src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwNone_GeomW4W4_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4W4_VF_avx_256_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Water4
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwNone_GeomW4W4_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr1;
  85     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     real *           vdwioffsetptr2;
  87     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     real *           vdwioffsetptr3;
  89     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  91     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  92     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  93     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  94     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  95     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  96     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  97     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  98     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  99     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 100     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 101     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 102     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 103     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 104     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 105     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 106     real             *charge;
 107     __m128i          ewitab;
 108     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 109     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 110     real             *ewtab;
 111     __m256d          dummy_mask,cutoff_mask;
 112     __m128           tmpmask0,tmpmask1;
 113     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 114     __m256d          one     = _mm256_set1_pd(1.0);
 115     __m256d          two     = _mm256_set1_pd(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm256_set1_pd(fr->epsfac);
 128     charge           = mdatoms->chargeA;
 129
 130     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 131     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 132     beta2            = _mm256_mul_pd(beta,beta);
 133     beta3            = _mm256_mul_pd(beta,beta2);
 134
 135     ewtab            = fr->ic->tabq_coul_FDV0;
 136     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 137     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 138
 139     /* Setup water-specific parameters */
 140     inr              = nlist->iinr[0];
 141     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 142     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 143     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 144
 145     jq1              = _mm256_set1_pd(charge[inr+1]);
 146     jq2              = _mm256_set1_pd(charge[inr+2]);
 147     jq3              = _mm256_set1_pd(charge[inr+3]);
 148     qq11             = _mm256_mul_pd(iq1,jq1);
 149     qq12             = _mm256_mul_pd(iq1,jq2);
 150     qq13             = _mm256_mul_pd(iq1,jq3);
 151     qq21             = _mm256_mul_pd(iq2,jq1);
 152     qq22             = _mm256_mul_pd(iq2,jq2);
 153     qq23             = _mm256_mul_pd(iq2,jq3);
 154     qq31             = _mm256_mul_pd(iq3,jq1);
 155     qq32             = _mm256_mul_pd(iq3,jq2);
 156     qq33             = _mm256_mul_pd(iq3,jq3);
 157
 158     /* Avoid stupid compiler warnings */
 159     jnrA = jnrB = jnrC = jnrD = 0;
 160     j_coord_offsetA = 0;
 161     j_coord_offsetB = 0;
 162     j_coord_offsetC = 0;
 163     j_coord_offsetD = 0;
 164
 165     outeriter        = 0;
 166     inneriter        = 0;
 167
 168     for(iidx=0;iidx<4*DIM;iidx++)
 169     {
 170         scratch[iidx] = 0.0;
 171     }
 172
 173     /* Start outer loop over neighborlists */
 174     for(iidx=0; iidx<nri; iidx++)
 175     {
 176         /* Load shift vector for this list */
 177         i_shift_offset   = DIM*shiftidx[iidx];
 178
 179         /* Load limits for loop over neighbors */
 180         j_index_start    = jindex[iidx];
 181         j_index_end      = jindex[iidx+1];
 182
 183         /* Get outer coordinate index */
 184         inr              = iinr[iidx];
 185         i_coord_offset   = DIM*inr;
 186
 187         /* Load i particle coords and add shift vector */
 188         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 189                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 190
 191         fix1             = _mm256_setzero_pd();
 192         fiy1             = _mm256_setzero_pd();
 193         fiz1             = _mm256_setzero_pd();
 194         fix2             = _mm256_setzero_pd();
 195         fiy2             = _mm256_setzero_pd();
 196         fiz2             = _mm256_setzero_pd();
 197         fix3             = _mm256_setzero_pd();
 198         fiy3             = _mm256_setzero_pd();
 199         fiz3             = _mm256_setzero_pd();
 200
 201         /* Reset potential sums */
 202         velecsum         = _mm256_setzero_pd();
 203
 204         /* Start inner kernel loop */
 205         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 206         {
 207
 208             /* Get j neighbor index, and coordinate index */
 209             jnrA             = jjnr[jidx];
 210             jnrB             = jjnr[jidx+1];
 211             jnrC             = jjnr[jidx+2];
 212             jnrD             = jjnr[jidx+3];
 213             j_coord_offsetA  = DIM*jnrA;
 214             j_coord_offsetB  = DIM*jnrB;
 215             j_coord_offsetC  = DIM*jnrC;
 216             j_coord_offsetD  = DIM*jnrD;
 217
 218             /* load j atom coordinates */
 219             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 220                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 221                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 222
 223             /* Calculate displacement vector */
 224             dx11             = _mm256_sub_pd(ix1,jx1);
 225             dy11             = _mm256_sub_pd(iy1,jy1);
 226             dz11             = _mm256_sub_pd(iz1,jz1);
 227             dx12             = _mm256_sub_pd(ix1,jx2);
 228             dy12             = _mm256_sub_pd(iy1,jy2);
 229             dz12             = _mm256_sub_pd(iz1,jz2);
 230             dx13             = _mm256_sub_pd(ix1,jx3);
 231             dy13             = _mm256_sub_pd(iy1,jy3);
 232             dz13             = _mm256_sub_pd(iz1,jz3);
 233             dx21             = _mm256_sub_pd(ix2,jx1);
 234             dy21             = _mm256_sub_pd(iy2,jy1);
 235             dz21             = _mm256_sub_pd(iz2,jz1);
 236             dx22             = _mm256_sub_pd(ix2,jx2);
 237             dy22             = _mm256_sub_pd(iy2,jy2);
 238             dz22             = _mm256_sub_pd(iz2,jz2);
 239             dx23             = _mm256_sub_pd(ix2,jx3);
 240             dy23             = _mm256_sub_pd(iy2,jy3);
 241             dz23             = _mm256_sub_pd(iz2,jz3);
 242             dx31             = _mm256_sub_pd(ix3,jx1);
 243             dy31             = _mm256_sub_pd(iy3,jy1);
 244             dz31             = _mm256_sub_pd(iz3,jz1);
 245             dx32             = _mm256_sub_pd(ix3,jx2);
 246             dy32             = _mm256_sub_pd(iy3,jy2);
 247             dz32             = _mm256_sub_pd(iz3,jz2);
 248             dx33             = _mm256_sub_pd(ix3,jx3);
 249             dy33             = _mm256_sub_pd(iy3,jy3);
 250             dz33             = _mm256_sub_pd(iz3,jz3);
 251
 252             /* Calculate squared distance and things based on it */
 253             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 254             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 255             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 256             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 257             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 258             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 259             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 260             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 261             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 262
 263             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 264             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 265             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 266             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 267             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 268             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 269             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 270             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 271             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 272
 273             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 274             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 275             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 276             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 277             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 278             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 279             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 280             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 281             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 282
 283             fjx1             = _mm256_setzero_pd();
 284             fjy1             = _mm256_setzero_pd();
 285             fjz1             = _mm256_setzero_pd();
 286             fjx2             = _mm256_setzero_pd();
 287             fjy2             = _mm256_setzero_pd();
 288             fjz2             = _mm256_setzero_pd();
 289             fjx3             = _mm256_setzero_pd();
 290             fjy3             = _mm256_setzero_pd();
 291             fjz3             = _mm256_setzero_pd();
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             r11              = _mm256_mul_pd(rsq11,rinv11);
 298
 299             /* EWALD ELECTROSTATICS */
 300
 301             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 302             ewrt             = _mm256_mul_pd(r11,ewtabscale);
 303             ewitab           = _mm256_cvttpd_epi32(ewrt);
 304             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 305             ewitab           = _mm_slli_epi32(ewitab,2);
 306             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 307             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 308             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 309             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 310             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 311             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 312             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 313             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
 314             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velecsum         = _mm256_add_pd(velecsum,velec);
 318
 319             fscal            = felec;
 320
 321             /* Calculate temporary vectorial force */
 322             tx               = _mm256_mul_pd(fscal,dx11);
 323             ty               = _mm256_mul_pd(fscal,dy11);
 324             tz               = _mm256_mul_pd(fscal,dz11);
 325
 326             /* Update vectorial force */
 327             fix1             = _mm256_add_pd(fix1,tx);
 328             fiy1             = _mm256_add_pd(fiy1,ty);
 329             fiz1             = _mm256_add_pd(fiz1,tz);
 330
 331             fjx1             = _mm256_add_pd(fjx1,tx);
 332             fjy1             = _mm256_add_pd(fjy1,ty);
 333             fjz1             = _mm256_add_pd(fjz1,tz);
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             r12              = _mm256_mul_pd(rsq12,rinv12);
 340
 341             /* EWALD ELECTROSTATICS */
 342
 343             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 344             ewrt             = _mm256_mul_pd(r12,ewtabscale);
 345             ewitab           = _mm256_cvttpd_epi32(ewrt);
 346             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 347             ewitab           = _mm_slli_epi32(ewitab,2);
 348             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 349             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 350             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 351             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 352             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 353             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 354             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 355             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
 356             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velecsum         = _mm256_add_pd(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm256_mul_pd(fscal,dx12);
 365             ty               = _mm256_mul_pd(fscal,dy12);
 366             tz               = _mm256_mul_pd(fscal,dz12);
 367
 368             /* Update vectorial force */
 369             fix1             = _mm256_add_pd(fix1,tx);
 370             fiy1             = _mm256_add_pd(fiy1,ty);
 371             fiz1             = _mm256_add_pd(fiz1,tz);
 372
 373             fjx2             = _mm256_add_pd(fjx2,tx);
 374             fjy2             = _mm256_add_pd(fjy2,ty);
 375             fjz2             = _mm256_add_pd(fjz2,tz);
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             r13              = _mm256_mul_pd(rsq13,rinv13);
 382
 383             /* EWALD ELECTROSTATICS */
 384
 385             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 386             ewrt             = _mm256_mul_pd(r13,ewtabscale);
 387             ewitab           = _mm256_cvttpd_epi32(ewrt);
 388             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 389             ewitab           = _mm_slli_epi32(ewitab,2);
 390             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 391             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 392             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 393             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 394             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 395             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 396             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 397             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(rinv13,velec));
 398             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             velecsum         = _mm256_add_pd(velecsum,velec);
 402
 403             fscal            = felec;
 404
 405             /* Calculate temporary vectorial force */
 406             tx               = _mm256_mul_pd(fscal,dx13);
 407             ty               = _mm256_mul_pd(fscal,dy13);
 408             tz               = _mm256_mul_pd(fscal,dz13);
 409
 410             /* Update vectorial force */
 411             fix1             = _mm256_add_pd(fix1,tx);
 412             fiy1             = _mm256_add_pd(fiy1,ty);
 413             fiz1             = _mm256_add_pd(fiz1,tz);
 414
 415             fjx3             = _mm256_add_pd(fjx3,tx);
 416             fjy3             = _mm256_add_pd(fjy3,ty);
 417             fjz3             = _mm256_add_pd(fjz3,tz);
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             r21              = _mm256_mul_pd(rsq21,rinv21);
 424
 425             /* EWALD ELECTROSTATICS */
 426
 427             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 428             ewrt             = _mm256_mul_pd(r21,ewtabscale);
 429             ewitab           = _mm256_cvttpd_epi32(ewrt);
 430             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 431             ewitab           = _mm_slli_epi32(ewitab,2);
 432             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 433             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 434             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 435             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 436             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 437             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 438             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 439             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
 440             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
 441
 442             /* Update potential sum for this i atom from the interaction with this j atom. */
 443             velecsum         = _mm256_add_pd(velecsum,velec);
 444
 445             fscal            = felec;
 446
 447             /* Calculate temporary vectorial force */
 448             tx               = _mm256_mul_pd(fscal,dx21);
 449             ty               = _mm256_mul_pd(fscal,dy21);
 450             tz               = _mm256_mul_pd(fscal,dz21);
 451
 452             /* Update vectorial force */
 453             fix2             = _mm256_add_pd(fix2,tx);
 454             fiy2             = _mm256_add_pd(fiy2,ty);
 455             fiz2             = _mm256_add_pd(fiz2,tz);
 456
 457             fjx1             = _mm256_add_pd(fjx1,tx);
 458             fjy1             = _mm256_add_pd(fjy1,ty);
 459             fjz1             = _mm256_add_pd(fjz1,tz);
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             r22              = _mm256_mul_pd(rsq22,rinv22);
 466
 467             /* EWALD ELECTROSTATICS */
 468
 469             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 470             ewrt             = _mm256_mul_pd(r22,ewtabscale);
 471             ewitab           = _mm256_cvttpd_epi32(ewrt);
 472             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 473             ewitab           = _mm_slli_epi32(ewitab,2);
 474             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 475             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 476             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 477             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 478             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 479             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 480             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 481             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
 482             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
 483
 484             /* Update potential sum for this i atom from the interaction with this j atom. */
 485             velecsum         = _mm256_add_pd(velecsum,velec);
 486
 487             fscal            = felec;
 488
 489             /* Calculate temporary vectorial force */
 490             tx               = _mm256_mul_pd(fscal,dx22);
 491             ty               = _mm256_mul_pd(fscal,dy22);
 492             tz               = _mm256_mul_pd(fscal,dz22);
 493
 494             /* Update vectorial force */
 495             fix2             = _mm256_add_pd(fix2,tx);
 496             fiy2             = _mm256_add_pd(fiy2,ty);
 497             fiz2             = _mm256_add_pd(fiz2,tz);
 498
 499             fjx2             = _mm256_add_pd(fjx2,tx);
 500             fjy2             = _mm256_add_pd(fjy2,ty);
 501             fjz2             = _mm256_add_pd(fjz2,tz);
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             r23              = _mm256_mul_pd(rsq23,rinv23);
 508
 509             /* EWALD ELECTROSTATICS */
 510
 511             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 512             ewrt             = _mm256_mul_pd(r23,ewtabscale);
 513             ewitab           = _mm256_cvttpd_epi32(ewrt);
 514             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 515             ewitab           = _mm_slli_epi32(ewitab,2);
 516             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 517             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 518             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 519             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 520             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 521             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 522             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 523             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(rinv23,velec));
 524             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
 525
 526             /* Update potential sum for this i atom from the interaction with this j atom. */
 527             velecsum         = _mm256_add_pd(velecsum,velec);
 528
 529             fscal            = felec;
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm256_mul_pd(fscal,dx23);
 533             ty               = _mm256_mul_pd(fscal,dy23);
 534             tz               = _mm256_mul_pd(fscal,dz23);
 535
 536             /* Update vectorial force */
 537             fix2             = _mm256_add_pd(fix2,tx);
 538             fiy2             = _mm256_add_pd(fiy2,ty);
 539             fiz2             = _mm256_add_pd(fiz2,tz);
 540
 541             fjx3             = _mm256_add_pd(fjx3,tx);
 542             fjy3             = _mm256_add_pd(fjy3,ty);
 543             fjz3             = _mm256_add_pd(fjz3,tz);
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             r31              = _mm256_mul_pd(rsq31,rinv31);
 550
 551             /* EWALD ELECTROSTATICS */
 552
 553             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 554             ewrt             = _mm256_mul_pd(r31,ewtabscale);
 555             ewitab           = _mm256_cvttpd_epi32(ewrt);
 556             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 557             ewitab           = _mm_slli_epi32(ewitab,2);
 558             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 559             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 560             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 561             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 562             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 563             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 564             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 565             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(rinv31,velec));
 566             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
 567
 568             /* Update potential sum for this i atom from the interaction with this j atom. */
 569             velecsum         = _mm256_add_pd(velecsum,velec);
 570
 571             fscal            = felec;
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm256_mul_pd(fscal,dx31);
 575             ty               = _mm256_mul_pd(fscal,dy31);
 576             tz               = _mm256_mul_pd(fscal,dz31);
 577
 578             /* Update vectorial force */
 579             fix3             = _mm256_add_pd(fix3,tx);
 580             fiy3             = _mm256_add_pd(fiy3,ty);
 581             fiz3             = _mm256_add_pd(fiz3,tz);
 582
 583             fjx1             = _mm256_add_pd(fjx1,tx);
 584             fjy1             = _mm256_add_pd(fjy1,ty);
 585             fjz1             = _mm256_add_pd(fjz1,tz);
 586
 587             /**************************
 588              * CALCULATE INTERACTIONS *
 589              **************************/
 590
 591             r32              = _mm256_mul_pd(rsq32,rinv32);
 592
 593             /* EWALD ELECTROSTATICS */
 594
 595             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 596             ewrt             = _mm256_mul_pd(r32,ewtabscale);
 597             ewitab           = _mm256_cvttpd_epi32(ewrt);
 598             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 599             ewitab           = _mm_slli_epi32(ewitab,2);
 600             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 601             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 602             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 603             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 604             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 605             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 606             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 607             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(rinv32,velec));
 608             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
 609
 610             /* Update potential sum for this i atom from the interaction with this j atom. */
 611             velecsum         = _mm256_add_pd(velecsum,velec);
 612
 613             fscal            = felec;
 614
 615             /* Calculate temporary vectorial force */
 616             tx               = _mm256_mul_pd(fscal,dx32);
 617             ty               = _mm256_mul_pd(fscal,dy32);
 618             tz               = _mm256_mul_pd(fscal,dz32);
 619
 620             /* Update vectorial force */
 621             fix3             = _mm256_add_pd(fix3,tx);
 622             fiy3             = _mm256_add_pd(fiy3,ty);
 623             fiz3             = _mm256_add_pd(fiz3,tz);
 624
 625             fjx2             = _mm256_add_pd(fjx2,tx);
 626             fjy2             = _mm256_add_pd(fjy2,ty);
 627             fjz2             = _mm256_add_pd(fjz2,tz);
 628
 629             /**************************
 630              * CALCULATE INTERACTIONS *
 631              **************************/
 632
 633             r33              = _mm256_mul_pd(rsq33,rinv33);
 634
 635             /* EWALD ELECTROSTATICS */
 636
 637             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 638             ewrt             = _mm256_mul_pd(r33,ewtabscale);
 639             ewitab           = _mm256_cvttpd_epi32(ewrt);
 640             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 641             ewitab           = _mm_slli_epi32(ewitab,2);
 642             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 643             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 644             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 645             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 646             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 647             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 648             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 649             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(rinv33,velec));
 650             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
 651
 652             /* Update potential sum for this i atom from the interaction with this j atom. */
 653             velecsum         = _mm256_add_pd(velecsum,velec);
 654
 655             fscal            = felec;
 656
 657             /* Calculate temporary vectorial force */
 658             tx               = _mm256_mul_pd(fscal,dx33);
 659             ty               = _mm256_mul_pd(fscal,dy33);
 660             tz               = _mm256_mul_pd(fscal,dz33);
 661
 662             /* Update vectorial force */
 663             fix3             = _mm256_add_pd(fix3,tx);
 664             fiy3             = _mm256_add_pd(fiy3,ty);
 665             fiz3             = _mm256_add_pd(fiz3,tz);
 666
 667             fjx3             = _mm256_add_pd(fjx3,tx);
 668             fjy3             = _mm256_add_pd(fjy3,ty);
 669             fjz3             = _mm256_add_pd(fjz3,tz);
 670
 671             fjptrA             = f+j_coord_offsetA;
 672             fjptrB             = f+j_coord_offsetB;
 673             fjptrC             = f+j_coord_offsetC;
 674             fjptrD             = f+j_coord_offsetD;
 675
 676             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 677                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 678
 679             /* Inner loop uses 369 flops */
 680         }
 681
 682         if(jidx<j_index_end)
 683         {
 684
 685             /* Get j neighbor index, and coordinate index */
 686             jnrlistA         = jjnr[jidx];
 687             jnrlistB         = jjnr[jidx+1];
 688             jnrlistC         = jjnr[jidx+2];
 689             jnrlistD         = jjnr[jidx+3];
 690             /* Sign of each element will be negative for non-real atoms.
 691              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 692              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 693              */
 694             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 695
 696             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 697             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 698             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 699
 700             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 701             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 702             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 703             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 704             j_coord_offsetA  = DIM*jnrA;
 705             j_coord_offsetB  = DIM*jnrB;
 706             j_coord_offsetC  = DIM*jnrC;
 707             j_coord_offsetD  = DIM*jnrD;
 708
 709             /* load j atom coordinates */
 710             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 711                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 712                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 713
 714             /* Calculate displacement vector */
 715             dx11             = _mm256_sub_pd(ix1,jx1);
 716             dy11             = _mm256_sub_pd(iy1,jy1);
 717             dz11             = _mm256_sub_pd(iz1,jz1);
 718             dx12             = _mm256_sub_pd(ix1,jx2);
 719             dy12             = _mm256_sub_pd(iy1,jy2);
 720             dz12             = _mm256_sub_pd(iz1,jz2);
 721             dx13             = _mm256_sub_pd(ix1,jx3);
 722             dy13             = _mm256_sub_pd(iy1,jy3);
 723             dz13             = _mm256_sub_pd(iz1,jz3);
 724             dx21             = _mm256_sub_pd(ix2,jx1);
 725             dy21             = _mm256_sub_pd(iy2,jy1);
 726             dz21             = _mm256_sub_pd(iz2,jz1);
 727             dx22             = _mm256_sub_pd(ix2,jx2);
 728             dy22             = _mm256_sub_pd(iy2,jy2);
 729             dz22             = _mm256_sub_pd(iz2,jz2);
 730             dx23             = _mm256_sub_pd(ix2,jx3);
 731             dy23             = _mm256_sub_pd(iy2,jy3);
 732             dz23             = _mm256_sub_pd(iz2,jz3);
 733             dx31             = _mm256_sub_pd(ix3,jx1);
 734             dy31             = _mm256_sub_pd(iy3,jy1);
 735             dz31             = _mm256_sub_pd(iz3,jz1);
 736             dx32             = _mm256_sub_pd(ix3,jx2);
 737             dy32             = _mm256_sub_pd(iy3,jy2);
 738             dz32             = _mm256_sub_pd(iz3,jz2);
 739             dx33             = _mm256_sub_pd(ix3,jx3);
 740             dy33             = _mm256_sub_pd(iy3,jy3);
 741             dz33             = _mm256_sub_pd(iz3,jz3);
 742
 743             /* Calculate squared distance and things based on it */
 744             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 745             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 746             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 747             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 748             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 749             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 750             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 751             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 752             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 753
 754             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 755             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 756             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 757             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 758             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 759             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 760             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 761             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 762             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 763
 764             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 765             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 766             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 767             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 768             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 769             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 770             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 771             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 772             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 773
 774             fjx1             = _mm256_setzero_pd();
 775             fjy1             = _mm256_setzero_pd();
 776             fjz1             = _mm256_setzero_pd();
 777             fjx2             = _mm256_setzero_pd();
 778             fjy2             = _mm256_setzero_pd();
 779             fjz2             = _mm256_setzero_pd();
 780             fjx3             = _mm256_setzero_pd();
 781             fjy3             = _mm256_setzero_pd();
 782             fjz3             = _mm256_setzero_pd();
 783
 784             /**************************
 785              * CALCULATE INTERACTIONS *
 786              **************************/
 787
 788             r11              = _mm256_mul_pd(rsq11,rinv11);
 789             r11              = _mm256_andnot_pd(dummy_mask,r11);
 790
 791             /* EWALD ELECTROSTATICS */
 792
 793             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 794             ewrt             = _mm256_mul_pd(r11,ewtabscale);
 795             ewitab           = _mm256_cvttpd_epi32(ewrt);
 796             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 797             ewitab           = _mm_slli_epi32(ewitab,2);
 798             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 799             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 800             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 801             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 802             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 803             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 804             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 805             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
 806             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
 807
 808             /* Update potential sum for this i atom from the interaction with this j atom. */
 809             velec            = _mm256_andnot_pd(dummy_mask,velec);
 810             velecsum         = _mm256_add_pd(velecsum,velec);
 811
 812             fscal            = felec;
 813
 814             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 815
 816             /* Calculate temporary vectorial force */
 817             tx               = _mm256_mul_pd(fscal,dx11);
 818             ty               = _mm256_mul_pd(fscal,dy11);
 819             tz               = _mm256_mul_pd(fscal,dz11);
 820
 821             /* Update vectorial force */
 822             fix1             = _mm256_add_pd(fix1,tx);
 823             fiy1             = _mm256_add_pd(fiy1,ty);
 824             fiz1             = _mm256_add_pd(fiz1,tz);
 825
 826             fjx1             = _mm256_add_pd(fjx1,tx);
 827             fjy1             = _mm256_add_pd(fjy1,ty);
 828             fjz1             = _mm256_add_pd(fjz1,tz);
 829
 830             /**************************
 831              * CALCULATE INTERACTIONS *
 832              **************************/
 833
 834             r12              = _mm256_mul_pd(rsq12,rinv12);
 835             r12              = _mm256_andnot_pd(dummy_mask,r12);
 836
 837             /* EWALD ELECTROSTATICS */
 838
 839             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 840             ewrt             = _mm256_mul_pd(r12,ewtabscale);
 841             ewitab           = _mm256_cvttpd_epi32(ewrt);
 842             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 843             ewitab           = _mm_slli_epi32(ewitab,2);
 844             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 845             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 846             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 847             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 848             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 849             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 850             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 851             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
 852             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
 853
 854             /* Update potential sum for this i atom from the interaction with this j atom. */
 855             velec            = _mm256_andnot_pd(dummy_mask,velec);
 856             velecsum         = _mm256_add_pd(velecsum,velec);
 857
 858             fscal            = felec;
 859
 860             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 861
 862             /* Calculate temporary vectorial force */
 863             tx               = _mm256_mul_pd(fscal,dx12);
 864             ty               = _mm256_mul_pd(fscal,dy12);
 865             tz               = _mm256_mul_pd(fscal,dz12);
 866
 867             /* Update vectorial force */
 868             fix1             = _mm256_add_pd(fix1,tx);
 869             fiy1             = _mm256_add_pd(fiy1,ty);
 870             fiz1             = _mm256_add_pd(fiz1,tz);
 871
 872             fjx2             = _mm256_add_pd(fjx2,tx);
 873             fjy2             = _mm256_add_pd(fjy2,ty);
 874             fjz2             = _mm256_add_pd(fjz2,tz);
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             r13              = _mm256_mul_pd(rsq13,rinv13);
 881             r13              = _mm256_andnot_pd(dummy_mask,r13);
 882
 883             /* EWALD ELECTROSTATICS */
 884
 885             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 886             ewrt             = _mm256_mul_pd(r13,ewtabscale);
 887             ewitab           = _mm256_cvttpd_epi32(ewrt);
 888             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 889             ewitab           = _mm_slli_epi32(ewitab,2);
 890             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 891             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 892             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 893             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 894             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 895             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 896             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 897             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(rinv13,velec));
 898             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
 899
 900             /* Update potential sum for this i atom from the interaction with this j atom. */
 901             velec            = _mm256_andnot_pd(dummy_mask,velec);
 902             velecsum         = _mm256_add_pd(velecsum,velec);
 903
 904             fscal            = felec;
 905
 906             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 907
 908             /* Calculate temporary vectorial force */
 909             tx               = _mm256_mul_pd(fscal,dx13);
 910             ty               = _mm256_mul_pd(fscal,dy13);
 911             tz               = _mm256_mul_pd(fscal,dz13);
 912
 913             /* Update vectorial force */
 914             fix1             = _mm256_add_pd(fix1,tx);
 915             fiy1             = _mm256_add_pd(fiy1,ty);
 916             fiz1             = _mm256_add_pd(fiz1,tz);
 917
 918             fjx3             = _mm256_add_pd(fjx3,tx);
 919             fjy3             = _mm256_add_pd(fjy3,ty);
 920             fjz3             = _mm256_add_pd(fjz3,tz);
 921
 922             /**************************
 923              * CALCULATE INTERACTIONS *
 924              **************************/
 925
 926             r21              = _mm256_mul_pd(rsq21,rinv21);
 927             r21              = _mm256_andnot_pd(dummy_mask,r21);
 928
 929             /* EWALD ELECTROSTATICS */
 930
 931             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 932             ewrt             = _mm256_mul_pd(r21,ewtabscale);
 933             ewitab           = _mm256_cvttpd_epi32(ewrt);
 934             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 935             ewitab           = _mm_slli_epi32(ewitab,2);
 936             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 937             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 938             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 939             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 940             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 941             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 942             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 943             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
 944             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
 945
 946             /* Update potential sum for this i atom from the interaction with this j atom. */
 947             velec            = _mm256_andnot_pd(dummy_mask,velec);
 948             velecsum         = _mm256_add_pd(velecsum,velec);
 949
 950             fscal            = felec;
 951
 952             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 953
 954             /* Calculate temporary vectorial force */
 955             tx               = _mm256_mul_pd(fscal,dx21);
 956             ty               = _mm256_mul_pd(fscal,dy21);
 957             tz               = _mm256_mul_pd(fscal,dz21);
 958
 959             /* Update vectorial force */
 960             fix2             = _mm256_add_pd(fix2,tx);
 961             fiy2             = _mm256_add_pd(fiy2,ty);
 962             fiz2             = _mm256_add_pd(fiz2,tz);
 963
 964             fjx1             = _mm256_add_pd(fjx1,tx);
 965             fjy1             = _mm256_add_pd(fjy1,ty);
 966             fjz1             = _mm256_add_pd(fjz1,tz);
 967
 968             /**************************
 969              * CALCULATE INTERACTIONS *
 970              **************************/
 971
 972             r22              = _mm256_mul_pd(rsq22,rinv22);
 973             r22              = _mm256_andnot_pd(dummy_mask,r22);
 974
 975             /* EWALD ELECTROSTATICS */
 976
 977             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 978             ewrt             = _mm256_mul_pd(r22,ewtabscale);
 979             ewitab           = _mm256_cvttpd_epi32(ewrt);
 980             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 981             ewitab           = _mm_slli_epi32(ewitab,2);
 982             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 983             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 984             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 985             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 986             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 987             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 988             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 989             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
 990             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
 991
 992             /* Update potential sum for this i atom from the interaction with this j atom. */
 993             velec            = _mm256_andnot_pd(dummy_mask,velec);
 994             velecsum         = _mm256_add_pd(velecsum,velec);
 995
 996             fscal            = felec;
 997
 998             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 999
1000             /* Calculate temporary vectorial force */
1001             tx               = _mm256_mul_pd(fscal,dx22);
1002             ty               = _mm256_mul_pd(fscal,dy22);
1003             tz               = _mm256_mul_pd(fscal,dz22);
1004
1005             /* Update vectorial force */
1006             fix2             = _mm256_add_pd(fix2,tx);
1007             fiy2             = _mm256_add_pd(fiy2,ty);
1008             fiz2             = _mm256_add_pd(fiz2,tz);
1009
1010             fjx2             = _mm256_add_pd(fjx2,tx);
1011             fjy2             = _mm256_add_pd(fjy2,ty);
1012             fjz2             = _mm256_add_pd(fjz2,tz);
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             r23              = _mm256_mul_pd(rsq23,rinv23);
1019             r23              = _mm256_andnot_pd(dummy_mask,r23);
1020
1021             /* EWALD ELECTROSTATICS */
1022
1023             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1024             ewrt             = _mm256_mul_pd(r23,ewtabscale);
1025             ewitab           = _mm256_cvttpd_epi32(ewrt);
1026             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1027             ewitab           = _mm_slli_epi32(ewitab,2);
1028             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1029             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1030             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1031             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1032             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1033             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1034             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1035             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(rinv23,velec));
1036             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
1037
1038             /* Update potential sum for this i atom from the interaction with this j atom. */
1039             velec            = _mm256_andnot_pd(dummy_mask,velec);
1040             velecsum         = _mm256_add_pd(velecsum,velec);
1041
1042             fscal            = felec;
1043
1044             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1045
1046             /* Calculate temporary vectorial force */
1047             tx               = _mm256_mul_pd(fscal,dx23);
1048             ty               = _mm256_mul_pd(fscal,dy23);
1049             tz               = _mm256_mul_pd(fscal,dz23);
1050
1051             /* Update vectorial force */
1052             fix2             = _mm256_add_pd(fix2,tx);
1053             fiy2             = _mm256_add_pd(fiy2,ty);
1054             fiz2             = _mm256_add_pd(fiz2,tz);
1055
1056             fjx3             = _mm256_add_pd(fjx3,tx);
1057             fjy3             = _mm256_add_pd(fjy3,ty);
1058             fjz3             = _mm256_add_pd(fjz3,tz);
1059
1060             /**************************
1061              * CALCULATE INTERACTIONS *
1062              **************************/
1063
1064             r31              = _mm256_mul_pd(rsq31,rinv31);
1065             r31              = _mm256_andnot_pd(dummy_mask,r31);
1066
1067             /* EWALD ELECTROSTATICS */
1068
1069             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1070             ewrt             = _mm256_mul_pd(r31,ewtabscale);
1071             ewitab           = _mm256_cvttpd_epi32(ewrt);
1072             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1073             ewitab           = _mm_slli_epi32(ewitab,2);
1074             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1075             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1076             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1077             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1078             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1079             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1080             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1081             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(rinv31,velec));
1082             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
1083
1084             /* Update potential sum for this i atom from the interaction with this j atom. */
1085             velec            = _mm256_andnot_pd(dummy_mask,velec);
1086             velecsum         = _mm256_add_pd(velecsum,velec);
1087
1088             fscal            = felec;
1089
1090             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1091
1092             /* Calculate temporary vectorial force */
1093             tx               = _mm256_mul_pd(fscal,dx31);
1094             ty               = _mm256_mul_pd(fscal,dy31);
1095             tz               = _mm256_mul_pd(fscal,dz31);
1096
1097             /* Update vectorial force */
1098             fix3             = _mm256_add_pd(fix3,tx);
1099             fiy3             = _mm256_add_pd(fiy3,ty);
1100             fiz3             = _mm256_add_pd(fiz3,tz);
1101
1102             fjx1             = _mm256_add_pd(fjx1,tx);
1103             fjy1             = _mm256_add_pd(fjy1,ty);
1104             fjz1             = _mm256_add_pd(fjz1,tz);
1105
1106             /**************************
1107              * CALCULATE INTERACTIONS *
1108              **************************/
1109
1110             r32              = _mm256_mul_pd(rsq32,rinv32);
1111             r32              = _mm256_andnot_pd(dummy_mask,r32);
1112
1113             /* EWALD ELECTROSTATICS */
1114
1115             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1116             ewrt             = _mm256_mul_pd(r32,ewtabscale);
1117             ewitab           = _mm256_cvttpd_epi32(ewrt);
1118             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1119             ewitab           = _mm_slli_epi32(ewitab,2);
1120             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1121             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1122             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1123             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1124             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1125             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1126             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1127             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(rinv32,velec));
1128             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
1129
1130             /* Update potential sum for this i atom from the interaction with this j atom. */
1131             velec            = _mm256_andnot_pd(dummy_mask,velec);
1132             velecsum         = _mm256_add_pd(velecsum,velec);
1133
1134             fscal            = felec;
1135
1136             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1137
1138             /* Calculate temporary vectorial force */
1139             tx               = _mm256_mul_pd(fscal,dx32);
1140             ty               = _mm256_mul_pd(fscal,dy32);
1141             tz               = _mm256_mul_pd(fscal,dz32);
1142
1143             /* Update vectorial force */
1144             fix3             = _mm256_add_pd(fix3,tx);
1145             fiy3             = _mm256_add_pd(fiy3,ty);
1146             fiz3             = _mm256_add_pd(fiz3,tz);
1147
1148             fjx2             = _mm256_add_pd(fjx2,tx);
1149             fjy2             = _mm256_add_pd(fjy2,ty);
1150             fjz2             = _mm256_add_pd(fjz2,tz);
1151
1152             /**************************
1153              * CALCULATE INTERACTIONS *
1154              **************************/
1155
1156             r33              = _mm256_mul_pd(rsq33,rinv33);
1157             r33              = _mm256_andnot_pd(dummy_mask,r33);
1158
1159             /* EWALD ELECTROSTATICS */
1160
1161             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1162             ewrt             = _mm256_mul_pd(r33,ewtabscale);
1163             ewitab           = _mm256_cvttpd_epi32(ewrt);
1164             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1165             ewitab           = _mm_slli_epi32(ewitab,2);
1166             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1167             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1168             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1169             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1170             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1171             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1172             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1173             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(rinv33,velec));
1174             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
1175
1176             /* Update potential sum for this i atom from the interaction with this j atom. */
1177             velec            = _mm256_andnot_pd(dummy_mask,velec);
1178             velecsum         = _mm256_add_pd(velecsum,velec);
1179
1180             fscal            = felec;
1181
1182             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1183
1184             /* Calculate temporary vectorial force */
1185             tx               = _mm256_mul_pd(fscal,dx33);
1186             ty               = _mm256_mul_pd(fscal,dy33);
1187             tz               = _mm256_mul_pd(fscal,dz33);
1188
1189             /* Update vectorial force */
1190             fix3             = _mm256_add_pd(fix3,tx);
1191             fiy3             = _mm256_add_pd(fiy3,ty);
1192             fiz3             = _mm256_add_pd(fiz3,tz);
1193
1194             fjx3             = _mm256_add_pd(fjx3,tx);
1195             fjy3             = _mm256_add_pd(fjy3,ty);
1196             fjz3             = _mm256_add_pd(fjz3,tz);
1197
1198             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1199             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1200             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1201             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1202
1203             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1204                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1205
1206             /* Inner loop uses 378 flops */
1207         }
1208
1209         /* End of innermost loop */
1210
1211         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1212                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1213
1214         ggid                        = gid[iidx];
1215         /* Update potential energies */
1216         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1217
1218         /* Increment number of inner iterations */
1219         inneriter                  += j_index_end - j_index_start;
1220
1221         /* Outer loop uses 19 flops */
1222     }
1223
1224     /* Increment number of outer iterations */
1225     outeriter        += nri;
1226
1227     /* Update outer/inner flops */
1228
1229     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*378);
1230 }
1231 /*
1232  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4W4_F_avx_256_double
1233  * Electrostatics interaction: Ewald
1234  * VdW interaction:            None
1235  * Geometry:                   Water4-Water4
1236  * Calculate force/pot:        Force
1237  */
1238 void
1239 nb_kernel_ElecEw_VdwNone_GeomW4W4_F_avx_256_double
1240                     (t_nblist                    * gmx_restrict       nlist,
1241                      rvec                        * gmx_restrict          xx,
1242                      rvec                        * gmx_restrict          ff,
1243                      t_forcerec                  * gmx_restrict          fr,
1244                      t_mdatoms                   * gmx_restrict     mdatoms,
1245                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1246                      t_nrnb                      * gmx_restrict        nrnb)
1247 {
1248     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1249      * just 0 for non-waters.
1250      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1251      * jnr indices corresponding to data put in the four positions in the SIMD register.
1252      */
1253     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1254     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1255     int              jnrA,jnrB,jnrC,jnrD;
1256     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1257     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1258     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1259     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1260     real             rcutoff_scalar;
1261     real             *shiftvec,*fshift,*x,*f;
1262     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1263     real             scratch[4*DIM];
1264     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1265     real *           vdwioffsetptr1;
1266     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1267     real *           vdwioffsetptr2;
1268     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1269     real *           vdwioffsetptr3;
1270     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1271     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1272     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1273     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1274     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1275     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1276     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1277     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1278     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1279     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1280     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1281     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1282     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1283     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1284     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1285     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1286     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1287     real             *charge;
1288     __m128i          ewitab;
1289     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1290     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1291     real             *ewtab;
1292     __m256d          dummy_mask,cutoff_mask;
1293     __m128           tmpmask0,tmpmask1;
1294     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1295     __m256d          one     = _mm256_set1_pd(1.0);
1296     __m256d          two     = _mm256_set1_pd(2.0);
1297     x                = xx[0];
1298     f                = ff[0];
1299
1300     nri              = nlist->nri;
1301     iinr             = nlist->iinr;
1302     jindex           = nlist->jindex;
1303     jjnr             = nlist->jjnr;
1304     shiftidx         = nlist->shift;
1305     gid              = nlist->gid;
1306     shiftvec         = fr->shift_vec[0];
1307     fshift           = fr->fshift[0];
1308     facel            = _mm256_set1_pd(fr->epsfac);
1309     charge           = mdatoms->chargeA;
1310
1311     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
1312     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
1313     beta2            = _mm256_mul_pd(beta,beta);
1314     beta3            = _mm256_mul_pd(beta,beta2);
1315
1316     ewtab            = fr->ic->tabq_coul_F;
1317     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
1318     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
1319
1320     /* Setup water-specific parameters */
1321     inr              = nlist->iinr[0];
1322     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1323     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1324     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1325
1326     jq1              = _mm256_set1_pd(charge[inr+1]);
1327     jq2              = _mm256_set1_pd(charge[inr+2]);
1328     jq3              = _mm256_set1_pd(charge[inr+3]);
1329     qq11             = _mm256_mul_pd(iq1,jq1);
1330     qq12             = _mm256_mul_pd(iq1,jq2);
1331     qq13             = _mm256_mul_pd(iq1,jq3);
1332     qq21             = _mm256_mul_pd(iq2,jq1);
1333     qq22             = _mm256_mul_pd(iq2,jq2);
1334     qq23             = _mm256_mul_pd(iq2,jq3);
1335     qq31             = _mm256_mul_pd(iq3,jq1);
1336     qq32             = _mm256_mul_pd(iq3,jq2);
1337     qq33             = _mm256_mul_pd(iq3,jq3);
1338
1339     /* Avoid stupid compiler warnings */
1340     jnrA = jnrB = jnrC = jnrD = 0;
1341     j_coord_offsetA = 0;
1342     j_coord_offsetB = 0;
1343     j_coord_offsetC = 0;
1344     j_coord_offsetD = 0;
1345
1346     outeriter        = 0;
1347     inneriter        = 0;
1348
1349     for(iidx=0;iidx<4*DIM;iidx++)
1350     {
1351         scratch[iidx] = 0.0;
1352     }
1353
1354     /* Start outer loop over neighborlists */
1355     for(iidx=0; iidx<nri; iidx++)
1356     {
1357         /* Load shift vector for this list */
1358         i_shift_offset   = DIM*shiftidx[iidx];
1359
1360         /* Load limits for loop over neighbors */
1361         j_index_start    = jindex[iidx];
1362         j_index_end      = jindex[iidx+1];
1363
1364         /* Get outer coordinate index */
1365         inr              = iinr[iidx];
1366         i_coord_offset   = DIM*inr;
1367
1368         /* Load i particle coords and add shift vector */
1369         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1370                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1371
1372         fix1             = _mm256_setzero_pd();
1373         fiy1             = _mm256_setzero_pd();
1374         fiz1             = _mm256_setzero_pd();
1375         fix2             = _mm256_setzero_pd();
1376         fiy2             = _mm256_setzero_pd();
1377         fiz2             = _mm256_setzero_pd();
1378         fix3             = _mm256_setzero_pd();
1379         fiy3             = _mm256_setzero_pd();
1380         fiz3             = _mm256_setzero_pd();
1381
1382         /* Start inner kernel loop */
1383         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1384         {
1385
1386             /* Get j neighbor index, and coordinate index */
1387             jnrA             = jjnr[jidx];
1388             jnrB             = jjnr[jidx+1];
1389             jnrC             = jjnr[jidx+2];
1390             jnrD             = jjnr[jidx+3];
1391             j_coord_offsetA  = DIM*jnrA;
1392             j_coord_offsetB  = DIM*jnrB;
1393             j_coord_offsetC  = DIM*jnrC;
1394             j_coord_offsetD  = DIM*jnrD;
1395
1396             /* load j atom coordinates */
1397             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1398                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1399                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1400
1401             /* Calculate displacement vector */
1402             dx11             = _mm256_sub_pd(ix1,jx1);
1403             dy11             = _mm256_sub_pd(iy1,jy1);
1404             dz11             = _mm256_sub_pd(iz1,jz1);
1405             dx12             = _mm256_sub_pd(ix1,jx2);
1406             dy12             = _mm256_sub_pd(iy1,jy2);
1407             dz12             = _mm256_sub_pd(iz1,jz2);
1408             dx13             = _mm256_sub_pd(ix1,jx3);
1409             dy13             = _mm256_sub_pd(iy1,jy3);
1410             dz13             = _mm256_sub_pd(iz1,jz3);
1411             dx21             = _mm256_sub_pd(ix2,jx1);
1412             dy21             = _mm256_sub_pd(iy2,jy1);
1413             dz21             = _mm256_sub_pd(iz2,jz1);
1414             dx22             = _mm256_sub_pd(ix2,jx2);
1415             dy22             = _mm256_sub_pd(iy2,jy2);
1416             dz22             = _mm256_sub_pd(iz2,jz2);
1417             dx23             = _mm256_sub_pd(ix2,jx3);
1418             dy23             = _mm256_sub_pd(iy2,jy3);
1419             dz23             = _mm256_sub_pd(iz2,jz3);
1420             dx31             = _mm256_sub_pd(ix3,jx1);
1421             dy31             = _mm256_sub_pd(iy3,jy1);
1422             dz31             = _mm256_sub_pd(iz3,jz1);
1423             dx32             = _mm256_sub_pd(ix3,jx2);
1424             dy32             = _mm256_sub_pd(iy3,jy2);
1425             dz32             = _mm256_sub_pd(iz3,jz2);
1426             dx33             = _mm256_sub_pd(ix3,jx3);
1427             dy33             = _mm256_sub_pd(iy3,jy3);
1428             dz33             = _mm256_sub_pd(iz3,jz3);
1429
1430             /* Calculate squared distance and things based on it */
1431             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1432             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1433             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1434             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1435             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1436             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1437             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1438             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1439             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1440
1441             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1442             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1443             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1444             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1445             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1446             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1447             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1448             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1449             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1450
1451             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1452             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1453             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1454             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1455             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1456             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1457             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1458             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1459             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1460
1461             fjx1             = _mm256_setzero_pd();
1462             fjy1             = _mm256_setzero_pd();
1463             fjz1             = _mm256_setzero_pd();
1464             fjx2             = _mm256_setzero_pd();
1465             fjy2             = _mm256_setzero_pd();
1466             fjz2             = _mm256_setzero_pd();
1467             fjx3             = _mm256_setzero_pd();
1468             fjy3             = _mm256_setzero_pd();
1469             fjz3             = _mm256_setzero_pd();
1470
1471             /**************************
1472              * CALCULATE INTERACTIONS *
1473              **************************/
1474
1475             r11              = _mm256_mul_pd(rsq11,rinv11);
1476
1477             /* EWALD ELECTROSTATICS */
1478
1479             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1480             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1481             ewitab           = _mm256_cvttpd_epi32(ewrt);
1482             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1483             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1484                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1485                                             &ewtabF,&ewtabFn);
1486             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1487             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1488
1489             fscal            = felec;
1490
1491             /* Calculate temporary vectorial force */
1492             tx               = _mm256_mul_pd(fscal,dx11);
1493             ty               = _mm256_mul_pd(fscal,dy11);
1494             tz               = _mm256_mul_pd(fscal,dz11);
1495
1496             /* Update vectorial force */
1497             fix1             = _mm256_add_pd(fix1,tx);
1498             fiy1             = _mm256_add_pd(fiy1,ty);
1499             fiz1             = _mm256_add_pd(fiz1,tz);
1500
1501             fjx1             = _mm256_add_pd(fjx1,tx);
1502             fjy1             = _mm256_add_pd(fjy1,ty);
1503             fjz1             = _mm256_add_pd(fjz1,tz);
1504
1505             /**************************
1506              * CALCULATE INTERACTIONS *
1507              **************************/
1508
1509             r12              = _mm256_mul_pd(rsq12,rinv12);
1510
1511             /* EWALD ELECTROSTATICS */
1512
1513             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1514             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1515             ewitab           = _mm256_cvttpd_epi32(ewrt);
1516             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1517             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1518                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1519                                             &ewtabF,&ewtabFn);
1520             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1521             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1522
1523             fscal            = felec;
1524
1525             /* Calculate temporary vectorial force */
1526             tx               = _mm256_mul_pd(fscal,dx12);
1527             ty               = _mm256_mul_pd(fscal,dy12);
1528             tz               = _mm256_mul_pd(fscal,dz12);
1529
1530             /* Update vectorial force */
1531             fix1             = _mm256_add_pd(fix1,tx);
1532             fiy1             = _mm256_add_pd(fiy1,ty);
1533             fiz1             = _mm256_add_pd(fiz1,tz);
1534
1535             fjx2             = _mm256_add_pd(fjx2,tx);
1536             fjy2             = _mm256_add_pd(fjy2,ty);
1537             fjz2             = _mm256_add_pd(fjz2,tz);
1538
1539             /**************************
1540              * CALCULATE INTERACTIONS *
1541              **************************/
1542
1543             r13              = _mm256_mul_pd(rsq13,rinv13);
1544
1545             /* EWALD ELECTROSTATICS */
1546
1547             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1548             ewrt             = _mm256_mul_pd(r13,ewtabscale);
1549             ewitab           = _mm256_cvttpd_epi32(ewrt);
1550             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1551             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1552                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1553                                             &ewtabF,&ewtabFn);
1554             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1555             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
1556
1557             fscal            = felec;
1558
1559             /* Calculate temporary vectorial force */
1560             tx               = _mm256_mul_pd(fscal,dx13);
1561             ty               = _mm256_mul_pd(fscal,dy13);
1562             tz               = _mm256_mul_pd(fscal,dz13);
1563
1564             /* Update vectorial force */
1565             fix1             = _mm256_add_pd(fix1,tx);
1566             fiy1             = _mm256_add_pd(fiy1,ty);
1567             fiz1             = _mm256_add_pd(fiz1,tz);
1568
1569             fjx3             = _mm256_add_pd(fjx3,tx);
1570             fjy3             = _mm256_add_pd(fjy3,ty);
1571             fjz3             = _mm256_add_pd(fjz3,tz);
1572
1573             /**************************
1574              * CALCULATE INTERACTIONS *
1575              **************************/
1576
1577             r21              = _mm256_mul_pd(rsq21,rinv21);
1578
1579             /* EWALD ELECTROSTATICS */
1580
1581             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1582             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1583             ewitab           = _mm256_cvttpd_epi32(ewrt);
1584             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1585             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1586                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1587                                             &ewtabF,&ewtabFn);
1588             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1589             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1590
1591             fscal            = felec;
1592
1593             /* Calculate temporary vectorial force */
1594             tx               = _mm256_mul_pd(fscal,dx21);
1595             ty               = _mm256_mul_pd(fscal,dy21);
1596             tz               = _mm256_mul_pd(fscal,dz21);
1597
1598             /* Update vectorial force */
1599             fix2             = _mm256_add_pd(fix2,tx);
1600             fiy2             = _mm256_add_pd(fiy2,ty);
1601             fiz2             = _mm256_add_pd(fiz2,tz);
1602
1603             fjx1             = _mm256_add_pd(fjx1,tx);
1604             fjy1             = _mm256_add_pd(fjy1,ty);
1605             fjz1             = _mm256_add_pd(fjz1,tz);
1606
1607             /**************************
1608              * CALCULATE INTERACTIONS *
1609              **************************/
1610
1611             r22              = _mm256_mul_pd(rsq22,rinv22);
1612
1613             /* EWALD ELECTROSTATICS */
1614
1615             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1616             ewrt             = _mm256_mul_pd(r22,ewtabscale);
1617             ewitab           = _mm256_cvttpd_epi32(ewrt);
1618             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1619             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1620                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1621                                             &ewtabF,&ewtabFn);
1622             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1623             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1624
1625             fscal            = felec;
1626
1627             /* Calculate temporary vectorial force */
1628             tx               = _mm256_mul_pd(fscal,dx22);
1629             ty               = _mm256_mul_pd(fscal,dy22);
1630             tz               = _mm256_mul_pd(fscal,dz22);
1631
1632             /* Update vectorial force */
1633             fix2             = _mm256_add_pd(fix2,tx);
1634             fiy2             = _mm256_add_pd(fiy2,ty);
1635             fiz2             = _mm256_add_pd(fiz2,tz);
1636
1637             fjx2             = _mm256_add_pd(fjx2,tx);
1638             fjy2             = _mm256_add_pd(fjy2,ty);
1639             fjz2             = _mm256_add_pd(fjz2,tz);
1640
1641             /**************************
1642              * CALCULATE INTERACTIONS *
1643              **************************/
1644
1645             r23              = _mm256_mul_pd(rsq23,rinv23);
1646
1647             /* EWALD ELECTROSTATICS */
1648
1649             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1650             ewrt             = _mm256_mul_pd(r23,ewtabscale);
1651             ewitab           = _mm256_cvttpd_epi32(ewrt);
1652             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1653             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1654                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1655                                             &ewtabF,&ewtabFn);
1656             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1657             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
1658
1659             fscal            = felec;
1660
1661             /* Calculate temporary vectorial force */
1662             tx               = _mm256_mul_pd(fscal,dx23);
1663             ty               = _mm256_mul_pd(fscal,dy23);
1664             tz               = _mm256_mul_pd(fscal,dz23);
1665
1666             /* Update vectorial force */
1667             fix2             = _mm256_add_pd(fix2,tx);
1668             fiy2             = _mm256_add_pd(fiy2,ty);
1669             fiz2             = _mm256_add_pd(fiz2,tz);
1670
1671             fjx3             = _mm256_add_pd(fjx3,tx);
1672             fjy3             = _mm256_add_pd(fjy3,ty);
1673             fjz3             = _mm256_add_pd(fjz3,tz);
1674
1675             /**************************
1676              * CALCULATE INTERACTIONS *
1677              **************************/
1678
1679             r31              = _mm256_mul_pd(rsq31,rinv31);
1680
1681             /* EWALD ELECTROSTATICS */
1682
1683             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1684             ewrt             = _mm256_mul_pd(r31,ewtabscale);
1685             ewitab           = _mm256_cvttpd_epi32(ewrt);
1686             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1687             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1688                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1689                                             &ewtabF,&ewtabFn);
1690             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1691             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
1692
1693             fscal            = felec;
1694
1695             /* Calculate temporary vectorial force */
1696             tx               = _mm256_mul_pd(fscal,dx31);
1697             ty               = _mm256_mul_pd(fscal,dy31);
1698             tz               = _mm256_mul_pd(fscal,dz31);
1699
1700             /* Update vectorial force */
1701             fix3             = _mm256_add_pd(fix3,tx);
1702             fiy3             = _mm256_add_pd(fiy3,ty);
1703             fiz3             = _mm256_add_pd(fiz3,tz);
1704
1705             fjx1             = _mm256_add_pd(fjx1,tx);
1706             fjy1             = _mm256_add_pd(fjy1,ty);
1707             fjz1             = _mm256_add_pd(fjz1,tz);
1708
1709             /**************************
1710              * CALCULATE INTERACTIONS *
1711              **************************/
1712
1713             r32              = _mm256_mul_pd(rsq32,rinv32);
1714
1715             /* EWALD ELECTROSTATICS */
1716
1717             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1718             ewrt             = _mm256_mul_pd(r32,ewtabscale);
1719             ewitab           = _mm256_cvttpd_epi32(ewrt);
1720             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1721             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1722                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1723                                             &ewtabF,&ewtabFn);
1724             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1725             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
1726
1727             fscal            = felec;
1728
1729             /* Calculate temporary vectorial force */
1730             tx               = _mm256_mul_pd(fscal,dx32);
1731             ty               = _mm256_mul_pd(fscal,dy32);
1732             tz               = _mm256_mul_pd(fscal,dz32);
1733
1734             /* Update vectorial force */
1735             fix3             = _mm256_add_pd(fix3,tx);
1736             fiy3             = _mm256_add_pd(fiy3,ty);
1737             fiz3             = _mm256_add_pd(fiz3,tz);
1738
1739             fjx2             = _mm256_add_pd(fjx2,tx);
1740             fjy2             = _mm256_add_pd(fjy2,ty);
1741             fjz2             = _mm256_add_pd(fjz2,tz);
1742
1743             /**************************
1744              * CALCULATE INTERACTIONS *
1745              **************************/
1746
1747             r33              = _mm256_mul_pd(rsq33,rinv33);
1748
1749             /* EWALD ELECTROSTATICS */
1750
1751             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1752             ewrt             = _mm256_mul_pd(r33,ewtabscale);
1753             ewitab           = _mm256_cvttpd_epi32(ewrt);
1754             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1755             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1756                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1757                                             &ewtabF,&ewtabFn);
1758             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1759             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
1760
1761             fscal            = felec;
1762
1763             /* Calculate temporary vectorial force */
1764             tx               = _mm256_mul_pd(fscal,dx33);
1765             ty               = _mm256_mul_pd(fscal,dy33);
1766             tz               = _mm256_mul_pd(fscal,dz33);
1767
1768             /* Update vectorial force */
1769             fix3             = _mm256_add_pd(fix3,tx);
1770             fiy3             = _mm256_add_pd(fiy3,ty);
1771             fiz3             = _mm256_add_pd(fiz3,tz);
1772
1773             fjx3             = _mm256_add_pd(fjx3,tx);
1774             fjy3             = _mm256_add_pd(fjy3,ty);
1775             fjz3             = _mm256_add_pd(fjz3,tz);
1776
1777             fjptrA             = f+j_coord_offsetA;
1778             fjptrB             = f+j_coord_offsetB;
1779             fjptrC             = f+j_coord_offsetC;
1780             fjptrD             = f+j_coord_offsetD;
1781
1782             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1783                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1784
1785             /* Inner loop uses 324 flops */
1786         }
1787
1788         if(jidx<j_index_end)
1789         {
1790
1791             /* Get j neighbor index, and coordinate index */
1792             jnrlistA         = jjnr[jidx];
1793             jnrlistB         = jjnr[jidx+1];
1794             jnrlistC         = jjnr[jidx+2];
1795             jnrlistD         = jjnr[jidx+3];
1796             /* Sign of each element will be negative for non-real atoms.
1797              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1798              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1799              */
1800             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1801
1802             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1803             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1804             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1805
1806             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1807             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1808             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1809             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1810             j_coord_offsetA  = DIM*jnrA;
1811             j_coord_offsetB  = DIM*jnrB;
1812             j_coord_offsetC  = DIM*jnrC;
1813             j_coord_offsetD  = DIM*jnrD;
1814
1815             /* load j atom coordinates */
1816             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1817                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1818                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1819
1820             /* Calculate displacement vector */
1821             dx11             = _mm256_sub_pd(ix1,jx1);
1822             dy11             = _mm256_sub_pd(iy1,jy1);
1823             dz11             = _mm256_sub_pd(iz1,jz1);
1824             dx12             = _mm256_sub_pd(ix1,jx2);
1825             dy12             = _mm256_sub_pd(iy1,jy2);
1826             dz12             = _mm256_sub_pd(iz1,jz2);
1827             dx13             = _mm256_sub_pd(ix1,jx3);
1828             dy13             = _mm256_sub_pd(iy1,jy3);
1829             dz13             = _mm256_sub_pd(iz1,jz3);
1830             dx21             = _mm256_sub_pd(ix2,jx1);
1831             dy21             = _mm256_sub_pd(iy2,jy1);
1832             dz21             = _mm256_sub_pd(iz2,jz1);
1833             dx22             = _mm256_sub_pd(ix2,jx2);
1834             dy22             = _mm256_sub_pd(iy2,jy2);
1835             dz22             = _mm256_sub_pd(iz2,jz2);
1836             dx23             = _mm256_sub_pd(ix2,jx3);
1837             dy23             = _mm256_sub_pd(iy2,jy3);
1838             dz23             = _mm256_sub_pd(iz2,jz3);
1839             dx31             = _mm256_sub_pd(ix3,jx1);
1840             dy31             = _mm256_sub_pd(iy3,jy1);
1841             dz31             = _mm256_sub_pd(iz3,jz1);
1842             dx32             = _mm256_sub_pd(ix3,jx2);
1843             dy32             = _mm256_sub_pd(iy3,jy2);
1844             dz32             = _mm256_sub_pd(iz3,jz2);
1845             dx33             = _mm256_sub_pd(ix3,jx3);
1846             dy33             = _mm256_sub_pd(iy3,jy3);
1847             dz33             = _mm256_sub_pd(iz3,jz3);
1848
1849             /* Calculate squared distance and things based on it */
1850             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1851             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1852             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1853             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1854             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1855             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1856             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1857             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1858             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1859
1860             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1861             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1862             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1863             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1864             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1865             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1866             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1867             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1868             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1869
1870             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1871             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1872             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1873             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1874             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1875             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1876             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1877             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1878             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1879
1880             fjx1             = _mm256_setzero_pd();
1881             fjy1             = _mm256_setzero_pd();
1882             fjz1             = _mm256_setzero_pd();
1883             fjx2             = _mm256_setzero_pd();
1884             fjy2             = _mm256_setzero_pd();
1885             fjz2             = _mm256_setzero_pd();
1886             fjx3             = _mm256_setzero_pd();
1887             fjy3             = _mm256_setzero_pd();
1888             fjz3             = _mm256_setzero_pd();
1889
1890             /**************************
1891              * CALCULATE INTERACTIONS *
1892              **************************/
1893
1894             r11              = _mm256_mul_pd(rsq11,rinv11);
1895             r11              = _mm256_andnot_pd(dummy_mask,r11);
1896
1897             /* EWALD ELECTROSTATICS */
1898
1899             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1900             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1901             ewitab           = _mm256_cvttpd_epi32(ewrt);
1902             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1903             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1904                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1905                                             &ewtabF,&ewtabFn);
1906             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1907             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1908
1909             fscal            = felec;
1910
1911             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1912
1913             /* Calculate temporary vectorial force */
1914             tx               = _mm256_mul_pd(fscal,dx11);
1915             ty               = _mm256_mul_pd(fscal,dy11);
1916             tz               = _mm256_mul_pd(fscal,dz11);
1917
1918             /* Update vectorial force */
1919             fix1             = _mm256_add_pd(fix1,tx);
1920             fiy1             = _mm256_add_pd(fiy1,ty);
1921             fiz1             = _mm256_add_pd(fiz1,tz);
1922
1923             fjx1             = _mm256_add_pd(fjx1,tx);
1924             fjy1             = _mm256_add_pd(fjy1,ty);
1925             fjz1             = _mm256_add_pd(fjz1,tz);
1926
1927             /**************************
1928              * CALCULATE INTERACTIONS *
1929              **************************/
1930
1931             r12              = _mm256_mul_pd(rsq12,rinv12);
1932             r12              = _mm256_andnot_pd(dummy_mask,r12);
1933
1934             /* EWALD ELECTROSTATICS */
1935
1936             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1937             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1938             ewitab           = _mm256_cvttpd_epi32(ewrt);
1939             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1940             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1941                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1942                                             &ewtabF,&ewtabFn);
1943             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1944             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1945
1946             fscal            = felec;
1947
1948             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1949
1950             /* Calculate temporary vectorial force */
1951             tx               = _mm256_mul_pd(fscal,dx12);
1952             ty               = _mm256_mul_pd(fscal,dy12);
1953             tz               = _mm256_mul_pd(fscal,dz12);
1954
1955             /* Update vectorial force */
1956             fix1             = _mm256_add_pd(fix1,tx);
1957             fiy1             = _mm256_add_pd(fiy1,ty);
1958             fiz1             = _mm256_add_pd(fiz1,tz);
1959
1960             fjx2             = _mm256_add_pd(fjx2,tx);
1961             fjy2             = _mm256_add_pd(fjy2,ty);
1962             fjz2             = _mm256_add_pd(fjz2,tz);
1963
1964             /**************************
1965              * CALCULATE INTERACTIONS *
1966              **************************/
1967
1968             r13              = _mm256_mul_pd(rsq13,rinv13);
1969             r13              = _mm256_andnot_pd(dummy_mask,r13);
1970
1971             /* EWALD ELECTROSTATICS */
1972
1973             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1974             ewrt             = _mm256_mul_pd(r13,ewtabscale);
1975             ewitab           = _mm256_cvttpd_epi32(ewrt);
1976             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1977             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1978                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1979                                             &ewtabF,&ewtabFn);
1980             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1981             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
1982
1983             fscal            = felec;
1984
1985             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1986
1987             /* Calculate temporary vectorial force */
1988             tx               = _mm256_mul_pd(fscal,dx13);
1989             ty               = _mm256_mul_pd(fscal,dy13);
1990             tz               = _mm256_mul_pd(fscal,dz13);
1991
1992             /* Update vectorial force */
1993             fix1             = _mm256_add_pd(fix1,tx);
1994             fiy1             = _mm256_add_pd(fiy1,ty);
1995             fiz1             = _mm256_add_pd(fiz1,tz);
1996
1997             fjx3             = _mm256_add_pd(fjx3,tx);
1998             fjy3             = _mm256_add_pd(fjy3,ty);
1999             fjz3             = _mm256_add_pd(fjz3,tz);
2000
2001             /**************************
2002              * CALCULATE INTERACTIONS *
2003              **************************/
2004
2005             r21              = _mm256_mul_pd(rsq21,rinv21);
2006             r21              = _mm256_andnot_pd(dummy_mask,r21);
2007
2008             /* EWALD ELECTROSTATICS */
2009
2010             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2011             ewrt             = _mm256_mul_pd(r21,ewtabscale);
2012             ewitab           = _mm256_cvttpd_epi32(ewrt);
2013             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2014             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2015                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2016                                             &ewtabF,&ewtabFn);
2017             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2018             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2019
2020             fscal            = felec;
2021
2022             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2023
2024             /* Calculate temporary vectorial force */
2025             tx               = _mm256_mul_pd(fscal,dx21);
2026             ty               = _mm256_mul_pd(fscal,dy21);
2027             tz               = _mm256_mul_pd(fscal,dz21);
2028
2029             /* Update vectorial force */
2030             fix2             = _mm256_add_pd(fix2,tx);
2031             fiy2             = _mm256_add_pd(fiy2,ty);
2032             fiz2             = _mm256_add_pd(fiz2,tz);
2033
2034             fjx1             = _mm256_add_pd(fjx1,tx);
2035             fjy1             = _mm256_add_pd(fjy1,ty);
2036             fjz1             = _mm256_add_pd(fjz1,tz);
2037
2038             /**************************
2039              * CALCULATE INTERACTIONS *
2040              **************************/
2041
2042             r22              = _mm256_mul_pd(rsq22,rinv22);
2043             r22              = _mm256_andnot_pd(dummy_mask,r22);
2044
2045             /* EWALD ELECTROSTATICS */
2046
2047             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2048             ewrt             = _mm256_mul_pd(r22,ewtabscale);
2049             ewitab           = _mm256_cvttpd_epi32(ewrt);
2050             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2051             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2052                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2053                                             &ewtabF,&ewtabFn);
2054             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2055             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2056
2057             fscal            = felec;
2058
2059             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2060
2061             /* Calculate temporary vectorial force */
2062             tx               = _mm256_mul_pd(fscal,dx22);
2063             ty               = _mm256_mul_pd(fscal,dy22);
2064             tz               = _mm256_mul_pd(fscal,dz22);
2065
2066             /* Update vectorial force */
2067             fix2             = _mm256_add_pd(fix2,tx);
2068             fiy2             = _mm256_add_pd(fiy2,ty);
2069             fiz2             = _mm256_add_pd(fiz2,tz);
2070
2071             fjx2             = _mm256_add_pd(fjx2,tx);
2072             fjy2             = _mm256_add_pd(fjy2,ty);
2073             fjz2             = _mm256_add_pd(fjz2,tz);
2074
2075             /**************************
2076              * CALCULATE INTERACTIONS *
2077              **************************/
2078
2079             r23              = _mm256_mul_pd(rsq23,rinv23);
2080             r23              = _mm256_andnot_pd(dummy_mask,r23);
2081
2082             /* EWALD ELECTROSTATICS */
2083
2084             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2085             ewrt             = _mm256_mul_pd(r23,ewtabscale);
2086             ewitab           = _mm256_cvttpd_epi32(ewrt);
2087             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2088             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2089                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2090                                             &ewtabF,&ewtabFn);
2091             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2092             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
2093
2094             fscal            = felec;
2095
2096             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2097
2098             /* Calculate temporary vectorial force */
2099             tx               = _mm256_mul_pd(fscal,dx23);
2100             ty               = _mm256_mul_pd(fscal,dy23);
2101             tz               = _mm256_mul_pd(fscal,dz23);
2102
2103             /* Update vectorial force */
2104             fix2             = _mm256_add_pd(fix2,tx);
2105             fiy2             = _mm256_add_pd(fiy2,ty);
2106             fiz2             = _mm256_add_pd(fiz2,tz);
2107
2108             fjx3             = _mm256_add_pd(fjx3,tx);
2109             fjy3             = _mm256_add_pd(fjy3,ty);
2110             fjz3             = _mm256_add_pd(fjz3,tz);
2111
2112             /**************************
2113              * CALCULATE INTERACTIONS *
2114              **************************/
2115
2116             r31              = _mm256_mul_pd(rsq31,rinv31);
2117             r31              = _mm256_andnot_pd(dummy_mask,r31);
2118
2119             /* EWALD ELECTROSTATICS */
2120
2121             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2122             ewrt             = _mm256_mul_pd(r31,ewtabscale);
2123             ewitab           = _mm256_cvttpd_epi32(ewrt);
2124             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2125             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2126                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2127                                             &ewtabF,&ewtabFn);
2128             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2129             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
2130
2131             fscal            = felec;
2132
2133             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2134
2135             /* Calculate temporary vectorial force */
2136             tx               = _mm256_mul_pd(fscal,dx31);
2137             ty               = _mm256_mul_pd(fscal,dy31);
2138             tz               = _mm256_mul_pd(fscal,dz31);
2139
2140             /* Update vectorial force */
2141             fix3             = _mm256_add_pd(fix3,tx);
2142             fiy3             = _mm256_add_pd(fiy3,ty);
2143             fiz3             = _mm256_add_pd(fiz3,tz);
2144
2145             fjx1             = _mm256_add_pd(fjx1,tx);
2146             fjy1             = _mm256_add_pd(fjy1,ty);
2147             fjz1             = _mm256_add_pd(fjz1,tz);
2148
2149             /**************************
2150              * CALCULATE INTERACTIONS *
2151              **************************/
2152
2153             r32              = _mm256_mul_pd(rsq32,rinv32);
2154             r32              = _mm256_andnot_pd(dummy_mask,r32);
2155
2156             /* EWALD ELECTROSTATICS */
2157
2158             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2159             ewrt             = _mm256_mul_pd(r32,ewtabscale);
2160             ewitab           = _mm256_cvttpd_epi32(ewrt);
2161             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2162             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2163                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2164                                             &ewtabF,&ewtabFn);
2165             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2166             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
2167
2168             fscal            = felec;
2169
2170             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2171
2172             /* Calculate temporary vectorial force */
2173             tx               = _mm256_mul_pd(fscal,dx32);
2174             ty               = _mm256_mul_pd(fscal,dy32);
2175             tz               = _mm256_mul_pd(fscal,dz32);
2176
2177             /* Update vectorial force */
2178             fix3             = _mm256_add_pd(fix3,tx);
2179             fiy3             = _mm256_add_pd(fiy3,ty);
2180             fiz3             = _mm256_add_pd(fiz3,tz);
2181
2182             fjx2             = _mm256_add_pd(fjx2,tx);
2183             fjy2             = _mm256_add_pd(fjy2,ty);
2184             fjz2             = _mm256_add_pd(fjz2,tz);
2185
2186             /**************************
2187              * CALCULATE INTERACTIONS *
2188              **************************/
2189
2190             r33              = _mm256_mul_pd(rsq33,rinv33);
2191             r33              = _mm256_andnot_pd(dummy_mask,r33);
2192
2193             /* EWALD ELECTROSTATICS */
2194
2195             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2196             ewrt             = _mm256_mul_pd(r33,ewtabscale);
2197             ewitab           = _mm256_cvttpd_epi32(ewrt);
2198             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2199             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2200                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2201                                             &ewtabF,&ewtabFn);
2202             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2203             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
2204
2205             fscal            = felec;
2206
2207             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2208
2209             /* Calculate temporary vectorial force */
2210             tx               = _mm256_mul_pd(fscal,dx33);
2211             ty               = _mm256_mul_pd(fscal,dy33);
2212             tz               = _mm256_mul_pd(fscal,dz33);
2213
2214             /* Update vectorial force */
2215             fix3             = _mm256_add_pd(fix3,tx);
2216             fiy3             = _mm256_add_pd(fiy3,ty);
2217             fiz3             = _mm256_add_pd(fiz3,tz);
2218
2219             fjx3             = _mm256_add_pd(fjx3,tx);
2220             fjy3             = _mm256_add_pd(fjy3,ty);
2221             fjz3             = _mm256_add_pd(fjz3,tz);
2222
2223             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2224             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2225             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2226             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2227
2228             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2229                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2230
2231             /* Inner loop uses 333 flops */
2232         }
2233
2234         /* End of innermost loop */
2235
2236         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2237                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
2238
2239         /* Increment number of inner iterations */
2240         inneriter                  += j_index_end - j_index_start;
2241
2242         /* Outer loop uses 18 flops */
2243     }
2244
2245     /* Increment number of outer iterations */
2246     outeriter        += nri;
2247
2248     /* Update outer/inner flops */
2249
2250     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*333);
2251 }