src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwNone_GeomW4W4_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single 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_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4W4_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Water4
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwNone_GeomW4W4_VF_avx_256_single
  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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr1;
  73     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     real *           vdwioffsetptr3;
  77     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  78     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  79     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  80     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  81     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  82     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
  83     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  84     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  85     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  86     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  87     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  88     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  89     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  90     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  91     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
  92     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
  93     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m256i          ewitab;
  96     __m128i          ewitab_lo,ewitab_hi;
  97     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  99     real             *ewtab;
 100     __m256           dummy_mask,cutoff_mask;
 101     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 102     __m256           one     = _mm256_set1_ps(1.0);
 103     __m256           two     = _mm256_set1_ps(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm256_set1_ps(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117
 118     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 119     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 120     beta2            = _mm256_mul_ps(beta,beta);
 121     beta3            = _mm256_mul_ps(beta,beta2);
 122
 123     ewtab            = fr->ic->tabq_coul_FDV0;
 124     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 125     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 130     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 131     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 132
 133     jq1              = _mm256_set1_ps(charge[inr+1]);
 134     jq2              = _mm256_set1_ps(charge[inr+2]);
 135     jq3              = _mm256_set1_ps(charge[inr+3]);
 136     qq11             = _mm256_mul_ps(iq1,jq1);
 137     qq12             = _mm256_mul_ps(iq1,jq2);
 138     qq13             = _mm256_mul_ps(iq1,jq3);
 139     qq21             = _mm256_mul_ps(iq2,jq1);
 140     qq22             = _mm256_mul_ps(iq2,jq2);
 141     qq23             = _mm256_mul_ps(iq2,jq3);
 142     qq31             = _mm256_mul_ps(iq3,jq1);
 143     qq32             = _mm256_mul_ps(iq3,jq2);
 144     qq33             = _mm256_mul_ps(iq3,jq3);
 145
 146     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 147     rcutoff_scalar   = fr->rcoulomb;
 148     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 149     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 150
 151     /* Avoid stupid compiler warnings */
 152     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 153     j_coord_offsetA = 0;
 154     j_coord_offsetB = 0;
 155     j_coord_offsetC = 0;
 156     j_coord_offsetD = 0;
 157     j_coord_offsetE = 0;
 158     j_coord_offsetF = 0;
 159     j_coord_offsetG = 0;
 160     j_coord_offsetH = 0;
 161
 162     outeriter        = 0;
 163     inneriter        = 0;
 164
 165     for(iidx=0;iidx<4*DIM;iidx++)
 166     {
 167         scratch[iidx] = 0.0;
 168     }
 169
 170     /* Start outer loop over neighborlists */
 171     for(iidx=0; iidx<nri; iidx++)
 172     {
 173         /* Load shift vector for this list */
 174         i_shift_offset   = DIM*shiftidx[iidx];
 175
 176         /* Load limits for loop over neighbors */
 177         j_index_start    = jindex[iidx];
 178         j_index_end      = jindex[iidx+1];
 179
 180         /* Get outer coordinate index */
 181         inr              = iinr[iidx];
 182         i_coord_offset   = DIM*inr;
 183
 184         /* Load i particle coords and add shift vector */
 185         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 186                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 187
 188         fix1             = _mm256_setzero_ps();
 189         fiy1             = _mm256_setzero_ps();
 190         fiz1             = _mm256_setzero_ps();
 191         fix2             = _mm256_setzero_ps();
 192         fiy2             = _mm256_setzero_ps();
 193         fiz2             = _mm256_setzero_ps();
 194         fix3             = _mm256_setzero_ps();
 195         fiy3             = _mm256_setzero_ps();
 196         fiz3             = _mm256_setzero_ps();
 197
 198         /* Reset potential sums */
 199         velecsum         = _mm256_setzero_ps();
 200
 201         /* Start inner kernel loop */
 202         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 203         {
 204
 205             /* Get j neighbor index, and coordinate index */
 206             jnrA             = jjnr[jidx];
 207             jnrB             = jjnr[jidx+1];
 208             jnrC             = jjnr[jidx+2];
 209             jnrD             = jjnr[jidx+3];
 210             jnrE             = jjnr[jidx+4];
 211             jnrF             = jjnr[jidx+5];
 212             jnrG             = jjnr[jidx+6];
 213             jnrH             = jjnr[jidx+7];
 214             j_coord_offsetA  = DIM*jnrA;
 215             j_coord_offsetB  = DIM*jnrB;
 216             j_coord_offsetC  = DIM*jnrC;
 217             j_coord_offsetD  = DIM*jnrD;
 218             j_coord_offsetE  = DIM*jnrE;
 219             j_coord_offsetF  = DIM*jnrF;
 220             j_coord_offsetG  = DIM*jnrG;
 221             j_coord_offsetH  = DIM*jnrH;
 222
 223             /* load j atom coordinates */
 224             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 225                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 226                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
 227                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
 228                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 229
 230             /* Calculate displacement vector */
 231             dx11             = _mm256_sub_ps(ix1,jx1);
 232             dy11             = _mm256_sub_ps(iy1,jy1);
 233             dz11             = _mm256_sub_ps(iz1,jz1);
 234             dx12             = _mm256_sub_ps(ix1,jx2);
 235             dy12             = _mm256_sub_ps(iy1,jy2);
 236             dz12             = _mm256_sub_ps(iz1,jz2);
 237             dx13             = _mm256_sub_ps(ix1,jx3);
 238             dy13             = _mm256_sub_ps(iy1,jy3);
 239             dz13             = _mm256_sub_ps(iz1,jz3);
 240             dx21             = _mm256_sub_ps(ix2,jx1);
 241             dy21             = _mm256_sub_ps(iy2,jy1);
 242             dz21             = _mm256_sub_ps(iz2,jz1);
 243             dx22             = _mm256_sub_ps(ix2,jx2);
 244             dy22             = _mm256_sub_ps(iy2,jy2);
 245             dz22             = _mm256_sub_ps(iz2,jz2);
 246             dx23             = _mm256_sub_ps(ix2,jx3);
 247             dy23             = _mm256_sub_ps(iy2,jy3);
 248             dz23             = _mm256_sub_ps(iz2,jz3);
 249             dx31             = _mm256_sub_ps(ix3,jx1);
 250             dy31             = _mm256_sub_ps(iy3,jy1);
 251             dz31             = _mm256_sub_ps(iz3,jz1);
 252             dx32             = _mm256_sub_ps(ix3,jx2);
 253             dy32             = _mm256_sub_ps(iy3,jy2);
 254             dz32             = _mm256_sub_ps(iz3,jz2);
 255             dx33             = _mm256_sub_ps(ix3,jx3);
 256             dy33             = _mm256_sub_ps(iy3,jy3);
 257             dz33             = _mm256_sub_ps(iz3,jz3);
 258
 259             /* Calculate squared distance and things based on it */
 260             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 261             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 262             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 263             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 264             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 265             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 266             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 267             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 268             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 269
 270             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 271             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 272             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 273             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 274             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 275             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 276             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 277             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 278             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 279
 280             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 281             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 282             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 283             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 284             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 285             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 286             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 287             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 288             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 289
 290             fjx1             = _mm256_setzero_ps();
 291             fjy1             = _mm256_setzero_ps();
 292             fjz1             = _mm256_setzero_ps();
 293             fjx2             = _mm256_setzero_ps();
 294             fjy2             = _mm256_setzero_ps();
 295             fjz2             = _mm256_setzero_ps();
 296             fjx3             = _mm256_setzero_ps();
 297             fjy3             = _mm256_setzero_ps();
 298             fjz3             = _mm256_setzero_ps();
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 305             {
 306
 307             r11              = _mm256_mul_ps(rsq11,rinv11);
 308
 309             /* EWALD ELECTROSTATICS */
 310
 311             /* Analytical PME correction */
 312             zeta2            = _mm256_mul_ps(beta2,rsq11);
 313             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 314             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 315             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 316             felec            = _mm256_mul_ps(qq11,felec);
 317             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 318             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 319             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
 320             velec            = _mm256_mul_ps(qq11,velec);
 321
 322             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velec            = _mm256_and_ps(velec,cutoff_mask);
 326             velecsum         = _mm256_add_ps(velecsum,velec);
 327
 328             fscal            = felec;
 329
 330             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm256_mul_ps(fscal,dx11);
 334             ty               = _mm256_mul_ps(fscal,dy11);
 335             tz               = _mm256_mul_ps(fscal,dz11);
 336
 337             /* Update vectorial force */
 338             fix1             = _mm256_add_ps(fix1,tx);
 339             fiy1             = _mm256_add_ps(fiy1,ty);
 340             fiz1             = _mm256_add_ps(fiz1,tz);
 341
 342             fjx1             = _mm256_add_ps(fjx1,tx);
 343             fjy1             = _mm256_add_ps(fjy1,ty);
 344             fjz1             = _mm256_add_ps(fjz1,tz);
 345
 346             }
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 353             {
 354
 355             r12              = _mm256_mul_ps(rsq12,rinv12);
 356
 357             /* EWALD ELECTROSTATICS */
 358
 359             /* Analytical PME correction */
 360             zeta2            = _mm256_mul_ps(beta2,rsq12);
 361             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 362             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 363             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 364             felec            = _mm256_mul_ps(qq12,felec);
 365             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 366             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 367             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
 368             velec            = _mm256_mul_ps(qq12,velec);
 369
 370             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velec            = _mm256_and_ps(velec,cutoff_mask);
 374             velecsum         = _mm256_add_ps(velecsum,velec);
 375
 376             fscal            = felec;
 377
 378             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 379
 380             /* Calculate temporary vectorial force */
 381             tx               = _mm256_mul_ps(fscal,dx12);
 382             ty               = _mm256_mul_ps(fscal,dy12);
 383             tz               = _mm256_mul_ps(fscal,dz12);
 384
 385             /* Update vectorial force */
 386             fix1             = _mm256_add_ps(fix1,tx);
 387             fiy1             = _mm256_add_ps(fiy1,ty);
 388             fiz1             = _mm256_add_ps(fiz1,tz);
 389
 390             fjx2             = _mm256_add_ps(fjx2,tx);
 391             fjy2             = _mm256_add_ps(fjy2,ty);
 392             fjz2             = _mm256_add_ps(fjz2,tz);
 393
 394             }
 395
 396             /**************************
 397              * CALCULATE INTERACTIONS *
 398              **************************/
 399
 400             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 401             {
 402
 403             r13              = _mm256_mul_ps(rsq13,rinv13);
 404
 405             /* EWALD ELECTROSTATICS */
 406
 407             /* Analytical PME correction */
 408             zeta2            = _mm256_mul_ps(beta2,rsq13);
 409             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
 410             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 411             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 412             felec            = _mm256_mul_ps(qq13,felec);
 413             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 414             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 415             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv13,sh_ewald),pmecorrV);
 416             velec            = _mm256_mul_ps(qq13,velec);
 417
 418             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
 419
 420             /* Update potential sum for this i atom from the interaction with this j atom. */
 421             velec            = _mm256_and_ps(velec,cutoff_mask);
 422             velecsum         = _mm256_add_ps(velecsum,velec);
 423
 424             fscal            = felec;
 425
 426             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 427
 428             /* Calculate temporary vectorial force */
 429             tx               = _mm256_mul_ps(fscal,dx13);
 430             ty               = _mm256_mul_ps(fscal,dy13);
 431             tz               = _mm256_mul_ps(fscal,dz13);
 432
 433             /* Update vectorial force */
 434             fix1             = _mm256_add_ps(fix1,tx);
 435             fiy1             = _mm256_add_ps(fiy1,ty);
 436             fiz1             = _mm256_add_ps(fiz1,tz);
 437
 438             fjx3             = _mm256_add_ps(fjx3,tx);
 439             fjy3             = _mm256_add_ps(fjy3,ty);
 440             fjz3             = _mm256_add_ps(fjz3,tz);
 441
 442             }
 443
 444             /**************************
 445              * CALCULATE INTERACTIONS *
 446              **************************/
 447
 448             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 449             {
 450
 451             r21              = _mm256_mul_ps(rsq21,rinv21);
 452
 453             /* EWALD ELECTROSTATICS */
 454
 455             /* Analytical PME correction */
 456             zeta2            = _mm256_mul_ps(beta2,rsq21);
 457             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 458             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 459             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 460             felec            = _mm256_mul_ps(qq21,felec);
 461             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 462             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 463             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
 464             velec            = _mm256_mul_ps(qq21,velec);
 465
 466             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
 467
 468             /* Update potential sum for this i atom from the interaction with this j atom. */
 469             velec            = _mm256_and_ps(velec,cutoff_mask);
 470             velecsum         = _mm256_add_ps(velecsum,velec);
 471
 472             fscal            = felec;
 473
 474             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 475
 476             /* Calculate temporary vectorial force */
 477             tx               = _mm256_mul_ps(fscal,dx21);
 478             ty               = _mm256_mul_ps(fscal,dy21);
 479             tz               = _mm256_mul_ps(fscal,dz21);
 480
 481             /* Update vectorial force */
 482             fix2             = _mm256_add_ps(fix2,tx);
 483             fiy2             = _mm256_add_ps(fiy2,ty);
 484             fiz2             = _mm256_add_ps(fiz2,tz);
 485
 486             fjx1             = _mm256_add_ps(fjx1,tx);
 487             fjy1             = _mm256_add_ps(fjy1,ty);
 488             fjz1             = _mm256_add_ps(fjz1,tz);
 489
 490             }
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 497             {
 498
 499             r22              = _mm256_mul_ps(rsq22,rinv22);
 500
 501             /* EWALD ELECTROSTATICS */
 502
 503             /* Analytical PME correction */
 504             zeta2            = _mm256_mul_ps(beta2,rsq22);
 505             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 506             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 507             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 508             felec            = _mm256_mul_ps(qq22,felec);
 509             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 510             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 511             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
 512             velec            = _mm256_mul_ps(qq22,velec);
 513
 514             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velec            = _mm256_and_ps(velec,cutoff_mask);
 518             velecsum         = _mm256_add_ps(velecsum,velec);
 519
 520             fscal            = felec;
 521
 522             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 523
 524             /* Calculate temporary vectorial force */
 525             tx               = _mm256_mul_ps(fscal,dx22);
 526             ty               = _mm256_mul_ps(fscal,dy22);
 527             tz               = _mm256_mul_ps(fscal,dz22);
 528
 529             /* Update vectorial force */
 530             fix2             = _mm256_add_ps(fix2,tx);
 531             fiy2             = _mm256_add_ps(fiy2,ty);
 532             fiz2             = _mm256_add_ps(fiz2,tz);
 533
 534             fjx2             = _mm256_add_ps(fjx2,tx);
 535             fjy2             = _mm256_add_ps(fjy2,ty);
 536             fjz2             = _mm256_add_ps(fjz2,tz);
 537
 538             }
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 545             {
 546
 547             r23              = _mm256_mul_ps(rsq23,rinv23);
 548
 549             /* EWALD ELECTROSTATICS */
 550
 551             /* Analytical PME correction */
 552             zeta2            = _mm256_mul_ps(beta2,rsq23);
 553             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
 554             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 555             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 556             felec            = _mm256_mul_ps(qq23,felec);
 557             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 558             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 559             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv23,sh_ewald),pmecorrV);
 560             velec            = _mm256_mul_ps(qq23,velec);
 561
 562             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
 563
 564             /* Update potential sum for this i atom from the interaction with this j atom. */
 565             velec            = _mm256_and_ps(velec,cutoff_mask);
 566             velecsum         = _mm256_add_ps(velecsum,velec);
 567
 568             fscal            = felec;
 569
 570             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 571
 572             /* Calculate temporary vectorial force */
 573             tx               = _mm256_mul_ps(fscal,dx23);
 574             ty               = _mm256_mul_ps(fscal,dy23);
 575             tz               = _mm256_mul_ps(fscal,dz23);
 576
 577             /* Update vectorial force */
 578             fix2             = _mm256_add_ps(fix2,tx);
 579             fiy2             = _mm256_add_ps(fiy2,ty);
 580             fiz2             = _mm256_add_ps(fiz2,tz);
 581
 582             fjx3             = _mm256_add_ps(fjx3,tx);
 583             fjy3             = _mm256_add_ps(fjy3,ty);
 584             fjz3             = _mm256_add_ps(fjz3,tz);
 585
 586             }
 587
 588             /**************************
 589              * CALCULATE INTERACTIONS *
 590              **************************/
 591
 592             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 593             {
 594
 595             r31              = _mm256_mul_ps(rsq31,rinv31);
 596
 597             /* EWALD ELECTROSTATICS */
 598
 599             /* Analytical PME correction */
 600             zeta2            = _mm256_mul_ps(beta2,rsq31);
 601             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
 602             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 603             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 604             felec            = _mm256_mul_ps(qq31,felec);
 605             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 606             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 607             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv31,sh_ewald),pmecorrV);
 608             velec            = _mm256_mul_ps(qq31,velec);
 609
 610             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
 611
 612             /* Update potential sum for this i atom from the interaction with this j atom. */
 613             velec            = _mm256_and_ps(velec,cutoff_mask);
 614             velecsum         = _mm256_add_ps(velecsum,velec);
 615
 616             fscal            = felec;
 617
 618             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 619
 620             /* Calculate temporary vectorial force */
 621             tx               = _mm256_mul_ps(fscal,dx31);
 622             ty               = _mm256_mul_ps(fscal,dy31);
 623             tz               = _mm256_mul_ps(fscal,dz31);
 624
 625             /* Update vectorial force */
 626             fix3             = _mm256_add_ps(fix3,tx);
 627             fiy3             = _mm256_add_ps(fiy3,ty);
 628             fiz3             = _mm256_add_ps(fiz3,tz);
 629
 630             fjx1             = _mm256_add_ps(fjx1,tx);
 631             fjy1             = _mm256_add_ps(fjy1,ty);
 632             fjz1             = _mm256_add_ps(fjz1,tz);
 633
 634             }
 635
 636             /**************************
 637              * CALCULATE INTERACTIONS *
 638              **************************/
 639
 640             if (gmx_mm256_any_lt(rsq32,rcutoff2))
 641             {
 642
 643             r32              = _mm256_mul_ps(rsq32,rinv32);
 644
 645             /* EWALD ELECTROSTATICS */
 646
 647             /* Analytical PME correction */
 648             zeta2            = _mm256_mul_ps(beta2,rsq32);
 649             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
 650             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 651             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 652             felec            = _mm256_mul_ps(qq32,felec);
 653             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 654             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 655             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv32,sh_ewald),pmecorrV);
 656             velec            = _mm256_mul_ps(qq32,velec);
 657
 658             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
 659
 660             /* Update potential sum for this i atom from the interaction with this j atom. */
 661             velec            = _mm256_and_ps(velec,cutoff_mask);
 662             velecsum         = _mm256_add_ps(velecsum,velec);
 663
 664             fscal            = felec;
 665
 666             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 667
 668             /* Calculate temporary vectorial force */
 669             tx               = _mm256_mul_ps(fscal,dx32);
 670             ty               = _mm256_mul_ps(fscal,dy32);
 671             tz               = _mm256_mul_ps(fscal,dz32);
 672
 673             /* Update vectorial force */
 674             fix3             = _mm256_add_ps(fix3,tx);
 675             fiy3             = _mm256_add_ps(fiy3,ty);
 676             fiz3             = _mm256_add_ps(fiz3,tz);
 677
 678             fjx2             = _mm256_add_ps(fjx2,tx);
 679             fjy2             = _mm256_add_ps(fjy2,ty);
 680             fjz2             = _mm256_add_ps(fjz2,tz);
 681
 682             }
 683
 684             /**************************
 685              * CALCULATE INTERACTIONS *
 686              **************************/
 687
 688             if (gmx_mm256_any_lt(rsq33,rcutoff2))
 689             {
 690
 691             r33              = _mm256_mul_ps(rsq33,rinv33);
 692
 693             /* EWALD ELECTROSTATICS */
 694
 695             /* Analytical PME correction */
 696             zeta2            = _mm256_mul_ps(beta2,rsq33);
 697             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
 698             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 699             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 700             felec            = _mm256_mul_ps(qq33,felec);
 701             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 702             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 703             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv33,sh_ewald),pmecorrV);
 704             velec            = _mm256_mul_ps(qq33,velec);
 705
 706             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
 707
 708             /* Update potential sum for this i atom from the interaction with this j atom. */
 709             velec            = _mm256_and_ps(velec,cutoff_mask);
 710             velecsum         = _mm256_add_ps(velecsum,velec);
 711
 712             fscal            = felec;
 713
 714             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 715
 716             /* Calculate temporary vectorial force */
 717             tx               = _mm256_mul_ps(fscal,dx33);
 718             ty               = _mm256_mul_ps(fscal,dy33);
 719             tz               = _mm256_mul_ps(fscal,dz33);
 720
 721             /* Update vectorial force */
 722             fix3             = _mm256_add_ps(fix3,tx);
 723             fiy3             = _mm256_add_ps(fiy3,ty);
 724             fiz3             = _mm256_add_ps(fiz3,tz);
 725
 726             fjx3             = _mm256_add_ps(fjx3,tx);
 727             fjy3             = _mm256_add_ps(fjy3,ty);
 728             fjz3             = _mm256_add_ps(fjz3,tz);
 729
 730             }
 731
 732             fjptrA             = f+j_coord_offsetA;
 733             fjptrB             = f+j_coord_offsetB;
 734             fjptrC             = f+j_coord_offsetC;
 735             fjptrD             = f+j_coord_offsetD;
 736             fjptrE             = f+j_coord_offsetE;
 737             fjptrF             = f+j_coord_offsetF;
 738             fjptrG             = f+j_coord_offsetG;
 739             fjptrH             = f+j_coord_offsetH;
 740
 741             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 742                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
 743                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 744
 745             /* Inner loop uses 981 flops */
 746         }
 747
 748         if(jidx<j_index_end)
 749         {
 750
 751             /* Get j neighbor index, and coordinate index */
 752             jnrlistA         = jjnr[jidx];
 753             jnrlistB         = jjnr[jidx+1];
 754             jnrlistC         = jjnr[jidx+2];
 755             jnrlistD         = jjnr[jidx+3];
 756             jnrlistE         = jjnr[jidx+4];
 757             jnrlistF         = jjnr[jidx+5];
 758             jnrlistG         = jjnr[jidx+6];
 759             jnrlistH         = jjnr[jidx+7];
 760             /* Sign of each element will be negative for non-real atoms.
 761              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 762              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 763              */
 764             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 765                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 766
 767             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 768             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 769             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 770             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 771             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 772             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 773             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 774             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 775             j_coord_offsetA  = DIM*jnrA;
 776             j_coord_offsetB  = DIM*jnrB;
 777             j_coord_offsetC  = DIM*jnrC;
 778             j_coord_offsetD  = DIM*jnrD;
 779             j_coord_offsetE  = DIM*jnrE;
 780             j_coord_offsetF  = DIM*jnrF;
 781             j_coord_offsetG  = DIM*jnrG;
 782             j_coord_offsetH  = DIM*jnrH;
 783
 784             /* load j atom coordinates */
 785             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 786                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 787                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
 788                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
 789                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 790
 791             /* Calculate displacement vector */
 792             dx11             = _mm256_sub_ps(ix1,jx1);
 793             dy11             = _mm256_sub_ps(iy1,jy1);
 794             dz11             = _mm256_sub_ps(iz1,jz1);
 795             dx12             = _mm256_sub_ps(ix1,jx2);
 796             dy12             = _mm256_sub_ps(iy1,jy2);
 797             dz12             = _mm256_sub_ps(iz1,jz2);
 798             dx13             = _mm256_sub_ps(ix1,jx3);
 799             dy13             = _mm256_sub_ps(iy1,jy3);
 800             dz13             = _mm256_sub_ps(iz1,jz3);
 801             dx21             = _mm256_sub_ps(ix2,jx1);
 802             dy21             = _mm256_sub_ps(iy2,jy1);
 803             dz21             = _mm256_sub_ps(iz2,jz1);
 804             dx22             = _mm256_sub_ps(ix2,jx2);
 805             dy22             = _mm256_sub_ps(iy2,jy2);
 806             dz22             = _mm256_sub_ps(iz2,jz2);
 807             dx23             = _mm256_sub_ps(ix2,jx3);
 808             dy23             = _mm256_sub_ps(iy2,jy3);
 809             dz23             = _mm256_sub_ps(iz2,jz3);
 810             dx31             = _mm256_sub_ps(ix3,jx1);
 811             dy31             = _mm256_sub_ps(iy3,jy1);
 812             dz31             = _mm256_sub_ps(iz3,jz1);
 813             dx32             = _mm256_sub_ps(ix3,jx2);
 814             dy32             = _mm256_sub_ps(iy3,jy2);
 815             dz32             = _mm256_sub_ps(iz3,jz2);
 816             dx33             = _mm256_sub_ps(ix3,jx3);
 817             dy33             = _mm256_sub_ps(iy3,jy3);
 818             dz33             = _mm256_sub_ps(iz3,jz3);
 819
 820             /* Calculate squared distance and things based on it */
 821             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 822             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 823             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 824             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 825             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 826             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 827             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 828             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 829             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 830
 831             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 832             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 833             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 834             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 835             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 836             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 837             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 838             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 839             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 840
 841             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 842             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 843             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 844             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 845             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 846             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 847             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 848             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 849             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 850
 851             fjx1             = _mm256_setzero_ps();
 852             fjy1             = _mm256_setzero_ps();
 853             fjz1             = _mm256_setzero_ps();
 854             fjx2             = _mm256_setzero_ps();
 855             fjy2             = _mm256_setzero_ps();
 856             fjz2             = _mm256_setzero_ps();
 857             fjx3             = _mm256_setzero_ps();
 858             fjy3             = _mm256_setzero_ps();
 859             fjz3             = _mm256_setzero_ps();
 860
 861             /**************************
 862              * CALCULATE INTERACTIONS *
 863              **************************/
 864
 865             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 866             {
 867
 868             r11              = _mm256_mul_ps(rsq11,rinv11);
 869             r11              = _mm256_andnot_ps(dummy_mask,r11);
 870
 871             /* EWALD ELECTROSTATICS */
 872
 873             /* Analytical PME correction */
 874             zeta2            = _mm256_mul_ps(beta2,rsq11);
 875             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 876             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 877             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 878             felec            = _mm256_mul_ps(qq11,felec);
 879             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 880             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 881             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
 882             velec            = _mm256_mul_ps(qq11,velec);
 883
 884             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 885
 886             /* Update potential sum for this i atom from the interaction with this j atom. */
 887             velec            = _mm256_and_ps(velec,cutoff_mask);
 888             velec            = _mm256_andnot_ps(dummy_mask,velec);
 889             velecsum         = _mm256_add_ps(velecsum,velec);
 890
 891             fscal            = felec;
 892
 893             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 894
 895             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 896
 897             /* Calculate temporary vectorial force */
 898             tx               = _mm256_mul_ps(fscal,dx11);
 899             ty               = _mm256_mul_ps(fscal,dy11);
 900             tz               = _mm256_mul_ps(fscal,dz11);
 901
 902             /* Update vectorial force */
 903             fix1             = _mm256_add_ps(fix1,tx);
 904             fiy1             = _mm256_add_ps(fiy1,ty);
 905             fiz1             = _mm256_add_ps(fiz1,tz);
 906
 907             fjx1             = _mm256_add_ps(fjx1,tx);
 908             fjy1             = _mm256_add_ps(fjy1,ty);
 909             fjz1             = _mm256_add_ps(fjz1,tz);
 910
 911             }
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 918             {
 919
 920             r12              = _mm256_mul_ps(rsq12,rinv12);
 921             r12              = _mm256_andnot_ps(dummy_mask,r12);
 922
 923             /* EWALD ELECTROSTATICS */
 924
 925             /* Analytical PME correction */
 926             zeta2            = _mm256_mul_ps(beta2,rsq12);
 927             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 928             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 929             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 930             felec            = _mm256_mul_ps(qq12,felec);
 931             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 932             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 933             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
 934             velec            = _mm256_mul_ps(qq12,velec);
 935
 936             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 937
 938             /* Update potential sum for this i atom from the interaction with this j atom. */
 939             velec            = _mm256_and_ps(velec,cutoff_mask);
 940             velec            = _mm256_andnot_ps(dummy_mask,velec);
 941             velecsum         = _mm256_add_ps(velecsum,velec);
 942
 943             fscal            = felec;
 944
 945             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 946
 947             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 948
 949             /* Calculate temporary vectorial force */
 950             tx               = _mm256_mul_ps(fscal,dx12);
 951             ty               = _mm256_mul_ps(fscal,dy12);
 952             tz               = _mm256_mul_ps(fscal,dz12);
 953
 954             /* Update vectorial force */
 955             fix1             = _mm256_add_ps(fix1,tx);
 956             fiy1             = _mm256_add_ps(fiy1,ty);
 957             fiz1             = _mm256_add_ps(fiz1,tz);
 958
 959             fjx2             = _mm256_add_ps(fjx2,tx);
 960             fjy2             = _mm256_add_ps(fjy2,ty);
 961             fjz2             = _mm256_add_ps(fjz2,tz);
 962
 963             }
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 970             {
 971
 972             r13              = _mm256_mul_ps(rsq13,rinv13);
 973             r13              = _mm256_andnot_ps(dummy_mask,r13);
 974
 975             /* EWALD ELECTROSTATICS */
 976
 977             /* Analytical PME correction */
 978             zeta2            = _mm256_mul_ps(beta2,rsq13);
 979             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
 980             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 981             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 982             felec            = _mm256_mul_ps(qq13,felec);
 983             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 984             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 985             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv13,sh_ewald),pmecorrV);
 986             velec            = _mm256_mul_ps(qq13,velec);
 987
 988             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
 989
 990             /* Update potential sum for this i atom from the interaction with this j atom. */
 991             velec            = _mm256_and_ps(velec,cutoff_mask);
 992             velec            = _mm256_andnot_ps(dummy_mask,velec);
 993             velecsum         = _mm256_add_ps(velecsum,velec);
 994
 995             fscal            = felec;
 996
 997             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 998
 999             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1000
1001             /* Calculate temporary vectorial force */
1002             tx               = _mm256_mul_ps(fscal,dx13);
1003             ty               = _mm256_mul_ps(fscal,dy13);
1004             tz               = _mm256_mul_ps(fscal,dz13);
1005
1006             /* Update vectorial force */
1007             fix1             = _mm256_add_ps(fix1,tx);
1008             fiy1             = _mm256_add_ps(fiy1,ty);
1009             fiz1             = _mm256_add_ps(fiz1,tz);
1010
1011             fjx3             = _mm256_add_ps(fjx3,tx);
1012             fjy3             = _mm256_add_ps(fjy3,ty);
1013             fjz3             = _mm256_add_ps(fjz3,tz);
1014
1015             }
1016
1017             /**************************
1018              * CALCULATE INTERACTIONS *
1019              **************************/
1020
1021             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1022             {
1023
1024             r21              = _mm256_mul_ps(rsq21,rinv21);
1025             r21              = _mm256_andnot_ps(dummy_mask,r21);
1026
1027             /* EWALD ELECTROSTATICS */
1028
1029             /* Analytical PME correction */
1030             zeta2            = _mm256_mul_ps(beta2,rsq21);
1031             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1032             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1033             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1034             felec            = _mm256_mul_ps(qq21,felec);
1035             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1036             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1037             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
1038             velec            = _mm256_mul_ps(qq21,velec);
1039
1040             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1041
1042             /* Update potential sum for this i atom from the interaction with this j atom. */
1043             velec            = _mm256_and_ps(velec,cutoff_mask);
1044             velec            = _mm256_andnot_ps(dummy_mask,velec);
1045             velecsum         = _mm256_add_ps(velecsum,velec);
1046
1047             fscal            = felec;
1048
1049             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1050
1051             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1052
1053             /* Calculate temporary vectorial force */
1054             tx               = _mm256_mul_ps(fscal,dx21);
1055             ty               = _mm256_mul_ps(fscal,dy21);
1056             tz               = _mm256_mul_ps(fscal,dz21);
1057
1058             /* Update vectorial force */
1059             fix2             = _mm256_add_ps(fix2,tx);
1060             fiy2             = _mm256_add_ps(fiy2,ty);
1061             fiz2             = _mm256_add_ps(fiz2,tz);
1062
1063             fjx1             = _mm256_add_ps(fjx1,tx);
1064             fjy1             = _mm256_add_ps(fjy1,ty);
1065             fjz1             = _mm256_add_ps(fjz1,tz);
1066
1067             }
1068
1069             /**************************
1070              * CALCULATE INTERACTIONS *
1071              **************************/
1072
1073             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1074             {
1075
1076             r22              = _mm256_mul_ps(rsq22,rinv22);
1077             r22              = _mm256_andnot_ps(dummy_mask,r22);
1078
1079             /* EWALD ELECTROSTATICS */
1080
1081             /* Analytical PME correction */
1082             zeta2            = _mm256_mul_ps(beta2,rsq22);
1083             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1084             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1085             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1086             felec            = _mm256_mul_ps(qq22,felec);
1087             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1088             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1089             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
1090             velec            = _mm256_mul_ps(qq22,velec);
1091
1092             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1093
1094             /* Update potential sum for this i atom from the interaction with this j atom. */
1095             velec            = _mm256_and_ps(velec,cutoff_mask);
1096             velec            = _mm256_andnot_ps(dummy_mask,velec);
1097             velecsum         = _mm256_add_ps(velecsum,velec);
1098
1099             fscal            = felec;
1100
1101             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1102
1103             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1104
1105             /* Calculate temporary vectorial force */
1106             tx               = _mm256_mul_ps(fscal,dx22);
1107             ty               = _mm256_mul_ps(fscal,dy22);
1108             tz               = _mm256_mul_ps(fscal,dz22);
1109
1110             /* Update vectorial force */
1111             fix2             = _mm256_add_ps(fix2,tx);
1112             fiy2             = _mm256_add_ps(fiy2,ty);
1113             fiz2             = _mm256_add_ps(fiz2,tz);
1114
1115             fjx2             = _mm256_add_ps(fjx2,tx);
1116             fjy2             = _mm256_add_ps(fjy2,ty);
1117             fjz2             = _mm256_add_ps(fjz2,tz);
1118
1119             }
1120
1121             /**************************
1122              * CALCULATE INTERACTIONS *
1123              **************************/
1124
1125             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1126             {
1127
1128             r23              = _mm256_mul_ps(rsq23,rinv23);
1129             r23              = _mm256_andnot_ps(dummy_mask,r23);
1130
1131             /* EWALD ELECTROSTATICS */
1132
1133             /* Analytical PME correction */
1134             zeta2            = _mm256_mul_ps(beta2,rsq23);
1135             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1136             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1137             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1138             felec            = _mm256_mul_ps(qq23,felec);
1139             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1140             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1141             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv23,sh_ewald),pmecorrV);
1142             velec            = _mm256_mul_ps(qq23,velec);
1143
1144             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
1145
1146             /* Update potential sum for this i atom from the interaction with this j atom. */
1147             velec            = _mm256_and_ps(velec,cutoff_mask);
1148             velec            = _mm256_andnot_ps(dummy_mask,velec);
1149             velecsum         = _mm256_add_ps(velecsum,velec);
1150
1151             fscal            = felec;
1152
1153             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1154
1155             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1156
1157             /* Calculate temporary vectorial force */
1158             tx               = _mm256_mul_ps(fscal,dx23);
1159             ty               = _mm256_mul_ps(fscal,dy23);
1160             tz               = _mm256_mul_ps(fscal,dz23);
1161
1162             /* Update vectorial force */
1163             fix2             = _mm256_add_ps(fix2,tx);
1164             fiy2             = _mm256_add_ps(fiy2,ty);
1165             fiz2             = _mm256_add_ps(fiz2,tz);
1166
1167             fjx3             = _mm256_add_ps(fjx3,tx);
1168             fjy3             = _mm256_add_ps(fjy3,ty);
1169             fjz3             = _mm256_add_ps(fjz3,tz);
1170
1171             }
1172
1173             /**************************
1174              * CALCULATE INTERACTIONS *
1175              **************************/
1176
1177             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1178             {
1179
1180             r31              = _mm256_mul_ps(rsq31,rinv31);
1181             r31              = _mm256_andnot_ps(dummy_mask,r31);
1182
1183             /* EWALD ELECTROSTATICS */
1184
1185             /* Analytical PME correction */
1186             zeta2            = _mm256_mul_ps(beta2,rsq31);
1187             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1188             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1189             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1190             felec            = _mm256_mul_ps(qq31,felec);
1191             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1192             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1193             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv31,sh_ewald),pmecorrV);
1194             velec            = _mm256_mul_ps(qq31,velec);
1195
1196             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
1197
1198             /* Update potential sum for this i atom from the interaction with this j atom. */
1199             velec            = _mm256_and_ps(velec,cutoff_mask);
1200             velec            = _mm256_andnot_ps(dummy_mask,velec);
1201             velecsum         = _mm256_add_ps(velecsum,velec);
1202
1203             fscal            = felec;
1204
1205             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1206
1207             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1208
1209             /* Calculate temporary vectorial force */
1210             tx               = _mm256_mul_ps(fscal,dx31);
1211             ty               = _mm256_mul_ps(fscal,dy31);
1212             tz               = _mm256_mul_ps(fscal,dz31);
1213
1214             /* Update vectorial force */
1215             fix3             = _mm256_add_ps(fix3,tx);
1216             fiy3             = _mm256_add_ps(fiy3,ty);
1217             fiz3             = _mm256_add_ps(fiz3,tz);
1218
1219             fjx1             = _mm256_add_ps(fjx1,tx);
1220             fjy1             = _mm256_add_ps(fjy1,ty);
1221             fjz1             = _mm256_add_ps(fjz1,tz);
1222
1223             }
1224
1225             /**************************
1226              * CALCULATE INTERACTIONS *
1227              **************************/
1228
1229             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1230             {
1231
1232             r32              = _mm256_mul_ps(rsq32,rinv32);
1233             r32              = _mm256_andnot_ps(dummy_mask,r32);
1234
1235             /* EWALD ELECTROSTATICS */
1236
1237             /* Analytical PME correction */
1238             zeta2            = _mm256_mul_ps(beta2,rsq32);
1239             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1240             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1241             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1242             felec            = _mm256_mul_ps(qq32,felec);
1243             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1244             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1245             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv32,sh_ewald),pmecorrV);
1246             velec            = _mm256_mul_ps(qq32,velec);
1247
1248             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
1249
1250             /* Update potential sum for this i atom from the interaction with this j atom. */
1251             velec            = _mm256_and_ps(velec,cutoff_mask);
1252             velec            = _mm256_andnot_ps(dummy_mask,velec);
1253             velecsum         = _mm256_add_ps(velecsum,velec);
1254
1255             fscal            = felec;
1256
1257             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1258
1259             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1260
1261             /* Calculate temporary vectorial force */
1262             tx               = _mm256_mul_ps(fscal,dx32);
1263             ty               = _mm256_mul_ps(fscal,dy32);
1264             tz               = _mm256_mul_ps(fscal,dz32);
1265
1266             /* Update vectorial force */
1267             fix3             = _mm256_add_ps(fix3,tx);
1268             fiy3             = _mm256_add_ps(fiy3,ty);
1269             fiz3             = _mm256_add_ps(fiz3,tz);
1270
1271             fjx2             = _mm256_add_ps(fjx2,tx);
1272             fjy2             = _mm256_add_ps(fjy2,ty);
1273             fjz2             = _mm256_add_ps(fjz2,tz);
1274
1275             }
1276
1277             /**************************
1278              * CALCULATE INTERACTIONS *
1279              **************************/
1280
1281             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1282             {
1283
1284             r33              = _mm256_mul_ps(rsq33,rinv33);
1285             r33              = _mm256_andnot_ps(dummy_mask,r33);
1286
1287             /* EWALD ELECTROSTATICS */
1288
1289             /* Analytical PME correction */
1290             zeta2            = _mm256_mul_ps(beta2,rsq33);
1291             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
1292             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1293             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1294             felec            = _mm256_mul_ps(qq33,felec);
1295             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1296             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1297             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv33,sh_ewald),pmecorrV);
1298             velec            = _mm256_mul_ps(qq33,velec);
1299
1300             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
1301
1302             /* Update potential sum for this i atom from the interaction with this j atom. */
1303             velec            = _mm256_and_ps(velec,cutoff_mask);
1304             velec            = _mm256_andnot_ps(dummy_mask,velec);
1305             velecsum         = _mm256_add_ps(velecsum,velec);
1306
1307             fscal            = felec;
1308
1309             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1310
1311             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1312
1313             /* Calculate temporary vectorial force */
1314             tx               = _mm256_mul_ps(fscal,dx33);
1315             ty               = _mm256_mul_ps(fscal,dy33);
1316             tz               = _mm256_mul_ps(fscal,dz33);
1317
1318             /* Update vectorial force */
1319             fix3             = _mm256_add_ps(fix3,tx);
1320             fiy3             = _mm256_add_ps(fiy3,ty);
1321             fiz3             = _mm256_add_ps(fiz3,tz);
1322
1323             fjx3             = _mm256_add_ps(fjx3,tx);
1324             fjy3             = _mm256_add_ps(fjy3,ty);
1325             fjz3             = _mm256_add_ps(fjz3,tz);
1326
1327             }
1328
1329             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1330             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1331             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1332             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1333             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1334             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1335             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1336             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1337
1338             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1339                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
1340                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1341
1342             /* Inner loop uses 990 flops */
1343         }
1344
1345         /* End of innermost loop */
1346
1347         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1348                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1349
1350         ggid                        = gid[iidx];
1351         /* Update potential energies */
1352         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1353
1354         /* Increment number of inner iterations */
1355         inneriter                  += j_index_end - j_index_start;
1356
1357         /* Outer loop uses 19 flops */
1358     }
1359
1360     /* Increment number of outer iterations */
1361     outeriter        += nri;
1362
1363     /* Update outer/inner flops */
1364
1365     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*990);
1366 }
1367 /*
1368  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4W4_F_avx_256_single
1369  * Electrostatics interaction: Ewald
1370  * VdW interaction:            None
1371  * Geometry:                   Water4-Water4
1372  * Calculate force/pot:        Force
1373  */
1374 void
1375 nb_kernel_ElecEwSh_VdwNone_GeomW4W4_F_avx_256_single
1376                     (t_nblist * gmx_restrict                nlist,
1377                      rvec * gmx_restrict                    xx,
1378                      rvec * gmx_restrict                    ff,
1379                      t_forcerec * gmx_restrict              fr,
1380                      t_mdatoms * gmx_restrict               mdatoms,
1381                      nb_kernel_data_t * gmx_restrict        kernel_data,
1382                      t_nrnb * gmx_restrict                  nrnb)
1383 {
1384     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1385      * just 0 for non-waters.
1386      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1387      * jnr indices corresponding to data put in the four positions in the SIMD register.
1388      */
1389     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1390     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1391     int              jnrA,jnrB,jnrC,jnrD;
1392     int              jnrE,jnrF,jnrG,jnrH;
1393     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1394     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1395     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1396     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1397     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1398     real             rcutoff_scalar;
1399     real             *shiftvec,*fshift,*x,*f;
1400     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1401     real             scratch[4*DIM];
1402     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1403     real *           vdwioffsetptr1;
1404     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1405     real *           vdwioffsetptr2;
1406     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1407     real *           vdwioffsetptr3;
1408     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1409     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1410     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1411     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1412     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1413     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1414     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1415     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1416     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1417     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1418     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1419     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1420     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1421     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1422     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1423     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1424     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1425     real             *charge;
1426     __m256i          ewitab;
1427     __m128i          ewitab_lo,ewitab_hi;
1428     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1429     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1430     real             *ewtab;
1431     __m256           dummy_mask,cutoff_mask;
1432     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1433     __m256           one     = _mm256_set1_ps(1.0);
1434     __m256           two     = _mm256_set1_ps(2.0);
1435     x                = xx[0];
1436     f                = ff[0];
1437
1438     nri              = nlist->nri;
1439     iinr             = nlist->iinr;
1440     jindex           = nlist->jindex;
1441     jjnr             = nlist->jjnr;
1442     shiftidx         = nlist->shift;
1443     gid              = nlist->gid;
1444     shiftvec         = fr->shift_vec[0];
1445     fshift           = fr->fshift[0];
1446     facel            = _mm256_set1_ps(fr->epsfac);
1447     charge           = mdatoms->chargeA;
1448
1449     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1450     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
1451     beta2            = _mm256_mul_ps(beta,beta);
1452     beta3            = _mm256_mul_ps(beta,beta2);
1453
1454     ewtab            = fr->ic->tabq_coul_F;
1455     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1456     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1457
1458     /* Setup water-specific parameters */
1459     inr              = nlist->iinr[0];
1460     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1461     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1462     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1463
1464     jq1              = _mm256_set1_ps(charge[inr+1]);
1465     jq2              = _mm256_set1_ps(charge[inr+2]);
1466     jq3              = _mm256_set1_ps(charge[inr+3]);
1467     qq11             = _mm256_mul_ps(iq1,jq1);
1468     qq12             = _mm256_mul_ps(iq1,jq2);
1469     qq13             = _mm256_mul_ps(iq1,jq3);
1470     qq21             = _mm256_mul_ps(iq2,jq1);
1471     qq22             = _mm256_mul_ps(iq2,jq2);
1472     qq23             = _mm256_mul_ps(iq2,jq3);
1473     qq31             = _mm256_mul_ps(iq3,jq1);
1474     qq32             = _mm256_mul_ps(iq3,jq2);
1475     qq33             = _mm256_mul_ps(iq3,jq3);
1476
1477     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1478     rcutoff_scalar   = fr->rcoulomb;
1479     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1480     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1481
1482     /* Avoid stupid compiler warnings */
1483     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1484     j_coord_offsetA = 0;
1485     j_coord_offsetB = 0;
1486     j_coord_offsetC = 0;
1487     j_coord_offsetD = 0;
1488     j_coord_offsetE = 0;
1489     j_coord_offsetF = 0;
1490     j_coord_offsetG = 0;
1491     j_coord_offsetH = 0;
1492
1493     outeriter        = 0;
1494     inneriter        = 0;
1495
1496     for(iidx=0;iidx<4*DIM;iidx++)
1497     {
1498         scratch[iidx] = 0.0;
1499     }
1500
1501     /* Start outer loop over neighborlists */
1502     for(iidx=0; iidx<nri; iidx++)
1503     {
1504         /* Load shift vector for this list */
1505         i_shift_offset   = DIM*shiftidx[iidx];
1506
1507         /* Load limits for loop over neighbors */
1508         j_index_start    = jindex[iidx];
1509         j_index_end      = jindex[iidx+1];
1510
1511         /* Get outer coordinate index */
1512         inr              = iinr[iidx];
1513         i_coord_offset   = DIM*inr;
1514
1515         /* Load i particle coords and add shift vector */
1516         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1517                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1518
1519         fix1             = _mm256_setzero_ps();
1520         fiy1             = _mm256_setzero_ps();
1521         fiz1             = _mm256_setzero_ps();
1522         fix2             = _mm256_setzero_ps();
1523         fiy2             = _mm256_setzero_ps();
1524         fiz2             = _mm256_setzero_ps();
1525         fix3             = _mm256_setzero_ps();
1526         fiy3             = _mm256_setzero_ps();
1527         fiz3             = _mm256_setzero_ps();
1528
1529         /* Start inner kernel loop */
1530         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1531         {
1532
1533             /* Get j neighbor index, and coordinate index */
1534             jnrA             = jjnr[jidx];
1535             jnrB             = jjnr[jidx+1];
1536             jnrC             = jjnr[jidx+2];
1537             jnrD             = jjnr[jidx+3];
1538             jnrE             = jjnr[jidx+4];
1539             jnrF             = jjnr[jidx+5];
1540             jnrG             = jjnr[jidx+6];
1541             jnrH             = jjnr[jidx+7];
1542             j_coord_offsetA  = DIM*jnrA;
1543             j_coord_offsetB  = DIM*jnrB;
1544             j_coord_offsetC  = DIM*jnrC;
1545             j_coord_offsetD  = DIM*jnrD;
1546             j_coord_offsetE  = DIM*jnrE;
1547             j_coord_offsetF  = DIM*jnrF;
1548             j_coord_offsetG  = DIM*jnrG;
1549             j_coord_offsetH  = DIM*jnrH;
1550
1551             /* load j atom coordinates */
1552             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1553                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1554                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
1555                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
1556                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1557
1558             /* Calculate displacement vector */
1559             dx11             = _mm256_sub_ps(ix1,jx1);
1560             dy11             = _mm256_sub_ps(iy1,jy1);
1561             dz11             = _mm256_sub_ps(iz1,jz1);
1562             dx12             = _mm256_sub_ps(ix1,jx2);
1563             dy12             = _mm256_sub_ps(iy1,jy2);
1564             dz12             = _mm256_sub_ps(iz1,jz2);
1565             dx13             = _mm256_sub_ps(ix1,jx3);
1566             dy13             = _mm256_sub_ps(iy1,jy3);
1567             dz13             = _mm256_sub_ps(iz1,jz3);
1568             dx21             = _mm256_sub_ps(ix2,jx1);
1569             dy21             = _mm256_sub_ps(iy2,jy1);
1570             dz21             = _mm256_sub_ps(iz2,jz1);
1571             dx22             = _mm256_sub_ps(ix2,jx2);
1572             dy22             = _mm256_sub_ps(iy2,jy2);
1573             dz22             = _mm256_sub_ps(iz2,jz2);
1574             dx23             = _mm256_sub_ps(ix2,jx3);
1575             dy23             = _mm256_sub_ps(iy2,jy3);
1576             dz23             = _mm256_sub_ps(iz2,jz3);
1577             dx31             = _mm256_sub_ps(ix3,jx1);
1578             dy31             = _mm256_sub_ps(iy3,jy1);
1579             dz31             = _mm256_sub_ps(iz3,jz1);
1580             dx32             = _mm256_sub_ps(ix3,jx2);
1581             dy32             = _mm256_sub_ps(iy3,jy2);
1582             dz32             = _mm256_sub_ps(iz3,jz2);
1583             dx33             = _mm256_sub_ps(ix3,jx3);
1584             dy33             = _mm256_sub_ps(iy3,jy3);
1585             dz33             = _mm256_sub_ps(iz3,jz3);
1586
1587             /* Calculate squared distance and things based on it */
1588             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1589             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1590             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1591             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1592             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1593             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1594             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1595             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1596             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1597
1598             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1599             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1600             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
1601             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1602             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1603             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
1604             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
1605             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
1606             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
1607
1608             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1609             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1610             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1611             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1612             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1613             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1614             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1615             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1616             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1617
1618             fjx1             = _mm256_setzero_ps();
1619             fjy1             = _mm256_setzero_ps();
1620             fjz1             = _mm256_setzero_ps();
1621             fjx2             = _mm256_setzero_ps();
1622             fjy2             = _mm256_setzero_ps();
1623             fjz2             = _mm256_setzero_ps();
1624             fjx3             = _mm256_setzero_ps();
1625             fjy3             = _mm256_setzero_ps();
1626             fjz3             = _mm256_setzero_ps();
1627
1628             /**************************
1629              * CALCULATE INTERACTIONS *
1630              **************************/
1631
1632             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1633             {
1634
1635             r11              = _mm256_mul_ps(rsq11,rinv11);
1636
1637             /* EWALD ELECTROSTATICS */
1638
1639             /* Analytical PME correction */
1640             zeta2            = _mm256_mul_ps(beta2,rsq11);
1641             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1642             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1643             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1644             felec            = _mm256_mul_ps(qq11,felec);
1645
1646             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1647
1648             fscal            = felec;
1649
1650             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1651
1652             /* Calculate temporary vectorial force */
1653             tx               = _mm256_mul_ps(fscal,dx11);
1654             ty               = _mm256_mul_ps(fscal,dy11);
1655             tz               = _mm256_mul_ps(fscal,dz11);
1656
1657             /* Update vectorial force */
1658             fix1             = _mm256_add_ps(fix1,tx);
1659             fiy1             = _mm256_add_ps(fiy1,ty);
1660             fiz1             = _mm256_add_ps(fiz1,tz);
1661
1662             fjx1             = _mm256_add_ps(fjx1,tx);
1663             fjy1             = _mm256_add_ps(fjy1,ty);
1664             fjz1             = _mm256_add_ps(fjz1,tz);
1665
1666             }
1667
1668             /**************************
1669              * CALCULATE INTERACTIONS *
1670              **************************/
1671
1672             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1673             {
1674
1675             r12              = _mm256_mul_ps(rsq12,rinv12);
1676
1677             /* EWALD ELECTROSTATICS */
1678
1679             /* Analytical PME correction */
1680             zeta2            = _mm256_mul_ps(beta2,rsq12);
1681             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1682             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1683             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1684             felec            = _mm256_mul_ps(qq12,felec);
1685
1686             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1687
1688             fscal            = felec;
1689
1690             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1691
1692             /* Calculate temporary vectorial force */
1693             tx               = _mm256_mul_ps(fscal,dx12);
1694             ty               = _mm256_mul_ps(fscal,dy12);
1695             tz               = _mm256_mul_ps(fscal,dz12);
1696
1697             /* Update vectorial force */
1698             fix1             = _mm256_add_ps(fix1,tx);
1699             fiy1             = _mm256_add_ps(fiy1,ty);
1700             fiz1             = _mm256_add_ps(fiz1,tz);
1701
1702             fjx2             = _mm256_add_ps(fjx2,tx);
1703             fjy2             = _mm256_add_ps(fjy2,ty);
1704             fjz2             = _mm256_add_ps(fjz2,tz);
1705
1706             }
1707
1708             /**************************
1709              * CALCULATE INTERACTIONS *
1710              **************************/
1711
1712             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1713             {
1714
1715             r13              = _mm256_mul_ps(rsq13,rinv13);
1716
1717             /* EWALD ELECTROSTATICS */
1718
1719             /* Analytical PME correction */
1720             zeta2            = _mm256_mul_ps(beta2,rsq13);
1721             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
1722             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1723             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1724             felec            = _mm256_mul_ps(qq13,felec);
1725
1726             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
1727
1728             fscal            = felec;
1729
1730             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1731
1732             /* Calculate temporary vectorial force */
1733             tx               = _mm256_mul_ps(fscal,dx13);
1734             ty               = _mm256_mul_ps(fscal,dy13);
1735             tz               = _mm256_mul_ps(fscal,dz13);
1736
1737             /* Update vectorial force */
1738             fix1             = _mm256_add_ps(fix1,tx);
1739             fiy1             = _mm256_add_ps(fiy1,ty);
1740             fiz1             = _mm256_add_ps(fiz1,tz);
1741
1742             fjx3             = _mm256_add_ps(fjx3,tx);
1743             fjy3             = _mm256_add_ps(fjy3,ty);
1744             fjz3             = _mm256_add_ps(fjz3,tz);
1745
1746             }
1747
1748             /**************************
1749              * CALCULATE INTERACTIONS *
1750              **************************/
1751
1752             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1753             {
1754
1755             r21              = _mm256_mul_ps(rsq21,rinv21);
1756
1757             /* EWALD ELECTROSTATICS */
1758
1759             /* Analytical PME correction */
1760             zeta2            = _mm256_mul_ps(beta2,rsq21);
1761             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1762             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1763             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1764             felec            = _mm256_mul_ps(qq21,felec);
1765
1766             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1767
1768             fscal            = felec;
1769
1770             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1771
1772             /* Calculate temporary vectorial force */
1773             tx               = _mm256_mul_ps(fscal,dx21);
1774             ty               = _mm256_mul_ps(fscal,dy21);
1775             tz               = _mm256_mul_ps(fscal,dz21);
1776
1777             /* Update vectorial force */
1778             fix2             = _mm256_add_ps(fix2,tx);
1779             fiy2             = _mm256_add_ps(fiy2,ty);
1780             fiz2             = _mm256_add_ps(fiz2,tz);
1781
1782             fjx1             = _mm256_add_ps(fjx1,tx);
1783             fjy1             = _mm256_add_ps(fjy1,ty);
1784             fjz1             = _mm256_add_ps(fjz1,tz);
1785
1786             }
1787
1788             /**************************
1789              * CALCULATE INTERACTIONS *
1790              **************************/
1791
1792             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1793             {
1794
1795             r22              = _mm256_mul_ps(rsq22,rinv22);
1796
1797             /* EWALD ELECTROSTATICS */
1798
1799             /* Analytical PME correction */
1800             zeta2            = _mm256_mul_ps(beta2,rsq22);
1801             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1802             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1803             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1804             felec            = _mm256_mul_ps(qq22,felec);
1805
1806             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1807
1808             fscal            = felec;
1809
1810             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1811
1812             /* Calculate temporary vectorial force */
1813             tx               = _mm256_mul_ps(fscal,dx22);
1814             ty               = _mm256_mul_ps(fscal,dy22);
1815             tz               = _mm256_mul_ps(fscal,dz22);
1816
1817             /* Update vectorial force */
1818             fix2             = _mm256_add_ps(fix2,tx);
1819             fiy2             = _mm256_add_ps(fiy2,ty);
1820             fiz2             = _mm256_add_ps(fiz2,tz);
1821
1822             fjx2             = _mm256_add_ps(fjx2,tx);
1823             fjy2             = _mm256_add_ps(fjy2,ty);
1824             fjz2             = _mm256_add_ps(fjz2,tz);
1825
1826             }
1827
1828             /**************************
1829              * CALCULATE INTERACTIONS *
1830              **************************/
1831
1832             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1833             {
1834
1835             r23              = _mm256_mul_ps(rsq23,rinv23);
1836
1837             /* EWALD ELECTROSTATICS */
1838
1839             /* Analytical PME correction */
1840             zeta2            = _mm256_mul_ps(beta2,rsq23);
1841             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1842             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1843             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1844             felec            = _mm256_mul_ps(qq23,felec);
1845
1846             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
1847
1848             fscal            = felec;
1849
1850             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1851
1852             /* Calculate temporary vectorial force */
1853             tx               = _mm256_mul_ps(fscal,dx23);
1854             ty               = _mm256_mul_ps(fscal,dy23);
1855             tz               = _mm256_mul_ps(fscal,dz23);
1856
1857             /* Update vectorial force */
1858             fix2             = _mm256_add_ps(fix2,tx);
1859             fiy2             = _mm256_add_ps(fiy2,ty);
1860             fiz2             = _mm256_add_ps(fiz2,tz);
1861
1862             fjx3             = _mm256_add_ps(fjx3,tx);
1863             fjy3             = _mm256_add_ps(fjy3,ty);
1864             fjz3             = _mm256_add_ps(fjz3,tz);
1865
1866             }
1867
1868             /**************************
1869              * CALCULATE INTERACTIONS *
1870              **************************/
1871
1872             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1873             {
1874
1875             r31              = _mm256_mul_ps(rsq31,rinv31);
1876
1877             /* EWALD ELECTROSTATICS */
1878
1879             /* Analytical PME correction */
1880             zeta2            = _mm256_mul_ps(beta2,rsq31);
1881             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1882             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1883             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1884             felec            = _mm256_mul_ps(qq31,felec);
1885
1886             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
1887
1888             fscal            = felec;
1889
1890             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1891
1892             /* Calculate temporary vectorial force */
1893             tx               = _mm256_mul_ps(fscal,dx31);
1894             ty               = _mm256_mul_ps(fscal,dy31);
1895             tz               = _mm256_mul_ps(fscal,dz31);
1896
1897             /* Update vectorial force */
1898             fix3             = _mm256_add_ps(fix3,tx);
1899             fiy3             = _mm256_add_ps(fiy3,ty);
1900             fiz3             = _mm256_add_ps(fiz3,tz);
1901
1902             fjx1             = _mm256_add_ps(fjx1,tx);
1903             fjy1             = _mm256_add_ps(fjy1,ty);
1904             fjz1             = _mm256_add_ps(fjz1,tz);
1905
1906             }
1907
1908             /**************************
1909              * CALCULATE INTERACTIONS *
1910              **************************/
1911
1912             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1913             {
1914
1915             r32              = _mm256_mul_ps(rsq32,rinv32);
1916
1917             /* EWALD ELECTROSTATICS */
1918
1919             /* Analytical PME correction */
1920             zeta2            = _mm256_mul_ps(beta2,rsq32);
1921             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1922             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1923             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1924             felec            = _mm256_mul_ps(qq32,felec);
1925
1926             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
1927
1928             fscal            = felec;
1929
1930             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1931
1932             /* Calculate temporary vectorial force */
1933             tx               = _mm256_mul_ps(fscal,dx32);
1934             ty               = _mm256_mul_ps(fscal,dy32);
1935             tz               = _mm256_mul_ps(fscal,dz32);
1936
1937             /* Update vectorial force */
1938             fix3             = _mm256_add_ps(fix3,tx);
1939             fiy3             = _mm256_add_ps(fiy3,ty);
1940             fiz3             = _mm256_add_ps(fiz3,tz);
1941
1942             fjx2             = _mm256_add_ps(fjx2,tx);
1943             fjy2             = _mm256_add_ps(fjy2,ty);
1944             fjz2             = _mm256_add_ps(fjz2,tz);
1945
1946             }
1947
1948             /**************************
1949              * CALCULATE INTERACTIONS *
1950              **************************/
1951
1952             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1953             {
1954
1955             r33              = _mm256_mul_ps(rsq33,rinv33);
1956
1957             /* EWALD ELECTROSTATICS */
1958
1959             /* Analytical PME correction */
1960             zeta2            = _mm256_mul_ps(beta2,rsq33);
1961             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
1962             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1963             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1964             felec            = _mm256_mul_ps(qq33,felec);
1965
1966             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
1967
1968             fscal            = felec;
1969
1970             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1971
1972             /* Calculate temporary vectorial force */
1973             tx               = _mm256_mul_ps(fscal,dx33);
1974             ty               = _mm256_mul_ps(fscal,dy33);
1975             tz               = _mm256_mul_ps(fscal,dz33);
1976
1977             /* Update vectorial force */
1978             fix3             = _mm256_add_ps(fix3,tx);
1979             fiy3             = _mm256_add_ps(fiy3,ty);
1980             fiz3             = _mm256_add_ps(fiz3,tz);
1981
1982             fjx3             = _mm256_add_ps(fjx3,tx);
1983             fjy3             = _mm256_add_ps(fjy3,ty);
1984             fjz3             = _mm256_add_ps(fjz3,tz);
1985
1986             }
1987
1988             fjptrA             = f+j_coord_offsetA;
1989             fjptrB             = f+j_coord_offsetB;
1990             fjptrC             = f+j_coord_offsetC;
1991             fjptrD             = f+j_coord_offsetD;
1992             fjptrE             = f+j_coord_offsetE;
1993             fjptrF             = f+j_coord_offsetF;
1994             fjptrG             = f+j_coord_offsetG;
1995             fjptrH             = f+j_coord_offsetH;
1996
1997             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1998                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
1999                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2000
2001             /* Inner loop uses 531 flops */
2002         }
2003
2004         if(jidx<j_index_end)
2005         {
2006
2007             /* Get j neighbor index, and coordinate index */
2008             jnrlistA         = jjnr[jidx];
2009             jnrlistB         = jjnr[jidx+1];
2010             jnrlistC         = jjnr[jidx+2];
2011             jnrlistD         = jjnr[jidx+3];
2012             jnrlistE         = jjnr[jidx+4];
2013             jnrlistF         = jjnr[jidx+5];
2014             jnrlistG         = jjnr[jidx+6];
2015             jnrlistH         = jjnr[jidx+7];
2016             /* Sign of each element will be negative for non-real atoms.
2017              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2018              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2019              */
2020             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2021                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2022
2023             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2024             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2025             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2026             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2027             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2028             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2029             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2030             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2031             j_coord_offsetA  = DIM*jnrA;
2032             j_coord_offsetB  = DIM*jnrB;
2033             j_coord_offsetC  = DIM*jnrC;
2034             j_coord_offsetD  = DIM*jnrD;
2035             j_coord_offsetE  = DIM*jnrE;
2036             j_coord_offsetF  = DIM*jnrF;
2037             j_coord_offsetG  = DIM*jnrG;
2038             j_coord_offsetH  = DIM*jnrH;
2039
2040             /* load j atom coordinates */
2041             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
2042                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
2043                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
2044                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
2045                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
2046
2047             /* Calculate displacement vector */
2048             dx11             = _mm256_sub_ps(ix1,jx1);
2049             dy11             = _mm256_sub_ps(iy1,jy1);
2050             dz11             = _mm256_sub_ps(iz1,jz1);
2051             dx12             = _mm256_sub_ps(ix1,jx2);
2052             dy12             = _mm256_sub_ps(iy1,jy2);
2053             dz12             = _mm256_sub_ps(iz1,jz2);
2054             dx13             = _mm256_sub_ps(ix1,jx3);
2055             dy13             = _mm256_sub_ps(iy1,jy3);
2056             dz13             = _mm256_sub_ps(iz1,jz3);
2057             dx21             = _mm256_sub_ps(ix2,jx1);
2058             dy21             = _mm256_sub_ps(iy2,jy1);
2059             dz21             = _mm256_sub_ps(iz2,jz1);
2060             dx22             = _mm256_sub_ps(ix2,jx2);
2061             dy22             = _mm256_sub_ps(iy2,jy2);
2062             dz22             = _mm256_sub_ps(iz2,jz2);
2063             dx23             = _mm256_sub_ps(ix2,jx3);
2064             dy23             = _mm256_sub_ps(iy2,jy3);
2065             dz23             = _mm256_sub_ps(iz2,jz3);
2066             dx31             = _mm256_sub_ps(ix3,jx1);
2067             dy31             = _mm256_sub_ps(iy3,jy1);
2068             dz31             = _mm256_sub_ps(iz3,jz1);
2069             dx32             = _mm256_sub_ps(ix3,jx2);
2070             dy32             = _mm256_sub_ps(iy3,jy2);
2071             dz32             = _mm256_sub_ps(iz3,jz2);
2072             dx33             = _mm256_sub_ps(ix3,jx3);
2073             dy33             = _mm256_sub_ps(iy3,jy3);
2074             dz33             = _mm256_sub_ps(iz3,jz3);
2075
2076             /* Calculate squared distance and things based on it */
2077             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2078             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2079             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
2080             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2081             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2082             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
2083             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
2084             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
2085             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
2086
2087             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2088             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2089             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
2090             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2091             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2092             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
2093             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
2094             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
2095             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
2096
2097             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2098             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2099             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
2100             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2101             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2102             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
2103             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
2104             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
2105             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
2106
2107             fjx1             = _mm256_setzero_ps();
2108             fjy1             = _mm256_setzero_ps();
2109             fjz1             = _mm256_setzero_ps();
2110             fjx2             = _mm256_setzero_ps();
2111             fjy2             = _mm256_setzero_ps();
2112             fjz2             = _mm256_setzero_ps();
2113             fjx3             = _mm256_setzero_ps();
2114             fjy3             = _mm256_setzero_ps();
2115             fjz3             = _mm256_setzero_ps();
2116
2117             /**************************
2118              * CALCULATE INTERACTIONS *
2119              **************************/
2120
2121             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2122             {
2123
2124             r11              = _mm256_mul_ps(rsq11,rinv11);
2125             r11              = _mm256_andnot_ps(dummy_mask,r11);
2126
2127             /* EWALD ELECTROSTATICS */
2128
2129             /* Analytical PME correction */
2130             zeta2            = _mm256_mul_ps(beta2,rsq11);
2131             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
2132             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2133             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2134             felec            = _mm256_mul_ps(qq11,felec);
2135
2136             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2137
2138             fscal            = felec;
2139
2140             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2141
2142             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2143
2144             /* Calculate temporary vectorial force */
2145             tx               = _mm256_mul_ps(fscal,dx11);
2146             ty               = _mm256_mul_ps(fscal,dy11);
2147             tz               = _mm256_mul_ps(fscal,dz11);
2148
2149             /* Update vectorial force */
2150             fix1             = _mm256_add_ps(fix1,tx);
2151             fiy1             = _mm256_add_ps(fiy1,ty);
2152             fiz1             = _mm256_add_ps(fiz1,tz);
2153
2154             fjx1             = _mm256_add_ps(fjx1,tx);
2155             fjy1             = _mm256_add_ps(fjy1,ty);
2156             fjz1             = _mm256_add_ps(fjz1,tz);
2157
2158             }
2159
2160             /**************************
2161              * CALCULATE INTERACTIONS *
2162              **************************/
2163
2164             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2165             {
2166
2167             r12              = _mm256_mul_ps(rsq12,rinv12);
2168             r12              = _mm256_andnot_ps(dummy_mask,r12);
2169
2170             /* EWALD ELECTROSTATICS */
2171
2172             /* Analytical PME correction */
2173             zeta2            = _mm256_mul_ps(beta2,rsq12);
2174             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2175             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2176             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2177             felec            = _mm256_mul_ps(qq12,felec);
2178
2179             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2180
2181             fscal            = felec;
2182
2183             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2184
2185             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2186
2187             /* Calculate temporary vectorial force */
2188             tx               = _mm256_mul_ps(fscal,dx12);
2189             ty               = _mm256_mul_ps(fscal,dy12);
2190             tz               = _mm256_mul_ps(fscal,dz12);
2191
2192             /* Update vectorial force */
2193             fix1             = _mm256_add_ps(fix1,tx);
2194             fiy1             = _mm256_add_ps(fiy1,ty);
2195             fiz1             = _mm256_add_ps(fiz1,tz);
2196
2197             fjx2             = _mm256_add_ps(fjx2,tx);
2198             fjy2             = _mm256_add_ps(fjy2,ty);
2199             fjz2             = _mm256_add_ps(fjz2,tz);
2200
2201             }
2202
2203             /**************************
2204              * CALCULATE INTERACTIONS *
2205              **************************/
2206
2207             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2208             {
2209
2210             r13              = _mm256_mul_ps(rsq13,rinv13);
2211             r13              = _mm256_andnot_ps(dummy_mask,r13);
2212
2213             /* EWALD ELECTROSTATICS */
2214
2215             /* Analytical PME correction */
2216             zeta2            = _mm256_mul_ps(beta2,rsq13);
2217             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
2218             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2219             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2220             felec            = _mm256_mul_ps(qq13,felec);
2221
2222             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
2223
2224             fscal            = felec;
2225
2226             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2227
2228             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2229
2230             /* Calculate temporary vectorial force */
2231             tx               = _mm256_mul_ps(fscal,dx13);
2232             ty               = _mm256_mul_ps(fscal,dy13);
2233             tz               = _mm256_mul_ps(fscal,dz13);
2234
2235             /* Update vectorial force */
2236             fix1             = _mm256_add_ps(fix1,tx);
2237             fiy1             = _mm256_add_ps(fiy1,ty);
2238             fiz1             = _mm256_add_ps(fiz1,tz);
2239
2240             fjx3             = _mm256_add_ps(fjx3,tx);
2241             fjy3             = _mm256_add_ps(fjy3,ty);
2242             fjz3             = _mm256_add_ps(fjz3,tz);
2243
2244             }
2245
2246             /**************************
2247              * CALCULATE INTERACTIONS *
2248              **************************/
2249
2250             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2251             {
2252
2253             r21              = _mm256_mul_ps(rsq21,rinv21);
2254             r21              = _mm256_andnot_ps(dummy_mask,r21);
2255
2256             /* EWALD ELECTROSTATICS */
2257
2258             /* Analytical PME correction */
2259             zeta2            = _mm256_mul_ps(beta2,rsq21);
2260             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2261             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2262             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2263             felec            = _mm256_mul_ps(qq21,felec);
2264
2265             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2266
2267             fscal            = felec;
2268
2269             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2270
2271             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2272
2273             /* Calculate temporary vectorial force */
2274             tx               = _mm256_mul_ps(fscal,dx21);
2275             ty               = _mm256_mul_ps(fscal,dy21);
2276             tz               = _mm256_mul_ps(fscal,dz21);
2277
2278             /* Update vectorial force */
2279             fix2             = _mm256_add_ps(fix2,tx);
2280             fiy2             = _mm256_add_ps(fiy2,ty);
2281             fiz2             = _mm256_add_ps(fiz2,tz);
2282
2283             fjx1             = _mm256_add_ps(fjx1,tx);
2284             fjy1             = _mm256_add_ps(fjy1,ty);
2285             fjz1             = _mm256_add_ps(fjz1,tz);
2286
2287             }
2288
2289             /**************************
2290              * CALCULATE INTERACTIONS *
2291              **************************/
2292
2293             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2294             {
2295
2296             r22              = _mm256_mul_ps(rsq22,rinv22);
2297             r22              = _mm256_andnot_ps(dummy_mask,r22);
2298
2299             /* EWALD ELECTROSTATICS */
2300
2301             /* Analytical PME correction */
2302             zeta2            = _mm256_mul_ps(beta2,rsq22);
2303             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2304             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2305             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2306             felec            = _mm256_mul_ps(qq22,felec);
2307
2308             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2309
2310             fscal            = felec;
2311
2312             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2313
2314             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2315
2316             /* Calculate temporary vectorial force */
2317             tx               = _mm256_mul_ps(fscal,dx22);
2318             ty               = _mm256_mul_ps(fscal,dy22);
2319             tz               = _mm256_mul_ps(fscal,dz22);
2320
2321             /* Update vectorial force */
2322             fix2             = _mm256_add_ps(fix2,tx);
2323             fiy2             = _mm256_add_ps(fiy2,ty);
2324             fiz2             = _mm256_add_ps(fiz2,tz);
2325
2326             fjx2             = _mm256_add_ps(fjx2,tx);
2327             fjy2             = _mm256_add_ps(fjy2,ty);
2328             fjz2             = _mm256_add_ps(fjz2,tz);
2329
2330             }
2331
2332             /**************************
2333              * CALCULATE INTERACTIONS *
2334              **************************/
2335
2336             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2337             {
2338
2339             r23              = _mm256_mul_ps(rsq23,rinv23);
2340             r23              = _mm256_andnot_ps(dummy_mask,r23);
2341
2342             /* EWALD ELECTROSTATICS */
2343
2344             /* Analytical PME correction */
2345             zeta2            = _mm256_mul_ps(beta2,rsq23);
2346             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
2347             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2348             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2349             felec            = _mm256_mul_ps(qq23,felec);
2350
2351             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
2352
2353             fscal            = felec;
2354
2355             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2356
2357             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2358
2359             /* Calculate temporary vectorial force */
2360             tx               = _mm256_mul_ps(fscal,dx23);
2361             ty               = _mm256_mul_ps(fscal,dy23);
2362             tz               = _mm256_mul_ps(fscal,dz23);
2363
2364             /* Update vectorial force */
2365             fix2             = _mm256_add_ps(fix2,tx);
2366             fiy2             = _mm256_add_ps(fiy2,ty);
2367             fiz2             = _mm256_add_ps(fiz2,tz);
2368
2369             fjx3             = _mm256_add_ps(fjx3,tx);
2370             fjy3             = _mm256_add_ps(fjy3,ty);
2371             fjz3             = _mm256_add_ps(fjz3,tz);
2372
2373             }
2374
2375             /**************************
2376              * CALCULATE INTERACTIONS *
2377              **************************/
2378
2379             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2380             {
2381
2382             r31              = _mm256_mul_ps(rsq31,rinv31);
2383             r31              = _mm256_andnot_ps(dummy_mask,r31);
2384
2385             /* EWALD ELECTROSTATICS */
2386
2387             /* Analytical PME correction */
2388             zeta2            = _mm256_mul_ps(beta2,rsq31);
2389             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
2390             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2391             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2392             felec            = _mm256_mul_ps(qq31,felec);
2393
2394             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
2395
2396             fscal            = felec;
2397
2398             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2399
2400             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2401
2402             /* Calculate temporary vectorial force */
2403             tx               = _mm256_mul_ps(fscal,dx31);
2404             ty               = _mm256_mul_ps(fscal,dy31);
2405             tz               = _mm256_mul_ps(fscal,dz31);
2406
2407             /* Update vectorial force */
2408             fix3             = _mm256_add_ps(fix3,tx);
2409             fiy3             = _mm256_add_ps(fiy3,ty);
2410             fiz3             = _mm256_add_ps(fiz3,tz);
2411
2412             fjx1             = _mm256_add_ps(fjx1,tx);
2413             fjy1             = _mm256_add_ps(fjy1,ty);
2414             fjz1             = _mm256_add_ps(fjz1,tz);
2415
2416             }
2417
2418             /**************************
2419              * CALCULATE INTERACTIONS *
2420              **************************/
2421
2422             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2423             {
2424
2425             r32              = _mm256_mul_ps(rsq32,rinv32);
2426             r32              = _mm256_andnot_ps(dummy_mask,r32);
2427
2428             /* EWALD ELECTROSTATICS */
2429
2430             /* Analytical PME correction */
2431             zeta2            = _mm256_mul_ps(beta2,rsq32);
2432             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
2433             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2434             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2435             felec            = _mm256_mul_ps(qq32,felec);
2436
2437             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
2438
2439             fscal            = felec;
2440
2441             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2442
2443             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2444
2445             /* Calculate temporary vectorial force */
2446             tx               = _mm256_mul_ps(fscal,dx32);
2447             ty               = _mm256_mul_ps(fscal,dy32);
2448             tz               = _mm256_mul_ps(fscal,dz32);
2449
2450             /* Update vectorial force */
2451             fix3             = _mm256_add_ps(fix3,tx);
2452             fiy3             = _mm256_add_ps(fiy3,ty);
2453             fiz3             = _mm256_add_ps(fiz3,tz);
2454
2455             fjx2             = _mm256_add_ps(fjx2,tx);
2456             fjy2             = _mm256_add_ps(fjy2,ty);
2457             fjz2             = _mm256_add_ps(fjz2,tz);
2458
2459             }
2460
2461             /**************************
2462              * CALCULATE INTERACTIONS *
2463              **************************/
2464
2465             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2466             {
2467
2468             r33              = _mm256_mul_ps(rsq33,rinv33);
2469             r33              = _mm256_andnot_ps(dummy_mask,r33);
2470
2471             /* EWALD ELECTROSTATICS */
2472
2473             /* Analytical PME correction */
2474             zeta2            = _mm256_mul_ps(beta2,rsq33);
2475             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
2476             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2477             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2478             felec            = _mm256_mul_ps(qq33,felec);
2479
2480             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
2481
2482             fscal            = felec;
2483
2484             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2485
2486             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2487
2488             /* Calculate temporary vectorial force */
2489             tx               = _mm256_mul_ps(fscal,dx33);
2490             ty               = _mm256_mul_ps(fscal,dy33);
2491             tz               = _mm256_mul_ps(fscal,dz33);
2492
2493             /* Update vectorial force */
2494             fix3             = _mm256_add_ps(fix3,tx);
2495             fiy3             = _mm256_add_ps(fiy3,ty);
2496             fiz3             = _mm256_add_ps(fiz3,tz);
2497
2498             fjx3             = _mm256_add_ps(fjx3,tx);
2499             fjy3             = _mm256_add_ps(fjy3,ty);
2500             fjz3             = _mm256_add_ps(fjz3,tz);
2501
2502             }
2503
2504             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2505             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2506             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2507             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2508             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2509             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2510             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2511             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2512
2513             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2514                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
2515                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2516
2517             /* Inner loop uses 540 flops */
2518         }
2519
2520         /* End of innermost loop */
2521
2522         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2523                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
2524
2525         /* Increment number of inner iterations */
2526         inneriter                  += j_index_end - j_index_start;
2527
2528         /* Outer loop uses 18 flops */
2529     }
2530
2531     /* Increment number of outer iterations */
2532     outeriter        += nri;
2533
2534     /* Update outer/inner flops */
2535
2536     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*540);
2537 }