src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwLJ_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3W3_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_GeomW3W3_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  95     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  97     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 100     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 101     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 102     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 105     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     int              nvdwtype;
 110     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 111     int              *vdwtype;
 112     real             *vdwparam;
 113     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 114     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 115     __m128i          ewitab;
 116     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 117     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 118     real             *ewtab;
 119     __m256d          dummy_mask,cutoff_mask;
 120     __m128           tmpmask0,tmpmask1;
 121     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 122     __m256d          one     = _mm256_set1_pd(1.0);
 123     __m256d          two     = _mm256_set1_pd(2.0);
 124     x                = xx[0];
 125     f                = ff[0];
 126
 127     nri              = nlist->nri;
 128     iinr             = nlist->iinr;
 129     jindex           = nlist->jindex;
 130     jjnr             = nlist->jjnr;
 131     shiftidx         = nlist->shift;
 132     gid              = nlist->gid;
 133     shiftvec         = fr->shift_vec[0];
 134     fshift           = fr->fshift[0];
 135     facel            = _mm256_set1_pd(fr->epsfac);
 136     charge           = mdatoms->chargeA;
 137     nvdwtype         = fr->ntype;
 138     vdwparam         = fr->nbfp;
 139     vdwtype          = mdatoms->typeA;
 140
 141     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 142     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 143     beta2            = _mm256_mul_pd(beta,beta);
 144     beta3            = _mm256_mul_pd(beta,beta2);
 145
 146     ewtab            = fr->ic->tabq_coul_FDV0;
 147     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 148     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 149
 150     /* Setup water-specific parameters */
 151     inr              = nlist->iinr[0];
 152     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 153     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 154     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 155     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 156
 157     jq0              = _mm256_set1_pd(charge[inr+0]);
 158     jq1              = _mm256_set1_pd(charge[inr+1]);
 159     jq2              = _mm256_set1_pd(charge[inr+2]);
 160     vdwjidx0A        = 2*vdwtype[inr+0];
 161     qq00             = _mm256_mul_pd(iq0,jq0);
 162     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 163     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 164     qq01             = _mm256_mul_pd(iq0,jq1);
 165     qq02             = _mm256_mul_pd(iq0,jq2);
 166     qq10             = _mm256_mul_pd(iq1,jq0);
 167     qq11             = _mm256_mul_pd(iq1,jq1);
 168     qq12             = _mm256_mul_pd(iq1,jq2);
 169     qq20             = _mm256_mul_pd(iq2,jq0);
 170     qq21             = _mm256_mul_pd(iq2,jq1);
 171     qq22             = _mm256_mul_pd(iq2,jq2);
 172
 173     /* Avoid stupid compiler warnings */
 174     jnrA = jnrB = jnrC = jnrD = 0;
 175     j_coord_offsetA = 0;
 176     j_coord_offsetB = 0;
 177     j_coord_offsetC = 0;
 178     j_coord_offsetD = 0;
 179
 180     outeriter        = 0;
 181     inneriter        = 0;
 182
 183     for(iidx=0;iidx<4*DIM;iidx++)
 184     {
 185         scratch[iidx] = 0.0;
 186     }
 187
 188     /* Start outer loop over neighborlists */
 189     for(iidx=0; iidx<nri; iidx++)
 190     {
 191         /* Load shift vector for this list */
 192         i_shift_offset   = DIM*shiftidx[iidx];
 193
 194         /* Load limits for loop over neighbors */
 195         j_index_start    = jindex[iidx];
 196         j_index_end      = jindex[iidx+1];
 197
 198         /* Get outer coordinate index */
 199         inr              = iinr[iidx];
 200         i_coord_offset   = DIM*inr;
 201
 202         /* Load i particle coords and add shift vector */
 203         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 204                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 205
 206         fix0             = _mm256_setzero_pd();
 207         fiy0             = _mm256_setzero_pd();
 208         fiz0             = _mm256_setzero_pd();
 209         fix1             = _mm256_setzero_pd();
 210         fiy1             = _mm256_setzero_pd();
 211         fiz1             = _mm256_setzero_pd();
 212         fix2             = _mm256_setzero_pd();
 213         fiy2             = _mm256_setzero_pd();
 214         fiz2             = _mm256_setzero_pd();
 215
 216         /* Reset potential sums */
 217         velecsum         = _mm256_setzero_pd();
 218         vvdwsum          = _mm256_setzero_pd();
 219
 220         /* Start inner kernel loop */
 221         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 222         {
 223
 224             /* Get j neighbor index, and coordinate index */
 225             jnrA             = jjnr[jidx];
 226             jnrB             = jjnr[jidx+1];
 227             jnrC             = jjnr[jidx+2];
 228             jnrD             = jjnr[jidx+3];
 229             j_coord_offsetA  = DIM*jnrA;
 230             j_coord_offsetB  = DIM*jnrB;
 231             j_coord_offsetC  = DIM*jnrC;
 232             j_coord_offsetD  = DIM*jnrD;
 233
 234             /* load j atom coordinates */
 235             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 236                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 237                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 238
 239             /* Calculate displacement vector */
 240             dx00             = _mm256_sub_pd(ix0,jx0);
 241             dy00             = _mm256_sub_pd(iy0,jy0);
 242             dz00             = _mm256_sub_pd(iz0,jz0);
 243             dx01             = _mm256_sub_pd(ix0,jx1);
 244             dy01             = _mm256_sub_pd(iy0,jy1);
 245             dz01             = _mm256_sub_pd(iz0,jz1);
 246             dx02             = _mm256_sub_pd(ix0,jx2);
 247             dy02             = _mm256_sub_pd(iy0,jy2);
 248             dz02             = _mm256_sub_pd(iz0,jz2);
 249             dx10             = _mm256_sub_pd(ix1,jx0);
 250             dy10             = _mm256_sub_pd(iy1,jy0);
 251             dz10             = _mm256_sub_pd(iz1,jz0);
 252             dx11             = _mm256_sub_pd(ix1,jx1);
 253             dy11             = _mm256_sub_pd(iy1,jy1);
 254             dz11             = _mm256_sub_pd(iz1,jz1);
 255             dx12             = _mm256_sub_pd(ix1,jx2);
 256             dy12             = _mm256_sub_pd(iy1,jy2);
 257             dz12             = _mm256_sub_pd(iz1,jz2);
 258             dx20             = _mm256_sub_pd(ix2,jx0);
 259             dy20             = _mm256_sub_pd(iy2,jy0);
 260             dz20             = _mm256_sub_pd(iz2,jz0);
 261             dx21             = _mm256_sub_pd(ix2,jx1);
 262             dy21             = _mm256_sub_pd(iy2,jy1);
 263             dz21             = _mm256_sub_pd(iz2,jz1);
 264             dx22             = _mm256_sub_pd(ix2,jx2);
 265             dy22             = _mm256_sub_pd(iy2,jy2);
 266             dz22             = _mm256_sub_pd(iz2,jz2);
 267
 268             /* Calculate squared distance and things based on it */
 269             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 270             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 271             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 272             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 273             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 274             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 275             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 276             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 277             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 278
 279             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 280             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 281             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 282             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 283             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 284             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 285             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 286             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 287             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 288
 289             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 290             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 291             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 292             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 293             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 294             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 295             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 296             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 297             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 298
 299             fjx0             = _mm256_setzero_pd();
 300             fjy0             = _mm256_setzero_pd();
 301             fjz0             = _mm256_setzero_pd();
 302             fjx1             = _mm256_setzero_pd();
 303             fjy1             = _mm256_setzero_pd();
 304             fjz1             = _mm256_setzero_pd();
 305             fjx2             = _mm256_setzero_pd();
 306             fjy2             = _mm256_setzero_pd();
 307             fjz2             = _mm256_setzero_pd();
 308
 309             /**************************
 310              * CALCULATE INTERACTIONS *
 311              **************************/
 312
 313             r00              = _mm256_mul_pd(rsq00,rinv00);
 314
 315             /* EWALD ELECTROSTATICS */
 316
 317             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 318             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 319             ewitab           = _mm256_cvttpd_epi32(ewrt);
 320             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 321             ewitab           = _mm_slli_epi32(ewitab,2);
 322             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 323             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 324             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 325             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 326             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 327             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 328             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 329             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 330             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 331
 332             /* LENNARD-JONES DISPERSION/REPULSION */
 333
 334             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 335             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 336             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 337             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 338             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velecsum         = _mm256_add_pd(velecsum,velec);
 342             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 343
 344             fscal            = _mm256_add_pd(felec,fvdw);
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm256_mul_pd(fscal,dx00);
 348             ty               = _mm256_mul_pd(fscal,dy00);
 349             tz               = _mm256_mul_pd(fscal,dz00);
 350
 351             /* Update vectorial force */
 352             fix0             = _mm256_add_pd(fix0,tx);
 353             fiy0             = _mm256_add_pd(fiy0,ty);
 354             fiz0             = _mm256_add_pd(fiz0,tz);
 355
 356             fjx0             = _mm256_add_pd(fjx0,tx);
 357             fjy0             = _mm256_add_pd(fjy0,ty);
 358             fjz0             = _mm256_add_pd(fjz0,tz);
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             r01              = _mm256_mul_pd(rsq01,rinv01);
 365
 366             /* EWALD ELECTROSTATICS */
 367
 368             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 369             ewrt             = _mm256_mul_pd(r01,ewtabscale);
 370             ewitab           = _mm256_cvttpd_epi32(ewrt);
 371             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 372             ewitab           = _mm_slli_epi32(ewitab,2);
 373             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 374             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 375             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 376             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 377             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 378             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 379             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 380             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(rinv01,velec));
 381             felec            = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
 382
 383             /* Update potential sum for this i atom from the interaction with this j atom. */
 384             velecsum         = _mm256_add_pd(velecsum,velec);
 385
 386             fscal            = felec;
 387
 388             /* Calculate temporary vectorial force */
 389             tx               = _mm256_mul_pd(fscal,dx01);
 390             ty               = _mm256_mul_pd(fscal,dy01);
 391             tz               = _mm256_mul_pd(fscal,dz01);
 392
 393             /* Update vectorial force */
 394             fix0             = _mm256_add_pd(fix0,tx);
 395             fiy0             = _mm256_add_pd(fiy0,ty);
 396             fiz0             = _mm256_add_pd(fiz0,tz);
 397
 398             fjx1             = _mm256_add_pd(fjx1,tx);
 399             fjy1             = _mm256_add_pd(fjy1,ty);
 400             fjz1             = _mm256_add_pd(fjz1,tz);
 401
 402             /**************************
 403              * CALCULATE INTERACTIONS *
 404              **************************/
 405
 406             r02              = _mm256_mul_pd(rsq02,rinv02);
 407
 408             /* EWALD ELECTROSTATICS */
 409
 410             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 411             ewrt             = _mm256_mul_pd(r02,ewtabscale);
 412             ewitab           = _mm256_cvttpd_epi32(ewrt);
 413             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 414             ewitab           = _mm_slli_epi32(ewitab,2);
 415             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 416             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 417             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 418             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 419             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 420             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 421             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 422             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(rinv02,velec));
 423             felec            = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
 424
 425             /* Update potential sum for this i atom from the interaction with this j atom. */
 426             velecsum         = _mm256_add_pd(velecsum,velec);
 427
 428             fscal            = felec;
 429
 430             /* Calculate temporary vectorial force */
 431             tx               = _mm256_mul_pd(fscal,dx02);
 432             ty               = _mm256_mul_pd(fscal,dy02);
 433             tz               = _mm256_mul_pd(fscal,dz02);
 434
 435             /* Update vectorial force */
 436             fix0             = _mm256_add_pd(fix0,tx);
 437             fiy0             = _mm256_add_pd(fiy0,ty);
 438             fiz0             = _mm256_add_pd(fiz0,tz);
 439
 440             fjx2             = _mm256_add_pd(fjx2,tx);
 441             fjy2             = _mm256_add_pd(fjy2,ty);
 442             fjz2             = _mm256_add_pd(fjz2,tz);
 443
 444             /**************************
 445              * CALCULATE INTERACTIONS *
 446              **************************/
 447
 448             r10              = _mm256_mul_pd(rsq10,rinv10);
 449
 450             /* EWALD ELECTROSTATICS */
 451
 452             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 453             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 454             ewitab           = _mm256_cvttpd_epi32(ewrt);
 455             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 456             ewitab           = _mm_slli_epi32(ewitab,2);
 457             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 458             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 459             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 460             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 461             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 462             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 463             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 464             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 465             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 466
 467             /* Update potential sum for this i atom from the interaction with this j atom. */
 468             velecsum         = _mm256_add_pd(velecsum,velec);
 469
 470             fscal            = felec;
 471
 472             /* Calculate temporary vectorial force */
 473             tx               = _mm256_mul_pd(fscal,dx10);
 474             ty               = _mm256_mul_pd(fscal,dy10);
 475             tz               = _mm256_mul_pd(fscal,dz10);
 476
 477             /* Update vectorial force */
 478             fix1             = _mm256_add_pd(fix1,tx);
 479             fiy1             = _mm256_add_pd(fiy1,ty);
 480             fiz1             = _mm256_add_pd(fiz1,tz);
 481
 482             fjx0             = _mm256_add_pd(fjx0,tx);
 483             fjy0             = _mm256_add_pd(fjy0,ty);
 484             fjz0             = _mm256_add_pd(fjz0,tz);
 485
 486             /**************************
 487              * CALCULATE INTERACTIONS *
 488              **************************/
 489
 490             r11              = _mm256_mul_pd(rsq11,rinv11);
 491
 492             /* EWALD ELECTROSTATICS */
 493
 494             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 495             ewrt             = _mm256_mul_pd(r11,ewtabscale);
 496             ewitab           = _mm256_cvttpd_epi32(ewrt);
 497             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 498             ewitab           = _mm_slli_epi32(ewitab,2);
 499             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 500             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 501             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 502             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 503             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 504             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 505             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 506             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
 507             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
 508
 509             /* Update potential sum for this i atom from the interaction with this j atom. */
 510             velecsum         = _mm256_add_pd(velecsum,velec);
 511
 512             fscal            = felec;
 513
 514             /* Calculate temporary vectorial force */
 515             tx               = _mm256_mul_pd(fscal,dx11);
 516             ty               = _mm256_mul_pd(fscal,dy11);
 517             tz               = _mm256_mul_pd(fscal,dz11);
 518
 519             /* Update vectorial force */
 520             fix1             = _mm256_add_pd(fix1,tx);
 521             fiy1             = _mm256_add_pd(fiy1,ty);
 522             fiz1             = _mm256_add_pd(fiz1,tz);
 523
 524             fjx1             = _mm256_add_pd(fjx1,tx);
 525             fjy1             = _mm256_add_pd(fjy1,ty);
 526             fjz1             = _mm256_add_pd(fjz1,tz);
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             r12              = _mm256_mul_pd(rsq12,rinv12);
 533
 534             /* EWALD ELECTROSTATICS */
 535
 536             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 537             ewrt             = _mm256_mul_pd(r12,ewtabscale);
 538             ewitab           = _mm256_cvttpd_epi32(ewrt);
 539             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 540             ewitab           = _mm_slli_epi32(ewitab,2);
 541             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 542             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 543             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 544             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 545             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 546             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 547             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 548             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
 549             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             velecsum         = _mm256_add_pd(velecsum,velec);
 553
 554             fscal            = felec;
 555
 556             /* Calculate temporary vectorial force */
 557             tx               = _mm256_mul_pd(fscal,dx12);
 558             ty               = _mm256_mul_pd(fscal,dy12);
 559             tz               = _mm256_mul_pd(fscal,dz12);
 560
 561             /* Update vectorial force */
 562             fix1             = _mm256_add_pd(fix1,tx);
 563             fiy1             = _mm256_add_pd(fiy1,ty);
 564             fiz1             = _mm256_add_pd(fiz1,tz);
 565
 566             fjx2             = _mm256_add_pd(fjx2,tx);
 567             fjy2             = _mm256_add_pd(fjy2,ty);
 568             fjz2             = _mm256_add_pd(fjz2,tz);
 569
 570             /**************************
 571              * CALCULATE INTERACTIONS *
 572              **************************/
 573
 574             r20              = _mm256_mul_pd(rsq20,rinv20);
 575
 576             /* EWALD ELECTROSTATICS */
 577
 578             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 579             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 580             ewitab           = _mm256_cvttpd_epi32(ewrt);
 581             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 582             ewitab           = _mm_slli_epi32(ewitab,2);
 583             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 584             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 585             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 586             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 587             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 588             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 589             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 590             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 591             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velecsum         = _mm256_add_pd(velecsum,velec);
 595
 596             fscal            = felec;
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm256_mul_pd(fscal,dx20);
 600             ty               = _mm256_mul_pd(fscal,dy20);
 601             tz               = _mm256_mul_pd(fscal,dz20);
 602
 603             /* Update vectorial force */
 604             fix2             = _mm256_add_pd(fix2,tx);
 605             fiy2             = _mm256_add_pd(fiy2,ty);
 606             fiz2             = _mm256_add_pd(fiz2,tz);
 607
 608             fjx0             = _mm256_add_pd(fjx0,tx);
 609             fjy0             = _mm256_add_pd(fjy0,ty);
 610             fjz0             = _mm256_add_pd(fjz0,tz);
 611
 612             /**************************
 613              * CALCULATE INTERACTIONS *
 614              **************************/
 615
 616             r21              = _mm256_mul_pd(rsq21,rinv21);
 617
 618             /* EWALD ELECTROSTATICS */
 619
 620             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 621             ewrt             = _mm256_mul_pd(r21,ewtabscale);
 622             ewitab           = _mm256_cvttpd_epi32(ewrt);
 623             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 624             ewitab           = _mm_slli_epi32(ewitab,2);
 625             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 626             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 627             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 628             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 629             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 630             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 631             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 632             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
 633             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
 634
 635             /* Update potential sum for this i atom from the interaction with this j atom. */
 636             velecsum         = _mm256_add_pd(velecsum,velec);
 637
 638             fscal            = felec;
 639
 640             /* Calculate temporary vectorial force */
 641             tx               = _mm256_mul_pd(fscal,dx21);
 642             ty               = _mm256_mul_pd(fscal,dy21);
 643             tz               = _mm256_mul_pd(fscal,dz21);
 644
 645             /* Update vectorial force */
 646             fix2             = _mm256_add_pd(fix2,tx);
 647             fiy2             = _mm256_add_pd(fiy2,ty);
 648             fiz2             = _mm256_add_pd(fiz2,tz);
 649
 650             fjx1             = _mm256_add_pd(fjx1,tx);
 651             fjy1             = _mm256_add_pd(fjy1,ty);
 652             fjz1             = _mm256_add_pd(fjz1,tz);
 653
 654             /**************************
 655              * CALCULATE INTERACTIONS *
 656              **************************/
 657
 658             r22              = _mm256_mul_pd(rsq22,rinv22);
 659
 660             /* EWALD ELECTROSTATICS */
 661
 662             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 663             ewrt             = _mm256_mul_pd(r22,ewtabscale);
 664             ewitab           = _mm256_cvttpd_epi32(ewrt);
 665             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 666             ewitab           = _mm_slli_epi32(ewitab,2);
 667             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 668             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 669             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 670             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 671             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 672             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 673             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 674             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
 675             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
 676
 677             /* Update potential sum for this i atom from the interaction with this j atom. */
 678             velecsum         = _mm256_add_pd(velecsum,velec);
 679
 680             fscal            = felec;
 681
 682             /* Calculate temporary vectorial force */
 683             tx               = _mm256_mul_pd(fscal,dx22);
 684             ty               = _mm256_mul_pd(fscal,dy22);
 685             tz               = _mm256_mul_pd(fscal,dz22);
 686
 687             /* Update vectorial force */
 688             fix2             = _mm256_add_pd(fix2,tx);
 689             fiy2             = _mm256_add_pd(fiy2,ty);
 690             fiz2             = _mm256_add_pd(fiz2,tz);
 691
 692             fjx2             = _mm256_add_pd(fjx2,tx);
 693             fjy2             = _mm256_add_pd(fjy2,ty);
 694             fjz2             = _mm256_add_pd(fjz2,tz);
 695
 696             fjptrA             = f+j_coord_offsetA;
 697             fjptrB             = f+j_coord_offsetB;
 698             fjptrC             = f+j_coord_offsetC;
 699             fjptrD             = f+j_coord_offsetD;
 700
 701             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 702                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 703
 704             /* Inner loop uses 381 flops */
 705         }
 706
 707         if(jidx<j_index_end)
 708         {
 709
 710             /* Get j neighbor index, and coordinate index */
 711             jnrlistA         = jjnr[jidx];
 712             jnrlistB         = jjnr[jidx+1];
 713             jnrlistC         = jjnr[jidx+2];
 714             jnrlistD         = jjnr[jidx+3];
 715             /* Sign of each element will be negative for non-real atoms.
 716              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 717              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 718              */
 719             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 720
 721             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 722             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 723             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 724
 725             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 726             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 727             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 728             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 729             j_coord_offsetA  = DIM*jnrA;
 730             j_coord_offsetB  = DIM*jnrB;
 731             j_coord_offsetC  = DIM*jnrC;
 732             j_coord_offsetD  = DIM*jnrD;
 733
 734             /* load j atom coordinates */
 735             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 736                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 737                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 738
 739             /* Calculate displacement vector */
 740             dx00             = _mm256_sub_pd(ix0,jx0);
 741             dy00             = _mm256_sub_pd(iy0,jy0);
 742             dz00             = _mm256_sub_pd(iz0,jz0);
 743             dx01             = _mm256_sub_pd(ix0,jx1);
 744             dy01             = _mm256_sub_pd(iy0,jy1);
 745             dz01             = _mm256_sub_pd(iz0,jz1);
 746             dx02             = _mm256_sub_pd(ix0,jx2);
 747             dy02             = _mm256_sub_pd(iy0,jy2);
 748             dz02             = _mm256_sub_pd(iz0,jz2);
 749             dx10             = _mm256_sub_pd(ix1,jx0);
 750             dy10             = _mm256_sub_pd(iy1,jy0);
 751             dz10             = _mm256_sub_pd(iz1,jz0);
 752             dx11             = _mm256_sub_pd(ix1,jx1);
 753             dy11             = _mm256_sub_pd(iy1,jy1);
 754             dz11             = _mm256_sub_pd(iz1,jz1);
 755             dx12             = _mm256_sub_pd(ix1,jx2);
 756             dy12             = _mm256_sub_pd(iy1,jy2);
 757             dz12             = _mm256_sub_pd(iz1,jz2);
 758             dx20             = _mm256_sub_pd(ix2,jx0);
 759             dy20             = _mm256_sub_pd(iy2,jy0);
 760             dz20             = _mm256_sub_pd(iz2,jz0);
 761             dx21             = _mm256_sub_pd(ix2,jx1);
 762             dy21             = _mm256_sub_pd(iy2,jy1);
 763             dz21             = _mm256_sub_pd(iz2,jz1);
 764             dx22             = _mm256_sub_pd(ix2,jx2);
 765             dy22             = _mm256_sub_pd(iy2,jy2);
 766             dz22             = _mm256_sub_pd(iz2,jz2);
 767
 768             /* Calculate squared distance and things based on it */
 769             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 770             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 771             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 772             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 773             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 774             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 775             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 776             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 777             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 778
 779             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 780             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 781             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 782             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 783             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 784             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 785             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 786             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 787             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 788
 789             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 790             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 791             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 792             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 793             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 794             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 795             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 796             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 797             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 798
 799             fjx0             = _mm256_setzero_pd();
 800             fjy0             = _mm256_setzero_pd();
 801             fjz0             = _mm256_setzero_pd();
 802             fjx1             = _mm256_setzero_pd();
 803             fjy1             = _mm256_setzero_pd();
 804             fjz1             = _mm256_setzero_pd();
 805             fjx2             = _mm256_setzero_pd();
 806             fjy2             = _mm256_setzero_pd();
 807             fjz2             = _mm256_setzero_pd();
 808
 809             /**************************
 810              * CALCULATE INTERACTIONS *
 811              **************************/
 812
 813             r00              = _mm256_mul_pd(rsq00,rinv00);
 814             r00              = _mm256_andnot_pd(dummy_mask,r00);
 815
 816             /* EWALD ELECTROSTATICS */
 817
 818             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 819             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 820             ewitab           = _mm256_cvttpd_epi32(ewrt);
 821             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 822             ewitab           = _mm_slli_epi32(ewitab,2);
 823             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 824             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 825             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 826             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 827             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 828             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 829             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 830             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 831             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 832
 833             /* LENNARD-JONES DISPERSION/REPULSION */
 834
 835             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 836             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 837             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 838             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 839             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 840
 841             /* Update potential sum for this i atom from the interaction with this j atom. */
 842             velec            = _mm256_andnot_pd(dummy_mask,velec);
 843             velecsum         = _mm256_add_pd(velecsum,velec);
 844             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 845             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 846
 847             fscal            = _mm256_add_pd(felec,fvdw);
 848
 849             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 850
 851             /* Calculate temporary vectorial force */
 852             tx               = _mm256_mul_pd(fscal,dx00);
 853             ty               = _mm256_mul_pd(fscal,dy00);
 854             tz               = _mm256_mul_pd(fscal,dz00);
 855
 856             /* Update vectorial force */
 857             fix0             = _mm256_add_pd(fix0,tx);
 858             fiy0             = _mm256_add_pd(fiy0,ty);
 859             fiz0             = _mm256_add_pd(fiz0,tz);
 860
 861             fjx0             = _mm256_add_pd(fjx0,tx);
 862             fjy0             = _mm256_add_pd(fjy0,ty);
 863             fjz0             = _mm256_add_pd(fjz0,tz);
 864
 865             /**************************
 866              * CALCULATE INTERACTIONS *
 867              **************************/
 868
 869             r01              = _mm256_mul_pd(rsq01,rinv01);
 870             r01              = _mm256_andnot_pd(dummy_mask,r01);
 871
 872             /* EWALD ELECTROSTATICS */
 873
 874             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 875             ewrt             = _mm256_mul_pd(r01,ewtabscale);
 876             ewitab           = _mm256_cvttpd_epi32(ewrt);
 877             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 878             ewitab           = _mm_slli_epi32(ewitab,2);
 879             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 880             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 881             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 882             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 883             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 884             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 885             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 886             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(rinv01,velec));
 887             felec            = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
 888
 889             /* Update potential sum for this i atom from the interaction with this j atom. */
 890             velec            = _mm256_andnot_pd(dummy_mask,velec);
 891             velecsum         = _mm256_add_pd(velecsum,velec);
 892
 893             fscal            = felec;
 894
 895             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 896
 897             /* Calculate temporary vectorial force */
 898             tx               = _mm256_mul_pd(fscal,dx01);
 899             ty               = _mm256_mul_pd(fscal,dy01);
 900             tz               = _mm256_mul_pd(fscal,dz01);
 901
 902             /* Update vectorial force */
 903             fix0             = _mm256_add_pd(fix0,tx);
 904             fiy0             = _mm256_add_pd(fiy0,ty);
 905             fiz0             = _mm256_add_pd(fiz0,tz);
 906
 907             fjx1             = _mm256_add_pd(fjx1,tx);
 908             fjy1             = _mm256_add_pd(fjy1,ty);
 909             fjz1             = _mm256_add_pd(fjz1,tz);
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             r02              = _mm256_mul_pd(rsq02,rinv02);
 916             r02              = _mm256_andnot_pd(dummy_mask,r02);
 917
 918             /* EWALD ELECTROSTATICS */
 919
 920             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 921             ewrt             = _mm256_mul_pd(r02,ewtabscale);
 922             ewitab           = _mm256_cvttpd_epi32(ewrt);
 923             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 924             ewitab           = _mm_slli_epi32(ewitab,2);
 925             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 926             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 927             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 928             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 929             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 930             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 931             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 932             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(rinv02,velec));
 933             felec            = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
 934
 935             /* Update potential sum for this i atom from the interaction with this j atom. */
 936             velec            = _mm256_andnot_pd(dummy_mask,velec);
 937             velecsum         = _mm256_add_pd(velecsum,velec);
 938
 939             fscal            = felec;
 940
 941             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 942
 943             /* Calculate temporary vectorial force */
 944             tx               = _mm256_mul_pd(fscal,dx02);
 945             ty               = _mm256_mul_pd(fscal,dy02);
 946             tz               = _mm256_mul_pd(fscal,dz02);
 947
 948             /* Update vectorial force */
 949             fix0             = _mm256_add_pd(fix0,tx);
 950             fiy0             = _mm256_add_pd(fiy0,ty);
 951             fiz0             = _mm256_add_pd(fiz0,tz);
 952
 953             fjx2             = _mm256_add_pd(fjx2,tx);
 954             fjy2             = _mm256_add_pd(fjy2,ty);
 955             fjz2             = _mm256_add_pd(fjz2,tz);
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             r10              = _mm256_mul_pd(rsq10,rinv10);
 962             r10              = _mm256_andnot_pd(dummy_mask,r10);
 963
 964             /* EWALD ELECTROSTATICS */
 965
 966             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 967             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 968             ewitab           = _mm256_cvttpd_epi32(ewrt);
 969             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 970             ewitab           = _mm_slli_epi32(ewitab,2);
 971             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 972             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 973             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 974             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 975             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 976             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 977             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 978             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 979             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 980
 981             /* Update potential sum for this i atom from the interaction with this j atom. */
 982             velec            = _mm256_andnot_pd(dummy_mask,velec);
 983             velecsum         = _mm256_add_pd(velecsum,velec);
 984
 985             fscal            = felec;
 986
 987             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 988
 989             /* Calculate temporary vectorial force */
 990             tx               = _mm256_mul_pd(fscal,dx10);
 991             ty               = _mm256_mul_pd(fscal,dy10);
 992             tz               = _mm256_mul_pd(fscal,dz10);
 993
 994             /* Update vectorial force */
 995             fix1             = _mm256_add_pd(fix1,tx);
 996             fiy1             = _mm256_add_pd(fiy1,ty);
 997             fiz1             = _mm256_add_pd(fiz1,tz);
 998
 999             fjx0             = _mm256_add_pd(fjx0,tx);
1000             fjy0             = _mm256_add_pd(fjy0,ty);
1001             fjz0             = _mm256_add_pd(fjz0,tz);
1002
1003             /**************************
1004              * CALCULATE INTERACTIONS *
1005              **************************/
1006
1007             r11              = _mm256_mul_pd(rsq11,rinv11);
1008             r11              = _mm256_andnot_pd(dummy_mask,r11);
1009
1010             /* EWALD ELECTROSTATICS */
1011
1012             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1013             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1014             ewitab           = _mm256_cvttpd_epi32(ewrt);
1015             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1016             ewitab           = _mm_slli_epi32(ewitab,2);
1017             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1018             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1019             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1020             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1021             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1022             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1023             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1024             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
1025             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1026
1027             /* Update potential sum for this i atom from the interaction with this j atom. */
1028             velec            = _mm256_andnot_pd(dummy_mask,velec);
1029             velecsum         = _mm256_add_pd(velecsum,velec);
1030
1031             fscal            = felec;
1032
1033             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1034
1035             /* Calculate temporary vectorial force */
1036             tx               = _mm256_mul_pd(fscal,dx11);
1037             ty               = _mm256_mul_pd(fscal,dy11);
1038             tz               = _mm256_mul_pd(fscal,dz11);
1039
1040             /* Update vectorial force */
1041             fix1             = _mm256_add_pd(fix1,tx);
1042             fiy1             = _mm256_add_pd(fiy1,ty);
1043             fiz1             = _mm256_add_pd(fiz1,tz);
1044
1045             fjx1             = _mm256_add_pd(fjx1,tx);
1046             fjy1             = _mm256_add_pd(fjy1,ty);
1047             fjz1             = _mm256_add_pd(fjz1,tz);
1048
1049             /**************************
1050              * CALCULATE INTERACTIONS *
1051              **************************/
1052
1053             r12              = _mm256_mul_pd(rsq12,rinv12);
1054             r12              = _mm256_andnot_pd(dummy_mask,r12);
1055
1056             /* EWALD ELECTROSTATICS */
1057
1058             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1059             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1060             ewitab           = _mm256_cvttpd_epi32(ewrt);
1061             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1062             ewitab           = _mm_slli_epi32(ewitab,2);
1063             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1064             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1065             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1066             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1067             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1068             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1069             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1070             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
1071             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1072
1073             /* Update potential sum for this i atom from the interaction with this j atom. */
1074             velec            = _mm256_andnot_pd(dummy_mask,velec);
1075             velecsum         = _mm256_add_pd(velecsum,velec);
1076
1077             fscal            = felec;
1078
1079             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1080
1081             /* Calculate temporary vectorial force */
1082             tx               = _mm256_mul_pd(fscal,dx12);
1083             ty               = _mm256_mul_pd(fscal,dy12);
1084             tz               = _mm256_mul_pd(fscal,dz12);
1085
1086             /* Update vectorial force */
1087             fix1             = _mm256_add_pd(fix1,tx);
1088             fiy1             = _mm256_add_pd(fiy1,ty);
1089             fiz1             = _mm256_add_pd(fiz1,tz);
1090
1091             fjx2             = _mm256_add_pd(fjx2,tx);
1092             fjy2             = _mm256_add_pd(fjy2,ty);
1093             fjz2             = _mm256_add_pd(fjz2,tz);
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             r20              = _mm256_mul_pd(rsq20,rinv20);
1100             r20              = _mm256_andnot_pd(dummy_mask,r20);
1101
1102             /* EWALD ELECTROSTATICS */
1103
1104             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1105             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1106             ewitab           = _mm256_cvttpd_epi32(ewrt);
1107             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1108             ewitab           = _mm_slli_epi32(ewitab,2);
1109             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1110             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1111             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1112             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1113             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1114             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1115             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1116             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
1117             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1118
1119             /* Update potential sum for this i atom from the interaction with this j atom. */
1120             velec            = _mm256_andnot_pd(dummy_mask,velec);
1121             velecsum         = _mm256_add_pd(velecsum,velec);
1122
1123             fscal            = felec;
1124
1125             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1126
1127             /* Calculate temporary vectorial force */
1128             tx               = _mm256_mul_pd(fscal,dx20);
1129             ty               = _mm256_mul_pd(fscal,dy20);
1130             tz               = _mm256_mul_pd(fscal,dz20);
1131
1132             /* Update vectorial force */
1133             fix2             = _mm256_add_pd(fix2,tx);
1134             fiy2             = _mm256_add_pd(fiy2,ty);
1135             fiz2             = _mm256_add_pd(fiz2,tz);
1136
1137             fjx0             = _mm256_add_pd(fjx0,tx);
1138             fjy0             = _mm256_add_pd(fjy0,ty);
1139             fjz0             = _mm256_add_pd(fjz0,tz);
1140
1141             /**************************
1142              * CALCULATE INTERACTIONS *
1143              **************************/
1144
1145             r21              = _mm256_mul_pd(rsq21,rinv21);
1146             r21              = _mm256_andnot_pd(dummy_mask,r21);
1147
1148             /* EWALD ELECTROSTATICS */
1149
1150             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1151             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1152             ewitab           = _mm256_cvttpd_epi32(ewrt);
1153             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1154             ewitab           = _mm_slli_epi32(ewitab,2);
1155             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1156             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1157             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1158             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1159             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1160             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1161             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1162             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
1163             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1164
1165             /* Update potential sum for this i atom from the interaction with this j atom. */
1166             velec            = _mm256_andnot_pd(dummy_mask,velec);
1167             velecsum         = _mm256_add_pd(velecsum,velec);
1168
1169             fscal            = felec;
1170
1171             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1172
1173             /* Calculate temporary vectorial force */
1174             tx               = _mm256_mul_pd(fscal,dx21);
1175             ty               = _mm256_mul_pd(fscal,dy21);
1176             tz               = _mm256_mul_pd(fscal,dz21);
1177
1178             /* Update vectorial force */
1179             fix2             = _mm256_add_pd(fix2,tx);
1180             fiy2             = _mm256_add_pd(fiy2,ty);
1181             fiz2             = _mm256_add_pd(fiz2,tz);
1182
1183             fjx1             = _mm256_add_pd(fjx1,tx);
1184             fjy1             = _mm256_add_pd(fjy1,ty);
1185             fjz1             = _mm256_add_pd(fjz1,tz);
1186
1187             /**************************
1188              * CALCULATE INTERACTIONS *
1189              **************************/
1190
1191             r22              = _mm256_mul_pd(rsq22,rinv22);
1192             r22              = _mm256_andnot_pd(dummy_mask,r22);
1193
1194             /* EWALD ELECTROSTATICS */
1195
1196             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1197             ewrt             = _mm256_mul_pd(r22,ewtabscale);
1198             ewitab           = _mm256_cvttpd_epi32(ewrt);
1199             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1200             ewitab           = _mm_slli_epi32(ewitab,2);
1201             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1202             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1203             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1204             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1205             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1206             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1207             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1208             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
1209             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1210
1211             /* Update potential sum for this i atom from the interaction with this j atom. */
1212             velec            = _mm256_andnot_pd(dummy_mask,velec);
1213             velecsum         = _mm256_add_pd(velecsum,velec);
1214
1215             fscal            = felec;
1216
1217             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1218
1219             /* Calculate temporary vectorial force */
1220             tx               = _mm256_mul_pd(fscal,dx22);
1221             ty               = _mm256_mul_pd(fscal,dy22);
1222             tz               = _mm256_mul_pd(fscal,dz22);
1223
1224             /* Update vectorial force */
1225             fix2             = _mm256_add_pd(fix2,tx);
1226             fiy2             = _mm256_add_pd(fiy2,ty);
1227             fiz2             = _mm256_add_pd(fiz2,tz);
1228
1229             fjx2             = _mm256_add_pd(fjx2,tx);
1230             fjy2             = _mm256_add_pd(fjy2,ty);
1231             fjz2             = _mm256_add_pd(fjz2,tz);
1232
1233             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1234             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1235             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1236             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1237
1238             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1239                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1240
1241             /* Inner loop uses 390 flops */
1242         }
1243
1244         /* End of innermost loop */
1245
1246         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1247                                                  f+i_coord_offset,fshift+i_shift_offset);
1248
1249         ggid                        = gid[iidx];
1250         /* Update potential energies */
1251         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1252         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1253
1254         /* Increment number of inner iterations */
1255         inneriter                  += j_index_end - j_index_start;
1256
1257         /* Outer loop uses 20 flops */
1258     }
1259
1260     /* Increment number of outer iterations */
1261     outeriter        += nri;
1262
1263     /* Update outer/inner flops */
1264
1265     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*390);
1266 }
1267 /*
1268  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3W3_F_avx_256_double
1269  * Electrostatics interaction: Ewald
1270  * VdW interaction:            LennardJones
1271  * Geometry:                   Water3-Water3
1272  * Calculate force/pot:        Force
1273  */
1274 void
1275 nb_kernel_ElecEw_VdwLJ_GeomW3W3_F_avx_256_double
1276                     (t_nblist                    * gmx_restrict       nlist,
1277                      rvec                        * gmx_restrict          xx,
1278                      rvec                        * gmx_restrict          ff,
1279                      t_forcerec                  * gmx_restrict          fr,
1280                      t_mdatoms                   * gmx_restrict     mdatoms,
1281                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1282                      t_nrnb                      * gmx_restrict        nrnb)
1283 {
1284     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1285      * just 0 for non-waters.
1286      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1287      * jnr indices corresponding to data put in the four positions in the SIMD register.
1288      */
1289     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1290     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1291     int              jnrA,jnrB,jnrC,jnrD;
1292     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1293     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1294     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1295     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1296     real             rcutoff_scalar;
1297     real             *shiftvec,*fshift,*x,*f;
1298     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1299     real             scratch[4*DIM];
1300     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1301     real *           vdwioffsetptr0;
1302     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1303     real *           vdwioffsetptr1;
1304     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1305     real *           vdwioffsetptr2;
1306     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1307     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1308     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1309     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1310     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1311     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1312     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1313     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1314     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1315     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1316     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1317     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1318     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1319     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1320     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1321     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1322     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1323     real             *charge;
1324     int              nvdwtype;
1325     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1326     int              *vdwtype;
1327     real             *vdwparam;
1328     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1329     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1330     __m128i          ewitab;
1331     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1332     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1333     real             *ewtab;
1334     __m256d          dummy_mask,cutoff_mask;
1335     __m128           tmpmask0,tmpmask1;
1336     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1337     __m256d          one     = _mm256_set1_pd(1.0);
1338     __m256d          two     = _mm256_set1_pd(2.0);
1339     x                = xx[0];
1340     f                = ff[0];
1341
1342     nri              = nlist->nri;
1343     iinr             = nlist->iinr;
1344     jindex           = nlist->jindex;
1345     jjnr             = nlist->jjnr;
1346     shiftidx         = nlist->shift;
1347     gid              = nlist->gid;
1348     shiftvec         = fr->shift_vec[0];
1349     fshift           = fr->fshift[0];
1350     facel            = _mm256_set1_pd(fr->epsfac);
1351     charge           = mdatoms->chargeA;
1352     nvdwtype         = fr->ntype;
1353     vdwparam         = fr->nbfp;
1354     vdwtype          = mdatoms->typeA;
1355
1356     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
1357     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
1358     beta2            = _mm256_mul_pd(beta,beta);
1359     beta3            = _mm256_mul_pd(beta,beta2);
1360
1361     ewtab            = fr->ic->tabq_coul_F;
1362     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
1363     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
1364
1365     /* Setup water-specific parameters */
1366     inr              = nlist->iinr[0];
1367     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1368     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1369     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1370     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1371
1372     jq0              = _mm256_set1_pd(charge[inr+0]);
1373     jq1              = _mm256_set1_pd(charge[inr+1]);
1374     jq2              = _mm256_set1_pd(charge[inr+2]);
1375     vdwjidx0A        = 2*vdwtype[inr+0];
1376     qq00             = _mm256_mul_pd(iq0,jq0);
1377     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1378     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1379     qq01             = _mm256_mul_pd(iq0,jq1);
1380     qq02             = _mm256_mul_pd(iq0,jq2);
1381     qq10             = _mm256_mul_pd(iq1,jq0);
1382     qq11             = _mm256_mul_pd(iq1,jq1);
1383     qq12             = _mm256_mul_pd(iq1,jq2);
1384     qq20             = _mm256_mul_pd(iq2,jq0);
1385     qq21             = _mm256_mul_pd(iq2,jq1);
1386     qq22             = _mm256_mul_pd(iq2,jq2);
1387
1388     /* Avoid stupid compiler warnings */
1389     jnrA = jnrB = jnrC = jnrD = 0;
1390     j_coord_offsetA = 0;
1391     j_coord_offsetB = 0;
1392     j_coord_offsetC = 0;
1393     j_coord_offsetD = 0;
1394
1395     outeriter        = 0;
1396     inneriter        = 0;
1397
1398     for(iidx=0;iidx<4*DIM;iidx++)
1399     {
1400         scratch[iidx] = 0.0;
1401     }
1402
1403     /* Start outer loop over neighborlists */
1404     for(iidx=0; iidx<nri; iidx++)
1405     {
1406         /* Load shift vector for this list */
1407         i_shift_offset   = DIM*shiftidx[iidx];
1408
1409         /* Load limits for loop over neighbors */
1410         j_index_start    = jindex[iidx];
1411         j_index_end      = jindex[iidx+1];
1412
1413         /* Get outer coordinate index */
1414         inr              = iinr[iidx];
1415         i_coord_offset   = DIM*inr;
1416
1417         /* Load i particle coords and add shift vector */
1418         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1419                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1420
1421         fix0             = _mm256_setzero_pd();
1422         fiy0             = _mm256_setzero_pd();
1423         fiz0             = _mm256_setzero_pd();
1424         fix1             = _mm256_setzero_pd();
1425         fiy1             = _mm256_setzero_pd();
1426         fiz1             = _mm256_setzero_pd();
1427         fix2             = _mm256_setzero_pd();
1428         fiy2             = _mm256_setzero_pd();
1429         fiz2             = _mm256_setzero_pd();
1430
1431         /* Start inner kernel loop */
1432         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1433         {
1434
1435             /* Get j neighbor index, and coordinate index */
1436             jnrA             = jjnr[jidx];
1437             jnrB             = jjnr[jidx+1];
1438             jnrC             = jjnr[jidx+2];
1439             jnrD             = jjnr[jidx+3];
1440             j_coord_offsetA  = DIM*jnrA;
1441             j_coord_offsetB  = DIM*jnrB;
1442             j_coord_offsetC  = DIM*jnrC;
1443             j_coord_offsetD  = DIM*jnrD;
1444
1445             /* load j atom coordinates */
1446             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1447                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1448                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1449
1450             /* Calculate displacement vector */
1451             dx00             = _mm256_sub_pd(ix0,jx0);
1452             dy00             = _mm256_sub_pd(iy0,jy0);
1453             dz00             = _mm256_sub_pd(iz0,jz0);
1454             dx01             = _mm256_sub_pd(ix0,jx1);
1455             dy01             = _mm256_sub_pd(iy0,jy1);
1456             dz01             = _mm256_sub_pd(iz0,jz1);
1457             dx02             = _mm256_sub_pd(ix0,jx2);
1458             dy02             = _mm256_sub_pd(iy0,jy2);
1459             dz02             = _mm256_sub_pd(iz0,jz2);
1460             dx10             = _mm256_sub_pd(ix1,jx0);
1461             dy10             = _mm256_sub_pd(iy1,jy0);
1462             dz10             = _mm256_sub_pd(iz1,jz0);
1463             dx11             = _mm256_sub_pd(ix1,jx1);
1464             dy11             = _mm256_sub_pd(iy1,jy1);
1465             dz11             = _mm256_sub_pd(iz1,jz1);
1466             dx12             = _mm256_sub_pd(ix1,jx2);
1467             dy12             = _mm256_sub_pd(iy1,jy2);
1468             dz12             = _mm256_sub_pd(iz1,jz2);
1469             dx20             = _mm256_sub_pd(ix2,jx0);
1470             dy20             = _mm256_sub_pd(iy2,jy0);
1471             dz20             = _mm256_sub_pd(iz2,jz0);
1472             dx21             = _mm256_sub_pd(ix2,jx1);
1473             dy21             = _mm256_sub_pd(iy2,jy1);
1474             dz21             = _mm256_sub_pd(iz2,jz1);
1475             dx22             = _mm256_sub_pd(ix2,jx2);
1476             dy22             = _mm256_sub_pd(iy2,jy2);
1477             dz22             = _mm256_sub_pd(iz2,jz2);
1478
1479             /* Calculate squared distance and things based on it */
1480             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1481             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1482             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1483             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1484             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1485             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1486             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1487             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1488             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1489
1490             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1491             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1492             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1493             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1494             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1495             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1496             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1497             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1498             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1499
1500             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1501             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1502             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1503             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1504             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1505             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1506             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1507             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1508             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1509
1510             fjx0             = _mm256_setzero_pd();
1511             fjy0             = _mm256_setzero_pd();
1512             fjz0             = _mm256_setzero_pd();
1513             fjx1             = _mm256_setzero_pd();
1514             fjy1             = _mm256_setzero_pd();
1515             fjz1             = _mm256_setzero_pd();
1516             fjx2             = _mm256_setzero_pd();
1517             fjy2             = _mm256_setzero_pd();
1518             fjz2             = _mm256_setzero_pd();
1519
1520             /**************************
1521              * CALCULATE INTERACTIONS *
1522              **************************/
1523
1524             r00              = _mm256_mul_pd(rsq00,rinv00);
1525
1526             /* EWALD ELECTROSTATICS */
1527
1528             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1529             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1530             ewitab           = _mm256_cvttpd_epi32(ewrt);
1531             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1532             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1533                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1534                                             &ewtabF,&ewtabFn);
1535             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1536             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1537
1538             /* LENNARD-JONES DISPERSION/REPULSION */
1539
1540             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1541             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1542
1543             fscal            = _mm256_add_pd(felec,fvdw);
1544
1545             /* Calculate temporary vectorial force */
1546             tx               = _mm256_mul_pd(fscal,dx00);
1547             ty               = _mm256_mul_pd(fscal,dy00);
1548             tz               = _mm256_mul_pd(fscal,dz00);
1549
1550             /* Update vectorial force */
1551             fix0             = _mm256_add_pd(fix0,tx);
1552             fiy0             = _mm256_add_pd(fiy0,ty);
1553             fiz0             = _mm256_add_pd(fiz0,tz);
1554
1555             fjx0             = _mm256_add_pd(fjx0,tx);
1556             fjy0             = _mm256_add_pd(fjy0,ty);
1557             fjz0             = _mm256_add_pd(fjz0,tz);
1558
1559             /**************************
1560              * CALCULATE INTERACTIONS *
1561              **************************/
1562
1563             r01              = _mm256_mul_pd(rsq01,rinv01);
1564
1565             /* EWALD ELECTROSTATICS */
1566
1567             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1568             ewrt             = _mm256_mul_pd(r01,ewtabscale);
1569             ewitab           = _mm256_cvttpd_epi32(ewrt);
1570             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1571             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1572                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1573                                             &ewtabF,&ewtabFn);
1574             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1575             felec            = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
1576
1577             fscal            = felec;
1578
1579             /* Calculate temporary vectorial force */
1580             tx               = _mm256_mul_pd(fscal,dx01);
1581             ty               = _mm256_mul_pd(fscal,dy01);
1582             tz               = _mm256_mul_pd(fscal,dz01);
1583
1584             /* Update vectorial force */
1585             fix0             = _mm256_add_pd(fix0,tx);
1586             fiy0             = _mm256_add_pd(fiy0,ty);
1587             fiz0             = _mm256_add_pd(fiz0,tz);
1588
1589             fjx1             = _mm256_add_pd(fjx1,tx);
1590             fjy1             = _mm256_add_pd(fjy1,ty);
1591             fjz1             = _mm256_add_pd(fjz1,tz);
1592
1593             /**************************
1594              * CALCULATE INTERACTIONS *
1595              **************************/
1596
1597             r02              = _mm256_mul_pd(rsq02,rinv02);
1598
1599             /* EWALD ELECTROSTATICS */
1600
1601             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1602             ewrt             = _mm256_mul_pd(r02,ewtabscale);
1603             ewitab           = _mm256_cvttpd_epi32(ewrt);
1604             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1605             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1606                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1607                                             &ewtabF,&ewtabFn);
1608             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1609             felec            = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
1610
1611             fscal            = felec;
1612
1613             /* Calculate temporary vectorial force */
1614             tx               = _mm256_mul_pd(fscal,dx02);
1615             ty               = _mm256_mul_pd(fscal,dy02);
1616             tz               = _mm256_mul_pd(fscal,dz02);
1617
1618             /* Update vectorial force */
1619             fix0             = _mm256_add_pd(fix0,tx);
1620             fiy0             = _mm256_add_pd(fiy0,ty);
1621             fiz0             = _mm256_add_pd(fiz0,tz);
1622
1623             fjx2             = _mm256_add_pd(fjx2,tx);
1624             fjy2             = _mm256_add_pd(fjy2,ty);
1625             fjz2             = _mm256_add_pd(fjz2,tz);
1626
1627             /**************************
1628              * CALCULATE INTERACTIONS *
1629              **************************/
1630
1631             r10              = _mm256_mul_pd(rsq10,rinv10);
1632
1633             /* EWALD ELECTROSTATICS */
1634
1635             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1636             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1637             ewitab           = _mm256_cvttpd_epi32(ewrt);
1638             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1639             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1640                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1641                                             &ewtabF,&ewtabFn);
1642             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1643             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1644
1645             fscal            = felec;
1646
1647             /* Calculate temporary vectorial force */
1648             tx               = _mm256_mul_pd(fscal,dx10);
1649             ty               = _mm256_mul_pd(fscal,dy10);
1650             tz               = _mm256_mul_pd(fscal,dz10);
1651
1652             /* Update vectorial force */
1653             fix1             = _mm256_add_pd(fix1,tx);
1654             fiy1             = _mm256_add_pd(fiy1,ty);
1655             fiz1             = _mm256_add_pd(fiz1,tz);
1656
1657             fjx0             = _mm256_add_pd(fjx0,tx);
1658             fjy0             = _mm256_add_pd(fjy0,ty);
1659             fjz0             = _mm256_add_pd(fjz0,tz);
1660
1661             /**************************
1662              * CALCULATE INTERACTIONS *
1663              **************************/
1664
1665             r11              = _mm256_mul_pd(rsq11,rinv11);
1666
1667             /* EWALD ELECTROSTATICS */
1668
1669             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1670             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1671             ewitab           = _mm256_cvttpd_epi32(ewrt);
1672             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1673             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1674                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1675                                             &ewtabF,&ewtabFn);
1676             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1677             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1678
1679             fscal            = felec;
1680
1681             /* Calculate temporary vectorial force */
1682             tx               = _mm256_mul_pd(fscal,dx11);
1683             ty               = _mm256_mul_pd(fscal,dy11);
1684             tz               = _mm256_mul_pd(fscal,dz11);
1685
1686             /* Update vectorial force */
1687             fix1             = _mm256_add_pd(fix1,tx);
1688             fiy1             = _mm256_add_pd(fiy1,ty);
1689             fiz1             = _mm256_add_pd(fiz1,tz);
1690
1691             fjx1             = _mm256_add_pd(fjx1,tx);
1692             fjy1             = _mm256_add_pd(fjy1,ty);
1693             fjz1             = _mm256_add_pd(fjz1,tz);
1694
1695             /**************************
1696              * CALCULATE INTERACTIONS *
1697              **************************/
1698
1699             r12              = _mm256_mul_pd(rsq12,rinv12);
1700
1701             /* EWALD ELECTROSTATICS */
1702
1703             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1704             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1705             ewitab           = _mm256_cvttpd_epi32(ewrt);
1706             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1707             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1708                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1709                                             &ewtabF,&ewtabFn);
1710             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1711             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1712
1713             fscal            = felec;
1714
1715             /* Calculate temporary vectorial force */
1716             tx               = _mm256_mul_pd(fscal,dx12);
1717             ty               = _mm256_mul_pd(fscal,dy12);
1718             tz               = _mm256_mul_pd(fscal,dz12);
1719
1720             /* Update vectorial force */
1721             fix1             = _mm256_add_pd(fix1,tx);
1722             fiy1             = _mm256_add_pd(fiy1,ty);
1723             fiz1             = _mm256_add_pd(fiz1,tz);
1724
1725             fjx2             = _mm256_add_pd(fjx2,tx);
1726             fjy2             = _mm256_add_pd(fjy2,ty);
1727             fjz2             = _mm256_add_pd(fjz2,tz);
1728
1729             /**************************
1730              * CALCULATE INTERACTIONS *
1731              **************************/
1732
1733             r20              = _mm256_mul_pd(rsq20,rinv20);
1734
1735             /* EWALD ELECTROSTATICS */
1736
1737             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1738             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1739             ewitab           = _mm256_cvttpd_epi32(ewrt);
1740             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1741             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1742                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1743                                             &ewtabF,&ewtabFn);
1744             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1745             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1746
1747             fscal            = felec;
1748
1749             /* Calculate temporary vectorial force */
1750             tx               = _mm256_mul_pd(fscal,dx20);
1751             ty               = _mm256_mul_pd(fscal,dy20);
1752             tz               = _mm256_mul_pd(fscal,dz20);
1753
1754             /* Update vectorial force */
1755             fix2             = _mm256_add_pd(fix2,tx);
1756             fiy2             = _mm256_add_pd(fiy2,ty);
1757             fiz2             = _mm256_add_pd(fiz2,tz);
1758
1759             fjx0             = _mm256_add_pd(fjx0,tx);
1760             fjy0             = _mm256_add_pd(fjy0,ty);
1761             fjz0             = _mm256_add_pd(fjz0,tz);
1762
1763             /**************************
1764              * CALCULATE INTERACTIONS *
1765              **************************/
1766
1767             r21              = _mm256_mul_pd(rsq21,rinv21);
1768
1769             /* EWALD ELECTROSTATICS */
1770
1771             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1772             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1773             ewitab           = _mm256_cvttpd_epi32(ewrt);
1774             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1775             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1776                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1777                                             &ewtabF,&ewtabFn);
1778             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1779             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1780
1781             fscal            = felec;
1782
1783             /* Calculate temporary vectorial force */
1784             tx               = _mm256_mul_pd(fscal,dx21);
1785             ty               = _mm256_mul_pd(fscal,dy21);
1786             tz               = _mm256_mul_pd(fscal,dz21);
1787
1788             /* Update vectorial force */
1789             fix2             = _mm256_add_pd(fix2,tx);
1790             fiy2             = _mm256_add_pd(fiy2,ty);
1791             fiz2             = _mm256_add_pd(fiz2,tz);
1792
1793             fjx1             = _mm256_add_pd(fjx1,tx);
1794             fjy1             = _mm256_add_pd(fjy1,ty);
1795             fjz1             = _mm256_add_pd(fjz1,tz);
1796
1797             /**************************
1798              * CALCULATE INTERACTIONS *
1799              **************************/
1800
1801             r22              = _mm256_mul_pd(rsq22,rinv22);
1802
1803             /* EWALD ELECTROSTATICS */
1804
1805             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1806             ewrt             = _mm256_mul_pd(r22,ewtabscale);
1807             ewitab           = _mm256_cvttpd_epi32(ewrt);
1808             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1809             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1810                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1811                                             &ewtabF,&ewtabFn);
1812             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1813             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1814
1815             fscal            = felec;
1816
1817             /* Calculate temporary vectorial force */
1818             tx               = _mm256_mul_pd(fscal,dx22);
1819             ty               = _mm256_mul_pd(fscal,dy22);
1820             tz               = _mm256_mul_pd(fscal,dz22);
1821
1822             /* Update vectorial force */
1823             fix2             = _mm256_add_pd(fix2,tx);
1824             fiy2             = _mm256_add_pd(fiy2,ty);
1825             fiz2             = _mm256_add_pd(fiz2,tz);
1826
1827             fjx2             = _mm256_add_pd(fjx2,tx);
1828             fjy2             = _mm256_add_pd(fjy2,ty);
1829             fjz2             = _mm256_add_pd(fjz2,tz);
1830
1831             fjptrA             = f+j_coord_offsetA;
1832             fjptrB             = f+j_coord_offsetB;
1833             fjptrC             = f+j_coord_offsetC;
1834             fjptrD             = f+j_coord_offsetD;
1835
1836             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1837                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1838
1839             /* Inner loop uses 331 flops */
1840         }
1841
1842         if(jidx<j_index_end)
1843         {
1844
1845             /* Get j neighbor index, and coordinate index */
1846             jnrlistA         = jjnr[jidx];
1847             jnrlistB         = jjnr[jidx+1];
1848             jnrlistC         = jjnr[jidx+2];
1849             jnrlistD         = jjnr[jidx+3];
1850             /* Sign of each element will be negative for non-real atoms.
1851              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1852              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1853              */
1854             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1855
1856             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1857             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1858             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1859
1860             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1861             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1862             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1863             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1864             j_coord_offsetA  = DIM*jnrA;
1865             j_coord_offsetB  = DIM*jnrB;
1866             j_coord_offsetC  = DIM*jnrC;
1867             j_coord_offsetD  = DIM*jnrD;
1868
1869             /* load j atom coordinates */
1870             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1871                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1872                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1873
1874             /* Calculate displacement vector */
1875             dx00             = _mm256_sub_pd(ix0,jx0);
1876             dy00             = _mm256_sub_pd(iy0,jy0);
1877             dz00             = _mm256_sub_pd(iz0,jz0);
1878             dx01             = _mm256_sub_pd(ix0,jx1);
1879             dy01             = _mm256_sub_pd(iy0,jy1);
1880             dz01             = _mm256_sub_pd(iz0,jz1);
1881             dx02             = _mm256_sub_pd(ix0,jx2);
1882             dy02             = _mm256_sub_pd(iy0,jy2);
1883             dz02             = _mm256_sub_pd(iz0,jz2);
1884             dx10             = _mm256_sub_pd(ix1,jx0);
1885             dy10             = _mm256_sub_pd(iy1,jy0);
1886             dz10             = _mm256_sub_pd(iz1,jz0);
1887             dx11             = _mm256_sub_pd(ix1,jx1);
1888             dy11             = _mm256_sub_pd(iy1,jy1);
1889             dz11             = _mm256_sub_pd(iz1,jz1);
1890             dx12             = _mm256_sub_pd(ix1,jx2);
1891             dy12             = _mm256_sub_pd(iy1,jy2);
1892             dz12             = _mm256_sub_pd(iz1,jz2);
1893             dx20             = _mm256_sub_pd(ix2,jx0);
1894             dy20             = _mm256_sub_pd(iy2,jy0);
1895             dz20             = _mm256_sub_pd(iz2,jz0);
1896             dx21             = _mm256_sub_pd(ix2,jx1);
1897             dy21             = _mm256_sub_pd(iy2,jy1);
1898             dz21             = _mm256_sub_pd(iz2,jz1);
1899             dx22             = _mm256_sub_pd(ix2,jx2);
1900             dy22             = _mm256_sub_pd(iy2,jy2);
1901             dz22             = _mm256_sub_pd(iz2,jz2);
1902
1903             /* Calculate squared distance and things based on it */
1904             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1905             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1906             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1907             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1908             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1909             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1910             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1911             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1912             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1913
1914             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1915             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1916             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1917             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1918             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1919             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1920             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1921             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1922             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1923
1924             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1925             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1926             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1927             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1928             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1929             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1930             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1931             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1932             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1933
1934             fjx0             = _mm256_setzero_pd();
1935             fjy0             = _mm256_setzero_pd();
1936             fjz0             = _mm256_setzero_pd();
1937             fjx1             = _mm256_setzero_pd();
1938             fjy1             = _mm256_setzero_pd();
1939             fjz1             = _mm256_setzero_pd();
1940             fjx2             = _mm256_setzero_pd();
1941             fjy2             = _mm256_setzero_pd();
1942             fjz2             = _mm256_setzero_pd();
1943
1944             /**************************
1945              * CALCULATE INTERACTIONS *
1946              **************************/
1947
1948             r00              = _mm256_mul_pd(rsq00,rinv00);
1949             r00              = _mm256_andnot_pd(dummy_mask,r00);
1950
1951             /* EWALD ELECTROSTATICS */
1952
1953             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1954             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1955             ewitab           = _mm256_cvttpd_epi32(ewrt);
1956             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1957             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1958                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1959                                             &ewtabF,&ewtabFn);
1960             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1961             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1962
1963             /* LENNARD-JONES DISPERSION/REPULSION */
1964
1965             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1966             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1967
1968             fscal            = _mm256_add_pd(felec,fvdw);
1969
1970             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1971
1972             /* Calculate temporary vectorial force */
1973             tx               = _mm256_mul_pd(fscal,dx00);
1974             ty               = _mm256_mul_pd(fscal,dy00);
1975             tz               = _mm256_mul_pd(fscal,dz00);
1976
1977             /* Update vectorial force */
1978             fix0             = _mm256_add_pd(fix0,tx);
1979             fiy0             = _mm256_add_pd(fiy0,ty);
1980             fiz0             = _mm256_add_pd(fiz0,tz);
1981
1982             fjx0             = _mm256_add_pd(fjx0,tx);
1983             fjy0             = _mm256_add_pd(fjy0,ty);
1984             fjz0             = _mm256_add_pd(fjz0,tz);
1985
1986             /**************************
1987              * CALCULATE INTERACTIONS *
1988              **************************/
1989
1990             r01              = _mm256_mul_pd(rsq01,rinv01);
1991             r01              = _mm256_andnot_pd(dummy_mask,r01);
1992
1993             /* EWALD ELECTROSTATICS */
1994
1995             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1996             ewrt             = _mm256_mul_pd(r01,ewtabscale);
1997             ewitab           = _mm256_cvttpd_epi32(ewrt);
1998             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1999             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2000                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2001                                             &ewtabF,&ewtabFn);
2002             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2003             felec            = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
2004
2005             fscal            = felec;
2006
2007             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2008
2009             /* Calculate temporary vectorial force */
2010             tx               = _mm256_mul_pd(fscal,dx01);
2011             ty               = _mm256_mul_pd(fscal,dy01);
2012             tz               = _mm256_mul_pd(fscal,dz01);
2013
2014             /* Update vectorial force */
2015             fix0             = _mm256_add_pd(fix0,tx);
2016             fiy0             = _mm256_add_pd(fiy0,ty);
2017             fiz0             = _mm256_add_pd(fiz0,tz);
2018
2019             fjx1             = _mm256_add_pd(fjx1,tx);
2020             fjy1             = _mm256_add_pd(fjy1,ty);
2021             fjz1             = _mm256_add_pd(fjz1,tz);
2022
2023             /**************************
2024              * CALCULATE INTERACTIONS *
2025              **************************/
2026
2027             r02              = _mm256_mul_pd(rsq02,rinv02);
2028             r02              = _mm256_andnot_pd(dummy_mask,r02);
2029
2030             /* EWALD ELECTROSTATICS */
2031
2032             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2033             ewrt             = _mm256_mul_pd(r02,ewtabscale);
2034             ewitab           = _mm256_cvttpd_epi32(ewrt);
2035             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2036             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2037                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2038                                             &ewtabF,&ewtabFn);
2039             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2040             felec            = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
2041
2042             fscal            = felec;
2043
2044             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2045
2046             /* Calculate temporary vectorial force */
2047             tx               = _mm256_mul_pd(fscal,dx02);
2048             ty               = _mm256_mul_pd(fscal,dy02);
2049             tz               = _mm256_mul_pd(fscal,dz02);
2050
2051             /* Update vectorial force */
2052             fix0             = _mm256_add_pd(fix0,tx);
2053             fiy0             = _mm256_add_pd(fiy0,ty);
2054             fiz0             = _mm256_add_pd(fiz0,tz);
2055
2056             fjx2             = _mm256_add_pd(fjx2,tx);
2057             fjy2             = _mm256_add_pd(fjy2,ty);
2058             fjz2             = _mm256_add_pd(fjz2,tz);
2059
2060             /**************************
2061              * CALCULATE INTERACTIONS *
2062              **************************/
2063
2064             r10              = _mm256_mul_pd(rsq10,rinv10);
2065             r10              = _mm256_andnot_pd(dummy_mask,r10);
2066
2067             /* EWALD ELECTROSTATICS */
2068
2069             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2070             ewrt             = _mm256_mul_pd(r10,ewtabscale);
2071             ewitab           = _mm256_cvttpd_epi32(ewrt);
2072             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2073             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2074                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2075                                             &ewtabF,&ewtabFn);
2076             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2077             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
2078
2079             fscal            = felec;
2080
2081             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2082
2083             /* Calculate temporary vectorial force */
2084             tx               = _mm256_mul_pd(fscal,dx10);
2085             ty               = _mm256_mul_pd(fscal,dy10);
2086             tz               = _mm256_mul_pd(fscal,dz10);
2087
2088             /* Update vectorial force */
2089             fix1             = _mm256_add_pd(fix1,tx);
2090             fiy1             = _mm256_add_pd(fiy1,ty);
2091             fiz1             = _mm256_add_pd(fiz1,tz);
2092
2093             fjx0             = _mm256_add_pd(fjx0,tx);
2094             fjy0             = _mm256_add_pd(fjy0,ty);
2095             fjz0             = _mm256_add_pd(fjz0,tz);
2096
2097             /**************************
2098              * CALCULATE INTERACTIONS *
2099              **************************/
2100
2101             r11              = _mm256_mul_pd(rsq11,rinv11);
2102             r11              = _mm256_andnot_pd(dummy_mask,r11);
2103
2104             /* EWALD ELECTROSTATICS */
2105
2106             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2107             ewrt             = _mm256_mul_pd(r11,ewtabscale);
2108             ewitab           = _mm256_cvttpd_epi32(ewrt);
2109             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2110             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2111                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2112                                             &ewtabF,&ewtabFn);
2113             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2114             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
2115
2116             fscal            = felec;
2117
2118             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2119
2120             /* Calculate temporary vectorial force */
2121             tx               = _mm256_mul_pd(fscal,dx11);
2122             ty               = _mm256_mul_pd(fscal,dy11);
2123             tz               = _mm256_mul_pd(fscal,dz11);
2124
2125             /* Update vectorial force */
2126             fix1             = _mm256_add_pd(fix1,tx);
2127             fiy1             = _mm256_add_pd(fiy1,ty);
2128             fiz1             = _mm256_add_pd(fiz1,tz);
2129
2130             fjx1             = _mm256_add_pd(fjx1,tx);
2131             fjy1             = _mm256_add_pd(fjy1,ty);
2132             fjz1             = _mm256_add_pd(fjz1,tz);
2133
2134             /**************************
2135              * CALCULATE INTERACTIONS *
2136              **************************/
2137
2138             r12              = _mm256_mul_pd(rsq12,rinv12);
2139             r12              = _mm256_andnot_pd(dummy_mask,r12);
2140
2141             /* EWALD ELECTROSTATICS */
2142
2143             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2144             ewrt             = _mm256_mul_pd(r12,ewtabscale);
2145             ewitab           = _mm256_cvttpd_epi32(ewrt);
2146             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2147             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2148                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2149                                             &ewtabF,&ewtabFn);
2150             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2151             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
2152
2153             fscal            = felec;
2154
2155             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2156
2157             /* Calculate temporary vectorial force */
2158             tx               = _mm256_mul_pd(fscal,dx12);
2159             ty               = _mm256_mul_pd(fscal,dy12);
2160             tz               = _mm256_mul_pd(fscal,dz12);
2161
2162             /* Update vectorial force */
2163             fix1             = _mm256_add_pd(fix1,tx);
2164             fiy1             = _mm256_add_pd(fiy1,ty);
2165             fiz1             = _mm256_add_pd(fiz1,tz);
2166
2167             fjx2             = _mm256_add_pd(fjx2,tx);
2168             fjy2             = _mm256_add_pd(fjy2,ty);
2169             fjz2             = _mm256_add_pd(fjz2,tz);
2170
2171             /**************************
2172              * CALCULATE INTERACTIONS *
2173              **************************/
2174
2175             r20              = _mm256_mul_pd(rsq20,rinv20);
2176             r20              = _mm256_andnot_pd(dummy_mask,r20);
2177
2178             /* EWALD ELECTROSTATICS */
2179
2180             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2181             ewrt             = _mm256_mul_pd(r20,ewtabscale);
2182             ewitab           = _mm256_cvttpd_epi32(ewrt);
2183             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2184             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2185                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2186                                             &ewtabF,&ewtabFn);
2187             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2188             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
2189
2190             fscal            = felec;
2191
2192             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2193
2194             /* Calculate temporary vectorial force */
2195             tx               = _mm256_mul_pd(fscal,dx20);
2196             ty               = _mm256_mul_pd(fscal,dy20);
2197             tz               = _mm256_mul_pd(fscal,dz20);
2198
2199             /* Update vectorial force */
2200             fix2             = _mm256_add_pd(fix2,tx);
2201             fiy2             = _mm256_add_pd(fiy2,ty);
2202             fiz2             = _mm256_add_pd(fiz2,tz);
2203
2204             fjx0             = _mm256_add_pd(fjx0,tx);
2205             fjy0             = _mm256_add_pd(fjy0,ty);
2206             fjz0             = _mm256_add_pd(fjz0,tz);
2207
2208             /**************************
2209              * CALCULATE INTERACTIONS *
2210              **************************/
2211
2212             r21              = _mm256_mul_pd(rsq21,rinv21);
2213             r21              = _mm256_andnot_pd(dummy_mask,r21);
2214
2215             /* EWALD ELECTROSTATICS */
2216
2217             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2218             ewrt             = _mm256_mul_pd(r21,ewtabscale);
2219             ewitab           = _mm256_cvttpd_epi32(ewrt);
2220             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2221             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2222                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2223                                             &ewtabF,&ewtabFn);
2224             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2225             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2226
2227             fscal            = felec;
2228
2229             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2230
2231             /* Calculate temporary vectorial force */
2232             tx               = _mm256_mul_pd(fscal,dx21);
2233             ty               = _mm256_mul_pd(fscal,dy21);
2234             tz               = _mm256_mul_pd(fscal,dz21);
2235
2236             /* Update vectorial force */
2237             fix2             = _mm256_add_pd(fix2,tx);
2238             fiy2             = _mm256_add_pd(fiy2,ty);
2239             fiz2             = _mm256_add_pd(fiz2,tz);
2240
2241             fjx1             = _mm256_add_pd(fjx1,tx);
2242             fjy1             = _mm256_add_pd(fjy1,ty);
2243             fjz1             = _mm256_add_pd(fjz1,tz);
2244
2245             /**************************
2246              * CALCULATE INTERACTIONS *
2247              **************************/
2248
2249             r22              = _mm256_mul_pd(rsq22,rinv22);
2250             r22              = _mm256_andnot_pd(dummy_mask,r22);
2251
2252             /* EWALD ELECTROSTATICS */
2253
2254             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2255             ewrt             = _mm256_mul_pd(r22,ewtabscale);
2256             ewitab           = _mm256_cvttpd_epi32(ewrt);
2257             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2258             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2259                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2260                                             &ewtabF,&ewtabFn);
2261             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2262             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2263
2264             fscal            = felec;
2265
2266             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2267
2268             /* Calculate temporary vectorial force */
2269             tx               = _mm256_mul_pd(fscal,dx22);
2270             ty               = _mm256_mul_pd(fscal,dy22);
2271             tz               = _mm256_mul_pd(fscal,dz22);
2272
2273             /* Update vectorial force */
2274             fix2             = _mm256_add_pd(fix2,tx);
2275             fiy2             = _mm256_add_pd(fiy2,ty);
2276             fiz2             = _mm256_add_pd(fiz2,tz);
2277
2278             fjx2             = _mm256_add_pd(fjx2,tx);
2279             fjy2             = _mm256_add_pd(fjy2,ty);
2280             fjz2             = _mm256_add_pd(fjz2,tz);
2281
2282             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2283             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2284             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2285             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2286
2287             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2288                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2289
2290             /* Inner loop uses 340 flops */
2291         }
2292
2293         /* End of innermost loop */
2294
2295         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2296                                                  f+i_coord_offset,fshift+i_shift_offset);
2297
2298         /* Increment number of inner iterations */
2299         inneriter                  += j_index_end - j_index_start;
2300
2301         /* Outer loop uses 18 flops */
2302     }
2303
2304     /* Increment number of outer iterations */
2305     outeriter        += nri;
2306
2307     /* Update outer/inner flops */
2308
2309     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*340);
2310 }