src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single 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_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_VF_avx_256_single
  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_single
  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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  81     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  82     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  83     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  84     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  86     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  87     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  88     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  89     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  90     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  92     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  93     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 100     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 101     __m256           dummy_mask,cutoff_mask;
 102     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 103     __m256           one     = _mm256_set1_ps(1.0);
 104     __m256           two     = _mm256_set1_ps(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm256_set1_ps(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118     krf              = _mm256_set1_ps(fr->ic->k_rf);
 119     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 120     crf              = _mm256_set1_ps(fr->ic->c_rf);
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     /* Setup water-specific parameters */
 126     inr              = nlist->iinr[0];
 127     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 128     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 129     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 130     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 131
 132     jq0              = _mm256_set1_ps(charge[inr+0]);
 133     jq1              = _mm256_set1_ps(charge[inr+1]);
 134     jq2              = _mm256_set1_ps(charge[inr+2]);
 135     vdwjidx0A        = 2*vdwtype[inr+0];
 136     qq00             = _mm256_mul_ps(iq0,jq0);
 137     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 138     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 139     qq01             = _mm256_mul_ps(iq0,jq1);
 140     qq02             = _mm256_mul_ps(iq0,jq2);
 141     qq10             = _mm256_mul_ps(iq1,jq0);
 142     qq11             = _mm256_mul_ps(iq1,jq1);
 143     qq12             = _mm256_mul_ps(iq1,jq2);
 144     qq20             = _mm256_mul_ps(iq2,jq0);
 145     qq21             = _mm256_mul_ps(iq2,jq1);
 146     qq22             = _mm256_mul_ps(iq2,jq2);
 147
 148     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 149     rcutoff_scalar   = fr->rcoulomb;
 150     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 151     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 152
 153     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 154     rvdw             = _mm256_set1_ps(fr->rvdw);
 155
 156     /* Avoid stupid compiler warnings */
 157     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 158     j_coord_offsetA = 0;
 159     j_coord_offsetB = 0;
 160     j_coord_offsetC = 0;
 161     j_coord_offsetD = 0;
 162     j_coord_offsetE = 0;
 163     j_coord_offsetF = 0;
 164     j_coord_offsetG = 0;
 165     j_coord_offsetH = 0;
 166
 167     outeriter        = 0;
 168     inneriter        = 0;
 169
 170     for(iidx=0;iidx<4*DIM;iidx++)
 171     {
 172         scratch[iidx] = 0.0;
 173     }
 174
 175     /* Start outer loop over neighborlists */
 176     for(iidx=0; iidx<nri; iidx++)
 177     {
 178         /* Load shift vector for this list */
 179         i_shift_offset   = DIM*shiftidx[iidx];
 180
 181         /* Load limits for loop over neighbors */
 182         j_index_start    = jindex[iidx];
 183         j_index_end      = jindex[iidx+1];
 184
 185         /* Get outer coordinate index */
 186         inr              = iinr[iidx];
 187         i_coord_offset   = DIM*inr;
 188
 189         /* Load i particle coords and add shift vector */
 190         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 191                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 192
 193         fix0             = _mm256_setzero_ps();
 194         fiy0             = _mm256_setzero_ps();
 195         fiz0             = _mm256_setzero_ps();
 196         fix1             = _mm256_setzero_ps();
 197         fiy1             = _mm256_setzero_ps();
 198         fiz1             = _mm256_setzero_ps();
 199         fix2             = _mm256_setzero_ps();
 200         fiy2             = _mm256_setzero_ps();
 201         fiz2             = _mm256_setzero_ps();
 202
 203         /* Reset potential sums */
 204         velecsum         = _mm256_setzero_ps();
 205         vvdwsum          = _mm256_setzero_ps();
 206
 207         /* Start inner kernel loop */
 208         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 209         {
 210
 211             /* Get j neighbor index, and coordinate index */
 212             jnrA             = jjnr[jidx];
 213             jnrB             = jjnr[jidx+1];
 214             jnrC             = jjnr[jidx+2];
 215             jnrD             = jjnr[jidx+3];
 216             jnrE             = jjnr[jidx+4];
 217             jnrF             = jjnr[jidx+5];
 218             jnrG             = jjnr[jidx+6];
 219             jnrH             = jjnr[jidx+7];
 220             j_coord_offsetA  = DIM*jnrA;
 221             j_coord_offsetB  = DIM*jnrB;
 222             j_coord_offsetC  = DIM*jnrC;
 223             j_coord_offsetD  = DIM*jnrD;
 224             j_coord_offsetE  = DIM*jnrE;
 225             j_coord_offsetF  = DIM*jnrF;
 226             j_coord_offsetG  = DIM*jnrG;
 227             j_coord_offsetH  = DIM*jnrH;
 228
 229             /* load j atom coordinates */
 230             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 231                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 232                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 233                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 234                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 235
 236             /* Calculate displacement vector */
 237             dx00             = _mm256_sub_ps(ix0,jx0);
 238             dy00             = _mm256_sub_ps(iy0,jy0);
 239             dz00             = _mm256_sub_ps(iz0,jz0);
 240             dx01             = _mm256_sub_ps(ix0,jx1);
 241             dy01             = _mm256_sub_ps(iy0,jy1);
 242             dz01             = _mm256_sub_ps(iz0,jz1);
 243             dx02             = _mm256_sub_ps(ix0,jx2);
 244             dy02             = _mm256_sub_ps(iy0,jy2);
 245             dz02             = _mm256_sub_ps(iz0,jz2);
 246             dx10             = _mm256_sub_ps(ix1,jx0);
 247             dy10             = _mm256_sub_ps(iy1,jy0);
 248             dz10             = _mm256_sub_ps(iz1,jz0);
 249             dx11             = _mm256_sub_ps(ix1,jx1);
 250             dy11             = _mm256_sub_ps(iy1,jy1);
 251             dz11             = _mm256_sub_ps(iz1,jz1);
 252             dx12             = _mm256_sub_ps(ix1,jx2);
 253             dy12             = _mm256_sub_ps(iy1,jy2);
 254             dz12             = _mm256_sub_ps(iz1,jz2);
 255             dx20             = _mm256_sub_ps(ix2,jx0);
 256             dy20             = _mm256_sub_ps(iy2,jy0);
 257             dz20             = _mm256_sub_ps(iz2,jz0);
 258             dx21             = _mm256_sub_ps(ix2,jx1);
 259             dy21             = _mm256_sub_ps(iy2,jy1);
 260             dz21             = _mm256_sub_ps(iz2,jz1);
 261             dx22             = _mm256_sub_ps(ix2,jx2);
 262             dy22             = _mm256_sub_ps(iy2,jy2);
 263             dz22             = _mm256_sub_ps(iz2,jz2);
 264
 265             /* Calculate squared distance and things based on it */
 266             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 267             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 268             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 269             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 270             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 271             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 272             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 273             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 274             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 275
 276             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 277             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 278             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 279             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 280             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 281             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 282             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 283             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 284             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 285
 286             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 287             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 288             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 289             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 290             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 291             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 292             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 293             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 294             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 295
 296             fjx0             = _mm256_setzero_ps();
 297             fjy0             = _mm256_setzero_ps();
 298             fjz0             = _mm256_setzero_ps();
 299             fjx1             = _mm256_setzero_ps();
 300             fjy1             = _mm256_setzero_ps();
 301             fjz1             = _mm256_setzero_ps();
 302             fjx2             = _mm256_setzero_ps();
 303             fjy2             = _mm256_setzero_ps();
 304             fjz2             = _mm256_setzero_ps();
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 311             {
 312
 313             /* REACTION-FIELD ELECTROSTATICS */
 314             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 315             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 316
 317             /* LENNARD-JONES DISPERSION/REPULSION */
 318
 319             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 320             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 321             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 322             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 323                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 324             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 325
 326             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velec            = _mm256_and_ps(velec,cutoff_mask);
 330             velecsum         = _mm256_add_ps(velecsum,velec);
 331             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 332             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 333
 334             fscal            = _mm256_add_ps(felec,fvdw);
 335
 336             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 337
 338             /* Calculate temporary vectorial force */
 339             tx               = _mm256_mul_ps(fscal,dx00);
 340             ty               = _mm256_mul_ps(fscal,dy00);
 341             tz               = _mm256_mul_ps(fscal,dz00);
 342
 343             /* Update vectorial force */
 344             fix0             = _mm256_add_ps(fix0,tx);
 345             fiy0             = _mm256_add_ps(fiy0,ty);
 346             fiz0             = _mm256_add_ps(fiz0,tz);
 347
 348             fjx0             = _mm256_add_ps(fjx0,tx);
 349             fjy0             = _mm256_add_ps(fjy0,ty);
 350             fjz0             = _mm256_add_ps(fjz0,tz);
 351
 352             }
 353
 354             /**************************
 355              * CALCULATE INTERACTIONS *
 356              **************************/
 357
 358             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 359             {
 360
 361             /* REACTION-FIELD ELECTROSTATICS */
 362             velec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
 363             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
 364
 365             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velec            = _mm256_and_ps(velec,cutoff_mask);
 369             velecsum         = _mm256_add_ps(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 374
 375             /* Calculate temporary vectorial force */
 376             tx               = _mm256_mul_ps(fscal,dx01);
 377             ty               = _mm256_mul_ps(fscal,dy01);
 378             tz               = _mm256_mul_ps(fscal,dz01);
 379
 380             /* Update vectorial force */
 381             fix0             = _mm256_add_ps(fix0,tx);
 382             fiy0             = _mm256_add_ps(fiy0,ty);
 383             fiz0             = _mm256_add_ps(fiz0,tz);
 384
 385             fjx1             = _mm256_add_ps(fjx1,tx);
 386             fjy1             = _mm256_add_ps(fjy1,ty);
 387             fjz1             = _mm256_add_ps(fjz1,tz);
 388
 389             }
 390
 391             /**************************
 392              * CALCULATE INTERACTIONS *
 393              **************************/
 394
 395             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 396             {
 397
 398             /* REACTION-FIELD ELECTROSTATICS */
 399             velec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
 400             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
 401
 402             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velec            = _mm256_and_ps(velec,cutoff_mask);
 406             velecsum         = _mm256_add_ps(velecsum,velec);
 407
 408             fscal            = felec;
 409
 410             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 411
 412             /* Calculate temporary vectorial force */
 413             tx               = _mm256_mul_ps(fscal,dx02);
 414             ty               = _mm256_mul_ps(fscal,dy02);
 415             tz               = _mm256_mul_ps(fscal,dz02);
 416
 417             /* Update vectorial force */
 418             fix0             = _mm256_add_ps(fix0,tx);
 419             fiy0             = _mm256_add_ps(fiy0,ty);
 420             fiz0             = _mm256_add_ps(fiz0,tz);
 421
 422             fjx2             = _mm256_add_ps(fjx2,tx);
 423             fjy2             = _mm256_add_ps(fjy2,ty);
 424             fjz2             = _mm256_add_ps(fjz2,tz);
 425
 426             }
 427
 428             /**************************
 429              * CALCULATE INTERACTIONS *
 430              **************************/
 431
 432             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 433             {
 434
 435             /* REACTION-FIELD ELECTROSTATICS */
 436             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 437             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 438
 439             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 440
 441             /* Update potential sum for this i atom from the interaction with this j atom. */
 442             velec            = _mm256_and_ps(velec,cutoff_mask);
 443             velecsum         = _mm256_add_ps(velecsum,velec);
 444
 445             fscal            = felec;
 446
 447             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 448
 449             /* Calculate temporary vectorial force */
 450             tx               = _mm256_mul_ps(fscal,dx10);
 451             ty               = _mm256_mul_ps(fscal,dy10);
 452             tz               = _mm256_mul_ps(fscal,dz10);
 453
 454             /* Update vectorial force */
 455             fix1             = _mm256_add_ps(fix1,tx);
 456             fiy1             = _mm256_add_ps(fiy1,ty);
 457             fiz1             = _mm256_add_ps(fiz1,tz);
 458
 459             fjx0             = _mm256_add_ps(fjx0,tx);
 460             fjy0             = _mm256_add_ps(fjy0,ty);
 461             fjz0             = _mm256_add_ps(fjz0,tz);
 462
 463             }
 464
 465             /**************************
 466              * CALCULATE INTERACTIONS *
 467              **************************/
 468
 469             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 470             {
 471
 472             /* REACTION-FIELD ELECTROSTATICS */
 473             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 474             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 475
 476             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 477
 478             /* Update potential sum for this i atom from the interaction with this j atom. */
 479             velec            = _mm256_and_ps(velec,cutoff_mask);
 480             velecsum         = _mm256_add_ps(velecsum,velec);
 481
 482             fscal            = felec;
 483
 484             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 485
 486             /* Calculate temporary vectorial force */
 487             tx               = _mm256_mul_ps(fscal,dx11);
 488             ty               = _mm256_mul_ps(fscal,dy11);
 489             tz               = _mm256_mul_ps(fscal,dz11);
 490
 491             /* Update vectorial force */
 492             fix1             = _mm256_add_ps(fix1,tx);
 493             fiy1             = _mm256_add_ps(fiy1,ty);
 494             fiz1             = _mm256_add_ps(fiz1,tz);
 495
 496             fjx1             = _mm256_add_ps(fjx1,tx);
 497             fjy1             = _mm256_add_ps(fjy1,ty);
 498             fjz1             = _mm256_add_ps(fjz1,tz);
 499
 500             }
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 507             {
 508
 509             /* REACTION-FIELD ELECTROSTATICS */
 510             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 511             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 512
 513             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 514
 515             /* Update potential sum for this i atom from the interaction with this j atom. */
 516             velec            = _mm256_and_ps(velec,cutoff_mask);
 517             velecsum         = _mm256_add_ps(velecsum,velec);
 518
 519             fscal            = felec;
 520
 521             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 522
 523             /* Calculate temporary vectorial force */
 524             tx               = _mm256_mul_ps(fscal,dx12);
 525             ty               = _mm256_mul_ps(fscal,dy12);
 526             tz               = _mm256_mul_ps(fscal,dz12);
 527
 528             /* Update vectorial force */
 529             fix1             = _mm256_add_ps(fix1,tx);
 530             fiy1             = _mm256_add_ps(fiy1,ty);
 531             fiz1             = _mm256_add_ps(fiz1,tz);
 532
 533             fjx2             = _mm256_add_ps(fjx2,tx);
 534             fjy2             = _mm256_add_ps(fjy2,ty);
 535             fjz2             = _mm256_add_ps(fjz2,tz);
 536
 537             }
 538
 539             /**************************
 540              * CALCULATE INTERACTIONS *
 541              **************************/
 542
 543             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 544             {
 545
 546             /* REACTION-FIELD ELECTROSTATICS */
 547             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 548             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 549
 550             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 551
 552             /* Update potential sum for this i atom from the interaction with this j atom. */
 553             velec            = _mm256_and_ps(velec,cutoff_mask);
 554             velecsum         = _mm256_add_ps(velecsum,velec);
 555
 556             fscal            = felec;
 557
 558             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 559
 560             /* Calculate temporary vectorial force */
 561             tx               = _mm256_mul_ps(fscal,dx20);
 562             ty               = _mm256_mul_ps(fscal,dy20);
 563             tz               = _mm256_mul_ps(fscal,dz20);
 564
 565             /* Update vectorial force */
 566             fix2             = _mm256_add_ps(fix2,tx);
 567             fiy2             = _mm256_add_ps(fiy2,ty);
 568             fiz2             = _mm256_add_ps(fiz2,tz);
 569
 570             fjx0             = _mm256_add_ps(fjx0,tx);
 571             fjy0             = _mm256_add_ps(fjy0,ty);
 572             fjz0             = _mm256_add_ps(fjz0,tz);
 573
 574             }
 575
 576             /**************************
 577              * CALCULATE INTERACTIONS *
 578              **************************/
 579
 580             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 581             {
 582
 583             /* REACTION-FIELD ELECTROSTATICS */
 584             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
 585             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
 586
 587             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
 588
 589             /* Update potential sum for this i atom from the interaction with this j atom. */
 590             velec            = _mm256_and_ps(velec,cutoff_mask);
 591             velecsum         = _mm256_add_ps(velecsum,velec);
 592
 593             fscal            = felec;
 594
 595             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 596
 597             /* Calculate temporary vectorial force */
 598             tx               = _mm256_mul_ps(fscal,dx21);
 599             ty               = _mm256_mul_ps(fscal,dy21);
 600             tz               = _mm256_mul_ps(fscal,dz21);
 601
 602             /* Update vectorial force */
 603             fix2             = _mm256_add_ps(fix2,tx);
 604             fiy2             = _mm256_add_ps(fiy2,ty);
 605             fiz2             = _mm256_add_ps(fiz2,tz);
 606
 607             fjx1             = _mm256_add_ps(fjx1,tx);
 608             fjy1             = _mm256_add_ps(fjy1,ty);
 609             fjz1             = _mm256_add_ps(fjz1,tz);
 610
 611             }
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 618             {
 619
 620             /* REACTION-FIELD ELECTROSTATICS */
 621             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
 622             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
 623
 624             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             velec            = _mm256_and_ps(velec,cutoff_mask);
 628             velecsum         = _mm256_add_ps(velecsum,velec);
 629
 630             fscal            = felec;
 631
 632             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 633
 634             /* Calculate temporary vectorial force */
 635             tx               = _mm256_mul_ps(fscal,dx22);
 636             ty               = _mm256_mul_ps(fscal,dy22);
 637             tz               = _mm256_mul_ps(fscal,dz22);
 638
 639             /* Update vectorial force */
 640             fix2             = _mm256_add_ps(fix2,tx);
 641             fiy2             = _mm256_add_ps(fiy2,ty);
 642             fiz2             = _mm256_add_ps(fiz2,tz);
 643
 644             fjx2             = _mm256_add_ps(fjx2,tx);
 645             fjy2             = _mm256_add_ps(fjy2,ty);
 646             fjz2             = _mm256_add_ps(fjz2,tz);
 647
 648             }
 649
 650             fjptrA             = f+j_coord_offsetA;
 651             fjptrB             = f+j_coord_offsetB;
 652             fjptrC             = f+j_coord_offsetC;
 653             fjptrD             = f+j_coord_offsetD;
 654             fjptrE             = f+j_coord_offsetE;
 655             fjptrF             = f+j_coord_offsetF;
 656             fjptrG             = f+j_coord_offsetG;
 657             fjptrH             = f+j_coord_offsetH;
 658
 659             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 660                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 661
 662             /* Inner loop uses 342 flops */
 663         }
 664
 665         if(jidx<j_index_end)
 666         {
 667
 668             /* Get j neighbor index, and coordinate index */
 669             jnrlistA         = jjnr[jidx];
 670             jnrlistB         = jjnr[jidx+1];
 671             jnrlistC         = jjnr[jidx+2];
 672             jnrlistD         = jjnr[jidx+3];
 673             jnrlistE         = jjnr[jidx+4];
 674             jnrlistF         = jjnr[jidx+5];
 675             jnrlistG         = jjnr[jidx+6];
 676             jnrlistH         = jjnr[jidx+7];
 677             /* Sign of each element will be negative for non-real atoms.
 678              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 679              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 680              */
 681             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 682                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 683
 684             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 685             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 686             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 687             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 688             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 689             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 690             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 691             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 692             j_coord_offsetA  = DIM*jnrA;
 693             j_coord_offsetB  = DIM*jnrB;
 694             j_coord_offsetC  = DIM*jnrC;
 695             j_coord_offsetD  = DIM*jnrD;
 696             j_coord_offsetE  = DIM*jnrE;
 697             j_coord_offsetF  = DIM*jnrF;
 698             j_coord_offsetG  = DIM*jnrG;
 699             j_coord_offsetH  = DIM*jnrH;
 700
 701             /* load j atom coordinates */
 702             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 703                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 704                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 705                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 706                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 707
 708             /* Calculate displacement vector */
 709             dx00             = _mm256_sub_ps(ix0,jx0);
 710             dy00             = _mm256_sub_ps(iy0,jy0);
 711             dz00             = _mm256_sub_ps(iz0,jz0);
 712             dx01             = _mm256_sub_ps(ix0,jx1);
 713             dy01             = _mm256_sub_ps(iy0,jy1);
 714             dz01             = _mm256_sub_ps(iz0,jz1);
 715             dx02             = _mm256_sub_ps(ix0,jx2);
 716             dy02             = _mm256_sub_ps(iy0,jy2);
 717             dz02             = _mm256_sub_ps(iz0,jz2);
 718             dx10             = _mm256_sub_ps(ix1,jx0);
 719             dy10             = _mm256_sub_ps(iy1,jy0);
 720             dz10             = _mm256_sub_ps(iz1,jz0);
 721             dx11             = _mm256_sub_ps(ix1,jx1);
 722             dy11             = _mm256_sub_ps(iy1,jy1);
 723             dz11             = _mm256_sub_ps(iz1,jz1);
 724             dx12             = _mm256_sub_ps(ix1,jx2);
 725             dy12             = _mm256_sub_ps(iy1,jy2);
 726             dz12             = _mm256_sub_ps(iz1,jz2);
 727             dx20             = _mm256_sub_ps(ix2,jx0);
 728             dy20             = _mm256_sub_ps(iy2,jy0);
 729             dz20             = _mm256_sub_ps(iz2,jz0);
 730             dx21             = _mm256_sub_ps(ix2,jx1);
 731             dy21             = _mm256_sub_ps(iy2,jy1);
 732             dz21             = _mm256_sub_ps(iz2,jz1);
 733             dx22             = _mm256_sub_ps(ix2,jx2);
 734             dy22             = _mm256_sub_ps(iy2,jy2);
 735             dz22             = _mm256_sub_ps(iz2,jz2);
 736
 737             /* Calculate squared distance and things based on it */
 738             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 739             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 740             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 741             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 742             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 743             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 744             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 745             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 746             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 747
 748             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 749             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 750             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 751             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 752             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 753             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 754             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 755             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 756             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 757
 758             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 759             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 760             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 761             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 762             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 763             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 764             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 765             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 766             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 767
 768             fjx0             = _mm256_setzero_ps();
 769             fjy0             = _mm256_setzero_ps();
 770             fjz0             = _mm256_setzero_ps();
 771             fjx1             = _mm256_setzero_ps();
 772             fjy1             = _mm256_setzero_ps();
 773             fjz1             = _mm256_setzero_ps();
 774             fjx2             = _mm256_setzero_ps();
 775             fjy2             = _mm256_setzero_ps();
 776             fjz2             = _mm256_setzero_ps();
 777
 778             /**************************
 779              * CALCULATE INTERACTIONS *
 780              **************************/
 781
 782             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 783             {
 784
 785             /* REACTION-FIELD ELECTROSTATICS */
 786             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 787             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 788
 789             /* LENNARD-JONES DISPERSION/REPULSION */
 790
 791             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 792             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 793             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 794             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 795                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 796             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 797
 798             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 799
 800             /* Update potential sum for this i atom from the interaction with this j atom. */
 801             velec            = _mm256_and_ps(velec,cutoff_mask);
 802             velec            = _mm256_andnot_ps(dummy_mask,velec);
 803             velecsum         = _mm256_add_ps(velecsum,velec);
 804             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 805             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 806             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 807
 808             fscal            = _mm256_add_ps(felec,fvdw);
 809
 810             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 811
 812             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 813
 814             /* Calculate temporary vectorial force */
 815             tx               = _mm256_mul_ps(fscal,dx00);
 816             ty               = _mm256_mul_ps(fscal,dy00);
 817             tz               = _mm256_mul_ps(fscal,dz00);
 818
 819             /* Update vectorial force */
 820             fix0             = _mm256_add_ps(fix0,tx);
 821             fiy0             = _mm256_add_ps(fiy0,ty);
 822             fiz0             = _mm256_add_ps(fiz0,tz);
 823
 824             fjx0             = _mm256_add_ps(fjx0,tx);
 825             fjy0             = _mm256_add_ps(fjy0,ty);
 826             fjz0             = _mm256_add_ps(fjz0,tz);
 827
 828             }
 829
 830             /**************************
 831              * CALCULATE INTERACTIONS *
 832              **************************/
 833
 834             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 835             {
 836
 837             /* REACTION-FIELD ELECTROSTATICS */
 838             velec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
 839             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
 840
 841             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 842
 843             /* Update potential sum for this i atom from the interaction with this j atom. */
 844             velec            = _mm256_and_ps(velec,cutoff_mask);
 845             velec            = _mm256_andnot_ps(dummy_mask,velec);
 846             velecsum         = _mm256_add_ps(velecsum,velec);
 847
 848             fscal            = felec;
 849
 850             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 851
 852             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 853
 854             /* Calculate temporary vectorial force */
 855             tx               = _mm256_mul_ps(fscal,dx01);
 856             ty               = _mm256_mul_ps(fscal,dy01);
 857             tz               = _mm256_mul_ps(fscal,dz01);
 858
 859             /* Update vectorial force */
 860             fix0             = _mm256_add_ps(fix0,tx);
 861             fiy0             = _mm256_add_ps(fiy0,ty);
 862             fiz0             = _mm256_add_ps(fiz0,tz);
 863
 864             fjx1             = _mm256_add_ps(fjx1,tx);
 865             fjy1             = _mm256_add_ps(fjy1,ty);
 866             fjz1             = _mm256_add_ps(fjz1,tz);
 867
 868             }
 869
 870             /**************************
 871              * CALCULATE INTERACTIONS *
 872              **************************/
 873
 874             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 875             {
 876
 877             /* REACTION-FIELD ELECTROSTATICS */
 878             velec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
 879             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
 880
 881             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
 882
 883             /* Update potential sum for this i atom from the interaction with this j atom. */
 884             velec            = _mm256_and_ps(velec,cutoff_mask);
 885             velec            = _mm256_andnot_ps(dummy_mask,velec);
 886             velecsum         = _mm256_add_ps(velecsum,velec);
 887
 888             fscal            = felec;
 889
 890             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 891
 892             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 893
 894             /* Calculate temporary vectorial force */
 895             tx               = _mm256_mul_ps(fscal,dx02);
 896             ty               = _mm256_mul_ps(fscal,dy02);
 897             tz               = _mm256_mul_ps(fscal,dz02);
 898
 899             /* Update vectorial force */
 900             fix0             = _mm256_add_ps(fix0,tx);
 901             fiy0             = _mm256_add_ps(fiy0,ty);
 902             fiz0             = _mm256_add_ps(fiz0,tz);
 903
 904             fjx2             = _mm256_add_ps(fjx2,tx);
 905             fjy2             = _mm256_add_ps(fjy2,ty);
 906             fjz2             = _mm256_add_ps(fjz2,tz);
 907
 908             }
 909
 910             /**************************
 911              * CALCULATE INTERACTIONS *
 912              **************************/
 913
 914             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 915             {
 916
 917             /* REACTION-FIELD ELECTROSTATICS */
 918             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 919             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 920
 921             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 922
 923             /* Update potential sum for this i atom from the interaction with this j atom. */
 924             velec            = _mm256_and_ps(velec,cutoff_mask);
 925             velec            = _mm256_andnot_ps(dummy_mask,velec);
 926             velecsum         = _mm256_add_ps(velecsum,velec);
 927
 928             fscal            = felec;
 929
 930             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 931
 932             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 933
 934             /* Calculate temporary vectorial force */
 935             tx               = _mm256_mul_ps(fscal,dx10);
 936             ty               = _mm256_mul_ps(fscal,dy10);
 937             tz               = _mm256_mul_ps(fscal,dz10);
 938
 939             /* Update vectorial force */
 940             fix1             = _mm256_add_ps(fix1,tx);
 941             fiy1             = _mm256_add_ps(fiy1,ty);
 942             fiz1             = _mm256_add_ps(fiz1,tz);
 943
 944             fjx0             = _mm256_add_ps(fjx0,tx);
 945             fjy0             = _mm256_add_ps(fjy0,ty);
 946             fjz0             = _mm256_add_ps(fjz0,tz);
 947
 948             }
 949
 950             /**************************
 951              * CALCULATE INTERACTIONS *
 952              **************************/
 953
 954             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 955             {
 956
 957             /* REACTION-FIELD ELECTROSTATICS */
 958             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 959             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 960
 961             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 962
 963             /* Update potential sum for this i atom from the interaction with this j atom. */
 964             velec            = _mm256_and_ps(velec,cutoff_mask);
 965             velec            = _mm256_andnot_ps(dummy_mask,velec);
 966             velecsum         = _mm256_add_ps(velecsum,velec);
 967
 968             fscal            = felec;
 969
 970             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 971
 972             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 973
 974             /* Calculate temporary vectorial force */
 975             tx               = _mm256_mul_ps(fscal,dx11);
 976             ty               = _mm256_mul_ps(fscal,dy11);
 977             tz               = _mm256_mul_ps(fscal,dz11);
 978
 979             /* Update vectorial force */
 980             fix1             = _mm256_add_ps(fix1,tx);
 981             fiy1             = _mm256_add_ps(fiy1,ty);
 982             fiz1             = _mm256_add_ps(fiz1,tz);
 983
 984             fjx1             = _mm256_add_ps(fjx1,tx);
 985             fjy1             = _mm256_add_ps(fjy1,ty);
 986             fjz1             = _mm256_add_ps(fjz1,tz);
 987
 988             }
 989
 990             /**************************
 991              * CALCULATE INTERACTIONS *
 992              **************************/
 993
 994             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 995             {
 996
 997             /* REACTION-FIELD ELECTROSTATICS */
 998             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 999             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1000
1001             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1002
1003             /* Update potential sum for this i atom from the interaction with this j atom. */
1004             velec            = _mm256_and_ps(velec,cutoff_mask);
1005             velec            = _mm256_andnot_ps(dummy_mask,velec);
1006             velecsum         = _mm256_add_ps(velecsum,velec);
1007
1008             fscal            = felec;
1009
1010             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1011
1012             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1013
1014             /* Calculate temporary vectorial force */
1015             tx               = _mm256_mul_ps(fscal,dx12);
1016             ty               = _mm256_mul_ps(fscal,dy12);
1017             tz               = _mm256_mul_ps(fscal,dz12);
1018
1019             /* Update vectorial force */
1020             fix1             = _mm256_add_ps(fix1,tx);
1021             fiy1             = _mm256_add_ps(fiy1,ty);
1022             fiz1             = _mm256_add_ps(fiz1,tz);
1023
1024             fjx2             = _mm256_add_ps(fjx2,tx);
1025             fjy2             = _mm256_add_ps(fjy2,ty);
1026             fjz2             = _mm256_add_ps(fjz2,tz);
1027
1028             }
1029
1030             /**************************
1031              * CALCULATE INTERACTIONS *
1032              **************************/
1033
1034             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1035             {
1036
1037             /* REACTION-FIELD ELECTROSTATICS */
1038             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
1039             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1040
1041             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1042
1043             /* Update potential sum for this i atom from the interaction with this j atom. */
1044             velec            = _mm256_and_ps(velec,cutoff_mask);
1045             velec            = _mm256_andnot_ps(dummy_mask,velec);
1046             velecsum         = _mm256_add_ps(velecsum,velec);
1047
1048             fscal            = felec;
1049
1050             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1051
1052             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1053
1054             /* Calculate temporary vectorial force */
1055             tx               = _mm256_mul_ps(fscal,dx20);
1056             ty               = _mm256_mul_ps(fscal,dy20);
1057             tz               = _mm256_mul_ps(fscal,dz20);
1058
1059             /* Update vectorial force */
1060             fix2             = _mm256_add_ps(fix2,tx);
1061             fiy2             = _mm256_add_ps(fiy2,ty);
1062             fiz2             = _mm256_add_ps(fiz2,tz);
1063
1064             fjx0             = _mm256_add_ps(fjx0,tx);
1065             fjy0             = _mm256_add_ps(fjy0,ty);
1066             fjz0             = _mm256_add_ps(fjz0,tz);
1067
1068             }
1069
1070             /**************************
1071              * CALCULATE INTERACTIONS *
1072              **************************/
1073
1074             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1075             {
1076
1077             /* REACTION-FIELD ELECTROSTATICS */
1078             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
1079             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1080
1081             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1082
1083             /* Update potential sum for this i atom from the interaction with this j atom. */
1084             velec            = _mm256_and_ps(velec,cutoff_mask);
1085             velec            = _mm256_andnot_ps(dummy_mask,velec);
1086             velecsum         = _mm256_add_ps(velecsum,velec);
1087
1088             fscal            = felec;
1089
1090             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1091
1092             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1093
1094             /* Calculate temporary vectorial force */
1095             tx               = _mm256_mul_ps(fscal,dx21);
1096             ty               = _mm256_mul_ps(fscal,dy21);
1097             tz               = _mm256_mul_ps(fscal,dz21);
1098
1099             /* Update vectorial force */
1100             fix2             = _mm256_add_ps(fix2,tx);
1101             fiy2             = _mm256_add_ps(fiy2,ty);
1102             fiz2             = _mm256_add_ps(fiz2,tz);
1103
1104             fjx1             = _mm256_add_ps(fjx1,tx);
1105             fjy1             = _mm256_add_ps(fjy1,ty);
1106             fjz1             = _mm256_add_ps(fjz1,tz);
1107
1108             }
1109
1110             /**************************
1111              * CALCULATE INTERACTIONS *
1112              **************************/
1113
1114             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1115             {
1116
1117             /* REACTION-FIELD ELECTROSTATICS */
1118             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
1119             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1120
1121             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1122
1123             /* Update potential sum for this i atom from the interaction with this j atom. */
1124             velec            = _mm256_and_ps(velec,cutoff_mask);
1125             velec            = _mm256_andnot_ps(dummy_mask,velec);
1126             velecsum         = _mm256_add_ps(velecsum,velec);
1127
1128             fscal            = felec;
1129
1130             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1131
1132             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1133
1134             /* Calculate temporary vectorial force */
1135             tx               = _mm256_mul_ps(fscal,dx22);
1136             ty               = _mm256_mul_ps(fscal,dy22);
1137             tz               = _mm256_mul_ps(fscal,dz22);
1138
1139             /* Update vectorial force */
1140             fix2             = _mm256_add_ps(fix2,tx);
1141             fiy2             = _mm256_add_ps(fiy2,ty);
1142             fiz2             = _mm256_add_ps(fiz2,tz);
1143
1144             fjx2             = _mm256_add_ps(fjx2,tx);
1145             fjy2             = _mm256_add_ps(fjy2,ty);
1146             fjz2             = _mm256_add_ps(fjz2,tz);
1147
1148             }
1149
1150             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1151             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1152             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1153             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1154             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1155             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1156             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1157             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1158
1159             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1160                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1161
1162             /* Inner loop uses 342 flops */
1163         }
1164
1165         /* End of innermost loop */
1166
1167         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1168                                                  f+i_coord_offset,fshift+i_shift_offset);
1169
1170         ggid                        = gid[iidx];
1171         /* Update potential energies */
1172         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1173         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1174
1175         /* Increment number of inner iterations */
1176         inneriter                  += j_index_end - j_index_start;
1177
1178         /* Outer loop uses 20 flops */
1179     }
1180
1181     /* Increment number of outer iterations */
1182     outeriter        += nri;
1183
1184     /* Update outer/inner flops */
1185
1186     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*342);
1187 }
1188 /*
1189  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_F_avx_256_single
1190  * Electrostatics interaction: ReactionField
1191  * VdW interaction:            LennardJones
1192  * Geometry:                   Water3-Water3
1193  * Calculate force/pot:        Force
1194  */
1195 void
1196 nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_F_avx_256_single
1197                     (t_nblist * gmx_restrict                nlist,
1198                      rvec * gmx_restrict                    xx,
1199                      rvec * gmx_restrict                    ff,
1200                      t_forcerec * gmx_restrict              fr,
1201                      t_mdatoms * gmx_restrict               mdatoms,
1202                      nb_kernel_data_t * gmx_restrict        kernel_data,
1203                      t_nrnb * gmx_restrict                  nrnb)
1204 {
1205     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1206      * just 0 for non-waters.
1207      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1208      * jnr indices corresponding to data put in the four positions in the SIMD register.
1209      */
1210     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1211     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1212     int              jnrA,jnrB,jnrC,jnrD;
1213     int              jnrE,jnrF,jnrG,jnrH;
1214     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1215     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1216     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1217     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1218     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1219     real             rcutoff_scalar;
1220     real             *shiftvec,*fshift,*x,*f;
1221     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1222     real             scratch[4*DIM];
1223     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1224     real *           vdwioffsetptr0;
1225     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1226     real *           vdwioffsetptr1;
1227     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1228     real *           vdwioffsetptr2;
1229     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1230     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1231     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1232     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1233     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1234     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1235     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1236     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1237     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1238     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1239     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1240     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1241     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1242     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1243     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1244     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1245     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1246     real             *charge;
1247     int              nvdwtype;
1248     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1249     int              *vdwtype;
1250     real             *vdwparam;
1251     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1252     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1253     __m256           dummy_mask,cutoff_mask;
1254     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1255     __m256           one     = _mm256_set1_ps(1.0);
1256     __m256           two     = _mm256_set1_ps(2.0);
1257     x                = xx[0];
1258     f                = ff[0];
1259
1260     nri              = nlist->nri;
1261     iinr             = nlist->iinr;
1262     jindex           = nlist->jindex;
1263     jjnr             = nlist->jjnr;
1264     shiftidx         = nlist->shift;
1265     gid              = nlist->gid;
1266     shiftvec         = fr->shift_vec[0];
1267     fshift           = fr->fshift[0];
1268     facel            = _mm256_set1_ps(fr->epsfac);
1269     charge           = mdatoms->chargeA;
1270     krf              = _mm256_set1_ps(fr->ic->k_rf);
1271     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
1272     crf              = _mm256_set1_ps(fr->ic->c_rf);
1273     nvdwtype         = fr->ntype;
1274     vdwparam         = fr->nbfp;
1275     vdwtype          = mdatoms->typeA;
1276
1277     /* Setup water-specific parameters */
1278     inr              = nlist->iinr[0];
1279     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1280     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1281     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1282     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1283
1284     jq0              = _mm256_set1_ps(charge[inr+0]);
1285     jq1              = _mm256_set1_ps(charge[inr+1]);
1286     jq2              = _mm256_set1_ps(charge[inr+2]);
1287     vdwjidx0A        = 2*vdwtype[inr+0];
1288     qq00             = _mm256_mul_ps(iq0,jq0);
1289     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1290     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1291     qq01             = _mm256_mul_ps(iq0,jq1);
1292     qq02             = _mm256_mul_ps(iq0,jq2);
1293     qq10             = _mm256_mul_ps(iq1,jq0);
1294     qq11             = _mm256_mul_ps(iq1,jq1);
1295     qq12             = _mm256_mul_ps(iq1,jq2);
1296     qq20             = _mm256_mul_ps(iq2,jq0);
1297     qq21             = _mm256_mul_ps(iq2,jq1);
1298     qq22             = _mm256_mul_ps(iq2,jq2);
1299
1300     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1301     rcutoff_scalar   = fr->rcoulomb;
1302     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1303     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1304
1305     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
1306     rvdw             = _mm256_set1_ps(fr->rvdw);
1307
1308     /* Avoid stupid compiler warnings */
1309     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1310     j_coord_offsetA = 0;
1311     j_coord_offsetB = 0;
1312     j_coord_offsetC = 0;
1313     j_coord_offsetD = 0;
1314     j_coord_offsetE = 0;
1315     j_coord_offsetF = 0;
1316     j_coord_offsetG = 0;
1317     j_coord_offsetH = 0;
1318
1319     outeriter        = 0;
1320     inneriter        = 0;
1321
1322     for(iidx=0;iidx<4*DIM;iidx++)
1323     {
1324         scratch[iidx] = 0.0;
1325     }
1326
1327     /* Start outer loop over neighborlists */
1328     for(iidx=0; iidx<nri; iidx++)
1329     {
1330         /* Load shift vector for this list */
1331         i_shift_offset   = DIM*shiftidx[iidx];
1332
1333         /* Load limits for loop over neighbors */
1334         j_index_start    = jindex[iidx];
1335         j_index_end      = jindex[iidx+1];
1336
1337         /* Get outer coordinate index */
1338         inr              = iinr[iidx];
1339         i_coord_offset   = DIM*inr;
1340
1341         /* Load i particle coords and add shift vector */
1342         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1343                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1344
1345         fix0             = _mm256_setzero_ps();
1346         fiy0             = _mm256_setzero_ps();
1347         fiz0             = _mm256_setzero_ps();
1348         fix1             = _mm256_setzero_ps();
1349         fiy1             = _mm256_setzero_ps();
1350         fiz1             = _mm256_setzero_ps();
1351         fix2             = _mm256_setzero_ps();
1352         fiy2             = _mm256_setzero_ps();
1353         fiz2             = _mm256_setzero_ps();
1354
1355         /* Start inner kernel loop */
1356         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1357         {
1358
1359             /* Get j neighbor index, and coordinate index */
1360             jnrA             = jjnr[jidx];
1361             jnrB             = jjnr[jidx+1];
1362             jnrC             = jjnr[jidx+2];
1363             jnrD             = jjnr[jidx+3];
1364             jnrE             = jjnr[jidx+4];
1365             jnrF             = jjnr[jidx+5];
1366             jnrG             = jjnr[jidx+6];
1367             jnrH             = jjnr[jidx+7];
1368             j_coord_offsetA  = DIM*jnrA;
1369             j_coord_offsetB  = DIM*jnrB;
1370             j_coord_offsetC  = DIM*jnrC;
1371             j_coord_offsetD  = DIM*jnrD;
1372             j_coord_offsetE  = DIM*jnrE;
1373             j_coord_offsetF  = DIM*jnrF;
1374             j_coord_offsetG  = DIM*jnrG;
1375             j_coord_offsetH  = DIM*jnrH;
1376
1377             /* load j atom coordinates */
1378             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1379                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1380                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1381                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1382                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1383
1384             /* Calculate displacement vector */
1385             dx00             = _mm256_sub_ps(ix0,jx0);
1386             dy00             = _mm256_sub_ps(iy0,jy0);
1387             dz00             = _mm256_sub_ps(iz0,jz0);
1388             dx01             = _mm256_sub_ps(ix0,jx1);
1389             dy01             = _mm256_sub_ps(iy0,jy1);
1390             dz01             = _mm256_sub_ps(iz0,jz1);
1391             dx02             = _mm256_sub_ps(ix0,jx2);
1392             dy02             = _mm256_sub_ps(iy0,jy2);
1393             dz02             = _mm256_sub_ps(iz0,jz2);
1394             dx10             = _mm256_sub_ps(ix1,jx0);
1395             dy10             = _mm256_sub_ps(iy1,jy0);
1396             dz10             = _mm256_sub_ps(iz1,jz0);
1397             dx11             = _mm256_sub_ps(ix1,jx1);
1398             dy11             = _mm256_sub_ps(iy1,jy1);
1399             dz11             = _mm256_sub_ps(iz1,jz1);
1400             dx12             = _mm256_sub_ps(ix1,jx2);
1401             dy12             = _mm256_sub_ps(iy1,jy2);
1402             dz12             = _mm256_sub_ps(iz1,jz2);
1403             dx20             = _mm256_sub_ps(ix2,jx0);
1404             dy20             = _mm256_sub_ps(iy2,jy0);
1405             dz20             = _mm256_sub_ps(iz2,jz0);
1406             dx21             = _mm256_sub_ps(ix2,jx1);
1407             dy21             = _mm256_sub_ps(iy2,jy1);
1408             dz21             = _mm256_sub_ps(iz2,jz1);
1409             dx22             = _mm256_sub_ps(ix2,jx2);
1410             dy22             = _mm256_sub_ps(iy2,jy2);
1411             dz22             = _mm256_sub_ps(iz2,jz2);
1412
1413             /* Calculate squared distance and things based on it */
1414             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1415             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1416             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1417             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1418             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1419             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1420             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1421             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1422             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1423
1424             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1425             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1426             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1427             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1428             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1429             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1430             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1431             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1432             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1433
1434             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1435             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1436             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1437             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1438             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1439             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1440             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1441             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1442             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1443
1444             fjx0             = _mm256_setzero_ps();
1445             fjy0             = _mm256_setzero_ps();
1446             fjz0             = _mm256_setzero_ps();
1447             fjx1             = _mm256_setzero_ps();
1448             fjy1             = _mm256_setzero_ps();
1449             fjz1             = _mm256_setzero_ps();
1450             fjx2             = _mm256_setzero_ps();
1451             fjy2             = _mm256_setzero_ps();
1452             fjz2             = _mm256_setzero_ps();
1453
1454             /**************************
1455              * CALCULATE INTERACTIONS *
1456              **************************/
1457
1458             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1459             {
1460
1461             /* REACTION-FIELD ELECTROSTATICS */
1462             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1463
1464             /* LENNARD-JONES DISPERSION/REPULSION */
1465
1466             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1467             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1468
1469             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1470
1471             fscal            = _mm256_add_ps(felec,fvdw);
1472
1473             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1474
1475             /* Calculate temporary vectorial force */
1476             tx               = _mm256_mul_ps(fscal,dx00);
1477             ty               = _mm256_mul_ps(fscal,dy00);
1478             tz               = _mm256_mul_ps(fscal,dz00);
1479
1480             /* Update vectorial force */
1481             fix0             = _mm256_add_ps(fix0,tx);
1482             fiy0             = _mm256_add_ps(fiy0,ty);
1483             fiz0             = _mm256_add_ps(fiz0,tz);
1484
1485             fjx0             = _mm256_add_ps(fjx0,tx);
1486             fjy0             = _mm256_add_ps(fjy0,ty);
1487             fjz0             = _mm256_add_ps(fjz0,tz);
1488
1489             }
1490
1491             /**************************
1492              * CALCULATE INTERACTIONS *
1493              **************************/
1494
1495             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1496             {
1497
1498             /* REACTION-FIELD ELECTROSTATICS */
1499             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1500
1501             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
1502
1503             fscal            = felec;
1504
1505             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1506
1507             /* Calculate temporary vectorial force */
1508             tx               = _mm256_mul_ps(fscal,dx01);
1509             ty               = _mm256_mul_ps(fscal,dy01);
1510             tz               = _mm256_mul_ps(fscal,dz01);
1511
1512             /* Update vectorial force */
1513             fix0             = _mm256_add_ps(fix0,tx);
1514             fiy0             = _mm256_add_ps(fiy0,ty);
1515             fiz0             = _mm256_add_ps(fiz0,tz);
1516
1517             fjx1             = _mm256_add_ps(fjx1,tx);
1518             fjy1             = _mm256_add_ps(fjy1,ty);
1519             fjz1             = _mm256_add_ps(fjz1,tz);
1520
1521             }
1522
1523             /**************************
1524              * CALCULATE INTERACTIONS *
1525              **************************/
1526
1527             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1528             {
1529
1530             /* REACTION-FIELD ELECTROSTATICS */
1531             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1532
1533             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1534
1535             fscal            = felec;
1536
1537             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1538
1539             /* Calculate temporary vectorial force */
1540             tx               = _mm256_mul_ps(fscal,dx02);
1541             ty               = _mm256_mul_ps(fscal,dy02);
1542             tz               = _mm256_mul_ps(fscal,dz02);
1543
1544             /* Update vectorial force */
1545             fix0             = _mm256_add_ps(fix0,tx);
1546             fiy0             = _mm256_add_ps(fiy0,ty);
1547             fiz0             = _mm256_add_ps(fiz0,tz);
1548
1549             fjx2             = _mm256_add_ps(fjx2,tx);
1550             fjy2             = _mm256_add_ps(fjy2,ty);
1551             fjz2             = _mm256_add_ps(fjz2,tz);
1552
1553             }
1554
1555             /**************************
1556              * CALCULATE INTERACTIONS *
1557              **************************/
1558
1559             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1560             {
1561
1562             /* REACTION-FIELD ELECTROSTATICS */
1563             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1564
1565             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1566
1567             fscal            = felec;
1568
1569             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1570
1571             /* Calculate temporary vectorial force */
1572             tx               = _mm256_mul_ps(fscal,dx10);
1573             ty               = _mm256_mul_ps(fscal,dy10);
1574             tz               = _mm256_mul_ps(fscal,dz10);
1575
1576             /* Update vectorial force */
1577             fix1             = _mm256_add_ps(fix1,tx);
1578             fiy1             = _mm256_add_ps(fiy1,ty);
1579             fiz1             = _mm256_add_ps(fiz1,tz);
1580
1581             fjx0             = _mm256_add_ps(fjx0,tx);
1582             fjy0             = _mm256_add_ps(fjy0,ty);
1583             fjz0             = _mm256_add_ps(fjz0,tz);
1584
1585             }
1586
1587             /**************************
1588              * CALCULATE INTERACTIONS *
1589              **************************/
1590
1591             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1592             {
1593
1594             /* REACTION-FIELD ELECTROSTATICS */
1595             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1596
1597             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1598
1599             fscal            = felec;
1600
1601             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1602
1603             /* Calculate temporary vectorial force */
1604             tx               = _mm256_mul_ps(fscal,dx11);
1605             ty               = _mm256_mul_ps(fscal,dy11);
1606             tz               = _mm256_mul_ps(fscal,dz11);
1607
1608             /* Update vectorial force */
1609             fix1             = _mm256_add_ps(fix1,tx);
1610             fiy1             = _mm256_add_ps(fiy1,ty);
1611             fiz1             = _mm256_add_ps(fiz1,tz);
1612
1613             fjx1             = _mm256_add_ps(fjx1,tx);
1614             fjy1             = _mm256_add_ps(fjy1,ty);
1615             fjz1             = _mm256_add_ps(fjz1,tz);
1616
1617             }
1618
1619             /**************************
1620              * CALCULATE INTERACTIONS *
1621              **************************/
1622
1623             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1624             {
1625
1626             /* REACTION-FIELD ELECTROSTATICS */
1627             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1628
1629             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1630
1631             fscal            = felec;
1632
1633             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1634
1635             /* Calculate temporary vectorial force */
1636             tx               = _mm256_mul_ps(fscal,dx12);
1637             ty               = _mm256_mul_ps(fscal,dy12);
1638             tz               = _mm256_mul_ps(fscal,dz12);
1639
1640             /* Update vectorial force */
1641             fix1             = _mm256_add_ps(fix1,tx);
1642             fiy1             = _mm256_add_ps(fiy1,ty);
1643             fiz1             = _mm256_add_ps(fiz1,tz);
1644
1645             fjx2             = _mm256_add_ps(fjx2,tx);
1646             fjy2             = _mm256_add_ps(fjy2,ty);
1647             fjz2             = _mm256_add_ps(fjz2,tz);
1648
1649             }
1650
1651             /**************************
1652              * CALCULATE INTERACTIONS *
1653              **************************/
1654
1655             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1656             {
1657
1658             /* REACTION-FIELD ELECTROSTATICS */
1659             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1660
1661             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1662
1663             fscal            = felec;
1664
1665             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1666
1667             /* Calculate temporary vectorial force */
1668             tx               = _mm256_mul_ps(fscal,dx20);
1669             ty               = _mm256_mul_ps(fscal,dy20);
1670             tz               = _mm256_mul_ps(fscal,dz20);
1671
1672             /* Update vectorial force */
1673             fix2             = _mm256_add_ps(fix2,tx);
1674             fiy2             = _mm256_add_ps(fiy2,ty);
1675             fiz2             = _mm256_add_ps(fiz2,tz);
1676
1677             fjx0             = _mm256_add_ps(fjx0,tx);
1678             fjy0             = _mm256_add_ps(fjy0,ty);
1679             fjz0             = _mm256_add_ps(fjz0,tz);
1680
1681             }
1682
1683             /**************************
1684              * CALCULATE INTERACTIONS *
1685              **************************/
1686
1687             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1688             {
1689
1690             /* REACTION-FIELD ELECTROSTATICS */
1691             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1692
1693             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1694
1695             fscal            = felec;
1696
1697             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1698
1699             /* Calculate temporary vectorial force */
1700             tx               = _mm256_mul_ps(fscal,dx21);
1701             ty               = _mm256_mul_ps(fscal,dy21);
1702             tz               = _mm256_mul_ps(fscal,dz21);
1703
1704             /* Update vectorial force */
1705             fix2             = _mm256_add_ps(fix2,tx);
1706             fiy2             = _mm256_add_ps(fiy2,ty);
1707             fiz2             = _mm256_add_ps(fiz2,tz);
1708
1709             fjx1             = _mm256_add_ps(fjx1,tx);
1710             fjy1             = _mm256_add_ps(fjy1,ty);
1711             fjz1             = _mm256_add_ps(fjz1,tz);
1712
1713             }
1714
1715             /**************************
1716              * CALCULATE INTERACTIONS *
1717              **************************/
1718
1719             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1720             {
1721
1722             /* REACTION-FIELD ELECTROSTATICS */
1723             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1724
1725             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1726
1727             fscal            = felec;
1728
1729             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1730
1731             /* Calculate temporary vectorial force */
1732             tx               = _mm256_mul_ps(fscal,dx22);
1733             ty               = _mm256_mul_ps(fscal,dy22);
1734             tz               = _mm256_mul_ps(fscal,dz22);
1735
1736             /* Update vectorial force */
1737             fix2             = _mm256_add_ps(fix2,tx);
1738             fiy2             = _mm256_add_ps(fiy2,ty);
1739             fiz2             = _mm256_add_ps(fiz2,tz);
1740
1741             fjx2             = _mm256_add_ps(fjx2,tx);
1742             fjy2             = _mm256_add_ps(fjy2,ty);
1743             fjz2             = _mm256_add_ps(fjz2,tz);
1744
1745             }
1746
1747             fjptrA             = f+j_coord_offsetA;
1748             fjptrB             = f+j_coord_offsetB;
1749             fjptrC             = f+j_coord_offsetC;
1750             fjptrD             = f+j_coord_offsetD;
1751             fjptrE             = f+j_coord_offsetE;
1752             fjptrF             = f+j_coord_offsetF;
1753             fjptrG             = f+j_coord_offsetG;
1754             fjptrH             = f+j_coord_offsetH;
1755
1756             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1757                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1758
1759             /* Inner loop uses 277 flops */
1760         }
1761
1762         if(jidx<j_index_end)
1763         {
1764
1765             /* Get j neighbor index, and coordinate index */
1766             jnrlistA         = jjnr[jidx];
1767             jnrlistB         = jjnr[jidx+1];
1768             jnrlistC         = jjnr[jidx+2];
1769             jnrlistD         = jjnr[jidx+3];
1770             jnrlistE         = jjnr[jidx+4];
1771             jnrlistF         = jjnr[jidx+5];
1772             jnrlistG         = jjnr[jidx+6];
1773             jnrlistH         = jjnr[jidx+7];
1774             /* Sign of each element will be negative for non-real atoms.
1775              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1776              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1777              */
1778             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1779                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1780
1781             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1782             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1783             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1784             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1785             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1786             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1787             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1788             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1789             j_coord_offsetA  = DIM*jnrA;
1790             j_coord_offsetB  = DIM*jnrB;
1791             j_coord_offsetC  = DIM*jnrC;
1792             j_coord_offsetD  = DIM*jnrD;
1793             j_coord_offsetE  = DIM*jnrE;
1794             j_coord_offsetF  = DIM*jnrF;
1795             j_coord_offsetG  = DIM*jnrG;
1796             j_coord_offsetH  = DIM*jnrH;
1797
1798             /* load j atom coordinates */
1799             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1800                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1801                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1802                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1803                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1804
1805             /* Calculate displacement vector */
1806             dx00             = _mm256_sub_ps(ix0,jx0);
1807             dy00             = _mm256_sub_ps(iy0,jy0);
1808             dz00             = _mm256_sub_ps(iz0,jz0);
1809             dx01             = _mm256_sub_ps(ix0,jx1);
1810             dy01             = _mm256_sub_ps(iy0,jy1);
1811             dz01             = _mm256_sub_ps(iz0,jz1);
1812             dx02             = _mm256_sub_ps(ix0,jx2);
1813             dy02             = _mm256_sub_ps(iy0,jy2);
1814             dz02             = _mm256_sub_ps(iz0,jz2);
1815             dx10             = _mm256_sub_ps(ix1,jx0);
1816             dy10             = _mm256_sub_ps(iy1,jy0);
1817             dz10             = _mm256_sub_ps(iz1,jz0);
1818             dx11             = _mm256_sub_ps(ix1,jx1);
1819             dy11             = _mm256_sub_ps(iy1,jy1);
1820             dz11             = _mm256_sub_ps(iz1,jz1);
1821             dx12             = _mm256_sub_ps(ix1,jx2);
1822             dy12             = _mm256_sub_ps(iy1,jy2);
1823             dz12             = _mm256_sub_ps(iz1,jz2);
1824             dx20             = _mm256_sub_ps(ix2,jx0);
1825             dy20             = _mm256_sub_ps(iy2,jy0);
1826             dz20             = _mm256_sub_ps(iz2,jz0);
1827             dx21             = _mm256_sub_ps(ix2,jx1);
1828             dy21             = _mm256_sub_ps(iy2,jy1);
1829             dz21             = _mm256_sub_ps(iz2,jz1);
1830             dx22             = _mm256_sub_ps(ix2,jx2);
1831             dy22             = _mm256_sub_ps(iy2,jy2);
1832             dz22             = _mm256_sub_ps(iz2,jz2);
1833
1834             /* Calculate squared distance and things based on it */
1835             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1836             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1837             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1838             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1839             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1840             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1841             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1842             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1843             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1844
1845             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1846             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1847             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1848             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1849             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1850             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1851             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1852             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1853             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1854
1855             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1856             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1857             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1858             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1859             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1860             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1861             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1862             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1863             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1864
1865             fjx0             = _mm256_setzero_ps();
1866             fjy0             = _mm256_setzero_ps();
1867             fjz0             = _mm256_setzero_ps();
1868             fjx1             = _mm256_setzero_ps();
1869             fjy1             = _mm256_setzero_ps();
1870             fjz1             = _mm256_setzero_ps();
1871             fjx2             = _mm256_setzero_ps();
1872             fjy2             = _mm256_setzero_ps();
1873             fjz2             = _mm256_setzero_ps();
1874
1875             /**************************
1876              * CALCULATE INTERACTIONS *
1877              **************************/
1878
1879             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1880             {
1881
1882             /* REACTION-FIELD ELECTROSTATICS */
1883             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1884
1885             /* LENNARD-JONES DISPERSION/REPULSION */
1886
1887             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1888             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1889
1890             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1891
1892             fscal            = _mm256_add_ps(felec,fvdw);
1893
1894             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1895
1896             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1897
1898             /* Calculate temporary vectorial force */
1899             tx               = _mm256_mul_ps(fscal,dx00);
1900             ty               = _mm256_mul_ps(fscal,dy00);
1901             tz               = _mm256_mul_ps(fscal,dz00);
1902
1903             /* Update vectorial force */
1904             fix0             = _mm256_add_ps(fix0,tx);
1905             fiy0             = _mm256_add_ps(fiy0,ty);
1906             fiz0             = _mm256_add_ps(fiz0,tz);
1907
1908             fjx0             = _mm256_add_ps(fjx0,tx);
1909             fjy0             = _mm256_add_ps(fjy0,ty);
1910             fjz0             = _mm256_add_ps(fjz0,tz);
1911
1912             }
1913
1914             /**************************
1915              * CALCULATE INTERACTIONS *
1916              **************************/
1917
1918             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1919             {
1920
1921             /* REACTION-FIELD ELECTROSTATICS */
1922             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1923
1924             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
1925
1926             fscal            = felec;
1927
1928             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1929
1930             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1931
1932             /* Calculate temporary vectorial force */
1933             tx               = _mm256_mul_ps(fscal,dx01);
1934             ty               = _mm256_mul_ps(fscal,dy01);
1935             tz               = _mm256_mul_ps(fscal,dz01);
1936
1937             /* Update vectorial force */
1938             fix0             = _mm256_add_ps(fix0,tx);
1939             fiy0             = _mm256_add_ps(fiy0,ty);
1940             fiz0             = _mm256_add_ps(fiz0,tz);
1941
1942             fjx1             = _mm256_add_ps(fjx1,tx);
1943             fjy1             = _mm256_add_ps(fjy1,ty);
1944             fjz1             = _mm256_add_ps(fjz1,tz);
1945
1946             }
1947
1948             /**************************
1949              * CALCULATE INTERACTIONS *
1950              **************************/
1951
1952             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1953             {
1954
1955             /* REACTION-FIELD ELECTROSTATICS */
1956             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1957
1958             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1959
1960             fscal            = felec;
1961
1962             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1963
1964             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1965
1966             /* Calculate temporary vectorial force */
1967             tx               = _mm256_mul_ps(fscal,dx02);
1968             ty               = _mm256_mul_ps(fscal,dy02);
1969             tz               = _mm256_mul_ps(fscal,dz02);
1970
1971             /* Update vectorial force */
1972             fix0             = _mm256_add_ps(fix0,tx);
1973             fiy0             = _mm256_add_ps(fiy0,ty);
1974             fiz0             = _mm256_add_ps(fiz0,tz);
1975
1976             fjx2             = _mm256_add_ps(fjx2,tx);
1977             fjy2             = _mm256_add_ps(fjy2,ty);
1978             fjz2             = _mm256_add_ps(fjz2,tz);
1979
1980             }
1981
1982             /**************************
1983              * CALCULATE INTERACTIONS *
1984              **************************/
1985
1986             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1987             {
1988
1989             /* REACTION-FIELD ELECTROSTATICS */
1990             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1991
1992             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1993
1994             fscal            = felec;
1995
1996             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1997
1998             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1999
2000             /* Calculate temporary vectorial force */
2001             tx               = _mm256_mul_ps(fscal,dx10);
2002             ty               = _mm256_mul_ps(fscal,dy10);
2003             tz               = _mm256_mul_ps(fscal,dz10);
2004
2005             /* Update vectorial force */
2006             fix1             = _mm256_add_ps(fix1,tx);
2007             fiy1             = _mm256_add_ps(fiy1,ty);
2008             fiz1             = _mm256_add_ps(fiz1,tz);
2009
2010             fjx0             = _mm256_add_ps(fjx0,tx);
2011             fjy0             = _mm256_add_ps(fjy0,ty);
2012             fjz0             = _mm256_add_ps(fjz0,tz);
2013
2014             }
2015
2016             /**************************
2017              * CALCULATE INTERACTIONS *
2018              **************************/
2019
2020             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2021             {
2022
2023             /* REACTION-FIELD ELECTROSTATICS */
2024             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
2025
2026             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2027
2028             fscal            = felec;
2029
2030             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2031
2032             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2033
2034             /* Calculate temporary vectorial force */
2035             tx               = _mm256_mul_ps(fscal,dx11);
2036             ty               = _mm256_mul_ps(fscal,dy11);
2037             tz               = _mm256_mul_ps(fscal,dz11);
2038
2039             /* Update vectorial force */
2040             fix1             = _mm256_add_ps(fix1,tx);
2041             fiy1             = _mm256_add_ps(fiy1,ty);
2042             fiz1             = _mm256_add_ps(fiz1,tz);
2043
2044             fjx1             = _mm256_add_ps(fjx1,tx);
2045             fjy1             = _mm256_add_ps(fjy1,ty);
2046             fjz1             = _mm256_add_ps(fjz1,tz);
2047
2048             }
2049
2050             /**************************
2051              * CALCULATE INTERACTIONS *
2052              **************************/
2053
2054             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2055             {
2056
2057             /* REACTION-FIELD ELECTROSTATICS */
2058             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
2059
2060             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2061
2062             fscal            = felec;
2063
2064             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2065
2066             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2067
2068             /* Calculate temporary vectorial force */
2069             tx               = _mm256_mul_ps(fscal,dx12);
2070             ty               = _mm256_mul_ps(fscal,dy12);
2071             tz               = _mm256_mul_ps(fscal,dz12);
2072
2073             /* Update vectorial force */
2074             fix1             = _mm256_add_ps(fix1,tx);
2075             fiy1             = _mm256_add_ps(fiy1,ty);
2076             fiz1             = _mm256_add_ps(fiz1,tz);
2077
2078             fjx2             = _mm256_add_ps(fjx2,tx);
2079             fjy2             = _mm256_add_ps(fjy2,ty);
2080             fjz2             = _mm256_add_ps(fjz2,tz);
2081
2082             }
2083
2084             /**************************
2085              * CALCULATE INTERACTIONS *
2086              **************************/
2087
2088             if (gmx_mm256_any_lt(rsq20,rcutoff2))
2089             {
2090
2091             /* REACTION-FIELD ELECTROSTATICS */
2092             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
2093
2094             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
2095
2096             fscal            = felec;
2097
2098             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2099
2100             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2101
2102             /* Calculate temporary vectorial force */
2103             tx               = _mm256_mul_ps(fscal,dx20);
2104             ty               = _mm256_mul_ps(fscal,dy20);
2105             tz               = _mm256_mul_ps(fscal,dz20);
2106
2107             /* Update vectorial force */
2108             fix2             = _mm256_add_ps(fix2,tx);
2109             fiy2             = _mm256_add_ps(fiy2,ty);
2110             fiz2             = _mm256_add_ps(fiz2,tz);
2111
2112             fjx0             = _mm256_add_ps(fjx0,tx);
2113             fjy0             = _mm256_add_ps(fjy0,ty);
2114             fjz0             = _mm256_add_ps(fjz0,tz);
2115
2116             }
2117
2118             /**************************
2119              * CALCULATE INTERACTIONS *
2120              **************************/
2121
2122             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2123             {
2124
2125             /* REACTION-FIELD ELECTROSTATICS */
2126             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
2127
2128             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2129
2130             fscal            = felec;
2131
2132             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2133
2134             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2135
2136             /* Calculate temporary vectorial force */
2137             tx               = _mm256_mul_ps(fscal,dx21);
2138             ty               = _mm256_mul_ps(fscal,dy21);
2139             tz               = _mm256_mul_ps(fscal,dz21);
2140
2141             /* Update vectorial force */
2142             fix2             = _mm256_add_ps(fix2,tx);
2143             fiy2             = _mm256_add_ps(fiy2,ty);
2144             fiz2             = _mm256_add_ps(fiz2,tz);
2145
2146             fjx1             = _mm256_add_ps(fjx1,tx);
2147             fjy1             = _mm256_add_ps(fjy1,ty);
2148             fjz1             = _mm256_add_ps(fjz1,tz);
2149
2150             }
2151
2152             /**************************
2153              * CALCULATE INTERACTIONS *
2154              **************************/
2155
2156             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2157             {
2158
2159             /* REACTION-FIELD ELECTROSTATICS */
2160             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
2161
2162             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2163
2164             fscal            = felec;
2165
2166             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2167
2168             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2169
2170             /* Calculate temporary vectorial force */
2171             tx               = _mm256_mul_ps(fscal,dx22);
2172             ty               = _mm256_mul_ps(fscal,dy22);
2173             tz               = _mm256_mul_ps(fscal,dz22);
2174
2175             /* Update vectorial force */
2176             fix2             = _mm256_add_ps(fix2,tx);
2177             fiy2             = _mm256_add_ps(fiy2,ty);
2178             fiz2             = _mm256_add_ps(fiz2,tz);
2179
2180             fjx2             = _mm256_add_ps(fjx2,tx);
2181             fjy2             = _mm256_add_ps(fjy2,ty);
2182             fjz2             = _mm256_add_ps(fjz2,tz);
2183
2184             }
2185
2186             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2187             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2188             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2189             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2190             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2191             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2192             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2193             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2194
2195             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2196                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2197
2198             /* Inner loop uses 277 flops */
2199         }
2200
2201         /* End of innermost loop */
2202
2203         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2204                                                  f+i_coord_offset,fshift+i_shift_offset);
2205
2206         /* Increment number of inner iterations */
2207         inneriter                  += j_index_end - j_index_start;
2208
2209         /* Outer loop uses 18 flops */
2210     }
2211
2212     /* Increment number of outer iterations */
2213     outeriter        += nri;
2214
2215     /* Update outer/inner flops */
2216
2217     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*277);
2218 }