src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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.
   8  *
   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.
  13  *
  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.
  18  *
  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.
  23  *
  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.
  31  *
  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.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_VF_avx_256_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  95     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  97     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 100     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 101     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 102     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 105     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     int              nvdwtype;
 110     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 111     int              *vdwtype;
 112     real             *vdwparam;
 113     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 114     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 115     __m256d          dummy_mask,cutoff_mask;
 116     __m128           tmpmask0,tmpmask1;
 117     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 118     __m256d          one     = _mm256_set1_pd(1.0);
 119     __m256d          two     = _mm256_set1_pd(2.0);
 120     x                = xx[0];
 121     f                = ff[0];
 122
 123     nri              = nlist->nri;
 124     iinr             = nlist->iinr;
 125     jindex           = nlist->jindex;
 126     jjnr             = nlist->jjnr;
 127     shiftidx         = nlist->shift;
 128     gid              = nlist->gid;
 129     shiftvec         = fr->shift_vec[0];
 130     fshift           = fr->fshift[0];
 131     facel            = _mm256_set1_pd(fr->epsfac);
 132     charge           = mdatoms->chargeA;
 133     krf              = _mm256_set1_pd(fr->ic->k_rf);
 134     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 135     crf              = _mm256_set1_pd(fr->ic->c_rf);
 136     nvdwtype         = fr->ntype;
 137     vdwparam         = fr->nbfp;
 138     vdwtype          = mdatoms->typeA;
 139
 140     /* Setup water-specific parameters */
 141     inr              = nlist->iinr[0];
 142     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 143     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 144     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 145     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 146
 147     jq0              = _mm256_set1_pd(charge[inr+0]);
 148     jq1              = _mm256_set1_pd(charge[inr+1]);
 149     jq2              = _mm256_set1_pd(charge[inr+2]);
 150     vdwjidx0A        = 2*vdwtype[inr+0];
 151     qq00             = _mm256_mul_pd(iq0,jq0);
 152     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 153     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 154     qq01             = _mm256_mul_pd(iq0,jq1);
 155     qq02             = _mm256_mul_pd(iq0,jq2);
 156     qq10             = _mm256_mul_pd(iq1,jq0);
 157     qq11             = _mm256_mul_pd(iq1,jq1);
 158     qq12             = _mm256_mul_pd(iq1,jq2);
 159     qq20             = _mm256_mul_pd(iq2,jq0);
 160     qq21             = _mm256_mul_pd(iq2,jq1);
 161     qq22             = _mm256_mul_pd(iq2,jq2);
 162
 163     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 164     rcutoff_scalar   = fr->rcoulomb;
 165     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 166     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 167
 168     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 169     rvdw             = _mm256_set1_pd(fr->rvdw);
 170
 171     /* Avoid stupid compiler warnings */
 172     jnrA = jnrB = jnrC = jnrD = 0;
 173     j_coord_offsetA = 0;
 174     j_coord_offsetB = 0;
 175     j_coord_offsetC = 0;
 176     j_coord_offsetD = 0;
 177
 178     outeriter        = 0;
 179     inneriter        = 0;
 180
 181     for(iidx=0;iidx<4*DIM;iidx++)
 182     {
 183         scratch[iidx] = 0.0;
 184     }
 185
 186     /* Start outer loop over neighborlists */
 187     for(iidx=0; iidx<nri; iidx++)
 188     {
 189         /* Load shift vector for this list */
 190         i_shift_offset   = DIM*shiftidx[iidx];
 191
 192         /* Load limits for loop over neighbors */
 193         j_index_start    = jindex[iidx];
 194         j_index_end      = jindex[iidx+1];
 195
 196         /* Get outer coordinate index */
 197         inr              = iinr[iidx];
 198         i_coord_offset   = DIM*inr;
 199
 200         /* Load i particle coords and add shift vector */
 201         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 202                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 203
 204         fix0             = _mm256_setzero_pd();
 205         fiy0             = _mm256_setzero_pd();
 206         fiz0             = _mm256_setzero_pd();
 207         fix1             = _mm256_setzero_pd();
 208         fiy1             = _mm256_setzero_pd();
 209         fiz1             = _mm256_setzero_pd();
 210         fix2             = _mm256_setzero_pd();
 211         fiy2             = _mm256_setzero_pd();
 212         fiz2             = _mm256_setzero_pd();
 213
 214         /* Reset potential sums */
 215         velecsum         = _mm256_setzero_pd();
 216         vvdwsum          = _mm256_setzero_pd();
 217
 218         /* Start inner kernel loop */
 219         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 220         {
 221
 222             /* Get j neighbor index, and coordinate index */
 223             jnrA             = jjnr[jidx];
 224             jnrB             = jjnr[jidx+1];
 225             jnrC             = jjnr[jidx+2];
 226             jnrD             = jjnr[jidx+3];
 227             j_coord_offsetA  = DIM*jnrA;
 228             j_coord_offsetB  = DIM*jnrB;
 229             j_coord_offsetC  = DIM*jnrC;
 230             j_coord_offsetD  = DIM*jnrD;
 231
 232             /* load j atom coordinates */
 233             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 234                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 235                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 236
 237             /* Calculate displacement vector */
 238             dx00             = _mm256_sub_pd(ix0,jx0);
 239             dy00             = _mm256_sub_pd(iy0,jy0);
 240             dz00             = _mm256_sub_pd(iz0,jz0);
 241             dx01             = _mm256_sub_pd(ix0,jx1);
 242             dy01             = _mm256_sub_pd(iy0,jy1);
 243             dz01             = _mm256_sub_pd(iz0,jz1);
 244             dx02             = _mm256_sub_pd(ix0,jx2);
 245             dy02             = _mm256_sub_pd(iy0,jy2);
 246             dz02             = _mm256_sub_pd(iz0,jz2);
 247             dx10             = _mm256_sub_pd(ix1,jx0);
 248             dy10             = _mm256_sub_pd(iy1,jy0);
 249             dz10             = _mm256_sub_pd(iz1,jz0);
 250             dx11             = _mm256_sub_pd(ix1,jx1);
 251             dy11             = _mm256_sub_pd(iy1,jy1);
 252             dz11             = _mm256_sub_pd(iz1,jz1);
 253             dx12             = _mm256_sub_pd(ix1,jx2);
 254             dy12             = _mm256_sub_pd(iy1,jy2);
 255             dz12             = _mm256_sub_pd(iz1,jz2);
 256             dx20             = _mm256_sub_pd(ix2,jx0);
 257             dy20             = _mm256_sub_pd(iy2,jy0);
 258             dz20             = _mm256_sub_pd(iz2,jz0);
 259             dx21             = _mm256_sub_pd(ix2,jx1);
 260             dy21             = _mm256_sub_pd(iy2,jy1);
 261             dz21             = _mm256_sub_pd(iz2,jz1);
 262             dx22             = _mm256_sub_pd(ix2,jx2);
 263             dy22             = _mm256_sub_pd(iy2,jy2);
 264             dz22             = _mm256_sub_pd(iz2,jz2);
 265
 266             /* Calculate squared distance and things based on it */
 267             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 268             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 269             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 270             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 271             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 272             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 273             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 274             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 275             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 276
 277             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 278             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 279             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 280             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 281             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 282             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 283             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 284             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 285             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 286
 287             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 288             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 289             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 290             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 291             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 292             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 293             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 294             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 295             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 296
 297             fjx0             = _mm256_setzero_pd();
 298             fjy0             = _mm256_setzero_pd();
 299             fjz0             = _mm256_setzero_pd();
 300             fjx1             = _mm256_setzero_pd();
 301             fjy1             = _mm256_setzero_pd();
 302             fjz1             = _mm256_setzero_pd();
 303             fjx2             = _mm256_setzero_pd();
 304             fjy2             = _mm256_setzero_pd();
 305             fjz2             = _mm256_setzero_pd();
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 312             {
 313
 314             /* REACTION-FIELD ELECTROSTATICS */
 315             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 316             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 317
 318             /* LENNARD-JONES DISPERSION/REPULSION */
 319
 320             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 321             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 322             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 323             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 324                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 325             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 326
 327             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 328
 329             /* Update potential sum for this i atom from the interaction with this j atom. */
 330             velec            = _mm256_and_pd(velec,cutoff_mask);
 331             velecsum         = _mm256_add_pd(velecsum,velec);
 332             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 333             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 334
 335             fscal            = _mm256_add_pd(felec,fvdw);
 336
 337             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 338
 339             /* Calculate temporary vectorial force */
 340             tx               = _mm256_mul_pd(fscal,dx00);
 341             ty               = _mm256_mul_pd(fscal,dy00);
 342             tz               = _mm256_mul_pd(fscal,dz00);
 343
 344             /* Update vectorial force */
 345             fix0             = _mm256_add_pd(fix0,tx);
 346             fiy0             = _mm256_add_pd(fiy0,ty);
 347             fiz0             = _mm256_add_pd(fiz0,tz);
 348
 349             fjx0             = _mm256_add_pd(fjx0,tx);
 350             fjy0             = _mm256_add_pd(fjy0,ty);
 351             fjz0             = _mm256_add_pd(fjz0,tz);
 352
 353             }
 354
 355             /**************************
 356              * CALCULATE INTERACTIONS *
 357              **************************/
 358
 359             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 360             {
 361
 362             /* REACTION-FIELD ELECTROSTATICS */
 363             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
 364             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
 365
 366             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velec            = _mm256_and_pd(velec,cutoff_mask);
 370             velecsum         = _mm256_add_pd(velecsum,velec);
 371
 372             fscal            = felec;
 373
 374             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm256_mul_pd(fscal,dx01);
 378             ty               = _mm256_mul_pd(fscal,dy01);
 379             tz               = _mm256_mul_pd(fscal,dz01);
 380
 381             /* Update vectorial force */
 382             fix0             = _mm256_add_pd(fix0,tx);
 383             fiy0             = _mm256_add_pd(fiy0,ty);
 384             fiz0             = _mm256_add_pd(fiz0,tz);
 385
 386             fjx1             = _mm256_add_pd(fjx1,tx);
 387             fjy1             = _mm256_add_pd(fjy1,ty);
 388             fjz1             = _mm256_add_pd(fjz1,tz);
 389
 390             }
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 397             {
 398
 399             /* REACTION-FIELD ELECTROSTATICS */
 400             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
 401             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
 402
 403             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
 404
 405             /* Update potential sum for this i atom from the interaction with this j atom. */
 406             velec            = _mm256_and_pd(velec,cutoff_mask);
 407             velecsum         = _mm256_add_pd(velecsum,velec);
 408
 409             fscal            = felec;
 410
 411             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 412
 413             /* Calculate temporary vectorial force */
 414             tx               = _mm256_mul_pd(fscal,dx02);
 415             ty               = _mm256_mul_pd(fscal,dy02);
 416             tz               = _mm256_mul_pd(fscal,dz02);
 417
 418             /* Update vectorial force */
 419             fix0             = _mm256_add_pd(fix0,tx);
 420             fiy0             = _mm256_add_pd(fiy0,ty);
 421             fiz0             = _mm256_add_pd(fiz0,tz);
 422
 423             fjx2             = _mm256_add_pd(fjx2,tx);
 424             fjy2             = _mm256_add_pd(fjy2,ty);
 425             fjz2             = _mm256_add_pd(fjz2,tz);
 426
 427             }
 428
 429             /**************************
 430              * CALCULATE INTERACTIONS *
 431              **************************/
 432
 433             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 434             {
 435
 436             /* REACTION-FIELD ELECTROSTATICS */
 437             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 438             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 439
 440             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 441
 442             /* Update potential sum for this i atom from the interaction with this j atom. */
 443             velec            = _mm256_and_pd(velec,cutoff_mask);
 444             velecsum         = _mm256_add_pd(velecsum,velec);
 445
 446             fscal            = felec;
 447
 448             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 449
 450             /* Calculate temporary vectorial force */
 451             tx               = _mm256_mul_pd(fscal,dx10);
 452             ty               = _mm256_mul_pd(fscal,dy10);
 453             tz               = _mm256_mul_pd(fscal,dz10);
 454
 455             /* Update vectorial force */
 456             fix1             = _mm256_add_pd(fix1,tx);
 457             fiy1             = _mm256_add_pd(fiy1,ty);
 458             fiz1             = _mm256_add_pd(fiz1,tz);
 459
 460             fjx0             = _mm256_add_pd(fjx0,tx);
 461             fjy0             = _mm256_add_pd(fjy0,ty);
 462             fjz0             = _mm256_add_pd(fjz0,tz);
 463
 464             }
 465
 466             /**************************
 467              * CALCULATE INTERACTIONS *
 468              **************************/
 469
 470             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 471             {
 472
 473             /* REACTION-FIELD ELECTROSTATICS */
 474             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 475             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 476
 477             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 478
 479             /* Update potential sum for this i atom from the interaction with this j atom. */
 480             velec            = _mm256_and_pd(velec,cutoff_mask);
 481             velecsum         = _mm256_add_pd(velecsum,velec);
 482
 483             fscal            = felec;
 484
 485             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 486
 487             /* Calculate temporary vectorial force */
 488             tx               = _mm256_mul_pd(fscal,dx11);
 489             ty               = _mm256_mul_pd(fscal,dy11);
 490             tz               = _mm256_mul_pd(fscal,dz11);
 491
 492             /* Update vectorial force */
 493             fix1             = _mm256_add_pd(fix1,tx);
 494             fiy1             = _mm256_add_pd(fiy1,ty);
 495             fiz1             = _mm256_add_pd(fiz1,tz);
 496
 497             fjx1             = _mm256_add_pd(fjx1,tx);
 498             fjy1             = _mm256_add_pd(fjy1,ty);
 499             fjz1             = _mm256_add_pd(fjz1,tz);
 500
 501             }
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 508             {
 509
 510             /* REACTION-FIELD ELECTROSTATICS */
 511             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 512             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 513
 514             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velec            = _mm256_and_pd(velec,cutoff_mask);
 518             velecsum         = _mm256_add_pd(velecsum,velec);
 519
 520             fscal            = felec;
 521
 522             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 523
 524             /* Calculate temporary vectorial force */
 525             tx               = _mm256_mul_pd(fscal,dx12);
 526             ty               = _mm256_mul_pd(fscal,dy12);
 527             tz               = _mm256_mul_pd(fscal,dz12);
 528
 529             /* Update vectorial force */
 530             fix1             = _mm256_add_pd(fix1,tx);
 531             fiy1             = _mm256_add_pd(fiy1,ty);
 532             fiz1             = _mm256_add_pd(fiz1,tz);
 533
 534             fjx2             = _mm256_add_pd(fjx2,tx);
 535             fjy2             = _mm256_add_pd(fjy2,ty);
 536             fjz2             = _mm256_add_pd(fjz2,tz);
 537
 538             }
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 545             {
 546
 547             /* REACTION-FIELD ELECTROSTATICS */
 548             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 549             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 550
 551             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 552
 553             /* Update potential sum for this i atom from the interaction with this j atom. */
 554             velec            = _mm256_and_pd(velec,cutoff_mask);
 555             velecsum         = _mm256_add_pd(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 560
 561             /* Calculate temporary vectorial force */
 562             tx               = _mm256_mul_pd(fscal,dx20);
 563             ty               = _mm256_mul_pd(fscal,dy20);
 564             tz               = _mm256_mul_pd(fscal,dz20);
 565
 566             /* Update vectorial force */
 567             fix2             = _mm256_add_pd(fix2,tx);
 568             fiy2             = _mm256_add_pd(fiy2,ty);
 569             fiz2             = _mm256_add_pd(fiz2,tz);
 570
 571             fjx0             = _mm256_add_pd(fjx0,tx);
 572             fjy0             = _mm256_add_pd(fjy0,ty);
 573             fjz0             = _mm256_add_pd(fjz0,tz);
 574
 575             }
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 582             {
 583
 584             /* REACTION-FIELD ELECTROSTATICS */
 585             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 586             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 587
 588             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
 589
 590             /* Update potential sum for this i atom from the interaction with this j atom. */
 591             velec            = _mm256_and_pd(velec,cutoff_mask);
 592             velecsum         = _mm256_add_pd(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm256_mul_pd(fscal,dx21);
 600             ty               = _mm256_mul_pd(fscal,dy21);
 601             tz               = _mm256_mul_pd(fscal,dz21);
 602
 603             /* Update vectorial force */
 604             fix2             = _mm256_add_pd(fix2,tx);
 605             fiy2             = _mm256_add_pd(fiy2,ty);
 606             fiz2             = _mm256_add_pd(fiz2,tz);
 607
 608             fjx1             = _mm256_add_pd(fjx1,tx);
 609             fjy1             = _mm256_add_pd(fjy1,ty);
 610             fjz1             = _mm256_add_pd(fjz1,tz);
 611
 612             }
 613
 614             /**************************
 615              * CALCULATE INTERACTIONS *
 616              **************************/
 617
 618             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 619             {
 620
 621             /* REACTION-FIELD ELECTROSTATICS */
 622             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 623             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 624
 625             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
 626
 627             /* Update potential sum for this i atom from the interaction with this j atom. */
 628             velec            = _mm256_and_pd(velec,cutoff_mask);
 629             velecsum         = _mm256_add_pd(velecsum,velec);
 630
 631             fscal            = felec;
 632
 633             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 634
 635             /* Calculate temporary vectorial force */
 636             tx               = _mm256_mul_pd(fscal,dx22);
 637             ty               = _mm256_mul_pd(fscal,dy22);
 638             tz               = _mm256_mul_pd(fscal,dz22);
 639
 640             /* Update vectorial force */
 641             fix2             = _mm256_add_pd(fix2,tx);
 642             fiy2             = _mm256_add_pd(fiy2,ty);
 643             fiz2             = _mm256_add_pd(fiz2,tz);
 644
 645             fjx2             = _mm256_add_pd(fjx2,tx);
 646             fjy2             = _mm256_add_pd(fjy2,ty);
 647             fjz2             = _mm256_add_pd(fjz2,tz);
 648
 649             }
 650
 651             fjptrA             = f+j_coord_offsetA;
 652             fjptrB             = f+j_coord_offsetB;
 653             fjptrC             = f+j_coord_offsetC;
 654             fjptrD             = f+j_coord_offsetD;
 655
 656             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 657                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 658
 659             /* Inner loop uses 342 flops */
 660         }
 661
 662         if(jidx<j_index_end)
 663         {
 664
 665             /* Get j neighbor index, and coordinate index */
 666             jnrlistA         = jjnr[jidx];
 667             jnrlistB         = jjnr[jidx+1];
 668             jnrlistC         = jjnr[jidx+2];
 669             jnrlistD         = jjnr[jidx+3];
 670             /* Sign of each element will be negative for non-real atoms.
 671              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 672              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 673              */
 674             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 675
 676             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 677             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 678             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 679
 680             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 681             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 682             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 683             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 684             j_coord_offsetA  = DIM*jnrA;
 685             j_coord_offsetB  = DIM*jnrB;
 686             j_coord_offsetC  = DIM*jnrC;
 687             j_coord_offsetD  = DIM*jnrD;
 688
 689             /* load j atom coordinates */
 690             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 691                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 692                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 693
 694             /* Calculate displacement vector */
 695             dx00             = _mm256_sub_pd(ix0,jx0);
 696             dy00             = _mm256_sub_pd(iy0,jy0);
 697             dz00             = _mm256_sub_pd(iz0,jz0);
 698             dx01             = _mm256_sub_pd(ix0,jx1);
 699             dy01             = _mm256_sub_pd(iy0,jy1);
 700             dz01             = _mm256_sub_pd(iz0,jz1);
 701             dx02             = _mm256_sub_pd(ix0,jx2);
 702             dy02             = _mm256_sub_pd(iy0,jy2);
 703             dz02             = _mm256_sub_pd(iz0,jz2);
 704             dx10             = _mm256_sub_pd(ix1,jx0);
 705             dy10             = _mm256_sub_pd(iy1,jy0);
 706             dz10             = _mm256_sub_pd(iz1,jz0);
 707             dx11             = _mm256_sub_pd(ix1,jx1);
 708             dy11             = _mm256_sub_pd(iy1,jy1);
 709             dz11             = _mm256_sub_pd(iz1,jz1);
 710             dx12             = _mm256_sub_pd(ix1,jx2);
 711             dy12             = _mm256_sub_pd(iy1,jy2);
 712             dz12             = _mm256_sub_pd(iz1,jz2);
 713             dx20             = _mm256_sub_pd(ix2,jx0);
 714             dy20             = _mm256_sub_pd(iy2,jy0);
 715             dz20             = _mm256_sub_pd(iz2,jz0);
 716             dx21             = _mm256_sub_pd(ix2,jx1);
 717             dy21             = _mm256_sub_pd(iy2,jy1);
 718             dz21             = _mm256_sub_pd(iz2,jz1);
 719             dx22             = _mm256_sub_pd(ix2,jx2);
 720             dy22             = _mm256_sub_pd(iy2,jy2);
 721             dz22             = _mm256_sub_pd(iz2,jz2);
 722
 723             /* Calculate squared distance and things based on it */
 724             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 725             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 726             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 727             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 728             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 729             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 730             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 731             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 732             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 733
 734             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 735             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 736             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 737             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 738             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 739             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 740             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 741             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 742             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 743
 744             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 745             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 746             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 747             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 748             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 749             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 750             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 751             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 752             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 753
 754             fjx0             = _mm256_setzero_pd();
 755             fjy0             = _mm256_setzero_pd();
 756             fjz0             = _mm256_setzero_pd();
 757             fjx1             = _mm256_setzero_pd();
 758             fjy1             = _mm256_setzero_pd();
 759             fjz1             = _mm256_setzero_pd();
 760             fjx2             = _mm256_setzero_pd();
 761             fjy2             = _mm256_setzero_pd();
 762             fjz2             = _mm256_setzero_pd();
 763
 764             /**************************
 765              * CALCULATE INTERACTIONS *
 766              **************************/
 767
 768             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 769             {
 770
 771             /* REACTION-FIELD ELECTROSTATICS */
 772             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 773             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 774
 775             /* LENNARD-JONES DISPERSION/REPULSION */
 776
 777             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 778             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 779             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 780             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 781                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 782             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 783
 784             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 785
 786             /* Update potential sum for this i atom from the interaction with this j atom. */
 787             velec            = _mm256_and_pd(velec,cutoff_mask);
 788             velec            = _mm256_andnot_pd(dummy_mask,velec);
 789             velecsum         = _mm256_add_pd(velecsum,velec);
 790             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 791             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 792             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 793
 794             fscal            = _mm256_add_pd(felec,fvdw);
 795
 796             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 797
 798             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 799
 800             /* Calculate temporary vectorial force */
 801             tx               = _mm256_mul_pd(fscal,dx00);
 802             ty               = _mm256_mul_pd(fscal,dy00);
 803             tz               = _mm256_mul_pd(fscal,dz00);
 804
 805             /* Update vectorial force */
 806             fix0             = _mm256_add_pd(fix0,tx);
 807             fiy0             = _mm256_add_pd(fiy0,ty);
 808             fiz0             = _mm256_add_pd(fiz0,tz);
 809
 810             fjx0             = _mm256_add_pd(fjx0,tx);
 811             fjy0             = _mm256_add_pd(fjy0,ty);
 812             fjz0             = _mm256_add_pd(fjz0,tz);
 813
 814             }
 815
 816             /**************************
 817              * CALCULATE INTERACTIONS *
 818              **************************/
 819
 820             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 821             {
 822
 823             /* REACTION-FIELD ELECTROSTATICS */
 824             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
 825             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
 826
 827             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
 828
 829             /* Update potential sum for this i atom from the interaction with this j atom. */
 830             velec            = _mm256_and_pd(velec,cutoff_mask);
 831             velec            = _mm256_andnot_pd(dummy_mask,velec);
 832             velecsum         = _mm256_add_pd(velecsum,velec);
 833
 834             fscal            = felec;
 835
 836             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 837
 838             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 839
 840             /* Calculate temporary vectorial force */
 841             tx               = _mm256_mul_pd(fscal,dx01);
 842             ty               = _mm256_mul_pd(fscal,dy01);
 843             tz               = _mm256_mul_pd(fscal,dz01);
 844
 845             /* Update vectorial force */
 846             fix0             = _mm256_add_pd(fix0,tx);
 847             fiy0             = _mm256_add_pd(fiy0,ty);
 848             fiz0             = _mm256_add_pd(fiz0,tz);
 849
 850             fjx1             = _mm256_add_pd(fjx1,tx);
 851             fjy1             = _mm256_add_pd(fjy1,ty);
 852             fjz1             = _mm256_add_pd(fjz1,tz);
 853
 854             }
 855
 856             /**************************
 857              * CALCULATE INTERACTIONS *
 858              **************************/
 859
 860             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 861             {
 862
 863             /* REACTION-FIELD ELECTROSTATICS */
 864             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
 865             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
 866
 867             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
 868
 869             /* Update potential sum for this i atom from the interaction with this j atom. */
 870             velec            = _mm256_and_pd(velec,cutoff_mask);
 871             velec            = _mm256_andnot_pd(dummy_mask,velec);
 872             velecsum         = _mm256_add_pd(velecsum,velec);
 873
 874             fscal            = felec;
 875
 876             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 877
 878             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 879
 880             /* Calculate temporary vectorial force */
 881             tx               = _mm256_mul_pd(fscal,dx02);
 882             ty               = _mm256_mul_pd(fscal,dy02);
 883             tz               = _mm256_mul_pd(fscal,dz02);
 884
 885             /* Update vectorial force */
 886             fix0             = _mm256_add_pd(fix0,tx);
 887             fiy0             = _mm256_add_pd(fiy0,ty);
 888             fiz0             = _mm256_add_pd(fiz0,tz);
 889
 890             fjx2             = _mm256_add_pd(fjx2,tx);
 891             fjy2             = _mm256_add_pd(fjy2,ty);
 892             fjz2             = _mm256_add_pd(fjz2,tz);
 893
 894             }
 895
 896             /**************************
 897              * CALCULATE INTERACTIONS *
 898              **************************/
 899
 900             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 901             {
 902
 903             /* REACTION-FIELD ELECTROSTATICS */
 904             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 905             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 906
 907             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 908
 909             /* Update potential sum for this i atom from the interaction with this j atom. */
 910             velec            = _mm256_and_pd(velec,cutoff_mask);
 911             velec            = _mm256_andnot_pd(dummy_mask,velec);
 912             velecsum         = _mm256_add_pd(velecsum,velec);
 913
 914             fscal            = felec;
 915
 916             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 917
 918             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 919
 920             /* Calculate temporary vectorial force */
 921             tx               = _mm256_mul_pd(fscal,dx10);
 922             ty               = _mm256_mul_pd(fscal,dy10);
 923             tz               = _mm256_mul_pd(fscal,dz10);
 924
 925             /* Update vectorial force */
 926             fix1             = _mm256_add_pd(fix1,tx);
 927             fiy1             = _mm256_add_pd(fiy1,ty);
 928             fiz1             = _mm256_add_pd(fiz1,tz);
 929
 930             fjx0             = _mm256_add_pd(fjx0,tx);
 931             fjy0             = _mm256_add_pd(fjy0,ty);
 932             fjz0             = _mm256_add_pd(fjz0,tz);
 933
 934             }
 935
 936             /**************************
 937              * CALCULATE INTERACTIONS *
 938              **************************/
 939
 940             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 941             {
 942
 943             /* REACTION-FIELD ELECTROSTATICS */
 944             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 945             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 946
 947             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 948
 949             /* Update potential sum for this i atom from the interaction with this j atom. */
 950             velec            = _mm256_and_pd(velec,cutoff_mask);
 951             velec            = _mm256_andnot_pd(dummy_mask,velec);
 952             velecsum         = _mm256_add_pd(velecsum,velec);
 953
 954             fscal            = felec;
 955
 956             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 957
 958             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 959
 960             /* Calculate temporary vectorial force */
 961             tx               = _mm256_mul_pd(fscal,dx11);
 962             ty               = _mm256_mul_pd(fscal,dy11);
 963             tz               = _mm256_mul_pd(fscal,dz11);
 964
 965             /* Update vectorial force */
 966             fix1             = _mm256_add_pd(fix1,tx);
 967             fiy1             = _mm256_add_pd(fiy1,ty);
 968             fiz1             = _mm256_add_pd(fiz1,tz);
 969
 970             fjx1             = _mm256_add_pd(fjx1,tx);
 971             fjy1             = _mm256_add_pd(fjy1,ty);
 972             fjz1             = _mm256_add_pd(fjz1,tz);
 973
 974             }
 975
 976             /**************************
 977              * CALCULATE INTERACTIONS *
 978              **************************/
 979
 980             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 981             {
 982
 983             /* REACTION-FIELD ELECTROSTATICS */
 984             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 985             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 986
 987             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 988
 989             /* Update potential sum for this i atom from the interaction with this j atom. */
 990             velec            = _mm256_and_pd(velec,cutoff_mask);
 991             velec            = _mm256_andnot_pd(dummy_mask,velec);
 992             velecsum         = _mm256_add_pd(velecsum,velec);
 993
 994             fscal            = felec;
 995
 996             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 997
 998             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 999
1000             /* Calculate temporary vectorial force */
1001             tx               = _mm256_mul_pd(fscal,dx12);
1002             ty               = _mm256_mul_pd(fscal,dy12);
1003             tz               = _mm256_mul_pd(fscal,dz12);
1004
1005             /* Update vectorial force */
1006             fix1             = _mm256_add_pd(fix1,tx);
1007             fiy1             = _mm256_add_pd(fiy1,ty);
1008             fiz1             = _mm256_add_pd(fiz1,tz);
1009
1010             fjx2             = _mm256_add_pd(fjx2,tx);
1011             fjy2             = _mm256_add_pd(fjy2,ty);
1012             fjz2             = _mm256_add_pd(fjz2,tz);
1013
1014             }
1015
1016             /**************************
1017              * CALCULATE INTERACTIONS *
1018              **************************/
1019
1020             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1021             {
1022
1023             /* REACTION-FIELD ELECTROSTATICS */
1024             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
1025             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1026
1027             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1028
1029             /* Update potential sum for this i atom from the interaction with this j atom. */
1030             velec            = _mm256_and_pd(velec,cutoff_mask);
1031             velec            = _mm256_andnot_pd(dummy_mask,velec);
1032             velecsum         = _mm256_add_pd(velecsum,velec);
1033
1034             fscal            = felec;
1035
1036             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1037
1038             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1039
1040             /* Calculate temporary vectorial force */
1041             tx               = _mm256_mul_pd(fscal,dx20);
1042             ty               = _mm256_mul_pd(fscal,dy20);
1043             tz               = _mm256_mul_pd(fscal,dz20);
1044
1045             /* Update vectorial force */
1046             fix2             = _mm256_add_pd(fix2,tx);
1047             fiy2             = _mm256_add_pd(fiy2,ty);
1048             fiz2             = _mm256_add_pd(fiz2,tz);
1049
1050             fjx0             = _mm256_add_pd(fjx0,tx);
1051             fjy0             = _mm256_add_pd(fjy0,ty);
1052             fjz0             = _mm256_add_pd(fjz0,tz);
1053
1054             }
1055
1056             /**************************
1057              * CALCULATE INTERACTIONS *
1058              **************************/
1059
1060             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1061             {
1062
1063             /* REACTION-FIELD ELECTROSTATICS */
1064             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
1065             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1066
1067             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1068
1069             /* Update potential sum for this i atom from the interaction with this j atom. */
1070             velec            = _mm256_and_pd(velec,cutoff_mask);
1071             velec            = _mm256_andnot_pd(dummy_mask,velec);
1072             velecsum         = _mm256_add_pd(velecsum,velec);
1073
1074             fscal            = felec;
1075
1076             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1077
1078             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1079
1080             /* Calculate temporary vectorial force */
1081             tx               = _mm256_mul_pd(fscal,dx21);
1082             ty               = _mm256_mul_pd(fscal,dy21);
1083             tz               = _mm256_mul_pd(fscal,dz21);
1084
1085             /* Update vectorial force */
1086             fix2             = _mm256_add_pd(fix2,tx);
1087             fiy2             = _mm256_add_pd(fiy2,ty);
1088             fiz2             = _mm256_add_pd(fiz2,tz);
1089
1090             fjx1             = _mm256_add_pd(fjx1,tx);
1091             fjy1             = _mm256_add_pd(fjy1,ty);
1092             fjz1             = _mm256_add_pd(fjz1,tz);
1093
1094             }
1095
1096             /**************************
1097              * CALCULATE INTERACTIONS *
1098              **************************/
1099
1100             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1101             {
1102
1103             /* REACTION-FIELD ELECTROSTATICS */
1104             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
1105             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1106
1107             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1108
1109             /* Update potential sum for this i atom from the interaction with this j atom. */
1110             velec            = _mm256_and_pd(velec,cutoff_mask);
1111             velec            = _mm256_andnot_pd(dummy_mask,velec);
1112             velecsum         = _mm256_add_pd(velecsum,velec);
1113
1114             fscal            = felec;
1115
1116             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1117
1118             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1119
1120             /* Calculate temporary vectorial force */
1121             tx               = _mm256_mul_pd(fscal,dx22);
1122             ty               = _mm256_mul_pd(fscal,dy22);
1123             tz               = _mm256_mul_pd(fscal,dz22);
1124
1125             /* Update vectorial force */
1126             fix2             = _mm256_add_pd(fix2,tx);
1127             fiy2             = _mm256_add_pd(fiy2,ty);
1128             fiz2             = _mm256_add_pd(fiz2,tz);
1129
1130             fjx2             = _mm256_add_pd(fjx2,tx);
1131             fjy2             = _mm256_add_pd(fjy2,ty);
1132             fjz2             = _mm256_add_pd(fjz2,tz);
1133
1134             }
1135
1136             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1137             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1138             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1139             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1140
1141             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1142                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1143
1144             /* Inner loop uses 342 flops */
1145         }
1146
1147         /* End of innermost loop */
1148
1149         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1150                                                  f+i_coord_offset,fshift+i_shift_offset);
1151
1152         ggid                        = gid[iidx];
1153         /* Update potential energies */
1154         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1155         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1156
1157         /* Increment number of inner iterations */
1158         inneriter                  += j_index_end - j_index_start;
1159
1160         /* Outer loop uses 20 flops */
1161     }
1162
1163     /* Increment number of outer iterations */
1164     outeriter        += nri;
1165
1166     /* Update outer/inner flops */
1167
1168     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*342);
1169 }
1170 /*
1171  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_F_avx_256_double
1172  * Electrostatics interaction: ReactionField
1173  * VdW interaction:            LennardJones
1174  * Geometry:                   Water3-Water3
1175  * Calculate force/pot:        Force
1176  */
1177 void
1178 nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_F_avx_256_double
1179                     (t_nblist                    * gmx_restrict       nlist,
1180                      rvec                        * gmx_restrict          xx,
1181                      rvec                        * gmx_restrict          ff,
1182                      t_forcerec                  * gmx_restrict          fr,
1183                      t_mdatoms                   * gmx_restrict     mdatoms,
1184                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1185                      t_nrnb                      * gmx_restrict        nrnb)
1186 {
1187     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1188      * just 0 for non-waters.
1189      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1190      * jnr indices corresponding to data put in the four positions in the SIMD register.
1191      */
1192     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1193     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1194     int              jnrA,jnrB,jnrC,jnrD;
1195     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1196     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1197     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1198     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1199     real             rcutoff_scalar;
1200     real             *shiftvec,*fshift,*x,*f;
1201     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1202     real             scratch[4*DIM];
1203     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1204     real *           vdwioffsetptr0;
1205     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1206     real *           vdwioffsetptr1;
1207     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1208     real *           vdwioffsetptr2;
1209     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1210     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1211     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1212     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1213     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1214     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1215     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1216     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1217     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1218     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1219     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1220     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1221     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1222     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1223     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1224     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1225     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1226     real             *charge;
1227     int              nvdwtype;
1228     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1229     int              *vdwtype;
1230     real             *vdwparam;
1231     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1232     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1233     __m256d          dummy_mask,cutoff_mask;
1234     __m128           tmpmask0,tmpmask1;
1235     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1236     __m256d          one     = _mm256_set1_pd(1.0);
1237     __m256d          two     = _mm256_set1_pd(2.0);
1238     x                = xx[0];
1239     f                = ff[0];
1240
1241     nri              = nlist->nri;
1242     iinr             = nlist->iinr;
1243     jindex           = nlist->jindex;
1244     jjnr             = nlist->jjnr;
1245     shiftidx         = nlist->shift;
1246     gid              = nlist->gid;
1247     shiftvec         = fr->shift_vec[0];
1248     fshift           = fr->fshift[0];
1249     facel            = _mm256_set1_pd(fr->epsfac);
1250     charge           = mdatoms->chargeA;
1251     krf              = _mm256_set1_pd(fr->ic->k_rf);
1252     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
1253     crf              = _mm256_set1_pd(fr->ic->c_rf);
1254     nvdwtype         = fr->ntype;
1255     vdwparam         = fr->nbfp;
1256     vdwtype          = mdatoms->typeA;
1257
1258     /* Setup water-specific parameters */
1259     inr              = nlist->iinr[0];
1260     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1261     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1262     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1263     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1264
1265     jq0              = _mm256_set1_pd(charge[inr+0]);
1266     jq1              = _mm256_set1_pd(charge[inr+1]);
1267     jq2              = _mm256_set1_pd(charge[inr+2]);
1268     vdwjidx0A        = 2*vdwtype[inr+0];
1269     qq00             = _mm256_mul_pd(iq0,jq0);
1270     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1271     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1272     qq01             = _mm256_mul_pd(iq0,jq1);
1273     qq02             = _mm256_mul_pd(iq0,jq2);
1274     qq10             = _mm256_mul_pd(iq1,jq0);
1275     qq11             = _mm256_mul_pd(iq1,jq1);
1276     qq12             = _mm256_mul_pd(iq1,jq2);
1277     qq20             = _mm256_mul_pd(iq2,jq0);
1278     qq21             = _mm256_mul_pd(iq2,jq1);
1279     qq22             = _mm256_mul_pd(iq2,jq2);
1280
1281     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1282     rcutoff_scalar   = fr->rcoulomb;
1283     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
1284     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
1285
1286     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
1287     rvdw             = _mm256_set1_pd(fr->rvdw);
1288
1289     /* Avoid stupid compiler warnings */
1290     jnrA = jnrB = jnrC = jnrD = 0;
1291     j_coord_offsetA = 0;
1292     j_coord_offsetB = 0;
1293     j_coord_offsetC = 0;
1294     j_coord_offsetD = 0;
1295
1296     outeriter        = 0;
1297     inneriter        = 0;
1298
1299     for(iidx=0;iidx<4*DIM;iidx++)
1300     {
1301         scratch[iidx] = 0.0;
1302     }
1303
1304     /* Start outer loop over neighborlists */
1305     for(iidx=0; iidx<nri; iidx++)
1306     {
1307         /* Load shift vector for this list */
1308         i_shift_offset   = DIM*shiftidx[iidx];
1309
1310         /* Load limits for loop over neighbors */
1311         j_index_start    = jindex[iidx];
1312         j_index_end      = jindex[iidx+1];
1313
1314         /* Get outer coordinate index */
1315         inr              = iinr[iidx];
1316         i_coord_offset   = DIM*inr;
1317
1318         /* Load i particle coords and add shift vector */
1319         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1320                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1321
1322         fix0             = _mm256_setzero_pd();
1323         fiy0             = _mm256_setzero_pd();
1324         fiz0             = _mm256_setzero_pd();
1325         fix1             = _mm256_setzero_pd();
1326         fiy1             = _mm256_setzero_pd();
1327         fiz1             = _mm256_setzero_pd();
1328         fix2             = _mm256_setzero_pd();
1329         fiy2             = _mm256_setzero_pd();
1330         fiz2             = _mm256_setzero_pd();
1331
1332         /* Start inner kernel loop */
1333         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1334         {
1335
1336             /* Get j neighbor index, and coordinate index */
1337             jnrA             = jjnr[jidx];
1338             jnrB             = jjnr[jidx+1];
1339             jnrC             = jjnr[jidx+2];
1340             jnrD             = jjnr[jidx+3];
1341             j_coord_offsetA  = DIM*jnrA;
1342             j_coord_offsetB  = DIM*jnrB;
1343             j_coord_offsetC  = DIM*jnrC;
1344             j_coord_offsetD  = DIM*jnrD;
1345
1346             /* load j atom coordinates */
1347             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1348                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1349                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1350
1351             /* Calculate displacement vector */
1352             dx00             = _mm256_sub_pd(ix0,jx0);
1353             dy00             = _mm256_sub_pd(iy0,jy0);
1354             dz00             = _mm256_sub_pd(iz0,jz0);
1355             dx01             = _mm256_sub_pd(ix0,jx1);
1356             dy01             = _mm256_sub_pd(iy0,jy1);
1357             dz01             = _mm256_sub_pd(iz0,jz1);
1358             dx02             = _mm256_sub_pd(ix0,jx2);
1359             dy02             = _mm256_sub_pd(iy0,jy2);
1360             dz02             = _mm256_sub_pd(iz0,jz2);
1361             dx10             = _mm256_sub_pd(ix1,jx0);
1362             dy10             = _mm256_sub_pd(iy1,jy0);
1363             dz10             = _mm256_sub_pd(iz1,jz0);
1364             dx11             = _mm256_sub_pd(ix1,jx1);
1365             dy11             = _mm256_sub_pd(iy1,jy1);
1366             dz11             = _mm256_sub_pd(iz1,jz1);
1367             dx12             = _mm256_sub_pd(ix1,jx2);
1368             dy12             = _mm256_sub_pd(iy1,jy2);
1369             dz12             = _mm256_sub_pd(iz1,jz2);
1370             dx20             = _mm256_sub_pd(ix2,jx0);
1371             dy20             = _mm256_sub_pd(iy2,jy0);
1372             dz20             = _mm256_sub_pd(iz2,jz0);
1373             dx21             = _mm256_sub_pd(ix2,jx1);
1374             dy21             = _mm256_sub_pd(iy2,jy1);
1375             dz21             = _mm256_sub_pd(iz2,jz1);
1376             dx22             = _mm256_sub_pd(ix2,jx2);
1377             dy22             = _mm256_sub_pd(iy2,jy2);
1378             dz22             = _mm256_sub_pd(iz2,jz2);
1379
1380             /* Calculate squared distance and things based on it */
1381             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1382             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1383             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1384             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1385             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1386             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1387             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1388             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1389             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1390
1391             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1392             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1393             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1394             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1395             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1396             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1397             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1398             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1399             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1400
1401             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1402             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1403             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1404             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1405             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1406             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1407             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1408             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1409             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1410
1411             fjx0             = _mm256_setzero_pd();
1412             fjy0             = _mm256_setzero_pd();
1413             fjz0             = _mm256_setzero_pd();
1414             fjx1             = _mm256_setzero_pd();
1415             fjy1             = _mm256_setzero_pd();
1416             fjz1             = _mm256_setzero_pd();
1417             fjx2             = _mm256_setzero_pd();
1418             fjy2             = _mm256_setzero_pd();
1419             fjz2             = _mm256_setzero_pd();
1420
1421             /**************************
1422              * CALCULATE INTERACTIONS *
1423              **************************/
1424
1425             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1426             {
1427
1428             /* REACTION-FIELD ELECTROSTATICS */
1429             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1430
1431             /* LENNARD-JONES DISPERSION/REPULSION */
1432
1433             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1434             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1435
1436             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1437
1438             fscal            = _mm256_add_pd(felec,fvdw);
1439
1440             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1441
1442             /* Calculate temporary vectorial force */
1443             tx               = _mm256_mul_pd(fscal,dx00);
1444             ty               = _mm256_mul_pd(fscal,dy00);
1445             tz               = _mm256_mul_pd(fscal,dz00);
1446
1447             /* Update vectorial force */
1448             fix0             = _mm256_add_pd(fix0,tx);
1449             fiy0             = _mm256_add_pd(fiy0,ty);
1450             fiz0             = _mm256_add_pd(fiz0,tz);
1451
1452             fjx0             = _mm256_add_pd(fjx0,tx);
1453             fjy0             = _mm256_add_pd(fjy0,ty);
1454             fjz0             = _mm256_add_pd(fjz0,tz);
1455
1456             }
1457
1458             /**************************
1459              * CALCULATE INTERACTIONS *
1460              **************************/
1461
1462             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1463             {
1464
1465             /* REACTION-FIELD ELECTROSTATICS */
1466             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1467
1468             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
1469
1470             fscal            = felec;
1471
1472             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1473
1474             /* Calculate temporary vectorial force */
1475             tx               = _mm256_mul_pd(fscal,dx01);
1476             ty               = _mm256_mul_pd(fscal,dy01);
1477             tz               = _mm256_mul_pd(fscal,dz01);
1478
1479             /* Update vectorial force */
1480             fix0             = _mm256_add_pd(fix0,tx);
1481             fiy0             = _mm256_add_pd(fiy0,ty);
1482             fiz0             = _mm256_add_pd(fiz0,tz);
1483
1484             fjx1             = _mm256_add_pd(fjx1,tx);
1485             fjy1             = _mm256_add_pd(fjy1,ty);
1486             fjz1             = _mm256_add_pd(fjz1,tz);
1487
1488             }
1489
1490             /**************************
1491              * CALCULATE INTERACTIONS *
1492              **************************/
1493
1494             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1495             {
1496
1497             /* REACTION-FIELD ELECTROSTATICS */
1498             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1499
1500             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
1501
1502             fscal            = felec;
1503
1504             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1505
1506             /* Calculate temporary vectorial force */
1507             tx               = _mm256_mul_pd(fscal,dx02);
1508             ty               = _mm256_mul_pd(fscal,dy02);
1509             tz               = _mm256_mul_pd(fscal,dz02);
1510
1511             /* Update vectorial force */
1512             fix0             = _mm256_add_pd(fix0,tx);
1513             fiy0             = _mm256_add_pd(fiy0,ty);
1514             fiz0             = _mm256_add_pd(fiz0,tz);
1515
1516             fjx2             = _mm256_add_pd(fjx2,tx);
1517             fjy2             = _mm256_add_pd(fjy2,ty);
1518             fjz2             = _mm256_add_pd(fjz2,tz);
1519
1520             }
1521
1522             /**************************
1523              * CALCULATE INTERACTIONS *
1524              **************************/
1525
1526             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1527             {
1528
1529             /* REACTION-FIELD ELECTROSTATICS */
1530             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1531
1532             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1533
1534             fscal            = felec;
1535
1536             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1537
1538             /* Calculate temporary vectorial force */
1539             tx               = _mm256_mul_pd(fscal,dx10);
1540             ty               = _mm256_mul_pd(fscal,dy10);
1541             tz               = _mm256_mul_pd(fscal,dz10);
1542
1543             /* Update vectorial force */
1544             fix1             = _mm256_add_pd(fix1,tx);
1545             fiy1             = _mm256_add_pd(fiy1,ty);
1546             fiz1             = _mm256_add_pd(fiz1,tz);
1547
1548             fjx0             = _mm256_add_pd(fjx0,tx);
1549             fjy0             = _mm256_add_pd(fjy0,ty);
1550             fjz0             = _mm256_add_pd(fjz0,tz);
1551
1552             }
1553
1554             /**************************
1555              * CALCULATE INTERACTIONS *
1556              **************************/
1557
1558             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1559             {
1560
1561             /* REACTION-FIELD ELECTROSTATICS */
1562             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1563
1564             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1565
1566             fscal            = felec;
1567
1568             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1569
1570             /* Calculate temporary vectorial force */
1571             tx               = _mm256_mul_pd(fscal,dx11);
1572             ty               = _mm256_mul_pd(fscal,dy11);
1573             tz               = _mm256_mul_pd(fscal,dz11);
1574
1575             /* Update vectorial force */
1576             fix1             = _mm256_add_pd(fix1,tx);
1577             fiy1             = _mm256_add_pd(fiy1,ty);
1578             fiz1             = _mm256_add_pd(fiz1,tz);
1579
1580             fjx1             = _mm256_add_pd(fjx1,tx);
1581             fjy1             = _mm256_add_pd(fjy1,ty);
1582             fjz1             = _mm256_add_pd(fjz1,tz);
1583
1584             }
1585
1586             /**************************
1587              * CALCULATE INTERACTIONS *
1588              **************************/
1589
1590             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1591             {
1592
1593             /* REACTION-FIELD ELECTROSTATICS */
1594             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1595
1596             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1597
1598             fscal            = felec;
1599
1600             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1601
1602             /* Calculate temporary vectorial force */
1603             tx               = _mm256_mul_pd(fscal,dx12);
1604             ty               = _mm256_mul_pd(fscal,dy12);
1605             tz               = _mm256_mul_pd(fscal,dz12);
1606
1607             /* Update vectorial force */
1608             fix1             = _mm256_add_pd(fix1,tx);
1609             fiy1             = _mm256_add_pd(fiy1,ty);
1610             fiz1             = _mm256_add_pd(fiz1,tz);
1611
1612             fjx2             = _mm256_add_pd(fjx2,tx);
1613             fjy2             = _mm256_add_pd(fjy2,ty);
1614             fjz2             = _mm256_add_pd(fjz2,tz);
1615
1616             }
1617
1618             /**************************
1619              * CALCULATE INTERACTIONS *
1620              **************************/
1621
1622             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1623             {
1624
1625             /* REACTION-FIELD ELECTROSTATICS */
1626             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1627
1628             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1629
1630             fscal            = felec;
1631
1632             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1633
1634             /* Calculate temporary vectorial force */
1635             tx               = _mm256_mul_pd(fscal,dx20);
1636             ty               = _mm256_mul_pd(fscal,dy20);
1637             tz               = _mm256_mul_pd(fscal,dz20);
1638
1639             /* Update vectorial force */
1640             fix2             = _mm256_add_pd(fix2,tx);
1641             fiy2             = _mm256_add_pd(fiy2,ty);
1642             fiz2             = _mm256_add_pd(fiz2,tz);
1643
1644             fjx0             = _mm256_add_pd(fjx0,tx);
1645             fjy0             = _mm256_add_pd(fjy0,ty);
1646             fjz0             = _mm256_add_pd(fjz0,tz);
1647
1648             }
1649
1650             /**************************
1651              * CALCULATE INTERACTIONS *
1652              **************************/
1653
1654             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1655             {
1656
1657             /* REACTION-FIELD ELECTROSTATICS */
1658             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1659
1660             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1661
1662             fscal            = felec;
1663
1664             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1665
1666             /* Calculate temporary vectorial force */
1667             tx               = _mm256_mul_pd(fscal,dx21);
1668             ty               = _mm256_mul_pd(fscal,dy21);
1669             tz               = _mm256_mul_pd(fscal,dz21);
1670
1671             /* Update vectorial force */
1672             fix2             = _mm256_add_pd(fix2,tx);
1673             fiy2             = _mm256_add_pd(fiy2,ty);
1674             fiz2             = _mm256_add_pd(fiz2,tz);
1675
1676             fjx1             = _mm256_add_pd(fjx1,tx);
1677             fjy1             = _mm256_add_pd(fjy1,ty);
1678             fjz1             = _mm256_add_pd(fjz1,tz);
1679
1680             }
1681
1682             /**************************
1683              * CALCULATE INTERACTIONS *
1684              **************************/
1685
1686             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1687             {
1688
1689             /* REACTION-FIELD ELECTROSTATICS */
1690             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1691
1692             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1693
1694             fscal            = felec;
1695
1696             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1697
1698             /* Calculate temporary vectorial force */
1699             tx               = _mm256_mul_pd(fscal,dx22);
1700             ty               = _mm256_mul_pd(fscal,dy22);
1701             tz               = _mm256_mul_pd(fscal,dz22);
1702
1703             /* Update vectorial force */
1704             fix2             = _mm256_add_pd(fix2,tx);
1705             fiy2             = _mm256_add_pd(fiy2,ty);
1706             fiz2             = _mm256_add_pd(fiz2,tz);
1707
1708             fjx2             = _mm256_add_pd(fjx2,tx);
1709             fjy2             = _mm256_add_pd(fjy2,ty);
1710             fjz2             = _mm256_add_pd(fjz2,tz);
1711
1712             }
1713
1714             fjptrA             = f+j_coord_offsetA;
1715             fjptrB             = f+j_coord_offsetB;
1716             fjptrC             = f+j_coord_offsetC;
1717             fjptrD             = f+j_coord_offsetD;
1718
1719             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1720                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1721
1722             /* Inner loop uses 277 flops */
1723         }
1724
1725         if(jidx<j_index_end)
1726         {
1727
1728             /* Get j neighbor index, and coordinate index */
1729             jnrlistA         = jjnr[jidx];
1730             jnrlistB         = jjnr[jidx+1];
1731             jnrlistC         = jjnr[jidx+2];
1732             jnrlistD         = jjnr[jidx+3];
1733             /* Sign of each element will be negative for non-real atoms.
1734              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1735              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1736              */
1737             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1738
1739             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1740             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1741             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1742
1743             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1744             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1745             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1746             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1747             j_coord_offsetA  = DIM*jnrA;
1748             j_coord_offsetB  = DIM*jnrB;
1749             j_coord_offsetC  = DIM*jnrC;
1750             j_coord_offsetD  = DIM*jnrD;
1751
1752             /* load j atom coordinates */
1753             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1754                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1755                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1756
1757             /* Calculate displacement vector */
1758             dx00             = _mm256_sub_pd(ix0,jx0);
1759             dy00             = _mm256_sub_pd(iy0,jy0);
1760             dz00             = _mm256_sub_pd(iz0,jz0);
1761             dx01             = _mm256_sub_pd(ix0,jx1);
1762             dy01             = _mm256_sub_pd(iy0,jy1);
1763             dz01             = _mm256_sub_pd(iz0,jz1);
1764             dx02             = _mm256_sub_pd(ix0,jx2);
1765             dy02             = _mm256_sub_pd(iy0,jy2);
1766             dz02             = _mm256_sub_pd(iz0,jz2);
1767             dx10             = _mm256_sub_pd(ix1,jx0);
1768             dy10             = _mm256_sub_pd(iy1,jy0);
1769             dz10             = _mm256_sub_pd(iz1,jz0);
1770             dx11             = _mm256_sub_pd(ix1,jx1);
1771             dy11             = _mm256_sub_pd(iy1,jy1);
1772             dz11             = _mm256_sub_pd(iz1,jz1);
1773             dx12             = _mm256_sub_pd(ix1,jx2);
1774             dy12             = _mm256_sub_pd(iy1,jy2);
1775             dz12             = _mm256_sub_pd(iz1,jz2);
1776             dx20             = _mm256_sub_pd(ix2,jx0);
1777             dy20             = _mm256_sub_pd(iy2,jy0);
1778             dz20             = _mm256_sub_pd(iz2,jz0);
1779             dx21             = _mm256_sub_pd(ix2,jx1);
1780             dy21             = _mm256_sub_pd(iy2,jy1);
1781             dz21             = _mm256_sub_pd(iz2,jz1);
1782             dx22             = _mm256_sub_pd(ix2,jx2);
1783             dy22             = _mm256_sub_pd(iy2,jy2);
1784             dz22             = _mm256_sub_pd(iz2,jz2);
1785
1786             /* Calculate squared distance and things based on it */
1787             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1788             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1789             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1790             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1791             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1792             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1793             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1794             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1795             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1796
1797             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1798             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1799             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1800             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1801             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1802             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1803             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1804             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1805             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1806
1807             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1808             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1809             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1810             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1811             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1812             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1813             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1814             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1815             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1816
1817             fjx0             = _mm256_setzero_pd();
1818             fjy0             = _mm256_setzero_pd();
1819             fjz0             = _mm256_setzero_pd();
1820             fjx1             = _mm256_setzero_pd();
1821             fjy1             = _mm256_setzero_pd();
1822             fjz1             = _mm256_setzero_pd();
1823             fjx2             = _mm256_setzero_pd();
1824             fjy2             = _mm256_setzero_pd();
1825             fjz2             = _mm256_setzero_pd();
1826
1827             /**************************
1828              * CALCULATE INTERACTIONS *
1829              **************************/
1830
1831             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1832             {
1833
1834             /* REACTION-FIELD ELECTROSTATICS */
1835             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1836
1837             /* LENNARD-JONES DISPERSION/REPULSION */
1838
1839             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1840             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1841
1842             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1843
1844             fscal            = _mm256_add_pd(felec,fvdw);
1845
1846             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1847
1848             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1849
1850             /* Calculate temporary vectorial force */
1851             tx               = _mm256_mul_pd(fscal,dx00);
1852             ty               = _mm256_mul_pd(fscal,dy00);
1853             tz               = _mm256_mul_pd(fscal,dz00);
1854
1855             /* Update vectorial force */
1856             fix0             = _mm256_add_pd(fix0,tx);
1857             fiy0             = _mm256_add_pd(fiy0,ty);
1858             fiz0             = _mm256_add_pd(fiz0,tz);
1859
1860             fjx0             = _mm256_add_pd(fjx0,tx);
1861             fjy0             = _mm256_add_pd(fjy0,ty);
1862             fjz0             = _mm256_add_pd(fjz0,tz);
1863
1864             }
1865
1866             /**************************
1867              * CALCULATE INTERACTIONS *
1868              **************************/
1869
1870             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1871             {
1872
1873             /* REACTION-FIELD ELECTROSTATICS */
1874             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1875
1876             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
1877
1878             fscal            = felec;
1879
1880             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1881
1882             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1883
1884             /* Calculate temporary vectorial force */
1885             tx               = _mm256_mul_pd(fscal,dx01);
1886             ty               = _mm256_mul_pd(fscal,dy01);
1887             tz               = _mm256_mul_pd(fscal,dz01);
1888
1889             /* Update vectorial force */
1890             fix0             = _mm256_add_pd(fix0,tx);
1891             fiy0             = _mm256_add_pd(fiy0,ty);
1892             fiz0             = _mm256_add_pd(fiz0,tz);
1893
1894             fjx1             = _mm256_add_pd(fjx1,tx);
1895             fjy1             = _mm256_add_pd(fjy1,ty);
1896             fjz1             = _mm256_add_pd(fjz1,tz);
1897
1898             }
1899
1900             /**************************
1901              * CALCULATE INTERACTIONS *
1902              **************************/
1903
1904             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1905             {
1906
1907             /* REACTION-FIELD ELECTROSTATICS */
1908             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1909
1910             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
1911
1912             fscal            = felec;
1913
1914             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1915
1916             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1917
1918             /* Calculate temporary vectorial force */
1919             tx               = _mm256_mul_pd(fscal,dx02);
1920             ty               = _mm256_mul_pd(fscal,dy02);
1921             tz               = _mm256_mul_pd(fscal,dz02);
1922
1923             /* Update vectorial force */
1924             fix0             = _mm256_add_pd(fix0,tx);
1925             fiy0             = _mm256_add_pd(fiy0,ty);
1926             fiz0             = _mm256_add_pd(fiz0,tz);
1927
1928             fjx2             = _mm256_add_pd(fjx2,tx);
1929             fjy2             = _mm256_add_pd(fjy2,ty);
1930             fjz2             = _mm256_add_pd(fjz2,tz);
1931
1932             }
1933
1934             /**************************
1935              * CALCULATE INTERACTIONS *
1936              **************************/
1937
1938             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1939             {
1940
1941             /* REACTION-FIELD ELECTROSTATICS */
1942             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1943
1944             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1945
1946             fscal            = felec;
1947
1948             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1949
1950             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1951
1952             /* Calculate temporary vectorial force */
1953             tx               = _mm256_mul_pd(fscal,dx10);
1954             ty               = _mm256_mul_pd(fscal,dy10);
1955             tz               = _mm256_mul_pd(fscal,dz10);
1956
1957             /* Update vectorial force */
1958             fix1             = _mm256_add_pd(fix1,tx);
1959             fiy1             = _mm256_add_pd(fiy1,ty);
1960             fiz1             = _mm256_add_pd(fiz1,tz);
1961
1962             fjx0             = _mm256_add_pd(fjx0,tx);
1963             fjy0             = _mm256_add_pd(fjy0,ty);
1964             fjz0             = _mm256_add_pd(fjz0,tz);
1965
1966             }
1967
1968             /**************************
1969              * CALCULATE INTERACTIONS *
1970              **************************/
1971
1972             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1973             {
1974
1975             /* REACTION-FIELD ELECTROSTATICS */
1976             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1977
1978             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1979
1980             fscal            = felec;
1981
1982             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1983
1984             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1985
1986             /* Calculate temporary vectorial force */
1987             tx               = _mm256_mul_pd(fscal,dx11);
1988             ty               = _mm256_mul_pd(fscal,dy11);
1989             tz               = _mm256_mul_pd(fscal,dz11);
1990
1991             /* Update vectorial force */
1992             fix1             = _mm256_add_pd(fix1,tx);
1993             fiy1             = _mm256_add_pd(fiy1,ty);
1994             fiz1             = _mm256_add_pd(fiz1,tz);
1995
1996             fjx1             = _mm256_add_pd(fjx1,tx);
1997             fjy1             = _mm256_add_pd(fjy1,ty);
1998             fjz1             = _mm256_add_pd(fjz1,tz);
1999
2000             }
2001
2002             /**************************
2003              * CALCULATE INTERACTIONS *
2004              **************************/
2005
2006             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2007             {
2008
2009             /* REACTION-FIELD ELECTROSTATICS */
2010             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
2011
2012             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
2013
2014             fscal            = felec;
2015
2016             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2017
2018             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2019
2020             /* Calculate temporary vectorial force */
2021             tx               = _mm256_mul_pd(fscal,dx12);
2022             ty               = _mm256_mul_pd(fscal,dy12);
2023             tz               = _mm256_mul_pd(fscal,dz12);
2024
2025             /* Update vectorial force */
2026             fix1             = _mm256_add_pd(fix1,tx);
2027             fiy1             = _mm256_add_pd(fiy1,ty);
2028             fiz1             = _mm256_add_pd(fiz1,tz);
2029
2030             fjx2             = _mm256_add_pd(fjx2,tx);
2031             fjy2             = _mm256_add_pd(fjy2,ty);
2032             fjz2             = _mm256_add_pd(fjz2,tz);
2033
2034             }
2035
2036             /**************************
2037              * CALCULATE INTERACTIONS *
2038              **************************/
2039
2040             if (gmx_mm256_any_lt(rsq20,rcutoff2))
2041             {
2042
2043             /* REACTION-FIELD ELECTROSTATICS */
2044             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
2045
2046             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
2047
2048             fscal            = felec;
2049
2050             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2051
2052             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2053
2054             /* Calculate temporary vectorial force */
2055             tx               = _mm256_mul_pd(fscal,dx20);
2056             ty               = _mm256_mul_pd(fscal,dy20);
2057             tz               = _mm256_mul_pd(fscal,dz20);
2058
2059             /* Update vectorial force */
2060             fix2             = _mm256_add_pd(fix2,tx);
2061             fiy2             = _mm256_add_pd(fiy2,ty);
2062             fiz2             = _mm256_add_pd(fiz2,tz);
2063
2064             fjx0             = _mm256_add_pd(fjx0,tx);
2065             fjy0             = _mm256_add_pd(fjy0,ty);
2066             fjz0             = _mm256_add_pd(fjz0,tz);
2067
2068             }
2069
2070             /**************************
2071              * CALCULATE INTERACTIONS *
2072              **************************/
2073
2074             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2075             {
2076
2077             /* REACTION-FIELD ELECTROSTATICS */
2078             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
2079
2080             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2081
2082             fscal            = felec;
2083
2084             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2085
2086             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2087
2088             /* Calculate temporary vectorial force */
2089             tx               = _mm256_mul_pd(fscal,dx21);
2090             ty               = _mm256_mul_pd(fscal,dy21);
2091             tz               = _mm256_mul_pd(fscal,dz21);
2092
2093             /* Update vectorial force */
2094             fix2             = _mm256_add_pd(fix2,tx);
2095             fiy2             = _mm256_add_pd(fiy2,ty);
2096             fiz2             = _mm256_add_pd(fiz2,tz);
2097
2098             fjx1             = _mm256_add_pd(fjx1,tx);
2099             fjy1             = _mm256_add_pd(fjy1,ty);
2100             fjz1             = _mm256_add_pd(fjz1,tz);
2101
2102             }
2103
2104             /**************************
2105              * CALCULATE INTERACTIONS *
2106              **************************/
2107
2108             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2109             {
2110
2111             /* REACTION-FIELD ELECTROSTATICS */
2112             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
2113
2114             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2115
2116             fscal            = felec;
2117
2118             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2119
2120             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2121
2122             /* Calculate temporary vectorial force */
2123             tx               = _mm256_mul_pd(fscal,dx22);
2124             ty               = _mm256_mul_pd(fscal,dy22);
2125             tz               = _mm256_mul_pd(fscal,dz22);
2126
2127             /* Update vectorial force */
2128             fix2             = _mm256_add_pd(fix2,tx);
2129             fiy2             = _mm256_add_pd(fiy2,ty);
2130             fiz2             = _mm256_add_pd(fiz2,tz);
2131
2132             fjx2             = _mm256_add_pd(fjx2,tx);
2133             fjy2             = _mm256_add_pd(fjy2,ty);
2134             fjz2             = _mm256_add_pd(fjz2,tz);
2135
2136             }
2137
2138             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2139             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2140             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2141             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2142
2143             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2144                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2145
2146             /* Inner loop uses 277 flops */
2147         }
2148
2149         /* End of innermost loop */
2150
2151         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2152                                                  f+i_coord_offset,fshift+i_shift_offset);
2153
2154         /* Increment number of inner iterations */
2155         inneriter                  += j_index_end - j_index_start;
2156
2157         /* Outer loop uses 18 flops */
2158     }
2159
2160     /* Increment number of outer iterations */
2161     outeriter        += nri;
2162
2163     /* Update outer/inner flops */
2164
2165     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*277);
2166 }