscale);
890 ewitab = _mm_cvttpd_epi32(ewrt);
891 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
892 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
894 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
895 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
899 /* Calculate temporary vectorial force */
900 tx = _mm_mul_pd(fscal,dx10);
901 ty = _mm_mul_pd(fscal,dy10);
902 tz = _mm_mul_pd(fscal,dz10);
904 /* Update vectorial force */
905 fix1 = _mm_add_pd(fix1,tx);
906 fiy1 = _mm_add_pd(fiy1,ty);
907 fiz1 = _mm_add_pd(fiz1,tz);
909 fjx0 = _mm_add_pd(fjx0,tx);
910 fjy0 = _mm_add_pd(fjy0,ty);
911 fjz0 = _mm_add_pd(fjz0,tz);
913 /**************************
914 * CALCULATE INTERACTIONS *
915 **************************/
917 r20 = _mm_mul_pd(rsq20,rinv20);
919 /* Compute parameters for interactions between i and j atoms */
920 qq20 = _mm_mul_pd(iq2,jq0);
922 /* EWALD ELECTROSTATICS */
924 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
925 ewrt = _mm_mul_pd(r20,ewtabscale);
926 ewitab = _mm_cvttpd_epi32(ewrt);
927 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
928 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
930 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
931 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
935 /* Calculate temporary vectorial force */
936 tx = _mm_mul_pd(fscal,dx20);
937 ty = _mm_mul_pd(fscal,dy20);
938 tz = _mm_mul_pd(fscal,dz20);
940 /* Update vectorial force */
941 fix2 = _mm_add_pd(fix2,tx);
942 fiy2 = _mm_add_pd(fiy2,ty);
943 fiz2 = _mm_add_pd(fiz2,tz);
945 fjx0 = _mm_add_pd(fjx0,tx);
946 fjy0 = _mm_add_pd(fjy0,ty);
947 fjz0 = _mm_add_pd(fjz0,tz);
949 /**************************
950 * CALCULATE INTERACTIONS *
951 **************************/
953 r30 = _mm_mul_pd(rsq30,rinv30);
955 /* Compute parameters for interactions between i and j atoms */
956 qq30 = _mm_mul_pd(iq3,jq0);
958 /* EWALD ELECTROSTATICS */
960 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
961 ewrt = _mm_mul_pd(r30,ewtabscale);
962 ewitab = _mm_cvttpd_epi32(ewrt);
963 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
964 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
966 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
967 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
971 /* Calculate temporary vectorial force */
972 tx = _mm_mul_pd(fscal,dx30);
973 ty = _mm_mul_pd(fscal,dy30);
974 tz = _mm_mul_pd(fscal,dz30);
976 /* Update vectorial force */
977 fix3 = _mm_add_pd(fix3,tx);
978 fiy3 = _mm_add_pd(fiy3,ty);
979 fiz3 = _mm_add_pd(fiz3,tz);
981 fjx0 = _mm_add_pd(fjx0,tx);
982 fjy0 = _mm_add_pd(fjy0,ty);
983 fjz0 = _mm_add_pd(fjz0,tz);
985 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
987 /* Inner loop uses 138 flops */
994 j_coord_offsetA = DIM*jnrA;
996 /* load j atom coordinates */
997 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1000 /* Calculate displacement vector */
1001 dx00 = _mm_sub_pd(ix0,jx0);
1002 dy00 = _mm_sub_pd(iy0,jy0);
1003 dz00 = _mm_sub_pd(iz0,jz0);
1004 dx10 = _mm_sub_pd(ix1,jx0);
1005 dy10 = _mm_sub_pd(iy1,jy0);
1006 dz10 = _mm_sub_pd(iz1,jz0);
1007 dx20 = _mm_sub_pd(ix2,jx0);
1008 dy20 = _mm_sub_pd(iy2,jy0);
1009 dz20 = _mm_sub_pd(iz2,jz0);
1010 dx30 = _mm_sub_pd(ix3,jx0);
1011 dy30 = _mm_sub_pd(iy3,jy0);
1012 dz30 = _mm_sub_pd(iz3,jz0);
1014 /* Calculate squared distance and things based on it */
1015 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1016 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1017 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1018 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1020 rinv10 = sse41_invsqrt_d(rsq10);
1021 rinv20 = sse41_invsqrt_d(rsq20);
1022 rinv30 = sse41_invsqrt_d(rsq30);
1024 rinvsq00 = sse41_inv_d(rsq00);
1025 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1026 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1027 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
1029 /* Load parameters for j particles */
1030 jq0 = _mm_load_sd(charge+jnrA+0);
1031 vdwjidx0A = 2*vdwtype[jnrA+0];
1033 fjx0 = _mm_setzero_pd();
1034 fjy0 = _mm_setzero_pd();
1035 fjz0 = _mm_setzero_pd();
1037 /**************************
1038 * CALCULATE INTERACTIONS *
1039 **************************/
1041 /* Compute parameters for interactions between i and j atoms */
1042 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1044 /* LENNARD-JONES DISPERSION/REPULSION */
1046 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1047 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1051 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1053 /* Calculate temporary vectorial force */
1054 tx = _mm_mul_pd(fscal,dx00);
1055 ty = _mm_mul_pd(fscal,dy00);
1056 tz = _mm_mul_pd(fscal,dz00);
1058 /* Update vectorial force */
1059 fix0 = _mm_add_pd(fix0,tx);
1060 fiy0 = _mm_add_pd(fiy0,ty);
1061 fiz0 = _mm_add_pd(fiz0,tz);
1063 fjx0 = _mm_add_pd(fjx0,tx);
1064 fjy0 = _mm_add_pd(fjy0,ty);
1065 fjz0 = _mm_add_pd(fjz0,tz);
1067 /**************************
1068 * CALCULATE INTERACTIONS *
1069 **************************/
1071 r10 = _mm_mul_pd(rsq10,rinv10);
1073 /* Compute parameters for interactions between i and j atoms */
1074 qq10 = _mm_mul_pd(iq1,jq0);
1076 /* EWALD ELECTROSTATICS */
1078 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1079 ewrt = _mm_mul_pd(r10,ewtabscale);
1080 ewitab = _mm_cvttpd_epi32(ewrt);
1081 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1082 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1083 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1084 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1088 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1090 /* Calculate temporary vectorial force */
1091 tx = _mm_mul_pd(fscal,dx10);
1092 ty = _mm_mul_pd(fscal,dy10);
1093 tz = _mm_mul_pd(fscal,dz10);
1095 /* Update vectorial force */
1096 fix1 = _mm_add_pd(fix1,tx);
1097 fiy1 = _mm_add_pd(fiy1,ty);
1098 fiz1 = _mm_add_pd(fiz1,tz);
1100 fjx0 = _mm_add_pd(fjx0,tx);
1101 fjy0 = _mm_add_pd(fjy0,ty);
1102 fjz0 = _mm_add_pd(fjz0,tz);
1104 /**************************
1105 * CALCULATE INTERACTIONS *
1106 **************************/
1108 r20 = _mm_mul_pd(rsq20,rinv20);
1110 /* Compute parameters for interactions between i and j atoms */
1111 qq20 = _mm_mul_pd(iq2,jq0);
1113 /* EWALD ELECTROSTATICS */
1115 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1116 ewrt = _mm_mul_pd(r20,ewtabscale);
1117 ewitab = _mm_cvttpd_epi32(ewrt);
1118 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1119 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1120 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1121 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1125 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1127 /* Calculate temporary vectorial force */
1128 tx = _mm_mul_pd(fscal,dx20);
1129 ty = _mm_mul_pd(fscal,dy20);
1130 tz = _mm_mul_pd(fscal,dz20);
1132 /* Update vectorial force */
1133 fix2 = _mm_add_pd(fix2,tx);
1134 fiy2 = _mm_add_pd(fiy2,ty);
1135 fiz2 = _mm_add_pd(fiz2,tz);
1137 fjx0 = _mm_add_pd(fjx0,tx);
1138 fjy0 = _mm_add_pd(fjy0,ty);
1139 fjz0 = _mm_add_pd(fjz0,tz);
1141 /**************************
1142 * CALCULATE INTERACTIONS *
1143 **************************/
1145 r30 = _mm_mul_pd(rsq30,rinv30);
1147 /* Compute parameters for interactions between i and j atoms */
1148 qq30 = _mm_mul_pd(iq3,jq0);
1150 /* EWALD ELECTROSTATICS */
1152 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1153 ewrt = _mm_mul_pd(r30,ewtabscale);
1154 ewitab = _mm_cvttpd_epi32(ewrt);
1155 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1156 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1157 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1158 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1162 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1164 /* Calculate temporary vectorial force */
1165 tx = _mm_mul_pd(fscal,dx30);
1166 ty = _mm_mul_pd(fscal,dy30);
1167 tz = _mm_mul_pd(fscal,dz30);
1169 /* Update vectorial force */
1170 fix3 = _mm_add_pd(fix3,tx);
1171 fiy3 = _mm_add_pd(fiy3,ty);
1172 fiz3 = _mm_add_pd(fiz3,tz);
1174 fjx0 = _mm_add_pd(fjx0,tx);
1175 fjy0 = _mm_add_pd(fjy0,ty);
1176 fjz0 = _mm_add_pd(fjz0,tz);
1178 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1180 /* Inner loop uses 138 flops */
1183 /* End of innermost loop */
1185 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1186 f+i_coord_offset,fshift+i_shift_offset);
1188 /* Increment number of inner iterations */
1189 inneriter += j_index_end - j_index_start;
1191 /* Outer loop uses 24 flops */
1194 /* Increment number of outer iterations */
1197 /* Update outer/inner flops */
1199 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*138);