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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4W4_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Water4
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwNone_GeomW4W4_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr1;
  87     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  93     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  94     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  95     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  96     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
  97     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  98     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  99     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 100     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 101     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 102     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 103     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 104     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 105     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 106     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 107     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     __m256i          ewitab;
 110     __m128i          ewitab_lo,ewitab_hi;
 111     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 112     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 113     real             *ewtab;
 114     __m256           dummy_mask,cutoff_mask;
 115     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 116     __m256           one     = _mm256_set1_ps(1.0);
 117     __m256           two     = _mm256_set1_ps(2.0);
 118     x                = xx[0];
 119     f                = ff[0];
 120
 121     nri              = nlist->nri;
 122     iinr             = nlist->iinr;
 123     jindex           = nlist->jindex;
 124     jjnr             = nlist->jjnr;
 125     shiftidx         = nlist->shift;
 126     gid              = nlist->gid;
 127     shiftvec         = fr->shift_vec[0];
 128     fshift           = fr->fshift[0];
 129     facel            = _mm256_set1_ps(fr->epsfac);
 130     charge           = mdatoms->chargeA;
 131
 132     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 133     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 134     beta2            = _mm256_mul_ps(beta,beta);
 135     beta3            = _mm256_mul_ps(beta,beta2);
 136
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 144     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 145     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 146
 147     jq1              = _mm256_set1_ps(charge[inr+1]);
 148     jq2              = _mm256_set1_ps(charge[inr+2]);
 149     jq3              = _mm256_set1_ps(charge[inr+3]);
 150     qq11             = _mm256_mul_ps(iq1,jq1);
 151     qq12             = _mm256_mul_ps(iq1,jq2);
 152     qq13             = _mm256_mul_ps(iq1,jq3);
 153     qq21             = _mm256_mul_ps(iq2,jq1);
 154     qq22             = _mm256_mul_ps(iq2,jq2);
 155     qq23             = _mm256_mul_ps(iq2,jq3);
 156     qq31             = _mm256_mul_ps(iq3,jq1);
 157     qq32             = _mm256_mul_ps(iq3,jq2);
 158     qq33             = _mm256_mul_ps(iq3,jq3);
 159
 160     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 161     rcutoff_scalar   = fr->rcoulomb;
 162     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 163     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 164
 165     /* Avoid stupid compiler warnings */
 166     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 167     j_coord_offsetA = 0;
 168     j_coord_offsetB = 0;
 169     j_coord_offsetC = 0;
 170     j_coord_offsetD = 0;
 171     j_coord_offsetE = 0;
 172     j_coord_offsetF = 0;
 173     j_coord_offsetG = 0;
 174     j_coord_offsetH = 0;
 175
 176     outeriter        = 0;
 177     inneriter        = 0;
 178
 179     for(iidx=0;iidx<4*DIM;iidx++)
 180     {
 181         scratch[iidx] = 0.0;
 182     }
 183
 184     /* Start outer loop over neighborlists */
 185     for(iidx=0; iidx<nri; iidx++)
 186     {
 187         /* Load shift vector for this list */
 188         i_shift_offset   = DIM*shiftidx[iidx];
 189
 190         /* Load limits for loop over neighbors */
 191         j_index_start    = jindex[iidx];
 192         j_index_end      = jindex[iidx+1];
 193
 194         /* Get outer coordinate index */
 195         inr              = iinr[iidx];
 196         i_coord_offset   = DIM*inr;
 197
 198         /* Load i particle coords and add shift vector */
 199         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 200                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 201
 202         fix1             = _mm256_setzero_ps();
 203         fiy1             = _mm256_setzero_ps();
 204         fiz1             = _mm256_setzero_ps();
 205         fix2             = _mm256_setzero_ps();
 206         fiy2             = _mm256_setzero_ps();
 207         fiz2             = _mm256_setzero_ps();
 208         fix3             = _mm256_setzero_ps();
 209         fiy3             = _mm256_setzero_ps();
 210         fiz3             = _mm256_setzero_ps();
 211
 212         /* Reset potential sums */
 213         velecsum         = _mm256_setzero_ps();
 214
 215         /* Start inner kernel loop */
 216         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 217         {
 218
 219             /* Get j neighbor index, and coordinate index */
 220             jnrA             = jjnr[jidx];
 221             jnrB             = jjnr[jidx+1];
 222             jnrC             = jjnr[jidx+2];
 223             jnrD             = jjnr[jidx+3];
 224             jnrE             = jjnr[jidx+4];
 225             jnrF             = jjnr[jidx+5];
 226             jnrG             = jjnr[jidx+6];
 227             jnrH             = jjnr[jidx+7];
 228             j_coord_offsetA  = DIM*jnrA;
 229             j_coord_offsetB  = DIM*jnrB;
 230             j_coord_offsetC  = DIM*jnrC;
 231             j_coord_offsetD  = DIM*jnrD;
 232             j_coord_offsetE  = DIM*jnrE;
 233             j_coord_offsetF  = DIM*jnrF;
 234             j_coord_offsetG  = DIM*jnrG;
 235             j_coord_offsetH  = DIM*jnrH;
 236
 237             /* load j atom coordinates */
 238             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 239                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 240                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
 241                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
 242                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 243
 244             /* Calculate displacement vector */
 245             dx11             = _mm256_sub_ps(ix1,jx1);
 246             dy11             = _mm256_sub_ps(iy1,jy1);
 247             dz11             = _mm256_sub_ps(iz1,jz1);
 248             dx12             = _mm256_sub_ps(ix1,jx2);
 249             dy12             = _mm256_sub_ps(iy1,jy2);
 250             dz12             = _mm256_sub_ps(iz1,jz2);
 251             dx13             = _mm256_sub_ps(ix1,jx3);
 252             dy13             = _mm256_sub_ps(iy1,jy3);
 253             dz13             = _mm256_sub_ps(iz1,jz3);
 254             dx21             = _mm256_sub_ps(ix2,jx1);
 255             dy21             = _mm256_sub_ps(iy2,jy1);
 256             dz21             = _mm256_sub_ps(iz2,jz1);
 257             dx22             = _mm256_sub_ps(ix2,jx2);
 258             dy22             = _mm256_sub_ps(iy2,jy2);
 259             dz22             = _mm256_sub_ps(iz2,jz2);
 260             dx23             = _mm256_sub_ps(ix2,jx3);
 261             dy23             = _mm256_sub_ps(iy2,jy3);
 262             dz23             = _mm256_sub_ps(iz2,jz3);
 263             dx31             = _mm256_sub_ps(ix3,jx1);
 264             dy31             = _mm256_sub_ps(iy3,jy1);
 265             dz31             = _mm256_sub_ps(iz3,jz1);
 266             dx32             = _mm256_sub_ps(ix3,jx2);
 267             dy32             = _mm256_sub_ps(iy3,jy2);
 268             dz32             = _mm256_sub_ps(iz3,jz2);
 269             dx33             = _mm256_sub_ps(ix3,jx3);
 270             dy33             = _mm256_sub_ps(iy3,jy3);
 271             dz33             = _mm256_sub_ps(iz3,jz3);
 272
 273             /* Calculate squared distance and things based on it */
 274             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 275             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 276             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 277             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 278             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 279             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 280             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 281             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 282             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 283
 284             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 285             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 286             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 287             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 288             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 289             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 290             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 291             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 292             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 293
 294             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 295             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 296             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 297             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 298             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 299             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 300             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 301             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 302             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 303
 304             fjx1             = _mm256_setzero_ps();
 305             fjy1             = _mm256_setzero_ps();
 306             fjz1             = _mm256_setzero_ps();
 307             fjx2             = _mm256_setzero_ps();
 308             fjy2             = _mm256_setzero_ps();
 309             fjz2             = _mm256_setzero_ps();
 310             fjx3             = _mm256_setzero_ps();
 311             fjy3             = _mm256_setzero_ps();
 312             fjz3             = _mm256_setzero_ps();
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 319             {
 320
 321             r11              = _mm256_mul_ps(rsq11,rinv11);
 322
 323             /* EWALD ELECTROSTATICS */
 324
 325             /* Analytical PME correction */
 326             zeta2            = _mm256_mul_ps(beta2,rsq11);
 327             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 328             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 329             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 330             felec            = _mm256_mul_ps(qq11,felec);
 331             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 332             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 333             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
 334             velec            = _mm256_mul_ps(qq11,velec);
 335
 336             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm256_and_ps(velec,cutoff_mask);
 340             velecsum         = _mm256_add_ps(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm256_mul_ps(fscal,dx11);
 348             ty               = _mm256_mul_ps(fscal,dy11);
 349             tz               = _mm256_mul_ps(fscal,dz11);
 350
 351             /* Update vectorial force */
 352             fix1             = _mm256_add_ps(fix1,tx);
 353             fiy1             = _mm256_add_ps(fiy1,ty);
 354             fiz1             = _mm256_add_ps(fiz1,tz);
 355
 356             fjx1             = _mm256_add_ps(fjx1,tx);
 357             fjy1             = _mm256_add_ps(fjy1,ty);
 358             fjz1             = _mm256_add_ps(fjz1,tz);
 359
 360             }
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 367             {
 368
 369             r12              = _mm256_mul_ps(rsq12,rinv12);
 370
 371             /* EWALD ELECTROSTATICS */
 372
 373             /* Analytical PME correction */
 374             zeta2            = _mm256_mul_ps(beta2,rsq12);
 375             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 376             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 377             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 378             felec            = _mm256_mul_ps(qq12,felec);
 379             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 380             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 381             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
 382             velec            = _mm256_mul_ps(qq12,velec);
 383
 384             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velec            = _mm256_and_ps(velec,cutoff_mask);
 388             velecsum         = _mm256_add_ps(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 393
 394             /* Calculate temporary vectorial force */
 395             tx               = _mm256_mul_ps(fscal,dx12);
 396             ty               = _mm256_mul_ps(fscal,dy12);
 397             tz               = _mm256_mul_ps(fscal,dz12);
 398
 399             /* Update vectorial force */
 400             fix1             = _mm256_add_ps(fix1,tx);
 401             fiy1             = _mm256_add_ps(fiy1,ty);
 402             fiz1             = _mm256_add_ps(fiz1,tz);
 403
 404             fjx2             = _mm256_add_ps(fjx2,tx);
 405             fjy2             = _mm256_add_ps(fjy2,ty);
 406             fjz2             = _mm256_add_ps(fjz2,tz);
 407
 408             }
 409
 410             /**************************
 411              * CALCULATE INTERACTIONS *
 412              **************************/
 413
 414             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 415             {
 416
 417             r13              = _mm256_mul_ps(rsq13,rinv13);
 418
 419             /* EWALD ELECTROSTATICS */
 420
 421             /* Analytical PME correction */
 422             zeta2            = _mm256_mul_ps(beta2,rsq13);
 423             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
 424             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 425             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 426             felec            = _mm256_mul_ps(qq13,felec);
 427             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 428             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 429             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv13,sh_ewald),pmecorrV);
 430             velec            = _mm256_mul_ps(qq13,velec);
 431
 432             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
 433
 434             /* Update potential sum for this i atom from the interaction with this j atom. */
 435             velec            = _mm256_and_ps(velec,cutoff_mask);
 436             velecsum         = _mm256_add_ps(velecsum,velec);
 437
 438             fscal            = felec;
 439
 440             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 441
 442             /* Calculate temporary vectorial force */
 443             tx               = _mm256_mul_ps(fscal,dx13);
 444             ty               = _mm256_mul_ps(fscal,dy13);
 445             tz               = _mm256_mul_ps(fscal,dz13);
 446
 447             /* Update vectorial force */
 448             fix1             = _mm256_add_ps(fix1,tx);
 449             fiy1             = _mm256_add_ps(fiy1,ty);
 450             fiz1             = _mm256_add_ps(fiz1,tz);
 451
 452             fjx3             = _mm256_add_ps(fjx3,tx);
 453             fjy3             = _mm256_add_ps(fjy3,ty);
 454             fjz3             = _mm256_add_ps(fjz3,tz);
 455
 456             }
 457
 458             /**************************
 459              * CALCULATE INTERACTIONS *
 460              **************************/
 461
 462             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 463             {
 464
 465             r21              = _mm256_mul_ps(rsq21,rinv21);
 466
 467             /* EWALD ELECTROSTATICS */
 468
 469             /* Analytical PME correction */
 470             zeta2            = _mm256_mul_ps(beta2,rsq21);
 471             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 472             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 473             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 474             felec            = _mm256_mul_ps(qq21,felec);
 475             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 476             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 477             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
 478             velec            = _mm256_mul_ps(qq21,velec);
 479
 480             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velec            = _mm256_and_ps(velec,cutoff_mask);
 484             velecsum         = _mm256_add_ps(velecsum,velec);
 485
 486             fscal            = felec;
 487
 488             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 489
 490             /* Calculate temporary vectorial force */
 491             tx               = _mm256_mul_ps(fscal,dx21);
 492             ty               = _mm256_mul_ps(fscal,dy21);
 493             tz               = _mm256_mul_ps(fscal,dz21);
 494
 495             /* Update vectorial force */
 496             fix2             = _mm256_add_ps(fix2,tx);
 497             fiy2             = _mm256_add_ps(fiy2,ty);
 498             fiz2             = _mm256_add_ps(fiz2,tz);
 499
 500             fjx1             = _mm256_add_ps(fjx1,tx);
 501             fjy1             = _mm256_add_ps(fjy1,ty);
 502             fjz1             = _mm256_add_ps(fjz1,tz);
 503
 504             }
 505
 506             /**************************
 507              * CALCULATE INTERACTIONS *
 508              **************************/
 509
 510             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 511             {
 512
 513             r22              = _mm256_mul_ps(rsq22,rinv22);
 514
 515             /* EWALD ELECTROSTATICS */
 516
 517             /* Analytical PME correction */
 518             zeta2            = _mm256_mul_ps(beta2,rsq22);
 519             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 520             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 521             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 522             felec            = _mm256_mul_ps(qq22,felec);
 523             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 524             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 525             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
 526             velec            = _mm256_mul_ps(qq22,velec);
 527
 528             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             velec            = _mm256_and_ps(velec,cutoff_mask);
 532             velecsum         = _mm256_add_ps(velecsum,velec);
 533
 534             fscal            = felec;
 535
 536             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 537
 538             /* Calculate temporary vectorial force */
 539             tx               = _mm256_mul_ps(fscal,dx22);
 540             ty               = _mm256_mul_ps(fscal,dy22);
 541             tz               = _mm256_mul_ps(fscal,dz22);
 542
 543             /* Update vectorial force */
 544             fix2             = _mm256_add_ps(fix2,tx);
 545             fiy2             = _mm256_add_ps(fiy2,ty);
 546             fiz2             = _mm256_add_ps(fiz2,tz);
 547
 548             fjx2             = _mm256_add_ps(fjx2,tx);
 549             fjy2             = _mm256_add_ps(fjy2,ty);
 550             fjz2             = _mm256_add_ps(fjz2,tz);
 551
 552             }
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 559             {
 560
 561             r23              = _mm256_mul_ps(rsq23,rinv23);
 562
 563             /* EWALD ELECTROSTATICS */
 564
 565             /* Analytical PME correction */
 566             zeta2            = _mm256_mul_ps(beta2,rsq23);
 567             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
 568             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 569             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 570             felec            = _mm256_mul_ps(qq23,felec);
 571             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 572             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 573             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv23,sh_ewald),pmecorrV);
 574             velec            = _mm256_mul_ps(qq23,velec);
 575
 576             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
 577
 578             /* Update potential sum for this i atom from the interaction with this j atom. */
 579             velec            = _mm256_and_ps(velec,cutoff_mask);
 580             velecsum         = _mm256_add_ps(velecsum,velec);
 581
 582             fscal            = felec;
 583
 584             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 585
 586             /* Calculate temporary vectorial force */
 587             tx               = _mm256_mul_ps(fscal,dx23);
 588             ty               = _mm256_mul_ps(fscal,dy23);
 589             tz               = _mm256_mul_ps(fscal,dz23);
 590
 591             /* Update vectorial force */
 592             fix2             = _mm256_add_ps(fix2,tx);
 593             fiy2             = _mm256_add_ps(fiy2,ty);
 594             fiz2             = _mm256_add_ps(fiz2,tz);
 595
 596             fjx3             = _mm256_add_ps(fjx3,tx);
 597             fjy3             = _mm256_add_ps(fjy3,ty);
 598             fjz3             = _mm256_add_ps(fjz3,tz);
 599
 600             }
 601
 602             /**************************
 603              * CALCULATE INTERACTIONS *
 604              **************************/
 605
 606             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 607             {
 608
 609             r31              = _mm256_mul_ps(rsq31,rinv31);
 610
 611             /* EWALD ELECTROSTATICS */
 612
 613             /* Analytical PME correction */
 614             zeta2            = _mm256_mul_ps(beta2,rsq31);
 615             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
 616             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 617             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 618             felec            = _mm256_mul_ps(qq31,felec);
 619             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 620             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 621             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv31,sh_ewald),pmecorrV);
 622             velec            = _mm256_mul_ps(qq31,velec);
 623
 624             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             velec            = _mm256_and_ps(velec,cutoff_mask);
 628             velecsum         = _mm256_add_ps(velecsum,velec);
 629
 630             fscal            = felec;
 631
 632             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 633
 634             /* Calculate temporary vectorial force */
 635             tx               = _mm256_mul_ps(fscal,dx31);
 636             ty               = _mm256_mul_ps(fscal,dy31);
 637             tz               = _mm256_mul_ps(fscal,dz31);
 638
 639             /* Update vectorial force */
 640             fix3             = _mm256_add_ps(fix3,tx);
 641             fiy3             = _mm256_add_ps(fiy3,ty);
 642             fiz3             = _mm256_add_ps(fiz3,tz);
 643
 644             fjx1             = _mm256_add_ps(fjx1,tx);
 645             fjy1             = _mm256_add_ps(fjy1,ty);
 646             fjz1             = _mm256_add_ps(fjz1,tz);
 647
 648             }
 649
 650             /**************************
 651              * CALCULATE INTERACTIONS *
 652              **************************/
 653
 654             if (gmx_mm256_any_lt(rsq32,rcutoff2))
 655             {
 656
 657             r32              = _mm256_mul_ps(rsq32,rinv32);
 658
 659             /* EWALD ELECTROSTATICS */
 660
 661             /* Analytical PME correction */
 662             zeta2            = _mm256_mul_ps(beta2,rsq32);
 663             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
 664             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 665             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 666             felec            = _mm256_mul_ps(qq32,felec);
 667             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 668             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 669             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv32,sh_ewald),pmecorrV);
 670             velec            = _mm256_mul_ps(qq32,velec);
 671
 672             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
 673
 674             /* Update potential sum for this i atom from the interaction with this j atom. */
 675             velec            = _mm256_and_ps(velec,cutoff_mask);
 676             velecsum         = _mm256_add_ps(velecsum,velec);
 677
 678             fscal            = felec;
 679
 680             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 681
 682             /* Calculate temporary vectorial force */
 683             tx               = _mm256_mul_ps(fscal,dx32);
 684             ty               = _mm256_mul_ps(fscal,dy32);
 685             tz               = _mm256_mul_ps(fscal,dz32);
 686
 687             /* Update vectorial force */
 688             fix3             = _mm256_add_ps(fix3,tx);
 689             fiy3             = _mm256_add_ps(fiy3,ty);
 690             fiz3             = _mm256_add_ps(fiz3,tz);
 691
 692             fjx2             = _mm256_add_ps(fjx2,tx);
 693             fjy2             = _mm256_add_ps(fjy2,ty);
 694             fjz2             = _mm256_add_ps(fjz2,tz);
 695
 696             }
 697
 698             /**************************
 699              * CALCULATE INTERACTIONS *
 700              **************************/
 701
 702             if (gmx_mm256_any_lt(rsq33,rcutoff2))
 703             {
 704
 705             r33              = _mm256_mul_ps(rsq33,rinv33);
 706
 707             /* EWALD ELECTROSTATICS */
 708
 709             /* Analytical PME correction */
 710             zeta2            = _mm256_mul_ps(beta2,rsq33);
 711             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
 712             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 713             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 714             felec            = _mm256_mul_ps(qq33,felec);
 715             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 716             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 717             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv33,sh_ewald),pmecorrV);
 718             velec            = _mm256_mul_ps(qq33,velec);
 719
 720             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
 721
 722             /* Update potential sum for this i atom from the interaction with this j atom. */
 723             velec            = _mm256_and_ps(velec,cutoff_mask);
 724             velecsum         = _mm256_add_ps(velecsum,velec);
 725
 726             fscal            = felec;
 727
 728             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 729
 730             /* Calculate temporary vectorial force */
 731             tx               = _mm256_mul_ps(fscal,dx33);
 732             ty               = _mm256_mul_ps(fscal,dy33);
 733             tz               = _mm256_mul_ps(fscal,dz33);
 734
 735             /* Update vectorial force */
 736             fix3             = _mm256_add_ps(fix3,tx);
 737             fiy3             = _mm256_add_ps(fiy3,ty);
 738             fiz3             = _mm256_add_ps(fiz3,tz);
 739
 740             fjx3             = _mm256_add_ps(fjx3,tx);
 741             fjy3             = _mm256_add_ps(fjy3,ty);
 742             fjz3             = _mm256_add_ps(fjz3,tz);
 743
 744             }
 745
 746             fjptrA             = f+j_coord_offsetA;
 747             fjptrB             = f+j_coord_offsetB;
 748             fjptrC             = f+j_coord_offsetC;
 749             fjptrD             = f+j_coord_offsetD;
 750             fjptrE             = f+j_coord_offsetE;
 751             fjptrF             = f+j_coord_offsetF;
 752             fjptrG             = f+j_coord_offsetG;
 753             fjptrH             = f+j_coord_offsetH;
 754
 755             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 756                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
 757                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 758
 759             /* Inner loop uses 981 flops */
 760         }
 761
 762         if(jidx<j_index_end)
 763         {
 764
 765             /* Get j neighbor index, and coordinate index */
 766             jnrlistA         = jjnr[jidx];
 767             jnrlistB         = jjnr[jidx+1];
 768             jnrlistC         = jjnr[jidx+2];
 769             jnrlistD         = jjnr[jidx+3];
 770             jnrlistE         = jjnr[jidx+4];
 771             jnrlistF         = jjnr[jidx+5];
 772             jnrlistG         = jjnr[jidx+6];
 773             jnrlistH         = jjnr[jidx+7];
 774             /* Sign of each element will be negative for non-real atoms.
 775              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 776              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 777              */
 778             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 779                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 780
 781             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 782             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 783             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 784             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 785             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 786             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 787             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 788             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 789             j_coord_offsetA  = DIM*jnrA;
 790             j_coord_offsetB  = DIM*jnrB;
 791             j_coord_offsetC  = DIM*jnrC;
 792             j_coord_offsetD  = DIM*jnrD;
 793             j_coord_offsetE  = DIM*jnrE;
 794             j_coord_offsetF  = DIM*jnrF;
 795             j_coord_offsetG  = DIM*jnrG;
 796             j_coord_offsetH  = DIM*jnrH;
 797
 798             /* load j atom coordinates */
 799             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 800                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 801                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
 802                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
 803                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 804
 805             /* Calculate displacement vector */
 806             dx11             = _mm256_sub_ps(ix1,jx1);
 807             dy11             = _mm256_sub_ps(iy1,jy1);
 808             dz11             = _mm256_sub_ps(iz1,jz1);
 809             dx12             = _mm256_sub_ps(ix1,jx2);
 810             dy12             = _mm256_sub_ps(iy1,jy2);
 811             dz12             = _mm256_sub_ps(iz1,jz2);
 812             dx13             = _mm256_sub_ps(ix1,jx3);
 813             dy13             = _mm256_sub_ps(iy1,jy3);
 814             dz13             = _mm256_sub_ps(iz1,jz3);
 815             dx21             = _mm256_sub_ps(ix2,jx1);
 816             dy21             = _mm256_sub_ps(iy2,jy1);
 817             dz21             = _mm256_sub_ps(iz2,jz1);
 818             dx22             = _mm256_sub_ps(ix2,jx2);
 819             dy22             = _mm256_sub_ps(iy2,jy2);
 820             dz22             = _mm256_sub_ps(iz2,jz2);
 821             dx23             = _mm256_sub_ps(ix2,jx3);
 822             dy23             = _mm256_sub_ps(iy2,jy3);
 823             dz23             = _mm256_sub_ps(iz2,jz3);
 824             dx31             = _mm256_sub_ps(ix3,jx1);
 825             dy31             = _mm256_sub_ps(iy3,jy1);
 826             dz31             = _mm256_sub_ps(iz3,jz1);
 827             dx32             = _mm256_sub_ps(ix3,jx2);
 828             dy32             = _mm256_sub_ps(iy3,jy2);
 829             dz32             = _mm256_sub_ps(iz3,jz2);
 830             dx33             = _mm256_sub_ps(ix3,jx3);
 831             dy33             = _mm256_sub_ps(iy3,jy3);
 832             dz33             = _mm256_sub_ps(iz3,jz3);
 833
 834             /* Calculate squared distance and things based on it */
 835             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 836             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 837             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 838             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 839             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 840             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 841             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 842             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 843             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 844
 845             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 846             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 847             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 848             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 849             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 850             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 851             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 852             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 853             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 854
 855             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 856             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 857             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 858             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 859             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 860             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 861             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 862             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 863             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 864
 865             fjx1             = _mm256_setzero_ps();
 866             fjy1             = _mm256_setzero_ps();
 867             fjz1             = _mm256_setzero_ps();
 868             fjx2             = _mm256_setzero_ps();
 869             fjy2             = _mm256_setzero_ps();
 870             fjz2             = _mm256_setzero_ps();
 871             fjx3             = _mm256_setzero_ps();
 872             fjy3             = _mm256_setzero_ps();
 873             fjz3             = _mm256_setzero_ps();
 874
 875             /**************************
 876              * CALCULATE INTERACTIONS *
 877              **************************/
 878
 879             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 880             {
 881
 882             r11              = _mm256_mul_ps(rsq11,rinv11);
 883             r11              = _mm256_andnot_ps(dummy_mask,r11);
 884
 885             /* EWALD ELECTROSTATICS */
 886
 887             /* Analytical PME correction */
 888             zeta2            = _mm256_mul_ps(beta2,rsq11);
 889             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 890             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 891             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 892             felec            = _mm256_mul_ps(qq11,felec);
 893             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 894             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 895             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
 896             velec            = _mm256_mul_ps(qq11,velec);
 897
 898             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 899
 900             /* Update potential sum for this i atom from the interaction with this j atom. */
 901             velec            = _mm256_and_ps(velec,cutoff_mask);
 902             velec            = _mm256_andnot_ps(dummy_mask,velec);
 903             velecsum         = _mm256_add_ps(velecsum,velec);
 904
 905             fscal            = felec;
 906
 907             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 908
 909             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 910
 911             /* Calculate temporary vectorial force */
 912             tx               = _mm256_mul_ps(fscal,dx11);
 913             ty               = _mm256_mul_ps(fscal,dy11);
 914             tz               = _mm256_mul_ps(fscal,dz11);
 915
 916             /* Update vectorial force */
 917             fix1             = _mm256_add_ps(fix1,tx);
 918             fiy1             = _mm256_add_ps(fiy1,ty);
 919             fiz1             = _mm256_add_ps(fiz1,tz);
 920
 921             fjx1             = _mm256_add_ps(fjx1,tx);
 922             fjy1             = _mm256_add_ps(fjy1,ty);
 923             fjz1             = _mm256_add_ps(fjz1,tz);
 924
 925             }
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 932             {
 933
 934             r12              = _mm256_mul_ps(rsq12,rinv12);
 935             r12              = _mm256_andnot_ps(dummy_mask,r12);
 936
 937             /* EWALD ELECTROSTATICS */
 938
 939             /* Analytical PME correction */
 940             zeta2            = _mm256_mul_ps(beta2,rsq12);
 941             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 942             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 943             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 944             felec            = _mm256_mul_ps(qq12,felec);
 945             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 946             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 947             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
 948             velec            = _mm256_mul_ps(qq12,velec);
 949
 950             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 951
 952             /* Update potential sum for this i atom from the interaction with this j atom. */
 953             velec            = _mm256_and_ps(velec,cutoff_mask);
 954             velec            = _mm256_andnot_ps(dummy_mask,velec);
 955             velecsum         = _mm256_add_ps(velecsum,velec);
 956
 957             fscal            = felec;
 958
 959             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 960
 961             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 962
 963             /* Calculate temporary vectorial force */
 964             tx               = _mm256_mul_ps(fscal,dx12);
 965             ty               = _mm256_mul_ps(fscal,dy12);
 966             tz               = _mm256_mul_ps(fscal,dz12);
 967
 968             /* Update vectorial force */
 969             fix1             = _mm256_add_ps(fix1,tx);
 970             fiy1             = _mm256_add_ps(fiy1,ty);
 971             fiz1             = _mm256_add_ps(fiz1,tz);
 972
 973             fjx2             = _mm256_add_ps(fjx2,tx);
 974             fjy2             = _mm256_add_ps(fjy2,ty);
 975             fjz2             = _mm256_add_ps(fjz2,tz);
 976
 977             }
 978
 979             /**************************
 980              * CALCULATE INTERACTIONS *
 981              **************************/
 982
 983             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 984             {
 985
 986             r13              = _mm256_mul_ps(rsq13,rinv13);
 987             r13              = _mm256_andnot_ps(dummy_mask,r13);
 988
 989             /* EWALD ELECTROSTATICS */
 990
 991             /* Analytical PME correction */
 992             zeta2            = _mm256_mul_ps(beta2,rsq13);
 993             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
 994             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 995             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 996             felec            = _mm256_mul_ps(qq13,felec);
 997             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 998             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 999             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv13,sh_ewald),pmecorrV);
1000             velec            = _mm256_mul_ps(qq13,velec);
1001
1002             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
1003
1004             /* Update potential sum for this i atom from the interaction with this j atom. */
1005             velec            = _mm256_and_ps(velec,cutoff_mask);
1006             velec            = _mm256_andnot_ps(dummy_mask,velec);
1007             velecsum         = _mm256_add_ps(velecsum,velec);
1008
1009             fscal            = felec;
1010
1011             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1012
1013             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1014
1015             /* Calculate temporary vectorial force */
1016             tx               = _mm256_mul_ps(fscal,dx13);
1017             ty               = _mm256_mul_ps(fscal,dy13);
1018             tz               = _mm256_mul_ps(fscal,dz13);
1019
1020             /* Update vectorial force */
1021             fix1             = _mm256_add_ps(fix1,tx);
1022             fiy1             = _mm256_add_ps(fiy1,ty);
1023             fiz1             = _mm256_add_ps(fiz1,tz);
1024
1025             fjx3             = _mm256_add_ps(fjx3,tx);
1026             fjy3             = _mm256_add_ps(fjy3,ty);
1027             fjz3             = _mm256_add_ps(fjz3,tz);
1028
1029             }
1030
1031             /**************************
1032              * CALCULATE INTERACTIONS *
1033              **************************/
1034
1035             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1036             {
1037
1038             r21              = _mm256_mul_ps(rsq21,rinv21);
1039             r21              = _mm256_andnot_ps(dummy_mask,r21);
1040
1041             /* EWALD ELECTROSTATICS */
1042
1043             /* Analytical PME correction */
1044             zeta2            = _mm256_mul_ps(beta2,rsq21);
1045             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1046             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1047             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1048             felec            = _mm256_mul_ps(qq21,felec);
1049             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1050             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1051             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
1052             velec            = _mm256_mul_ps(qq21,velec);
1053
1054             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1055
1056             /* Update potential sum for this i atom from the interaction with this j atom. */
1057             velec            = _mm256_and_ps(velec,cutoff_mask);
1058             velec            = _mm256_andnot_ps(dummy_mask,velec);
1059             velecsum         = _mm256_add_ps(velecsum,velec);
1060
1061             fscal            = felec;
1062
1063             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1064
1065             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1066
1067             /* Calculate temporary vectorial force */
1068             tx               = _mm256_mul_ps(fscal,dx21);
1069             ty               = _mm256_mul_ps(fscal,dy21);
1070             tz               = _mm256_mul_ps(fscal,dz21);
1071
1072             /* Update vectorial force */
1073             fix2             = _mm256_add_ps(fix2,tx);
1074             fiy2             = _mm256_add_ps(fiy2,ty);
1075             fiz2             = _mm256_add_ps(fiz2,tz);
1076
1077             fjx1             = _mm256_add_ps(fjx1,tx);
1078             fjy1             = _mm256_add_ps(fjy1,ty);
1079             fjz1             = _mm256_add_ps(fjz1,tz);
1080
1081             }
1082
1083             /**************************
1084              * CALCULATE INTERACTIONS *
1085              **************************/
1086
1087             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1088             {
1089
1090             r22              = _mm256_mul_ps(rsq22,rinv22);
1091             r22              = _mm256_andnot_ps(dummy_mask,r22);
1092
1093             /* EWALD ELECTROSTATICS */
1094
1095             /* Analytical PME correction */
1096             zeta2            = _mm256_mul_ps(beta2,rsq22);
1097             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1098             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1099             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1100             felec            = _mm256_mul_ps(qq22,felec);
1101             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1102             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1103             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
1104             velec            = _mm256_mul_ps(qq22,velec);
1105
1106             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1107
1108             /* Update potential sum for this i atom from the interaction with this j atom. */
1109             velec            = _mm256_and_ps(velec,cutoff_mask);
1110             velec            = _mm256_andnot_ps(dummy_mask,velec);
1111             velecsum         = _mm256_add_ps(velecsum,velec);
1112
1113             fscal            = felec;
1114
1115             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1116
1117             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1118
1119             /* Calculate temporary vectorial force */
1120             tx               = _mm256_mul_ps(fscal,dx22);
1121             ty               = _mm256_mul_ps(fscal,dy22);
1122             tz               = _mm256_mul_ps(fscal,dz22);
1123
1124             /* Update vectorial force */
1125             fix2             = _mm256_add_ps(fix2,tx);
1126             fiy2             = _mm256_add_ps(fiy2,ty);
1127             fiz2             = _mm256_add_ps(fiz2,tz);
1128
1129             fjx2             = _mm256_add_ps(fjx2,tx);
1130             fjy2             = _mm256_add_ps(fjy2,ty);
1131             fjz2             = _mm256_add_ps(fjz2,tz);
1132
1133             }
1134
1135             /**************************
1136              * CALCULATE INTERACTIONS *
1137              **************************/
1138
1139             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1140             {
1141
1142             r23              = _mm256_mul_ps(rsq23,rinv23);
1143             r23              = _mm256_andnot_ps(dummy_mask,r23);
1144
1145             /* EWALD ELECTROSTATICS */
1146
1147             /* Analytical PME correction */
1148             zeta2            = _mm256_mul_ps(beta2,rsq23);
1149             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1150             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1151             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1152             felec            = _mm256_mul_ps(qq23,felec);
1153             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1154             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1155             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv23,sh_ewald),pmecorrV);
1156             velec            = _mm256_mul_ps(qq23,velec);
1157
1158             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
1159
1160             /* Update potential sum for this i atom from the interaction with this j atom. */
1161             velec            = _mm256_and_ps(velec,cutoff_mask);
1162             velec            = _mm256_andnot_ps(dummy_mask,velec);
1163             velecsum         = _mm256_add_ps(velecsum,velec);
1164
1165             fscal            = felec;
1166
1167             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1168
1169             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1170
1171             /* Calculate temporary vectorial force */
1172             tx               = _mm256_mul_ps(fscal,dx23);
1173             ty               = _mm256_mul_ps(fscal,dy23);
1174             tz               = _mm256_mul_ps(fscal,dz23);
1175
1176             /* Update vectorial force */
1177             fix2             = _mm256_add_ps(fix2,tx);
1178             fiy2             = _mm256_add_ps(fiy2,ty);
1179             fiz2             = _mm256_add_ps(fiz2,tz);
1180
1181             fjx3             = _mm256_add_ps(fjx3,tx);
1182             fjy3             = _mm256_add_ps(fjy3,ty);
1183             fjz3             = _mm256_add_ps(fjz3,tz);
1184
1185             }
1186
1187             /**************************
1188              * CALCULATE INTERACTIONS *
1189              **************************/
1190
1191             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1192             {
1193
1194             r31              = _mm256_mul_ps(rsq31,rinv31);
1195             r31              = _mm256_andnot_ps(dummy_mask,r31);
1196
1197             /* EWALD ELECTROSTATICS */
1198
1199             /* Analytical PME correction */
1200             zeta2            = _mm256_mul_ps(beta2,rsq31);
1201             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1202             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1203             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1204             felec            = _mm256_mul_ps(qq31,felec);
1205             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1206             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1207             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv31,sh_ewald),pmecorrV);
1208             velec            = _mm256_mul_ps(qq31,velec);
1209
1210             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
1211
1212             /* Update potential sum for this i atom from the interaction with this j atom. */
1213             velec            = _mm256_and_ps(velec,cutoff_mask);
1214             velec            = _mm256_andnot_ps(dummy_mask,velec);
1215             velecsum         = _mm256_add_ps(velecsum,velec);
1216
1217             fscal            = felec;
1218
1219             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1220
1221             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1222
1223             /* Calculate temporary vectorial force */
1224             tx               = _mm256_mul_ps(fscal,dx31);
1225             ty               = _mm256_mul_ps(fscal,dy31);
1226             tz               = _mm256_mul_ps(fscal,dz31);
1227
1228             /* Update vectorial force */
1229             fix3             = _mm256_add_ps(fix3,tx);
1230             fiy3             = _mm256_add_ps(fiy3,ty);
1231             fiz3             = _mm256_add_ps(fiz3,tz);
1232
1233             fjx1             = _mm256_add_ps(fjx1,tx);
1234             fjy1             = _mm256_add_ps(fjy1,ty);
1235             fjz1             = _mm256_add_ps(fjz1,tz);
1236
1237             }
1238
1239             /**************************
1240              * CALCULATE INTERACTIONS *
1241              **************************/
1242
1243             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1244             {
1245
1246             r32              = _mm256_mul_ps(rsq32,rinv32);
1247             r32              = _mm256_andnot_ps(dummy_mask,r32);
1248
1249             /* EWALD ELECTROSTATICS */
1250
1251             /* Analytical PME correction */
1252             zeta2            = _mm256_mul_ps(beta2,rsq32);
1253             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1254             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1255             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1256             felec            = _mm256_mul_ps(qq32,felec);
1257             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1258             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1259             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv32,sh_ewald),pmecorrV);
1260             velec            = _mm256_mul_ps(qq32,velec);
1261
1262             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
1263
1264             /* Update potential sum for this i atom from the interaction with this j atom. */
1265             velec            = _mm256_and_ps(velec,cutoff_mask);
1266             velec            = _mm256_andnot_ps(dummy_mask,velec);
1267             velecsum         = _mm256_add_ps(velecsum,velec);
1268
1269             fscal            = felec;
1270
1271             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1272
1273             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1274
1275             /* Calculate temporary vectorial force */
1276             tx               = _mm256_mul_ps(fscal,dx32);
1277             ty               = _mm256_mul_ps(fscal,dy32);
1278             tz               = _mm256_mul_ps(fscal,dz32);
1279
1280             /* Update vectorial force */
1281             fix3             = _mm256_add_ps(fix3,tx);
1282             fiy3             = _mm256_add_ps(fiy3,ty);
1283             fiz3             = _mm256_add_ps(fiz3,tz);
1284
1285             fjx2             = _mm256_add_ps(fjx2,tx);
1286             fjy2             = _mm256_add_ps(fjy2,ty);
1287             fjz2             = _mm256_add_ps(fjz2,tz);
1288
1289             }
1290
1291             /**************************
1292              * CALCULATE INTERACTIONS *
1293              **************************/
1294
1295             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1296             {
1297
1298             r33              = _mm256_mul_ps(rsq33,rinv33);
1299             r33              = _mm256_andnot_ps(dummy_mask,r33);
1300
1301             /* EWALD ELECTROSTATICS */
1302
1303             /* Analytical PME correction */
1304             zeta2            = _mm256_mul_ps(beta2,rsq33);
1305             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
1306             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1307             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1308             felec            = _mm256_mul_ps(qq33,felec);
1309             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1310             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1311             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv33,sh_ewald),pmecorrV);
1312             velec            = _mm256_mul_ps(qq33,velec);
1313
1314             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
1315
1316             /* Update potential sum for this i atom from the interaction with this j atom. */
1317             velec            = _mm256_and_ps(velec,cutoff_mask);
1318             velec            = _mm256_andnot_ps(dummy_mask,velec);
1319             velecsum         = _mm256_add_ps(velecsum,velec);
1320
1321             fscal            = felec;
1322
1323             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1324
1325             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1326
1327             /* Calculate temporary vectorial force */
1328             tx               = _mm256_mul_ps(fscal,dx33);
1329             ty               = _mm256_mul_ps(fscal,dy33);
1330             tz               = _mm256_mul_ps(fscal,dz33);
1331
1332             /* Update vectorial force */
1333             fix3             = _mm256_add_ps(fix3,tx);
1334             fiy3             = _mm256_add_ps(fiy3,ty);
1335             fiz3             = _mm256_add_ps(fiz3,tz);
1336
1337             fjx3             = _mm256_add_ps(fjx3,tx);
1338             fjy3             = _mm256_add_ps(fjy3,ty);
1339             fjz3             = _mm256_add_ps(fjz3,tz);
1340
1341             }
1342
1343             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1344             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1345             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1346             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1347             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1348             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1349             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1350             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1351
1352             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1353                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
1354                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1355
1356             /* Inner loop uses 990 flops */
1357         }
1358
1359         /* End of innermost loop */
1360
1361         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1362                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1363
1364         ggid                        = gid[iidx];
1365         /* Update potential energies */
1366         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1367
1368         /* Increment number of inner iterations */
1369         inneriter                  += j_index_end - j_index_start;
1370
1371         /* Outer loop uses 19 flops */
1372     }
1373
1374     /* Increment number of outer iterations */
1375     outeriter        += nri;
1376
1377     /* Update outer/inner flops */
1378
1379     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*990);
1380 }
1381 /*
1382  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4W4_F_avx_256_single
1383  * Electrostatics interaction: Ewald
1384  * VdW interaction:            None
1385  * Geometry:                   Water4-Water4
1386  * Calculate force/pot:        Force
1387  */
1388 void
1389 nb_kernel_ElecEwSh_VdwNone_GeomW4W4_F_avx_256_single
1390                     (t_nblist                    * gmx_restrict       nlist,
1391                      rvec                        * gmx_restrict          xx,
1392                      rvec                        * gmx_restrict          ff,
1393                      t_forcerec                  * gmx_restrict          fr,
1394                      t_mdatoms                   * gmx_restrict     mdatoms,
1395                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1396                      t_nrnb                      * gmx_restrict        nrnb)
1397 {
1398     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1399      * just 0 for non-waters.
1400      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1401      * jnr indices corresponding to data put in the four positions in the SIMD register.
1402      */
1403     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1404     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1405     int              jnrA,jnrB,jnrC,jnrD;
1406     int              jnrE,jnrF,jnrG,jnrH;
1407     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1408     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1409     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1410     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1411     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1412     real             rcutoff_scalar;
1413     real             *shiftvec,*fshift,*x,*f;
1414     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1415     real             scratch[4*DIM];
1416     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1417     real *           vdwioffsetptr1;
1418     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1419     real *           vdwioffsetptr2;
1420     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1421     real *           vdwioffsetptr3;
1422     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1423     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1424     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1425     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1426     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1427     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1428     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1429     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1430     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1431     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1432     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1433     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1434     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1435     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1436     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1437     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1438     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1439     real             *charge;
1440     __m256i          ewitab;
1441     __m128i          ewitab_lo,ewitab_hi;
1442     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1443     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1444     real             *ewtab;
1445     __m256           dummy_mask,cutoff_mask;
1446     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1447     __m256           one     = _mm256_set1_ps(1.0);
1448     __m256           two     = _mm256_set1_ps(2.0);
1449     x                = xx[0];
1450     f                = ff[0];
1451
1452     nri              = nlist->nri;
1453     iinr             = nlist->iinr;
1454     jindex           = nlist->jindex;
1455     jjnr             = nlist->jjnr;
1456     shiftidx         = nlist->shift;
1457     gid              = nlist->gid;
1458     shiftvec         = fr->shift_vec[0];
1459     fshift           = fr->fshift[0];
1460     facel            = _mm256_set1_ps(fr->epsfac);
1461     charge           = mdatoms->chargeA;
1462
1463     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1464     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1465     beta2            = _mm256_mul_ps(beta,beta);
1466     beta3            = _mm256_mul_ps(beta,beta2);
1467
1468     ewtab            = fr->ic->tabq_coul_F;
1469     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1470     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1471
1472     /* Setup water-specific parameters */
1473     inr              = nlist->iinr[0];
1474     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1475     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1476     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1477
1478     jq1              = _mm256_set1_ps(charge[inr+1]);
1479     jq2              = _mm256_set1_ps(charge[inr+2]);
1480     jq3              = _mm256_set1_ps(charge[inr+3]);
1481     qq11             = _mm256_mul_ps(iq1,jq1);
1482     qq12             = _mm256_mul_ps(iq1,jq2);
1483     qq13             = _mm256_mul_ps(iq1,jq3);
1484     qq21             = _mm256_mul_ps(iq2,jq1);
1485     qq22             = _mm256_mul_ps(iq2,jq2);
1486     qq23             = _mm256_mul_ps(iq2,jq3);
1487     qq31             = _mm256_mul_ps(iq3,jq1);
1488     qq32             = _mm256_mul_ps(iq3,jq2);
1489     qq33             = _mm256_mul_ps(iq3,jq3);
1490
1491     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1492     rcutoff_scalar   = fr->rcoulomb;
1493     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1494     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1495
1496     /* Avoid stupid compiler warnings */
1497     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1498     j_coord_offsetA = 0;
1499     j_coord_offsetB = 0;
1500     j_coord_offsetC = 0;
1501     j_coord_offsetD = 0;
1502     j_coord_offsetE = 0;
1503     j_coord_offsetF = 0;
1504     j_coord_offsetG = 0;
1505     j_coord_offsetH = 0;
1506
1507     outeriter        = 0;
1508     inneriter        = 0;
1509
1510     for(iidx=0;iidx<4*DIM;iidx++)
1511     {
1512         scratch[iidx] = 0.0;
1513     }
1514
1515     /* Start outer loop over neighborlists */
1516     for(iidx=0; iidx<nri; iidx++)
1517     {
1518         /* Load shift vector for this list */
1519         i_shift_offset   = DIM*shiftidx[iidx];
1520
1521         /* Load limits for loop over neighbors */
1522         j_index_start    = jindex[iidx];
1523         j_index_end      = jindex[iidx+1];
1524
1525         /* Get outer coordinate index */
1526         inr              = iinr[iidx];
1527         i_coord_offset   = DIM*inr;
1528
1529         /* Load i particle coords and add shift vector */
1530         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1531                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1532
1533         fix1             = _mm256_setzero_ps();
1534         fiy1             = _mm256_setzero_ps();
1535         fiz1             = _mm256_setzero_ps();
1536         fix2             = _mm256_setzero_ps();
1537         fiy2             = _mm256_setzero_ps();
1538         fiz2             = _mm256_setzero_ps();
1539         fix3             = _mm256_setzero_ps();
1540         fiy3             = _mm256_setzero_ps();
1541         fiz3             = _mm256_setzero_ps();
1542
1543         /* Start inner kernel loop */
1544         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1545         {
1546
1547             /* Get j neighbor index, and coordinate index */
1548             jnrA             = jjnr[jidx];
1549             jnrB             = jjnr[jidx+1];
1550             jnrC             = jjnr[jidx+2];
1551             jnrD             = jjnr[jidx+3];
1552             jnrE             = jjnr[jidx+4];
1553             jnrF             = jjnr[jidx+5];
1554             jnrG             = jjnr[jidx+6];
1555             jnrH             = jjnr[jidx+7];
1556             j_coord_offsetA  = DIM*jnrA;
1557             j_coord_offsetB  = DIM*jnrB;
1558             j_coord_offsetC  = DIM*jnrC;
1559             j_coord_offsetD  = DIM*jnrD;
1560             j_coord_offsetE  = DIM*jnrE;
1561             j_coord_offsetF  = DIM*jnrF;
1562             j_coord_offsetG  = DIM*jnrG;
1563             j_coord_offsetH  = DIM*jnrH;
1564
1565             /* load j atom coordinates */
1566             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1567                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1568                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
1569                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
1570                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1571
1572             /* Calculate displacement vector */
1573             dx11             = _mm256_sub_ps(ix1,jx1);
1574             dy11             = _mm256_sub_ps(iy1,jy1);
1575             dz11             = _mm256_sub_ps(iz1,jz1);
1576             dx12             = _mm256_sub_ps(ix1,jx2);
1577             dy12             = _mm256_sub_ps(iy1,jy2);
1578             dz12             = _mm256_sub_ps(iz1,jz2);
1579             dx13             = _mm256_sub_ps(ix1,jx3);
1580             dy13             = _mm256_sub_ps(iy1,jy3);
1581             dz13             = _mm256_sub_ps(iz1,jz3);
1582             dx21             = _mm256_sub_ps(ix2,jx1);
1583             dy21             = _mm256_sub_ps(iy2,jy1);
1584             dz21             = _mm256_sub_ps(iz2,jz1);
1585             dx22             = _mm256_sub_ps(ix2,jx2);
1586             dy22             = _mm256_sub_ps(iy2,jy2);
1587             dz22             = _mm256_sub_ps(iz2,jz2);
1588             dx23             = _mm256_sub_ps(ix2,jx3);
1589             dy23             = _mm256_sub_ps(iy2,jy3);
1590             dz23             = _mm256_sub_ps(iz2,jz3);
1591             dx31             = _mm256_sub_ps(ix3,jx1);
1592             dy31             = _mm256_sub_ps(iy3,jy1);
1593             dz31             = _mm256_sub_ps(iz3,jz1);
1594             dx32             = _mm256_sub_ps(ix3,jx2);
1595             dy32             = _mm256_sub_ps(iy3,jy2);
1596             dz32             = _mm256_sub_ps(iz3,jz2);
1597             dx33             = _mm256_sub_ps(ix3,jx3);
1598             dy33             = _mm256_sub_ps(iy3,jy3);
1599             dz33             = _mm256_sub_ps(iz3,jz3);
1600
1601             /* Calculate squared distance and things based on it */
1602             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1603             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1604             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1605             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1606             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1607             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1608             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1609             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1610             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1611
1612             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1613             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1614             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
1615             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1616             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1617             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
1618             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
1619             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
1620             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
1621
1622             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1623             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1624             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1625             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1626             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1627             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1628             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1629             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1630             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1631
1632             fjx1             = _mm256_setzero_ps();
1633             fjy1             = _mm256_setzero_ps();
1634             fjz1             = _mm256_setzero_ps();
1635             fjx2             = _mm256_setzero_ps();
1636             fjy2             = _mm256_setzero_ps();
1637             fjz2             = _mm256_setzero_ps();
1638             fjx3             = _mm256_setzero_ps();
1639             fjy3             = _mm256_setzero_ps();
1640             fjz3             = _mm256_setzero_ps();
1641
1642             /**************************
1643              * CALCULATE INTERACTIONS *
1644              **************************/
1645
1646             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1647             {
1648
1649             r11              = _mm256_mul_ps(rsq11,rinv11);
1650
1651             /* EWALD ELECTROSTATICS */
1652
1653             /* Analytical PME correction */
1654             zeta2            = _mm256_mul_ps(beta2,rsq11);
1655             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1656             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1657             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1658             felec            = _mm256_mul_ps(qq11,felec);
1659
1660             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1661
1662             fscal            = felec;
1663
1664             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1665
1666             /* Calculate temporary vectorial force */
1667             tx               = _mm256_mul_ps(fscal,dx11);
1668             ty               = _mm256_mul_ps(fscal,dy11);
1669             tz               = _mm256_mul_ps(fscal,dz11);
1670
1671             /* Update vectorial force */
1672             fix1             = _mm256_add_ps(fix1,tx);
1673             fiy1             = _mm256_add_ps(fiy1,ty);
1674             fiz1             = _mm256_add_ps(fiz1,tz);
1675
1676             fjx1             = _mm256_add_ps(fjx1,tx);
1677             fjy1             = _mm256_add_ps(fjy1,ty);
1678             fjz1             = _mm256_add_ps(fjz1,tz);
1679
1680             }
1681
1682             /**************************
1683              * CALCULATE INTERACTIONS *
1684              **************************/
1685
1686             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1687             {
1688
1689             r12              = _mm256_mul_ps(rsq12,rinv12);
1690
1691             /* EWALD ELECTROSTATICS */
1692
1693             /* Analytical PME correction */
1694             zeta2            = _mm256_mul_ps(beta2,rsq12);
1695             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1696             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1697             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1698             felec            = _mm256_mul_ps(qq12,felec);
1699
1700             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1701
1702             fscal            = felec;
1703
1704             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1705
1706             /* Calculate temporary vectorial force */
1707             tx               = _mm256_mul_ps(fscal,dx12);
1708             ty               = _mm256_mul_ps(fscal,dy12);
1709             tz               = _mm256_mul_ps(fscal,dz12);
1710
1711             /* Update vectorial force */
1712             fix1             = _mm256_add_ps(fix1,tx);
1713             fiy1             = _mm256_add_ps(fiy1,ty);
1714             fiz1             = _mm256_add_ps(fiz1,tz);
1715
1716             fjx2             = _mm256_add_ps(fjx2,tx);
1717             fjy2             = _mm256_add_ps(fjy2,ty);
1718             fjz2             = _mm256_add_ps(fjz2,tz);
1719
1720             }
1721
1722             /**************************
1723              * CALCULATE INTERACTIONS *
1724              **************************/
1725
1726             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1727             {
1728
1729             r13              = _mm256_mul_ps(rsq13,rinv13);
1730
1731             /* EWALD ELECTROSTATICS */
1732
1733             /* Analytical PME correction */
1734             zeta2            = _mm256_mul_ps(beta2,rsq13);
1735             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
1736             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1737             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1738             felec            = _mm256_mul_ps(qq13,felec);
1739
1740             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
1741
1742             fscal            = felec;
1743
1744             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1745
1746             /* Calculate temporary vectorial force */
1747             tx               = _mm256_mul_ps(fscal,dx13);
1748             ty               = _mm256_mul_ps(fscal,dy13);
1749             tz               = _mm256_mul_ps(fscal,dz13);
1750
1751             /* Update vectorial force */
1752             fix1             = _mm256_add_ps(fix1,tx);
1753             fiy1             = _mm256_add_ps(fiy1,ty);
1754             fiz1             = _mm256_add_ps(fiz1,tz);
1755
1756             fjx3             = _mm256_add_ps(fjx3,tx);
1757             fjy3             = _mm256_add_ps(fjy3,ty);
1758             fjz3             = _mm256_add_ps(fjz3,tz);
1759
1760             }
1761
1762             /**************************
1763              * CALCULATE INTERACTIONS *
1764              **************************/
1765
1766             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1767             {
1768
1769             r21              = _mm256_mul_ps(rsq21,rinv21);
1770
1771             /* EWALD ELECTROSTATICS */
1772
1773             /* Analytical PME correction */
1774             zeta2            = _mm256_mul_ps(beta2,rsq21);
1775             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1776             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1777             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1778             felec            = _mm256_mul_ps(qq21,felec);
1779
1780             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1781
1782             fscal            = felec;
1783
1784             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1785
1786             /* Calculate temporary vectorial force */
1787             tx               = _mm256_mul_ps(fscal,dx21);
1788             ty               = _mm256_mul_ps(fscal,dy21);
1789             tz               = _mm256_mul_ps(fscal,dz21);
1790
1791             /* Update vectorial force */
1792             fix2             = _mm256_add_ps(fix2,tx);
1793             fiy2             = _mm256_add_ps(fiy2,ty);
1794             fiz2             = _mm256_add_ps(fiz2,tz);
1795
1796             fjx1             = _mm256_add_ps(fjx1,tx);
1797             fjy1             = _mm256_add_ps(fjy1,ty);
1798             fjz1             = _mm256_add_ps(fjz1,tz);
1799
1800             }
1801
1802             /**************************
1803              * CALCULATE INTERACTIONS *
1804              **************************/
1805
1806             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1807             {
1808
1809             r22              = _mm256_mul_ps(rsq22,rinv22);
1810
1811             /* EWALD ELECTROSTATICS */
1812
1813             /* Analytical PME correction */
1814             zeta2            = _mm256_mul_ps(beta2,rsq22);
1815             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1816             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1817             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1818             felec            = _mm256_mul_ps(qq22,felec);
1819
1820             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1821
1822             fscal            = felec;
1823
1824             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1825
1826             /* Calculate temporary vectorial force */
1827             tx               = _mm256_mul_ps(fscal,dx22);
1828             ty               = _mm256_mul_ps(fscal,dy22);
1829             tz               = _mm256_mul_ps(fscal,dz22);
1830
1831             /* Update vectorial force */
1832             fix2             = _mm256_add_ps(fix2,tx);
1833             fiy2             = _mm256_add_ps(fiy2,ty);
1834             fiz2             = _mm256_add_ps(fiz2,tz);
1835
1836             fjx2             = _mm256_add_ps(fjx2,tx);
1837             fjy2             = _mm256_add_ps(fjy2,ty);
1838             fjz2             = _mm256_add_ps(fjz2,tz);
1839
1840             }
1841
1842             /**************************
1843              * CALCULATE INTERACTIONS *
1844              **************************/
1845
1846             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1847             {
1848
1849             r23              = _mm256_mul_ps(rsq23,rinv23);
1850
1851             /* EWALD ELECTROSTATICS */
1852
1853             /* Analytical PME correction */
1854             zeta2            = _mm256_mul_ps(beta2,rsq23);
1855             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1856             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1857             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1858             felec            = _mm256_mul_ps(qq23,felec);
1859
1860             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
1861
1862             fscal            = felec;
1863
1864             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1865
1866             /* Calculate temporary vectorial force */
1867             tx               = _mm256_mul_ps(fscal,dx23);
1868             ty               = _mm256_mul_ps(fscal,dy23);
1869             tz               = _mm256_mul_ps(fscal,dz23);
1870
1871             /* Update vectorial force */
1872             fix2             = _mm256_add_ps(fix2,tx);
1873             fiy2             = _mm256_add_ps(fiy2,ty);
1874             fiz2             = _mm256_add_ps(fiz2,tz);
1875
1876             fjx3             = _mm256_add_ps(fjx3,tx);
1877             fjy3             = _mm256_add_ps(fjy3,ty);
1878             fjz3             = _mm256_add_ps(fjz3,tz);
1879
1880             }
1881
1882             /**************************
1883              * CALCULATE INTERACTIONS *
1884              **************************/
1885
1886             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1887             {
1888
1889             r31              = _mm256_mul_ps(rsq31,rinv31);
1890
1891             /* EWALD ELECTROSTATICS */
1892
1893             /* Analytical PME correction */
1894             zeta2            = _mm256_mul_ps(beta2,rsq31);
1895             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1896             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1897             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1898             felec            = _mm256_mul_ps(qq31,felec);
1899
1900             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
1901
1902             fscal            = felec;
1903
1904             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1905
1906             /* Calculate temporary vectorial force */
1907             tx               = _mm256_mul_ps(fscal,dx31);
1908             ty               = _mm256_mul_ps(fscal,dy31);
1909             tz               = _mm256_mul_ps(fscal,dz31);
1910
1911             /* Update vectorial force */
1912             fix3             = _mm256_add_ps(fix3,tx);
1913             fiy3             = _mm256_add_ps(fiy3,ty);
1914             fiz3             = _mm256_add_ps(fiz3,tz);
1915
1916             fjx1             = _mm256_add_ps(fjx1,tx);
1917             fjy1             = _mm256_add_ps(fjy1,ty);
1918             fjz1             = _mm256_add_ps(fjz1,tz);
1919
1920             }
1921
1922             /**************************
1923              * CALCULATE INTERACTIONS *
1924              **************************/
1925
1926             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1927             {
1928
1929             r32              = _mm256_mul_ps(rsq32,rinv32);
1930
1931             /* EWALD ELECTROSTATICS */
1932
1933             /* Analytical PME correction */
1934             zeta2            = _mm256_mul_ps(beta2,rsq32);
1935             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1936             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1937             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1938             felec            = _mm256_mul_ps(qq32,felec);
1939
1940             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
1941
1942             fscal            = felec;
1943
1944             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1945
1946             /* Calculate temporary vectorial force */
1947             tx               = _mm256_mul_ps(fscal,dx32);
1948             ty               = _mm256_mul_ps(fscal,dy32);
1949             tz               = _mm256_mul_ps(fscal,dz32);
1950
1951             /* Update vectorial force */
1952             fix3             = _mm256_add_ps(fix3,tx);
1953             fiy3             = _mm256_add_ps(fiy3,ty);
1954             fiz3             = _mm256_add_ps(fiz3,tz);
1955
1956             fjx2             = _mm256_add_ps(fjx2,tx);
1957             fjy2             = _mm256_add_ps(fjy2,ty);
1958             fjz2             = _mm256_add_ps(fjz2,tz);
1959
1960             }
1961
1962             /**************************
1963              * CALCULATE INTERACTIONS *
1964              **************************/
1965
1966             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1967             {
1968
1969             r33              = _mm256_mul_ps(rsq33,rinv33);
1970
1971             /* EWALD ELECTROSTATICS */
1972
1973             /* Analytical PME correction */
1974             zeta2            = _mm256_mul_ps(beta2,rsq33);
1975             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
1976             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1977             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1978             felec            = _mm256_mul_ps(qq33,felec);
1979
1980             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
1981
1982             fscal            = felec;
1983
1984             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1985
1986             /* Calculate temporary vectorial force */
1987             tx               = _mm256_mul_ps(fscal,dx33);
1988             ty               = _mm256_mul_ps(fscal,dy33);
1989             tz               = _mm256_mul_ps(fscal,dz33);
1990
1991             /* Update vectorial force */
1992             fix3             = _mm256_add_ps(fix3,tx);
1993             fiy3             = _mm256_add_ps(fiy3,ty);
1994             fiz3             = _mm256_add_ps(fiz3,tz);
1995
1996             fjx3             = _mm256_add_ps(fjx3,tx);
1997             fjy3             = _mm256_add_ps(fjy3,ty);
1998             fjz3             = _mm256_add_ps(fjz3,tz);
1999
2000             }
2001
2002             fjptrA             = f+j_coord_offsetA;
2003             fjptrB             = f+j_coord_offsetB;
2004             fjptrC             = f+j_coord_offsetC;
2005             fjptrD             = f+j_coord_offsetD;
2006             fjptrE             = f+j_coord_offsetE;
2007             fjptrF             = f+j_coord_offsetF;
2008             fjptrG             = f+j_coord_offsetG;
2009             fjptrH             = f+j_coord_offsetH;
2010
2011             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2012                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
2013                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2014
2015             /* Inner loop uses 531 flops */
2016         }
2017
2018         if(jidx<j_index_end)
2019         {
2020
2021             /* Get j neighbor index, and coordinate index */
2022             jnrlistA         = jjnr[jidx];
2023             jnrlistB         = jjnr[jidx+1];
2024             jnrlistC         = jjnr[jidx+2];
2025             jnrlistD         = jjnr[jidx+3];
2026             jnrlistE         = jjnr[jidx+4];
2027             jnrlistF         = jjnr[jidx+5];
2028             jnrlistG         = jjnr[jidx+6];
2029             jnrlistH         = jjnr[jidx+7];
2030             /* Sign of each element will be negative for non-real atoms.
2031              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2032              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2033              */
2034             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2035                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2036
2037             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2038             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2039             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2040             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2041             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2042             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2043             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2044             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2045             j_coord_offsetA  = DIM*jnrA;
2046             j_coord_offsetB  = DIM*jnrB;
2047             j_coord_offsetC  = DIM*jnrC;
2048             j_coord_offsetD  = DIM*jnrD;
2049             j_coord_offsetE  = DIM*jnrE;
2050             j_coord_offsetF  = DIM*jnrF;
2051             j_coord_offsetG  = DIM*jnrG;
2052             j_coord_offsetH  = DIM*jnrH;
2053
2054             /* load j atom coordinates */
2055             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
2056                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
2057                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
2058                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
2059                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
2060
2061             /* Calculate displacement vector */
2062             dx11             = _mm256_sub_ps(ix1,jx1);
2063             dy11             = _mm256_sub_ps(iy1,jy1);
2064             dz11             = _mm256_sub_ps(iz1,jz1);
2065             dx12             = _mm256_sub_ps(ix1,jx2);
2066             dy12             = _mm256_sub_ps(iy1,jy2);
2067             dz12             = _mm256_sub_ps(iz1,jz2);
2068             dx13             = _mm256_sub_ps(ix1,jx3);
2069             dy13             = _mm256_sub_ps(iy1,jy3);
2070             dz13             = _mm256_sub_ps(iz1,jz3);
2071             dx21             = _mm256_sub_ps(ix2,jx1);
2072             dy21             = _mm256_sub_ps(iy2,jy1);
2073             dz21             = _mm256_sub_ps(iz2,jz1);
2074             dx22             = _mm256_sub_ps(ix2,jx2);
2075             dy22             = _mm256_sub_ps(iy2,jy2);
2076             dz22             = _mm256_sub_ps(iz2,jz2);
2077             dx23             = _mm256_sub_ps(ix2,jx3);
2078             dy23             = _mm256_sub_ps(iy2,jy3);
2079             dz23             = _mm256_sub_ps(iz2,jz3);
2080             dx31             = _mm256_sub_ps(ix3,jx1);
2081             dy31             = _mm256_sub_ps(iy3,jy1);
2082             dz31             = _mm256_sub_ps(iz3,jz1);
2083             dx32             = _mm256_sub_ps(ix3,jx2);
2084             dy32             = _mm256_sub_ps(iy3,jy2);
2085             dz32             = _mm256_sub_ps(iz3,jz2);
2086             dx33             = _mm256_sub_ps(ix3,jx3);
2087             dy33             = _mm256_sub_ps(iy3,jy3);
2088             dz33             = _mm256_sub_ps(iz3,jz3);
2089
2090             /* Calculate squared distance and things based on it */
2091             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2092             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2093             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
2094             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2095             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2096             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
2097             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
2098             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
2099             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
2100
2101             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2102             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2103             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
2104             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2105             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2106             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
2107             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
2108             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
2109             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
2110
2111             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2112             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2113             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
2114             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2115             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2116             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
2117             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
2118             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
2119             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
2120
2121             fjx1             = _mm256_setzero_ps();
2122             fjy1             = _mm256_setzero_ps();
2123             fjz1             = _mm256_setzero_ps();
2124             fjx2             = _mm256_setzero_ps();
2125             fjy2             = _mm256_setzero_ps();
2126             fjz2             = _mm256_setzero_ps();
2127             fjx3             = _mm256_setzero_ps();
2128             fjy3             = _mm256_setzero_ps();
2129             fjz3             = _mm256_setzero_ps();
2130
2131             /**************************
2132              * CALCULATE INTERACTIONS *
2133              **************************/
2134
2135             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2136             {
2137
2138             r11              = _mm256_mul_ps(rsq11,rinv11);
2139             r11              = _mm256_andnot_ps(dummy_mask,r11);
2140
2141             /* EWALD ELECTROSTATICS */
2142
2143             /* Analytical PME correction */
2144             zeta2            = _mm256_mul_ps(beta2,rsq11);
2145             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
2146             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2147             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2148             felec            = _mm256_mul_ps(qq11,felec);
2149
2150             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2151
2152             fscal            = felec;
2153
2154             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2155
2156             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2157
2158             /* Calculate temporary vectorial force */
2159             tx               = _mm256_mul_ps(fscal,dx11);
2160             ty               = _mm256_mul_ps(fscal,dy11);
2161             tz               = _mm256_mul_ps(fscal,dz11);
2162
2163             /* Update vectorial force */
2164             fix1             = _mm256_add_ps(fix1,tx);
2165             fiy1             = _mm256_add_ps(fiy1,ty);
2166             fiz1             = _mm256_add_ps(fiz1,tz);
2167
2168             fjx1             = _mm256_add_ps(fjx1,tx);
2169             fjy1             = _mm256_add_ps(fjy1,ty);
2170             fjz1             = _mm256_add_ps(fjz1,tz);
2171
2172             }
2173
2174             /**************************
2175              * CALCULATE INTERACTIONS *
2176              **************************/
2177
2178             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2179             {
2180
2181             r12              = _mm256_mul_ps(rsq12,rinv12);
2182             r12              = _mm256_andnot_ps(dummy_mask,r12);
2183
2184             /* EWALD ELECTROSTATICS */
2185
2186             /* Analytical PME correction */
2187             zeta2            = _mm256_mul_ps(beta2,rsq12);
2188             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2189             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2190             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2191             felec            = _mm256_mul_ps(qq12,felec);
2192
2193             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2194
2195             fscal            = felec;
2196
2197             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2198
2199             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2200
2201             /* Calculate temporary vectorial force */
2202             tx               = _mm256_mul_ps(fscal,dx12);
2203             ty               = _mm256_mul_ps(fscal,dy12);
2204             tz               = _mm256_mul_ps(fscal,dz12);
2205
2206             /* Update vectorial force */
2207             fix1             = _mm256_add_ps(fix1,tx);
2208             fiy1             = _mm256_add_ps(fiy1,ty);
2209             fiz1             = _mm256_add_ps(fiz1,tz);
2210
2211             fjx2             = _mm256_add_ps(fjx2,tx);
2212             fjy2             = _mm256_add_ps(fjy2,ty);
2213             fjz2             = _mm256_add_ps(fjz2,tz);
2214
2215             }
2216
2217             /**************************
2218              * CALCULATE INTERACTIONS *
2219              **************************/
2220
2221             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2222             {
2223
2224             r13              = _mm256_mul_ps(rsq13,rinv13);
2225             r13              = _mm256_andnot_ps(dummy_mask,r13);
2226
2227             /* EWALD ELECTROSTATICS */
2228
2229             /* Analytical PME correction */
2230             zeta2            = _mm256_mul_ps(beta2,rsq13);
2231             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
2232             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2233             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2234             felec            = _mm256_mul_ps(qq13,felec);
2235
2236             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
2237
2238             fscal            = felec;
2239
2240             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2241
2242             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2243
2244             /* Calculate temporary vectorial force */
2245             tx               = _mm256_mul_ps(fscal,dx13);
2246             ty               = _mm256_mul_ps(fscal,dy13);
2247             tz               = _mm256_mul_ps(fscal,dz13);
2248
2249             /* Update vectorial force */
2250             fix1             = _mm256_add_ps(fix1,tx);
2251             fiy1             = _mm256_add_ps(fiy1,ty);
2252             fiz1             = _mm256_add_ps(fiz1,tz);
2253
2254             fjx3             = _mm256_add_ps(fjx3,tx);
2255             fjy3             = _mm256_add_ps(fjy3,ty);
2256             fjz3             = _mm256_add_ps(fjz3,tz);
2257
2258             }
2259
2260             /**************************
2261              * CALCULATE INTERACTIONS *
2262              **************************/
2263
2264             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2265             {
2266
2267             r21              = _mm256_mul_ps(rsq21,rinv21);
2268             r21              = _mm256_andnot_ps(dummy_mask,r21);
2269
2270             /* EWALD ELECTROSTATICS */
2271
2272             /* Analytical PME correction */
2273             zeta2            = _mm256_mul_ps(beta2,rsq21);
2274             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2275             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2276             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2277             felec            = _mm256_mul_ps(qq21,felec);
2278
2279             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2280
2281             fscal            = felec;
2282
2283             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2284
2285             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2286
2287             /* Calculate temporary vectorial force */
2288             tx               = _mm256_mul_ps(fscal,dx21);
2289             ty               = _mm256_mul_ps(fscal,dy21);
2290             tz               = _mm256_mul_ps(fscal,dz21);
2291
2292             /* Update vectorial force */
2293             fix2             = _mm256_add_ps(fix2,tx);
2294             fiy2             = _mm256_add_ps(fiy2,ty);
2295             fiz2             = _mm256_add_ps(fiz2,tz);
2296
2297             fjx1             = _mm256_add_ps(fjx1,tx);
2298             fjy1             = _mm256_add_ps(fjy1,ty);
2299             fjz1             = _mm256_add_ps(fjz1,tz);
2300
2301             }
2302
2303             /**************************
2304              * CALCULATE INTERACTIONS *
2305              **************************/
2306
2307             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2308             {
2309
2310             r22              = _mm256_mul_ps(rsq22,rinv22);
2311             r22              = _mm256_andnot_ps(dummy_mask,r22);
2312
2313             /* EWALD ELECTROSTATICS */
2314
2315             /* Analytical PME correction */
2316             zeta2            = _mm256_mul_ps(beta2,rsq22);
2317             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2318             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2319             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2320             felec            = _mm256_mul_ps(qq22,felec);
2321
2322             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2323
2324             fscal            = felec;
2325
2326             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2327
2328             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2329
2330             /* Calculate temporary vectorial force */
2331             tx               = _mm256_mul_ps(fscal,dx22);
2332             ty               = _mm256_mul_ps(fscal,dy22);
2333             tz               = _mm256_mul_ps(fscal,dz22);
2334
2335             /* Update vectorial force */
2336             fix2             = _mm256_add_ps(fix2,tx);
2337             fiy2             = _mm256_add_ps(fiy2,ty);
2338             fiz2             = _mm256_add_ps(fiz2,tz);
2339
2340             fjx2             = _mm256_add_ps(fjx2,tx);
2341             fjy2             = _mm256_add_ps(fjy2,ty);
2342             fjz2             = _mm256_add_ps(fjz2,tz);
2343
2344             }
2345
2346             /**************************
2347              * CALCULATE INTERACTIONS *
2348              **************************/
2349
2350             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2351             {
2352
2353             r23              = _mm256_mul_ps(rsq23,rinv23);
2354             r23              = _mm256_andnot_ps(dummy_mask,r23);
2355
2356             /* EWALD ELECTROSTATICS */
2357
2358             /* Analytical PME correction */
2359             zeta2            = _mm256_mul_ps(beta2,rsq23);
2360             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
2361             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2362             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2363             felec            = _mm256_mul_ps(qq23,felec);
2364
2365             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
2366
2367             fscal            = felec;
2368
2369             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2370
2371             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2372
2373             /* Calculate temporary vectorial force */
2374             tx               = _mm256_mul_ps(fscal,dx23);
2375             ty               = _mm256_mul_ps(fscal,dy23);
2376             tz               = _mm256_mul_ps(fscal,dz23);
2377
2378             /* Update vectorial force */
2379             fix2             = _mm256_add_ps(fix2,tx);
2380             fiy2             = _mm256_add_ps(fiy2,ty);
2381             fiz2             = _mm256_add_ps(fiz2,tz);
2382
2383             fjx3             = _mm256_add_ps(fjx3,tx);
2384             fjy3             = _mm256_add_ps(fjy3,ty);
2385             fjz3             = _mm256_add_ps(fjz3,tz);
2386
2387             }
2388
2389             /**************************
2390              * CALCULATE INTERACTIONS *
2391              **************************/
2392
2393             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2394             {
2395
2396             r31              = _mm256_mul_ps(rsq31,rinv31);
2397             r31              = _mm256_andnot_ps(dummy_mask,r31);
2398
2399             /* EWALD ELECTROSTATICS */
2400
2401             /* Analytical PME correction */
2402             zeta2            = _mm256_mul_ps(beta2,rsq31);
2403             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
2404             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2405             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2406             felec            = _mm256_mul_ps(qq31,felec);
2407
2408             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
2409
2410             fscal            = felec;
2411
2412             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2413
2414             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2415
2416             /* Calculate temporary vectorial force */
2417             tx               = _mm256_mul_ps(fscal,dx31);
2418             ty               = _mm256_mul_ps(fscal,dy31);
2419             tz               = _mm256_mul_ps(fscal,dz31);
2420
2421             /* Update vectorial force */
2422             fix3             = _mm256_add_ps(fix3,tx);
2423             fiy3             = _mm256_add_ps(fiy3,ty);
2424             fiz3             = _mm256_add_ps(fiz3,tz);
2425
2426             fjx1             = _mm256_add_ps(fjx1,tx);
2427             fjy1             = _mm256_add_ps(fjy1,ty);
2428             fjz1             = _mm256_add_ps(fjz1,tz);
2429
2430             }
2431
2432             /**************************
2433              * CALCULATE INTERACTIONS *
2434              **************************/
2435
2436             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2437             {
2438
2439             r32              = _mm256_mul_ps(rsq32,rinv32);
2440             r32              = _mm256_andnot_ps(dummy_mask,r32);
2441
2442             /* EWALD ELECTROSTATICS */
2443
2444             /* Analytical PME correction */
2445             zeta2            = _mm256_mul_ps(beta2,rsq32);
2446             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
2447             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2448             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2449             felec            = _mm256_mul_ps(qq32,felec);
2450
2451             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
2452
2453             fscal            = felec;
2454
2455             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2456
2457             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2458
2459             /* Calculate temporary vectorial force */
2460             tx               = _mm256_mul_ps(fscal,dx32);
2461             ty               = _mm256_mul_ps(fscal,dy32);
2462             tz               = _mm256_mul_ps(fscal,dz32);
2463
2464             /* Update vectorial force */
2465             fix3             = _mm256_add_ps(fix3,tx);
2466             fiy3             = _mm256_add_ps(fiy3,ty);
2467             fiz3             = _mm256_add_ps(fiz3,tz);
2468
2469             fjx2             = _mm256_add_ps(fjx2,tx);
2470             fjy2             = _mm256_add_ps(fjy2,ty);
2471             fjz2             = _mm256_add_ps(fjz2,tz);
2472
2473             }
2474
2475             /**************************
2476              * CALCULATE INTERACTIONS *
2477              **************************/
2478
2479             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2480             {
2481
2482             r33              = _mm256_mul_ps(rsq33,rinv33);
2483             r33              = _mm256_andnot_ps(dummy_mask,r33);
2484
2485             /* EWALD ELECTROSTATICS */
2486
2487             /* Analytical PME correction */
2488             zeta2            = _mm256_mul_ps(beta2,rsq33);
2489             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
2490             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2491             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2492             felec            = _mm256_mul_ps(qq33,felec);
2493
2494             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
2495
2496             fscal            = felec;
2497
2498             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2499
2500             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2501
2502             /* Calculate temporary vectorial force */
2503             tx               = _mm256_mul_ps(fscal,dx33);
2504             ty               = _mm256_mul_ps(fscal,dy33);
2505             tz               = _mm256_mul_ps(fscal,dz33);
2506
2507             /* Update vectorial force */
2508             fix3             = _mm256_add_ps(fix3,tx);
2509             fiy3             = _mm256_add_ps(fiy3,ty);
2510             fiz3             = _mm256_add_ps(fiz3,tz);
2511
2512             fjx3             = _mm256_add_ps(fjx3,tx);
2513             fjy3             = _mm256_add_ps(fjy3,ty);
2514             fjz3             = _mm256_add_ps(fjz3,tz);
2515
2516             }
2517
2518             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2519             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2520             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2521             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2522             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2523             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2524             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2525             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2526
2527             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2528                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
2529                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2530
2531             /* Inner loop uses 540 flops */
2532         }
2533
2534         /* End of innermost loop */
2535
2536         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2537                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
2538
2539         /* Increment number of inner iterations */
2540         inneriter                  += j_index_end - j_index_start;
2541
2542         /* Outer loop uses 18 flops */
2543     }
2544
2545     /* Increment number of outer iterations */
2546     outeriter        += nri;
2547
2548     /* Update outer/inner flops */
2549
2550     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*540);
2551 }