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

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