src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_VdwCSTab_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/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_ElecRF_VdwCSTab_GeomW3W3_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwCSTab_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     __m128i          vfitab;
 114     __m128i          ifour       = _mm_set1_epi32(4);
 115     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 116     real             *vftab;
 117     __m256d          dummy_mask,cutoff_mask;
 118     __m128           tmpmask0,tmpmask1;
 119     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 120     __m256d          one     = _mm256_set1_pd(1.0);
 121     __m256d          two     = _mm256_set1_pd(2.0);
 122     x                = xx[0];
 123     f                = ff[0];
 124
 125     nri              = nlist->nri;
 126     iinr             = nlist->iinr;
 127     jindex           = nlist->jindex;
 128     jjnr             = nlist->jjnr;
 129     shiftidx         = nlist->shift;
 130     gid              = nlist->gid;
 131     shiftvec         = fr->shift_vec[0];
 132     fshift           = fr->fshift[0];
 133     facel            = _mm256_set1_pd(fr->epsfac);
 134     charge           = mdatoms->chargeA;
 135     krf              = _mm256_set1_pd(fr->ic->k_rf);
 136     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 137     crf              = _mm256_set1_pd(fr->ic->c_rf);
 138     nvdwtype         = fr->ntype;
 139     vdwparam         = fr->nbfp;
 140     vdwtype          = mdatoms->typeA;
 141
 142     vftab            = kernel_data->table_vdw->data;
 143     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 144
 145     /* Setup water-specific parameters */
 146     inr              = nlist->iinr[0];
 147     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 148     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 149     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 150     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 151
 152     jq0              = _mm256_set1_pd(charge[inr+0]);
 153     jq1              = _mm256_set1_pd(charge[inr+1]);
 154     jq2              = _mm256_set1_pd(charge[inr+2]);
 155     vdwjidx0A        = 2*vdwtype[inr+0];
 156     qq00             = _mm256_mul_pd(iq0,jq0);
 157     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 158     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 159     qq01             = _mm256_mul_pd(iq0,jq1);
 160     qq02             = _mm256_mul_pd(iq0,jq2);
 161     qq10             = _mm256_mul_pd(iq1,jq0);
 162     qq11             = _mm256_mul_pd(iq1,jq1);
 163     qq12             = _mm256_mul_pd(iq1,jq2);
 164     qq20             = _mm256_mul_pd(iq2,jq0);
 165     qq21             = _mm256_mul_pd(iq2,jq1);
 166     qq22             = _mm256_mul_pd(iq2,jq2);
 167
 168     /* Avoid stupid compiler warnings */
 169     jnrA = jnrB = jnrC = jnrD = 0;
 170     j_coord_offsetA = 0;
 171     j_coord_offsetB = 0;
 172     j_coord_offsetC = 0;
 173     j_coord_offsetD = 0;
 174
 175     outeriter        = 0;
 176     inneriter        = 0;
 177
 178     for(iidx=0;iidx<4*DIM;iidx++)
 179     {
 180         scratch[iidx] = 0.0;
 181     }
 182
 183     /* Start outer loop over neighborlists */
 184     for(iidx=0; iidx<nri; iidx++)
 185     {
 186         /* Load shift vector for this list */
 187         i_shift_offset   = DIM*shiftidx[iidx];
 188
 189         /* Load limits for loop over neighbors */
 190         j_index_start    = jindex[iidx];
 191         j_index_end      = jindex[iidx+1];
 192
 193         /* Get outer coordinate index */
 194         inr              = iinr[iidx];
 195         i_coord_offset   = DIM*inr;
 196
 197         /* Load i particle coords and add shift vector */
 198         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 199                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 200
 201         fix0             = _mm256_setzero_pd();
 202         fiy0             = _mm256_setzero_pd();
 203         fiz0             = _mm256_setzero_pd();
 204         fix1             = _mm256_setzero_pd();
 205         fiy1             = _mm256_setzero_pd();
 206         fiz1             = _mm256_setzero_pd();
 207         fix2             = _mm256_setzero_pd();
 208         fiy2             = _mm256_setzero_pd();
 209         fiz2             = _mm256_setzero_pd();
 210
 211         /* Reset potential sums */
 212         velecsum         = _mm256_setzero_pd();
 213         vvdwsum          = _mm256_setzero_pd();
 214
 215         /* Start inner kernel loop */
 216         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 217         {
 218
 219             /* Get j neighbor index, and coordinate index */
 220             jnrA             = jjnr[jidx];
 221             jnrB             = jjnr[jidx+1];
 222             jnrC             = jjnr[jidx+2];
 223             jnrD             = jjnr[jidx+3];
 224             j_coord_offsetA  = DIM*jnrA;
 225             j_coord_offsetB  = DIM*jnrB;
 226             j_coord_offsetC  = DIM*jnrC;
 227             j_coord_offsetD  = DIM*jnrD;
 228
 229             /* load j atom coordinates */
 230             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 231                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 232                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 233
 234             /* Calculate displacement vector */
 235             dx00             = _mm256_sub_pd(ix0,jx0);
 236             dy00             = _mm256_sub_pd(iy0,jy0);
 237             dz00             = _mm256_sub_pd(iz0,jz0);
 238             dx01             = _mm256_sub_pd(ix0,jx1);
 239             dy01             = _mm256_sub_pd(iy0,jy1);
 240             dz01             = _mm256_sub_pd(iz0,jz1);
 241             dx02             = _mm256_sub_pd(ix0,jx2);
 242             dy02             = _mm256_sub_pd(iy0,jy2);
 243             dz02             = _mm256_sub_pd(iz0,jz2);
 244             dx10             = _mm256_sub_pd(ix1,jx0);
 245             dy10             = _mm256_sub_pd(iy1,jy0);
 246             dz10             = _mm256_sub_pd(iz1,jz0);
 247             dx11             = _mm256_sub_pd(ix1,jx1);
 248             dy11             = _mm256_sub_pd(iy1,jy1);
 249             dz11             = _mm256_sub_pd(iz1,jz1);
 250             dx12             = _mm256_sub_pd(ix1,jx2);
 251             dy12             = _mm256_sub_pd(iy1,jy2);
 252             dz12             = _mm256_sub_pd(iz1,jz2);
 253             dx20             = _mm256_sub_pd(ix2,jx0);
 254             dy20             = _mm256_sub_pd(iy2,jy0);
 255             dz20             = _mm256_sub_pd(iz2,jz0);
 256             dx21             = _mm256_sub_pd(ix2,jx1);
 257             dy21             = _mm256_sub_pd(iy2,jy1);
 258             dz21             = _mm256_sub_pd(iz2,jz1);
 259             dx22             = _mm256_sub_pd(ix2,jx2);
 260             dy22             = _mm256_sub_pd(iy2,jy2);
 261             dz22             = _mm256_sub_pd(iz2,jz2);
 262
 263             /* Calculate squared distance and things based on it */
 264             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 265             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 266             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 267             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 268             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 269             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 270             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 271             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 272             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 273
 274             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 275             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 276             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 277             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 278             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 279             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 280             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 281             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 282             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 283
 284             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 285             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 286             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 287             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 288             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 289             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 290             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 291             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 292             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 293
 294             fjx0             = _mm256_setzero_pd();
 295             fjy0             = _mm256_setzero_pd();
 296             fjz0             = _mm256_setzero_pd();
 297             fjx1             = _mm256_setzero_pd();
 298             fjy1             = _mm256_setzero_pd();
 299             fjz1             = _mm256_setzero_pd();
 300             fjx2             = _mm256_setzero_pd();
 301             fjy2             = _mm256_setzero_pd();
 302             fjz2             = _mm256_setzero_pd();
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             r00              = _mm256_mul_pd(rsq00,rinv00);
 309
 310             /* Calculate table index by multiplying r with table scale and truncate to integer */
 311             rt               = _mm256_mul_pd(r00,vftabscale);
 312             vfitab           = _mm256_cvttpd_epi32(rt);
 313             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 314             vfitab           = _mm_slli_epi32(vfitab,3);
 315
 316             /* REACTION-FIELD ELECTROSTATICS */
 317             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 318             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 319
 320             /* CUBIC SPLINE TABLE DISPERSION */
 321             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 322             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 323             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 324             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 325             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 326             Heps             = _mm256_mul_pd(vfeps,H);
 327             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 328             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 329             vvdw6            = _mm256_mul_pd(c6_00,VV);
 330             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 331             fvdw6            = _mm256_mul_pd(c6_00,FF);
 332
 333             /* CUBIC SPLINE TABLE REPULSION */
 334             vfitab           = _mm_add_epi32(vfitab,ifour);
 335             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 336             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 337             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 338             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 339             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 340             Heps             = _mm256_mul_pd(vfeps,H);
 341             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 342             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 343             vvdw12           = _mm256_mul_pd(c12_00,VV);
 344             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 345             fvdw12           = _mm256_mul_pd(c12_00,FF);
 346             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 347             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velecsum         = _mm256_add_pd(velecsum,velec);
 351             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 352
 353             fscal            = _mm256_add_pd(felec,fvdw);
 354
 355             /* Calculate temporary vectorial force */
 356             tx               = _mm256_mul_pd(fscal,dx00);
 357             ty               = _mm256_mul_pd(fscal,dy00);
 358             tz               = _mm256_mul_pd(fscal,dz00);
 359
 360             /* Update vectorial force */
 361             fix0             = _mm256_add_pd(fix0,tx);
 362             fiy0             = _mm256_add_pd(fiy0,ty);
 363             fiz0             = _mm256_add_pd(fiz0,tz);
 364
 365             fjx0             = _mm256_add_pd(fjx0,tx);
 366             fjy0             = _mm256_add_pd(fjy0,ty);
 367             fjz0             = _mm256_add_pd(fjz0,tz);
 368
 369             /**************************
 370              * CALCULATE INTERACTIONS *
 371              **************************/
 372
 373             /* REACTION-FIELD ELECTROSTATICS */
 374             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
 375             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velecsum         = _mm256_add_pd(velecsum,velec);
 379
 380             fscal            = felec;
 381
 382             /* Calculate temporary vectorial force */
 383             tx               = _mm256_mul_pd(fscal,dx01);
 384             ty               = _mm256_mul_pd(fscal,dy01);
 385             tz               = _mm256_mul_pd(fscal,dz01);
 386
 387             /* Update vectorial force */
 388             fix0             = _mm256_add_pd(fix0,tx);
 389             fiy0             = _mm256_add_pd(fiy0,ty);
 390             fiz0             = _mm256_add_pd(fiz0,tz);
 391
 392             fjx1             = _mm256_add_pd(fjx1,tx);
 393             fjy1             = _mm256_add_pd(fjy1,ty);
 394             fjz1             = _mm256_add_pd(fjz1,tz);
 395
 396             /**************************
 397              * CALCULATE INTERACTIONS *
 398              **************************/
 399
 400             /* REACTION-FIELD ELECTROSTATICS */
 401             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
 402             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velecsum         = _mm256_add_pd(velecsum,velec);
 406
 407             fscal            = felec;
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm256_mul_pd(fscal,dx02);
 411             ty               = _mm256_mul_pd(fscal,dy02);
 412             tz               = _mm256_mul_pd(fscal,dz02);
 413
 414             /* Update vectorial force */
 415             fix0             = _mm256_add_pd(fix0,tx);
 416             fiy0             = _mm256_add_pd(fiy0,ty);
 417             fiz0             = _mm256_add_pd(fiz0,tz);
 418
 419             fjx2             = _mm256_add_pd(fjx2,tx);
 420             fjy2             = _mm256_add_pd(fjy2,ty);
 421             fjz2             = _mm256_add_pd(fjz2,tz);
 422
 423             /**************************
 424              * CALCULATE INTERACTIONS *
 425              **************************/
 426
 427             /* REACTION-FIELD ELECTROSTATICS */
 428             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 429             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 430
 431             /* Update potential sum for this i atom from the interaction with this j atom. */
 432             velecsum         = _mm256_add_pd(velecsum,velec);
 433
 434             fscal            = felec;
 435
 436             /* Calculate temporary vectorial force */
 437             tx               = _mm256_mul_pd(fscal,dx10);
 438             ty               = _mm256_mul_pd(fscal,dy10);
 439             tz               = _mm256_mul_pd(fscal,dz10);
 440
 441             /* Update vectorial force */
 442             fix1             = _mm256_add_pd(fix1,tx);
 443             fiy1             = _mm256_add_pd(fiy1,ty);
 444             fiz1             = _mm256_add_pd(fiz1,tz);
 445
 446             fjx0             = _mm256_add_pd(fjx0,tx);
 447             fjy0             = _mm256_add_pd(fjy0,ty);
 448             fjz0             = _mm256_add_pd(fjz0,tz);
 449
 450             /**************************
 451              * CALCULATE INTERACTIONS *
 452              **************************/
 453
 454             /* REACTION-FIELD ELECTROSTATICS */
 455             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 456             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velecsum         = _mm256_add_pd(velecsum,velec);
 460
 461             fscal            = felec;
 462
 463             /* Calculate temporary vectorial force */
 464             tx               = _mm256_mul_pd(fscal,dx11);
 465             ty               = _mm256_mul_pd(fscal,dy11);
 466             tz               = _mm256_mul_pd(fscal,dz11);
 467
 468             /* Update vectorial force */
 469             fix1             = _mm256_add_pd(fix1,tx);
 470             fiy1             = _mm256_add_pd(fiy1,ty);
 471             fiz1             = _mm256_add_pd(fiz1,tz);
 472
 473             fjx1             = _mm256_add_pd(fjx1,tx);
 474             fjy1             = _mm256_add_pd(fjy1,ty);
 475             fjz1             = _mm256_add_pd(fjz1,tz);
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             /* REACTION-FIELD ELECTROSTATICS */
 482             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 483             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 484
 485             /* Update potential sum for this i atom from the interaction with this j atom. */
 486             velecsum         = _mm256_add_pd(velecsum,velec);
 487
 488             fscal            = felec;
 489
 490             /* Calculate temporary vectorial force */
 491             tx               = _mm256_mul_pd(fscal,dx12);
 492             ty               = _mm256_mul_pd(fscal,dy12);
 493             tz               = _mm256_mul_pd(fscal,dz12);
 494
 495             /* Update vectorial force */
 496             fix1             = _mm256_add_pd(fix1,tx);
 497             fiy1             = _mm256_add_pd(fiy1,ty);
 498             fiz1             = _mm256_add_pd(fiz1,tz);
 499
 500             fjx2             = _mm256_add_pd(fjx2,tx);
 501             fjy2             = _mm256_add_pd(fjy2,ty);
 502             fjz2             = _mm256_add_pd(fjz2,tz);
 503
 504             /**************************
 505              * CALCULATE INTERACTIONS *
 506              **************************/
 507
 508             /* REACTION-FIELD ELECTROSTATICS */
 509             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 510             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 511
 512             /* Update potential sum for this i atom from the interaction with this j atom. */
 513             velecsum         = _mm256_add_pd(velecsum,velec);
 514
 515             fscal            = felec;
 516
 517             /* Calculate temporary vectorial force */
 518             tx               = _mm256_mul_pd(fscal,dx20);
 519             ty               = _mm256_mul_pd(fscal,dy20);
 520             tz               = _mm256_mul_pd(fscal,dz20);
 521
 522             /* Update vectorial force */
 523             fix2             = _mm256_add_pd(fix2,tx);
 524             fiy2             = _mm256_add_pd(fiy2,ty);
 525             fiz2             = _mm256_add_pd(fiz2,tz);
 526
 527             fjx0             = _mm256_add_pd(fjx0,tx);
 528             fjy0             = _mm256_add_pd(fjy0,ty);
 529             fjz0             = _mm256_add_pd(fjz0,tz);
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             /* REACTION-FIELD ELECTROSTATICS */
 536             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 537             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 538
 539             /* Update potential sum for this i atom from the interaction with this j atom. */
 540             velecsum         = _mm256_add_pd(velecsum,velec);
 541
 542             fscal            = felec;
 543
 544             /* Calculate temporary vectorial force */
 545             tx               = _mm256_mul_pd(fscal,dx21);
 546             ty               = _mm256_mul_pd(fscal,dy21);
 547             tz               = _mm256_mul_pd(fscal,dz21);
 548
 549             /* Update vectorial force */
 550             fix2             = _mm256_add_pd(fix2,tx);
 551             fiy2             = _mm256_add_pd(fiy2,ty);
 552             fiz2             = _mm256_add_pd(fiz2,tz);
 553
 554             fjx1             = _mm256_add_pd(fjx1,tx);
 555             fjy1             = _mm256_add_pd(fjy1,ty);
 556             fjz1             = _mm256_add_pd(fjz1,tz);
 557
 558             /**************************
 559              * CALCULATE INTERACTIONS *
 560              **************************/
 561
 562             /* REACTION-FIELD ELECTROSTATICS */
 563             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 564             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 565
 566             /* Update potential sum for this i atom from the interaction with this j atom. */
 567             velecsum         = _mm256_add_pd(velecsum,velec);
 568
 569             fscal            = felec;
 570
 571             /* Calculate temporary vectorial force */
 572             tx               = _mm256_mul_pd(fscal,dx22);
 573             ty               = _mm256_mul_pd(fscal,dy22);
 574             tz               = _mm256_mul_pd(fscal,dz22);
 575
 576             /* Update vectorial force */
 577             fix2             = _mm256_add_pd(fix2,tx);
 578             fiy2             = _mm256_add_pd(fiy2,ty);
 579             fiz2             = _mm256_add_pd(fiz2,tz);
 580
 581             fjx2             = _mm256_add_pd(fjx2,tx);
 582             fjy2             = _mm256_add_pd(fjy2,ty);
 583             fjz2             = _mm256_add_pd(fjz2,tz);
 584
 585             fjptrA             = f+j_coord_offsetA;
 586             fjptrB             = f+j_coord_offsetB;
 587             fjptrC             = f+j_coord_offsetC;
 588             fjptrD             = f+j_coord_offsetD;
 589
 590             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 591                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 592
 593             /* Inner loop uses 323 flops */
 594         }
 595
 596         if(jidx<j_index_end)
 597         {
 598
 599             /* Get j neighbor index, and coordinate index */
 600             jnrlistA         = jjnr[jidx];
 601             jnrlistB         = jjnr[jidx+1];
 602             jnrlistC         = jjnr[jidx+2];
 603             jnrlistD         = jjnr[jidx+3];
 604             /* Sign of each element will be negative for non-real atoms.
 605              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 606              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 607              */
 608             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 609
 610             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 611             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 612             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 613
 614             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 615             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 616             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 617             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 618             j_coord_offsetA  = DIM*jnrA;
 619             j_coord_offsetB  = DIM*jnrB;
 620             j_coord_offsetC  = DIM*jnrC;
 621             j_coord_offsetD  = DIM*jnrD;
 622
 623             /* load j atom coordinates */
 624             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 625                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 626                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 627
 628             /* Calculate displacement vector */
 629             dx00             = _mm256_sub_pd(ix0,jx0);
 630             dy00             = _mm256_sub_pd(iy0,jy0);
 631             dz00             = _mm256_sub_pd(iz0,jz0);
 632             dx01             = _mm256_sub_pd(ix0,jx1);
 633             dy01             = _mm256_sub_pd(iy0,jy1);
 634             dz01             = _mm256_sub_pd(iz0,jz1);
 635             dx02             = _mm256_sub_pd(ix0,jx2);
 636             dy02             = _mm256_sub_pd(iy0,jy2);
 637             dz02             = _mm256_sub_pd(iz0,jz2);
 638             dx10             = _mm256_sub_pd(ix1,jx0);
 639             dy10             = _mm256_sub_pd(iy1,jy0);
 640             dz10             = _mm256_sub_pd(iz1,jz0);
 641             dx11             = _mm256_sub_pd(ix1,jx1);
 642             dy11             = _mm256_sub_pd(iy1,jy1);
 643             dz11             = _mm256_sub_pd(iz1,jz1);
 644             dx12             = _mm256_sub_pd(ix1,jx2);
 645             dy12             = _mm256_sub_pd(iy1,jy2);
 646             dz12             = _mm256_sub_pd(iz1,jz2);
 647             dx20             = _mm256_sub_pd(ix2,jx0);
 648             dy20             = _mm256_sub_pd(iy2,jy0);
 649             dz20             = _mm256_sub_pd(iz2,jz0);
 650             dx21             = _mm256_sub_pd(ix2,jx1);
 651             dy21             = _mm256_sub_pd(iy2,jy1);
 652             dz21             = _mm256_sub_pd(iz2,jz1);
 653             dx22             = _mm256_sub_pd(ix2,jx2);
 654             dy22             = _mm256_sub_pd(iy2,jy2);
 655             dz22             = _mm256_sub_pd(iz2,jz2);
 656
 657             /* Calculate squared distance and things based on it */
 658             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 659             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 660             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 661             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 662             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 663             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 664             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 665             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 666             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 667
 668             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 669             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 670             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 671             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 672             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 673             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 674             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 675             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 676             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 677
 678             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 679             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 680             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 681             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 682             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 683             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 684             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 685             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 686             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 687
 688             fjx0             = _mm256_setzero_pd();
 689             fjy0             = _mm256_setzero_pd();
 690             fjz0             = _mm256_setzero_pd();
 691             fjx1             = _mm256_setzero_pd();
 692             fjy1             = _mm256_setzero_pd();
 693             fjz1             = _mm256_setzero_pd();
 694             fjx2             = _mm256_setzero_pd();
 695             fjy2             = _mm256_setzero_pd();
 696             fjz2             = _mm256_setzero_pd();
 697
 698             /**************************
 699              * CALCULATE INTERACTIONS *
 700              **************************/
 701
 702             r00              = _mm256_mul_pd(rsq00,rinv00);
 703             r00              = _mm256_andnot_pd(dummy_mask,r00);
 704
 705             /* Calculate table index by multiplying r with table scale and truncate to integer */
 706             rt               = _mm256_mul_pd(r00,vftabscale);
 707             vfitab           = _mm256_cvttpd_epi32(rt);
 708             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 709             vfitab           = _mm_slli_epi32(vfitab,3);
 710
 711             /* REACTION-FIELD ELECTROSTATICS */
 712             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 713             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 714
 715             /* CUBIC SPLINE TABLE DISPERSION */
 716             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 717             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 718             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 719             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 720             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 721             Heps             = _mm256_mul_pd(vfeps,H);
 722             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 723             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 724             vvdw6            = _mm256_mul_pd(c6_00,VV);
 725             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 726             fvdw6            = _mm256_mul_pd(c6_00,FF);
 727
 728             /* CUBIC SPLINE TABLE REPULSION */
 729             vfitab           = _mm_add_epi32(vfitab,ifour);
 730             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 731             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 732             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 733             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 734             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 735             Heps             = _mm256_mul_pd(vfeps,H);
 736             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 737             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 738             vvdw12           = _mm256_mul_pd(c12_00,VV);
 739             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 740             fvdw12           = _mm256_mul_pd(c12_00,FF);
 741             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 742             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 743
 744             /* Update potential sum for this i atom from the interaction with this j atom. */
 745             velec            = _mm256_andnot_pd(dummy_mask,velec);
 746             velecsum         = _mm256_add_pd(velecsum,velec);
 747             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 748             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 749
 750             fscal            = _mm256_add_pd(felec,fvdw);
 751
 752             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 753
 754             /* Calculate temporary vectorial force */
 755             tx               = _mm256_mul_pd(fscal,dx00);
 756             ty               = _mm256_mul_pd(fscal,dy00);
 757             tz               = _mm256_mul_pd(fscal,dz00);
 758
 759             /* Update vectorial force */
 760             fix0             = _mm256_add_pd(fix0,tx);
 761             fiy0             = _mm256_add_pd(fiy0,ty);
 762             fiz0             = _mm256_add_pd(fiz0,tz);
 763
 764             fjx0             = _mm256_add_pd(fjx0,tx);
 765             fjy0             = _mm256_add_pd(fjy0,ty);
 766             fjz0             = _mm256_add_pd(fjz0,tz);
 767
 768             /**************************
 769              * CALCULATE INTERACTIONS *
 770              **************************/
 771
 772             /* REACTION-FIELD ELECTROSTATICS */
 773             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
 774             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
 775
 776             /* Update potential sum for this i atom from the interaction with this j atom. */
 777             velec            = _mm256_andnot_pd(dummy_mask,velec);
 778             velecsum         = _mm256_add_pd(velecsum,velec);
 779
 780             fscal            = felec;
 781
 782             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 783
 784             /* Calculate temporary vectorial force */
 785             tx               = _mm256_mul_pd(fscal,dx01);
 786             ty               = _mm256_mul_pd(fscal,dy01);
 787             tz               = _mm256_mul_pd(fscal,dz01);
 788
 789             /* Update vectorial force */
 790             fix0             = _mm256_add_pd(fix0,tx);
 791             fiy0             = _mm256_add_pd(fiy0,ty);
 792             fiz0             = _mm256_add_pd(fiz0,tz);
 793
 794             fjx1             = _mm256_add_pd(fjx1,tx);
 795             fjy1             = _mm256_add_pd(fjy1,ty);
 796             fjz1             = _mm256_add_pd(fjz1,tz);
 797
 798             /**************************
 799              * CALCULATE INTERACTIONS *
 800              **************************/
 801
 802             /* REACTION-FIELD ELECTROSTATICS */
 803             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
 804             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
 805
 806             /* Update potential sum for this i atom from the interaction with this j atom. */
 807             velec            = _mm256_andnot_pd(dummy_mask,velec);
 808             velecsum         = _mm256_add_pd(velecsum,velec);
 809
 810             fscal            = felec;
 811
 812             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 813
 814             /* Calculate temporary vectorial force */
 815             tx               = _mm256_mul_pd(fscal,dx02);
 816             ty               = _mm256_mul_pd(fscal,dy02);
 817             tz               = _mm256_mul_pd(fscal,dz02);
 818
 819             /* Update vectorial force */
 820             fix0             = _mm256_add_pd(fix0,tx);
 821             fiy0             = _mm256_add_pd(fiy0,ty);
 822             fiz0             = _mm256_add_pd(fiz0,tz);
 823
 824             fjx2             = _mm256_add_pd(fjx2,tx);
 825             fjy2             = _mm256_add_pd(fjy2,ty);
 826             fjz2             = _mm256_add_pd(fjz2,tz);
 827
 828             /**************************
 829              * CALCULATE INTERACTIONS *
 830              **************************/
 831
 832             /* REACTION-FIELD ELECTROSTATICS */
 833             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 834             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 835
 836             /* Update potential sum for this i atom from the interaction with this j atom. */
 837             velec            = _mm256_andnot_pd(dummy_mask,velec);
 838             velecsum         = _mm256_add_pd(velecsum,velec);
 839
 840             fscal            = felec;
 841
 842             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 843
 844             /* Calculate temporary vectorial force */
 845             tx               = _mm256_mul_pd(fscal,dx10);
 846             ty               = _mm256_mul_pd(fscal,dy10);
 847             tz               = _mm256_mul_pd(fscal,dz10);
 848
 849             /* Update vectorial force */
 850             fix1             = _mm256_add_pd(fix1,tx);
 851             fiy1             = _mm256_add_pd(fiy1,ty);
 852             fiz1             = _mm256_add_pd(fiz1,tz);
 853
 854             fjx0             = _mm256_add_pd(fjx0,tx);
 855             fjy0             = _mm256_add_pd(fjy0,ty);
 856             fjz0             = _mm256_add_pd(fjz0,tz);
 857
 858             /**************************
 859              * CALCULATE INTERACTIONS *
 860              **************************/
 861
 862             /* REACTION-FIELD ELECTROSTATICS */
 863             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 864             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 865
 866             /* Update potential sum for this i atom from the interaction with this j atom. */
 867             velec            = _mm256_andnot_pd(dummy_mask,velec);
 868             velecsum         = _mm256_add_pd(velecsum,velec);
 869
 870             fscal            = felec;
 871
 872             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 873
 874             /* Calculate temporary vectorial force */
 875             tx               = _mm256_mul_pd(fscal,dx11);
 876             ty               = _mm256_mul_pd(fscal,dy11);
 877             tz               = _mm256_mul_pd(fscal,dz11);
 878
 879             /* Update vectorial force */
 880             fix1             = _mm256_add_pd(fix1,tx);
 881             fiy1             = _mm256_add_pd(fiy1,ty);
 882             fiz1             = _mm256_add_pd(fiz1,tz);
 883
 884             fjx1             = _mm256_add_pd(fjx1,tx);
 885             fjy1             = _mm256_add_pd(fjy1,ty);
 886             fjz1             = _mm256_add_pd(fjz1,tz);
 887
 888             /**************************
 889              * CALCULATE INTERACTIONS *
 890              **************************/
 891
 892             /* REACTION-FIELD ELECTROSTATICS */
 893             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 894             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 895
 896             /* Update potential sum for this i atom from the interaction with this j atom. */
 897             velec            = _mm256_andnot_pd(dummy_mask,velec);
 898             velecsum         = _mm256_add_pd(velecsum,velec);
 899
 900             fscal            = felec;
 901
 902             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 903
 904             /* Calculate temporary vectorial force */
 905             tx               = _mm256_mul_pd(fscal,dx12);
 906             ty               = _mm256_mul_pd(fscal,dy12);
 907             tz               = _mm256_mul_pd(fscal,dz12);
 908
 909             /* Update vectorial force */
 910             fix1             = _mm256_add_pd(fix1,tx);
 911             fiy1             = _mm256_add_pd(fiy1,ty);
 912             fiz1             = _mm256_add_pd(fiz1,tz);
 913
 914             fjx2             = _mm256_add_pd(fjx2,tx);
 915             fjy2             = _mm256_add_pd(fjy2,ty);
 916             fjz2             = _mm256_add_pd(fjz2,tz);
 917
 918             /**************************
 919              * CALCULATE INTERACTIONS *
 920              **************************/
 921
 922             /* REACTION-FIELD ELECTROSTATICS */
 923             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 924             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 925
 926             /* Update potential sum for this i atom from the interaction with this j atom. */
 927             velec            = _mm256_andnot_pd(dummy_mask,velec);
 928             velecsum         = _mm256_add_pd(velecsum,velec);
 929
 930             fscal            = felec;
 931
 932             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 933
 934             /* Calculate temporary vectorial force */
 935             tx               = _mm256_mul_pd(fscal,dx20);
 936             ty               = _mm256_mul_pd(fscal,dy20);
 937             tz               = _mm256_mul_pd(fscal,dz20);
 938
 939             /* Update vectorial force */
 940             fix2             = _mm256_add_pd(fix2,tx);
 941             fiy2             = _mm256_add_pd(fiy2,ty);
 942             fiz2             = _mm256_add_pd(fiz2,tz);
 943
 944             fjx0             = _mm256_add_pd(fjx0,tx);
 945             fjy0             = _mm256_add_pd(fjy0,ty);
 946             fjz0             = _mm256_add_pd(fjz0,tz);
 947
 948             /**************************
 949              * CALCULATE INTERACTIONS *
 950              **************************/
 951
 952             /* REACTION-FIELD ELECTROSTATICS */
 953             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 954             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 955
 956             /* Update potential sum for this i atom from the interaction with this j atom. */
 957             velec            = _mm256_andnot_pd(dummy_mask,velec);
 958             velecsum         = _mm256_add_pd(velecsum,velec);
 959
 960             fscal            = felec;
 961
 962             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 963
 964             /* Calculate temporary vectorial force */
 965             tx               = _mm256_mul_pd(fscal,dx21);
 966             ty               = _mm256_mul_pd(fscal,dy21);
 967             tz               = _mm256_mul_pd(fscal,dz21);
 968
 969             /* Update vectorial force */
 970             fix2             = _mm256_add_pd(fix2,tx);
 971             fiy2             = _mm256_add_pd(fiy2,ty);
 972             fiz2             = _mm256_add_pd(fiz2,tz);
 973
 974             fjx1             = _mm256_add_pd(fjx1,tx);
 975             fjy1             = _mm256_add_pd(fjy1,ty);
 976             fjz1             = _mm256_add_pd(fjz1,tz);
 977
 978             /**************************
 979              * CALCULATE INTERACTIONS *
 980              **************************/
 981
 982             /* REACTION-FIELD ELECTROSTATICS */
 983             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 984             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 985
 986             /* Update potential sum for this i atom from the interaction with this j atom. */
 987             velec            = _mm256_andnot_pd(dummy_mask,velec);
 988             velecsum         = _mm256_add_pd(velecsum,velec);
 989
 990             fscal            = felec;
 991
 992             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 993
 994             /* Calculate temporary vectorial force */
 995             tx               = _mm256_mul_pd(fscal,dx22);
 996             ty               = _mm256_mul_pd(fscal,dy22);
 997             tz               = _mm256_mul_pd(fscal,dz22);
 998
 999             /* Update vectorial force */
1000             fix2             = _mm256_add_pd(fix2,tx);
1001             fiy2             = _mm256_add_pd(fiy2,ty);
1002             fiz2             = _mm256_add_pd(fiz2,tz);
1003
1004             fjx2             = _mm256_add_pd(fjx2,tx);
1005             fjy2             = _mm256_add_pd(fjy2,ty);
1006             fjz2             = _mm256_add_pd(fjz2,tz);
1007
1008             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1009             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1010             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1011             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1012
1013             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1014                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1015
1016             /* Inner loop uses 324 flops */
1017         }
1018
1019         /* End of innermost loop */
1020
1021         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1022                                                  f+i_coord_offset,fshift+i_shift_offset);
1023
1024         ggid                        = gid[iidx];
1025         /* Update potential energies */
1026         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1027         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1028
1029         /* Increment number of inner iterations */
1030         inneriter                  += j_index_end - j_index_start;
1031
1032         /* Outer loop uses 20 flops */
1033     }
1034
1035     /* Increment number of outer iterations */
1036     outeriter        += nri;
1037
1038     /* Update outer/inner flops */
1039
1040     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*324);
1041 }
1042 /*
1043  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3W3_F_avx_256_double
1044  * Electrostatics interaction: ReactionField
1045  * VdW interaction:            CubicSplineTable
1046  * Geometry:                   Water3-Water3
1047  * Calculate force/pot:        Force
1048  */
1049 void
1050 nb_kernel_ElecRF_VdwCSTab_GeomW3W3_F_avx_256_double
1051                     (t_nblist                    * gmx_restrict       nlist,
1052                      rvec                        * gmx_restrict          xx,
1053                      rvec                        * gmx_restrict          ff,
1054                      t_forcerec                  * gmx_restrict          fr,
1055                      t_mdatoms                   * gmx_restrict     mdatoms,
1056                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1057                      t_nrnb                      * gmx_restrict        nrnb)
1058 {
1059     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1060      * just 0 for non-waters.
1061      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1062      * jnr indices corresponding to data put in the four positions in the SIMD register.
1063      */
1064     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1065     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1066     int              jnrA,jnrB,jnrC,jnrD;
1067     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1068     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1069     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1070     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1071     real             rcutoff_scalar;
1072     real             *shiftvec,*fshift,*x,*f;
1073     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1074     real             scratch[4*DIM];
1075     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1076     real *           vdwioffsetptr0;
1077     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1078     real *           vdwioffsetptr1;
1079     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1080     real *           vdwioffsetptr2;
1081     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1082     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1083     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1084     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1085     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1086     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1087     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1088     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1089     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1090     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1091     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1092     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1093     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1094     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1095     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1096     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1097     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1098     real             *charge;
1099     int              nvdwtype;
1100     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1101     int              *vdwtype;
1102     real             *vdwparam;
1103     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1104     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1105     __m128i          vfitab;
1106     __m128i          ifour       = _mm_set1_epi32(4);
1107     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1108     real             *vftab;
1109     __m256d          dummy_mask,cutoff_mask;
1110     __m128           tmpmask0,tmpmask1;
1111     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1112     __m256d          one     = _mm256_set1_pd(1.0);
1113     __m256d          two     = _mm256_set1_pd(2.0);
1114     x                = xx[0];
1115     f                = ff[0];
1116
1117     nri              = nlist->nri;
1118     iinr             = nlist->iinr;
1119     jindex           = nlist->jindex;
1120     jjnr             = nlist->jjnr;
1121     shiftidx         = nlist->shift;
1122     gid              = nlist->gid;
1123     shiftvec         = fr->shift_vec[0];
1124     fshift           = fr->fshift[0];
1125     facel            = _mm256_set1_pd(fr->epsfac);
1126     charge           = mdatoms->chargeA;
1127     krf              = _mm256_set1_pd(fr->ic->k_rf);
1128     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
1129     crf              = _mm256_set1_pd(fr->ic->c_rf);
1130     nvdwtype         = fr->ntype;
1131     vdwparam         = fr->nbfp;
1132     vdwtype          = mdatoms->typeA;
1133
1134     vftab            = kernel_data->table_vdw->data;
1135     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
1136
1137     /* Setup water-specific parameters */
1138     inr              = nlist->iinr[0];
1139     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1140     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1141     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1142     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1143
1144     jq0              = _mm256_set1_pd(charge[inr+0]);
1145     jq1              = _mm256_set1_pd(charge[inr+1]);
1146     jq2              = _mm256_set1_pd(charge[inr+2]);
1147     vdwjidx0A        = 2*vdwtype[inr+0];
1148     qq00             = _mm256_mul_pd(iq0,jq0);
1149     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1150     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1151     qq01             = _mm256_mul_pd(iq0,jq1);
1152     qq02             = _mm256_mul_pd(iq0,jq2);
1153     qq10             = _mm256_mul_pd(iq1,jq0);
1154     qq11             = _mm256_mul_pd(iq1,jq1);
1155     qq12             = _mm256_mul_pd(iq1,jq2);
1156     qq20             = _mm256_mul_pd(iq2,jq0);
1157     qq21             = _mm256_mul_pd(iq2,jq1);
1158     qq22             = _mm256_mul_pd(iq2,jq2);
1159
1160     /* Avoid stupid compiler warnings */
1161     jnrA = jnrB = jnrC = jnrD = 0;
1162     j_coord_offsetA = 0;
1163     j_coord_offsetB = 0;
1164     j_coord_offsetC = 0;
1165     j_coord_offsetD = 0;
1166
1167     outeriter        = 0;
1168     inneriter        = 0;
1169
1170     for(iidx=0;iidx<4*DIM;iidx++)
1171     {
1172         scratch[iidx] = 0.0;
1173     }
1174
1175     /* Start outer loop over neighborlists */
1176     for(iidx=0; iidx<nri; iidx++)
1177     {
1178         /* Load shift vector for this list */
1179         i_shift_offset   = DIM*shiftidx[iidx];
1180
1181         /* Load limits for loop over neighbors */
1182         j_index_start    = jindex[iidx];
1183         j_index_end      = jindex[iidx+1];
1184
1185         /* Get outer coordinate index */
1186         inr              = iinr[iidx];
1187         i_coord_offset   = DIM*inr;
1188
1189         /* Load i particle coords and add shift vector */
1190         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1191                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1192
1193         fix0             = _mm256_setzero_pd();
1194         fiy0             = _mm256_setzero_pd();
1195         fiz0             = _mm256_setzero_pd();
1196         fix1             = _mm256_setzero_pd();
1197         fiy1             = _mm256_setzero_pd();
1198         fiz1             = _mm256_setzero_pd();
1199         fix2             = _mm256_setzero_pd();
1200         fiy2             = _mm256_setzero_pd();
1201         fiz2             = _mm256_setzero_pd();
1202
1203         /* Start inner kernel loop */
1204         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1205         {
1206
1207             /* Get j neighbor index, and coordinate index */
1208             jnrA             = jjnr[jidx];
1209             jnrB             = jjnr[jidx+1];
1210             jnrC             = jjnr[jidx+2];
1211             jnrD             = jjnr[jidx+3];
1212             j_coord_offsetA  = DIM*jnrA;
1213             j_coord_offsetB  = DIM*jnrB;
1214             j_coord_offsetC  = DIM*jnrC;
1215             j_coord_offsetD  = DIM*jnrD;
1216
1217             /* load j atom coordinates */
1218             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1219                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1220                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1221
1222             /* Calculate displacement vector */
1223             dx00             = _mm256_sub_pd(ix0,jx0);
1224             dy00             = _mm256_sub_pd(iy0,jy0);
1225             dz00             = _mm256_sub_pd(iz0,jz0);
1226             dx01             = _mm256_sub_pd(ix0,jx1);
1227             dy01             = _mm256_sub_pd(iy0,jy1);
1228             dz01             = _mm256_sub_pd(iz0,jz1);
1229             dx02             = _mm256_sub_pd(ix0,jx2);
1230             dy02             = _mm256_sub_pd(iy0,jy2);
1231             dz02             = _mm256_sub_pd(iz0,jz2);
1232             dx10             = _mm256_sub_pd(ix1,jx0);
1233             dy10             = _mm256_sub_pd(iy1,jy0);
1234             dz10             = _mm256_sub_pd(iz1,jz0);
1235             dx11             = _mm256_sub_pd(ix1,jx1);
1236             dy11             = _mm256_sub_pd(iy1,jy1);
1237             dz11             = _mm256_sub_pd(iz1,jz1);
1238             dx12             = _mm256_sub_pd(ix1,jx2);
1239             dy12             = _mm256_sub_pd(iy1,jy2);
1240             dz12             = _mm256_sub_pd(iz1,jz2);
1241             dx20             = _mm256_sub_pd(ix2,jx0);
1242             dy20             = _mm256_sub_pd(iy2,jy0);
1243             dz20             = _mm256_sub_pd(iz2,jz0);
1244             dx21             = _mm256_sub_pd(ix2,jx1);
1245             dy21             = _mm256_sub_pd(iy2,jy1);
1246             dz21             = _mm256_sub_pd(iz2,jz1);
1247             dx22             = _mm256_sub_pd(ix2,jx2);
1248             dy22             = _mm256_sub_pd(iy2,jy2);
1249             dz22             = _mm256_sub_pd(iz2,jz2);
1250
1251             /* Calculate squared distance and things based on it */
1252             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1253             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1254             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1255             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1256             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1257             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1258             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1259             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1260             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1261
1262             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1263             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1264             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1265             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1266             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1267             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1268             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1269             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1270             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1271
1272             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1273             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1274             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1275             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1276             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1277             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1278             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1279             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1280             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1281
1282             fjx0             = _mm256_setzero_pd();
1283             fjy0             = _mm256_setzero_pd();
1284             fjz0             = _mm256_setzero_pd();
1285             fjx1             = _mm256_setzero_pd();
1286             fjy1             = _mm256_setzero_pd();
1287             fjz1             = _mm256_setzero_pd();
1288             fjx2             = _mm256_setzero_pd();
1289             fjy2             = _mm256_setzero_pd();
1290             fjz2             = _mm256_setzero_pd();
1291
1292             /**************************
1293              * CALCULATE INTERACTIONS *
1294              **************************/
1295
1296             r00              = _mm256_mul_pd(rsq00,rinv00);
1297
1298             /* Calculate table index by multiplying r with table scale and truncate to integer */
1299             rt               = _mm256_mul_pd(r00,vftabscale);
1300             vfitab           = _mm256_cvttpd_epi32(rt);
1301             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1302             vfitab           = _mm_slli_epi32(vfitab,3);
1303
1304             /* REACTION-FIELD ELECTROSTATICS */
1305             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1306
1307             /* CUBIC SPLINE TABLE DISPERSION */
1308             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1309             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1310             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1311             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1312             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1313             Heps             = _mm256_mul_pd(vfeps,H);
1314             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1315             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1316             fvdw6            = _mm256_mul_pd(c6_00,FF);
1317
1318             /* CUBIC SPLINE TABLE REPULSION */
1319             vfitab           = _mm_add_epi32(vfitab,ifour);
1320             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1321             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1322             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1323             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1324             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1325             Heps             = _mm256_mul_pd(vfeps,H);
1326             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1327             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1328             fvdw12           = _mm256_mul_pd(c12_00,FF);
1329             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1330
1331             fscal            = _mm256_add_pd(felec,fvdw);
1332
1333             /* Calculate temporary vectorial force */
1334             tx               = _mm256_mul_pd(fscal,dx00);
1335             ty               = _mm256_mul_pd(fscal,dy00);
1336             tz               = _mm256_mul_pd(fscal,dz00);
1337
1338             /* Update vectorial force */
1339             fix0             = _mm256_add_pd(fix0,tx);
1340             fiy0             = _mm256_add_pd(fiy0,ty);
1341             fiz0             = _mm256_add_pd(fiz0,tz);
1342
1343             fjx0             = _mm256_add_pd(fjx0,tx);
1344             fjy0             = _mm256_add_pd(fjy0,ty);
1345             fjz0             = _mm256_add_pd(fjz0,tz);
1346
1347             /**************************
1348              * CALCULATE INTERACTIONS *
1349              **************************/
1350
1351             /* REACTION-FIELD ELECTROSTATICS */
1352             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1353
1354             fscal            = felec;
1355
1356             /* Calculate temporary vectorial force */
1357             tx               = _mm256_mul_pd(fscal,dx01);
1358             ty               = _mm256_mul_pd(fscal,dy01);
1359             tz               = _mm256_mul_pd(fscal,dz01);
1360
1361             /* Update vectorial force */
1362             fix0             = _mm256_add_pd(fix0,tx);
1363             fiy0             = _mm256_add_pd(fiy0,ty);
1364             fiz0             = _mm256_add_pd(fiz0,tz);
1365
1366             fjx1             = _mm256_add_pd(fjx1,tx);
1367             fjy1             = _mm256_add_pd(fjy1,ty);
1368             fjz1             = _mm256_add_pd(fjz1,tz);
1369
1370             /**************************
1371              * CALCULATE INTERACTIONS *
1372              **************************/
1373
1374             /* REACTION-FIELD ELECTROSTATICS */
1375             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1376
1377             fscal            = felec;
1378
1379             /* Calculate temporary vectorial force */
1380             tx               = _mm256_mul_pd(fscal,dx02);
1381             ty               = _mm256_mul_pd(fscal,dy02);
1382             tz               = _mm256_mul_pd(fscal,dz02);
1383
1384             /* Update vectorial force */
1385             fix0             = _mm256_add_pd(fix0,tx);
1386             fiy0             = _mm256_add_pd(fiy0,ty);
1387             fiz0             = _mm256_add_pd(fiz0,tz);
1388
1389             fjx2             = _mm256_add_pd(fjx2,tx);
1390             fjy2             = _mm256_add_pd(fjy2,ty);
1391             fjz2             = _mm256_add_pd(fjz2,tz);
1392
1393             /**************************
1394              * CALCULATE INTERACTIONS *
1395              **************************/
1396
1397             /* REACTION-FIELD ELECTROSTATICS */
1398             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1399
1400             fscal            = felec;
1401
1402             /* Calculate temporary vectorial force */
1403             tx               = _mm256_mul_pd(fscal,dx10);
1404             ty               = _mm256_mul_pd(fscal,dy10);
1405             tz               = _mm256_mul_pd(fscal,dz10);
1406
1407             /* Update vectorial force */
1408             fix1             = _mm256_add_pd(fix1,tx);
1409             fiy1             = _mm256_add_pd(fiy1,ty);
1410             fiz1             = _mm256_add_pd(fiz1,tz);
1411
1412             fjx0             = _mm256_add_pd(fjx0,tx);
1413             fjy0             = _mm256_add_pd(fjy0,ty);
1414             fjz0             = _mm256_add_pd(fjz0,tz);
1415
1416             /**************************
1417              * CALCULATE INTERACTIONS *
1418              **************************/
1419
1420             /* REACTION-FIELD ELECTROSTATICS */
1421             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1422
1423             fscal            = felec;
1424
1425             /* Calculate temporary vectorial force */
1426             tx               = _mm256_mul_pd(fscal,dx11);
1427             ty               = _mm256_mul_pd(fscal,dy11);
1428             tz               = _mm256_mul_pd(fscal,dz11);
1429
1430             /* Update vectorial force */
1431             fix1             = _mm256_add_pd(fix1,tx);
1432             fiy1             = _mm256_add_pd(fiy1,ty);
1433             fiz1             = _mm256_add_pd(fiz1,tz);
1434
1435             fjx1             = _mm256_add_pd(fjx1,tx);
1436             fjy1             = _mm256_add_pd(fjy1,ty);
1437             fjz1             = _mm256_add_pd(fjz1,tz);
1438
1439             /**************************
1440              * CALCULATE INTERACTIONS *
1441              **************************/
1442
1443             /* REACTION-FIELD ELECTROSTATICS */
1444             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1445
1446             fscal            = felec;
1447
1448             /* Calculate temporary vectorial force */
1449             tx               = _mm256_mul_pd(fscal,dx12);
1450             ty               = _mm256_mul_pd(fscal,dy12);
1451             tz               = _mm256_mul_pd(fscal,dz12);
1452
1453             /* Update vectorial force */
1454             fix1             = _mm256_add_pd(fix1,tx);
1455             fiy1             = _mm256_add_pd(fiy1,ty);
1456             fiz1             = _mm256_add_pd(fiz1,tz);
1457
1458             fjx2             = _mm256_add_pd(fjx2,tx);
1459             fjy2             = _mm256_add_pd(fjy2,ty);
1460             fjz2             = _mm256_add_pd(fjz2,tz);
1461
1462             /**************************
1463              * CALCULATE INTERACTIONS *
1464              **************************/
1465
1466             /* REACTION-FIELD ELECTROSTATICS */
1467             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1468
1469             fscal            = felec;
1470
1471             /* Calculate temporary vectorial force */
1472             tx               = _mm256_mul_pd(fscal,dx20);
1473             ty               = _mm256_mul_pd(fscal,dy20);
1474             tz               = _mm256_mul_pd(fscal,dz20);
1475
1476             /* Update vectorial force */
1477             fix2             = _mm256_add_pd(fix2,tx);
1478             fiy2             = _mm256_add_pd(fiy2,ty);
1479             fiz2             = _mm256_add_pd(fiz2,tz);
1480
1481             fjx0             = _mm256_add_pd(fjx0,tx);
1482             fjy0             = _mm256_add_pd(fjy0,ty);
1483             fjz0             = _mm256_add_pd(fjz0,tz);
1484
1485             /**************************
1486              * CALCULATE INTERACTIONS *
1487              **************************/
1488
1489             /* REACTION-FIELD ELECTROSTATICS */
1490             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1491
1492             fscal            = felec;
1493
1494             /* Calculate temporary vectorial force */
1495             tx               = _mm256_mul_pd(fscal,dx21);
1496             ty               = _mm256_mul_pd(fscal,dy21);
1497             tz               = _mm256_mul_pd(fscal,dz21);
1498
1499             /* Update vectorial force */
1500             fix2             = _mm256_add_pd(fix2,tx);
1501             fiy2             = _mm256_add_pd(fiy2,ty);
1502             fiz2             = _mm256_add_pd(fiz2,tz);
1503
1504             fjx1             = _mm256_add_pd(fjx1,tx);
1505             fjy1             = _mm256_add_pd(fjy1,ty);
1506             fjz1             = _mm256_add_pd(fjz1,tz);
1507
1508             /**************************
1509              * CALCULATE INTERACTIONS *
1510              **************************/
1511
1512             /* REACTION-FIELD ELECTROSTATICS */
1513             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1514
1515             fscal            = felec;
1516
1517             /* Calculate temporary vectorial force */
1518             tx               = _mm256_mul_pd(fscal,dx22);
1519             ty               = _mm256_mul_pd(fscal,dy22);
1520             tz               = _mm256_mul_pd(fscal,dz22);
1521
1522             /* Update vectorial force */
1523             fix2             = _mm256_add_pd(fix2,tx);
1524             fiy2             = _mm256_add_pd(fiy2,ty);
1525             fiz2             = _mm256_add_pd(fiz2,tz);
1526
1527             fjx2             = _mm256_add_pd(fjx2,tx);
1528             fjy2             = _mm256_add_pd(fjy2,ty);
1529             fjz2             = _mm256_add_pd(fjz2,tz);
1530
1531             fjptrA             = f+j_coord_offsetA;
1532             fjptrB             = f+j_coord_offsetB;
1533             fjptrC             = f+j_coord_offsetC;
1534             fjptrD             = f+j_coord_offsetD;
1535
1536             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1537                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1538
1539             /* Inner loop uses 270 flops */
1540         }
1541
1542         if(jidx<j_index_end)
1543         {
1544
1545             /* Get j neighbor index, and coordinate index */
1546             jnrlistA         = jjnr[jidx];
1547             jnrlistB         = jjnr[jidx+1];
1548             jnrlistC         = jjnr[jidx+2];
1549             jnrlistD         = jjnr[jidx+3];
1550             /* Sign of each element will be negative for non-real atoms.
1551              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1552              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1553              */
1554             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1555
1556             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1557             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1558             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1559
1560             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1561             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1562             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1563             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1564             j_coord_offsetA  = DIM*jnrA;
1565             j_coord_offsetB  = DIM*jnrB;
1566             j_coord_offsetC  = DIM*jnrC;
1567             j_coord_offsetD  = DIM*jnrD;
1568
1569             /* load j atom coordinates */
1570             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1571                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1572                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1573
1574             /* Calculate displacement vector */
1575             dx00             = _mm256_sub_pd(ix0,jx0);
1576             dy00             = _mm256_sub_pd(iy0,jy0);
1577             dz00             = _mm256_sub_pd(iz0,jz0);
1578             dx01             = _mm256_sub_pd(ix0,jx1);
1579             dy01             = _mm256_sub_pd(iy0,jy1);
1580             dz01             = _mm256_sub_pd(iz0,jz1);
1581             dx02             = _mm256_sub_pd(ix0,jx2);
1582             dy02             = _mm256_sub_pd(iy0,jy2);
1583             dz02             = _mm256_sub_pd(iz0,jz2);
1584             dx10             = _mm256_sub_pd(ix1,jx0);
1585             dy10             = _mm256_sub_pd(iy1,jy0);
1586             dz10             = _mm256_sub_pd(iz1,jz0);
1587             dx11             = _mm256_sub_pd(ix1,jx1);
1588             dy11             = _mm256_sub_pd(iy1,jy1);
1589             dz11             = _mm256_sub_pd(iz1,jz1);
1590             dx12             = _mm256_sub_pd(ix1,jx2);
1591             dy12             = _mm256_sub_pd(iy1,jy2);
1592             dz12             = _mm256_sub_pd(iz1,jz2);
1593             dx20             = _mm256_sub_pd(ix2,jx0);
1594             dy20             = _mm256_sub_pd(iy2,jy0);
1595             dz20             = _mm256_sub_pd(iz2,jz0);
1596             dx21             = _mm256_sub_pd(ix2,jx1);
1597             dy21             = _mm256_sub_pd(iy2,jy1);
1598             dz21             = _mm256_sub_pd(iz2,jz1);
1599             dx22             = _mm256_sub_pd(ix2,jx2);
1600             dy22             = _mm256_sub_pd(iy2,jy2);
1601             dz22             = _mm256_sub_pd(iz2,jz2);
1602
1603             /* Calculate squared distance and things based on it */
1604             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1605             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1606             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1607             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1608             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1609             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1610             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1611             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1612             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1613
1614             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1615             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1616             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1617             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1618             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1619             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1620             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1621             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1622             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1623
1624             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1625             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1626             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1627             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1628             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1629             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1630             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1631             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1632             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1633
1634             fjx0             = _mm256_setzero_pd();
1635             fjy0             = _mm256_setzero_pd();
1636             fjz0             = _mm256_setzero_pd();
1637             fjx1             = _mm256_setzero_pd();
1638             fjy1             = _mm256_setzero_pd();
1639             fjz1             = _mm256_setzero_pd();
1640             fjx2             = _mm256_setzero_pd();
1641             fjy2             = _mm256_setzero_pd();
1642             fjz2             = _mm256_setzero_pd();
1643
1644             /**************************
1645              * CALCULATE INTERACTIONS *
1646              **************************/
1647
1648             r00              = _mm256_mul_pd(rsq00,rinv00);
1649             r00              = _mm256_andnot_pd(dummy_mask,r00);
1650
1651             /* Calculate table index by multiplying r with table scale and truncate to integer */
1652             rt               = _mm256_mul_pd(r00,vftabscale);
1653             vfitab           = _mm256_cvttpd_epi32(rt);
1654             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1655             vfitab           = _mm_slli_epi32(vfitab,3);
1656
1657             /* REACTION-FIELD ELECTROSTATICS */
1658             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1659
1660             /* CUBIC SPLINE TABLE DISPERSION */
1661             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1662             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1663             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1664             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1665             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1666             Heps             = _mm256_mul_pd(vfeps,H);
1667             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1668             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1669             fvdw6            = _mm256_mul_pd(c6_00,FF);
1670
1671             /* CUBIC SPLINE TABLE REPULSION */
1672             vfitab           = _mm_add_epi32(vfitab,ifour);
1673             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1674             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1675             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1676             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1677             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1678             Heps             = _mm256_mul_pd(vfeps,H);
1679             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1680             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1681             fvdw12           = _mm256_mul_pd(c12_00,FF);
1682             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1683
1684             fscal            = _mm256_add_pd(felec,fvdw);
1685
1686             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1687
1688             /* Calculate temporary vectorial force */
1689             tx               = _mm256_mul_pd(fscal,dx00);
1690             ty               = _mm256_mul_pd(fscal,dy00);
1691             tz               = _mm256_mul_pd(fscal,dz00);
1692
1693             /* Update vectorial force */
1694             fix0             = _mm256_add_pd(fix0,tx);
1695             fiy0             = _mm256_add_pd(fiy0,ty);
1696             fiz0             = _mm256_add_pd(fiz0,tz);
1697
1698             fjx0             = _mm256_add_pd(fjx0,tx);
1699             fjy0             = _mm256_add_pd(fjy0,ty);
1700             fjz0             = _mm256_add_pd(fjz0,tz);
1701
1702             /**************************
1703              * CALCULATE INTERACTIONS *
1704              **************************/
1705
1706             /* REACTION-FIELD ELECTROSTATICS */
1707             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1708
1709             fscal            = felec;
1710
1711             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1712
1713             /* Calculate temporary vectorial force */
1714             tx               = _mm256_mul_pd(fscal,dx01);
1715             ty               = _mm256_mul_pd(fscal,dy01);
1716             tz               = _mm256_mul_pd(fscal,dz01);
1717
1718             /* Update vectorial force */
1719             fix0             = _mm256_add_pd(fix0,tx);
1720             fiy0             = _mm256_add_pd(fiy0,ty);
1721             fiz0             = _mm256_add_pd(fiz0,tz);
1722
1723             fjx1             = _mm256_add_pd(fjx1,tx);
1724             fjy1             = _mm256_add_pd(fjy1,ty);
1725             fjz1             = _mm256_add_pd(fjz1,tz);
1726
1727             /**************************
1728              * CALCULATE INTERACTIONS *
1729              **************************/
1730
1731             /* REACTION-FIELD ELECTROSTATICS */
1732             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1733
1734             fscal            = felec;
1735
1736             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1737
1738             /* Calculate temporary vectorial force */
1739             tx               = _mm256_mul_pd(fscal,dx02);
1740             ty               = _mm256_mul_pd(fscal,dy02);
1741             tz               = _mm256_mul_pd(fscal,dz02);
1742
1743             /* Update vectorial force */
1744             fix0             = _mm256_add_pd(fix0,tx);
1745             fiy0             = _mm256_add_pd(fiy0,ty);
1746             fiz0             = _mm256_add_pd(fiz0,tz);
1747
1748             fjx2             = _mm256_add_pd(fjx2,tx);
1749             fjy2             = _mm256_add_pd(fjy2,ty);
1750             fjz2             = _mm256_add_pd(fjz2,tz);
1751
1752             /**************************
1753              * CALCULATE INTERACTIONS *
1754              **************************/
1755
1756             /* REACTION-FIELD ELECTROSTATICS */
1757             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1758
1759             fscal            = felec;
1760
1761             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1762
1763             /* Calculate temporary vectorial force */
1764             tx               = _mm256_mul_pd(fscal,dx10);
1765             ty               = _mm256_mul_pd(fscal,dy10);
1766             tz               = _mm256_mul_pd(fscal,dz10);
1767
1768             /* Update vectorial force */
1769             fix1             = _mm256_add_pd(fix1,tx);
1770             fiy1             = _mm256_add_pd(fiy1,ty);
1771             fiz1             = _mm256_add_pd(fiz1,tz);
1772
1773             fjx0             = _mm256_add_pd(fjx0,tx);
1774             fjy0             = _mm256_add_pd(fjy0,ty);
1775             fjz0             = _mm256_add_pd(fjz0,tz);
1776
1777             /**************************
1778              * CALCULATE INTERACTIONS *
1779              **************************/
1780
1781             /* REACTION-FIELD ELECTROSTATICS */
1782             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1783
1784             fscal            = felec;
1785
1786             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1787
1788             /* Calculate temporary vectorial force */
1789             tx               = _mm256_mul_pd(fscal,dx11);
1790             ty               = _mm256_mul_pd(fscal,dy11);
1791             tz               = _mm256_mul_pd(fscal,dz11);
1792
1793             /* Update vectorial force */
1794             fix1             = _mm256_add_pd(fix1,tx);
1795             fiy1             = _mm256_add_pd(fiy1,ty);
1796             fiz1             = _mm256_add_pd(fiz1,tz);
1797
1798             fjx1             = _mm256_add_pd(fjx1,tx);
1799             fjy1             = _mm256_add_pd(fjy1,ty);
1800             fjz1             = _mm256_add_pd(fjz1,tz);
1801
1802             /**************************
1803              * CALCULATE INTERACTIONS *
1804              **************************/
1805
1806             /* REACTION-FIELD ELECTROSTATICS */
1807             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1808
1809             fscal            = felec;
1810
1811             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1812
1813             /* Calculate temporary vectorial force */
1814             tx               = _mm256_mul_pd(fscal,dx12);
1815             ty               = _mm256_mul_pd(fscal,dy12);
1816             tz               = _mm256_mul_pd(fscal,dz12);
1817
1818             /* Update vectorial force */
1819             fix1             = _mm256_add_pd(fix1,tx);
1820             fiy1             = _mm256_add_pd(fiy1,ty);
1821             fiz1             = _mm256_add_pd(fiz1,tz);
1822
1823             fjx2             = _mm256_add_pd(fjx2,tx);
1824             fjy2             = _mm256_add_pd(fjy2,ty);
1825             fjz2             = _mm256_add_pd(fjz2,tz);
1826
1827             /**************************
1828              * CALCULATE INTERACTIONS *
1829              **************************/
1830
1831             /* REACTION-FIELD ELECTROSTATICS */
1832             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1833
1834             fscal            = felec;
1835
1836             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1837
1838             /* Calculate temporary vectorial force */
1839             tx               = _mm256_mul_pd(fscal,dx20);
1840             ty               = _mm256_mul_pd(fscal,dy20);
1841             tz               = _mm256_mul_pd(fscal,dz20);
1842
1843             /* Update vectorial force */
1844             fix2             = _mm256_add_pd(fix2,tx);
1845             fiy2             = _mm256_add_pd(fiy2,ty);
1846             fiz2             = _mm256_add_pd(fiz2,tz);
1847
1848             fjx0             = _mm256_add_pd(fjx0,tx);
1849             fjy0             = _mm256_add_pd(fjy0,ty);
1850             fjz0             = _mm256_add_pd(fjz0,tz);
1851
1852             /**************************
1853              * CALCULATE INTERACTIONS *
1854              **************************/
1855
1856             /* REACTION-FIELD ELECTROSTATICS */
1857             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1858
1859             fscal            = felec;
1860
1861             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1862
1863             /* Calculate temporary vectorial force */
1864             tx               = _mm256_mul_pd(fscal,dx21);
1865             ty               = _mm256_mul_pd(fscal,dy21);
1866             tz               = _mm256_mul_pd(fscal,dz21);
1867
1868             /* Update vectorial force */
1869             fix2             = _mm256_add_pd(fix2,tx);
1870             fiy2             = _mm256_add_pd(fiy2,ty);
1871             fiz2             = _mm256_add_pd(fiz2,tz);
1872
1873             fjx1             = _mm256_add_pd(fjx1,tx);
1874             fjy1             = _mm256_add_pd(fjy1,ty);
1875             fjz1             = _mm256_add_pd(fjz1,tz);
1876
1877             /**************************
1878              * CALCULATE INTERACTIONS *
1879              **************************/
1880
1881             /* REACTION-FIELD ELECTROSTATICS */
1882             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1883
1884             fscal            = felec;
1885
1886             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1887
1888             /* Calculate temporary vectorial force */
1889             tx               = _mm256_mul_pd(fscal,dx22);
1890             ty               = _mm256_mul_pd(fscal,dy22);
1891             tz               = _mm256_mul_pd(fscal,dz22);
1892
1893             /* Update vectorial force */
1894             fix2             = _mm256_add_pd(fix2,tx);
1895             fiy2             = _mm256_add_pd(fiy2,ty);
1896             fiz2             = _mm256_add_pd(fiz2,tz);
1897
1898             fjx2             = _mm256_add_pd(fjx2,tx);
1899             fjy2             = _mm256_add_pd(fjy2,ty);
1900             fjz2             = _mm256_add_pd(fjz2,tz);
1901
1902             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1903             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1904             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1905             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1906
1907             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1908                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1909
1910             /* Inner loop uses 271 flops */
1911         }
1912
1913         /* End of innermost loop */
1914
1915         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1916                                                  f+i_coord_offset,fshift+i_shift_offset);
1917
1918         /* Increment number of inner iterations */
1919         inneriter                  += j_index_end - j_index_start;
1920
1921         /* Outer loop uses 18 flops */
1922     }
1923
1924     /* Increment number of outer iterations */
1925     outeriter        += nri;
1926
1927     /* Update outer/inner flops */
1928
1929     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*271);
1930 }