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 avx_128_fma_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
48 #include "kernelutil_x86_avx_128_fma_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_128_fma_double
52 * Electrostatics interaction: Ewald
53 * VdW interaction: LennardJones
54 * Geometry: Water3-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_128_fma_double
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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
75 int j_coord_offsetA,j_coord_offsetB;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
83 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
86 int vdwjidx0A,vdwjidx0B;
87 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
88 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
89 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
90 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
91 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
94 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
97 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
98 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
100 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
102 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
103 real rswitch_scalar,d_scalar;
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->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 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
133 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
134 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
135 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
137 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
138 rcutoff_scalar = fr->rcoulomb;
139 rcutoff = _mm_set1_pd(rcutoff_scalar);
140 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
142 rswitch_scalar = fr->rcoulomb_switch;
143 rswitch = _mm_set1_pd(rswitch_scalar);
144 /* Setup switch parameters */
145 d_scalar = rcutoff_scalar-rswitch_scalar;
146 d = _mm_set1_pd(d_scalar);
147 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
148 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
149 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
150 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
151 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
152 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
154 /* Avoid stupid compiler warnings */
162 /* Start outer loop over neighborlists */
163 for(iidx=0; iidx<nri; iidx++)
165 /* Load shift vector for this list */
166 i_shift_offset = DIM*shiftidx[iidx];
168 /* Load limits for loop over neighbors */
169 j_index_start = jindex[iidx];
170 j_index_end = jindex[iidx+1];
172 /* Get outer coordinate index */
174 i_coord_offset = DIM*inr;
176 /* Load i particle coords and add shift vector */
177 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
178 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
180 fix0 = _mm_setzero_pd();
181 fiy0 = _mm_setzero_pd();
182 fiz0 = _mm_setzero_pd();
183 fix1 = _mm_setzero_pd();
184 fiy1 = _mm_setzero_pd();
185 fiz1 = _mm_setzero_pd();
186 fix2 = _mm_setzero_pd();
187 fiy2 = _mm_setzero_pd();
188 fiz2 = _mm_setzero_pd();
190 /* Reset potential sums */
191 velecsum = _mm_setzero_pd();
192 vvdwsum = _mm_setzero_pd();
194 /* Start inner kernel loop */
195 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
198 /* Get j neighbor index, and coordinate index */
201 j_coord_offsetA = DIM*jnrA;
202 j_coord_offsetB = DIM*jnrB;
204 /* load j atom coordinates */
205 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
208 /* Calculate displacement vector */
209 dx00 = _mm_sub_pd(ix0,jx0);
210 dy00 = _mm_sub_pd(iy0,jy0);
211 dz00 = _mm_sub_pd(iz0,jz0);
212 dx10 = _mm_sub_pd(ix1,jx0);
213 dy10 = _mm_sub_pd(iy1,jy0);
214 dz10 = _mm_sub_pd(iz1,jz0);
215 dx20 = _mm_sub_pd(ix2,jx0);
216 dy20 = _mm_sub_pd(iy2,jy0);
217 dz20 = _mm_sub_pd(iz2,jz0);
219 /* Calculate squared distance and things based on it */
220 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
221 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
222 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
224 rinv00 = gmx_mm_invsqrt_pd(rsq00);
225 rinv10 = gmx_mm_invsqrt_pd(rsq10);
226 rinv20 = gmx_mm_invsqrt_pd(rsq20);
228 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
229 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
230 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
232 /* Load parameters for j particles */
233 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
234 vdwjidx0A = 2*vdwtype[jnrA+0];
235 vdwjidx0B = 2*vdwtype[jnrB+0];
237 fjx0 = _mm_setzero_pd();
238 fjy0 = _mm_setzero_pd();
239 fjz0 = _mm_setzero_pd();
241 /**************************
242 * CALCULATE INTERACTIONS *
243 **************************/
245 if (gmx_mm_any_lt(rsq00,rcutoff2))
248 r00 = _mm_mul_pd(rsq00,rinv00);
250 /* Compute parameters for interactions between i and j atoms */
251 qq00 = _mm_mul_pd(iq0,jq0);
252 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
253 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
255 /* EWALD ELECTROSTATICS */
257 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
258 ewrt = _mm_mul_pd(r00,ewtabscale);
259 ewitab = _mm_cvttpd_epi32(ewrt);
261 eweps = _mm_frcz_pd(ewrt);
263 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
265 twoeweps = _mm_add_pd(eweps,eweps);
266 ewitab = _mm_slli_epi32(ewitab,2);
267 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
268 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
269 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
270 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
271 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
272 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
273 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
274 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
275 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
276 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
278 /* LENNARD-JONES DISPERSION/REPULSION */
280 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
281 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
282 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
283 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
284 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
286 d = _mm_sub_pd(r00,rswitch);
287 d = _mm_max_pd(d,_mm_setzero_pd());
288 d2 = _mm_mul_pd(d,d);
289 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
291 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
293 /* Evaluate switch function */
294 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
295 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
296 fvdw = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
297 velec = _mm_mul_pd(velec,sw);
298 vvdw = _mm_mul_pd(vvdw,sw);
299 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
301 /* Update potential sum for this i atom from the interaction with this j atom. */
302 velec = _mm_and_pd(velec,cutoff_mask);
303 velecsum = _mm_add_pd(velecsum,velec);
304 vvdw = _mm_and_pd(vvdw,cutoff_mask);
305 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
307 fscal = _mm_add_pd(felec,fvdw);
309 fscal = _mm_and_pd(fscal,cutoff_mask);
311 /* Update vectorial force */
312 fix0 = _mm_macc_pd(dx00,fscal,fix0);
313 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
314 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
316 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
317 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
318 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
322 /**************************
323 * CALCULATE INTERACTIONS *
324 **************************/
326 if (gmx_mm_any_lt(rsq10,rcutoff2))
329 r10 = _mm_mul_pd(rsq10,rinv10);
331 /* Compute parameters for interactions between i and j atoms */
332 qq10 = _mm_mul_pd(iq1,jq0);
334 /* EWALD ELECTROSTATICS */
336 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
337 ewrt = _mm_mul_pd(r10,ewtabscale);
338 ewitab = _mm_cvttpd_epi32(ewrt);
340 eweps = _mm_frcz_pd(ewrt);
342 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
344 twoeweps = _mm_add_pd(eweps,eweps);
345 ewitab = _mm_slli_epi32(ewitab,2);
346 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
347 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
348 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
349 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
350 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
351 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
352 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
353 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
354 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
355 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
357 d = _mm_sub_pd(r10,rswitch);
358 d = _mm_max_pd(d,_mm_setzero_pd());
359 d2 = _mm_mul_pd(d,d);
360 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
362 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
364 /* Evaluate switch function */
365 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
366 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
367 velec = _mm_mul_pd(velec,sw);
368 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
370 /* Update potential sum for this i atom from the interaction with this j atom. */
371 velec = _mm_and_pd(velec,cutoff_mask);
372 velecsum = _mm_add_pd(velecsum,velec);
376 fscal = _mm_and_pd(fscal,cutoff_mask);
378 /* Update vectorial force */
379 fix1 = _mm_macc_pd(dx10,fscal,fix1);
380 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
381 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
383 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
384 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
385 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
389 /**************************
390 * CALCULATE INTERACTIONS *
391 **************************/
393 if (gmx_mm_any_lt(rsq20,rcutoff2))
396 r20 = _mm_mul_pd(rsq20,rinv20);
398 /* Compute parameters for interactions between i and j atoms */
399 qq20 = _mm_mul_pd(iq2,jq0);
401 /* EWALD ELECTROSTATICS */
403 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
404 ewrt = _mm_mul_pd(r20,ewtabscale);
405 ewitab = _mm_cvttpd_epi32(ewrt);
407 eweps = _mm_frcz_pd(ewrt);
409 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
411 twoeweps = _mm_add_pd(eweps,eweps);
412 ewitab = _mm_slli_epi32(ewitab,2);
413 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
414 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
415 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
416 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
417 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
418 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
419 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
420 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
421 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
422 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
424 d = _mm_sub_pd(r20,rswitch);
425 d = _mm_max_pd(d,_mm_setzero_pd());
426 d2 = _mm_mul_pd(d,d);
427 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
429 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
431 /* Evaluate switch function */
432 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
433 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
434 velec = _mm_mul_pd(velec,sw);
435 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
437 /* Update potential sum for this i atom from the interaction with this j atom. */
438 velec = _mm_and_pd(velec,cutoff_mask);
439 velecsum = _mm_add_pd(velecsum,velec);
443 fscal = _mm_and_pd(fscal,cutoff_mask);
445 /* Update vectorial force */
446 fix2 = _mm_macc_pd(dx20,fscal,fix2);
447 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
448 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
450 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
451 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
452 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
456 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
458 /* Inner loop uses 225 flops */
465 j_coord_offsetA = DIM*jnrA;
467 /* load j atom coordinates */
468 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
471 /* Calculate displacement vector */
472 dx00 = _mm_sub_pd(ix0,jx0);
473 dy00 = _mm_sub_pd(iy0,jy0);
474 dz00 = _mm_sub_pd(iz0,jz0);
475 dx10 = _mm_sub_pd(ix1,jx0);
476 dy10 = _mm_sub_pd(iy1,jy0);
477 dz10 = _mm_sub_pd(iz1,jz0);
478 dx20 = _mm_sub_pd(ix2,jx0);
479 dy20 = _mm_sub_pd(iy2,jy0);
480 dz20 = _mm_sub_pd(iz2,jz0);
482 /* Calculate squared distance and things based on it */
483 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
484 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
485 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
487 rinv00 = gmx_mm_invsqrt_pd(rsq00);
488 rinv10 = gmx_mm_invsqrt_pd(rsq10);
489 rinv20 = gmx_mm_invsqrt_pd(rsq20);
491 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
492 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
493 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
495 /* Load parameters for j particles */
496 jq0 = _mm_load_sd(charge+jnrA+0);
497 vdwjidx0A = 2*vdwtype[jnrA+0];
499 fjx0 = _mm_setzero_pd();
500 fjy0 = _mm_setzero_pd();
501 fjz0 = _mm_setzero_pd();
503 /**************************
504 * CALCULATE INTERACTIONS *
505 **************************/
507 if (gmx_mm_any_lt(rsq00,rcutoff2))
510 r00 = _mm_mul_pd(rsq00,rinv00);
512 /* Compute parameters for interactions between i and j atoms */
513 qq00 = _mm_mul_pd(iq0,jq0);
514 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
516 /* EWALD ELECTROSTATICS */
518 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
519 ewrt = _mm_mul_pd(r00,ewtabscale);
520 ewitab = _mm_cvttpd_epi32(ewrt);
522 eweps = _mm_frcz_pd(ewrt);
524 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
526 twoeweps = _mm_add_pd(eweps,eweps);
527 ewitab = _mm_slli_epi32(ewitab,2);
528 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
529 ewtabD = _mm_setzero_pd();
530 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
531 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
532 ewtabFn = _mm_setzero_pd();
533 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
534 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
535 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
536 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
537 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
539 /* LENNARD-JONES DISPERSION/REPULSION */
541 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
542 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
543 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
544 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
545 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
547 d = _mm_sub_pd(r00,rswitch);
548 d = _mm_max_pd(d,_mm_setzero_pd());
549 d2 = _mm_mul_pd(d,d);
550 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
552 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
554 /* Evaluate switch function */
555 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
556 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
557 fvdw = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
558 velec = _mm_mul_pd(velec,sw);
559 vvdw = _mm_mul_pd(vvdw,sw);
560 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
562 /* Update potential sum for this i atom from the interaction with this j atom. */
563 velec = _mm_and_pd(velec,cutoff_mask);
564 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
565 velecsum = _mm_add_pd(velecsum,velec);
566 vvdw = _mm_and_pd(vvdw,cutoff_mask);
567 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
568 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
570 fscal = _mm_add_pd(felec,fvdw);
572 fscal = _mm_and_pd(fscal,cutoff_mask);
574 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
576 /* Update vectorial force */
577 fix0 = _mm_macc_pd(dx00,fscal,fix0);
578 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
579 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
581 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
582 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
583 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
587 /**************************
588 * CALCULATE INTERACTIONS *
589 **************************/
591 if (gmx_mm_any_lt(rsq10,rcutoff2))
594 r10 = _mm_mul_pd(rsq10,rinv10);
596 /* Compute parameters for interactions between i and j atoms */
597 qq10 = _mm_mul_pd(iq1,jq0);
599 /* EWALD ELECTROSTATICS */
601 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
602 ewrt = _mm_mul_pd(r10,ewtabscale);
603 ewitab = _mm_cvttpd_epi32(ewrt);
605 eweps = _mm_frcz_pd(ewrt);
607 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
609 twoeweps = _mm_add_pd(eweps,eweps);
610 ewitab = _mm_slli_epi32(ewitab,2);
611 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
612 ewtabD = _mm_setzero_pd();
613 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
614 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
615 ewtabFn = _mm_setzero_pd();
616 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
617 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
618 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
619 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
620 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
622 d = _mm_sub_pd(r10,rswitch);
623 d = _mm_max_pd(d,_mm_setzero_pd());
624 d2 = _mm_mul_pd(d,d);
625 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
627 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
629 /* Evaluate switch function */
630 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
631 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
632 velec = _mm_mul_pd(velec,sw);
633 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
635 /* Update potential sum for this i atom from the interaction with this j atom. */
636 velec = _mm_and_pd(velec,cutoff_mask);
637 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
638 velecsum = _mm_add_pd(velecsum,velec);
642 fscal = _mm_and_pd(fscal,cutoff_mask);
644 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
646 /* Update vectorial force */
647 fix1 = _mm_macc_pd(dx10,fscal,fix1);
648 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
649 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
651 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
652 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
653 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
657 /**************************
658 * CALCULATE INTERACTIONS *
659 **************************/
661 if (gmx_mm_any_lt(rsq20,rcutoff2))
664 r20 = _mm_mul_pd(rsq20,rinv20);
666 /* Compute parameters for interactions between i and j atoms */
667 qq20 = _mm_mul_pd(iq2,jq0);
669 /* EWALD ELECTROSTATICS */
671 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
672 ewrt = _mm_mul_pd(r20,ewtabscale);
673 ewitab = _mm_cvttpd_epi32(ewrt);
675 eweps = _mm_frcz_pd(ewrt);
677 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
679 twoeweps = _mm_add_pd(eweps,eweps);
680 ewitab = _mm_slli_epi32(ewitab,2);
681 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
682 ewtabD = _mm_setzero_pd();
683 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
684 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
685 ewtabFn = _mm_setzero_pd();
686 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
687 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
688 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
689 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
690 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
692 d = _mm_sub_pd(r20,rswitch);
693 d = _mm_max_pd(d,_mm_setzero_pd());
694 d2 = _mm_mul_pd(d,d);
695 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
697 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
699 /* Evaluate switch function */
700 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
701 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
702 velec = _mm_mul_pd(velec,sw);
703 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
705 /* Update potential sum for this i atom from the interaction with this j atom. */
706 velec = _mm_and_pd(velec,cutoff_mask);
707 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
708 velecsum = _mm_add_pd(velecsum,velec);
712 fscal = _mm_and_pd(fscal,cutoff_mask);
714 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
716 /* Update vectorial force */
717 fix2 = _mm_macc_pd(dx20,fscal,fix2);
718 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
719 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
721 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
722 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
723 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
727 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
729 /* Inner loop uses 225 flops */
732 /* End of innermost loop */
734 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
735 f+i_coord_offset,fshift+i_shift_offset);
738 /* Update potential energies */
739 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
740 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
742 /* Increment number of inner iterations */
743 inneriter += j_index_end - j_index_start;
745 /* Outer loop uses 20 flops */
748 /* Increment number of outer iterations */
751 /* Update outer/inner flops */
753 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*225);
756 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_128_fma_double
757 * Electrostatics interaction: Ewald
758 * VdW interaction: LennardJones
759 * Geometry: Water3-Particle
760 * Calculate force/pot: Force
763 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_128_fma_double
764 (t_nblist * gmx_restrict nlist,
765 rvec * gmx_restrict xx,
766 rvec * gmx_restrict ff,
767 t_forcerec * gmx_restrict fr,
768 t_mdatoms * gmx_restrict mdatoms,
769 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
770 t_nrnb * gmx_restrict nrnb)
772 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
773 * just 0 for non-waters.
774 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
775 * jnr indices corresponding to data put in the four positions in the SIMD register.
777 int i_shift_offset,i_coord_offset,outeriter,inneriter;
778 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
780 int j_coord_offsetA,j_coord_offsetB;
781 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
783 real *shiftvec,*fshift,*x,*f;
784 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
786 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
788 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
790 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
791 int vdwjidx0A,vdwjidx0B;
792 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
793 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
794 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
795 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
796 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
799 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
802 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
803 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
805 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
807 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
808 real rswitch_scalar,d_scalar;
809 __m128d dummy_mask,cutoff_mask;
810 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
811 __m128d one = _mm_set1_pd(1.0);
812 __m128d two = _mm_set1_pd(2.0);
818 jindex = nlist->jindex;
820 shiftidx = nlist->shift;
822 shiftvec = fr->shift_vec[0];
823 fshift = fr->fshift[0];
824 facel = _mm_set1_pd(fr->epsfac);
825 charge = mdatoms->chargeA;
826 nvdwtype = fr->ntype;
828 vdwtype = mdatoms->typeA;
830 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
831 ewtab = fr->ic->tabq_coul_FDV0;
832 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
833 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
835 /* Setup water-specific parameters */
836 inr = nlist->iinr[0];
837 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
838 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
839 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
840 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
842 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
843 rcutoff_scalar = fr->rcoulomb;
844 rcutoff = _mm_set1_pd(rcutoff_scalar);
845 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
847 rswitch_scalar = fr->rcoulomb_switch;
848 rswitch = _mm_set1_pd(rswitch_scalar);
849 /* Setup switch parameters */
850 d_scalar = rcutoff_scalar-rswitch_scalar;
851 d = _mm_set1_pd(d_scalar);
852 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
853 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
854 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
855 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
856 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
857 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
859 /* Avoid stupid compiler warnings */
867 /* Start outer loop over neighborlists */
868 for(iidx=0; iidx<nri; iidx++)
870 /* Load shift vector for this list */
871 i_shift_offset = DIM*shiftidx[iidx];
873 /* Load limits for loop over neighbors */
874 j_index_start = jindex[iidx];
875 j_index_end = jindex[iidx+1];
877 /* Get outer coordinate index */
879 i_coord_offset = DIM*inr;
881 /* Load i particle coords and add shift vector */
882 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
883 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
885 fix0 = _mm_setzero_pd();
886 fiy0 = _mm_setzero_pd();
887 fiz0 = _mm_setzero_pd();
888 fix1 = _mm_setzero_pd();
889 fiy1 = _mm_setzero_pd();
890 fiz1 = _mm_setzero_pd();
891 fix2 = _mm_setzero_pd();
892 fiy2 = _mm_setzero_pd();
893 fiz2 = _mm_setzero_pd();
895 /* Start inner kernel loop */
896 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
899 /* Get j neighbor index, and coordinate index */
902 j_coord_offsetA = DIM*jnrA;
903 j_coord_offsetB = DIM*jnrB;
905 /* load j atom coordinates */
906 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
909 /* Calculate displacement vector */
910 dx00 = _mm_sub_pd(ix0,jx0);
911 dy00 = _mm_sub_pd(iy0,jy0);
912 dz00 = _mm_sub_pd(iz0,jz0);
913 dx10 = _mm_sub_pd(ix1,jx0);
914 dy10 = _mm_sub_pd(iy1,jy0);
915 dz10 = _mm_sub_pd(iz1,jz0);
916 dx20 = _mm_sub_pd(ix2,jx0);
917 dy20 = _mm_sub_pd(iy2,jy0);
918 dz20 = _mm_sub_pd(iz2,jz0);
920 /* Calculate squared distance and things based on it */
921 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
922 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
923 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
925 rinv00 = gmx_mm_invsqrt_pd(rsq00);
926 rinv10 = gmx_mm_invsqrt_pd(rsq10);
927 rinv20 = gmx_mm_invsqrt_pd(rsq20);
929 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
930 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
931 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
933 /* Load parameters for j particles */
934 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
935 vdwjidx0A = 2*vdwtype[jnrA+0];
936 vdwjidx0B = 2*vdwtype[jnrB+0];
938 fjx0 = _mm_setzero_pd();
939 fjy0 = _mm_setzero_pd();
940 fjz0 = _mm_setzero_pd();
942 /**************************
943 * CALCULATE INTERACTIONS *
944 **************************/
946 if (gmx_mm_any_lt(rsq00,rcutoff2))
949 r00 = _mm_mul_pd(rsq00,rinv00);
951 /* Compute parameters for interactions between i and j atoms */
952 qq00 = _mm_mul_pd(iq0,jq0);
953 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
954 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
956 /* EWALD ELECTROSTATICS */
958 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
959 ewrt = _mm_mul_pd(r00,ewtabscale);
960 ewitab = _mm_cvttpd_epi32(ewrt);
962 eweps = _mm_frcz_pd(ewrt);
964 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
966 twoeweps = _mm_add_pd(eweps,eweps);
967 ewitab = _mm_slli_epi32(ewitab,2);
968 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
969 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
970 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
971 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
972 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
973 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
974 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
975 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
976 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
977 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
979 /* LENNARD-JONES DISPERSION/REPULSION */
981 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
982 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
983 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
984 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
985 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
987 d = _mm_sub_pd(r00,rswitch);
988 d = _mm_max_pd(d,_mm_setzero_pd());
989 d2 = _mm_mul_pd(d,d);
990 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
992 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
994 /* Evaluate switch function */
995 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
996 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
997 fvdw = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
998 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1000 fscal = _mm_add_pd(felec,fvdw);
1002 fscal = _mm_and_pd(fscal,cutoff_mask);
1004 /* Update vectorial force */
1005 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1006 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1007 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1009 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1010 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1011 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1015 /**************************
1016 * CALCULATE INTERACTIONS *
1017 **************************/
1019 if (gmx_mm_any_lt(rsq10,rcutoff2))
1022 r10 = _mm_mul_pd(rsq10,rinv10);
1024 /* Compute parameters for interactions between i and j atoms */
1025 qq10 = _mm_mul_pd(iq1,jq0);
1027 /* EWALD ELECTROSTATICS */
1029 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1030 ewrt = _mm_mul_pd(r10,ewtabscale);
1031 ewitab = _mm_cvttpd_epi32(ewrt);
1033 eweps = _mm_frcz_pd(ewrt);
1035 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1037 twoeweps = _mm_add_pd(eweps,eweps);
1038 ewitab = _mm_slli_epi32(ewitab,2);
1039 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1040 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1041 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1042 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1043 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1044 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1045 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1046 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1047 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1048 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1050 d = _mm_sub_pd(r10,rswitch);
1051 d = _mm_max_pd(d,_mm_setzero_pd());
1052 d2 = _mm_mul_pd(d,d);
1053 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1055 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1057 /* Evaluate switch function */
1058 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1059 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1060 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1064 fscal = _mm_and_pd(fscal,cutoff_mask);
1066 /* Update vectorial force */
1067 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1068 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1069 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1071 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1072 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1073 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1077 /**************************
1078 * CALCULATE INTERACTIONS *
1079 **************************/
1081 if (gmx_mm_any_lt(rsq20,rcutoff2))
1084 r20 = _mm_mul_pd(rsq20,rinv20);
1086 /* Compute parameters for interactions between i and j atoms */
1087 qq20 = _mm_mul_pd(iq2,jq0);
1089 /* EWALD ELECTROSTATICS */
1091 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1092 ewrt = _mm_mul_pd(r20,ewtabscale);
1093 ewitab = _mm_cvttpd_epi32(ewrt);
1095 eweps = _mm_frcz_pd(ewrt);
1097 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1099 twoeweps = _mm_add_pd(eweps,eweps);
1100 ewitab = _mm_slli_epi32(ewitab,2);
1101 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1102 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1103 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1104 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1105 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1106 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1107 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1108 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1109 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1110 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1112 d = _mm_sub_pd(r20,rswitch);
1113 d = _mm_max_pd(d,_mm_setzero_pd());
1114 d2 = _mm_mul_pd(d,d);
1115 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1117 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1119 /* Evaluate switch function */
1120 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1121 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1122 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1126 fscal = _mm_and_pd(fscal,cutoff_mask);
1128 /* Update vectorial force */
1129 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1130 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1131 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1133 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1134 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1135 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1139 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1141 /* Inner loop uses 213 flops */
1144 if(jidx<j_index_end)
1148 j_coord_offsetA = DIM*jnrA;
1150 /* load j atom coordinates */
1151 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1154 /* Calculate displacement vector */
1155 dx00 = _mm_sub_pd(ix0,jx0);
1156 dy00 = _mm_sub_pd(iy0,jy0);
1157 dz00 = _mm_sub_pd(iz0,jz0);
1158 dx10 = _mm_sub_pd(ix1,jx0);
1159 dy10 = _mm_sub_pd(iy1,jy0);
1160 dz10 = _mm_sub_pd(iz1,jz0);
1161 dx20 = _mm_sub_pd(ix2,jx0);
1162 dy20 = _mm_sub_pd(iy2,jy0);
1163 dz20 = _mm_sub_pd(iz2,jz0);
1165 /* Calculate squared distance and things based on it */
1166 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1167 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1168 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1170 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1171 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1172 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1174 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1175 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1176 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1178 /* Load parameters for j particles */
1179 jq0 = _mm_load_sd(charge+jnrA+0);
1180 vdwjidx0A = 2*vdwtype[jnrA+0];
1182 fjx0 = _mm_setzero_pd();
1183 fjy0 = _mm_setzero_pd();
1184 fjz0 = _mm_setzero_pd();
1186 /**************************
1187 * CALCULATE INTERACTIONS *
1188 **************************/
1190 if (gmx_mm_any_lt(rsq00,rcutoff2))
1193 r00 = _mm_mul_pd(rsq00,rinv00);
1195 /* Compute parameters for interactions between i and j atoms */
1196 qq00 = _mm_mul_pd(iq0,jq0);
1197 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1199 /* EWALD ELECTROSTATICS */
1201 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1202 ewrt = _mm_mul_pd(r00,ewtabscale);
1203 ewitab = _mm_cvttpd_epi32(ewrt);
1205 eweps = _mm_frcz_pd(ewrt);
1207 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1209 twoeweps = _mm_add_pd(eweps,eweps);
1210 ewitab = _mm_slli_epi32(ewitab,2);
1211 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1212 ewtabD = _mm_setzero_pd();
1213 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1214 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1215 ewtabFn = _mm_setzero_pd();
1216 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1217 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1218 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1219 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
1220 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1222 /* LENNARD-JONES DISPERSION/REPULSION */
1224 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1225 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
1226 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1227 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
1228 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1230 d = _mm_sub_pd(r00,rswitch);
1231 d = _mm_max_pd(d,_mm_setzero_pd());
1232 d2 = _mm_mul_pd(d,d);
1233 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1235 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
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 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
1240 fvdw = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1241 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1243 fscal = _mm_add_pd(felec,fvdw);
1245 fscal = _mm_and_pd(fscal,cutoff_mask);
1247 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1249 /* Update vectorial force */
1250 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1251 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1252 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1254 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1255 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1256 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1260 /**************************
1261 * CALCULATE INTERACTIONS *
1262 **************************/
1264 if (gmx_mm_any_lt(rsq10,rcutoff2))
1267 r10 = _mm_mul_pd(rsq10,rinv10);
1269 /* Compute parameters for interactions between i and j atoms */
1270 qq10 = _mm_mul_pd(iq1,jq0);
1272 /* EWALD ELECTROSTATICS */
1274 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1275 ewrt = _mm_mul_pd(r10,ewtabscale);
1276 ewitab = _mm_cvttpd_epi32(ewrt);
1278 eweps = _mm_frcz_pd(ewrt);
1280 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1282 twoeweps = _mm_add_pd(eweps,eweps);
1283 ewitab = _mm_slli_epi32(ewitab,2);
1284 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1285 ewtabD = _mm_setzero_pd();
1286 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1287 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1288 ewtabFn = _mm_setzero_pd();
1289 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1290 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1291 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1292 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1293 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1295 d = _mm_sub_pd(r10,rswitch);
1296 d = _mm_max_pd(d,_mm_setzero_pd());
1297 d2 = _mm_mul_pd(d,d);
1298 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1300 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1302 /* Evaluate switch function */
1303 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1304 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1305 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1309 fscal = _mm_and_pd(fscal,cutoff_mask);
1311 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1313 /* Update vectorial force */
1314 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1315 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1316 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1318 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1319 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1320 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1324 /**************************
1325 * CALCULATE INTERACTIONS *
1326 **************************/
1328 if (gmx_mm_any_lt(rsq20,rcutoff2))
1331 r20 = _mm_mul_pd(rsq20,rinv20);
1333 /* Compute parameters for interactions between i and j atoms */
1334 qq20 = _mm_mul_pd(iq2,jq0);
1336 /* EWALD ELECTROSTATICS */
1338 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1339 ewrt = _mm_mul_pd(r20,ewtabscale);
1340 ewitab = _mm_cvttpd_epi32(ewrt);
1342 eweps = _mm_frcz_pd(ewrt);
1344 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1346 twoeweps = _mm_add_pd(eweps,eweps);
1347 ewitab = _mm_slli_epi32(ewitab,2);
1348 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1349 ewtabD = _mm_setzero_pd();
1350 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1351 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1352 ewtabFn = _mm_setzero_pd();
1353 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1354 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1355 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1356 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1357 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1359 d = _mm_sub_pd(r20,rswitch);
1360 d = _mm_max_pd(d,_mm_setzero_pd());
1361 d2 = _mm_mul_pd(d,d);
1362 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1364 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1366 /* Evaluate switch function */
1367 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1368 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1369 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1373 fscal = _mm_and_pd(fscal,cutoff_mask);
1375 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1377 /* Update vectorial force */
1378 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1379 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1380 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1382 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1383 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1384 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1388 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1390 /* Inner loop uses 213 flops */
1393 /* End of innermost loop */
1395 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1396 f+i_coord_offset,fshift+i_shift_offset);
1398 /* Increment number of inner iterations */
1399 inneriter += j_index_end - j_index_start;
1401 /* Outer loop uses 18 flops */
1404 /* Increment number of outer iterations */
1407 /* Update outer/inner flops */
1409 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*213);