2 * Note: this file was generated by the Gromacs sse4_1_double kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_sse4_1_double.h"
34 #include "kernelutil_x86_sse4_1_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_sse4_1_double
38 * Electrostatics interaction: Ewald
39 * VdW interaction: LennardJones
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_sse4_1_double
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
61 int j_coord_offsetA,j_coord_offsetB;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
64 real *shiftvec,*fshift,*x,*f;
65 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
69 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
71 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
72 int vdwjidx0A,vdwjidx0B;
73 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
74 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
75 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
76 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
77 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
80 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
83 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
84 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
86 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
88 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
89 real rswitch_scalar,d_scalar;
90 __m128d dummy_mask,cutoff_mask;
91 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
92 __m128d one = _mm_set1_pd(1.0);
93 __m128d two = _mm_set1_pd(2.0);
99 jindex = nlist->jindex;
101 shiftidx = nlist->shift;
103 shiftvec = fr->shift_vec[0];
104 fshift = fr->fshift[0];
105 facel = _mm_set1_pd(fr->epsfac);
106 charge = mdatoms->chargeA;
107 nvdwtype = fr->ntype;
109 vdwtype = mdatoms->typeA;
111 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
112 ewtab = fr->ic->tabq_coul_FDV0;
113 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
114 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
116 /* Setup water-specific parameters */
117 inr = nlist->iinr[0];
118 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
119 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
120 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
121 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
123 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
124 rcutoff_scalar = fr->rcoulomb;
125 rcutoff = _mm_set1_pd(rcutoff_scalar);
126 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
128 rswitch_scalar = fr->rcoulomb_switch;
129 rswitch = _mm_set1_pd(rswitch_scalar);
130 /* Setup switch parameters */
131 d_scalar = rcutoff_scalar-rswitch_scalar;
132 d = _mm_set1_pd(d_scalar);
133 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
134 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
135 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
136 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
137 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
138 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
140 /* Avoid stupid compiler warnings */
148 /* Start outer loop over neighborlists */
149 for(iidx=0; iidx<nri; iidx++)
151 /* Load shift vector for this list */
152 i_shift_offset = DIM*shiftidx[iidx];
154 /* Load limits for loop over neighbors */
155 j_index_start = jindex[iidx];
156 j_index_end = jindex[iidx+1];
158 /* Get outer coordinate index */
160 i_coord_offset = DIM*inr;
162 /* Load i particle coords and add shift vector */
163 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
164 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
166 fix0 = _mm_setzero_pd();
167 fiy0 = _mm_setzero_pd();
168 fiz0 = _mm_setzero_pd();
169 fix1 = _mm_setzero_pd();
170 fiy1 = _mm_setzero_pd();
171 fiz1 = _mm_setzero_pd();
172 fix2 = _mm_setzero_pd();
173 fiy2 = _mm_setzero_pd();
174 fiz2 = _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);
205 /* Calculate squared distance and things based on it */
206 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
207 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
208 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
210 rinv00 = gmx_mm_invsqrt_pd(rsq00);
211 rinv10 = gmx_mm_invsqrt_pd(rsq10);
212 rinv20 = gmx_mm_invsqrt_pd(rsq20);
214 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
215 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
216 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
218 /* Load parameters for j particles */
219 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
220 vdwjidx0A = 2*vdwtype[jnrA+0];
221 vdwjidx0B = 2*vdwtype[jnrB+0];
223 fjx0 = _mm_setzero_pd();
224 fjy0 = _mm_setzero_pd();
225 fjz0 = _mm_setzero_pd();
227 /**************************
228 * CALCULATE INTERACTIONS *
229 **************************/
231 if (gmx_mm_any_lt(rsq00,rcutoff2))
234 r00 = _mm_mul_pd(rsq00,rinv00);
236 /* Compute parameters for interactions between i and j atoms */
237 qq00 = _mm_mul_pd(iq0,jq0);
238 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
239 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
241 /* EWALD ELECTROSTATICS */
243 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
244 ewrt = _mm_mul_pd(r00,ewtabscale);
245 ewitab = _mm_cvttpd_epi32(ewrt);
246 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
247 ewitab = _mm_slli_epi32(ewitab,2);
248 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
249 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
250 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
251 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
252 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
253 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
254 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
255 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
256 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
257 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
259 /* LENNARD-JONES DISPERSION/REPULSION */
261 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
262 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
263 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
264 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
265 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
267 d = _mm_sub_pd(r00,rswitch);
268 d = _mm_max_pd(d,_mm_setzero_pd());
269 d2 = _mm_mul_pd(d,d);
270 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
272 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
274 /* Evaluate switch function */
275 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
276 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
277 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
278 velec = _mm_mul_pd(velec,sw);
279 vvdw = _mm_mul_pd(vvdw,sw);
280 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
282 /* Update potential sum for this i atom from the interaction with this j atom. */
283 velec = _mm_and_pd(velec,cutoff_mask);
284 velecsum = _mm_add_pd(velecsum,velec);
285 vvdw = _mm_and_pd(vvdw,cutoff_mask);
286 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
288 fscal = _mm_add_pd(felec,fvdw);
290 fscal = _mm_and_pd(fscal,cutoff_mask);
292 /* Calculate temporary vectorial force */
293 tx = _mm_mul_pd(fscal,dx00);
294 ty = _mm_mul_pd(fscal,dy00);
295 tz = _mm_mul_pd(fscal,dz00);
297 /* Update vectorial force */
298 fix0 = _mm_add_pd(fix0,tx);
299 fiy0 = _mm_add_pd(fiy0,ty);
300 fiz0 = _mm_add_pd(fiz0,tz);
302 fjx0 = _mm_add_pd(fjx0,tx);
303 fjy0 = _mm_add_pd(fjy0,ty);
304 fjz0 = _mm_add_pd(fjz0,tz);
308 /**************************
309 * CALCULATE INTERACTIONS *
310 **************************/
312 if (gmx_mm_any_lt(rsq10,rcutoff2))
315 r10 = _mm_mul_pd(rsq10,rinv10);
317 /* Compute parameters for interactions between i and j atoms */
318 qq10 = _mm_mul_pd(iq1,jq0);
320 /* EWALD ELECTROSTATICS */
322 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
323 ewrt = _mm_mul_pd(r10,ewtabscale);
324 ewitab = _mm_cvttpd_epi32(ewrt);
325 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
326 ewitab = _mm_slli_epi32(ewitab,2);
327 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
328 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
329 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
330 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
331 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
332 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
333 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
334 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
335 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
336 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
338 d = _mm_sub_pd(r10,rswitch);
339 d = _mm_max_pd(d,_mm_setzero_pd());
340 d2 = _mm_mul_pd(d,d);
341 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
343 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
345 /* Evaluate switch function */
346 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
347 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
348 velec = _mm_mul_pd(velec,sw);
349 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
351 /* Update potential sum for this i atom from the interaction with this j atom. */
352 velec = _mm_and_pd(velec,cutoff_mask);
353 velecsum = _mm_add_pd(velecsum,velec);
357 fscal = _mm_and_pd(fscal,cutoff_mask);
359 /* Calculate temporary vectorial force */
360 tx = _mm_mul_pd(fscal,dx10);
361 ty = _mm_mul_pd(fscal,dy10);
362 tz = _mm_mul_pd(fscal,dz10);
364 /* Update vectorial force */
365 fix1 = _mm_add_pd(fix1,tx);
366 fiy1 = _mm_add_pd(fiy1,ty);
367 fiz1 = _mm_add_pd(fiz1,tz);
369 fjx0 = _mm_add_pd(fjx0,tx);
370 fjy0 = _mm_add_pd(fjy0,ty);
371 fjz0 = _mm_add_pd(fjz0,tz);
375 /**************************
376 * CALCULATE INTERACTIONS *
377 **************************/
379 if (gmx_mm_any_lt(rsq20,rcutoff2))
382 r20 = _mm_mul_pd(rsq20,rinv20);
384 /* Compute parameters for interactions between i and j atoms */
385 qq20 = _mm_mul_pd(iq2,jq0);
387 /* EWALD ELECTROSTATICS */
389 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
390 ewrt = _mm_mul_pd(r20,ewtabscale);
391 ewitab = _mm_cvttpd_epi32(ewrt);
392 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
393 ewitab = _mm_slli_epi32(ewitab,2);
394 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
395 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
396 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
397 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
398 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
399 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
400 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
401 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
402 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
403 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
405 d = _mm_sub_pd(r20,rswitch);
406 d = _mm_max_pd(d,_mm_setzero_pd());
407 d2 = _mm_mul_pd(d,d);
408 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
410 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
412 /* Evaluate switch function */
413 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
414 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
415 velec = _mm_mul_pd(velec,sw);
416 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
418 /* Update potential sum for this i atom from the interaction with this j atom. */
419 velec = _mm_and_pd(velec,cutoff_mask);
420 velecsum = _mm_add_pd(velecsum,velec);
424 fscal = _mm_and_pd(fscal,cutoff_mask);
426 /* Calculate temporary vectorial force */
427 tx = _mm_mul_pd(fscal,dx20);
428 ty = _mm_mul_pd(fscal,dy20);
429 tz = _mm_mul_pd(fscal,dz20);
431 /* Update vectorial force */
432 fix2 = _mm_add_pd(fix2,tx);
433 fiy2 = _mm_add_pd(fiy2,ty);
434 fiz2 = _mm_add_pd(fiz2,tz);
436 fjx0 = _mm_add_pd(fjx0,tx);
437 fjy0 = _mm_add_pd(fjy0,ty);
438 fjz0 = _mm_add_pd(fjz0,tz);
442 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
444 /* Inner loop uses 216 flops */
451 j_coord_offsetA = DIM*jnrA;
453 /* load j atom coordinates */
454 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
457 /* Calculate displacement vector */
458 dx00 = _mm_sub_pd(ix0,jx0);
459 dy00 = _mm_sub_pd(iy0,jy0);
460 dz00 = _mm_sub_pd(iz0,jz0);
461 dx10 = _mm_sub_pd(ix1,jx0);
462 dy10 = _mm_sub_pd(iy1,jy0);
463 dz10 = _mm_sub_pd(iz1,jz0);
464 dx20 = _mm_sub_pd(ix2,jx0);
465 dy20 = _mm_sub_pd(iy2,jy0);
466 dz20 = _mm_sub_pd(iz2,jz0);
468 /* Calculate squared distance and things based on it */
469 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
470 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
471 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
473 rinv00 = gmx_mm_invsqrt_pd(rsq00);
474 rinv10 = gmx_mm_invsqrt_pd(rsq10);
475 rinv20 = gmx_mm_invsqrt_pd(rsq20);
477 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
478 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
479 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
481 /* Load parameters for j particles */
482 jq0 = _mm_load_sd(charge+jnrA+0);
483 vdwjidx0A = 2*vdwtype[jnrA+0];
485 fjx0 = _mm_setzero_pd();
486 fjy0 = _mm_setzero_pd();
487 fjz0 = _mm_setzero_pd();
489 /**************************
490 * CALCULATE INTERACTIONS *
491 **************************/
493 if (gmx_mm_any_lt(rsq00,rcutoff2))
496 r00 = _mm_mul_pd(rsq00,rinv00);
498 /* Compute parameters for interactions between i and j atoms */
499 qq00 = _mm_mul_pd(iq0,jq0);
500 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
502 /* EWALD ELECTROSTATICS */
504 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
505 ewrt = _mm_mul_pd(r00,ewtabscale);
506 ewitab = _mm_cvttpd_epi32(ewrt);
507 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
508 ewitab = _mm_slli_epi32(ewitab,2);
509 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
510 ewtabD = _mm_setzero_pd();
511 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
512 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
513 ewtabFn = _mm_setzero_pd();
514 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
515 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
516 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
517 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
518 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
520 /* LENNARD-JONES DISPERSION/REPULSION */
522 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
523 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
524 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
525 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
526 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
528 d = _mm_sub_pd(r00,rswitch);
529 d = _mm_max_pd(d,_mm_setzero_pd());
530 d2 = _mm_mul_pd(d,d);
531 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
533 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
535 /* Evaluate switch function */
536 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
537 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
538 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
539 velec = _mm_mul_pd(velec,sw);
540 vvdw = _mm_mul_pd(vvdw,sw);
541 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
543 /* Update potential sum for this i atom from the interaction with this j atom. */
544 velec = _mm_and_pd(velec,cutoff_mask);
545 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
546 velecsum = _mm_add_pd(velecsum,velec);
547 vvdw = _mm_and_pd(vvdw,cutoff_mask);
548 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
549 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
551 fscal = _mm_add_pd(felec,fvdw);
553 fscal = _mm_and_pd(fscal,cutoff_mask);
555 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
557 /* Calculate temporary vectorial force */
558 tx = _mm_mul_pd(fscal,dx00);
559 ty = _mm_mul_pd(fscal,dy00);
560 tz = _mm_mul_pd(fscal,dz00);
562 /* Update vectorial force */
563 fix0 = _mm_add_pd(fix0,tx);
564 fiy0 = _mm_add_pd(fiy0,ty);
565 fiz0 = _mm_add_pd(fiz0,tz);
567 fjx0 = _mm_add_pd(fjx0,tx);
568 fjy0 = _mm_add_pd(fjy0,ty);
569 fjz0 = _mm_add_pd(fjz0,tz);
573 /**************************
574 * CALCULATE INTERACTIONS *
575 **************************/
577 if (gmx_mm_any_lt(rsq10,rcutoff2))
580 r10 = _mm_mul_pd(rsq10,rinv10);
582 /* Compute parameters for interactions between i and j atoms */
583 qq10 = _mm_mul_pd(iq1,jq0);
585 /* EWALD ELECTROSTATICS */
587 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
588 ewrt = _mm_mul_pd(r10,ewtabscale);
589 ewitab = _mm_cvttpd_epi32(ewrt);
590 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
591 ewitab = _mm_slli_epi32(ewitab,2);
592 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
593 ewtabD = _mm_setzero_pd();
594 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
595 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
596 ewtabFn = _mm_setzero_pd();
597 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
598 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
599 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
600 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
601 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
603 d = _mm_sub_pd(r10,rswitch);
604 d = _mm_max_pd(d,_mm_setzero_pd());
605 d2 = _mm_mul_pd(d,d);
606 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
608 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
610 /* Evaluate switch function */
611 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
612 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
613 velec = _mm_mul_pd(velec,sw);
614 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
616 /* Update potential sum for this i atom from the interaction with this j atom. */
617 velec = _mm_and_pd(velec,cutoff_mask);
618 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
619 velecsum = _mm_add_pd(velecsum,velec);
623 fscal = _mm_and_pd(fscal,cutoff_mask);
625 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
627 /* Calculate temporary vectorial force */
628 tx = _mm_mul_pd(fscal,dx10);
629 ty = _mm_mul_pd(fscal,dy10);
630 tz = _mm_mul_pd(fscal,dz10);
632 /* Update vectorial force */
633 fix1 = _mm_add_pd(fix1,tx);
634 fiy1 = _mm_add_pd(fiy1,ty);
635 fiz1 = _mm_add_pd(fiz1,tz);
637 fjx0 = _mm_add_pd(fjx0,tx);
638 fjy0 = _mm_add_pd(fjy0,ty);
639 fjz0 = _mm_add_pd(fjz0,tz);
643 /**************************
644 * CALCULATE INTERACTIONS *
645 **************************/
647 if (gmx_mm_any_lt(rsq20,rcutoff2))
650 r20 = _mm_mul_pd(rsq20,rinv20);
652 /* Compute parameters for interactions between i and j atoms */
653 qq20 = _mm_mul_pd(iq2,jq0);
655 /* EWALD ELECTROSTATICS */
657 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
658 ewrt = _mm_mul_pd(r20,ewtabscale);
659 ewitab = _mm_cvttpd_epi32(ewrt);
660 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
661 ewitab = _mm_slli_epi32(ewitab,2);
662 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
663 ewtabD = _mm_setzero_pd();
664 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
665 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
666 ewtabFn = _mm_setzero_pd();
667 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
668 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
669 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
670 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
671 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
673 d = _mm_sub_pd(r20,rswitch);
674 d = _mm_max_pd(d,_mm_setzero_pd());
675 d2 = _mm_mul_pd(d,d);
676 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
678 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
680 /* Evaluate switch function */
681 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
682 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
683 velec = _mm_mul_pd(velec,sw);
684 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
686 /* Update potential sum for this i atom from the interaction with this j atom. */
687 velec = _mm_and_pd(velec,cutoff_mask);
688 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
689 velecsum = _mm_add_pd(velecsum,velec);
693 fscal = _mm_and_pd(fscal,cutoff_mask);
695 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
697 /* Calculate temporary vectorial force */
698 tx = _mm_mul_pd(fscal,dx20);
699 ty = _mm_mul_pd(fscal,dy20);
700 tz = _mm_mul_pd(fscal,dz20);
702 /* Update vectorial force */
703 fix2 = _mm_add_pd(fix2,tx);
704 fiy2 = _mm_add_pd(fiy2,ty);
705 fiz2 = _mm_add_pd(fiz2,tz);
707 fjx0 = _mm_add_pd(fjx0,tx);
708 fjy0 = _mm_add_pd(fjy0,ty);
709 fjz0 = _mm_add_pd(fjz0,tz);
713 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
715 /* Inner loop uses 216 flops */
718 /* End of innermost loop */
720 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
721 f+i_coord_offset,fshift+i_shift_offset);
724 /* Update potential energies */
725 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
726 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
728 /* Increment number of inner iterations */
729 inneriter += j_index_end - j_index_start;
731 /* Outer loop uses 20 flops */
734 /* Increment number of outer iterations */
737 /* Update outer/inner flops */
739 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*216);
742 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_sse4_1_double
743 * Electrostatics interaction: Ewald
744 * VdW interaction: LennardJones
745 * Geometry: Water3-Particle
746 * Calculate force/pot: Force
749 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_sse4_1_double
750 (t_nblist * gmx_restrict nlist,
751 rvec * gmx_restrict xx,
752 rvec * gmx_restrict ff,
753 t_forcerec * gmx_restrict fr,
754 t_mdatoms * gmx_restrict mdatoms,
755 nb_kernel_data_t * gmx_restrict kernel_data,
756 t_nrnb * gmx_restrict nrnb)
758 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
759 * just 0 for non-waters.
760 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
761 * jnr indices corresponding to data put in the four positions in the SIMD register.
763 int i_shift_offset,i_coord_offset,outeriter,inneriter;
764 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
766 int j_coord_offsetA,j_coord_offsetB;
767 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
769 real *shiftvec,*fshift,*x,*f;
770 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
772 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
774 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
776 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
777 int vdwjidx0A,vdwjidx0B;
778 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
779 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
780 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
781 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
782 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
785 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
788 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
789 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
791 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
793 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
794 real rswitch_scalar,d_scalar;
795 __m128d dummy_mask,cutoff_mask;
796 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
797 __m128d one = _mm_set1_pd(1.0);
798 __m128d two = _mm_set1_pd(2.0);
804 jindex = nlist->jindex;
806 shiftidx = nlist->shift;
808 shiftvec = fr->shift_vec[0];
809 fshift = fr->fshift[0];
810 facel = _mm_set1_pd(fr->epsfac);
811 charge = mdatoms->chargeA;
812 nvdwtype = fr->ntype;
814 vdwtype = mdatoms->typeA;
816 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
817 ewtab = fr->ic->tabq_coul_FDV0;
818 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
819 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
821 /* Setup water-specific parameters */
822 inr = nlist->iinr[0];
823 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
824 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
825 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
826 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
828 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
829 rcutoff_scalar = fr->rcoulomb;
830 rcutoff = _mm_set1_pd(rcutoff_scalar);
831 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
833 rswitch_scalar = fr->rcoulomb_switch;
834 rswitch = _mm_set1_pd(rswitch_scalar);
835 /* Setup switch parameters */
836 d_scalar = rcutoff_scalar-rswitch_scalar;
837 d = _mm_set1_pd(d_scalar);
838 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
839 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
840 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
841 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
842 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
843 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
845 /* Avoid stupid compiler warnings */
853 /* Start outer loop over neighborlists */
854 for(iidx=0; iidx<nri; iidx++)
856 /* Load shift vector for this list */
857 i_shift_offset = DIM*shiftidx[iidx];
859 /* Load limits for loop over neighbors */
860 j_index_start = jindex[iidx];
861 j_index_end = jindex[iidx+1];
863 /* Get outer coordinate index */
865 i_coord_offset = DIM*inr;
867 /* Load i particle coords and add shift vector */
868 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
869 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
871 fix0 = _mm_setzero_pd();
872 fiy0 = _mm_setzero_pd();
873 fiz0 = _mm_setzero_pd();
874 fix1 = _mm_setzero_pd();
875 fiy1 = _mm_setzero_pd();
876 fiz1 = _mm_setzero_pd();
877 fix2 = _mm_setzero_pd();
878 fiy2 = _mm_setzero_pd();
879 fiz2 = _mm_setzero_pd();
881 /* Start inner kernel loop */
882 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
885 /* Get j neighbor index, and coordinate index */
888 j_coord_offsetA = DIM*jnrA;
889 j_coord_offsetB = DIM*jnrB;
891 /* load j atom coordinates */
892 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
895 /* Calculate displacement vector */
896 dx00 = _mm_sub_pd(ix0,jx0);
897 dy00 = _mm_sub_pd(iy0,jy0);
898 dz00 = _mm_sub_pd(iz0,jz0);
899 dx10 = _mm_sub_pd(ix1,jx0);
900 dy10 = _mm_sub_pd(iy1,jy0);
901 dz10 = _mm_sub_pd(iz1,jz0);
902 dx20 = _mm_sub_pd(ix2,jx0);
903 dy20 = _mm_sub_pd(iy2,jy0);
904 dz20 = _mm_sub_pd(iz2,jz0);
906 /* Calculate squared distance and things based on it */
907 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
908 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
909 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
911 rinv00 = gmx_mm_invsqrt_pd(rsq00);
912 rinv10 = gmx_mm_invsqrt_pd(rsq10);
913 rinv20 = gmx_mm_invsqrt_pd(rsq20);
915 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
916 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
917 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
919 /* Load parameters for j particles */
920 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
921 vdwjidx0A = 2*vdwtype[jnrA+0];
922 vdwjidx0B = 2*vdwtype[jnrB+0];
924 fjx0 = _mm_setzero_pd();
925 fjy0 = _mm_setzero_pd();
926 fjz0 = _mm_setzero_pd();
928 /**************************
929 * CALCULATE INTERACTIONS *
930 **************************/
932 if (gmx_mm_any_lt(rsq00,rcutoff2))
935 r00 = _mm_mul_pd(rsq00,rinv00);
937 /* Compute parameters for interactions between i and j atoms */
938 qq00 = _mm_mul_pd(iq0,jq0);
939 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
940 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
942 /* EWALD ELECTROSTATICS */
944 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
945 ewrt = _mm_mul_pd(r00,ewtabscale);
946 ewitab = _mm_cvttpd_epi32(ewrt);
947 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
948 ewitab = _mm_slli_epi32(ewitab,2);
949 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
950 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
951 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
952 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
953 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
954 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
955 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
956 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
957 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
958 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
960 /* LENNARD-JONES DISPERSION/REPULSION */
962 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
963 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
964 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
965 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
966 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
968 d = _mm_sub_pd(r00,rswitch);
969 d = _mm_max_pd(d,_mm_setzero_pd());
970 d2 = _mm_mul_pd(d,d);
971 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
973 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
975 /* Evaluate switch function */
976 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
977 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
978 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
979 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
981 fscal = _mm_add_pd(felec,fvdw);
983 fscal = _mm_and_pd(fscal,cutoff_mask);
985 /* Calculate temporary vectorial force */
986 tx = _mm_mul_pd(fscal,dx00);
987 ty = _mm_mul_pd(fscal,dy00);
988 tz = _mm_mul_pd(fscal,dz00);
990 /* Update vectorial force */
991 fix0 = _mm_add_pd(fix0,tx);
992 fiy0 = _mm_add_pd(fiy0,ty);
993 fiz0 = _mm_add_pd(fiz0,tz);
995 fjx0 = _mm_add_pd(fjx0,tx);
996 fjy0 = _mm_add_pd(fjy0,ty);
997 fjz0 = _mm_add_pd(fjz0,tz);
1001 /**************************
1002 * CALCULATE INTERACTIONS *
1003 **************************/
1005 if (gmx_mm_any_lt(rsq10,rcutoff2))
1008 r10 = _mm_mul_pd(rsq10,rinv10);
1010 /* Compute parameters for interactions between i and j atoms */
1011 qq10 = _mm_mul_pd(iq1,jq0);
1013 /* EWALD ELECTROSTATICS */
1015 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1016 ewrt = _mm_mul_pd(r10,ewtabscale);
1017 ewitab = _mm_cvttpd_epi32(ewrt);
1018 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1019 ewitab = _mm_slli_epi32(ewitab,2);
1020 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1021 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1022 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1023 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1024 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1025 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1026 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1027 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1028 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1029 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1031 d = _mm_sub_pd(r10,rswitch);
1032 d = _mm_max_pd(d,_mm_setzero_pd());
1033 d2 = _mm_mul_pd(d,d);
1034 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1036 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1038 /* Evaluate switch function */
1039 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1040 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1041 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1045 fscal = _mm_and_pd(fscal,cutoff_mask);
1047 /* Calculate temporary vectorial force */
1048 tx = _mm_mul_pd(fscal,dx10);
1049 ty = _mm_mul_pd(fscal,dy10);
1050 tz = _mm_mul_pd(fscal,dz10);
1052 /* Update vectorial force */
1053 fix1 = _mm_add_pd(fix1,tx);
1054 fiy1 = _mm_add_pd(fiy1,ty);
1055 fiz1 = _mm_add_pd(fiz1,tz);
1057 fjx0 = _mm_add_pd(fjx0,tx);
1058 fjy0 = _mm_add_pd(fjy0,ty);
1059 fjz0 = _mm_add_pd(fjz0,tz);
1063 /**************************
1064 * CALCULATE INTERACTIONS *
1065 **************************/
1067 if (gmx_mm_any_lt(rsq20,rcutoff2))
1070 r20 = _mm_mul_pd(rsq20,rinv20);
1072 /* Compute parameters for interactions between i and j atoms */
1073 qq20 = _mm_mul_pd(iq2,jq0);
1075 /* EWALD ELECTROSTATICS */
1077 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1078 ewrt = _mm_mul_pd(r20,ewtabscale);
1079 ewitab = _mm_cvttpd_epi32(ewrt);
1080 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1081 ewitab = _mm_slli_epi32(ewitab,2);
1082 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1083 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1084 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1085 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1086 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1087 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1088 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1089 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1090 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1091 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1093 d = _mm_sub_pd(r20,rswitch);
1094 d = _mm_max_pd(d,_mm_setzero_pd());
1095 d2 = _mm_mul_pd(d,d);
1096 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1098 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1100 /* Evaluate switch function */
1101 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1102 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1103 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1107 fscal = _mm_and_pd(fscal,cutoff_mask);
1109 /* Calculate temporary vectorial force */
1110 tx = _mm_mul_pd(fscal,dx20);
1111 ty = _mm_mul_pd(fscal,dy20);
1112 tz = _mm_mul_pd(fscal,dz20);
1114 /* Update vectorial force */
1115 fix2 = _mm_add_pd(fix2,tx);
1116 fiy2 = _mm_add_pd(fiy2,ty);
1117 fiz2 = _mm_add_pd(fiz2,tz);
1119 fjx0 = _mm_add_pd(fjx0,tx);
1120 fjy0 = _mm_add_pd(fjy0,ty);
1121 fjz0 = _mm_add_pd(fjz0,tz);
1125 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1127 /* Inner loop uses 204 flops */
1130 if(jidx<j_index_end)
1134 j_coord_offsetA = DIM*jnrA;
1136 /* load j atom coordinates */
1137 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1140 /* Calculate displacement vector */
1141 dx00 = _mm_sub_pd(ix0,jx0);
1142 dy00 = _mm_sub_pd(iy0,jy0);
1143 dz00 = _mm_sub_pd(iz0,jz0);
1144 dx10 = _mm_sub_pd(ix1,jx0);
1145 dy10 = _mm_sub_pd(iy1,jy0);
1146 dz10 = _mm_sub_pd(iz1,jz0);
1147 dx20 = _mm_sub_pd(ix2,jx0);
1148 dy20 = _mm_sub_pd(iy2,jy0);
1149 dz20 = _mm_sub_pd(iz2,jz0);
1151 /* Calculate squared distance and things based on it */
1152 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1153 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1154 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1156 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1157 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1158 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1160 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1161 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1162 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1164 /* Load parameters for j particles */
1165 jq0 = _mm_load_sd(charge+jnrA+0);
1166 vdwjidx0A = 2*vdwtype[jnrA+0];
1168 fjx0 = _mm_setzero_pd();
1169 fjy0 = _mm_setzero_pd();
1170 fjz0 = _mm_setzero_pd();
1172 /**************************
1173 * CALCULATE INTERACTIONS *
1174 **************************/
1176 if (gmx_mm_any_lt(rsq00,rcutoff2))
1179 r00 = _mm_mul_pd(rsq00,rinv00);
1181 /* Compute parameters for interactions between i and j atoms */
1182 qq00 = _mm_mul_pd(iq0,jq0);
1183 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1185 /* EWALD ELECTROSTATICS */
1187 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1188 ewrt = _mm_mul_pd(r00,ewtabscale);
1189 ewitab = _mm_cvttpd_epi32(ewrt);
1190 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1191 ewitab = _mm_slli_epi32(ewitab,2);
1192 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1193 ewtabD = _mm_setzero_pd();
1194 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1195 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1196 ewtabFn = _mm_setzero_pd();
1197 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1198 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1199 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1200 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
1201 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1203 /* LENNARD-JONES DISPERSION/REPULSION */
1205 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1206 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
1207 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1208 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
1209 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1211 d = _mm_sub_pd(r00,rswitch);
1212 d = _mm_max_pd(d,_mm_setzero_pd());
1213 d2 = _mm_mul_pd(d,d);
1214 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1216 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1218 /* Evaluate switch function */
1219 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1220 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
1221 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1222 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1224 fscal = _mm_add_pd(felec,fvdw);
1226 fscal = _mm_and_pd(fscal,cutoff_mask);
1228 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1230 /* Calculate temporary vectorial force */
1231 tx = _mm_mul_pd(fscal,dx00);
1232 ty = _mm_mul_pd(fscal,dy00);
1233 tz = _mm_mul_pd(fscal,dz00);
1235 /* Update vectorial force */
1236 fix0 = _mm_add_pd(fix0,tx);
1237 fiy0 = _mm_add_pd(fiy0,ty);
1238 fiz0 = _mm_add_pd(fiz0,tz);
1240 fjx0 = _mm_add_pd(fjx0,tx);
1241 fjy0 = _mm_add_pd(fjy0,ty);
1242 fjz0 = _mm_add_pd(fjz0,tz);
1246 /**************************
1247 * CALCULATE INTERACTIONS *
1248 **************************/
1250 if (gmx_mm_any_lt(rsq10,rcutoff2))
1253 r10 = _mm_mul_pd(rsq10,rinv10);
1255 /* Compute parameters for interactions between i and j atoms */
1256 qq10 = _mm_mul_pd(iq1,jq0);
1258 /* EWALD ELECTROSTATICS */
1260 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1261 ewrt = _mm_mul_pd(r10,ewtabscale);
1262 ewitab = _mm_cvttpd_epi32(ewrt);
1263 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1264 ewitab = _mm_slli_epi32(ewitab,2);
1265 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1266 ewtabD = _mm_setzero_pd();
1267 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1268 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1269 ewtabFn = _mm_setzero_pd();
1270 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1271 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1272 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1273 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1274 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1276 d = _mm_sub_pd(r10,rswitch);
1277 d = _mm_max_pd(d,_mm_setzero_pd());
1278 d2 = _mm_mul_pd(d,d);
1279 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1281 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1283 /* Evaluate switch function */
1284 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1285 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1286 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1290 fscal = _mm_and_pd(fscal,cutoff_mask);
1292 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1294 /* Calculate temporary vectorial force */
1295 tx = _mm_mul_pd(fscal,dx10);
1296 ty = _mm_mul_pd(fscal,dy10);
1297 tz = _mm_mul_pd(fscal,dz10);
1299 /* Update vectorial force */
1300 fix1 = _mm_add_pd(fix1,tx);
1301 fiy1 = _mm_add_pd(fiy1,ty);
1302 fiz1 = _mm_add_pd(fiz1,tz);
1304 fjx0 = _mm_add_pd(fjx0,tx);
1305 fjy0 = _mm_add_pd(fjy0,ty);
1306 fjz0 = _mm_add_pd(fjz0,tz);
1310 /**************************
1311 * CALCULATE INTERACTIONS *
1312 **************************/
1314 if (gmx_mm_any_lt(rsq20,rcutoff2))
1317 r20 = _mm_mul_pd(rsq20,rinv20);
1319 /* Compute parameters for interactions between i and j atoms */
1320 qq20 = _mm_mul_pd(iq2,jq0);
1322 /* EWALD ELECTROSTATICS */
1324 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1325 ewrt = _mm_mul_pd(r20,ewtabscale);
1326 ewitab = _mm_cvttpd_epi32(ewrt);
1327 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1328 ewitab = _mm_slli_epi32(ewitab,2);
1329 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1330 ewtabD = _mm_setzero_pd();
1331 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1332 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1333 ewtabFn = _mm_setzero_pd();
1334 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1335 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1336 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1337 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1338 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1340 d = _mm_sub_pd(r20,rswitch);
1341 d = _mm_max_pd(d,_mm_setzero_pd());
1342 d2 = _mm_mul_pd(d,d);
1343 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1345 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1347 /* Evaluate switch function */
1348 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1349 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1350 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1354 fscal = _mm_and_pd(fscal,cutoff_mask);
1356 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1358 /* Calculate temporary vectorial force */
1359 tx = _mm_mul_pd(fscal,dx20);
1360 ty = _mm_mul_pd(fscal,dy20);
1361 tz = _mm_mul_pd(fscal,dz20);
1363 /* Update vectorial force */
1364 fix2 = _mm_add_pd(fix2,tx);
1365 fiy2 = _mm_add_pd(fiy2,ty);
1366 fiz2 = _mm_add_pd(fiz2,tz);
1368 fjx0 = _mm_add_pd(fjx0,tx);
1369 fjy0 = _mm_add_pd(fjy0,ty);
1370 fjz0 = _mm_add_pd(fjz0,tz);
1374 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1376 /* Inner loop uses 204 flops */
1379 /* End of innermost loop */
1381 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1382 f+i_coord_offset,fshift+i_shift_offset);
1384 /* Increment number of inner iterations */
1385 inneriter += j_index_end - j_index_start;
1387 /* Outer loop uses 18 flops */
1390 /* Increment number of outer iterations */
1393 /* Update outer/inner flops */
1395 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*204);