src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_VF_avx_256_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Water3
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_VF_avx_256_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  77     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  78     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  79     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  80     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  81     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  82     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  83     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  84     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  85     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  86     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  87     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  88     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  89     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  90     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  91     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  98     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  99     __m256d          dummy_mask,cutoff_mask;
 100     __m128           tmpmask0,tmpmask1;
 101     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 102     __m256d          one     = _mm256_set1_pd(1.0);
 103     __m256d          two     = _mm256_set1_pd(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm256_set1_pd(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117     krf              = _mm256_set1_pd(fr->ic->k_rf);
 118     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 119     crf              = _mm256_set1_pd(fr->ic->c_rf);
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 127     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 128     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 129     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 130
 131     jq0              = _mm256_set1_pd(charge[inr+0]);
 132     jq1              = _mm256_set1_pd(charge[inr+1]);
 133     jq2              = _mm256_set1_pd(charge[inr+2]);
 134     vdwjidx0A        = 2*vdwtype[inr+0];
 135     qq00             = _mm256_mul_pd(iq0,jq0);
 136     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 137     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 138     qq01             = _mm256_mul_pd(iq0,jq1);
 139     qq02             = _mm256_mul_pd(iq0,jq2);
 140     qq10             = _mm256_mul_pd(iq1,jq0);
 141     qq11             = _mm256_mul_pd(iq1,jq1);
 142     qq12             = _mm256_mul_pd(iq1,jq2);
 143     qq20             = _mm256_mul_pd(iq2,jq0);
 144     qq21             = _mm256_mul_pd(iq2,jq1);
 145     qq22             = _mm256_mul_pd(iq2,jq2);
 146
 147     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 148     rcutoff_scalar   = fr->rcoulomb;
 149     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 150     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 151
 152     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 153     rvdw             = _mm256_set1_pd(fr->rvdw);
 154
 155     /* Avoid stupid compiler warnings */
 156     jnrA = jnrB = jnrC = jnrD = 0;
 157     j_coord_offsetA = 0;
 158     j_coord_offsetB = 0;
 159     j_coord_offsetC = 0;
 160     j_coord_offsetD = 0;
 161
 162     outeriter        = 0;
 163     inneriter        = 0;
 164
 165     for(iidx=0;iidx<4*DIM;iidx++)
 166     {
 167         scratch[iidx] = 0.0;
 168     }
 169
 170     /* Start outer loop over neighborlists */
 171     for(iidx=0; iidx<nri; iidx++)
 172     {
 173         /* Load shift vector for this list */
 174         i_shift_offset   = DIM*shiftidx[iidx];
 175
 176         /* Load limits for loop over neighbors */
 177         j_index_start    = jindex[iidx];
 178         j_index_end      = jindex[iidx+1];
 179
 180         /* Get outer coordinate index */
 181         inr              = iinr[iidx];
 182         i_coord_offset   = DIM*inr;
 183
 184         /* Load i particle coords and add shift vector */
 185         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 186                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 187
 188         fix0             = _mm256_setzero_pd();
 189         fiy0             = _mm256_setzero_pd();
 190         fiz0             = _mm256_setzero_pd();
 191         fix1             = _mm256_setzero_pd();
 192         fiy1             = _mm256_setzero_pd();
 193         fiz1             = _mm256_setzero_pd();
 194         fix2             = _mm256_setzero_pd();
 195         fiy2             = _mm256_setzero_pd();
 196         fiz2             = _mm256_setzero_pd();
 197
 198         /* Reset potential sums */
 199         velecsum         = _mm256_setzero_pd();
 200         vvdwsum          = _mm256_setzero_pd();
 201
 202         /* Start inner kernel loop */
 203         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 204         {
 205
 206             /* Get j neighbor index, and coordinate index */
 207             jnrA             = jjnr[jidx];
 208             jnrB             = jjnr[jidx+1];
 209             jnrC             = jjnr[jidx+2];
 210             jnrD             = jjnr[jidx+3];
 211             j_coord_offsetA  = DIM*jnrA;
 212             j_coord_offsetB  = DIM*jnrB;
 213             j_coord_offsetC  = DIM*jnrC;
 214             j_coord_offsetD  = DIM*jnrD;
 215
 216             /* load j atom coordinates */
 217             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 218                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 219                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm256_sub_pd(ix0,jx0);
 223             dy00             = _mm256_sub_pd(iy0,jy0);
 224             dz00             = _mm256_sub_pd(iz0,jz0);
 225             dx01             = _mm256_sub_pd(ix0,jx1);
 226             dy01             = _mm256_sub_pd(iy0,jy1);
 227             dz01             = _mm256_sub_pd(iz0,jz1);
 228             dx02             = _mm256_sub_pd(ix0,jx2);
 229             dy02             = _mm256_sub_pd(iy0,jy2);
 230             dz02             = _mm256_sub_pd(iz0,jz2);
 231             dx10             = _mm256_sub_pd(ix1,jx0);
 232             dy10             = _mm256_sub_pd(iy1,jy0);
 233             dz10             = _mm256_sub_pd(iz1,jz0);
 234             dx11             = _mm256_sub_pd(ix1,jx1);
 235             dy11             = _mm256_sub_pd(iy1,jy1);
 236             dz11             = _mm256_sub_pd(iz1,jz1);
 237             dx12             = _mm256_sub_pd(ix1,jx2);
 238             dy12             = _mm256_sub_pd(iy1,jy2);
 239             dz12             = _mm256_sub_pd(iz1,jz2);
 240             dx20             = _mm256_sub_pd(ix2,jx0);
 241             dy20             = _mm256_sub_pd(iy2,jy0);
 242             dz20             = _mm256_sub_pd(iz2,jz0);
 243             dx21             = _mm256_sub_pd(ix2,jx1);
 244             dy21             = _mm256_sub_pd(iy2,jy1);
 245             dz21             = _mm256_sub_pd(iz2,jz1);
 246             dx22             = _mm256_sub_pd(ix2,jx2);
 247             dy22             = _mm256_sub_pd(iy2,jy2);
 248             dz22             = _mm256_sub_pd(iz2,jz2);
 249
 250             /* Calculate squared distance and things based on it */
 251             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 252             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 253             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 254             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 255             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 256             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 257             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 258             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 259             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 260
 261             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 262             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 263             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 264             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 265             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 266             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 267             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 268             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 269             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 270
 271             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 272             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 273             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 274             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 275             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 276             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 277             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 278             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 279             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 280
 281             fjx0             = _mm256_setzero_pd();
 282             fjy0             = _mm256_setzero_pd();
 283             fjz0             = _mm256_setzero_pd();
 284             fjx1             = _mm256_setzero_pd();
 285             fjy1             = _mm256_setzero_pd();
 286             fjz1             = _mm256_setzero_pd();
 287             fjx2             = _mm256_setzero_pd();
 288             fjy2             = _mm256_setzero_pd();
 289             fjz2             = _mm256_setzero_pd();
 290
 291             /**************************
 292              * CALCULATE INTERACTIONS *
 293              **************************/
 294
 295             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 296             {
 297
 298             /* REACTION-FIELD ELECTROSTATICS */
 299             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 300             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 301
 302             /* LENNARD-JONES DISPERSION/REPULSION */
 303
 304             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 305             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 306             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 307             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) ,
 308                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 309             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 310
 311             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velec            = _mm256_and_pd(velec,cutoff_mask);
 315             velecsum         = _mm256_add_pd(velecsum,velec);
 316             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 317             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 318
 319             fscal            = _mm256_add_pd(felec,fvdw);
 320
 321             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm256_mul_pd(fscal,dx00);
 325             ty               = _mm256_mul_pd(fscal,dy00);
 326             tz               = _mm256_mul_pd(fscal,dz00);
 327
 328             /* Update vectorial force */
 329             fix0             = _mm256_add_pd(fix0,tx);
 330             fiy0             = _mm256_add_pd(fiy0,ty);
 331             fiz0             = _mm256_add_pd(fiz0,tz);
 332
 333             fjx0             = _mm256_add_pd(fjx0,tx);
 334             fjy0             = _mm256_add_pd(fjy0,ty);
 335             fjz0             = _mm256_add_pd(fjz0,tz);
 336
 337             }
 338
 339             /**************************
 340              * CALCULATE INTERACTIONS *
 341              **************************/
 342
 343             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 344             {
 345
 346             /* REACTION-FIELD ELECTROSTATICS */
 347             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
 348             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
 349
 350             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velec            = _mm256_and_pd(velec,cutoff_mask);
 354             velecsum         = _mm256_add_pd(velecsum,velec);
 355
 356             fscal            = felec;
 357
 358             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm256_mul_pd(fscal,dx01);
 362             ty               = _mm256_mul_pd(fscal,dy01);
 363             tz               = _mm256_mul_pd(fscal,dz01);
 364
 365             /* Update vectorial force */
 366             fix0             = _mm256_add_pd(fix0,tx);
 367             fiy0             = _mm256_add_pd(fiy0,ty);
 368             fiz0             = _mm256_add_pd(fiz0,tz);
 369
 370             fjx1             = _mm256_add_pd(fjx1,tx);
 371             fjy1             = _mm256_add_pd(fjy1,ty);
 372             fjz1             = _mm256_add_pd(fjz1,tz);
 373
 374             }
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 381             {
 382
 383             /* REACTION-FIELD ELECTROSTATICS */
 384             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
 385             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
 386
 387             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velec            = _mm256_and_pd(velec,cutoff_mask);
 391             velecsum         = _mm256_add_pd(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 396
 397             /* Calculate temporary vectorial force */
 398             tx               = _mm256_mul_pd(fscal,dx02);
 399             ty               = _mm256_mul_pd(fscal,dy02);
 400             tz               = _mm256_mul_pd(fscal,dz02);
 401
 402             /* Update vectorial force */
 403             fix0             = _mm256_add_pd(fix0,tx);
 404             fiy0             = _mm256_add_pd(fiy0,ty);
 405             fiz0             = _mm256_add_pd(fiz0,tz);
 406
 407             fjx2             = _mm256_add_pd(fjx2,tx);
 408             fjy2             = _mm256_add_pd(fjy2,ty);
 409             fjz2             = _mm256_add_pd(fjz2,tz);
 410
 411             }
 412
 413             /**************************
 414              * CALCULATE INTERACTIONS *
 415              **************************/
 416
 417             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 418             {
 419
 420             /* REACTION-FIELD ELECTROSTATICS */
 421             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 422             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 423
 424             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 425
 426             /* Update potential sum for this i atom from the interaction with this j atom. */
 427             velec            = _mm256_and_pd(velec,cutoff_mask);
 428             velecsum         = _mm256_add_pd(velecsum,velec);
 429
 430             fscal            = felec;
 431
 432             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 433
 434             /* Calculate temporary vectorial force */
 435             tx               = _mm256_mul_pd(fscal,dx10);
 436             ty               = _mm256_mul_pd(fscal,dy10);
 437             tz               = _mm256_mul_pd(fscal,dz10);
 438
 439             /* Update vectorial force */
 440             fix1             = _mm256_add_pd(fix1,tx);
 441             fiy1             = _mm256_add_pd(fiy1,ty);
 442             fiz1             = _mm256_add_pd(fiz1,tz);
 443
 444             fjx0             = _mm256_add_pd(fjx0,tx);
 445             fjy0             = _mm256_add_pd(fjy0,ty);
 446             fjz0             = _mm256_add_pd(fjz0,tz);
 447
 448             }
 449
 450             /**************************
 451              * CALCULATE INTERACTIONS *
 452              **************************/
 453
 454             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 455             {
 456
 457             /* REACTION-FIELD ELECTROSTATICS */
 458             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 459             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 460
 461             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 462
 463             /* Update potential sum for this i atom from the interaction with this j atom. */
 464             velec            = _mm256_and_pd(velec,cutoff_mask);
 465             velecsum         = _mm256_add_pd(velecsum,velec);
 466
 467             fscal            = felec;
 468
 469             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 470
 471             /* Calculate temporary vectorial force */
 472             tx               = _mm256_mul_pd(fscal,dx11);
 473             ty               = _mm256_mul_pd(fscal,dy11);
 474             tz               = _mm256_mul_pd(fscal,dz11);
 475
 476             /* Update vectorial force */
 477             fix1             = _mm256_add_pd(fix1,tx);
 478             fiy1             = _mm256_add_pd(fiy1,ty);
 479             fiz1             = _mm256_add_pd(fiz1,tz);
 480
 481             fjx1             = _mm256_add_pd(fjx1,tx);
 482             fjy1             = _mm256_add_pd(fjy1,ty);
 483             fjz1             = _mm256_add_pd(fjz1,tz);
 484
 485             }
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 489              **************************/
 490
 491             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 492             {
 493
 494             /* REACTION-FIELD ELECTROSTATICS */
 495             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 496             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 497
 498             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 499
 500             /* Update potential sum for this i atom from the interaction with this j atom. */
 501             velec            = _mm256_and_pd(velec,cutoff_mask);
 502             velecsum         = _mm256_add_pd(velecsum,velec);
 503
 504             fscal            = felec;
 505
 506             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 507
 508             /* Calculate temporary vectorial force */
 509             tx               = _mm256_mul_pd(fscal,dx12);
 510             ty               = _mm256_mul_pd(fscal,dy12);
 511             tz               = _mm256_mul_pd(fscal,dz12);
 512
 513             /* Update vectorial force */
 514             fix1             = _mm256_add_pd(fix1,tx);
 515             fiy1             = _mm256_add_pd(fiy1,ty);
 516             fiz1             = _mm256_add_pd(fiz1,tz);
 517
 518             fjx2             = _mm256_add_pd(fjx2,tx);
 519             fjy2             = _mm256_add_pd(fjy2,ty);
 520             fjz2             = _mm256_add_pd(fjz2,tz);
 521
 522             }
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 529             {
 530
 531             /* REACTION-FIELD ELECTROSTATICS */
 532             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 533             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 534
 535             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velec            = _mm256_and_pd(velec,cutoff_mask);
 539             velecsum         = _mm256_add_pd(velecsum,velec);
 540
 541             fscal            = felec;
 542
 543             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 544
 545             /* Calculate temporary vectorial force */
 546             tx               = _mm256_mul_pd(fscal,dx20);
 547             ty               = _mm256_mul_pd(fscal,dy20);
 548             tz               = _mm256_mul_pd(fscal,dz20);
 549
 550             /* Update vectorial force */
 551             fix2             = _mm256_add_pd(fix2,tx);
 552             fiy2             = _mm256_add_pd(fiy2,ty);
 553             fiz2             = _mm256_add_pd(fiz2,tz);
 554
 555             fjx0             = _mm256_add_pd(fjx0,tx);
 556             fjy0             = _mm256_add_pd(fjy0,ty);
 557             fjz0             = _mm256_add_pd(fjz0,tz);
 558
 559             }
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 566             {
 567
 568             /* REACTION-FIELD ELECTROSTATICS */
 569             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 570             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 571
 572             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
 573
 574             /* Update potential sum for this i atom from the interaction with this j atom. */
 575             velec            = _mm256_and_pd(velec,cutoff_mask);
 576             velecsum         = _mm256_add_pd(velecsum,velec);
 577
 578             fscal            = felec;
 579
 580             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm256_mul_pd(fscal,dx21);
 584             ty               = _mm256_mul_pd(fscal,dy21);
 585             tz               = _mm256_mul_pd(fscal,dz21);
 586
 587             /* Update vectorial force */
 588             fix2             = _mm256_add_pd(fix2,tx);
 589             fiy2             = _mm256_add_pd(fiy2,ty);
 590             fiz2             = _mm256_add_pd(fiz2,tz);
 591
 592             fjx1             = _mm256_add_pd(fjx1,tx);
 593             fjy1             = _mm256_add_pd(fjy1,ty);
 594             fjz1             = _mm256_add_pd(fjz1,tz);
 595
 596             }
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 603             {
 604
 605             /* REACTION-FIELD ELECTROSTATICS */
 606             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 607             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 608
 609             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
 610
 611             /* Update potential sum for this i atom from the interaction with this j atom. */
 612             velec            = _mm256_and_pd(velec,cutoff_mask);
 613             velecsum         = _mm256_add_pd(velecsum,velec);
 614
 615             fscal            = felec;
 616
 617             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 618
 619             /* Calculate temporary vectorial force */
 620             tx               = _mm256_mul_pd(fscal,dx22);
 621             ty               = _mm256_mul_pd(fscal,dy22);
 622             tz               = _mm256_mul_pd(fscal,dz22);
 623
 624             /* Update vectorial force */
 625             fix2             = _mm256_add_pd(fix2,tx);
 626             fiy2             = _mm256_add_pd(fiy2,ty);
 627             fiz2             = _mm256_add_pd(fiz2,tz);
 628
 629             fjx2             = _mm256_add_pd(fjx2,tx);
 630             fjy2             = _mm256_add_pd(fjy2,ty);
 631             fjz2             = _mm256_add_pd(fjz2,tz);
 632
 633             }
 634
 635             fjptrA             = f+j_coord_offsetA;
 636             fjptrB             = f+j_coord_offsetB;
 637             fjptrC             = f+j_coord_offsetC;
 638             fjptrD             = f+j_coord_offsetD;
 639
 640             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 641                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 642
 643             /* Inner loop uses 342 flops */
 644         }
 645
 646         if(jidx<j_index_end)
 647         {
 648
 649             /* Get j neighbor index, and coordinate index */
 650             jnrlistA         = jjnr[jidx];
 651             jnrlistB         = jjnr[jidx+1];
 652             jnrlistC         = jjnr[jidx+2];
 653             jnrlistD         = jjnr[jidx+3];
 654             /* Sign of each element will be negative for non-real atoms.
 655              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 656              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 657              */
 658             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 659
 660             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 661             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 662             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 663
 664             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 665             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 666             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 667             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 668             j_coord_offsetA  = DIM*jnrA;
 669             j_coord_offsetB  = DIM*jnrB;
 670             j_coord_offsetC  = DIM*jnrC;
 671             j_coord_offsetD  = DIM*jnrD;
 672
 673             /* load j atom coordinates */
 674             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 675                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 676                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 677
 678             /* Calculate displacement vector */
 679             dx00             = _mm256_sub_pd(ix0,jx0);
 680             dy00             = _mm256_sub_pd(iy0,jy0);
 681             dz00             = _mm256_sub_pd(iz0,jz0);
 682             dx01             = _mm256_sub_pd(ix0,jx1);
 683             dy01             = _mm256_sub_pd(iy0,jy1);
 684             dz01             = _mm256_sub_pd(iz0,jz1);
 685             dx02             = _mm256_sub_pd(ix0,jx2);
 686             dy02             = _mm256_sub_pd(iy0,jy2);
 687             dz02             = _mm256_sub_pd(iz0,jz2);
 688             dx10             = _mm256_sub_pd(ix1,jx0);
 689             dy10             = _mm256_sub_pd(iy1,jy0);
 690             dz10             = _mm256_sub_pd(iz1,jz0);
 691             dx11             = _mm256_sub_pd(ix1,jx1);
 692             dy11             = _mm256_sub_pd(iy1,jy1);
 693             dz11             = _mm256_sub_pd(iz1,jz1);
 694             dx12             = _mm256_sub_pd(ix1,jx2);
 695             dy12             = _mm256_sub_pd(iy1,jy2);
 696             dz12             = _mm256_sub_pd(iz1,jz2);
 697             dx20             = _mm256_sub_pd(ix2,jx0);
 698             dy20             = _mm256_sub_pd(iy2,jy0);
 699             dz20             = _mm256_sub_pd(iz2,jz0);
 700             dx21             = _mm256_sub_pd(ix2,jx1);
 701             dy21             = _mm256_sub_pd(iy2,jy1);
 702             dz21             = _mm256_sub_pd(iz2,jz1);
 703             dx22             = _mm256_sub_pd(ix2,jx2);
 704             dy22             = _mm256_sub_pd(iy2,jy2);
 705             dz22             = _mm256_sub_pd(iz2,jz2);
 706
 707             /* Calculate squared distance and things based on it */
 708             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 709             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 710             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 711             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 712             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 713             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 714             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 715             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 716             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 717
 718             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 719             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 720             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 721             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 722             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 723             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 724             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 725             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 726             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 727
 728             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 729             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 730             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 731             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 732             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 733             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 734             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 735             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 736             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 737
 738             fjx0             = _mm256_setzero_pd();
 739             fjy0             = _mm256_setzero_pd();
 740             fjz0             = _mm256_setzero_pd();
 741             fjx1             = _mm256_setzero_pd();
 742             fjy1             = _mm256_setzero_pd();
 743             fjz1             = _mm256_setzero_pd();
 744             fjx2             = _mm256_setzero_pd();
 745             fjy2             = _mm256_setzero_pd();
 746             fjz2             = _mm256_setzero_pd();
 747
 748             /**************************
 749              * CALCULATE INTERACTIONS *
 750              **************************/
 751
 752             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 753             {
 754
 755             /* REACTION-FIELD ELECTROSTATICS */
 756             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 757             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 758
 759             /* LENNARD-JONES DISPERSION/REPULSION */
 760
 761             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 762             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 763             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 764             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) ,
 765                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 766             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 767
 768             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 769
 770             /* Update potential sum for this i atom from the interaction with this j atom. */
 771             velec            = _mm256_and_pd(velec,cutoff_mask);
 772             velec            = _mm256_andnot_pd(dummy_mask,velec);
 773             velecsum         = _mm256_add_pd(velecsum,velec);
 774             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 775             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 776             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 777
 778             fscal            = _mm256_add_pd(felec,fvdw);
 779
 780             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 781
 782             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 783
 784             /* Calculate temporary vectorial force */
 785             tx               = _mm256_mul_pd(fscal,dx00);
 786             ty               = _mm256_mul_pd(fscal,dy00);
 787             tz               = _mm256_mul_pd(fscal,dz00);
 788
 789             /* Update vectorial force */
 790             fix0             = _mm256_add_pd(fix0,tx);
 791             fiy0             = _mm256_add_pd(fiy0,ty);
 792             fiz0             = _mm256_add_pd(fiz0,tz);
 793
 794             fjx0             = _mm256_add_pd(fjx0,tx);
 795             fjy0             = _mm256_add_pd(fjy0,ty);
 796             fjz0             = _mm256_add_pd(fjz0,tz);
 797
 798             }
 799
 800             /**************************
 801              * CALCULATE INTERACTIONS *
 802              **************************/
 803
 804             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 805             {
 806
 807             /* REACTION-FIELD ELECTROSTATICS */
 808             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
 809             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
 810
 811             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
 812
 813             /* Update potential sum for this i atom from the interaction with this j atom. */
 814             velec            = _mm256_and_pd(velec,cutoff_mask);
 815             velec            = _mm256_andnot_pd(dummy_mask,velec);
 816             velecsum         = _mm256_add_pd(velecsum,velec);
 817
 818             fscal            = felec;
 819
 820             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 821
 822             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 823
 824             /* Calculate temporary vectorial force */
 825             tx               = _mm256_mul_pd(fscal,dx01);
 826             ty               = _mm256_mul_pd(fscal,dy01);
 827             tz               = _mm256_mul_pd(fscal,dz01);
 828
 829             /* Update vectorial force */
 830             fix0             = _mm256_add_pd(fix0,tx);
 831             fiy0             = _mm256_add_pd(fiy0,ty);
 832             fiz0             = _mm256_add_pd(fiz0,tz);
 833
 834             fjx1             = _mm256_add_pd(fjx1,tx);
 835             fjy1             = _mm256_add_pd(fjy1,ty);
 836             fjz1             = _mm256_add_pd(fjz1,tz);
 837
 838             }
 839
 840             /**************************
 841              * CALCULATE INTERACTIONS *
 842              **************************/
 843
 844             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 845             {
 846
 847             /* REACTION-FIELD ELECTROSTATICS */
 848             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
 849             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
 850
 851             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
 852
 853             /* Update potential sum for this i atom from the interaction with this j atom. */
 854             velec            = _mm256_and_pd(velec,cutoff_mask);
 855             velec            = _mm256_andnot_pd(dummy_mask,velec);
 856             velecsum         = _mm256_add_pd(velecsum,velec);
 857
 858             fscal            = felec;
 859
 860             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 861
 862             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 863
 864             /* Calculate temporary vectorial force */
 865             tx               = _mm256_mul_pd(fscal,dx02);
 866             ty               = _mm256_mul_pd(fscal,dy02);
 867             tz               = _mm256_mul_pd(fscal,dz02);
 868
 869             /* Update vectorial force */
 870             fix0             = _mm256_add_pd(fix0,tx);
 871             fiy0             = _mm256_add_pd(fiy0,ty);
 872             fiz0             = _mm256_add_pd(fiz0,tz);
 873
 874             fjx2             = _mm256_add_pd(fjx2,tx);
 875             fjy2             = _mm256_add_pd(fjy2,ty);
 876             fjz2             = _mm256_add_pd(fjz2,tz);
 877
 878             }
 879
 880             /**************************
 881              * CALCULATE INTERACTIONS *
 882              **************************/
 883
 884             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 885             {
 886
 887             /* REACTION-FIELD ELECTROSTATICS */
 888             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 889             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 890
 891             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 892
 893             /* Update potential sum for this i atom from the interaction with this j atom. */
 894             velec            = _mm256_and_pd(velec,cutoff_mask);
 895             velec            = _mm256_andnot_pd(dummy_mask,velec);
 896             velecsum         = _mm256_add_pd(velecsum,velec);
 897
 898             fscal            = felec;
 899
 900             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 901
 902             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 903
 904             /* Calculate temporary vectorial force */
 905             tx               = _mm256_mul_pd(fscal,dx10);
 906             ty               = _mm256_mul_pd(fscal,dy10);
 907             tz               = _mm256_mul_pd(fscal,dz10);
 908
 909             /* Update vectorial force */
 910             fix1             = _mm256_add_pd(fix1,tx);
 911             fiy1             = _mm256_add_pd(fiy1,ty);
 912             fiz1             = _mm256_add_pd(fiz1,tz);
 913
 914             fjx0             = _mm256_add_pd(fjx0,tx);
 915             fjy0             = _mm256_add_pd(fjy0,ty);
 916             fjz0             = _mm256_add_pd(fjz0,tz);
 917
 918             }
 919
 920             /**************************
 921              * CALCULATE INTERACTIONS *
 922              **************************/
 923
 924             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 925             {
 926
 927             /* REACTION-FIELD ELECTROSTATICS */
 928             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 929             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 930
 931             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 932
 933             /* Update potential sum for this i atom from the interaction with this j atom. */
 934             velec            = _mm256_and_pd(velec,cutoff_mask);
 935             velec            = _mm256_andnot_pd(dummy_mask,velec);
 936             velecsum         = _mm256_add_pd(velecsum,velec);
 937
 938             fscal            = felec;
 939
 940             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 941
 942             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 943
 944             /* Calculate temporary vectorial force */
 945             tx               = _mm256_mul_pd(fscal,dx11);
 946             ty               = _mm256_mul_pd(fscal,dy11);
 947             tz               = _mm256_mul_pd(fscal,dz11);
 948
 949             /* Update vectorial force */
 950             fix1             = _mm256_add_pd(fix1,tx);
 951             fiy1             = _mm256_add_pd(fiy1,ty);
 952             fiz1             = _mm256_add_pd(fiz1,tz);
 953
 954             fjx1             = _mm256_add_pd(fjx1,tx);
 955             fjy1             = _mm256_add_pd(fjy1,ty);
 956             fjz1             = _mm256_add_pd(fjz1,tz);
 957
 958             }
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 965             {
 966
 967             /* REACTION-FIELD ELECTROSTATICS */
 968             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 969             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 970
 971             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 972
 973             /* Update potential sum for this i atom from the interaction with this j atom. */
 974             velec            = _mm256_and_pd(velec,cutoff_mask);
 975             velec            = _mm256_andnot_pd(dummy_mask,velec);
 976             velecsum         = _mm256_add_pd(velecsum,velec);
 977
 978             fscal            = felec;
 979
 980             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 981
 982             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 983
 984             /* Calculate temporary vectorial force */
 985             tx               = _mm256_mul_pd(fscal,dx12);
 986             ty               = _mm256_mul_pd(fscal,dy12);
 987             tz               = _mm256_mul_pd(fscal,dz12);
 988
 989             /* Update vectorial force */
 990             fix1             = _mm256_add_pd(fix1,tx);
 991             fiy1             = _mm256_add_pd(fiy1,ty);
 992             fiz1             = _mm256_add_pd(fiz1,tz);
 993
 994             fjx2             = _mm256_add_pd(fjx2,tx);
 995             fjy2             = _mm256_add_pd(fjy2,ty);
 996             fjz2             = _mm256_add_pd(fjz2,tz);
 997
 998             }
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1005             {
1006
1007             /* REACTION-FIELD ELECTROSTATICS */
1008             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
1009             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1010
1011             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1012
1013             /* Update potential sum for this i atom from the interaction with this j atom. */
1014             velec            = _mm256_and_pd(velec,cutoff_mask);
1015             velec            = _mm256_andnot_pd(dummy_mask,velec);
1016             velecsum         = _mm256_add_pd(velecsum,velec);
1017
1018             fscal            = felec;
1019
1020             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1021
1022             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1023
1024             /* Calculate temporary vectorial force */
1025             tx               = _mm256_mul_pd(fscal,dx20);
1026             ty               = _mm256_mul_pd(fscal,dy20);
1027             tz               = _mm256_mul_pd(fscal,dz20);
1028
1029             /* Update vectorial force */
1030             fix2             = _mm256_add_pd(fix2,tx);
1031             fiy2             = _mm256_add_pd(fiy2,ty);
1032             fiz2             = _mm256_add_pd(fiz2,tz);
1033
1034             fjx0             = _mm256_add_pd(fjx0,tx);
1035             fjy0             = _mm256_add_pd(fjy0,ty);
1036             fjz0             = _mm256_add_pd(fjz0,tz);
1037
1038             }
1039
1040             /**************************
1041              * CALCULATE INTERACTIONS *
1042              **************************/
1043
1044             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1045             {
1046
1047             /* REACTION-FIELD ELECTROSTATICS */
1048             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
1049             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1050
1051             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1052
1053             /* Update potential sum for this i atom from the interaction with this j atom. */
1054             velec            = _mm256_and_pd(velec,cutoff_mask);
1055             velec            = _mm256_andnot_pd(dummy_mask,velec);
1056             velecsum         = _mm256_add_pd(velecsum,velec);
1057
1058             fscal            = felec;
1059
1060             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1061
1062             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1063
1064             /* Calculate temporary vectorial force */
1065             tx               = _mm256_mul_pd(fscal,dx21);
1066             ty               = _mm256_mul_pd(fscal,dy21);
1067             tz               = _mm256_mul_pd(fscal,dz21);
1068
1069             /* Update vectorial force */
1070             fix2             = _mm256_add_pd(fix2,tx);
1071             fiy2             = _mm256_add_pd(fiy2,ty);
1072             fiz2             = _mm256_add_pd(fiz2,tz);
1073
1074             fjx1             = _mm256_add_pd(fjx1,tx);
1075             fjy1             = _mm256_add_pd(fjy1,ty);
1076             fjz1             = _mm256_add_pd(fjz1,tz);
1077
1078             }
1079
1080             /**************************
1081              * CALCULATE INTERACTIONS *
1082              **************************/
1083
1084             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1085             {
1086
1087             /* REACTION-FIELD ELECTROSTATICS */
1088             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
1089             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1090
1091             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1092
1093             /* Update potential sum for this i atom from the interaction with this j atom. */
1094             velec            = _mm256_and_pd(velec,cutoff_mask);
1095             velec            = _mm256_andnot_pd(dummy_mask,velec);
1096             velecsum         = _mm256_add_pd(velecsum,velec);
1097
1098             fscal            = felec;
1099
1100             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1101
1102             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1103
1104             /* Calculate temporary vectorial force */
1105             tx               = _mm256_mul_pd(fscal,dx22);
1106             ty               = _mm256_mul_pd(fscal,dy22);
1107             tz               = _mm256_mul_pd(fscal,dz22);
1108
1109             /* Update vectorial force */
1110             fix2             = _mm256_add_pd(fix2,tx);
1111             fiy2             = _mm256_add_pd(fiy2,ty);
1112             fiz2             = _mm256_add_pd(fiz2,tz);
1113
1114             fjx2             = _mm256_add_pd(fjx2,tx);
1115             fjy2             = _mm256_add_pd(fjy2,ty);
1116             fjz2             = _mm256_add_pd(fjz2,tz);
1117
1118             }
1119
1120             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1121             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1122             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1123             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1124
1125             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1126                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1127
1128             /* Inner loop uses 342 flops */
1129         }
1130
1131         /* End of innermost loop */
1132
1133         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1134                                                  f+i_coord_offset,fshift+i_shift_offset);
1135
1136         ggid                        = gid[iidx];
1137         /* Update potential energies */
1138         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1139         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1140
1141         /* Increment number of inner iterations */
1142         inneriter                  += j_index_end - j_index_start;
1143
1144         /* Outer loop uses 20 flops */
1145     }
1146
1147     /* Increment number of outer iterations */
1148     outeriter        += nri;
1149
1150     /* Update outer/inner flops */
1151
1152     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*342);
1153 }
1154 /*
1155  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_F_avx_256_double
1156  * Electrostatics interaction: ReactionField
1157  * VdW interaction:            LennardJones
1158  * Geometry:                   Water3-Water3
1159  * Calculate force/pot:        Force
1160  */
1161 void
1162 nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_F_avx_256_double
1163                     (t_nblist * gmx_restrict                nlist,
1164                      rvec * gmx_restrict                    xx,
1165                      rvec * gmx_restrict                    ff,
1166                      t_forcerec * gmx_restrict              fr,
1167                      t_mdatoms * gmx_restrict               mdatoms,
1168                      nb_kernel_data_t * gmx_restrict        kernel_data,
1169                      t_nrnb * gmx_restrict                  nrnb)
1170 {
1171     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1172      * just 0 for non-waters.
1173      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1174      * jnr indices corresponding to data put in the four positions in the SIMD register.
1175      */
1176     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1177     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1178     int              jnrA,jnrB,jnrC,jnrD;
1179     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1180     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1181     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1182     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1183     real             rcutoff_scalar;
1184     real             *shiftvec,*fshift,*x,*f;
1185     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1186     real             scratch[4*DIM];
1187     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1188     real *           vdwioffsetptr0;
1189     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1190     real *           vdwioffsetptr1;
1191     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1192     real *           vdwioffsetptr2;
1193     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1194     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1195     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1196     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1197     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1198     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1199     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1200     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1201     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1202     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1203     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1204     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1205     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1206     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1207     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1208     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1209     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1210     real             *charge;
1211     int              nvdwtype;
1212     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1213     int              *vdwtype;
1214     real             *vdwparam;
1215     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1216     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1217     __m256d          dummy_mask,cutoff_mask;
1218     __m128           tmpmask0,tmpmask1;
1219     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1220     __m256d          one     = _mm256_set1_pd(1.0);
1221     __m256d          two     = _mm256_set1_pd(2.0);
1222     x                = xx[0];
1223     f                = ff[0];
1224
1225     nri              = nlist->nri;
1226     iinr             = nlist->iinr;
1227     jindex           = nlist->jindex;
1228     jjnr             = nlist->jjnr;
1229     shiftidx         = nlist->shift;
1230     gid              = nlist->gid;
1231     shiftvec         = fr->shift_vec[0];
1232     fshift           = fr->fshift[0];
1233     facel            = _mm256_set1_pd(fr->epsfac);
1234     charge           = mdatoms->chargeA;
1235     krf              = _mm256_set1_pd(fr->ic->k_rf);
1236     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
1237     crf              = _mm256_set1_pd(fr->ic->c_rf);
1238     nvdwtype         = fr->ntype;
1239     vdwparam         = fr->nbfp;
1240     vdwtype          = mdatoms->typeA;
1241
1242     /* Setup water-specific parameters */
1243     inr              = nlist->iinr[0];
1244     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1245     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1246     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1247     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1248
1249     jq0              = _mm256_set1_pd(charge[inr+0]);
1250     jq1              = _mm256_set1_pd(charge[inr+1]);
1251     jq2              = _mm256_set1_pd(charge[inr+2]);
1252     vdwjidx0A        = 2*vdwtype[inr+0];
1253     qq00             = _mm256_mul_pd(iq0,jq0);
1254     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1255     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1256     qq01             = _mm256_mul_pd(iq0,jq1);
1257     qq02             = _mm256_mul_pd(iq0,jq2);
1258     qq10             = _mm256_mul_pd(iq1,jq0);
1259     qq11             = _mm256_mul_pd(iq1,jq1);
1260     qq12             = _mm256_mul_pd(iq1,jq2);
1261     qq20             = _mm256_mul_pd(iq2,jq0);
1262     qq21             = _mm256_mul_pd(iq2,jq1);
1263     qq22             = _mm256_mul_pd(iq2,jq2);
1264
1265     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1266     rcutoff_scalar   = fr->rcoulomb;
1267     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
1268     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
1269
1270     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
1271     rvdw             = _mm256_set1_pd(fr->rvdw);
1272
1273     /* Avoid stupid compiler warnings */
1274     jnrA = jnrB = jnrC = jnrD = 0;
1275     j_coord_offsetA = 0;
1276     j_coord_offsetB = 0;
1277     j_coord_offsetC = 0;
1278     j_coord_offsetD = 0;
1279
1280     outeriter        = 0;
1281     inneriter        = 0;
1282
1283     for(iidx=0;iidx<4*DIM;iidx++)
1284     {
1285         scratch[iidx] = 0.0;
1286     }
1287
1288     /* Start outer loop over neighborlists */
1289     for(iidx=0; iidx<nri; iidx++)
1290     {
1291         /* Load shift vector for this list */
1292         i_shift_offset   = DIM*shiftidx[iidx];
1293
1294         /* Load limits for loop over neighbors */
1295         j_index_start    = jindex[iidx];
1296         j_index_end      = jindex[iidx+1];
1297
1298         /* Get outer coordinate index */
1299         inr              = iinr[iidx];
1300         i_coord_offset   = DIM*inr;
1301
1302         /* Load i particle coords and add shift vector */
1303         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1304                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1305
1306         fix0             = _mm256_setzero_pd();
1307         fiy0             = _mm256_setzero_pd();
1308         fiz0             = _mm256_setzero_pd();
1309         fix1             = _mm256_setzero_pd();
1310         fiy1             = _mm256_setzero_pd();
1311         fiz1             = _mm256_setzero_pd();
1312         fix2             = _mm256_setzero_pd();
1313         fiy2             = _mm256_setzero_pd();
1314         fiz2             = _mm256_setzero_pd();
1315
1316         /* Start inner kernel loop */
1317         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1318         {
1319
1320             /* Get j neighbor index, and coordinate index */
1321             jnrA             = jjnr[jidx];
1322             jnrB             = jjnr[jidx+1];
1323             jnrC             = jjnr[jidx+2];
1324             jnrD             = jjnr[jidx+3];
1325             j_coord_offsetA  = DIM*jnrA;
1326             j_coord_offsetB  = DIM*jnrB;
1327             j_coord_offsetC  = DIM*jnrC;
1328             j_coord_offsetD  = DIM*jnrD;
1329
1330             /* load j atom coordinates */
1331             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1332                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1333                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1334
1335             /* Calculate displacement vector */
1336             dx00             = _mm256_sub_pd(ix0,jx0);
1337             dy00             = _mm256_sub_pd(iy0,jy0);
1338             dz00             = _mm256_sub_pd(iz0,jz0);
1339             dx01             = _mm256_sub_pd(ix0,jx1);
1340             dy01             = _mm256_sub_pd(iy0,jy1);
1341             dz01             = _mm256_sub_pd(iz0,jz1);
1342             dx02             = _mm256_sub_pd(ix0,jx2);
1343             dy02             = _mm256_sub_pd(iy0,jy2);
1344             dz02             = _mm256_sub_pd(iz0,jz2);
1345             dx10             = _mm256_sub_pd(ix1,jx0);
1346             dy10             = _mm256_sub_pd(iy1,jy0);
1347             dz10             = _mm256_sub_pd(iz1,jz0);
1348             dx11             = _mm256_sub_pd(ix1,jx1);
1349             dy11             = _mm256_sub_pd(iy1,jy1);
1350             dz11             = _mm256_sub_pd(iz1,jz1);
1351             dx12             = _mm256_sub_pd(ix1,jx2);
1352             dy12             = _mm256_sub_pd(iy1,jy2);
1353             dz12             = _mm256_sub_pd(iz1,jz2);
1354             dx20             = _mm256_sub_pd(ix2,jx0);
1355             dy20             = _mm256_sub_pd(iy2,jy0);
1356             dz20             = _mm256_sub_pd(iz2,jz0);
1357             dx21             = _mm256_sub_pd(ix2,jx1);
1358             dy21             = _mm256_sub_pd(iy2,jy1);
1359             dz21             = _mm256_sub_pd(iz2,jz1);
1360             dx22             = _mm256_sub_pd(ix2,jx2);
1361             dy22             = _mm256_sub_pd(iy2,jy2);
1362             dz22             = _mm256_sub_pd(iz2,jz2);
1363
1364             /* Calculate squared distance and things based on it */
1365             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1366             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1367             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1368             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1369             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1370             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1371             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1372             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1373             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1374
1375             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1376             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1377             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1378             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1379             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1380             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1381             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1382             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1383             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1384
1385             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1386             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1387             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1388             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1389             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1390             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1391             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1392             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1393             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1394
1395             fjx0             = _mm256_setzero_pd();
1396             fjy0             = _mm256_setzero_pd();
1397             fjz0             = _mm256_setzero_pd();
1398             fjx1             = _mm256_setzero_pd();
1399             fjy1             = _mm256_setzero_pd();
1400             fjz1             = _mm256_setzero_pd();
1401             fjx2             = _mm256_setzero_pd();
1402             fjy2             = _mm256_setzero_pd();
1403             fjz2             = _mm256_setzero_pd();
1404
1405             /**************************
1406              * CALCULATE INTERACTIONS *
1407              **************************/
1408
1409             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1410             {
1411
1412             /* REACTION-FIELD ELECTROSTATICS */
1413             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1414
1415             /* LENNARD-JONES DISPERSION/REPULSION */
1416
1417             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1418             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1419
1420             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1421
1422             fscal            = _mm256_add_pd(felec,fvdw);
1423
1424             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1425
1426             /* Calculate temporary vectorial force */
1427             tx               = _mm256_mul_pd(fscal,dx00);
1428             ty               = _mm256_mul_pd(fscal,dy00);
1429             tz               = _mm256_mul_pd(fscal,dz00);
1430
1431             /* Update vectorial force */
1432             fix0             = _mm256_add_pd(fix0,tx);
1433             fiy0             = _mm256_add_pd(fiy0,ty);
1434             fiz0             = _mm256_add_pd(fiz0,tz);
1435
1436             fjx0             = _mm256_add_pd(fjx0,tx);
1437             fjy0             = _mm256_add_pd(fjy0,ty);
1438             fjz0             = _mm256_add_pd(fjz0,tz);
1439
1440             }
1441
1442             /**************************
1443              * CALCULATE INTERACTIONS *
1444              **************************/
1445
1446             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1447             {
1448
1449             /* REACTION-FIELD ELECTROSTATICS */
1450             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1451
1452             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
1453
1454             fscal            = felec;
1455
1456             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1457
1458             /* Calculate temporary vectorial force */
1459             tx               = _mm256_mul_pd(fscal,dx01);
1460             ty               = _mm256_mul_pd(fscal,dy01);
1461             tz               = _mm256_mul_pd(fscal,dz01);
1462
1463             /* Update vectorial force */
1464             fix0             = _mm256_add_pd(fix0,tx);
1465             fiy0             = _mm256_add_pd(fiy0,ty);
1466             fiz0             = _mm256_add_pd(fiz0,tz);
1467
1468             fjx1             = _mm256_add_pd(fjx1,tx);
1469             fjy1             = _mm256_add_pd(fjy1,ty);
1470             fjz1             = _mm256_add_pd(fjz1,tz);
1471
1472             }
1473
1474             /**************************
1475              * CALCULATE INTERACTIONS *
1476              **************************/
1477
1478             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1479             {
1480
1481             /* REACTION-FIELD ELECTROSTATICS */
1482             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1483
1484             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
1485
1486             fscal            = felec;
1487
1488             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1489
1490             /* Calculate temporary vectorial force */
1491             tx               = _mm256_mul_pd(fscal,dx02);
1492             ty               = _mm256_mul_pd(fscal,dy02);
1493             tz               = _mm256_mul_pd(fscal,dz02);
1494
1495             /* Update vectorial force */
1496             fix0             = _mm256_add_pd(fix0,tx);
1497             fiy0             = _mm256_add_pd(fiy0,ty);
1498             fiz0             = _mm256_add_pd(fiz0,tz);
1499
1500             fjx2             = _mm256_add_pd(fjx2,tx);
1501             fjy2             = _mm256_add_pd(fjy2,ty);
1502             fjz2             = _mm256_add_pd(fjz2,tz);
1503
1504             }
1505
1506             /**************************
1507              * CALCULATE INTERACTIONS *
1508              **************************/
1509
1510             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1511             {
1512
1513             /* REACTION-FIELD ELECTROSTATICS */
1514             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1515
1516             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1517
1518             fscal            = felec;
1519
1520             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1521
1522             /* Calculate temporary vectorial force */
1523             tx               = _mm256_mul_pd(fscal,dx10);
1524             ty               = _mm256_mul_pd(fscal,dy10);
1525             tz               = _mm256_mul_pd(fscal,dz10);
1526
1527             /* Update vectorial force */
1528             fix1             = _mm256_add_pd(fix1,tx);
1529             fiy1             = _mm256_add_pd(fiy1,ty);
1530             fiz1             = _mm256_add_pd(fiz1,tz);
1531
1532             fjx0             = _mm256_add_pd(fjx0,tx);
1533             fjy0             = _mm256_add_pd(fjy0,ty);
1534             fjz0             = _mm256_add_pd(fjz0,tz);
1535
1536             }
1537
1538             /**************************
1539              * CALCULATE INTERACTIONS *
1540              **************************/
1541
1542             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1543             {
1544
1545             /* REACTION-FIELD ELECTROSTATICS */
1546             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1547
1548             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1549
1550             fscal            = felec;
1551
1552             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1553
1554             /* Calculate temporary vectorial force */
1555             tx               = _mm256_mul_pd(fscal,dx11);
1556             ty               = _mm256_mul_pd(fscal,dy11);
1557             tz               = _mm256_mul_pd(fscal,dz11);
1558
1559             /* Update vectorial force */
1560             fix1             = _mm256_add_pd(fix1,tx);
1561             fiy1             = _mm256_add_pd(fiy1,ty);
1562             fiz1             = _mm256_add_pd(fiz1,tz);
1563
1564             fjx1             = _mm256_add_pd(fjx1,tx);
1565             fjy1             = _mm256_add_pd(fjy1,ty);
1566             fjz1             = _mm256_add_pd(fjz1,tz);
1567
1568             }
1569
1570             /**************************
1571              * CALCULATE INTERACTIONS *
1572              **************************/
1573
1574             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1575             {
1576
1577             /* REACTION-FIELD ELECTROSTATICS */
1578             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1579
1580             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1581
1582             fscal            = felec;
1583
1584             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1585
1586             /* Calculate temporary vectorial force */
1587             tx               = _mm256_mul_pd(fscal,dx12);
1588             ty               = _mm256_mul_pd(fscal,dy12);
1589             tz               = _mm256_mul_pd(fscal,dz12);
1590
1591             /* Update vectorial force */
1592             fix1             = _mm256_add_pd(fix1,tx);
1593             fiy1             = _mm256_add_pd(fiy1,ty);
1594             fiz1             = _mm256_add_pd(fiz1,tz);
1595
1596             fjx2             = _mm256_add_pd(fjx2,tx);
1597             fjy2             = _mm256_add_pd(fjy2,ty);
1598             fjz2             = _mm256_add_pd(fjz2,tz);
1599
1600             }
1601
1602             /**************************
1603              * CALCULATE INTERACTIONS *
1604              **************************/
1605
1606             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1607             {
1608
1609             /* REACTION-FIELD ELECTROSTATICS */
1610             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1611
1612             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1613
1614             fscal            = felec;
1615
1616             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1617
1618             /* Calculate temporary vectorial force */
1619             tx               = _mm256_mul_pd(fscal,dx20);
1620             ty               = _mm256_mul_pd(fscal,dy20);
1621             tz               = _mm256_mul_pd(fscal,dz20);
1622
1623             /* Update vectorial force */
1624             fix2             = _mm256_add_pd(fix2,tx);
1625             fiy2             = _mm256_add_pd(fiy2,ty);
1626             fiz2             = _mm256_add_pd(fiz2,tz);
1627
1628             fjx0             = _mm256_add_pd(fjx0,tx);
1629             fjy0             = _mm256_add_pd(fjy0,ty);
1630             fjz0             = _mm256_add_pd(fjz0,tz);
1631
1632             }
1633
1634             /**************************
1635              * CALCULATE INTERACTIONS *
1636              **************************/
1637
1638             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1639             {
1640
1641             /* REACTION-FIELD ELECTROSTATICS */
1642             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1643
1644             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1645
1646             fscal            = felec;
1647
1648             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1649
1650             /* Calculate temporary vectorial force */
1651             tx               = _mm256_mul_pd(fscal,dx21);
1652             ty               = _mm256_mul_pd(fscal,dy21);
1653             tz               = _mm256_mul_pd(fscal,dz21);
1654
1655             /* Update vectorial force */
1656             fix2             = _mm256_add_pd(fix2,tx);
1657             fiy2             = _mm256_add_pd(fiy2,ty);
1658             fiz2             = _mm256_add_pd(fiz2,tz);
1659
1660             fjx1             = _mm256_add_pd(fjx1,tx);
1661             fjy1             = _mm256_add_pd(fjy1,ty);
1662             fjz1             = _mm256_add_pd(fjz1,tz);
1663
1664             }
1665
1666             /**************************
1667              * CALCULATE INTERACTIONS *
1668              **************************/
1669
1670             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1671             {
1672
1673             /* REACTION-FIELD ELECTROSTATICS */
1674             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1675
1676             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1677
1678             fscal            = felec;
1679
1680             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1681
1682             /* Calculate temporary vectorial force */
1683             tx               = _mm256_mul_pd(fscal,dx22);
1684             ty               = _mm256_mul_pd(fscal,dy22);
1685             tz               = _mm256_mul_pd(fscal,dz22);
1686
1687             /* Update vectorial force */
1688             fix2             = _mm256_add_pd(fix2,tx);
1689             fiy2             = _mm256_add_pd(fiy2,ty);
1690             fiz2             = _mm256_add_pd(fiz2,tz);
1691
1692             fjx2             = _mm256_add_pd(fjx2,tx);
1693             fjy2             = _mm256_add_pd(fjy2,ty);
1694             fjz2             = _mm256_add_pd(fjz2,tz);
1695
1696             }
1697
1698             fjptrA             = f+j_coord_offsetA;
1699             fjptrB             = f+j_coord_offsetB;
1700             fjptrC             = f+j_coord_offsetC;
1701             fjptrD             = f+j_coord_offsetD;
1702
1703             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1704                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1705
1706             /* Inner loop uses 277 flops */
1707         }
1708
1709         if(jidx<j_index_end)
1710         {
1711
1712             /* Get j neighbor index, and coordinate index */
1713             jnrlistA         = jjnr[jidx];
1714             jnrlistB         = jjnr[jidx+1];
1715             jnrlistC         = jjnr[jidx+2];
1716             jnrlistD         = jjnr[jidx+3];
1717             /* Sign of each element will be negative for non-real atoms.
1718              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1719              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1720              */
1721             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1722
1723             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1724             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1725             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1726
1727             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1728             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1729             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1730             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1731             j_coord_offsetA  = DIM*jnrA;
1732             j_coord_offsetB  = DIM*jnrB;
1733             j_coord_offsetC  = DIM*jnrC;
1734             j_coord_offsetD  = DIM*jnrD;
1735
1736             /* load j atom coordinates */
1737             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1738                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1739                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1740
1741             /* Calculate displacement vector */
1742             dx00             = _mm256_sub_pd(ix0,jx0);
1743             dy00             = _mm256_sub_pd(iy0,jy0);
1744             dz00             = _mm256_sub_pd(iz0,jz0);
1745             dx01             = _mm256_sub_pd(ix0,jx1);
1746             dy01             = _mm256_sub_pd(iy0,jy1);
1747             dz01             = _mm256_sub_pd(iz0,jz1);
1748             dx02             = _mm256_sub_pd(ix0,jx2);
1749             dy02             = _mm256_sub_pd(iy0,jy2);
1750             dz02             = _mm256_sub_pd(iz0,jz2);
1751             dx10             = _mm256_sub_pd(ix1,jx0);
1752             dy10             = _mm256_sub_pd(iy1,jy0);
1753             dz10             = _mm256_sub_pd(iz1,jz0);
1754             dx11             = _mm256_sub_pd(ix1,jx1);
1755             dy11             = _mm256_sub_pd(iy1,jy1);
1756             dz11             = _mm256_sub_pd(iz1,jz1);
1757             dx12             = _mm256_sub_pd(ix1,jx2);
1758             dy12             = _mm256_sub_pd(iy1,jy2);
1759             dz12             = _mm256_sub_pd(iz1,jz2);
1760             dx20             = _mm256_sub_pd(ix2,jx0);
1761             dy20             = _mm256_sub_pd(iy2,jy0);
1762             dz20             = _mm256_sub_pd(iz2,jz0);
1763             dx21             = _mm256_sub_pd(ix2,jx1);
1764             dy21             = _mm256_sub_pd(iy2,jy1);
1765             dz21             = _mm256_sub_pd(iz2,jz1);
1766             dx22             = _mm256_sub_pd(ix2,jx2);
1767             dy22             = _mm256_sub_pd(iy2,jy2);
1768             dz22             = _mm256_sub_pd(iz2,jz2);
1769
1770             /* Calculate squared distance and things based on it */
1771             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1772             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1773             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1774             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1775             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1776             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1777             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1778             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1779             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1780
1781             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1782             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1783             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1784             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1785             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1786             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1787             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1788             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1789             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1790
1791             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1792             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1793             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1794             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1795             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1796             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1797             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1798             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1799             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1800
1801             fjx0             = _mm256_setzero_pd();
1802             fjy0             = _mm256_setzero_pd();
1803             fjz0             = _mm256_setzero_pd();
1804             fjx1             = _mm256_setzero_pd();
1805             fjy1             = _mm256_setzero_pd();
1806             fjz1             = _mm256_setzero_pd();
1807             fjx2             = _mm256_setzero_pd();
1808             fjy2             = _mm256_setzero_pd();
1809             fjz2             = _mm256_setzero_pd();
1810
1811             /**************************
1812              * CALCULATE INTERACTIONS *
1813              **************************/
1814
1815             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1816             {
1817
1818             /* REACTION-FIELD ELECTROSTATICS */
1819             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1820
1821             /* LENNARD-JONES DISPERSION/REPULSION */
1822
1823             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1824             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1825
1826             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1827
1828             fscal            = _mm256_add_pd(felec,fvdw);
1829
1830             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1831
1832             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1833
1834             /* Calculate temporary vectorial force */
1835             tx               = _mm256_mul_pd(fscal,dx00);
1836             ty               = _mm256_mul_pd(fscal,dy00);
1837             tz               = _mm256_mul_pd(fscal,dz00);
1838
1839             /* Update vectorial force */
1840             fix0             = _mm256_add_pd(fix0,tx);
1841             fiy0             = _mm256_add_pd(fiy0,ty);
1842             fiz0             = _mm256_add_pd(fiz0,tz);
1843
1844             fjx0             = _mm256_add_pd(fjx0,tx);
1845             fjy0             = _mm256_add_pd(fjy0,ty);
1846             fjz0             = _mm256_add_pd(fjz0,tz);
1847
1848             }
1849
1850             /**************************
1851              * CALCULATE INTERACTIONS *
1852              **************************/
1853
1854             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1855             {
1856
1857             /* REACTION-FIELD ELECTROSTATICS */
1858             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1859
1860             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
1861
1862             fscal            = felec;
1863
1864             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1865
1866             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1867
1868             /* Calculate temporary vectorial force */
1869             tx               = _mm256_mul_pd(fscal,dx01);
1870             ty               = _mm256_mul_pd(fscal,dy01);
1871             tz               = _mm256_mul_pd(fscal,dz01);
1872
1873             /* Update vectorial force */
1874             fix0             = _mm256_add_pd(fix0,tx);
1875             fiy0             = _mm256_add_pd(fiy0,ty);
1876             fiz0             = _mm256_add_pd(fiz0,tz);
1877
1878             fjx1             = _mm256_add_pd(fjx1,tx);
1879             fjy1             = _mm256_add_pd(fjy1,ty);
1880             fjz1             = _mm256_add_pd(fjz1,tz);
1881
1882             }
1883
1884             /**************************
1885              * CALCULATE INTERACTIONS *
1886              **************************/
1887
1888             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1889             {
1890
1891             /* REACTION-FIELD ELECTROSTATICS */
1892             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1893
1894             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
1895
1896             fscal            = felec;
1897
1898             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1899
1900             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1901
1902             /* Calculate temporary vectorial force */
1903             tx               = _mm256_mul_pd(fscal,dx02);
1904             ty               = _mm256_mul_pd(fscal,dy02);
1905             tz               = _mm256_mul_pd(fscal,dz02);
1906
1907             /* Update vectorial force */
1908             fix0             = _mm256_add_pd(fix0,tx);
1909             fiy0             = _mm256_add_pd(fiy0,ty);
1910             fiz0             = _mm256_add_pd(fiz0,tz);
1911
1912             fjx2             = _mm256_add_pd(fjx2,tx);
1913             fjy2             = _mm256_add_pd(fjy2,ty);
1914             fjz2             = _mm256_add_pd(fjz2,tz);
1915
1916             }
1917
1918             /**************************
1919              * CALCULATE INTERACTIONS *
1920              **************************/
1921
1922             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1923             {
1924
1925             /* REACTION-FIELD ELECTROSTATICS */
1926             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1927
1928             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1929
1930             fscal            = felec;
1931
1932             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1933
1934             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1935
1936             /* Calculate temporary vectorial force */
1937             tx               = _mm256_mul_pd(fscal,dx10);
1938             ty               = _mm256_mul_pd(fscal,dy10);
1939             tz               = _mm256_mul_pd(fscal,dz10);
1940
1941             /* Update vectorial force */
1942             fix1             = _mm256_add_pd(fix1,tx);
1943             fiy1             = _mm256_add_pd(fiy1,ty);
1944             fiz1             = _mm256_add_pd(fiz1,tz);
1945
1946             fjx0             = _mm256_add_pd(fjx0,tx);
1947             fjy0             = _mm256_add_pd(fjy0,ty);
1948             fjz0             = _mm256_add_pd(fjz0,tz);
1949
1950             }
1951
1952             /**************************
1953              * CALCULATE INTERACTIONS *
1954              **************************/
1955
1956             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1957             {
1958
1959             /* REACTION-FIELD ELECTROSTATICS */
1960             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1961
1962             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1963
1964             fscal            = felec;
1965
1966             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1967
1968             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1969
1970             /* Calculate temporary vectorial force */
1971             tx               = _mm256_mul_pd(fscal,dx11);
1972             ty               = _mm256_mul_pd(fscal,dy11);
1973             tz               = _mm256_mul_pd(fscal,dz11);
1974
1975             /* Update vectorial force */
1976             fix1             = _mm256_add_pd(fix1,tx);
1977             fiy1             = _mm256_add_pd(fiy1,ty);
1978             fiz1             = _mm256_add_pd(fiz1,tz);
1979
1980             fjx1             = _mm256_add_pd(fjx1,tx);
1981             fjy1             = _mm256_add_pd(fjy1,ty);
1982             fjz1             = _mm256_add_pd(fjz1,tz);
1983
1984             }
1985
1986             /**************************
1987              * CALCULATE INTERACTIONS *
1988              **************************/
1989
1990             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1991             {
1992
1993             /* REACTION-FIELD ELECTROSTATICS */
1994             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1995
1996             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1997
1998             fscal            = felec;
1999
2000             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2001
2002             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2003
2004             /* Calculate temporary vectorial force */
2005             tx               = _mm256_mul_pd(fscal,dx12);
2006             ty               = _mm256_mul_pd(fscal,dy12);
2007             tz               = _mm256_mul_pd(fscal,dz12);
2008
2009             /* Update vectorial force */
2010             fix1             = _mm256_add_pd(fix1,tx);
2011             fiy1             = _mm256_add_pd(fiy1,ty);
2012             fiz1             = _mm256_add_pd(fiz1,tz);
2013
2014             fjx2             = _mm256_add_pd(fjx2,tx);
2015             fjy2             = _mm256_add_pd(fjy2,ty);
2016             fjz2             = _mm256_add_pd(fjz2,tz);
2017
2018             }
2019
2020             /**************************
2021              * CALCULATE INTERACTIONS *
2022              **************************/
2023
2024             if (gmx_mm256_any_lt(rsq20,rcutoff2))
2025             {
2026
2027             /* REACTION-FIELD ELECTROSTATICS */
2028             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
2029
2030             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
2031
2032             fscal            = felec;
2033
2034             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2035
2036             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2037
2038             /* Calculate temporary vectorial force */
2039             tx               = _mm256_mul_pd(fscal,dx20);
2040             ty               = _mm256_mul_pd(fscal,dy20);
2041             tz               = _mm256_mul_pd(fscal,dz20);
2042
2043             /* Update vectorial force */
2044             fix2             = _mm256_add_pd(fix2,tx);
2045             fiy2             = _mm256_add_pd(fiy2,ty);
2046             fiz2             = _mm256_add_pd(fiz2,tz);
2047
2048             fjx0             = _mm256_add_pd(fjx0,tx);
2049             fjy0             = _mm256_add_pd(fjy0,ty);
2050             fjz0             = _mm256_add_pd(fjz0,tz);
2051
2052             }
2053
2054             /**************************
2055              * CALCULATE INTERACTIONS *
2056              **************************/
2057
2058             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2059             {
2060
2061             /* REACTION-FIELD ELECTROSTATICS */
2062             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
2063
2064             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2065
2066             fscal            = felec;
2067
2068             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2069
2070             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2071
2072             /* Calculate temporary vectorial force */
2073             tx               = _mm256_mul_pd(fscal,dx21);
2074             ty               = _mm256_mul_pd(fscal,dy21);
2075             tz               = _mm256_mul_pd(fscal,dz21);
2076
2077             /* Update vectorial force */
2078             fix2             = _mm256_add_pd(fix2,tx);
2079             fiy2             = _mm256_add_pd(fiy2,ty);
2080             fiz2             = _mm256_add_pd(fiz2,tz);
2081
2082             fjx1             = _mm256_add_pd(fjx1,tx);
2083             fjy1             = _mm256_add_pd(fjy1,ty);
2084             fjz1             = _mm256_add_pd(fjz1,tz);
2085
2086             }
2087
2088             /**************************
2089              * CALCULATE INTERACTIONS *
2090              **************************/
2091
2092             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2093             {
2094
2095             /* REACTION-FIELD ELECTROSTATICS */
2096             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
2097
2098             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2099
2100             fscal            = felec;
2101
2102             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2103
2104             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2105
2106             /* Calculate temporary vectorial force */
2107             tx               = _mm256_mul_pd(fscal,dx22);
2108             ty               = _mm256_mul_pd(fscal,dy22);
2109             tz               = _mm256_mul_pd(fscal,dz22);
2110
2111             /* Update vectorial force */
2112             fix2             = _mm256_add_pd(fix2,tx);
2113             fiy2             = _mm256_add_pd(fiy2,ty);
2114             fiz2             = _mm256_add_pd(fiz2,tz);
2115
2116             fjx2             = _mm256_add_pd(fjx2,tx);
2117             fjy2             = _mm256_add_pd(fjy2,ty);
2118             fjz2             = _mm256_add_pd(fjz2,tz);
2119
2120             }
2121
2122             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2123             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2124             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2125             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2126
2127             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2128                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2129
2130             /* Inner loop uses 277 flops */
2131         }
2132
2133         /* End of innermost loop */
2134
2135         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2136                                                  f+i_coord_offset,fshift+i_shift_offset);
2137
2138         /* Increment number of inner iterations */
2139         inneriter                  += j_index_end - j_index_start;
2140
2141         /* Outer loop uses 18 flops */
2142     }
2143
2144     /* Increment number of outer iterations */
2145     outeriter        += nri;
2146
2147     /* Update outer/inner flops */
2148
2149     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*277);
2150 }