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

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4W4_VF_avx_256_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Water4
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwNone_GeomW4W4_VF_avx_256_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr1;
  71     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  72     real *           vdwioffsetptr2;
  73     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  74     real *           vdwioffsetptr3;
  75     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  76     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  77     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  78     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  79     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  80     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  81     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  82     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  83     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  84     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  85     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  86     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  87     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  88     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  89     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
  90     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
  91     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m128i          ewitab;
  94     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  95     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  96     real             *ewtab;
  97     __m256d          dummy_mask,cutoff_mask;
  98     __m128           tmpmask0,tmpmask1;
  99     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 100     __m256d          one     = _mm256_set1_pd(1.0);
 101     __m256d          two     = _mm256_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm256_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115
 116     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 117     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 118     beta2            = _mm256_mul_pd(beta,beta);
 119     beta3            = _mm256_mul_pd(beta,beta2);
 120
 121     ewtab            = fr->ic->tabq_coul_FDV0;
 122     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 123     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 124
 125     /* Setup water-specific parameters */
 126     inr              = nlist->iinr[0];
 127     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 128     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 129     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 130
 131     jq1              = _mm256_set1_pd(charge[inr+1]);
 132     jq2              = _mm256_set1_pd(charge[inr+2]);
 133     jq3              = _mm256_set1_pd(charge[inr+3]);
 134     qq11             = _mm256_mul_pd(iq1,jq1);
 135     qq12             = _mm256_mul_pd(iq1,jq2);
 136     qq13             = _mm256_mul_pd(iq1,jq3);
 137     qq21             = _mm256_mul_pd(iq2,jq1);
 138     qq22             = _mm256_mul_pd(iq2,jq2);
 139     qq23             = _mm256_mul_pd(iq2,jq3);
 140     qq31             = _mm256_mul_pd(iq3,jq1);
 141     qq32             = _mm256_mul_pd(iq3,jq2);
 142     qq33             = _mm256_mul_pd(iq3,jq3);
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 175                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 176
 177         fix1             = _mm256_setzero_pd();
 178         fiy1             = _mm256_setzero_pd();
 179         fiz1             = _mm256_setzero_pd();
 180         fix2             = _mm256_setzero_pd();
 181         fiy2             = _mm256_setzero_pd();
 182         fiz2             = _mm256_setzero_pd();
 183         fix3             = _mm256_setzero_pd();
 184         fiy3             = _mm256_setzero_pd();
 185         fiz3             = _mm256_setzero_pd();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm256_setzero_pd();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203
 204             /* load j atom coordinates */
 205             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 206                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 207                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 208
 209             /* Calculate displacement vector */
 210             dx11             = _mm256_sub_pd(ix1,jx1);
 211             dy11             = _mm256_sub_pd(iy1,jy1);
 212             dz11             = _mm256_sub_pd(iz1,jz1);
 213             dx12             = _mm256_sub_pd(ix1,jx2);
 214             dy12             = _mm256_sub_pd(iy1,jy2);
 215             dz12             = _mm256_sub_pd(iz1,jz2);
 216             dx13             = _mm256_sub_pd(ix1,jx3);
 217             dy13             = _mm256_sub_pd(iy1,jy3);
 218             dz13             = _mm256_sub_pd(iz1,jz3);
 219             dx21             = _mm256_sub_pd(ix2,jx1);
 220             dy21             = _mm256_sub_pd(iy2,jy1);
 221             dz21             = _mm256_sub_pd(iz2,jz1);
 222             dx22             = _mm256_sub_pd(ix2,jx2);
 223             dy22             = _mm256_sub_pd(iy2,jy2);
 224             dz22             = _mm256_sub_pd(iz2,jz2);
 225             dx23             = _mm256_sub_pd(ix2,jx3);
 226             dy23             = _mm256_sub_pd(iy2,jy3);
 227             dz23             = _mm256_sub_pd(iz2,jz3);
 228             dx31             = _mm256_sub_pd(ix3,jx1);
 229             dy31             = _mm256_sub_pd(iy3,jy1);
 230             dz31             = _mm256_sub_pd(iz3,jz1);
 231             dx32             = _mm256_sub_pd(ix3,jx2);
 232             dy32             = _mm256_sub_pd(iy3,jy2);
 233             dz32             = _mm256_sub_pd(iz3,jz2);
 234             dx33             = _mm256_sub_pd(ix3,jx3);
 235             dy33             = _mm256_sub_pd(iy3,jy3);
 236             dz33             = _mm256_sub_pd(iz3,jz3);
 237
 238             /* Calculate squared distance and things based on it */
 239             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 240             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 241             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 242             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 243             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 244             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 245             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 246             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 247             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 248
 249             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 250             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 251             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 252             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 253             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 254             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 255             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 256             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 257             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 258
 259             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 260             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 261             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 262             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 263             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 264             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 265             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 266             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 267             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 268
 269             fjx1             = _mm256_setzero_pd();
 270             fjy1             = _mm256_setzero_pd();
 271             fjz1             = _mm256_setzero_pd();
 272             fjx2             = _mm256_setzero_pd();
 273             fjy2             = _mm256_setzero_pd();
 274             fjz2             = _mm256_setzero_pd();
 275             fjx3             = _mm256_setzero_pd();
 276             fjy3             = _mm256_setzero_pd();
 277             fjz3             = _mm256_setzero_pd();
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             r11              = _mm256_mul_pd(rsq11,rinv11);
 284
 285             /* EWALD ELECTROSTATICS */
 286
 287             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 288             ewrt             = _mm256_mul_pd(r11,ewtabscale);
 289             ewitab           = _mm256_cvttpd_epi32(ewrt);
 290             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 291             ewitab           = _mm_slli_epi32(ewitab,2);
 292             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 293             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 294             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 295             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 296             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 297             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 298             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 299             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
 300             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             velecsum         = _mm256_add_pd(velecsum,velec);
 304
 305             fscal            = felec;
 306
 307             /* Calculate temporary vectorial force */
 308             tx               = _mm256_mul_pd(fscal,dx11);
 309             ty               = _mm256_mul_pd(fscal,dy11);
 310             tz               = _mm256_mul_pd(fscal,dz11);
 311
 312             /* Update vectorial force */
 313             fix1             = _mm256_add_pd(fix1,tx);
 314             fiy1             = _mm256_add_pd(fiy1,ty);
 315             fiz1             = _mm256_add_pd(fiz1,tz);
 316
 317             fjx1             = _mm256_add_pd(fjx1,tx);
 318             fjy1             = _mm256_add_pd(fjy1,ty);
 319             fjz1             = _mm256_add_pd(fjz1,tz);
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             r12              = _mm256_mul_pd(rsq12,rinv12);
 326
 327             /* EWALD ELECTROSTATICS */
 328
 329             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 330             ewrt             = _mm256_mul_pd(r12,ewtabscale);
 331             ewitab           = _mm256_cvttpd_epi32(ewrt);
 332             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 333             ewitab           = _mm_slli_epi32(ewitab,2);
 334             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 335             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 336             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 337             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 338             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 339             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 340             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 341             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
 342             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm256_add_pd(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm256_mul_pd(fscal,dx12);
 351             ty               = _mm256_mul_pd(fscal,dy12);
 352             tz               = _mm256_mul_pd(fscal,dz12);
 353
 354             /* Update vectorial force */
 355             fix1             = _mm256_add_pd(fix1,tx);
 356             fiy1             = _mm256_add_pd(fiy1,ty);
 357             fiz1             = _mm256_add_pd(fiz1,tz);
 358
 359             fjx2             = _mm256_add_pd(fjx2,tx);
 360             fjy2             = _mm256_add_pd(fjy2,ty);
 361             fjz2             = _mm256_add_pd(fjz2,tz);
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             r13              = _mm256_mul_pd(rsq13,rinv13);
 368
 369             /* EWALD ELECTROSTATICS */
 370
 371             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 372             ewrt             = _mm256_mul_pd(r13,ewtabscale);
 373             ewitab           = _mm256_cvttpd_epi32(ewrt);
 374             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 375             ewitab           = _mm_slli_epi32(ewitab,2);
 376             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 377             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 378             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 379             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 380             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 381             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 382             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 383             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(rinv13,velec));
 384             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velecsum         = _mm256_add_pd(velecsum,velec);
 388
 389             fscal            = felec;
 390
 391             /* Calculate temporary vectorial force */
 392             tx               = _mm256_mul_pd(fscal,dx13);
 393             ty               = _mm256_mul_pd(fscal,dy13);
 394             tz               = _mm256_mul_pd(fscal,dz13);
 395
 396             /* Update vectorial force */
 397             fix1             = _mm256_add_pd(fix1,tx);
 398             fiy1             = _mm256_add_pd(fiy1,ty);
 399             fiz1             = _mm256_add_pd(fiz1,tz);
 400
 401             fjx3             = _mm256_add_pd(fjx3,tx);
 402             fjy3             = _mm256_add_pd(fjy3,ty);
 403             fjz3             = _mm256_add_pd(fjz3,tz);
 404
 405             /**************************
 406              * CALCULATE INTERACTIONS *
 407              **************************/
 408
 409             r21              = _mm256_mul_pd(rsq21,rinv21);
 410
 411             /* EWALD ELECTROSTATICS */
 412
 413             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 414             ewrt             = _mm256_mul_pd(r21,ewtabscale);
 415             ewitab           = _mm256_cvttpd_epi32(ewrt);
 416             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 417             ewitab           = _mm_slli_epi32(ewitab,2);
 418             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 419             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 420             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 421             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 422             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 423             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 424             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 425             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
 426             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
 427
 428             /* Update potential sum for this i atom from the interaction with this j atom. */
 429             velecsum         = _mm256_add_pd(velecsum,velec);
 430
 431             fscal            = felec;
 432
 433             /* Calculate temporary vectorial force */
 434             tx               = _mm256_mul_pd(fscal,dx21);
 435             ty               = _mm256_mul_pd(fscal,dy21);
 436             tz               = _mm256_mul_pd(fscal,dz21);
 437
 438             /* Update vectorial force */
 439             fix2             = _mm256_add_pd(fix2,tx);
 440             fiy2             = _mm256_add_pd(fiy2,ty);
 441             fiz2             = _mm256_add_pd(fiz2,tz);
 442
 443             fjx1             = _mm256_add_pd(fjx1,tx);
 444             fjy1             = _mm256_add_pd(fjy1,ty);
 445             fjz1             = _mm256_add_pd(fjz1,tz);
 446
 447             /**************************
 448              * CALCULATE INTERACTIONS *
 449              **************************/
 450
 451             r22              = _mm256_mul_pd(rsq22,rinv22);
 452
 453             /* EWALD ELECTROSTATICS */
 454
 455             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 456             ewrt             = _mm256_mul_pd(r22,ewtabscale);
 457             ewitab           = _mm256_cvttpd_epi32(ewrt);
 458             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 459             ewitab           = _mm_slli_epi32(ewitab,2);
 460             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 461             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 462             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 463             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 464             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 465             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 466             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 467             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
 468             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
 469
 470             /* Update potential sum for this i atom from the interaction with this j atom. */
 471             velecsum         = _mm256_add_pd(velecsum,velec);
 472
 473             fscal            = felec;
 474
 475             /* Calculate temporary vectorial force */
 476             tx               = _mm256_mul_pd(fscal,dx22);
 477             ty               = _mm256_mul_pd(fscal,dy22);
 478             tz               = _mm256_mul_pd(fscal,dz22);
 479
 480             /* Update vectorial force */
 481             fix2             = _mm256_add_pd(fix2,tx);
 482             fiy2             = _mm256_add_pd(fiy2,ty);
 483             fiz2             = _mm256_add_pd(fiz2,tz);
 484
 485             fjx2             = _mm256_add_pd(fjx2,tx);
 486             fjy2             = _mm256_add_pd(fjy2,ty);
 487             fjz2             = _mm256_add_pd(fjz2,tz);
 488
 489             /**************************
 490              * CALCULATE INTERACTIONS *
 491              **************************/
 492
 493             r23              = _mm256_mul_pd(rsq23,rinv23);
 494
 495             /* EWALD ELECTROSTATICS */
 496
 497             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 498             ewrt             = _mm256_mul_pd(r23,ewtabscale);
 499             ewitab           = _mm256_cvttpd_epi32(ewrt);
 500             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 501             ewitab           = _mm_slli_epi32(ewitab,2);
 502             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 503             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 504             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 505             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 506             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 507             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 508             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 509             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(rinv23,velec));
 510             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
 511
 512             /* Update potential sum for this i atom from the interaction with this j atom. */
 513             velecsum         = _mm256_add_pd(velecsum,velec);
 514
 515             fscal            = felec;
 516
 517             /* Calculate temporary vectorial force */
 518             tx               = _mm256_mul_pd(fscal,dx23);
 519             ty               = _mm256_mul_pd(fscal,dy23);
 520             tz               = _mm256_mul_pd(fscal,dz23);
 521
 522             /* Update vectorial force */
 523             fix2             = _mm256_add_pd(fix2,tx);
 524             fiy2             = _mm256_add_pd(fiy2,ty);
 525             fiz2             = _mm256_add_pd(fiz2,tz);
 526
 527             fjx3             = _mm256_add_pd(fjx3,tx);
 528             fjy3             = _mm256_add_pd(fjy3,ty);
 529             fjz3             = _mm256_add_pd(fjz3,tz);
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             r31              = _mm256_mul_pd(rsq31,rinv31);
 536
 537             /* EWALD ELECTROSTATICS */
 538
 539             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 540             ewrt             = _mm256_mul_pd(r31,ewtabscale);
 541             ewitab           = _mm256_cvttpd_epi32(ewrt);
 542             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 543             ewitab           = _mm_slli_epi32(ewitab,2);
 544             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 545             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 546             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 547             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 548             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 549             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 550             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 551             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(rinv31,velec));
 552             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velecsum         = _mm256_add_pd(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm256_mul_pd(fscal,dx31);
 561             ty               = _mm256_mul_pd(fscal,dy31);
 562             tz               = _mm256_mul_pd(fscal,dz31);
 563
 564             /* Update vectorial force */
 565             fix3             = _mm256_add_pd(fix3,tx);
 566             fiy3             = _mm256_add_pd(fiy3,ty);
 567             fiz3             = _mm256_add_pd(fiz3,tz);
 568
 569             fjx1             = _mm256_add_pd(fjx1,tx);
 570             fjy1             = _mm256_add_pd(fjy1,ty);
 571             fjz1             = _mm256_add_pd(fjz1,tz);
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             r32              = _mm256_mul_pd(rsq32,rinv32);
 578
 579             /* EWALD ELECTROSTATICS */
 580
 581             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 582             ewrt             = _mm256_mul_pd(r32,ewtabscale);
 583             ewitab           = _mm256_cvttpd_epi32(ewrt);
 584             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 585             ewitab           = _mm_slli_epi32(ewitab,2);
 586             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 587             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 588             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 589             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 590             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 591             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 592             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 593             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(rinv32,velec));
 594             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
 595
 596             /* Update potential sum for this i atom from the interaction with this j atom. */
 597             velecsum         = _mm256_add_pd(velecsum,velec);
 598
 599             fscal            = felec;
 600
 601             /* Calculate temporary vectorial force */
 602             tx               = _mm256_mul_pd(fscal,dx32);
 603             ty               = _mm256_mul_pd(fscal,dy32);
 604             tz               = _mm256_mul_pd(fscal,dz32);
 605
 606             /* Update vectorial force */
 607             fix3             = _mm256_add_pd(fix3,tx);
 608             fiy3             = _mm256_add_pd(fiy3,ty);
 609             fiz3             = _mm256_add_pd(fiz3,tz);
 610
 611             fjx2             = _mm256_add_pd(fjx2,tx);
 612             fjy2             = _mm256_add_pd(fjy2,ty);
 613             fjz2             = _mm256_add_pd(fjz2,tz);
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             r33              = _mm256_mul_pd(rsq33,rinv33);
 620
 621             /* EWALD ELECTROSTATICS */
 622
 623             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 624             ewrt             = _mm256_mul_pd(r33,ewtabscale);
 625             ewitab           = _mm256_cvttpd_epi32(ewrt);
 626             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 627             ewitab           = _mm_slli_epi32(ewitab,2);
 628             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 629             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 630             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 631             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 632             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 633             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 634             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 635             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(rinv33,velec));
 636             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
 637
 638             /* Update potential sum for this i atom from the interaction with this j atom. */
 639             velecsum         = _mm256_add_pd(velecsum,velec);
 640
 641             fscal            = felec;
 642
 643             /* Calculate temporary vectorial force */
 644             tx               = _mm256_mul_pd(fscal,dx33);
 645             ty               = _mm256_mul_pd(fscal,dy33);
 646             tz               = _mm256_mul_pd(fscal,dz33);
 647
 648             /* Update vectorial force */
 649             fix3             = _mm256_add_pd(fix3,tx);
 650             fiy3             = _mm256_add_pd(fiy3,ty);
 651             fiz3             = _mm256_add_pd(fiz3,tz);
 652
 653             fjx3             = _mm256_add_pd(fjx3,tx);
 654             fjy3             = _mm256_add_pd(fjy3,ty);
 655             fjz3             = _mm256_add_pd(fjz3,tz);
 656
 657             fjptrA             = f+j_coord_offsetA;
 658             fjptrB             = f+j_coord_offsetB;
 659             fjptrC             = f+j_coord_offsetC;
 660             fjptrD             = f+j_coord_offsetD;
 661
 662             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 663                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 664
 665             /* Inner loop uses 369 flops */
 666         }
 667
 668         if(jidx<j_index_end)
 669         {
 670
 671             /* Get j neighbor index, and coordinate index */
 672             jnrlistA         = jjnr[jidx];
 673             jnrlistB         = jjnr[jidx+1];
 674             jnrlistC         = jjnr[jidx+2];
 675             jnrlistD         = jjnr[jidx+3];
 676             /* Sign of each element will be negative for non-real atoms.
 677              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 678              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 679              */
 680             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 681
 682             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 683             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 684             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 685
 686             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 687             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 688             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 689             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 690             j_coord_offsetA  = DIM*jnrA;
 691             j_coord_offsetB  = DIM*jnrB;
 692             j_coord_offsetC  = DIM*jnrC;
 693             j_coord_offsetD  = DIM*jnrD;
 694
 695             /* load j atom coordinates */
 696             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 697                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 698                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 699
 700             /* Calculate displacement vector */
 701             dx11             = _mm256_sub_pd(ix1,jx1);
 702             dy11             = _mm256_sub_pd(iy1,jy1);
 703             dz11             = _mm256_sub_pd(iz1,jz1);
 704             dx12             = _mm256_sub_pd(ix1,jx2);
 705             dy12             = _mm256_sub_pd(iy1,jy2);
 706             dz12             = _mm256_sub_pd(iz1,jz2);
 707             dx13             = _mm256_sub_pd(ix1,jx3);
 708             dy13             = _mm256_sub_pd(iy1,jy3);
 709             dz13             = _mm256_sub_pd(iz1,jz3);
 710             dx21             = _mm256_sub_pd(ix2,jx1);
 711             dy21             = _mm256_sub_pd(iy2,jy1);
 712             dz21             = _mm256_sub_pd(iz2,jz1);
 713             dx22             = _mm256_sub_pd(ix2,jx2);
 714             dy22             = _mm256_sub_pd(iy2,jy2);
 715             dz22             = _mm256_sub_pd(iz2,jz2);
 716             dx23             = _mm256_sub_pd(ix2,jx3);
 717             dy23             = _mm256_sub_pd(iy2,jy3);
 718             dz23             = _mm256_sub_pd(iz2,jz3);
 719             dx31             = _mm256_sub_pd(ix3,jx1);
 720             dy31             = _mm256_sub_pd(iy3,jy1);
 721             dz31             = _mm256_sub_pd(iz3,jz1);
 722             dx32             = _mm256_sub_pd(ix3,jx2);
 723             dy32             = _mm256_sub_pd(iy3,jy2);
 724             dz32             = _mm256_sub_pd(iz3,jz2);
 725             dx33             = _mm256_sub_pd(ix3,jx3);
 726             dy33             = _mm256_sub_pd(iy3,jy3);
 727             dz33             = _mm256_sub_pd(iz3,jz3);
 728
 729             /* Calculate squared distance and things based on it */
 730             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 731             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 732             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 733             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 734             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 735             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 736             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 737             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 738             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 739
 740             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 741             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 742             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 743             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 744             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 745             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 746             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 747             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 748             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 749
 750             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 751             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 752             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 753             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 754             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 755             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 756             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 757             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 758             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 759
 760             fjx1             = _mm256_setzero_pd();
 761             fjy1             = _mm256_setzero_pd();
 762             fjz1             = _mm256_setzero_pd();
 763             fjx2             = _mm256_setzero_pd();
 764             fjy2             = _mm256_setzero_pd();
 765             fjz2             = _mm256_setzero_pd();
 766             fjx3             = _mm256_setzero_pd();
 767             fjy3             = _mm256_setzero_pd();
 768             fjz3             = _mm256_setzero_pd();
 769
 770             /**************************
 771              * CALCULATE INTERACTIONS *
 772              **************************/
 773
 774             r11              = _mm256_mul_pd(rsq11,rinv11);
 775             r11              = _mm256_andnot_pd(dummy_mask,r11);
 776
 777             /* EWALD ELECTROSTATICS */
 778
 779             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 780             ewrt             = _mm256_mul_pd(r11,ewtabscale);
 781             ewitab           = _mm256_cvttpd_epi32(ewrt);
 782             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 783             ewitab           = _mm_slli_epi32(ewitab,2);
 784             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 785             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 786             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 787             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 788             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 789             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 790             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 791             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
 792             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
 793
 794             /* Update potential sum for this i atom from the interaction with this j atom. */
 795             velec            = _mm256_andnot_pd(dummy_mask,velec);
 796             velecsum         = _mm256_add_pd(velecsum,velec);
 797
 798             fscal            = felec;
 799
 800             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 801
 802             /* Calculate temporary vectorial force */
 803             tx               = _mm256_mul_pd(fscal,dx11);
 804             ty               = _mm256_mul_pd(fscal,dy11);
 805             tz               = _mm256_mul_pd(fscal,dz11);
 806
 807             /* Update vectorial force */
 808             fix1             = _mm256_add_pd(fix1,tx);
 809             fiy1             = _mm256_add_pd(fiy1,ty);
 810             fiz1             = _mm256_add_pd(fiz1,tz);
 811
 812             fjx1             = _mm256_add_pd(fjx1,tx);
 813             fjy1             = _mm256_add_pd(fjy1,ty);
 814             fjz1             = _mm256_add_pd(fjz1,tz);
 815
 816             /**************************
 817              * CALCULATE INTERACTIONS *
 818              **************************/
 819
 820             r12              = _mm256_mul_pd(rsq12,rinv12);
 821             r12              = _mm256_andnot_pd(dummy_mask,r12);
 822
 823             /* EWALD ELECTROSTATICS */
 824
 825             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 826             ewrt             = _mm256_mul_pd(r12,ewtabscale);
 827             ewitab           = _mm256_cvttpd_epi32(ewrt);
 828             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 829             ewitab           = _mm_slli_epi32(ewitab,2);
 830             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 831             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 832             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 833             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 834             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 835             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 836             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 837             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
 838             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
 839
 840             /* Update potential sum for this i atom from the interaction with this j atom. */
 841             velec            = _mm256_andnot_pd(dummy_mask,velec);
 842             velecsum         = _mm256_add_pd(velecsum,velec);
 843
 844             fscal            = felec;
 845
 846             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 847
 848             /* Calculate temporary vectorial force */
 849             tx               = _mm256_mul_pd(fscal,dx12);
 850             ty               = _mm256_mul_pd(fscal,dy12);
 851             tz               = _mm256_mul_pd(fscal,dz12);
 852
 853             /* Update vectorial force */
 854             fix1             = _mm256_add_pd(fix1,tx);
 855             fiy1             = _mm256_add_pd(fiy1,ty);
 856             fiz1             = _mm256_add_pd(fiz1,tz);
 857
 858             fjx2             = _mm256_add_pd(fjx2,tx);
 859             fjy2             = _mm256_add_pd(fjy2,ty);
 860             fjz2             = _mm256_add_pd(fjz2,tz);
 861
 862             /**************************
 863              * CALCULATE INTERACTIONS *
 864              **************************/
 865
 866             r13              = _mm256_mul_pd(rsq13,rinv13);
 867             r13              = _mm256_andnot_pd(dummy_mask,r13);
 868
 869             /* EWALD ELECTROSTATICS */
 870
 871             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 872             ewrt             = _mm256_mul_pd(r13,ewtabscale);
 873             ewitab           = _mm256_cvttpd_epi32(ewrt);
 874             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 875             ewitab           = _mm_slli_epi32(ewitab,2);
 876             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 877             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 878             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 879             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 880             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 881             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 882             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 883             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(rinv13,velec));
 884             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
 885
 886             /* Update potential sum for this i atom from the interaction with this j atom. */
 887             velec            = _mm256_andnot_pd(dummy_mask,velec);
 888             velecsum         = _mm256_add_pd(velecsum,velec);
 889
 890             fscal            = felec;
 891
 892             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 893
 894             /* Calculate temporary vectorial force */
 895             tx               = _mm256_mul_pd(fscal,dx13);
 896             ty               = _mm256_mul_pd(fscal,dy13);
 897             tz               = _mm256_mul_pd(fscal,dz13);
 898
 899             /* Update vectorial force */
 900             fix1             = _mm256_add_pd(fix1,tx);
 901             fiy1             = _mm256_add_pd(fiy1,ty);
 902             fiz1             = _mm256_add_pd(fiz1,tz);
 903
 904             fjx3             = _mm256_add_pd(fjx3,tx);
 905             fjy3             = _mm256_add_pd(fjy3,ty);
 906             fjz3             = _mm256_add_pd(fjz3,tz);
 907
 908             /**************************
 909              * CALCULATE INTERACTIONS *
 910              **************************/
 911
 912             r21              = _mm256_mul_pd(rsq21,rinv21);
 913             r21              = _mm256_andnot_pd(dummy_mask,r21);
 914
 915             /* EWALD ELECTROSTATICS */
 916
 917             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 918             ewrt             = _mm256_mul_pd(r21,ewtabscale);
 919             ewitab           = _mm256_cvttpd_epi32(ewrt);
 920             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 921             ewitab           = _mm_slli_epi32(ewitab,2);
 922             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 923             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 924             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 925             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 926             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 927             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 928             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 929             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
 930             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
 931
 932             /* Update potential sum for this i atom from the interaction with this j atom. */
 933             velec            = _mm256_andnot_pd(dummy_mask,velec);
 934             velecsum         = _mm256_add_pd(velecsum,velec);
 935
 936             fscal            = felec;
 937
 938             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 939
 940             /* Calculate temporary vectorial force */
 941             tx               = _mm256_mul_pd(fscal,dx21);
 942             ty               = _mm256_mul_pd(fscal,dy21);
 943             tz               = _mm256_mul_pd(fscal,dz21);
 944
 945             /* Update vectorial force */
 946             fix2             = _mm256_add_pd(fix2,tx);
 947             fiy2             = _mm256_add_pd(fiy2,ty);
 948             fiz2             = _mm256_add_pd(fiz2,tz);
 949
 950             fjx1             = _mm256_add_pd(fjx1,tx);
 951             fjy1             = _mm256_add_pd(fjy1,ty);
 952             fjz1             = _mm256_add_pd(fjz1,tz);
 953
 954             /**************************
 955              * CALCULATE INTERACTIONS *
 956              **************************/
 957
 958             r22              = _mm256_mul_pd(rsq22,rinv22);
 959             r22              = _mm256_andnot_pd(dummy_mask,r22);
 960
 961             /* EWALD ELECTROSTATICS */
 962
 963             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 964             ewrt             = _mm256_mul_pd(r22,ewtabscale);
 965             ewitab           = _mm256_cvttpd_epi32(ewrt);
 966             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 967             ewitab           = _mm_slli_epi32(ewitab,2);
 968             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 969             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 970             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 971             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 972             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 973             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 974             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 975             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
 976             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
 977
 978             /* Update potential sum for this i atom from the interaction with this j atom. */
 979             velec            = _mm256_andnot_pd(dummy_mask,velec);
 980             velecsum         = _mm256_add_pd(velecsum,velec);
 981
 982             fscal            = felec;
 983
 984             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 985
 986             /* Calculate temporary vectorial force */
 987             tx               = _mm256_mul_pd(fscal,dx22);
 988             ty               = _mm256_mul_pd(fscal,dy22);
 989             tz               = _mm256_mul_pd(fscal,dz22);
 990
 991             /* Update vectorial force */
 992             fix2             = _mm256_add_pd(fix2,tx);
 993             fiy2             = _mm256_add_pd(fiy2,ty);
 994             fiz2             = _mm256_add_pd(fiz2,tz);
 995
 996             fjx2             = _mm256_add_pd(fjx2,tx);
 997             fjy2             = _mm256_add_pd(fjy2,ty);
 998             fjz2             = _mm256_add_pd(fjz2,tz);
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             r23              = _mm256_mul_pd(rsq23,rinv23);
1005             r23              = _mm256_andnot_pd(dummy_mask,r23);
1006
1007             /* EWALD ELECTROSTATICS */
1008
1009             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1010             ewrt             = _mm256_mul_pd(r23,ewtabscale);
1011             ewitab           = _mm256_cvttpd_epi32(ewrt);
1012             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1013             ewitab           = _mm_slli_epi32(ewitab,2);
1014             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1015             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1016             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1017             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1018             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1019             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1020             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1021             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(rinv23,velec));
1022             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
1023
1024             /* Update potential sum for this i atom from the interaction with this j atom. */
1025             velec            = _mm256_andnot_pd(dummy_mask,velec);
1026             velecsum         = _mm256_add_pd(velecsum,velec);
1027
1028             fscal            = felec;
1029
1030             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1031
1032             /* Calculate temporary vectorial force */
1033             tx               = _mm256_mul_pd(fscal,dx23);
1034             ty               = _mm256_mul_pd(fscal,dy23);
1035             tz               = _mm256_mul_pd(fscal,dz23);
1036
1037             /* Update vectorial force */
1038             fix2             = _mm256_add_pd(fix2,tx);
1039             fiy2             = _mm256_add_pd(fiy2,ty);
1040             fiz2             = _mm256_add_pd(fiz2,tz);
1041
1042             fjx3             = _mm256_add_pd(fjx3,tx);
1043             fjy3             = _mm256_add_pd(fjy3,ty);
1044             fjz3             = _mm256_add_pd(fjz3,tz);
1045
1046             /**************************
1047              * CALCULATE INTERACTIONS *
1048              **************************/
1049
1050             r31              = _mm256_mul_pd(rsq31,rinv31);
1051             r31              = _mm256_andnot_pd(dummy_mask,r31);
1052
1053             /* EWALD ELECTROSTATICS */
1054
1055             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1056             ewrt             = _mm256_mul_pd(r31,ewtabscale);
1057             ewitab           = _mm256_cvttpd_epi32(ewrt);
1058             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1059             ewitab           = _mm_slli_epi32(ewitab,2);
1060             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1061             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1062             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1063             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1064             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1065             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1066             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1067             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(rinv31,velec));
1068             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
1069
1070             /* Update potential sum for this i atom from the interaction with this j atom. */
1071             velec            = _mm256_andnot_pd(dummy_mask,velec);
1072             velecsum         = _mm256_add_pd(velecsum,velec);
1073
1074             fscal            = felec;
1075
1076             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1077
1078             /* Calculate temporary vectorial force */
1079             tx               = _mm256_mul_pd(fscal,dx31);
1080             ty               = _mm256_mul_pd(fscal,dy31);
1081             tz               = _mm256_mul_pd(fscal,dz31);
1082
1083             /* Update vectorial force */
1084             fix3             = _mm256_add_pd(fix3,tx);
1085             fiy3             = _mm256_add_pd(fiy3,ty);
1086             fiz3             = _mm256_add_pd(fiz3,tz);
1087
1088             fjx1             = _mm256_add_pd(fjx1,tx);
1089             fjy1             = _mm256_add_pd(fjy1,ty);
1090             fjz1             = _mm256_add_pd(fjz1,tz);
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             r32              = _mm256_mul_pd(rsq32,rinv32);
1097             r32              = _mm256_andnot_pd(dummy_mask,r32);
1098
1099             /* EWALD ELECTROSTATICS */
1100
1101             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1102             ewrt             = _mm256_mul_pd(r32,ewtabscale);
1103             ewitab           = _mm256_cvttpd_epi32(ewrt);
1104             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1105             ewitab           = _mm_slli_epi32(ewitab,2);
1106             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1107             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1108             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1109             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1110             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1111             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1112             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1113             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(rinv32,velec));
1114             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
1115
1116             /* Update potential sum for this i atom from the interaction with this j atom. */
1117             velec            = _mm256_andnot_pd(dummy_mask,velec);
1118             velecsum         = _mm256_add_pd(velecsum,velec);
1119
1120             fscal            = felec;
1121
1122             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1123
1124             /* Calculate temporary vectorial force */
1125             tx               = _mm256_mul_pd(fscal,dx32);
1126             ty               = _mm256_mul_pd(fscal,dy32);
1127             tz               = _mm256_mul_pd(fscal,dz32);
1128
1129             /* Update vectorial force */
1130             fix3             = _mm256_add_pd(fix3,tx);
1131             fiy3             = _mm256_add_pd(fiy3,ty);
1132             fiz3             = _mm256_add_pd(fiz3,tz);
1133
1134             fjx2             = _mm256_add_pd(fjx2,tx);
1135             fjy2             = _mm256_add_pd(fjy2,ty);
1136             fjz2             = _mm256_add_pd(fjz2,tz);
1137
1138             /**************************
1139              * CALCULATE INTERACTIONS *
1140              **************************/
1141
1142             r33              = _mm256_mul_pd(rsq33,rinv33);
1143             r33              = _mm256_andnot_pd(dummy_mask,r33);
1144
1145             /* EWALD ELECTROSTATICS */
1146
1147             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1148             ewrt             = _mm256_mul_pd(r33,ewtabscale);
1149             ewitab           = _mm256_cvttpd_epi32(ewrt);
1150             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1151             ewitab           = _mm_slli_epi32(ewitab,2);
1152             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1153             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1154             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1155             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1156             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1157             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1158             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1159             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(rinv33,velec));
1160             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
1161
1162             /* Update potential sum for this i atom from the interaction with this j atom. */
1163             velec            = _mm256_andnot_pd(dummy_mask,velec);
1164             velecsum         = _mm256_add_pd(velecsum,velec);
1165
1166             fscal            = felec;
1167
1168             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1169
1170             /* Calculate temporary vectorial force */
1171             tx               = _mm256_mul_pd(fscal,dx33);
1172             ty               = _mm256_mul_pd(fscal,dy33);
1173             tz               = _mm256_mul_pd(fscal,dz33);
1174
1175             /* Update vectorial force */
1176             fix3             = _mm256_add_pd(fix3,tx);
1177             fiy3             = _mm256_add_pd(fiy3,ty);
1178             fiz3             = _mm256_add_pd(fiz3,tz);
1179
1180             fjx3             = _mm256_add_pd(fjx3,tx);
1181             fjy3             = _mm256_add_pd(fjy3,ty);
1182             fjz3             = _mm256_add_pd(fjz3,tz);
1183
1184             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1185             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1186             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1187             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1188
1189             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1190                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1191
1192             /* Inner loop uses 378 flops */
1193         }
1194
1195         /* End of innermost loop */
1196
1197         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1198                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1199
1200         ggid                        = gid[iidx];
1201         /* Update potential energies */
1202         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1203
1204         /* Increment number of inner iterations */
1205         inneriter                  += j_index_end - j_index_start;
1206
1207         /* Outer loop uses 19 flops */
1208     }
1209
1210     /* Increment number of outer iterations */
1211     outeriter        += nri;
1212
1213     /* Update outer/inner flops */
1214
1215     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*378);
1216 }
1217 /*
1218  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4W4_F_avx_256_double
1219  * Electrostatics interaction: Ewald
1220  * VdW interaction:            None
1221  * Geometry:                   Water4-Water4
1222  * Calculate force/pot:        Force
1223  */
1224 void
1225 nb_kernel_ElecEw_VdwNone_GeomW4W4_F_avx_256_double
1226                     (t_nblist * gmx_restrict                nlist,
1227                      rvec * gmx_restrict                    xx,
1228                      rvec * gmx_restrict                    ff,
1229                      t_forcerec * gmx_restrict              fr,
1230                      t_mdatoms * gmx_restrict               mdatoms,
1231                      nb_kernel_data_t * gmx_restrict        kernel_data,
1232                      t_nrnb * gmx_restrict                  nrnb)
1233 {
1234     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1235      * just 0 for non-waters.
1236      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1237      * jnr indices corresponding to data put in the four positions in the SIMD register.
1238      */
1239     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1240     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1241     int              jnrA,jnrB,jnrC,jnrD;
1242     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1243     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1244     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1245     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1246     real             rcutoff_scalar;
1247     real             *shiftvec,*fshift,*x,*f;
1248     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1249     real             scratch[4*DIM];
1250     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1251     real *           vdwioffsetptr1;
1252     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1253     real *           vdwioffsetptr2;
1254     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1255     real *           vdwioffsetptr3;
1256     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1257     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1258     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1259     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1260     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1261     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1262     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1263     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1264     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1265     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1266     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1267     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1268     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1269     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1270     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1271     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1272     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1273     real             *charge;
1274     __m128i          ewitab;
1275     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1276     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1277     real             *ewtab;
1278     __m256d          dummy_mask,cutoff_mask;
1279     __m128           tmpmask0,tmpmask1;
1280     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1281     __m256d          one     = _mm256_set1_pd(1.0);
1282     __m256d          two     = _mm256_set1_pd(2.0);
1283     x                = xx[0];
1284     f                = ff[0];
1285
1286     nri              = nlist->nri;
1287     iinr             = nlist->iinr;
1288     jindex           = nlist->jindex;
1289     jjnr             = nlist->jjnr;
1290     shiftidx         = nlist->shift;
1291     gid              = nlist->gid;
1292     shiftvec         = fr->shift_vec[0];
1293     fshift           = fr->fshift[0];
1294     facel            = _mm256_set1_pd(fr->epsfac);
1295     charge           = mdatoms->chargeA;
1296
1297     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
1298     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
1299     beta2            = _mm256_mul_pd(beta,beta);
1300     beta3            = _mm256_mul_pd(beta,beta2);
1301
1302     ewtab            = fr->ic->tabq_coul_F;
1303     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
1304     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
1305
1306     /* Setup water-specific parameters */
1307     inr              = nlist->iinr[0];
1308     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1309     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1310     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1311
1312     jq1              = _mm256_set1_pd(charge[inr+1]);
1313     jq2              = _mm256_set1_pd(charge[inr+2]);
1314     jq3              = _mm256_set1_pd(charge[inr+3]);
1315     qq11             = _mm256_mul_pd(iq1,jq1);
1316     qq12             = _mm256_mul_pd(iq1,jq2);
1317     qq13             = _mm256_mul_pd(iq1,jq3);
1318     qq21             = _mm256_mul_pd(iq2,jq1);
1319     qq22             = _mm256_mul_pd(iq2,jq2);
1320     qq23             = _mm256_mul_pd(iq2,jq3);
1321     qq31             = _mm256_mul_pd(iq3,jq1);
1322     qq32             = _mm256_mul_pd(iq3,jq2);
1323     qq33             = _mm256_mul_pd(iq3,jq3);
1324
1325     /* Avoid stupid compiler warnings */
1326     jnrA = jnrB = jnrC = jnrD = 0;
1327     j_coord_offsetA = 0;
1328     j_coord_offsetB = 0;
1329     j_coord_offsetC = 0;
1330     j_coord_offsetD = 0;
1331
1332     outeriter        = 0;
1333     inneriter        = 0;
1334
1335     for(iidx=0;iidx<4*DIM;iidx++)
1336     {
1337         scratch[iidx] = 0.0;
1338     }
1339
1340     /* Start outer loop over neighborlists */
1341     for(iidx=0; iidx<nri; iidx++)
1342     {
1343         /* Load shift vector for this list */
1344         i_shift_offset   = DIM*shiftidx[iidx];
1345
1346         /* Load limits for loop over neighbors */
1347         j_index_start    = jindex[iidx];
1348         j_index_end      = jindex[iidx+1];
1349
1350         /* Get outer coordinate index */
1351         inr              = iinr[iidx];
1352         i_coord_offset   = DIM*inr;
1353
1354         /* Load i particle coords and add shift vector */
1355         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1356                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1357
1358         fix1             = _mm256_setzero_pd();
1359         fiy1             = _mm256_setzero_pd();
1360         fiz1             = _mm256_setzero_pd();
1361         fix2             = _mm256_setzero_pd();
1362         fiy2             = _mm256_setzero_pd();
1363         fiz2             = _mm256_setzero_pd();
1364         fix3             = _mm256_setzero_pd();
1365         fiy3             = _mm256_setzero_pd();
1366         fiz3             = _mm256_setzero_pd();
1367
1368         /* Start inner kernel loop */
1369         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1370         {
1371
1372             /* Get j neighbor index, and coordinate index */
1373             jnrA             = jjnr[jidx];
1374             jnrB             = jjnr[jidx+1];
1375             jnrC             = jjnr[jidx+2];
1376             jnrD             = jjnr[jidx+3];
1377             j_coord_offsetA  = DIM*jnrA;
1378             j_coord_offsetB  = DIM*jnrB;
1379             j_coord_offsetC  = DIM*jnrC;
1380             j_coord_offsetD  = DIM*jnrD;
1381
1382             /* load j atom coordinates */
1383             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1384                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1385                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1386
1387             /* Calculate displacement vector */
1388             dx11             = _mm256_sub_pd(ix1,jx1);
1389             dy11             = _mm256_sub_pd(iy1,jy1);
1390             dz11             = _mm256_sub_pd(iz1,jz1);
1391             dx12             = _mm256_sub_pd(ix1,jx2);
1392             dy12             = _mm256_sub_pd(iy1,jy2);
1393             dz12             = _mm256_sub_pd(iz1,jz2);
1394             dx13             = _mm256_sub_pd(ix1,jx3);
1395             dy13             = _mm256_sub_pd(iy1,jy3);
1396             dz13             = _mm256_sub_pd(iz1,jz3);
1397             dx21             = _mm256_sub_pd(ix2,jx1);
1398             dy21             = _mm256_sub_pd(iy2,jy1);
1399             dz21             = _mm256_sub_pd(iz2,jz1);
1400             dx22             = _mm256_sub_pd(ix2,jx2);
1401             dy22             = _mm256_sub_pd(iy2,jy2);
1402             dz22             = _mm256_sub_pd(iz2,jz2);
1403             dx23             = _mm256_sub_pd(ix2,jx3);
1404             dy23             = _mm256_sub_pd(iy2,jy3);
1405             dz23             = _mm256_sub_pd(iz2,jz3);
1406             dx31             = _mm256_sub_pd(ix3,jx1);
1407             dy31             = _mm256_sub_pd(iy3,jy1);
1408             dz31             = _mm256_sub_pd(iz3,jz1);
1409             dx32             = _mm256_sub_pd(ix3,jx2);
1410             dy32             = _mm256_sub_pd(iy3,jy2);
1411             dz32             = _mm256_sub_pd(iz3,jz2);
1412             dx33             = _mm256_sub_pd(ix3,jx3);
1413             dy33             = _mm256_sub_pd(iy3,jy3);
1414             dz33             = _mm256_sub_pd(iz3,jz3);
1415
1416             /* Calculate squared distance and things based on it */
1417             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1418             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1419             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1420             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1421             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1422             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1423             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1424             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1425             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1426
1427             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1428             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1429             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1430             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1431             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1432             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1433             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1434             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1435             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1436
1437             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1438             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1439             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1440             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1441             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1442             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1443             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1444             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1445             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1446
1447             fjx1             = _mm256_setzero_pd();
1448             fjy1             = _mm256_setzero_pd();
1449             fjz1             = _mm256_setzero_pd();
1450             fjx2             = _mm256_setzero_pd();
1451             fjy2             = _mm256_setzero_pd();
1452             fjz2             = _mm256_setzero_pd();
1453             fjx3             = _mm256_setzero_pd();
1454             fjy3             = _mm256_setzero_pd();
1455             fjz3             = _mm256_setzero_pd();
1456
1457             /**************************
1458              * CALCULATE INTERACTIONS *
1459              **************************/
1460
1461             r11              = _mm256_mul_pd(rsq11,rinv11);
1462
1463             /* EWALD ELECTROSTATICS */
1464
1465             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1466             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1467             ewitab           = _mm256_cvttpd_epi32(ewrt);
1468             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1469             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1470                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1471                                             &ewtabF,&ewtabFn);
1472             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1473             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1474
1475             fscal            = felec;
1476
1477             /* Calculate temporary vectorial force */
1478             tx               = _mm256_mul_pd(fscal,dx11);
1479             ty               = _mm256_mul_pd(fscal,dy11);
1480             tz               = _mm256_mul_pd(fscal,dz11);
1481
1482             /* Update vectorial force */
1483             fix1             = _mm256_add_pd(fix1,tx);
1484             fiy1             = _mm256_add_pd(fiy1,ty);
1485             fiz1             = _mm256_add_pd(fiz1,tz);
1486
1487             fjx1             = _mm256_add_pd(fjx1,tx);
1488             fjy1             = _mm256_add_pd(fjy1,ty);
1489             fjz1             = _mm256_add_pd(fjz1,tz);
1490
1491             /**************************
1492              * CALCULATE INTERACTIONS *
1493              **************************/
1494
1495             r12              = _mm256_mul_pd(rsq12,rinv12);
1496
1497             /* EWALD ELECTROSTATICS */
1498
1499             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1500             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1501             ewitab           = _mm256_cvttpd_epi32(ewrt);
1502             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1503             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1504                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1505                                             &ewtabF,&ewtabFn);
1506             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1507             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1508
1509             fscal            = felec;
1510
1511             /* Calculate temporary vectorial force */
1512             tx               = _mm256_mul_pd(fscal,dx12);
1513             ty               = _mm256_mul_pd(fscal,dy12);
1514             tz               = _mm256_mul_pd(fscal,dz12);
1515
1516             /* Update vectorial force */
1517             fix1             = _mm256_add_pd(fix1,tx);
1518             fiy1             = _mm256_add_pd(fiy1,ty);
1519             fiz1             = _mm256_add_pd(fiz1,tz);
1520
1521             fjx2             = _mm256_add_pd(fjx2,tx);
1522             fjy2             = _mm256_add_pd(fjy2,ty);
1523             fjz2             = _mm256_add_pd(fjz2,tz);
1524
1525             /**************************
1526              * CALCULATE INTERACTIONS *
1527              **************************/
1528
1529             r13              = _mm256_mul_pd(rsq13,rinv13);
1530
1531             /* EWALD ELECTROSTATICS */
1532
1533             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1534             ewrt             = _mm256_mul_pd(r13,ewtabscale);
1535             ewitab           = _mm256_cvttpd_epi32(ewrt);
1536             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1537             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1538                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1539                                             &ewtabF,&ewtabFn);
1540             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1541             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
1542
1543             fscal            = felec;
1544
1545             /* Calculate temporary vectorial force */
1546             tx               = _mm256_mul_pd(fscal,dx13);
1547             ty               = _mm256_mul_pd(fscal,dy13);
1548             tz               = _mm256_mul_pd(fscal,dz13);
1549
1550             /* Update vectorial force */
1551             fix1             = _mm256_add_pd(fix1,tx);
1552             fiy1             = _mm256_add_pd(fiy1,ty);
1553             fiz1             = _mm256_add_pd(fiz1,tz);
1554
1555             fjx3             = _mm256_add_pd(fjx3,tx);
1556             fjy3             = _mm256_add_pd(fjy3,ty);
1557             fjz3             = _mm256_add_pd(fjz3,tz);
1558
1559             /**************************
1560              * CALCULATE INTERACTIONS *
1561              **************************/
1562
1563             r21              = _mm256_mul_pd(rsq21,rinv21);
1564
1565             /* EWALD ELECTROSTATICS */
1566
1567             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1568             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1569             ewitab           = _mm256_cvttpd_epi32(ewrt);
1570             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1571             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1572                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1573                                             &ewtabF,&ewtabFn);
1574             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1575             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1576
1577             fscal            = felec;
1578
1579             /* Calculate temporary vectorial force */
1580             tx               = _mm256_mul_pd(fscal,dx21);
1581             ty               = _mm256_mul_pd(fscal,dy21);
1582             tz               = _mm256_mul_pd(fscal,dz21);
1583
1584             /* Update vectorial force */
1585             fix2             = _mm256_add_pd(fix2,tx);
1586             fiy2             = _mm256_add_pd(fiy2,ty);
1587             fiz2             = _mm256_add_pd(fiz2,tz);
1588
1589             fjx1             = _mm256_add_pd(fjx1,tx);
1590             fjy1             = _mm256_add_pd(fjy1,ty);
1591             fjz1             = _mm256_add_pd(fjz1,tz);
1592
1593             /**************************
1594              * CALCULATE INTERACTIONS *
1595              **************************/
1596
1597             r22              = _mm256_mul_pd(rsq22,rinv22);
1598
1599             /* EWALD ELECTROSTATICS */
1600
1601             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1602             ewrt             = _mm256_mul_pd(r22,ewtabscale);
1603             ewitab           = _mm256_cvttpd_epi32(ewrt);
1604             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1605             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1606                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1607                                             &ewtabF,&ewtabFn);
1608             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1609             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1610
1611             fscal            = felec;
1612
1613             /* Calculate temporary vectorial force */
1614             tx               = _mm256_mul_pd(fscal,dx22);
1615             ty               = _mm256_mul_pd(fscal,dy22);
1616             tz               = _mm256_mul_pd(fscal,dz22);
1617
1618             /* Update vectorial force */
1619             fix2             = _mm256_add_pd(fix2,tx);
1620             fiy2             = _mm256_add_pd(fiy2,ty);
1621             fiz2             = _mm256_add_pd(fiz2,tz);
1622
1623             fjx2             = _mm256_add_pd(fjx2,tx);
1624             fjy2             = _mm256_add_pd(fjy2,ty);
1625             fjz2             = _mm256_add_pd(fjz2,tz);
1626
1627             /**************************
1628              * CALCULATE INTERACTIONS *
1629              **************************/
1630
1631             r23              = _mm256_mul_pd(rsq23,rinv23);
1632
1633             /* EWALD ELECTROSTATICS */
1634
1635             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1636             ewrt             = _mm256_mul_pd(r23,ewtabscale);
1637             ewitab           = _mm256_cvttpd_epi32(ewrt);
1638             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1639             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1640                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1641                                             &ewtabF,&ewtabFn);
1642             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1643             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
1644
1645             fscal            = felec;
1646
1647             /* Calculate temporary vectorial force */
1648             tx               = _mm256_mul_pd(fscal,dx23);
1649             ty               = _mm256_mul_pd(fscal,dy23);
1650             tz               = _mm256_mul_pd(fscal,dz23);
1651
1652             /* Update vectorial force */
1653             fix2             = _mm256_add_pd(fix2,tx);
1654             fiy2             = _mm256_add_pd(fiy2,ty);
1655             fiz2             = _mm256_add_pd(fiz2,tz);
1656
1657             fjx3             = _mm256_add_pd(fjx3,tx);
1658             fjy3             = _mm256_add_pd(fjy3,ty);
1659             fjz3             = _mm256_add_pd(fjz3,tz);
1660
1661             /**************************
1662              * CALCULATE INTERACTIONS *
1663              **************************/
1664
1665             r31              = _mm256_mul_pd(rsq31,rinv31);
1666
1667             /* EWALD ELECTROSTATICS */
1668
1669             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1670             ewrt             = _mm256_mul_pd(r31,ewtabscale);
1671             ewitab           = _mm256_cvttpd_epi32(ewrt);
1672             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1673             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1674                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1675                                             &ewtabF,&ewtabFn);
1676             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1677             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
1678
1679             fscal            = felec;
1680
1681             /* Calculate temporary vectorial force */
1682             tx               = _mm256_mul_pd(fscal,dx31);
1683             ty               = _mm256_mul_pd(fscal,dy31);
1684             tz               = _mm256_mul_pd(fscal,dz31);
1685
1686             /* Update vectorial force */
1687             fix3             = _mm256_add_pd(fix3,tx);
1688             fiy3             = _mm256_add_pd(fiy3,ty);
1689             fiz3             = _mm256_add_pd(fiz3,tz);
1690
1691             fjx1             = _mm256_add_pd(fjx1,tx);
1692             fjy1             = _mm256_add_pd(fjy1,ty);
1693             fjz1             = _mm256_add_pd(fjz1,tz);
1694
1695             /**************************
1696              * CALCULATE INTERACTIONS *
1697              **************************/
1698
1699             r32              = _mm256_mul_pd(rsq32,rinv32);
1700
1701             /* EWALD ELECTROSTATICS */
1702
1703             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1704             ewrt             = _mm256_mul_pd(r32,ewtabscale);
1705             ewitab           = _mm256_cvttpd_epi32(ewrt);
1706             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1707             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1708                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1709                                             &ewtabF,&ewtabFn);
1710             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1711             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
1712
1713             fscal            = felec;
1714
1715             /* Calculate temporary vectorial force */
1716             tx               = _mm256_mul_pd(fscal,dx32);
1717             ty               = _mm256_mul_pd(fscal,dy32);
1718             tz               = _mm256_mul_pd(fscal,dz32);
1719
1720             /* Update vectorial force */
1721             fix3             = _mm256_add_pd(fix3,tx);
1722             fiy3             = _mm256_add_pd(fiy3,ty);
1723             fiz3             = _mm256_add_pd(fiz3,tz);
1724
1725             fjx2             = _mm256_add_pd(fjx2,tx);
1726             fjy2             = _mm256_add_pd(fjy2,ty);
1727             fjz2             = _mm256_add_pd(fjz2,tz);
1728
1729             /**************************
1730              * CALCULATE INTERACTIONS *
1731              **************************/
1732
1733             r33              = _mm256_mul_pd(rsq33,rinv33);
1734
1735             /* EWALD ELECTROSTATICS */
1736
1737             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1738             ewrt             = _mm256_mul_pd(r33,ewtabscale);
1739             ewitab           = _mm256_cvttpd_epi32(ewrt);
1740             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1741             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1742                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1743                                             &ewtabF,&ewtabFn);
1744             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1745             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
1746
1747             fscal            = felec;
1748
1749             /* Calculate temporary vectorial force */
1750             tx               = _mm256_mul_pd(fscal,dx33);
1751             ty               = _mm256_mul_pd(fscal,dy33);
1752             tz               = _mm256_mul_pd(fscal,dz33);
1753
1754             /* Update vectorial force */
1755             fix3             = _mm256_add_pd(fix3,tx);
1756             fiy3             = _mm256_add_pd(fiy3,ty);
1757             fiz3             = _mm256_add_pd(fiz3,tz);
1758
1759             fjx3             = _mm256_add_pd(fjx3,tx);
1760             fjy3             = _mm256_add_pd(fjy3,ty);
1761             fjz3             = _mm256_add_pd(fjz3,tz);
1762
1763             fjptrA             = f+j_coord_offsetA;
1764             fjptrB             = f+j_coord_offsetB;
1765             fjptrC             = f+j_coord_offsetC;
1766             fjptrD             = f+j_coord_offsetD;
1767
1768             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1769                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1770
1771             /* Inner loop uses 324 flops */
1772         }
1773
1774         if(jidx<j_index_end)
1775         {
1776
1777             /* Get j neighbor index, and coordinate index */
1778             jnrlistA         = jjnr[jidx];
1779             jnrlistB         = jjnr[jidx+1];
1780             jnrlistC         = jjnr[jidx+2];
1781             jnrlistD         = jjnr[jidx+3];
1782             /* Sign of each element will be negative for non-real atoms.
1783              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1784              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1785              */
1786             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1787
1788             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1789             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1790             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1791
1792             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1793             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1794             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1795             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1796             j_coord_offsetA  = DIM*jnrA;
1797             j_coord_offsetB  = DIM*jnrB;
1798             j_coord_offsetC  = DIM*jnrC;
1799             j_coord_offsetD  = DIM*jnrD;
1800
1801             /* load j atom coordinates */
1802             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1803                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1804                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1805
1806             /* Calculate displacement vector */
1807             dx11             = _mm256_sub_pd(ix1,jx1);
1808             dy11             = _mm256_sub_pd(iy1,jy1);
1809             dz11             = _mm256_sub_pd(iz1,jz1);
1810             dx12             = _mm256_sub_pd(ix1,jx2);
1811             dy12             = _mm256_sub_pd(iy1,jy2);
1812             dz12             = _mm256_sub_pd(iz1,jz2);
1813             dx13             = _mm256_sub_pd(ix1,jx3);
1814             dy13             = _mm256_sub_pd(iy1,jy3);
1815             dz13             = _mm256_sub_pd(iz1,jz3);
1816             dx21             = _mm256_sub_pd(ix2,jx1);
1817             dy21             = _mm256_sub_pd(iy2,jy1);
1818             dz21             = _mm256_sub_pd(iz2,jz1);
1819             dx22             = _mm256_sub_pd(ix2,jx2);
1820             dy22             = _mm256_sub_pd(iy2,jy2);
1821             dz22             = _mm256_sub_pd(iz2,jz2);
1822             dx23             = _mm256_sub_pd(ix2,jx3);
1823             dy23             = _mm256_sub_pd(iy2,jy3);
1824             dz23             = _mm256_sub_pd(iz2,jz3);
1825             dx31             = _mm256_sub_pd(ix3,jx1);
1826             dy31             = _mm256_sub_pd(iy3,jy1);
1827             dz31             = _mm256_sub_pd(iz3,jz1);
1828             dx32             = _mm256_sub_pd(ix3,jx2);
1829             dy32             = _mm256_sub_pd(iy3,jy2);
1830             dz32             = _mm256_sub_pd(iz3,jz2);
1831             dx33             = _mm256_sub_pd(ix3,jx3);
1832             dy33             = _mm256_sub_pd(iy3,jy3);
1833             dz33             = _mm256_sub_pd(iz3,jz3);
1834
1835             /* Calculate squared distance and things based on it */
1836             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1837             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1838             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1839             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1840             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1841             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1842             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1843             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1844             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1845
1846             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1847             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1848             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1849             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1850             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1851             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1852             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1853             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1854             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1855
1856             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1857             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1858             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1859             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1860             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1861             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1862             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1863             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1864             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1865
1866             fjx1             = _mm256_setzero_pd();
1867             fjy1             = _mm256_setzero_pd();
1868             fjz1             = _mm256_setzero_pd();
1869             fjx2             = _mm256_setzero_pd();
1870             fjy2             = _mm256_setzero_pd();
1871             fjz2             = _mm256_setzero_pd();
1872             fjx3             = _mm256_setzero_pd();
1873             fjy3             = _mm256_setzero_pd();
1874             fjz3             = _mm256_setzero_pd();
1875
1876             /**************************
1877              * CALCULATE INTERACTIONS *
1878              **************************/
1879
1880             r11              = _mm256_mul_pd(rsq11,rinv11);
1881             r11              = _mm256_andnot_pd(dummy_mask,r11);
1882
1883             /* EWALD ELECTROSTATICS */
1884
1885             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1886             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1887             ewitab           = _mm256_cvttpd_epi32(ewrt);
1888             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1889             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1890                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1891                                             &ewtabF,&ewtabFn);
1892             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1893             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1894
1895             fscal            = felec;
1896
1897             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1898
1899             /* Calculate temporary vectorial force */
1900             tx               = _mm256_mul_pd(fscal,dx11);
1901             ty               = _mm256_mul_pd(fscal,dy11);
1902             tz               = _mm256_mul_pd(fscal,dz11);
1903
1904             /* Update vectorial force */
1905             fix1             = _mm256_add_pd(fix1,tx);
1906             fiy1             = _mm256_add_pd(fiy1,ty);
1907             fiz1             = _mm256_add_pd(fiz1,tz);
1908
1909             fjx1             = _mm256_add_pd(fjx1,tx);
1910             fjy1             = _mm256_add_pd(fjy1,ty);
1911             fjz1             = _mm256_add_pd(fjz1,tz);
1912
1913             /**************************
1914              * CALCULATE INTERACTIONS *
1915              **************************/
1916
1917             r12              = _mm256_mul_pd(rsq12,rinv12);
1918             r12              = _mm256_andnot_pd(dummy_mask,r12);
1919
1920             /* EWALD ELECTROSTATICS */
1921
1922             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1923             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1924             ewitab           = _mm256_cvttpd_epi32(ewrt);
1925             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1926             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1927                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1928                                             &ewtabF,&ewtabFn);
1929             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1930             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1931
1932             fscal            = felec;
1933
1934             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1935
1936             /* Calculate temporary vectorial force */
1937             tx               = _mm256_mul_pd(fscal,dx12);
1938             ty               = _mm256_mul_pd(fscal,dy12);
1939             tz               = _mm256_mul_pd(fscal,dz12);
1940
1941             /* Update vectorial force */
1942             fix1             = _mm256_add_pd(fix1,tx);
1943             fiy1             = _mm256_add_pd(fiy1,ty);
1944             fiz1             = _mm256_add_pd(fiz1,tz);
1945
1946             fjx2             = _mm256_add_pd(fjx2,tx);
1947             fjy2             = _mm256_add_pd(fjy2,ty);
1948             fjz2             = _mm256_add_pd(fjz2,tz);
1949
1950             /**************************
1951              * CALCULATE INTERACTIONS *
1952              **************************/
1953
1954             r13              = _mm256_mul_pd(rsq13,rinv13);
1955             r13              = _mm256_andnot_pd(dummy_mask,r13);
1956
1957             /* EWALD ELECTROSTATICS */
1958
1959             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1960             ewrt             = _mm256_mul_pd(r13,ewtabscale);
1961             ewitab           = _mm256_cvttpd_epi32(ewrt);
1962             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1963             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1964                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1965                                             &ewtabF,&ewtabFn);
1966             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1967             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
1968
1969             fscal            = felec;
1970
1971             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1972
1973             /* Calculate temporary vectorial force */
1974             tx               = _mm256_mul_pd(fscal,dx13);
1975             ty               = _mm256_mul_pd(fscal,dy13);
1976             tz               = _mm256_mul_pd(fscal,dz13);
1977
1978             /* Update vectorial force */
1979             fix1             = _mm256_add_pd(fix1,tx);
1980             fiy1             = _mm256_add_pd(fiy1,ty);
1981             fiz1             = _mm256_add_pd(fiz1,tz);
1982
1983             fjx3             = _mm256_add_pd(fjx3,tx);
1984             fjy3             = _mm256_add_pd(fjy3,ty);
1985             fjz3             = _mm256_add_pd(fjz3,tz);
1986
1987             /**************************
1988              * CALCULATE INTERACTIONS *
1989              **************************/
1990
1991             r21              = _mm256_mul_pd(rsq21,rinv21);
1992             r21              = _mm256_andnot_pd(dummy_mask,r21);
1993
1994             /* EWALD ELECTROSTATICS */
1995
1996             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1997             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1998             ewitab           = _mm256_cvttpd_epi32(ewrt);
1999             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2000             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2001                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2002                                             &ewtabF,&ewtabFn);
2003             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2004             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2005
2006             fscal            = felec;
2007
2008             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2009
2010             /* Calculate temporary vectorial force */
2011             tx               = _mm256_mul_pd(fscal,dx21);
2012             ty               = _mm256_mul_pd(fscal,dy21);
2013             tz               = _mm256_mul_pd(fscal,dz21);
2014
2015             /* Update vectorial force */
2016             fix2             = _mm256_add_pd(fix2,tx);
2017             fiy2             = _mm256_add_pd(fiy2,ty);
2018             fiz2             = _mm256_add_pd(fiz2,tz);
2019
2020             fjx1             = _mm256_add_pd(fjx1,tx);
2021             fjy1             = _mm256_add_pd(fjy1,ty);
2022             fjz1             = _mm256_add_pd(fjz1,tz);
2023
2024             /**************************
2025              * CALCULATE INTERACTIONS *
2026              **************************/
2027
2028             r22              = _mm256_mul_pd(rsq22,rinv22);
2029             r22              = _mm256_andnot_pd(dummy_mask,r22);
2030
2031             /* EWALD ELECTROSTATICS */
2032
2033             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2034             ewrt             = _mm256_mul_pd(r22,ewtabscale);
2035             ewitab           = _mm256_cvttpd_epi32(ewrt);
2036             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2037             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2038                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2039                                             &ewtabF,&ewtabFn);
2040             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2041             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2042
2043             fscal            = felec;
2044
2045             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2046
2047             /* Calculate temporary vectorial force */
2048             tx               = _mm256_mul_pd(fscal,dx22);
2049             ty               = _mm256_mul_pd(fscal,dy22);
2050             tz               = _mm256_mul_pd(fscal,dz22);
2051
2052             /* Update vectorial force */
2053             fix2             = _mm256_add_pd(fix2,tx);
2054             fiy2             = _mm256_add_pd(fiy2,ty);
2055             fiz2             = _mm256_add_pd(fiz2,tz);
2056
2057             fjx2             = _mm256_add_pd(fjx2,tx);
2058             fjy2             = _mm256_add_pd(fjy2,ty);
2059             fjz2             = _mm256_add_pd(fjz2,tz);
2060
2061             /**************************
2062              * CALCULATE INTERACTIONS *
2063              **************************/
2064
2065             r23              = _mm256_mul_pd(rsq23,rinv23);
2066             r23              = _mm256_andnot_pd(dummy_mask,r23);
2067
2068             /* EWALD ELECTROSTATICS */
2069
2070             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2071             ewrt             = _mm256_mul_pd(r23,ewtabscale);
2072             ewitab           = _mm256_cvttpd_epi32(ewrt);
2073             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2074             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2075                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2076                                             &ewtabF,&ewtabFn);
2077             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2078             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
2079
2080             fscal            = felec;
2081
2082             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2083
2084             /* Calculate temporary vectorial force */
2085             tx               = _mm256_mul_pd(fscal,dx23);
2086             ty               = _mm256_mul_pd(fscal,dy23);
2087             tz               = _mm256_mul_pd(fscal,dz23);
2088
2089             /* Update vectorial force */
2090             fix2             = _mm256_add_pd(fix2,tx);
2091             fiy2             = _mm256_add_pd(fiy2,ty);
2092             fiz2             = _mm256_add_pd(fiz2,tz);
2093
2094             fjx3             = _mm256_add_pd(fjx3,tx);
2095             fjy3             = _mm256_add_pd(fjy3,ty);
2096             fjz3             = _mm256_add_pd(fjz3,tz);
2097
2098             /**************************
2099              * CALCULATE INTERACTIONS *
2100              **************************/
2101
2102             r31              = _mm256_mul_pd(rsq31,rinv31);
2103             r31              = _mm256_andnot_pd(dummy_mask,r31);
2104
2105             /* EWALD ELECTROSTATICS */
2106
2107             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2108             ewrt             = _mm256_mul_pd(r31,ewtabscale);
2109             ewitab           = _mm256_cvttpd_epi32(ewrt);
2110             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2111             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2112                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2113                                             &ewtabF,&ewtabFn);
2114             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2115             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
2116
2117             fscal            = felec;
2118
2119             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2120
2121             /* Calculate temporary vectorial force */
2122             tx               = _mm256_mul_pd(fscal,dx31);
2123             ty               = _mm256_mul_pd(fscal,dy31);
2124             tz               = _mm256_mul_pd(fscal,dz31);
2125
2126             /* Update vectorial force */
2127             fix3             = _mm256_add_pd(fix3,tx);
2128             fiy3             = _mm256_add_pd(fiy3,ty);
2129             fiz3             = _mm256_add_pd(fiz3,tz);
2130
2131             fjx1             = _mm256_add_pd(fjx1,tx);
2132             fjy1             = _mm256_add_pd(fjy1,ty);
2133             fjz1             = _mm256_add_pd(fjz1,tz);
2134
2135             /**************************
2136              * CALCULATE INTERACTIONS *
2137              **************************/
2138
2139             r32              = _mm256_mul_pd(rsq32,rinv32);
2140             r32              = _mm256_andnot_pd(dummy_mask,r32);
2141
2142             /* EWALD ELECTROSTATICS */
2143
2144             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2145             ewrt             = _mm256_mul_pd(r32,ewtabscale);
2146             ewitab           = _mm256_cvttpd_epi32(ewrt);
2147             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2148             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2149                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2150                                             &ewtabF,&ewtabFn);
2151             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2152             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
2153
2154             fscal            = felec;
2155
2156             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2157
2158             /* Calculate temporary vectorial force */
2159             tx               = _mm256_mul_pd(fscal,dx32);
2160             ty               = _mm256_mul_pd(fscal,dy32);
2161             tz               = _mm256_mul_pd(fscal,dz32);
2162
2163             /* Update vectorial force */
2164             fix3             = _mm256_add_pd(fix3,tx);
2165             fiy3             = _mm256_add_pd(fiy3,ty);
2166             fiz3             = _mm256_add_pd(fiz3,tz);
2167
2168             fjx2             = _mm256_add_pd(fjx2,tx);
2169             fjy2             = _mm256_add_pd(fjy2,ty);
2170             fjz2             = _mm256_add_pd(fjz2,tz);
2171
2172             /**************************
2173              * CALCULATE INTERACTIONS *
2174              **************************/
2175
2176             r33              = _mm256_mul_pd(rsq33,rinv33);
2177             r33              = _mm256_andnot_pd(dummy_mask,r33);
2178
2179             /* EWALD ELECTROSTATICS */
2180
2181             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2182             ewrt             = _mm256_mul_pd(r33,ewtabscale);
2183             ewitab           = _mm256_cvttpd_epi32(ewrt);
2184             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2185             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2186                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2187                                             &ewtabF,&ewtabFn);
2188             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2189             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
2190
2191             fscal            = felec;
2192
2193             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2194
2195             /* Calculate temporary vectorial force */
2196             tx               = _mm256_mul_pd(fscal,dx33);
2197             ty               = _mm256_mul_pd(fscal,dy33);
2198             tz               = _mm256_mul_pd(fscal,dz33);
2199
2200             /* Update vectorial force */
2201             fix3             = _mm256_add_pd(fix3,tx);
2202             fiy3             = _mm256_add_pd(fiy3,ty);
2203             fiz3             = _mm256_add_pd(fiz3,tz);
2204
2205             fjx3             = _mm256_add_pd(fjx3,tx);
2206             fjy3             = _mm256_add_pd(fjy3,ty);
2207             fjz3             = _mm256_add_pd(fjz3,tz);
2208
2209             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2210             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2211             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2212             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2213
2214             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2215                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2216
2217             /* Inner loop uses 333 flops */
2218         }
2219
2220         /* End of innermost loop */
2221
2222         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2223                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
2224
2225         /* Increment number of inner iterations */
2226         inneriter                  += j_index_end - j_index_start;
2227
2228         /* Outer loop uses 18 flops */
2229     }
2230
2231     /* Increment number of outer iterations */
2232     outeriter        += nri;
2233
2234     /* Update outer/inner flops */
2235
2236     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*333);
2237 }