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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_VF_avx_256_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  97     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  98     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  99     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 100     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 101     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 102     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 103     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 104     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 105     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 106     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 107     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 108     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 109     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 110     real             *charge;
 111     int              nvdwtype;
 112     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 113     int              *vdwtype;
 114     real             *vdwparam;
 115     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 116     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 117     __m256           dummy_mask,cutoff_mask;
 118     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 119     __m256           one     = _mm256_set1_ps(1.0);
 120     __m256           two     = _mm256_set1_ps(2.0);
 121     x                = xx[0];
 122     f                = ff[0];
 123
 124     nri              = nlist->nri;
 125     iinr             = nlist->iinr;
 126     jindex           = nlist->jindex;
 127     jjnr             = nlist->jjnr;
 128     shiftidx         = nlist->shift;
 129     gid              = nlist->gid;
 130     shiftvec         = fr->shift_vec[0];
 131     fshift           = fr->fshift[0];
 132     facel            = _mm256_set1_ps(fr->epsfac);
 133     charge           = mdatoms->chargeA;
 134     krf              = _mm256_set1_ps(fr->ic->k_rf);
 135     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 136     crf              = _mm256_set1_ps(fr->ic->c_rf);
 137     nvdwtype         = fr->ntype;
 138     vdwparam         = fr->nbfp;
 139     vdwtype          = mdatoms->typeA;
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 144     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 145     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 146     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 147
 148     jq0              = _mm256_set1_ps(charge[inr+0]);
 149     jq1              = _mm256_set1_ps(charge[inr+1]);
 150     jq2              = _mm256_set1_ps(charge[inr+2]);
 151     vdwjidx0A        = 2*vdwtype[inr+0];
 152     qq00             = _mm256_mul_ps(iq0,jq0);
 153     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 154     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 155     qq01             = _mm256_mul_ps(iq0,jq1);
 156     qq02             = _mm256_mul_ps(iq0,jq2);
 157     qq10             = _mm256_mul_ps(iq1,jq0);
 158     qq11             = _mm256_mul_ps(iq1,jq1);
 159     qq12             = _mm256_mul_ps(iq1,jq2);
 160     qq20             = _mm256_mul_ps(iq2,jq0);
 161     qq21             = _mm256_mul_ps(iq2,jq1);
 162     qq22             = _mm256_mul_ps(iq2,jq2);
 163
 164     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 165     rcutoff_scalar   = fr->rcoulomb;
 166     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 167     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 168
 169     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 170     rvdw             = _mm256_set1_ps(fr->rvdw);
 171
 172     /* Avoid stupid compiler warnings */
 173     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 174     j_coord_offsetA = 0;
 175     j_coord_offsetB = 0;
 176     j_coord_offsetC = 0;
 177     j_coord_offsetD = 0;
 178     j_coord_offsetE = 0;
 179     j_coord_offsetF = 0;
 180     j_coord_offsetG = 0;
 181     j_coord_offsetH = 0;
 182
 183     outeriter        = 0;
 184     inneriter        = 0;
 185
 186     for(iidx=0;iidx<4*DIM;iidx++)
 187     {
 188         scratch[iidx] = 0.0;
 189     }
 190
 191     /* Start outer loop over neighborlists */
 192     for(iidx=0; iidx<nri; iidx++)
 193     {
 194         /* Load shift vector for this list */
 195         i_shift_offset   = DIM*shiftidx[iidx];
 196
 197         /* Load limits for loop over neighbors */
 198         j_index_start    = jindex[iidx];
 199         j_index_end      = jindex[iidx+1];
 200
 201         /* Get outer coordinate index */
 202         inr              = iinr[iidx];
 203         i_coord_offset   = DIM*inr;
 204
 205         /* Load i particle coords and add shift vector */
 206         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 207                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 208
 209         fix0             = _mm256_setzero_ps();
 210         fiy0             = _mm256_setzero_ps();
 211         fiz0             = _mm256_setzero_ps();
 212         fix1             = _mm256_setzero_ps();
 213         fiy1             = _mm256_setzero_ps();
 214         fiz1             = _mm256_setzero_ps();
 215         fix2             = _mm256_setzero_ps();
 216         fiy2             = _mm256_setzero_ps();
 217         fiz2             = _mm256_setzero_ps();
 218
 219         /* Reset potential sums */
 220         velecsum         = _mm256_setzero_ps();
 221         vvdwsum          = _mm256_setzero_ps();
 222
 223         /* Start inner kernel loop */
 224         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 225         {
 226
 227             /* Get j neighbor index, and coordinate index */
 228             jnrA             = jjnr[jidx];
 229             jnrB             = jjnr[jidx+1];
 230             jnrC             = jjnr[jidx+2];
 231             jnrD             = jjnr[jidx+3];
 232             jnrE             = jjnr[jidx+4];
 233             jnrF             = jjnr[jidx+5];
 234             jnrG             = jjnr[jidx+6];
 235             jnrH             = jjnr[jidx+7];
 236             j_coord_offsetA  = DIM*jnrA;
 237             j_coord_offsetB  = DIM*jnrB;
 238             j_coord_offsetC  = DIM*jnrC;
 239             j_coord_offsetD  = DIM*jnrD;
 240             j_coord_offsetE  = DIM*jnrE;
 241             j_coord_offsetF  = DIM*jnrF;
 242             j_coord_offsetG  = DIM*jnrG;
 243             j_coord_offsetH  = DIM*jnrH;
 244
 245             /* load j atom coordinates */
 246             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 247                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 248                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 249                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 250                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 251
 252             /* Calculate displacement vector */
 253             dx00             = _mm256_sub_ps(ix0,jx0);
 254             dy00             = _mm256_sub_ps(iy0,jy0);
 255             dz00             = _mm256_sub_ps(iz0,jz0);
 256             dx01             = _mm256_sub_ps(ix0,jx1);
 257             dy01             = _mm256_sub_ps(iy0,jy1);
 258             dz01             = _mm256_sub_ps(iz0,jz1);
 259             dx02             = _mm256_sub_ps(ix0,jx2);
 260             dy02             = _mm256_sub_ps(iy0,jy2);
 261             dz02             = _mm256_sub_ps(iz0,jz2);
 262             dx10             = _mm256_sub_ps(ix1,jx0);
 263             dy10             = _mm256_sub_ps(iy1,jy0);
 264             dz10             = _mm256_sub_ps(iz1,jz0);
 265             dx11             = _mm256_sub_ps(ix1,jx1);
 266             dy11             = _mm256_sub_ps(iy1,jy1);
 267             dz11             = _mm256_sub_ps(iz1,jz1);
 268             dx12             = _mm256_sub_ps(ix1,jx2);
 269             dy12             = _mm256_sub_ps(iy1,jy2);
 270             dz12             = _mm256_sub_ps(iz1,jz2);
 271             dx20             = _mm256_sub_ps(ix2,jx0);
 272             dy20             = _mm256_sub_ps(iy2,jy0);
 273             dz20             = _mm256_sub_ps(iz2,jz0);
 274             dx21             = _mm256_sub_ps(ix2,jx1);
 275             dy21             = _mm256_sub_ps(iy2,jy1);
 276             dz21             = _mm256_sub_ps(iz2,jz1);
 277             dx22             = _mm256_sub_ps(ix2,jx2);
 278             dy22             = _mm256_sub_ps(iy2,jy2);
 279             dz22             = _mm256_sub_ps(iz2,jz2);
 280
 281             /* Calculate squared distance and things based on it */
 282             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 283             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 284             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 285             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 286             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 287             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 288             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 289             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 290             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 291
 292             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 293             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 294             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 295             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 296             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 297             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 298             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 299             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 300             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 301
 302             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 303             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 304             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 305             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 306             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 307             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 308             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 309             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 310             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 311
 312             fjx0             = _mm256_setzero_ps();
 313             fjy0             = _mm256_setzero_ps();
 314             fjz0             = _mm256_setzero_ps();
 315             fjx1             = _mm256_setzero_ps();
 316             fjy1             = _mm256_setzero_ps();
 317             fjz1             = _mm256_setzero_ps();
 318             fjx2             = _mm256_setzero_ps();
 319             fjy2             = _mm256_setzero_ps();
 320             fjz2             = _mm256_setzero_ps();
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 327             {
 328
 329             /* REACTION-FIELD ELECTROSTATICS */
 330             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 331             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 332
 333             /* LENNARD-JONES DISPERSION/REPULSION */
 334
 335             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 336             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 337             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 338             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) ,
 339                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 340             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 341
 342             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velec            = _mm256_and_ps(velec,cutoff_mask);
 346             velecsum         = _mm256_add_ps(velecsum,velec);
 347             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 348             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 349
 350             fscal            = _mm256_add_ps(felec,fvdw);
 351
 352             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 353
 354             /* Calculate temporary vectorial force */
 355             tx               = _mm256_mul_ps(fscal,dx00);
 356             ty               = _mm256_mul_ps(fscal,dy00);
 357             tz               = _mm256_mul_ps(fscal,dz00);
 358
 359             /* Update vectorial force */
 360             fix0             = _mm256_add_ps(fix0,tx);
 361             fiy0             = _mm256_add_ps(fiy0,ty);
 362             fiz0             = _mm256_add_ps(fiz0,tz);
 363
 364             fjx0             = _mm256_add_ps(fjx0,tx);
 365             fjy0             = _mm256_add_ps(fjy0,ty);
 366             fjz0             = _mm256_add_ps(fjz0,tz);
 367
 368             }
 369
 370             /**************************
 371              * CALCULATE INTERACTIONS *
 372              **************************/
 373
 374             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 375             {
 376
 377             /* REACTION-FIELD ELECTROSTATICS */
 378             velec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
 379             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
 380
 381             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 382
 383             /* Update potential sum for this i atom from the interaction with this j atom. */
 384             velec            = _mm256_and_ps(velec,cutoff_mask);
 385             velecsum         = _mm256_add_ps(velecsum,velec);
 386
 387             fscal            = felec;
 388
 389             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 390
 391             /* Calculate temporary vectorial force */
 392             tx               = _mm256_mul_ps(fscal,dx01);
 393             ty               = _mm256_mul_ps(fscal,dy01);
 394             tz               = _mm256_mul_ps(fscal,dz01);
 395
 396             /* Update vectorial force */
 397             fix0             = _mm256_add_ps(fix0,tx);
 398             fiy0             = _mm256_add_ps(fiy0,ty);
 399             fiz0             = _mm256_add_ps(fiz0,tz);
 400
 401             fjx1             = _mm256_add_ps(fjx1,tx);
 402             fjy1             = _mm256_add_ps(fjy1,ty);
 403             fjz1             = _mm256_add_ps(fjz1,tz);
 404
 405             }
 406
 407             /**************************
 408              * CALCULATE INTERACTIONS *
 409              **************************/
 410
 411             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 412             {
 413
 414             /* REACTION-FIELD ELECTROSTATICS */
 415             velec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
 416             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
 417
 418             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
 419
 420             /* Update potential sum for this i atom from the interaction with this j atom. */
 421             velec            = _mm256_and_ps(velec,cutoff_mask);
 422             velecsum         = _mm256_add_ps(velecsum,velec);
 423
 424             fscal            = felec;
 425
 426             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 427
 428             /* Calculate temporary vectorial force */
 429             tx               = _mm256_mul_ps(fscal,dx02);
 430             ty               = _mm256_mul_ps(fscal,dy02);
 431             tz               = _mm256_mul_ps(fscal,dz02);
 432
 433             /* Update vectorial force */
 434             fix0             = _mm256_add_ps(fix0,tx);
 435             fiy0             = _mm256_add_ps(fiy0,ty);
 436             fiz0             = _mm256_add_ps(fiz0,tz);
 437
 438             fjx2             = _mm256_add_ps(fjx2,tx);
 439             fjy2             = _mm256_add_ps(fjy2,ty);
 440             fjz2             = _mm256_add_ps(fjz2,tz);
 441
 442             }
 443
 444             /**************************
 445              * CALCULATE INTERACTIONS *
 446              **************************/
 447
 448             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 449             {
 450
 451             /* REACTION-FIELD ELECTROSTATICS */
 452             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 453             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 454
 455             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 456
 457             /* Update potential sum for this i atom from the interaction with this j atom. */
 458             velec            = _mm256_and_ps(velec,cutoff_mask);
 459             velecsum         = _mm256_add_ps(velecsum,velec);
 460
 461             fscal            = felec;
 462
 463             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 464
 465             /* Calculate temporary vectorial force */
 466             tx               = _mm256_mul_ps(fscal,dx10);
 467             ty               = _mm256_mul_ps(fscal,dy10);
 468             tz               = _mm256_mul_ps(fscal,dz10);
 469
 470             /* Update vectorial force */
 471             fix1             = _mm256_add_ps(fix1,tx);
 472             fiy1             = _mm256_add_ps(fiy1,ty);
 473             fiz1             = _mm256_add_ps(fiz1,tz);
 474
 475             fjx0             = _mm256_add_ps(fjx0,tx);
 476             fjy0             = _mm256_add_ps(fjy0,ty);
 477             fjz0             = _mm256_add_ps(fjz0,tz);
 478
 479             }
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 486             {
 487
 488             /* REACTION-FIELD ELECTROSTATICS */
 489             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 490             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 491
 492             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 493
 494             /* Update potential sum for this i atom from the interaction with this j atom. */
 495             velec            = _mm256_and_ps(velec,cutoff_mask);
 496             velecsum         = _mm256_add_ps(velecsum,velec);
 497
 498             fscal            = felec;
 499
 500             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 501
 502             /* Calculate temporary vectorial force */
 503             tx               = _mm256_mul_ps(fscal,dx11);
 504             ty               = _mm256_mul_ps(fscal,dy11);
 505             tz               = _mm256_mul_ps(fscal,dz11);
 506
 507             /* Update vectorial force */
 508             fix1             = _mm256_add_ps(fix1,tx);
 509             fiy1             = _mm256_add_ps(fiy1,ty);
 510             fiz1             = _mm256_add_ps(fiz1,tz);
 511
 512             fjx1             = _mm256_add_ps(fjx1,tx);
 513             fjy1             = _mm256_add_ps(fjy1,ty);
 514             fjz1             = _mm256_add_ps(fjz1,tz);
 515
 516             }
 517
 518             /**************************
 519              * CALCULATE INTERACTIONS *
 520              **************************/
 521
 522             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 523             {
 524
 525             /* REACTION-FIELD ELECTROSTATICS */
 526             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 527             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 528
 529             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 530
 531             /* Update potential sum for this i atom from the interaction with this j atom. */
 532             velec            = _mm256_and_ps(velec,cutoff_mask);
 533             velecsum         = _mm256_add_ps(velecsum,velec);
 534
 535             fscal            = felec;
 536
 537             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 538
 539             /* Calculate temporary vectorial force */
 540             tx               = _mm256_mul_ps(fscal,dx12);
 541             ty               = _mm256_mul_ps(fscal,dy12);
 542             tz               = _mm256_mul_ps(fscal,dz12);
 543
 544             /* Update vectorial force */
 545             fix1             = _mm256_add_ps(fix1,tx);
 546             fiy1             = _mm256_add_ps(fiy1,ty);
 547             fiz1             = _mm256_add_ps(fiz1,tz);
 548
 549             fjx2             = _mm256_add_ps(fjx2,tx);
 550             fjy2             = _mm256_add_ps(fjy2,ty);
 551             fjz2             = _mm256_add_ps(fjz2,tz);
 552
 553             }
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 560             {
 561
 562             /* REACTION-FIELD ELECTROSTATICS */
 563             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 564             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 565
 566             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 567
 568             /* Update potential sum for this i atom from the interaction with this j atom. */
 569             velec            = _mm256_and_ps(velec,cutoff_mask);
 570             velecsum         = _mm256_add_ps(velecsum,velec);
 571
 572             fscal            = felec;
 573
 574             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 575
 576             /* Calculate temporary vectorial force */
 577             tx               = _mm256_mul_ps(fscal,dx20);
 578             ty               = _mm256_mul_ps(fscal,dy20);
 579             tz               = _mm256_mul_ps(fscal,dz20);
 580
 581             /* Update vectorial force */
 582             fix2             = _mm256_add_ps(fix2,tx);
 583             fiy2             = _mm256_add_ps(fiy2,ty);
 584             fiz2             = _mm256_add_ps(fiz2,tz);
 585
 586             fjx0             = _mm256_add_ps(fjx0,tx);
 587             fjy0             = _mm256_add_ps(fjy0,ty);
 588             fjz0             = _mm256_add_ps(fjz0,tz);
 589
 590             }
 591
 592             /**************************
 593              * CALCULATE INTERACTIONS *
 594              **************************/
 595
 596             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 597             {
 598
 599             /* REACTION-FIELD ELECTROSTATICS */
 600             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
 601             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
 602
 603             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _mm256_and_ps(velec,cutoff_mask);
 607             velecsum         = _mm256_add_ps(velecsum,velec);
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm256_mul_ps(fscal,dx21);
 615             ty               = _mm256_mul_ps(fscal,dy21);
 616             tz               = _mm256_mul_ps(fscal,dz21);
 617
 618             /* Update vectorial force */
 619             fix2             = _mm256_add_ps(fix2,tx);
 620             fiy2             = _mm256_add_ps(fiy2,ty);
 621             fiz2             = _mm256_add_ps(fiz2,tz);
 622
 623             fjx1             = _mm256_add_ps(fjx1,tx);
 624             fjy1             = _mm256_add_ps(fjy1,ty);
 625             fjz1             = _mm256_add_ps(fjz1,tz);
 626
 627             }
 628
 629             /**************************
 630              * CALCULATE INTERACTIONS *
 631              **************************/
 632
 633             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 634             {
 635
 636             /* REACTION-FIELD ELECTROSTATICS */
 637             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
 638             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
 639
 640             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
 641
 642             /* Update potential sum for this i atom from the interaction with this j atom. */
 643             velec            = _mm256_and_ps(velec,cutoff_mask);
 644             velecsum         = _mm256_add_ps(velecsum,velec);
 645
 646             fscal            = felec;
 647
 648             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 649
 650             /* Calculate temporary vectorial force */
 651             tx               = _mm256_mul_ps(fscal,dx22);
 652             ty               = _mm256_mul_ps(fscal,dy22);
 653             tz               = _mm256_mul_ps(fscal,dz22);
 654
 655             /* Update vectorial force */
 656             fix2             = _mm256_add_ps(fix2,tx);
 657             fiy2             = _mm256_add_ps(fiy2,ty);
 658             fiz2             = _mm256_add_ps(fiz2,tz);
 659
 660             fjx2             = _mm256_add_ps(fjx2,tx);
 661             fjy2             = _mm256_add_ps(fjy2,ty);
 662             fjz2             = _mm256_add_ps(fjz2,tz);
 663
 664             }
 665
 666             fjptrA             = f+j_coord_offsetA;
 667             fjptrB             = f+j_coord_offsetB;
 668             fjptrC             = f+j_coord_offsetC;
 669             fjptrD             = f+j_coord_offsetD;
 670             fjptrE             = f+j_coord_offsetE;
 671             fjptrF             = f+j_coord_offsetF;
 672             fjptrG             = f+j_coord_offsetG;
 673             fjptrH             = f+j_coord_offsetH;
 674
 675             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 676                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 677
 678             /* Inner loop uses 342 flops */
 679         }
 680
 681         if(jidx<j_index_end)
 682         {
 683
 684             /* Get j neighbor index, and coordinate index */
 685             jnrlistA         = jjnr[jidx];
 686             jnrlistB         = jjnr[jidx+1];
 687             jnrlistC         = jjnr[jidx+2];
 688             jnrlistD         = jjnr[jidx+3];
 689             jnrlistE         = jjnr[jidx+4];
 690             jnrlistF         = jjnr[jidx+5];
 691             jnrlistG         = jjnr[jidx+6];
 692             jnrlistH         = jjnr[jidx+7];
 693             /* Sign of each element will be negative for non-real atoms.
 694              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 695              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 696              */
 697             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 698                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 699
 700             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 701             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 702             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 703             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 704             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 705             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 706             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 707             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 708             j_coord_offsetA  = DIM*jnrA;
 709             j_coord_offsetB  = DIM*jnrB;
 710             j_coord_offsetC  = DIM*jnrC;
 711             j_coord_offsetD  = DIM*jnrD;
 712             j_coord_offsetE  = DIM*jnrE;
 713             j_coord_offsetF  = DIM*jnrF;
 714             j_coord_offsetG  = DIM*jnrG;
 715             j_coord_offsetH  = DIM*jnrH;
 716
 717             /* load j atom coordinates */
 718             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 719                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 720                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 721                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 722                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 723
 724             /* Calculate displacement vector */
 725             dx00             = _mm256_sub_ps(ix0,jx0);
 726             dy00             = _mm256_sub_ps(iy0,jy0);
 727             dz00             = _mm256_sub_ps(iz0,jz0);
 728             dx01             = _mm256_sub_ps(ix0,jx1);
 729             dy01             = _mm256_sub_ps(iy0,jy1);
 730             dz01             = _mm256_sub_ps(iz0,jz1);
 731             dx02             = _mm256_sub_ps(ix0,jx2);
 732             dy02             = _mm256_sub_ps(iy0,jy2);
 733             dz02             = _mm256_sub_ps(iz0,jz2);
 734             dx10             = _mm256_sub_ps(ix1,jx0);
 735             dy10             = _mm256_sub_ps(iy1,jy0);
 736             dz10             = _mm256_sub_ps(iz1,jz0);
 737             dx11             = _mm256_sub_ps(ix1,jx1);
 738             dy11             = _mm256_sub_ps(iy1,jy1);
 739             dz11             = _mm256_sub_ps(iz1,jz1);
 740             dx12             = _mm256_sub_ps(ix1,jx2);
 741             dy12             = _mm256_sub_ps(iy1,jy2);
 742             dz12             = _mm256_sub_ps(iz1,jz2);
 743             dx20             = _mm256_sub_ps(ix2,jx0);
 744             dy20             = _mm256_sub_ps(iy2,jy0);
 745             dz20             = _mm256_sub_ps(iz2,jz0);
 746             dx21             = _mm256_sub_ps(ix2,jx1);
 747             dy21             = _mm256_sub_ps(iy2,jy1);
 748             dz21             = _mm256_sub_ps(iz2,jz1);
 749             dx22             = _mm256_sub_ps(ix2,jx2);
 750             dy22             = _mm256_sub_ps(iy2,jy2);
 751             dz22             = _mm256_sub_ps(iz2,jz2);
 752
 753             /* Calculate squared distance and things based on it */
 754             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 755             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 756             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 757             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 758             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 759             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 760             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 761             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 762             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 763
 764             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 765             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 766             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 767             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 768             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 769             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 770             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 771             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 772             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 773
 774             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 775             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 776             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 777             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 778             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 779             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 780             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 781             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 782             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 783
 784             fjx0             = _mm256_setzero_ps();
 785             fjy0             = _mm256_setzero_ps();
 786             fjz0             = _mm256_setzero_ps();
 787             fjx1             = _mm256_setzero_ps();
 788             fjy1             = _mm256_setzero_ps();
 789             fjz1             = _mm256_setzero_ps();
 790             fjx2             = _mm256_setzero_ps();
 791             fjy2             = _mm256_setzero_ps();
 792             fjz2             = _mm256_setzero_ps();
 793
 794             /**************************
 795              * CALCULATE INTERACTIONS *
 796              **************************/
 797
 798             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 799             {
 800
 801             /* REACTION-FIELD ELECTROSTATICS */
 802             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 803             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 804
 805             /* LENNARD-JONES DISPERSION/REPULSION */
 806
 807             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 808             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 809             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 810             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) ,
 811                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 812             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 813
 814             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 815
 816             /* Update potential sum for this i atom from the interaction with this j atom. */
 817             velec            = _mm256_and_ps(velec,cutoff_mask);
 818             velec            = _mm256_andnot_ps(dummy_mask,velec);
 819             velecsum         = _mm256_add_ps(velecsum,velec);
 820             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 821             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 822             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 823
 824             fscal            = _mm256_add_ps(felec,fvdw);
 825
 826             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 827
 828             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 829
 830             /* Calculate temporary vectorial force */
 831             tx               = _mm256_mul_ps(fscal,dx00);
 832             ty               = _mm256_mul_ps(fscal,dy00);
 833             tz               = _mm256_mul_ps(fscal,dz00);
 834
 835             /* Update vectorial force */
 836             fix0             = _mm256_add_ps(fix0,tx);
 837             fiy0             = _mm256_add_ps(fiy0,ty);
 838             fiz0             = _mm256_add_ps(fiz0,tz);
 839
 840             fjx0             = _mm256_add_ps(fjx0,tx);
 841             fjy0             = _mm256_add_ps(fjy0,ty);
 842             fjz0             = _mm256_add_ps(fjz0,tz);
 843
 844             }
 845
 846             /**************************
 847              * CALCULATE INTERACTIONS *
 848              **************************/
 849
 850             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 851             {
 852
 853             /* REACTION-FIELD ELECTROSTATICS */
 854             velec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
 855             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
 856
 857             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 858
 859             /* Update potential sum for this i atom from the interaction with this j atom. */
 860             velec            = _mm256_and_ps(velec,cutoff_mask);
 861             velec            = _mm256_andnot_ps(dummy_mask,velec);
 862             velecsum         = _mm256_add_ps(velecsum,velec);
 863
 864             fscal            = felec;
 865
 866             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 867
 868             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 869
 870             /* Calculate temporary vectorial force */
 871             tx               = _mm256_mul_ps(fscal,dx01);
 872             ty               = _mm256_mul_ps(fscal,dy01);
 873             tz               = _mm256_mul_ps(fscal,dz01);
 874
 875             /* Update vectorial force */
 876             fix0             = _mm256_add_ps(fix0,tx);
 877             fiy0             = _mm256_add_ps(fiy0,ty);
 878             fiz0             = _mm256_add_ps(fiz0,tz);
 879
 880             fjx1             = _mm256_add_ps(fjx1,tx);
 881             fjy1             = _mm256_add_ps(fjy1,ty);
 882             fjz1             = _mm256_add_ps(fjz1,tz);
 883
 884             }
 885
 886             /**************************
 887              * CALCULATE INTERACTIONS *
 888              **************************/
 889
 890             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 891             {
 892
 893             /* REACTION-FIELD ELECTROSTATICS */
 894             velec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
 895             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
 896
 897             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
 898
 899             /* Update potential sum for this i atom from the interaction with this j atom. */
 900             velec            = _mm256_and_ps(velec,cutoff_mask);
 901             velec            = _mm256_andnot_ps(dummy_mask,velec);
 902             velecsum         = _mm256_add_ps(velecsum,velec);
 903
 904             fscal            = felec;
 905
 906             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 907
 908             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 909
 910             /* Calculate temporary vectorial force */
 911             tx               = _mm256_mul_ps(fscal,dx02);
 912             ty               = _mm256_mul_ps(fscal,dy02);
 913             tz               = _mm256_mul_ps(fscal,dz02);
 914
 915             /* Update vectorial force */
 916             fix0             = _mm256_add_ps(fix0,tx);
 917             fiy0             = _mm256_add_ps(fiy0,ty);
 918             fiz0             = _mm256_add_ps(fiz0,tz);
 919
 920             fjx2             = _mm256_add_ps(fjx2,tx);
 921             fjy2             = _mm256_add_ps(fjy2,ty);
 922             fjz2             = _mm256_add_ps(fjz2,tz);
 923
 924             }
 925
 926             /**************************
 927              * CALCULATE INTERACTIONS *
 928              **************************/
 929
 930             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 931             {
 932
 933             /* REACTION-FIELD ELECTROSTATICS */
 934             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 935             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 936
 937             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 938
 939             /* Update potential sum for this i atom from the interaction with this j atom. */
 940             velec            = _mm256_and_ps(velec,cutoff_mask);
 941             velec            = _mm256_andnot_ps(dummy_mask,velec);
 942             velecsum         = _mm256_add_ps(velecsum,velec);
 943
 944             fscal            = felec;
 945
 946             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 947
 948             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 949
 950             /* Calculate temporary vectorial force */
 951             tx               = _mm256_mul_ps(fscal,dx10);
 952             ty               = _mm256_mul_ps(fscal,dy10);
 953             tz               = _mm256_mul_ps(fscal,dz10);
 954
 955             /* Update vectorial force */
 956             fix1             = _mm256_add_ps(fix1,tx);
 957             fiy1             = _mm256_add_ps(fiy1,ty);
 958             fiz1             = _mm256_add_ps(fiz1,tz);
 959
 960             fjx0             = _mm256_add_ps(fjx0,tx);
 961             fjy0             = _mm256_add_ps(fjy0,ty);
 962             fjz0             = _mm256_add_ps(fjz0,tz);
 963
 964             }
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 971             {
 972
 973             /* REACTION-FIELD ELECTROSTATICS */
 974             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 975             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 976
 977             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 978
 979             /* Update potential sum for this i atom from the interaction with this j atom. */
 980             velec            = _mm256_and_ps(velec,cutoff_mask);
 981             velec            = _mm256_andnot_ps(dummy_mask,velec);
 982             velecsum         = _mm256_add_ps(velecsum,velec);
 983
 984             fscal            = felec;
 985
 986             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 987
 988             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 989
 990             /* Calculate temporary vectorial force */
 991             tx               = _mm256_mul_ps(fscal,dx11);
 992             ty               = _mm256_mul_ps(fscal,dy11);
 993             tz               = _mm256_mul_ps(fscal,dz11);
 994
 995             /* Update vectorial force */
 996             fix1             = _mm256_add_ps(fix1,tx);
 997             fiy1             = _mm256_add_ps(fiy1,ty);
 998             fiz1             = _mm256_add_ps(fiz1,tz);
 999
1000             fjx1             = _mm256_add_ps(fjx1,tx);
1001             fjy1             = _mm256_add_ps(fjy1,ty);
1002             fjz1             = _mm256_add_ps(fjz1,tz);
1003
1004             }
1005
1006             /**************************
1007              * CALCULATE INTERACTIONS *
1008              **************************/
1009
1010             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1011             {
1012
1013             /* REACTION-FIELD ELECTROSTATICS */
1014             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
1015             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1016
1017             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1018
1019             /* Update potential sum for this i atom from the interaction with this j atom. */
1020             velec            = _mm256_and_ps(velec,cutoff_mask);
1021             velec            = _mm256_andnot_ps(dummy_mask,velec);
1022             velecsum         = _mm256_add_ps(velecsum,velec);
1023
1024             fscal            = felec;
1025
1026             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1027
1028             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1029
1030             /* Calculate temporary vectorial force */
1031             tx               = _mm256_mul_ps(fscal,dx12);
1032             ty               = _mm256_mul_ps(fscal,dy12);
1033             tz               = _mm256_mul_ps(fscal,dz12);
1034
1035             /* Update vectorial force */
1036             fix1             = _mm256_add_ps(fix1,tx);
1037             fiy1             = _mm256_add_ps(fiy1,ty);
1038             fiz1             = _mm256_add_ps(fiz1,tz);
1039
1040             fjx2             = _mm256_add_ps(fjx2,tx);
1041             fjy2             = _mm256_add_ps(fjy2,ty);
1042             fjz2             = _mm256_add_ps(fjz2,tz);
1043
1044             }
1045
1046             /**************************
1047              * CALCULATE INTERACTIONS *
1048              **************************/
1049
1050             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1051             {
1052
1053             /* REACTION-FIELD ELECTROSTATICS */
1054             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
1055             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1056
1057             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1058
1059             /* Update potential sum for this i atom from the interaction with this j atom. */
1060             velec            = _mm256_and_ps(velec,cutoff_mask);
1061             velec            = _mm256_andnot_ps(dummy_mask,velec);
1062             velecsum         = _mm256_add_ps(velecsum,velec);
1063
1064             fscal            = felec;
1065
1066             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1067
1068             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1069
1070             /* Calculate temporary vectorial force */
1071             tx               = _mm256_mul_ps(fscal,dx20);
1072             ty               = _mm256_mul_ps(fscal,dy20);
1073             tz               = _mm256_mul_ps(fscal,dz20);
1074
1075             /* Update vectorial force */
1076             fix2             = _mm256_add_ps(fix2,tx);
1077             fiy2             = _mm256_add_ps(fiy2,ty);
1078             fiz2             = _mm256_add_ps(fiz2,tz);
1079
1080             fjx0             = _mm256_add_ps(fjx0,tx);
1081             fjy0             = _mm256_add_ps(fjy0,ty);
1082             fjz0             = _mm256_add_ps(fjz0,tz);
1083
1084             }
1085
1086             /**************************
1087              * CALCULATE INTERACTIONS *
1088              **************************/
1089
1090             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1091             {
1092
1093             /* REACTION-FIELD ELECTROSTATICS */
1094             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
1095             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1096
1097             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1098
1099             /* Update potential sum for this i atom from the interaction with this j atom. */
1100             velec            = _mm256_and_ps(velec,cutoff_mask);
1101             velec            = _mm256_andnot_ps(dummy_mask,velec);
1102             velecsum         = _mm256_add_ps(velecsum,velec);
1103
1104             fscal            = felec;
1105
1106             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1107
1108             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1109
1110             /* Calculate temporary vectorial force */
1111             tx               = _mm256_mul_ps(fscal,dx21);
1112             ty               = _mm256_mul_ps(fscal,dy21);
1113             tz               = _mm256_mul_ps(fscal,dz21);
1114
1115             /* Update vectorial force */
1116             fix2             = _mm256_add_ps(fix2,tx);
1117             fiy2             = _mm256_add_ps(fiy2,ty);
1118             fiz2             = _mm256_add_ps(fiz2,tz);
1119
1120             fjx1             = _mm256_add_ps(fjx1,tx);
1121             fjy1             = _mm256_add_ps(fjy1,ty);
1122             fjz1             = _mm256_add_ps(fjz1,tz);
1123
1124             }
1125
1126             /**************************
1127              * CALCULATE INTERACTIONS *
1128              **************************/
1129
1130             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1131             {
1132
1133             /* REACTION-FIELD ELECTROSTATICS */
1134             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
1135             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1136
1137             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1138
1139             /* Update potential sum for this i atom from the interaction with this j atom. */
1140             velec            = _mm256_and_ps(velec,cutoff_mask);
1141             velec            = _mm256_andnot_ps(dummy_mask,velec);
1142             velecsum         = _mm256_add_ps(velecsum,velec);
1143
1144             fscal            = felec;
1145
1146             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1147
1148             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1149
1150             /* Calculate temporary vectorial force */
1151             tx               = _mm256_mul_ps(fscal,dx22);
1152             ty               = _mm256_mul_ps(fscal,dy22);
1153             tz               = _mm256_mul_ps(fscal,dz22);
1154
1155             /* Update vectorial force */
1156             fix2             = _mm256_add_ps(fix2,tx);
1157             fiy2             = _mm256_add_ps(fiy2,ty);
1158             fiz2             = _mm256_add_ps(fiz2,tz);
1159
1160             fjx2             = _mm256_add_ps(fjx2,tx);
1161             fjy2             = _mm256_add_ps(fjy2,ty);
1162             fjz2             = _mm256_add_ps(fjz2,tz);
1163
1164             }
1165
1166             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1167             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1168             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1169             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1170             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1171             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1172             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1173             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1174
1175             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1176                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1177
1178             /* Inner loop uses 342 flops */
1179         }
1180
1181         /* End of innermost loop */
1182
1183         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1184                                                  f+i_coord_offset,fshift+i_shift_offset);
1185
1186         ggid                        = gid[iidx];
1187         /* Update potential energies */
1188         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1189         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1190
1191         /* Increment number of inner iterations */
1192         inneriter                  += j_index_end - j_index_start;
1193
1194         /* Outer loop uses 20 flops */
1195     }
1196
1197     /* Increment number of outer iterations */
1198     outeriter        += nri;
1199
1200     /* Update outer/inner flops */
1201
1202     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*342);
1203 }
1204 /*
1205  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_F_avx_256_single
1206  * Electrostatics interaction: ReactionField
1207  * VdW interaction:            LennardJones
1208  * Geometry:                   Water3-Water3
1209  * Calculate force/pot:        Force
1210  */
1211 void
1212 nb_kernel_ElecRFCut_VdwLJSh_GeomW3W3_F_avx_256_single
1213                     (t_nblist                    * gmx_restrict       nlist,
1214                      rvec                        * gmx_restrict          xx,
1215                      rvec                        * gmx_restrict          ff,
1216                      t_forcerec                  * gmx_restrict          fr,
1217                      t_mdatoms                   * gmx_restrict     mdatoms,
1218                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1219                      t_nrnb                      * gmx_restrict        nrnb)
1220 {
1221     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1222      * just 0 for non-waters.
1223      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1224      * jnr indices corresponding to data put in the four positions in the SIMD register.
1225      */
1226     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1227     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1228     int              jnrA,jnrB,jnrC,jnrD;
1229     int              jnrE,jnrF,jnrG,jnrH;
1230     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1231     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1232     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1233     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1234     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1235     real             rcutoff_scalar;
1236     real             *shiftvec,*fshift,*x,*f;
1237     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1238     real             scratch[4*DIM];
1239     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1240     real *           vdwioffsetptr0;
1241     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1242     real *           vdwioffsetptr1;
1243     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1244     real *           vdwioffsetptr2;
1245     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1246     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1247     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1248     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1249     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1250     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1251     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1252     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1253     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1254     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1255     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1256     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1257     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1258     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1259     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1260     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1261     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1262     real             *charge;
1263     int              nvdwtype;
1264     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1265     int              *vdwtype;
1266     real             *vdwparam;
1267     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1268     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1269     __m256           dummy_mask,cutoff_mask;
1270     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1271     __m256           one     = _mm256_set1_ps(1.0);
1272     __m256           two     = _mm256_set1_ps(2.0);
1273     x                = xx[0];
1274     f                = ff[0];
1275
1276     nri              = nlist->nri;
1277     iinr             = nlist->iinr;
1278     jindex           = nlist->jindex;
1279     jjnr             = nlist->jjnr;
1280     shiftidx         = nlist->shift;
1281     gid              = nlist->gid;
1282     shiftvec         = fr->shift_vec[0];
1283     fshift           = fr->fshift[0];
1284     facel            = _mm256_set1_ps(fr->epsfac);
1285     charge           = mdatoms->chargeA;
1286     krf              = _mm256_set1_ps(fr->ic->k_rf);
1287     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
1288     crf              = _mm256_set1_ps(fr->ic->c_rf);
1289     nvdwtype         = fr->ntype;
1290     vdwparam         = fr->nbfp;
1291     vdwtype          = mdatoms->typeA;
1292
1293     /* Setup water-specific parameters */
1294     inr              = nlist->iinr[0];
1295     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1296     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1297     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1298     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1299
1300     jq0              = _mm256_set1_ps(charge[inr+0]);
1301     jq1              = _mm256_set1_ps(charge[inr+1]);
1302     jq2              = _mm256_set1_ps(charge[inr+2]);
1303     vdwjidx0A        = 2*vdwtype[inr+0];
1304     qq00             = _mm256_mul_ps(iq0,jq0);
1305     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1306     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1307     qq01             = _mm256_mul_ps(iq0,jq1);
1308     qq02             = _mm256_mul_ps(iq0,jq2);
1309     qq10             = _mm256_mul_ps(iq1,jq0);
1310     qq11             = _mm256_mul_ps(iq1,jq1);
1311     qq12             = _mm256_mul_ps(iq1,jq2);
1312     qq20             = _mm256_mul_ps(iq2,jq0);
1313     qq21             = _mm256_mul_ps(iq2,jq1);
1314     qq22             = _mm256_mul_ps(iq2,jq2);
1315
1316     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1317     rcutoff_scalar   = fr->rcoulomb;
1318     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1319     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1320
1321     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
1322     rvdw             = _mm256_set1_ps(fr->rvdw);
1323
1324     /* Avoid stupid compiler warnings */
1325     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1326     j_coord_offsetA = 0;
1327     j_coord_offsetB = 0;
1328     j_coord_offsetC = 0;
1329     j_coord_offsetD = 0;
1330     j_coord_offsetE = 0;
1331     j_coord_offsetF = 0;
1332     j_coord_offsetG = 0;
1333     j_coord_offsetH = 0;
1334
1335     outeriter        = 0;
1336     inneriter        = 0;
1337
1338     for(iidx=0;iidx<4*DIM;iidx++)
1339     {
1340         scratch[iidx] = 0.0;
1341     }
1342
1343     /* Start outer loop over neighborlists */
1344     for(iidx=0; iidx<nri; iidx++)
1345     {
1346         /* Load shift vector for this list */
1347         i_shift_offset   = DIM*shiftidx[iidx];
1348
1349         /* Load limits for loop over neighbors */
1350         j_index_start    = jindex[iidx];
1351         j_index_end      = jindex[iidx+1];
1352
1353         /* Get outer coordinate index */
1354         inr              = iinr[iidx];
1355         i_coord_offset   = DIM*inr;
1356
1357         /* Load i particle coords and add shift vector */
1358         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1359                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1360
1361         fix0             = _mm256_setzero_ps();
1362         fiy0             = _mm256_setzero_ps();
1363         fiz0             = _mm256_setzero_ps();
1364         fix1             = _mm256_setzero_ps();
1365         fiy1             = _mm256_setzero_ps();
1366         fiz1             = _mm256_setzero_ps();
1367         fix2             = _mm256_setzero_ps();
1368         fiy2             = _mm256_setzero_ps();
1369         fiz2             = _mm256_setzero_ps();
1370
1371         /* Start inner kernel loop */
1372         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1373         {
1374
1375             /* Get j neighbor index, and coordinate index */
1376             jnrA             = jjnr[jidx];
1377             jnrB             = jjnr[jidx+1];
1378             jnrC             = jjnr[jidx+2];
1379             jnrD             = jjnr[jidx+3];
1380             jnrE             = jjnr[jidx+4];
1381             jnrF             = jjnr[jidx+5];
1382             jnrG             = jjnr[jidx+6];
1383             jnrH             = jjnr[jidx+7];
1384             j_coord_offsetA  = DIM*jnrA;
1385             j_coord_offsetB  = DIM*jnrB;
1386             j_coord_offsetC  = DIM*jnrC;
1387             j_coord_offsetD  = DIM*jnrD;
1388             j_coord_offsetE  = DIM*jnrE;
1389             j_coord_offsetF  = DIM*jnrF;
1390             j_coord_offsetG  = DIM*jnrG;
1391             j_coord_offsetH  = DIM*jnrH;
1392
1393             /* load j atom coordinates */
1394             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1395                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1396                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1397                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1398                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1399
1400             /* Calculate displacement vector */
1401             dx00             = _mm256_sub_ps(ix0,jx0);
1402             dy00             = _mm256_sub_ps(iy0,jy0);
1403             dz00             = _mm256_sub_ps(iz0,jz0);
1404             dx01             = _mm256_sub_ps(ix0,jx1);
1405             dy01             = _mm256_sub_ps(iy0,jy1);
1406             dz01             = _mm256_sub_ps(iz0,jz1);
1407             dx02             = _mm256_sub_ps(ix0,jx2);
1408             dy02             = _mm256_sub_ps(iy0,jy2);
1409             dz02             = _mm256_sub_ps(iz0,jz2);
1410             dx10             = _mm256_sub_ps(ix1,jx0);
1411             dy10             = _mm256_sub_ps(iy1,jy0);
1412             dz10             = _mm256_sub_ps(iz1,jz0);
1413             dx11             = _mm256_sub_ps(ix1,jx1);
1414             dy11             = _mm256_sub_ps(iy1,jy1);
1415             dz11             = _mm256_sub_ps(iz1,jz1);
1416             dx12             = _mm256_sub_ps(ix1,jx2);
1417             dy12             = _mm256_sub_ps(iy1,jy2);
1418             dz12             = _mm256_sub_ps(iz1,jz2);
1419             dx20             = _mm256_sub_ps(ix2,jx0);
1420             dy20             = _mm256_sub_ps(iy2,jy0);
1421             dz20             = _mm256_sub_ps(iz2,jz0);
1422             dx21             = _mm256_sub_ps(ix2,jx1);
1423             dy21             = _mm256_sub_ps(iy2,jy1);
1424             dz21             = _mm256_sub_ps(iz2,jz1);
1425             dx22             = _mm256_sub_ps(ix2,jx2);
1426             dy22             = _mm256_sub_ps(iy2,jy2);
1427             dz22             = _mm256_sub_ps(iz2,jz2);
1428
1429             /* Calculate squared distance and things based on it */
1430             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1431             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1432             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1433             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1434             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1435             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1436             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1437             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1438             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1439
1440             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1441             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1442             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1443             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1444             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1445             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1446             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1447             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1448             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1449
1450             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1451             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1452             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1453             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1454             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1455             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1456             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1457             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1458             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1459
1460             fjx0             = _mm256_setzero_ps();
1461             fjy0             = _mm256_setzero_ps();
1462             fjz0             = _mm256_setzero_ps();
1463             fjx1             = _mm256_setzero_ps();
1464             fjy1             = _mm256_setzero_ps();
1465             fjz1             = _mm256_setzero_ps();
1466             fjx2             = _mm256_setzero_ps();
1467             fjy2             = _mm256_setzero_ps();
1468             fjz2             = _mm256_setzero_ps();
1469
1470             /**************************
1471              * CALCULATE INTERACTIONS *
1472              **************************/
1473
1474             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1475             {
1476
1477             /* REACTION-FIELD ELECTROSTATICS */
1478             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1479
1480             /* LENNARD-JONES DISPERSION/REPULSION */
1481
1482             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1483             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1484
1485             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1486
1487             fscal            = _mm256_add_ps(felec,fvdw);
1488
1489             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1490
1491             /* Calculate temporary vectorial force */
1492             tx               = _mm256_mul_ps(fscal,dx00);
1493             ty               = _mm256_mul_ps(fscal,dy00);
1494             tz               = _mm256_mul_ps(fscal,dz00);
1495
1496             /* Update vectorial force */
1497             fix0             = _mm256_add_ps(fix0,tx);
1498             fiy0             = _mm256_add_ps(fiy0,ty);
1499             fiz0             = _mm256_add_ps(fiz0,tz);
1500
1501             fjx0             = _mm256_add_ps(fjx0,tx);
1502             fjy0             = _mm256_add_ps(fjy0,ty);
1503             fjz0             = _mm256_add_ps(fjz0,tz);
1504
1505             }
1506
1507             /**************************
1508              * CALCULATE INTERACTIONS *
1509              **************************/
1510
1511             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1512             {
1513
1514             /* REACTION-FIELD ELECTROSTATICS */
1515             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1516
1517             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
1518
1519             fscal            = felec;
1520
1521             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1522
1523             /* Calculate temporary vectorial force */
1524             tx               = _mm256_mul_ps(fscal,dx01);
1525             ty               = _mm256_mul_ps(fscal,dy01);
1526             tz               = _mm256_mul_ps(fscal,dz01);
1527
1528             /* Update vectorial force */
1529             fix0             = _mm256_add_ps(fix0,tx);
1530             fiy0             = _mm256_add_ps(fiy0,ty);
1531             fiz0             = _mm256_add_ps(fiz0,tz);
1532
1533             fjx1             = _mm256_add_ps(fjx1,tx);
1534             fjy1             = _mm256_add_ps(fjy1,ty);
1535             fjz1             = _mm256_add_ps(fjz1,tz);
1536
1537             }
1538
1539             /**************************
1540              * CALCULATE INTERACTIONS *
1541              **************************/
1542
1543             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1544             {
1545
1546             /* REACTION-FIELD ELECTROSTATICS */
1547             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1548
1549             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1550
1551             fscal            = felec;
1552
1553             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1554
1555             /* Calculate temporary vectorial force */
1556             tx               = _mm256_mul_ps(fscal,dx02);
1557             ty               = _mm256_mul_ps(fscal,dy02);
1558             tz               = _mm256_mul_ps(fscal,dz02);
1559
1560             /* Update vectorial force */
1561             fix0             = _mm256_add_ps(fix0,tx);
1562             fiy0             = _mm256_add_ps(fiy0,ty);
1563             fiz0             = _mm256_add_ps(fiz0,tz);
1564
1565             fjx2             = _mm256_add_ps(fjx2,tx);
1566             fjy2             = _mm256_add_ps(fjy2,ty);
1567             fjz2             = _mm256_add_ps(fjz2,tz);
1568
1569             }
1570
1571             /**************************
1572              * CALCULATE INTERACTIONS *
1573              **************************/
1574
1575             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1576             {
1577
1578             /* REACTION-FIELD ELECTROSTATICS */
1579             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1580
1581             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1582
1583             fscal            = felec;
1584
1585             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1586
1587             /* Calculate temporary vectorial force */
1588             tx               = _mm256_mul_ps(fscal,dx10);
1589             ty               = _mm256_mul_ps(fscal,dy10);
1590             tz               = _mm256_mul_ps(fscal,dz10);
1591
1592             /* Update vectorial force */
1593             fix1             = _mm256_add_ps(fix1,tx);
1594             fiy1             = _mm256_add_ps(fiy1,ty);
1595             fiz1             = _mm256_add_ps(fiz1,tz);
1596
1597             fjx0             = _mm256_add_ps(fjx0,tx);
1598             fjy0             = _mm256_add_ps(fjy0,ty);
1599             fjz0             = _mm256_add_ps(fjz0,tz);
1600
1601             }
1602
1603             /**************************
1604              * CALCULATE INTERACTIONS *
1605              **************************/
1606
1607             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1608             {
1609
1610             /* REACTION-FIELD ELECTROSTATICS */
1611             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1612
1613             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1614
1615             fscal            = felec;
1616
1617             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1618
1619             /* Calculate temporary vectorial force */
1620             tx               = _mm256_mul_ps(fscal,dx11);
1621             ty               = _mm256_mul_ps(fscal,dy11);
1622             tz               = _mm256_mul_ps(fscal,dz11);
1623
1624             /* Update vectorial force */
1625             fix1             = _mm256_add_ps(fix1,tx);
1626             fiy1             = _mm256_add_ps(fiy1,ty);
1627             fiz1             = _mm256_add_ps(fiz1,tz);
1628
1629             fjx1             = _mm256_add_ps(fjx1,tx);
1630             fjy1             = _mm256_add_ps(fjy1,ty);
1631             fjz1             = _mm256_add_ps(fjz1,tz);
1632
1633             }
1634
1635             /**************************
1636              * CALCULATE INTERACTIONS *
1637              **************************/
1638
1639             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1640             {
1641
1642             /* REACTION-FIELD ELECTROSTATICS */
1643             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1644
1645             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1646
1647             fscal            = felec;
1648
1649             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1650
1651             /* Calculate temporary vectorial force */
1652             tx               = _mm256_mul_ps(fscal,dx12);
1653             ty               = _mm256_mul_ps(fscal,dy12);
1654             tz               = _mm256_mul_ps(fscal,dz12);
1655
1656             /* Update vectorial force */
1657             fix1             = _mm256_add_ps(fix1,tx);
1658             fiy1             = _mm256_add_ps(fiy1,ty);
1659             fiz1             = _mm256_add_ps(fiz1,tz);
1660
1661             fjx2             = _mm256_add_ps(fjx2,tx);
1662             fjy2             = _mm256_add_ps(fjy2,ty);
1663             fjz2             = _mm256_add_ps(fjz2,tz);
1664
1665             }
1666
1667             /**************************
1668              * CALCULATE INTERACTIONS *
1669              **************************/
1670
1671             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1672             {
1673
1674             /* REACTION-FIELD ELECTROSTATICS */
1675             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1676
1677             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1678
1679             fscal            = felec;
1680
1681             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1682
1683             /* Calculate temporary vectorial force */
1684             tx               = _mm256_mul_ps(fscal,dx20);
1685             ty               = _mm256_mul_ps(fscal,dy20);
1686             tz               = _mm256_mul_ps(fscal,dz20);
1687
1688             /* Update vectorial force */
1689             fix2             = _mm256_add_ps(fix2,tx);
1690             fiy2             = _mm256_add_ps(fiy2,ty);
1691             fiz2             = _mm256_add_ps(fiz2,tz);
1692
1693             fjx0             = _mm256_add_ps(fjx0,tx);
1694             fjy0             = _mm256_add_ps(fjy0,ty);
1695             fjz0             = _mm256_add_ps(fjz0,tz);
1696
1697             }
1698
1699             /**************************
1700              * CALCULATE INTERACTIONS *
1701              **************************/
1702
1703             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1704             {
1705
1706             /* REACTION-FIELD ELECTROSTATICS */
1707             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1708
1709             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1710
1711             fscal            = felec;
1712
1713             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1714
1715             /* Calculate temporary vectorial force */
1716             tx               = _mm256_mul_ps(fscal,dx21);
1717             ty               = _mm256_mul_ps(fscal,dy21);
1718             tz               = _mm256_mul_ps(fscal,dz21);
1719
1720             /* Update vectorial force */
1721             fix2             = _mm256_add_ps(fix2,tx);
1722             fiy2             = _mm256_add_ps(fiy2,ty);
1723             fiz2             = _mm256_add_ps(fiz2,tz);
1724
1725             fjx1             = _mm256_add_ps(fjx1,tx);
1726             fjy1             = _mm256_add_ps(fjy1,ty);
1727             fjz1             = _mm256_add_ps(fjz1,tz);
1728
1729             }
1730
1731             /**************************
1732              * CALCULATE INTERACTIONS *
1733              **************************/
1734
1735             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1736             {
1737
1738             /* REACTION-FIELD ELECTROSTATICS */
1739             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1740
1741             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1742
1743             fscal            = felec;
1744
1745             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1746
1747             /* Calculate temporary vectorial force */
1748             tx               = _mm256_mul_ps(fscal,dx22);
1749             ty               = _mm256_mul_ps(fscal,dy22);
1750             tz               = _mm256_mul_ps(fscal,dz22);
1751
1752             /* Update vectorial force */
1753             fix2             = _mm256_add_ps(fix2,tx);
1754             fiy2             = _mm256_add_ps(fiy2,ty);
1755             fiz2             = _mm256_add_ps(fiz2,tz);
1756
1757             fjx2             = _mm256_add_ps(fjx2,tx);
1758             fjy2             = _mm256_add_ps(fjy2,ty);
1759             fjz2             = _mm256_add_ps(fjz2,tz);
1760
1761             }
1762
1763             fjptrA             = f+j_coord_offsetA;
1764             fjptrB             = f+j_coord_offsetB;
1765             fjptrC             = f+j_coord_offsetC;
1766             fjptrD             = f+j_coord_offsetD;
1767             fjptrE             = f+j_coord_offsetE;
1768             fjptrF             = f+j_coord_offsetF;
1769             fjptrG             = f+j_coord_offsetG;
1770             fjptrH             = f+j_coord_offsetH;
1771
1772             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1773                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1774
1775             /* Inner loop uses 277 flops */
1776         }
1777
1778         if(jidx<j_index_end)
1779         {
1780
1781             /* Get j neighbor index, and coordinate index */
1782             jnrlistA         = jjnr[jidx];
1783             jnrlistB         = jjnr[jidx+1];
1784             jnrlistC         = jjnr[jidx+2];
1785             jnrlistD         = jjnr[jidx+3];
1786             jnrlistE         = jjnr[jidx+4];
1787             jnrlistF         = jjnr[jidx+5];
1788             jnrlistG         = jjnr[jidx+6];
1789             jnrlistH         = jjnr[jidx+7];
1790             /* Sign of each element will be negative for non-real atoms.
1791              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1792              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1793              */
1794             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1795                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1796
1797             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1798             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1799             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1800             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1801             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1802             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1803             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1804             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1805             j_coord_offsetA  = DIM*jnrA;
1806             j_coord_offsetB  = DIM*jnrB;
1807             j_coord_offsetC  = DIM*jnrC;
1808             j_coord_offsetD  = DIM*jnrD;
1809             j_coord_offsetE  = DIM*jnrE;
1810             j_coord_offsetF  = DIM*jnrF;
1811             j_coord_offsetG  = DIM*jnrG;
1812             j_coord_offsetH  = DIM*jnrH;
1813
1814             /* load j atom coordinates */
1815             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1816                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1817                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1818                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1819                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1820
1821             /* Calculate displacement vector */
1822             dx00             = _mm256_sub_ps(ix0,jx0);
1823             dy00             = _mm256_sub_ps(iy0,jy0);
1824             dz00             = _mm256_sub_ps(iz0,jz0);
1825             dx01             = _mm256_sub_ps(ix0,jx1);
1826             dy01             = _mm256_sub_ps(iy0,jy1);
1827             dz01             = _mm256_sub_ps(iz0,jz1);
1828             dx02             = _mm256_sub_ps(ix0,jx2);
1829             dy02             = _mm256_sub_ps(iy0,jy2);
1830             dz02             = _mm256_sub_ps(iz0,jz2);
1831             dx10             = _mm256_sub_ps(ix1,jx0);
1832             dy10             = _mm256_sub_ps(iy1,jy0);
1833             dz10             = _mm256_sub_ps(iz1,jz0);
1834             dx11             = _mm256_sub_ps(ix1,jx1);
1835             dy11             = _mm256_sub_ps(iy1,jy1);
1836             dz11             = _mm256_sub_ps(iz1,jz1);
1837             dx12             = _mm256_sub_ps(ix1,jx2);
1838             dy12             = _mm256_sub_ps(iy1,jy2);
1839             dz12             = _mm256_sub_ps(iz1,jz2);
1840             dx20             = _mm256_sub_ps(ix2,jx0);
1841             dy20             = _mm256_sub_ps(iy2,jy0);
1842             dz20             = _mm256_sub_ps(iz2,jz0);
1843             dx21             = _mm256_sub_ps(ix2,jx1);
1844             dy21             = _mm256_sub_ps(iy2,jy1);
1845             dz21             = _mm256_sub_ps(iz2,jz1);
1846             dx22             = _mm256_sub_ps(ix2,jx2);
1847             dy22             = _mm256_sub_ps(iy2,jy2);
1848             dz22             = _mm256_sub_ps(iz2,jz2);
1849
1850             /* Calculate squared distance and things based on it */
1851             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1852             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1853             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1854             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1855             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1856             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1857             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1858             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1859             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1860
1861             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1862             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1863             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1864             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1865             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1866             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1867             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1868             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1869             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1870
1871             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1872             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1873             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1874             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1875             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1876             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1877             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1878             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1879             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1880
1881             fjx0             = _mm256_setzero_ps();
1882             fjy0             = _mm256_setzero_ps();
1883             fjz0             = _mm256_setzero_ps();
1884             fjx1             = _mm256_setzero_ps();
1885             fjy1             = _mm256_setzero_ps();
1886             fjz1             = _mm256_setzero_ps();
1887             fjx2             = _mm256_setzero_ps();
1888             fjy2             = _mm256_setzero_ps();
1889             fjz2             = _mm256_setzero_ps();
1890
1891             /**************************
1892              * CALCULATE INTERACTIONS *
1893              **************************/
1894
1895             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1896             {
1897
1898             /* REACTION-FIELD ELECTROSTATICS */
1899             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1900
1901             /* LENNARD-JONES DISPERSION/REPULSION */
1902
1903             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1904             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1905
1906             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1907
1908             fscal            = _mm256_add_ps(felec,fvdw);
1909
1910             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1911
1912             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1913
1914             /* Calculate temporary vectorial force */
1915             tx               = _mm256_mul_ps(fscal,dx00);
1916             ty               = _mm256_mul_ps(fscal,dy00);
1917             tz               = _mm256_mul_ps(fscal,dz00);
1918
1919             /* Update vectorial force */
1920             fix0             = _mm256_add_ps(fix0,tx);
1921             fiy0             = _mm256_add_ps(fiy0,ty);
1922             fiz0             = _mm256_add_ps(fiz0,tz);
1923
1924             fjx0             = _mm256_add_ps(fjx0,tx);
1925             fjy0             = _mm256_add_ps(fjy0,ty);
1926             fjz0             = _mm256_add_ps(fjz0,tz);
1927
1928             }
1929
1930             /**************************
1931              * CALCULATE INTERACTIONS *
1932              **************************/
1933
1934             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1935             {
1936
1937             /* REACTION-FIELD ELECTROSTATICS */
1938             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1939
1940             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
1941
1942             fscal            = felec;
1943
1944             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1945
1946             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1947
1948             /* Calculate temporary vectorial force */
1949             tx               = _mm256_mul_ps(fscal,dx01);
1950             ty               = _mm256_mul_ps(fscal,dy01);
1951             tz               = _mm256_mul_ps(fscal,dz01);
1952
1953             /* Update vectorial force */
1954             fix0             = _mm256_add_ps(fix0,tx);
1955             fiy0             = _mm256_add_ps(fiy0,ty);
1956             fiz0             = _mm256_add_ps(fiz0,tz);
1957
1958             fjx1             = _mm256_add_ps(fjx1,tx);
1959             fjy1             = _mm256_add_ps(fjy1,ty);
1960             fjz1             = _mm256_add_ps(fjz1,tz);
1961
1962             }
1963
1964             /**************************
1965              * CALCULATE INTERACTIONS *
1966              **************************/
1967
1968             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1969             {
1970
1971             /* REACTION-FIELD ELECTROSTATICS */
1972             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1973
1974             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1975
1976             fscal            = felec;
1977
1978             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1979
1980             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1981
1982             /* Calculate temporary vectorial force */
1983             tx               = _mm256_mul_ps(fscal,dx02);
1984             ty               = _mm256_mul_ps(fscal,dy02);
1985             tz               = _mm256_mul_ps(fscal,dz02);
1986
1987             /* Update vectorial force */
1988             fix0             = _mm256_add_ps(fix0,tx);
1989             fiy0             = _mm256_add_ps(fiy0,ty);
1990             fiz0             = _mm256_add_ps(fiz0,tz);
1991
1992             fjx2             = _mm256_add_ps(fjx2,tx);
1993             fjy2             = _mm256_add_ps(fjy2,ty);
1994             fjz2             = _mm256_add_ps(fjz2,tz);
1995
1996             }
1997
1998             /**************************
1999              * CALCULATE INTERACTIONS *
2000              **************************/
2001
2002             if (gmx_mm256_any_lt(rsq10,rcutoff2))
2003             {
2004
2005             /* REACTION-FIELD ELECTROSTATICS */
2006             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
2007
2008             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
2009
2010             fscal            = felec;
2011
2012             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2013
2014             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2015
2016             /* Calculate temporary vectorial force */
2017             tx               = _mm256_mul_ps(fscal,dx10);
2018             ty               = _mm256_mul_ps(fscal,dy10);
2019             tz               = _mm256_mul_ps(fscal,dz10);
2020
2021             /* Update vectorial force */
2022             fix1             = _mm256_add_ps(fix1,tx);
2023             fiy1             = _mm256_add_ps(fiy1,ty);
2024             fiz1             = _mm256_add_ps(fiz1,tz);
2025
2026             fjx0             = _mm256_add_ps(fjx0,tx);
2027             fjy0             = _mm256_add_ps(fjy0,ty);
2028             fjz0             = _mm256_add_ps(fjz0,tz);
2029
2030             }
2031
2032             /**************************
2033              * CALCULATE INTERACTIONS *
2034              **************************/
2035
2036             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2037             {
2038
2039             /* REACTION-FIELD ELECTROSTATICS */
2040             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
2041
2042             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2043
2044             fscal            = felec;
2045
2046             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2047
2048             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2049
2050             /* Calculate temporary vectorial force */
2051             tx               = _mm256_mul_ps(fscal,dx11);
2052             ty               = _mm256_mul_ps(fscal,dy11);
2053             tz               = _mm256_mul_ps(fscal,dz11);
2054
2055             /* Update vectorial force */
2056             fix1             = _mm256_add_ps(fix1,tx);
2057             fiy1             = _mm256_add_ps(fiy1,ty);
2058             fiz1             = _mm256_add_ps(fiz1,tz);
2059
2060             fjx1             = _mm256_add_ps(fjx1,tx);
2061             fjy1             = _mm256_add_ps(fjy1,ty);
2062             fjz1             = _mm256_add_ps(fjz1,tz);
2063
2064             }
2065
2066             /**************************
2067              * CALCULATE INTERACTIONS *
2068              **************************/
2069
2070             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2071             {
2072
2073             /* REACTION-FIELD ELECTROSTATICS */
2074             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
2075
2076             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2077
2078             fscal            = felec;
2079
2080             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2081
2082             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2083
2084             /* Calculate temporary vectorial force */
2085             tx               = _mm256_mul_ps(fscal,dx12);
2086             ty               = _mm256_mul_ps(fscal,dy12);
2087             tz               = _mm256_mul_ps(fscal,dz12);
2088
2089             /* Update vectorial force */
2090             fix1             = _mm256_add_ps(fix1,tx);
2091             fiy1             = _mm256_add_ps(fiy1,ty);
2092             fiz1             = _mm256_add_ps(fiz1,tz);
2093
2094             fjx2             = _mm256_add_ps(fjx2,tx);
2095             fjy2             = _mm256_add_ps(fjy2,ty);
2096             fjz2             = _mm256_add_ps(fjz2,tz);
2097
2098             }
2099
2100             /**************************
2101              * CALCULATE INTERACTIONS *
2102              **************************/
2103
2104             if (gmx_mm256_any_lt(rsq20,rcutoff2))
2105             {
2106
2107             /* REACTION-FIELD ELECTROSTATICS */
2108             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
2109
2110             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
2111
2112             fscal            = felec;
2113
2114             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2115
2116             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2117
2118             /* Calculate temporary vectorial force */
2119             tx               = _mm256_mul_ps(fscal,dx20);
2120             ty               = _mm256_mul_ps(fscal,dy20);
2121             tz               = _mm256_mul_ps(fscal,dz20);
2122
2123             /* Update vectorial force */
2124             fix2             = _mm256_add_ps(fix2,tx);
2125             fiy2             = _mm256_add_ps(fiy2,ty);
2126             fiz2             = _mm256_add_ps(fiz2,tz);
2127
2128             fjx0             = _mm256_add_ps(fjx0,tx);
2129             fjy0             = _mm256_add_ps(fjy0,ty);
2130             fjz0             = _mm256_add_ps(fjz0,tz);
2131
2132             }
2133
2134             /**************************
2135              * CALCULATE INTERACTIONS *
2136              **************************/
2137
2138             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2139             {
2140
2141             /* REACTION-FIELD ELECTROSTATICS */
2142             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
2143
2144             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2145
2146             fscal            = felec;
2147
2148             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2149
2150             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2151
2152             /* Calculate temporary vectorial force */
2153             tx               = _mm256_mul_ps(fscal,dx21);
2154             ty               = _mm256_mul_ps(fscal,dy21);
2155             tz               = _mm256_mul_ps(fscal,dz21);
2156
2157             /* Update vectorial force */
2158             fix2             = _mm256_add_ps(fix2,tx);
2159             fiy2             = _mm256_add_ps(fiy2,ty);
2160             fiz2             = _mm256_add_ps(fiz2,tz);
2161
2162             fjx1             = _mm256_add_ps(fjx1,tx);
2163             fjy1             = _mm256_add_ps(fjy1,ty);
2164             fjz1             = _mm256_add_ps(fjz1,tz);
2165
2166             }
2167
2168             /**************************
2169              * CALCULATE INTERACTIONS *
2170              **************************/
2171
2172             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2173             {
2174
2175             /* REACTION-FIELD ELECTROSTATICS */
2176             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
2177
2178             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2179
2180             fscal            = felec;
2181
2182             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2183
2184             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2185
2186             /* Calculate temporary vectorial force */
2187             tx               = _mm256_mul_ps(fscal,dx22);
2188             ty               = _mm256_mul_ps(fscal,dy22);
2189             tz               = _mm256_mul_ps(fscal,dz22);
2190
2191             /* Update vectorial force */
2192             fix2             = _mm256_add_ps(fix2,tx);
2193             fiy2             = _mm256_add_ps(fiy2,ty);
2194             fiz2             = _mm256_add_ps(fiz2,tz);
2195
2196             fjx2             = _mm256_add_ps(fjx2,tx);
2197             fjy2             = _mm256_add_ps(fjy2,ty);
2198             fjz2             = _mm256_add_ps(fjz2,tz);
2199
2200             }
2201
2202             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2203             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2204             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2205             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2206             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2207             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2208             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2209             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2210
2211             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2212                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2213
2214             /* Inner loop uses 277 flops */
2215         }
2216
2217         /* End of innermost loop */
2218
2219         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2220                                                  f+i_coord_offset,fshift+i_shift_offset);
2221
2222         /* Increment number of inner iterations */
2223         inneriter                  += j_index_end - j_index_start;
2224
2225         /* Outer loop uses 18 flops */
2226     }
2227
2228     /* Increment number of outer iterations */
2229     outeriter        += nri;
2230
2231     /* Update outer/inner flops */
2232
2233     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*277);
2234 }