src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwNone_GeomW4W4_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4W4_VF_avx_256_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Water4-Water4
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwNone_GeomW4W4_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr1;
  86     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     real *           vdwioffsetptr2;
  88     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     real *           vdwioffsetptr3;
  90     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  92     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  93     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  94     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  95     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
  96     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  97     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  98     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  99     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 100     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 101     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 102     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 103     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 104     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 105     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 106     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 107     real             *charge;
 108     __m256i          ewitab;
 109     __m128i          ewitab_lo,ewitab_hi;
 110     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 112     real             *ewtab;
 113     __m256           dummy_mask,cutoff_mask;
 114     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 115     __m256           one     = _mm256_set1_ps(1.0);
 116     __m256           two     = _mm256_set1_ps(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_ps(fr->ic->epsfac);
 129     charge           = mdatoms->chargeA;
 130
 131     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 132     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 133     beta2            = _mm256_mul_ps(beta,beta);
 134     beta3            = _mm256_mul_ps(beta,beta2);
 135
 136     ewtab            = fr->ic->tabq_coul_FDV0;
 137     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 138     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 139
 140     /* Setup water-specific parameters */
 141     inr              = nlist->iinr[0];
 142     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 143     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 144     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 145
 146     jq1              = _mm256_set1_ps(charge[inr+1]);
 147     jq2              = _mm256_set1_ps(charge[inr+2]);
 148     jq3              = _mm256_set1_ps(charge[inr+3]);
 149     qq11             = _mm256_mul_ps(iq1,jq1);
 150     qq12             = _mm256_mul_ps(iq1,jq2);
 151     qq13             = _mm256_mul_ps(iq1,jq3);
 152     qq21             = _mm256_mul_ps(iq2,jq1);
 153     qq22             = _mm256_mul_ps(iq2,jq2);
 154     qq23             = _mm256_mul_ps(iq2,jq3);
 155     qq31             = _mm256_mul_ps(iq3,jq1);
 156     qq32             = _mm256_mul_ps(iq3,jq2);
 157     qq33             = _mm256_mul_ps(iq3,jq3);
 158
 159     /* Avoid stupid compiler warnings */
 160     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 161     j_coord_offsetA = 0;
 162     j_coord_offsetB = 0;
 163     j_coord_offsetC = 0;
 164     j_coord_offsetD = 0;
 165     j_coord_offsetE = 0;
 166     j_coord_offsetF = 0;
 167     j_coord_offsetG = 0;
 168     j_coord_offsetH = 0;
 169
 170     outeriter        = 0;
 171     inneriter        = 0;
 172
 173     for(iidx=0;iidx<4*DIM;iidx++)
 174     {
 175         scratch[iidx] = 0.0;
 176     }
 177
 178     /* Start outer loop over neighborlists */
 179     for(iidx=0; iidx<nri; iidx++)
 180     {
 181         /* Load shift vector for this list */
 182         i_shift_offset   = DIM*shiftidx[iidx];
 183
 184         /* Load limits for loop over neighbors */
 185         j_index_start    = jindex[iidx];
 186         j_index_end      = jindex[iidx+1];
 187
 188         /* Get outer coordinate index */
 189         inr              = iinr[iidx];
 190         i_coord_offset   = DIM*inr;
 191
 192         /* Load i particle coords and add shift vector */
 193         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 194                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 195
 196         fix1             = _mm256_setzero_ps();
 197         fiy1             = _mm256_setzero_ps();
 198         fiz1             = _mm256_setzero_ps();
 199         fix2             = _mm256_setzero_ps();
 200         fiy2             = _mm256_setzero_ps();
 201         fiz2             = _mm256_setzero_ps();
 202         fix3             = _mm256_setzero_ps();
 203         fiy3             = _mm256_setzero_ps();
 204         fiz3             = _mm256_setzero_ps();
 205
 206         /* Reset potential sums */
 207         velecsum         = _mm256_setzero_ps();
 208
 209         /* Start inner kernel loop */
 210         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 211         {
 212
 213             /* Get j neighbor index, and coordinate index */
 214             jnrA             = jjnr[jidx];
 215             jnrB             = jjnr[jidx+1];
 216             jnrC             = jjnr[jidx+2];
 217             jnrD             = jjnr[jidx+3];
 218             jnrE             = jjnr[jidx+4];
 219             jnrF             = jjnr[jidx+5];
 220             jnrG             = jjnr[jidx+6];
 221             jnrH             = jjnr[jidx+7];
 222             j_coord_offsetA  = DIM*jnrA;
 223             j_coord_offsetB  = DIM*jnrB;
 224             j_coord_offsetC  = DIM*jnrC;
 225             j_coord_offsetD  = DIM*jnrD;
 226             j_coord_offsetE  = DIM*jnrE;
 227             j_coord_offsetF  = DIM*jnrF;
 228             j_coord_offsetG  = DIM*jnrG;
 229             j_coord_offsetH  = DIM*jnrH;
 230
 231             /* load j atom coordinates */
 232             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 233                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 234                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
 235                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
 236                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 237
 238             /* Calculate displacement vector */
 239             dx11             = _mm256_sub_ps(ix1,jx1);
 240             dy11             = _mm256_sub_ps(iy1,jy1);
 241             dz11             = _mm256_sub_ps(iz1,jz1);
 242             dx12             = _mm256_sub_ps(ix1,jx2);
 243             dy12             = _mm256_sub_ps(iy1,jy2);
 244             dz12             = _mm256_sub_ps(iz1,jz2);
 245             dx13             = _mm256_sub_ps(ix1,jx3);
 246             dy13             = _mm256_sub_ps(iy1,jy3);
 247             dz13             = _mm256_sub_ps(iz1,jz3);
 248             dx21             = _mm256_sub_ps(ix2,jx1);
 249             dy21             = _mm256_sub_ps(iy2,jy1);
 250             dz21             = _mm256_sub_ps(iz2,jz1);
 251             dx22             = _mm256_sub_ps(ix2,jx2);
 252             dy22             = _mm256_sub_ps(iy2,jy2);
 253             dz22             = _mm256_sub_ps(iz2,jz2);
 254             dx23             = _mm256_sub_ps(ix2,jx3);
 255             dy23             = _mm256_sub_ps(iy2,jy3);
 256             dz23             = _mm256_sub_ps(iz2,jz3);
 257             dx31             = _mm256_sub_ps(ix3,jx1);
 258             dy31             = _mm256_sub_ps(iy3,jy1);
 259             dz31             = _mm256_sub_ps(iz3,jz1);
 260             dx32             = _mm256_sub_ps(ix3,jx2);
 261             dy32             = _mm256_sub_ps(iy3,jy2);
 262             dz32             = _mm256_sub_ps(iz3,jz2);
 263             dx33             = _mm256_sub_ps(ix3,jx3);
 264             dy33             = _mm256_sub_ps(iy3,jy3);
 265             dz33             = _mm256_sub_ps(iz3,jz3);
 266
 267             /* Calculate squared distance and things based on it */
 268             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 269             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 270             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 271             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 272             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 273             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 274             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 275             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 276             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 277
 278             rinv11           = avx256_invsqrt_f(rsq11);
 279             rinv12           = avx256_invsqrt_f(rsq12);
 280             rinv13           = avx256_invsqrt_f(rsq13);
 281             rinv21           = avx256_invsqrt_f(rsq21);
 282             rinv22           = avx256_invsqrt_f(rsq22);
 283             rinv23           = avx256_invsqrt_f(rsq23);
 284             rinv31           = avx256_invsqrt_f(rsq31);
 285             rinv32           = avx256_invsqrt_f(rsq32);
 286             rinv33           = avx256_invsqrt_f(rsq33);
 287
 288             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 289             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 290             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 291             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 292             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 293             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 294             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 295             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 296             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 297
 298             fjx1             = _mm256_setzero_ps();
 299             fjy1             = _mm256_setzero_ps();
 300             fjz1             = _mm256_setzero_ps();
 301             fjx2             = _mm256_setzero_ps();
 302             fjy2             = _mm256_setzero_ps();
 303             fjz2             = _mm256_setzero_ps();
 304             fjx3             = _mm256_setzero_ps();
 305             fjy3             = _mm256_setzero_ps();
 306             fjz3             = _mm256_setzero_ps();
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             r11              = _mm256_mul_ps(rsq11,rinv11);
 313
 314             /* EWALD ELECTROSTATICS */
 315
 316             /* Analytical PME correction */
 317             zeta2            = _mm256_mul_ps(beta2,rsq11);
 318             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 319             pmecorrF         = avx256_pmecorrF_f(zeta2);
 320             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 321             felec            = _mm256_mul_ps(qq11,felec);
 322             pmecorrV         = avx256_pmecorrV_f(zeta2);
 323             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 324             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 325             velec            = _mm256_mul_ps(qq11,velec);
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velecsum         = _mm256_add_ps(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm256_mul_ps(fscal,dx11);
 334             ty               = _mm256_mul_ps(fscal,dy11);
 335             tz               = _mm256_mul_ps(fscal,dz11);
 336
 337             /* Update vectorial force */
 338             fix1             = _mm256_add_ps(fix1,tx);
 339             fiy1             = _mm256_add_ps(fiy1,ty);
 340             fiz1             = _mm256_add_ps(fiz1,tz);
 341
 342             fjx1             = _mm256_add_ps(fjx1,tx);
 343             fjy1             = _mm256_add_ps(fjy1,ty);
 344             fjz1             = _mm256_add_ps(fjz1,tz);
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             r12              = _mm256_mul_ps(rsq12,rinv12);
 351
 352             /* EWALD ELECTROSTATICS */
 353
 354             /* Analytical PME correction */
 355             zeta2            = _mm256_mul_ps(beta2,rsq12);
 356             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 357             pmecorrF         = avx256_pmecorrF_f(zeta2);
 358             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 359             felec            = _mm256_mul_ps(qq12,felec);
 360             pmecorrV         = avx256_pmecorrV_f(zeta2);
 361             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 362             velec            = _mm256_sub_ps(rinv12,pmecorrV);
 363             velec            = _mm256_mul_ps(qq12,velec);
 364
 365             /* Update potential sum for this i atom from the interaction with this j atom. */
 366             velecsum         = _mm256_add_ps(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             /* Calculate temporary vectorial force */
 371             tx               = _mm256_mul_ps(fscal,dx12);
 372             ty               = _mm256_mul_ps(fscal,dy12);
 373             tz               = _mm256_mul_ps(fscal,dz12);
 374
 375             /* Update vectorial force */
 376             fix1             = _mm256_add_ps(fix1,tx);
 377             fiy1             = _mm256_add_ps(fiy1,ty);
 378             fiz1             = _mm256_add_ps(fiz1,tz);
 379
 380             fjx2             = _mm256_add_ps(fjx2,tx);
 381             fjy2             = _mm256_add_ps(fjy2,ty);
 382             fjz2             = _mm256_add_ps(fjz2,tz);
 383
 384             /**************************
 385              * CALCULATE INTERACTIONS *
 386              **************************/
 387
 388             r13              = _mm256_mul_ps(rsq13,rinv13);
 389
 390             /* EWALD ELECTROSTATICS */
 391
 392             /* Analytical PME correction */
 393             zeta2            = _mm256_mul_ps(beta2,rsq13);
 394             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
 395             pmecorrF         = avx256_pmecorrF_f(zeta2);
 396             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 397             felec            = _mm256_mul_ps(qq13,felec);
 398             pmecorrV         = avx256_pmecorrV_f(zeta2);
 399             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 400             velec            = _mm256_sub_ps(rinv13,pmecorrV);
 401             velec            = _mm256_mul_ps(qq13,velec);
 402
 403             /* Update potential sum for this i atom from the interaction with this j atom. */
 404             velecsum         = _mm256_add_ps(velecsum,velec);
 405
 406             fscal            = felec;
 407
 408             /* Calculate temporary vectorial force */
 409             tx               = _mm256_mul_ps(fscal,dx13);
 410             ty               = _mm256_mul_ps(fscal,dy13);
 411             tz               = _mm256_mul_ps(fscal,dz13);
 412
 413             /* Update vectorial force */
 414             fix1             = _mm256_add_ps(fix1,tx);
 415             fiy1             = _mm256_add_ps(fiy1,ty);
 416             fiz1             = _mm256_add_ps(fiz1,tz);
 417
 418             fjx3             = _mm256_add_ps(fjx3,tx);
 419             fjy3             = _mm256_add_ps(fjy3,ty);
 420             fjz3             = _mm256_add_ps(fjz3,tz);
 421
 422             /**************************
 423              * CALCULATE INTERACTIONS *
 424              **************************/
 425
 426             r21              = _mm256_mul_ps(rsq21,rinv21);
 427
 428             /* EWALD ELECTROSTATICS */
 429
 430             /* Analytical PME correction */
 431             zeta2            = _mm256_mul_ps(beta2,rsq21);
 432             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 433             pmecorrF         = avx256_pmecorrF_f(zeta2);
 434             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 435             felec            = _mm256_mul_ps(qq21,felec);
 436             pmecorrV         = avx256_pmecorrV_f(zeta2);
 437             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 438             velec            = _mm256_sub_ps(rinv21,pmecorrV);
 439             velec            = _mm256_mul_ps(qq21,velec);
 440
 441             /* Update potential sum for this i atom from the interaction with this j atom. */
 442             velecsum         = _mm256_add_ps(velecsum,velec);
 443
 444             fscal            = felec;
 445
 446             /* Calculate temporary vectorial force */
 447             tx               = _mm256_mul_ps(fscal,dx21);
 448             ty               = _mm256_mul_ps(fscal,dy21);
 449             tz               = _mm256_mul_ps(fscal,dz21);
 450
 451             /* Update vectorial force */
 452             fix2             = _mm256_add_ps(fix2,tx);
 453             fiy2             = _mm256_add_ps(fiy2,ty);
 454             fiz2             = _mm256_add_ps(fiz2,tz);
 455
 456             fjx1             = _mm256_add_ps(fjx1,tx);
 457             fjy1             = _mm256_add_ps(fjy1,ty);
 458             fjz1             = _mm256_add_ps(fjz1,tz);
 459
 460             /**************************
 461              * CALCULATE INTERACTIONS *
 462              **************************/
 463
 464             r22              = _mm256_mul_ps(rsq22,rinv22);
 465
 466             /* EWALD ELECTROSTATICS */
 467
 468             /* Analytical PME correction */
 469             zeta2            = _mm256_mul_ps(beta2,rsq22);
 470             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 471             pmecorrF         = avx256_pmecorrF_f(zeta2);
 472             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 473             felec            = _mm256_mul_ps(qq22,felec);
 474             pmecorrV         = avx256_pmecorrV_f(zeta2);
 475             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 476             velec            = _mm256_sub_ps(rinv22,pmecorrV);
 477             velec            = _mm256_mul_ps(qq22,velec);
 478
 479             /* Update potential sum for this i atom from the interaction with this j atom. */
 480             velecsum         = _mm256_add_ps(velecsum,velec);
 481
 482             fscal            = felec;
 483
 484             /* Calculate temporary vectorial force */
 485             tx               = _mm256_mul_ps(fscal,dx22);
 486             ty               = _mm256_mul_ps(fscal,dy22);
 487             tz               = _mm256_mul_ps(fscal,dz22);
 488
 489             /* Update vectorial force */
 490             fix2             = _mm256_add_ps(fix2,tx);
 491             fiy2             = _mm256_add_ps(fiy2,ty);
 492             fiz2             = _mm256_add_ps(fiz2,tz);
 493
 494             fjx2             = _mm256_add_ps(fjx2,tx);
 495             fjy2             = _mm256_add_ps(fjy2,ty);
 496             fjz2             = _mm256_add_ps(fjz2,tz);
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             r23              = _mm256_mul_ps(rsq23,rinv23);
 503
 504             /* EWALD ELECTROSTATICS */
 505
 506             /* Analytical PME correction */
 507             zeta2            = _mm256_mul_ps(beta2,rsq23);
 508             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
 509             pmecorrF         = avx256_pmecorrF_f(zeta2);
 510             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 511             felec            = _mm256_mul_ps(qq23,felec);
 512             pmecorrV         = avx256_pmecorrV_f(zeta2);
 513             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 514             velec            = _mm256_sub_ps(rinv23,pmecorrV);
 515             velec            = _mm256_mul_ps(qq23,velec);
 516
 517             /* Update potential sum for this i atom from the interaction with this j atom. */
 518             velecsum         = _mm256_add_ps(velecsum,velec);
 519
 520             fscal            = felec;
 521
 522             /* Calculate temporary vectorial force */
 523             tx               = _mm256_mul_ps(fscal,dx23);
 524             ty               = _mm256_mul_ps(fscal,dy23);
 525             tz               = _mm256_mul_ps(fscal,dz23);
 526
 527             /* Update vectorial force */
 528             fix2             = _mm256_add_ps(fix2,tx);
 529             fiy2             = _mm256_add_ps(fiy2,ty);
 530             fiz2             = _mm256_add_ps(fiz2,tz);
 531
 532             fjx3             = _mm256_add_ps(fjx3,tx);
 533             fjy3             = _mm256_add_ps(fjy3,ty);
 534             fjz3             = _mm256_add_ps(fjz3,tz);
 535
 536             /**************************
 537              * CALCULATE INTERACTIONS *
 538              **************************/
 539
 540             r31              = _mm256_mul_ps(rsq31,rinv31);
 541
 542             /* EWALD ELECTROSTATICS */
 543
 544             /* Analytical PME correction */
 545             zeta2            = _mm256_mul_ps(beta2,rsq31);
 546             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
 547             pmecorrF         = avx256_pmecorrF_f(zeta2);
 548             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 549             felec            = _mm256_mul_ps(qq31,felec);
 550             pmecorrV         = avx256_pmecorrV_f(zeta2);
 551             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 552             velec            = _mm256_sub_ps(rinv31,pmecorrV);
 553             velec            = _mm256_mul_ps(qq31,velec);
 554
 555             /* Update potential sum for this i atom from the interaction with this j atom. */
 556             velecsum         = _mm256_add_ps(velecsum,velec);
 557
 558             fscal            = felec;
 559
 560             /* Calculate temporary vectorial force */
 561             tx               = _mm256_mul_ps(fscal,dx31);
 562             ty               = _mm256_mul_ps(fscal,dy31);
 563             tz               = _mm256_mul_ps(fscal,dz31);
 564
 565             /* Update vectorial force */
 566             fix3             = _mm256_add_ps(fix3,tx);
 567             fiy3             = _mm256_add_ps(fiy3,ty);
 568             fiz3             = _mm256_add_ps(fiz3,tz);
 569
 570             fjx1             = _mm256_add_ps(fjx1,tx);
 571             fjy1             = _mm256_add_ps(fjy1,ty);
 572             fjz1             = _mm256_add_ps(fjz1,tz);
 573
 574             /**************************
 575              * CALCULATE INTERACTIONS *
 576              **************************/
 577
 578             r32              = _mm256_mul_ps(rsq32,rinv32);
 579
 580             /* EWALD ELECTROSTATICS */
 581
 582             /* Analytical PME correction */
 583             zeta2            = _mm256_mul_ps(beta2,rsq32);
 584             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
 585             pmecorrF         = avx256_pmecorrF_f(zeta2);
 586             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 587             felec            = _mm256_mul_ps(qq32,felec);
 588             pmecorrV         = avx256_pmecorrV_f(zeta2);
 589             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 590             velec            = _mm256_sub_ps(rinv32,pmecorrV);
 591             velec            = _mm256_mul_ps(qq32,velec);
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velecsum         = _mm256_add_ps(velecsum,velec);
 595
 596             fscal            = felec;
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm256_mul_ps(fscal,dx32);
 600             ty               = _mm256_mul_ps(fscal,dy32);
 601             tz               = _mm256_mul_ps(fscal,dz32);
 602
 603             /* Update vectorial force */
 604             fix3             = _mm256_add_ps(fix3,tx);
 605             fiy3             = _mm256_add_ps(fiy3,ty);
 606             fiz3             = _mm256_add_ps(fiz3,tz);
 607
 608             fjx2             = _mm256_add_ps(fjx2,tx);
 609             fjy2             = _mm256_add_ps(fjy2,ty);
 610             fjz2             = _mm256_add_ps(fjz2,tz);
 611
 612             /**************************
 613              * CALCULATE INTERACTIONS *
 614              **************************/
 615
 616             r33              = _mm256_mul_ps(rsq33,rinv33);
 617
 618             /* EWALD ELECTROSTATICS */
 619
 620             /* Analytical PME correction */
 621             zeta2            = _mm256_mul_ps(beta2,rsq33);
 622             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
 623             pmecorrF         = avx256_pmecorrF_f(zeta2);
 624             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 625             felec            = _mm256_mul_ps(qq33,felec);
 626             pmecorrV         = avx256_pmecorrV_f(zeta2);
 627             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 628             velec            = _mm256_sub_ps(rinv33,pmecorrV);
 629             velec            = _mm256_mul_ps(qq33,velec);
 630
 631             /* Update potential sum for this i atom from the interaction with this j atom. */
 632             velecsum         = _mm256_add_ps(velecsum,velec);
 633
 634             fscal            = felec;
 635
 636             /* Calculate temporary vectorial force */
 637             tx               = _mm256_mul_ps(fscal,dx33);
 638             ty               = _mm256_mul_ps(fscal,dy33);
 639             tz               = _mm256_mul_ps(fscal,dz33);
 640
 641             /* Update vectorial force */
 642             fix3             = _mm256_add_ps(fix3,tx);
 643             fiy3             = _mm256_add_ps(fiy3,ty);
 644             fiz3             = _mm256_add_ps(fiz3,tz);
 645
 646             fjx3             = _mm256_add_ps(fjx3,tx);
 647             fjy3             = _mm256_add_ps(fjy3,ty);
 648             fjz3             = _mm256_add_ps(fjz3,tz);
 649
 650             fjptrA             = f+j_coord_offsetA;
 651             fjptrB             = f+j_coord_offsetB;
 652             fjptrC             = f+j_coord_offsetC;
 653             fjptrD             = f+j_coord_offsetD;
 654             fjptrE             = f+j_coord_offsetE;
 655             fjptrF             = f+j_coord_offsetF;
 656             fjptrG             = f+j_coord_offsetG;
 657             fjptrH             = f+j_coord_offsetH;
 658
 659             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 660                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
 661                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 662
 663             /* Inner loop uses 756 flops */
 664         }
 665
 666         if(jidx<j_index_end)
 667         {
 668
 669             /* Get j neighbor index, and coordinate index */
 670             jnrlistA         = jjnr[jidx];
 671             jnrlistB         = jjnr[jidx+1];
 672             jnrlistC         = jjnr[jidx+2];
 673             jnrlistD         = jjnr[jidx+3];
 674             jnrlistE         = jjnr[jidx+4];
 675             jnrlistF         = jjnr[jidx+5];
 676             jnrlistG         = jjnr[jidx+6];
 677             jnrlistH         = jjnr[jidx+7];
 678             /* Sign of each element will be negative for non-real atoms.
 679              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 680              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 681              */
 682             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 683                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 684
 685             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 686             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 687             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 688             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 689             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 690             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 691             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 692             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 693             j_coord_offsetA  = DIM*jnrA;
 694             j_coord_offsetB  = DIM*jnrB;
 695             j_coord_offsetC  = DIM*jnrC;
 696             j_coord_offsetD  = DIM*jnrD;
 697             j_coord_offsetE  = DIM*jnrE;
 698             j_coord_offsetF  = DIM*jnrF;
 699             j_coord_offsetG  = DIM*jnrG;
 700             j_coord_offsetH  = DIM*jnrH;
 701
 702             /* load j atom coordinates */
 703             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 704                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 705                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
 706                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
 707                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 708
 709             /* Calculate displacement vector */
 710             dx11             = _mm256_sub_ps(ix1,jx1);
 711             dy11             = _mm256_sub_ps(iy1,jy1);
 712             dz11             = _mm256_sub_ps(iz1,jz1);
 713             dx12             = _mm256_sub_ps(ix1,jx2);
 714             dy12             = _mm256_sub_ps(iy1,jy2);
 715             dz12             = _mm256_sub_ps(iz1,jz2);
 716             dx13             = _mm256_sub_ps(ix1,jx3);
 717             dy13             = _mm256_sub_ps(iy1,jy3);
 718             dz13             = _mm256_sub_ps(iz1,jz3);
 719             dx21             = _mm256_sub_ps(ix2,jx1);
 720             dy21             = _mm256_sub_ps(iy2,jy1);
 721             dz21             = _mm256_sub_ps(iz2,jz1);
 722             dx22             = _mm256_sub_ps(ix2,jx2);
 723             dy22             = _mm256_sub_ps(iy2,jy2);
 724             dz22             = _mm256_sub_ps(iz2,jz2);
 725             dx23             = _mm256_sub_ps(ix2,jx3);
 726             dy23             = _mm256_sub_ps(iy2,jy3);
 727             dz23             = _mm256_sub_ps(iz2,jz3);
 728             dx31             = _mm256_sub_ps(ix3,jx1);
 729             dy31             = _mm256_sub_ps(iy3,jy1);
 730             dz31             = _mm256_sub_ps(iz3,jz1);
 731             dx32             = _mm256_sub_ps(ix3,jx2);
 732             dy32             = _mm256_sub_ps(iy3,jy2);
 733             dz32             = _mm256_sub_ps(iz3,jz2);
 734             dx33             = _mm256_sub_ps(ix3,jx3);
 735             dy33             = _mm256_sub_ps(iy3,jy3);
 736             dz33             = _mm256_sub_ps(iz3,jz3);
 737
 738             /* Calculate squared distance and things based on it */
 739             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 740             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 741             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 742             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 743             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 744             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 745             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 746             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 747             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 748
 749             rinv11           = avx256_invsqrt_f(rsq11);
 750             rinv12           = avx256_invsqrt_f(rsq12);
 751             rinv13           = avx256_invsqrt_f(rsq13);
 752             rinv21           = avx256_invsqrt_f(rsq21);
 753             rinv22           = avx256_invsqrt_f(rsq22);
 754             rinv23           = avx256_invsqrt_f(rsq23);
 755             rinv31           = avx256_invsqrt_f(rsq31);
 756             rinv32           = avx256_invsqrt_f(rsq32);
 757             rinv33           = avx256_invsqrt_f(rsq33);
 758
 759             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 760             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 761             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 762             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 763             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 764             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 765             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 766             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 767             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 768
 769             fjx1             = _mm256_setzero_ps();
 770             fjy1             = _mm256_setzero_ps();
 771             fjz1             = _mm256_setzero_ps();
 772             fjx2             = _mm256_setzero_ps();
 773             fjy2             = _mm256_setzero_ps();
 774             fjz2             = _mm256_setzero_ps();
 775             fjx3             = _mm256_setzero_ps();
 776             fjy3             = _mm256_setzero_ps();
 777             fjz3             = _mm256_setzero_ps();
 778
 779             /**************************
 780              * CALCULATE INTERACTIONS *
 781              **************************/
 782
 783             r11              = _mm256_mul_ps(rsq11,rinv11);
 784             r11              = _mm256_andnot_ps(dummy_mask,r11);
 785
 786             /* EWALD ELECTROSTATICS */
 787
 788             /* Analytical PME correction */
 789             zeta2            = _mm256_mul_ps(beta2,rsq11);
 790             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 791             pmecorrF         = avx256_pmecorrF_f(zeta2);
 792             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 793             felec            = _mm256_mul_ps(qq11,felec);
 794             pmecorrV         = avx256_pmecorrV_f(zeta2);
 795             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 796             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 797             velec            = _mm256_mul_ps(qq11,velec);
 798
 799             /* Update potential sum for this i atom from the interaction with this j atom. */
 800             velec            = _mm256_andnot_ps(dummy_mask,velec);
 801             velecsum         = _mm256_add_ps(velecsum,velec);
 802
 803             fscal            = felec;
 804
 805             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 806
 807             /* Calculate temporary vectorial force */
 808             tx               = _mm256_mul_ps(fscal,dx11);
 809             ty               = _mm256_mul_ps(fscal,dy11);
 810             tz               = _mm256_mul_ps(fscal,dz11);
 811
 812             /* Update vectorial force */
 813             fix1             = _mm256_add_ps(fix1,tx);
 814             fiy1             = _mm256_add_ps(fiy1,ty);
 815             fiz1             = _mm256_add_ps(fiz1,tz);
 816
 817             fjx1             = _mm256_add_ps(fjx1,tx);
 818             fjy1             = _mm256_add_ps(fjy1,ty);
 819             fjz1             = _mm256_add_ps(fjz1,tz);
 820
 821             /**************************
 822              * CALCULATE INTERACTIONS *
 823              **************************/
 824
 825             r12              = _mm256_mul_ps(rsq12,rinv12);
 826             r12              = _mm256_andnot_ps(dummy_mask,r12);
 827
 828             /* EWALD ELECTROSTATICS */
 829
 830             /* Analytical PME correction */
 831             zeta2            = _mm256_mul_ps(beta2,rsq12);
 832             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 833             pmecorrF         = avx256_pmecorrF_f(zeta2);
 834             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 835             felec            = _mm256_mul_ps(qq12,felec);
 836             pmecorrV         = avx256_pmecorrV_f(zeta2);
 837             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 838             velec            = _mm256_sub_ps(rinv12,pmecorrV);
 839             velec            = _mm256_mul_ps(qq12,velec);
 840
 841             /* Update potential sum for this i atom from the interaction with this j atom. */
 842             velec            = _mm256_andnot_ps(dummy_mask,velec);
 843             velecsum         = _mm256_add_ps(velecsum,velec);
 844
 845             fscal            = felec;
 846
 847             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 848
 849             /* Calculate temporary vectorial force */
 850             tx               = _mm256_mul_ps(fscal,dx12);
 851             ty               = _mm256_mul_ps(fscal,dy12);
 852             tz               = _mm256_mul_ps(fscal,dz12);
 853
 854             /* Update vectorial force */
 855             fix1             = _mm256_add_ps(fix1,tx);
 856             fiy1             = _mm256_add_ps(fiy1,ty);
 857             fiz1             = _mm256_add_ps(fiz1,tz);
 858
 859             fjx2             = _mm256_add_ps(fjx2,tx);
 860             fjy2             = _mm256_add_ps(fjy2,ty);
 861             fjz2             = _mm256_add_ps(fjz2,tz);
 862
 863             /**************************
 864              * CALCULATE INTERACTIONS *
 865              **************************/
 866
 867             r13              = _mm256_mul_ps(rsq13,rinv13);
 868             r13              = _mm256_andnot_ps(dummy_mask,r13);
 869
 870             /* EWALD ELECTROSTATICS */
 871
 872             /* Analytical PME correction */
 873             zeta2            = _mm256_mul_ps(beta2,rsq13);
 874             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
 875             pmecorrF         = avx256_pmecorrF_f(zeta2);
 876             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 877             felec            = _mm256_mul_ps(qq13,felec);
 878             pmecorrV         = avx256_pmecorrV_f(zeta2);
 879             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 880             velec            = _mm256_sub_ps(rinv13,pmecorrV);
 881             velec            = _mm256_mul_ps(qq13,velec);
 882
 883             /* Update potential sum for this i atom from the interaction with this j atom. */
 884             velec            = _mm256_andnot_ps(dummy_mask,velec);
 885             velecsum         = _mm256_add_ps(velecsum,velec);
 886
 887             fscal            = felec;
 888
 889             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 890
 891             /* Calculate temporary vectorial force */
 892             tx               = _mm256_mul_ps(fscal,dx13);
 893             ty               = _mm256_mul_ps(fscal,dy13);
 894             tz               = _mm256_mul_ps(fscal,dz13);
 895
 896             /* Update vectorial force */
 897             fix1             = _mm256_add_ps(fix1,tx);
 898             fiy1             = _mm256_add_ps(fiy1,ty);
 899             fiz1             = _mm256_add_ps(fiz1,tz);
 900
 901             fjx3             = _mm256_add_ps(fjx3,tx);
 902             fjy3             = _mm256_add_ps(fjy3,ty);
 903             fjz3             = _mm256_add_ps(fjz3,tz);
 904
 905             /**************************
 906              * CALCULATE INTERACTIONS *
 907              **************************/
 908
 909             r21              = _mm256_mul_ps(rsq21,rinv21);
 910             r21              = _mm256_andnot_ps(dummy_mask,r21);
 911
 912             /* EWALD ELECTROSTATICS */
 913
 914             /* Analytical PME correction */
 915             zeta2            = _mm256_mul_ps(beta2,rsq21);
 916             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 917             pmecorrF         = avx256_pmecorrF_f(zeta2);
 918             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 919             felec            = _mm256_mul_ps(qq21,felec);
 920             pmecorrV         = avx256_pmecorrV_f(zeta2);
 921             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 922             velec            = _mm256_sub_ps(rinv21,pmecorrV);
 923             velec            = _mm256_mul_ps(qq21,velec);
 924
 925             /* Update potential sum for this i atom from the interaction with this j atom. */
 926             velec            = _mm256_andnot_ps(dummy_mask,velec);
 927             velecsum         = _mm256_add_ps(velecsum,velec);
 928
 929             fscal            = felec;
 930
 931             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 932
 933             /* Calculate temporary vectorial force */
 934             tx               = _mm256_mul_ps(fscal,dx21);
 935             ty               = _mm256_mul_ps(fscal,dy21);
 936             tz               = _mm256_mul_ps(fscal,dz21);
 937
 938             /* Update vectorial force */
 939             fix2             = _mm256_add_ps(fix2,tx);
 940             fiy2             = _mm256_add_ps(fiy2,ty);
 941             fiz2             = _mm256_add_ps(fiz2,tz);
 942
 943             fjx1             = _mm256_add_ps(fjx1,tx);
 944             fjy1             = _mm256_add_ps(fjy1,ty);
 945             fjz1             = _mm256_add_ps(fjz1,tz);
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             r22              = _mm256_mul_ps(rsq22,rinv22);
 952             r22              = _mm256_andnot_ps(dummy_mask,r22);
 953
 954             /* EWALD ELECTROSTATICS */
 955
 956             /* Analytical PME correction */
 957             zeta2            = _mm256_mul_ps(beta2,rsq22);
 958             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 959             pmecorrF         = avx256_pmecorrF_f(zeta2);
 960             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 961             felec            = _mm256_mul_ps(qq22,felec);
 962             pmecorrV         = avx256_pmecorrV_f(zeta2);
 963             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 964             velec            = _mm256_sub_ps(rinv22,pmecorrV);
 965             velec            = _mm256_mul_ps(qq22,velec);
 966
 967             /* Update potential sum for this i atom from the interaction with this j atom. */
 968             velec            = _mm256_andnot_ps(dummy_mask,velec);
 969             velecsum         = _mm256_add_ps(velecsum,velec);
 970
 971             fscal            = felec;
 972
 973             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 974
 975             /* Calculate temporary vectorial force */
 976             tx               = _mm256_mul_ps(fscal,dx22);
 977             ty               = _mm256_mul_ps(fscal,dy22);
 978             tz               = _mm256_mul_ps(fscal,dz22);
 979
 980             /* Update vectorial force */
 981             fix2             = _mm256_add_ps(fix2,tx);
 982             fiy2             = _mm256_add_ps(fiy2,ty);
 983             fiz2             = _mm256_add_ps(fiz2,tz);
 984
 985             fjx2             = _mm256_add_ps(fjx2,tx);
 986             fjy2             = _mm256_add_ps(fjy2,ty);
 987             fjz2             = _mm256_add_ps(fjz2,tz);
 988
 989             /**************************
 990              * CALCULATE INTERACTIONS *
 991              **************************/
 992
 993             r23              = _mm256_mul_ps(rsq23,rinv23);
 994             r23              = _mm256_andnot_ps(dummy_mask,r23);
 995
 996             /* EWALD ELECTROSTATICS */
 997
 998             /* Analytical PME correction */
 999             zeta2            = _mm256_mul_ps(beta2,rsq23);
1000             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1001             pmecorrF         = avx256_pmecorrF_f(zeta2);
1002             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1003             felec            = _mm256_mul_ps(qq23,felec);
1004             pmecorrV         = avx256_pmecorrV_f(zeta2);
1005             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1006             velec            = _mm256_sub_ps(rinv23,pmecorrV);
1007             velec            = _mm256_mul_ps(qq23,velec);
1008
1009             /* Update potential sum for this i atom from the interaction with this j atom. */
1010             velec            = _mm256_andnot_ps(dummy_mask,velec);
1011             velecsum         = _mm256_add_ps(velecsum,velec);
1012
1013             fscal            = felec;
1014
1015             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1016
1017             /* Calculate temporary vectorial force */
1018             tx               = _mm256_mul_ps(fscal,dx23);
1019             ty               = _mm256_mul_ps(fscal,dy23);
1020             tz               = _mm256_mul_ps(fscal,dz23);
1021
1022             /* Update vectorial force */
1023             fix2             = _mm256_add_ps(fix2,tx);
1024             fiy2             = _mm256_add_ps(fiy2,ty);
1025             fiz2             = _mm256_add_ps(fiz2,tz);
1026
1027             fjx3             = _mm256_add_ps(fjx3,tx);
1028             fjy3             = _mm256_add_ps(fjy3,ty);
1029             fjz3             = _mm256_add_ps(fjz3,tz);
1030
1031             /**************************
1032              * CALCULATE INTERACTIONS *
1033              **************************/
1034
1035             r31              = _mm256_mul_ps(rsq31,rinv31);
1036             r31              = _mm256_andnot_ps(dummy_mask,r31);
1037
1038             /* EWALD ELECTROSTATICS */
1039
1040             /* Analytical PME correction */
1041             zeta2            = _mm256_mul_ps(beta2,rsq31);
1042             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1043             pmecorrF         = avx256_pmecorrF_f(zeta2);
1044             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1045             felec            = _mm256_mul_ps(qq31,felec);
1046             pmecorrV         = avx256_pmecorrV_f(zeta2);
1047             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1048             velec            = _mm256_sub_ps(rinv31,pmecorrV);
1049             velec            = _mm256_mul_ps(qq31,velec);
1050
1051             /* Update potential sum for this i atom from the interaction with this j atom. */
1052             velec            = _mm256_andnot_ps(dummy_mask,velec);
1053             velecsum         = _mm256_add_ps(velecsum,velec);
1054
1055             fscal            = felec;
1056
1057             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1058
1059             /* Calculate temporary vectorial force */
1060             tx               = _mm256_mul_ps(fscal,dx31);
1061             ty               = _mm256_mul_ps(fscal,dy31);
1062             tz               = _mm256_mul_ps(fscal,dz31);
1063
1064             /* Update vectorial force */
1065             fix3             = _mm256_add_ps(fix3,tx);
1066             fiy3             = _mm256_add_ps(fiy3,ty);
1067             fiz3             = _mm256_add_ps(fiz3,tz);
1068
1069             fjx1             = _mm256_add_ps(fjx1,tx);
1070             fjy1             = _mm256_add_ps(fjy1,ty);
1071             fjz1             = _mm256_add_ps(fjz1,tz);
1072
1073             /**************************
1074              * CALCULATE INTERACTIONS *
1075              **************************/
1076
1077             r32              = _mm256_mul_ps(rsq32,rinv32);
1078             r32              = _mm256_andnot_ps(dummy_mask,r32);
1079
1080             /* EWALD ELECTROSTATICS */
1081
1082             /* Analytical PME correction */
1083             zeta2            = _mm256_mul_ps(beta2,rsq32);
1084             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1085             pmecorrF         = avx256_pmecorrF_f(zeta2);
1086             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1087             felec            = _mm256_mul_ps(qq32,felec);
1088             pmecorrV         = avx256_pmecorrV_f(zeta2);
1089             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1090             velec            = _mm256_sub_ps(rinv32,pmecorrV);
1091             velec            = _mm256_mul_ps(qq32,velec);
1092
1093             /* Update potential sum for this i atom from the interaction with this j atom. */
1094             velec            = _mm256_andnot_ps(dummy_mask,velec);
1095             velecsum         = _mm256_add_ps(velecsum,velec);
1096
1097             fscal            = felec;
1098
1099             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1100
1101             /* Calculate temporary vectorial force */
1102             tx               = _mm256_mul_ps(fscal,dx32);
1103             ty               = _mm256_mul_ps(fscal,dy32);
1104             tz               = _mm256_mul_ps(fscal,dz32);
1105
1106             /* Update vectorial force */
1107             fix3             = _mm256_add_ps(fix3,tx);
1108             fiy3             = _mm256_add_ps(fiy3,ty);
1109             fiz3             = _mm256_add_ps(fiz3,tz);
1110
1111             fjx2             = _mm256_add_ps(fjx2,tx);
1112             fjy2             = _mm256_add_ps(fjy2,ty);
1113             fjz2             = _mm256_add_ps(fjz2,tz);
1114
1115             /**************************
1116              * CALCULATE INTERACTIONS *
1117              **************************/
1118
1119             r33              = _mm256_mul_ps(rsq33,rinv33);
1120             r33              = _mm256_andnot_ps(dummy_mask,r33);
1121
1122             /* EWALD ELECTROSTATICS */
1123
1124             /* Analytical PME correction */
1125             zeta2            = _mm256_mul_ps(beta2,rsq33);
1126             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
1127             pmecorrF         = avx256_pmecorrF_f(zeta2);
1128             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1129             felec            = _mm256_mul_ps(qq33,felec);
1130             pmecorrV         = avx256_pmecorrV_f(zeta2);
1131             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1132             velec            = _mm256_sub_ps(rinv33,pmecorrV);
1133             velec            = _mm256_mul_ps(qq33,velec);
1134
1135             /* Update potential sum for this i atom from the interaction with this j atom. */
1136             velec            = _mm256_andnot_ps(dummy_mask,velec);
1137             velecsum         = _mm256_add_ps(velecsum,velec);
1138
1139             fscal            = felec;
1140
1141             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1142
1143             /* Calculate temporary vectorial force */
1144             tx               = _mm256_mul_ps(fscal,dx33);
1145             ty               = _mm256_mul_ps(fscal,dy33);
1146             tz               = _mm256_mul_ps(fscal,dz33);
1147
1148             /* Update vectorial force */
1149             fix3             = _mm256_add_ps(fix3,tx);
1150             fiy3             = _mm256_add_ps(fiy3,ty);
1151             fiz3             = _mm256_add_ps(fiz3,tz);
1152
1153             fjx3             = _mm256_add_ps(fjx3,tx);
1154             fjy3             = _mm256_add_ps(fjy3,ty);
1155             fjz3             = _mm256_add_ps(fjz3,tz);
1156
1157             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1158             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1159             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1160             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1161             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1162             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1163             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1164             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1165
1166             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1167                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
1168                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1169
1170             /* Inner loop uses 765 flops */
1171         }
1172
1173         /* End of innermost loop */
1174
1175         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1176                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1177
1178         ggid                        = gid[iidx];
1179         /* Update potential energies */
1180         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1181
1182         /* Increment number of inner iterations */
1183         inneriter                  += j_index_end - j_index_start;
1184
1185         /* Outer loop uses 19 flops */
1186     }
1187
1188     /* Increment number of outer iterations */
1189     outeriter        += nri;
1190
1191     /* Update outer/inner flops */
1192
1193     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*765);
1194 }
1195 /*
1196  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4W4_F_avx_256_single
1197  * Electrostatics interaction: Ewald
1198  * VdW interaction:            None
1199  * Geometry:                   Water4-Water4
1200  * Calculate force/pot:        Force
1201  */
1202 void
1203 nb_kernel_ElecEw_VdwNone_GeomW4W4_F_avx_256_single
1204                     (t_nblist                    * gmx_restrict       nlist,
1205                      rvec                        * gmx_restrict          xx,
1206                      rvec                        * gmx_restrict          ff,
1207                      struct t_forcerec           * gmx_restrict          fr,
1208                      t_mdatoms                   * gmx_restrict     mdatoms,
1209                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1210                      t_nrnb                      * gmx_restrict        nrnb)
1211 {
1212     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1213      * just 0 for non-waters.
1214      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1215      * jnr indices corresponding to data put in the four positions in the SIMD register.
1216      */
1217     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1218     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1219     int              jnrA,jnrB,jnrC,jnrD;
1220     int              jnrE,jnrF,jnrG,jnrH;
1221     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1222     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1223     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1224     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1225     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1226     real             rcutoff_scalar;
1227     real             *shiftvec,*fshift,*x,*f;
1228     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1229     real             scratch[4*DIM];
1230     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1231     real *           vdwioffsetptr1;
1232     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1233     real *           vdwioffsetptr2;
1234     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1235     real *           vdwioffsetptr3;
1236     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1237     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1238     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1239     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1240     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1241     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1242     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1243     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1244     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1245     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1246     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1247     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1248     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1249     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1250     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1251     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1252     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1253     real             *charge;
1254     __m256i          ewitab;
1255     __m128i          ewitab_lo,ewitab_hi;
1256     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1257     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1258     real             *ewtab;
1259     __m256           dummy_mask,cutoff_mask;
1260     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1261     __m256           one     = _mm256_set1_ps(1.0);
1262     __m256           two     = _mm256_set1_ps(2.0);
1263     x                = xx[0];
1264     f                = ff[0];
1265
1266     nri              = nlist->nri;
1267     iinr             = nlist->iinr;
1268     jindex           = nlist->jindex;
1269     jjnr             = nlist->jjnr;
1270     shiftidx         = nlist->shift;
1271     gid              = nlist->gid;
1272     shiftvec         = fr->shift_vec[0];
1273     fshift           = fr->fshift[0];
1274     facel            = _mm256_set1_ps(fr->ic->epsfac);
1275     charge           = mdatoms->chargeA;
1276
1277     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1278     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1279     beta2            = _mm256_mul_ps(beta,beta);
1280     beta3            = _mm256_mul_ps(beta,beta2);
1281
1282     ewtab            = fr->ic->tabq_coul_F;
1283     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1284     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1285
1286     /* Setup water-specific parameters */
1287     inr              = nlist->iinr[0];
1288     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1289     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1290     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1291
1292     jq1              = _mm256_set1_ps(charge[inr+1]);
1293     jq2              = _mm256_set1_ps(charge[inr+2]);
1294     jq3              = _mm256_set1_ps(charge[inr+3]);
1295     qq11             = _mm256_mul_ps(iq1,jq1);
1296     qq12             = _mm256_mul_ps(iq1,jq2);
1297     qq13             = _mm256_mul_ps(iq1,jq3);
1298     qq21             = _mm256_mul_ps(iq2,jq1);
1299     qq22             = _mm256_mul_ps(iq2,jq2);
1300     qq23             = _mm256_mul_ps(iq2,jq3);
1301     qq31             = _mm256_mul_ps(iq3,jq1);
1302     qq32             = _mm256_mul_ps(iq3,jq2);
1303     qq33             = _mm256_mul_ps(iq3,jq3);
1304
1305     /* Avoid stupid compiler warnings */
1306     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1307     j_coord_offsetA = 0;
1308     j_coord_offsetB = 0;
1309     j_coord_offsetC = 0;
1310     j_coord_offsetD = 0;
1311     j_coord_offsetE = 0;
1312     j_coord_offsetF = 0;
1313     j_coord_offsetG = 0;
1314     j_coord_offsetH = 0;
1315
1316     outeriter        = 0;
1317     inneriter        = 0;
1318
1319     for(iidx=0;iidx<4*DIM;iidx++)
1320     {
1321         scratch[iidx] = 0.0;
1322     }
1323
1324     /* Start outer loop over neighborlists */
1325     for(iidx=0; iidx<nri; iidx++)
1326     {
1327         /* Load shift vector for this list */
1328         i_shift_offset   = DIM*shiftidx[iidx];
1329
1330         /* Load limits for loop over neighbors */
1331         j_index_start    = jindex[iidx];
1332         j_index_end      = jindex[iidx+1];
1333
1334         /* Get outer coordinate index */
1335         inr              = iinr[iidx];
1336         i_coord_offset   = DIM*inr;
1337
1338         /* Load i particle coords and add shift vector */
1339         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1340                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1341
1342         fix1             = _mm256_setzero_ps();
1343         fiy1             = _mm256_setzero_ps();
1344         fiz1             = _mm256_setzero_ps();
1345         fix2             = _mm256_setzero_ps();
1346         fiy2             = _mm256_setzero_ps();
1347         fiz2             = _mm256_setzero_ps();
1348         fix3             = _mm256_setzero_ps();
1349         fiy3             = _mm256_setzero_ps();
1350         fiz3             = _mm256_setzero_ps();
1351
1352         /* Start inner kernel loop */
1353         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1354         {
1355
1356             /* Get j neighbor index, and coordinate index */
1357             jnrA             = jjnr[jidx];
1358             jnrB             = jjnr[jidx+1];
1359             jnrC             = jjnr[jidx+2];
1360             jnrD             = jjnr[jidx+3];
1361             jnrE             = jjnr[jidx+4];
1362             jnrF             = jjnr[jidx+5];
1363             jnrG             = jjnr[jidx+6];
1364             jnrH             = jjnr[jidx+7];
1365             j_coord_offsetA  = DIM*jnrA;
1366             j_coord_offsetB  = DIM*jnrB;
1367             j_coord_offsetC  = DIM*jnrC;
1368             j_coord_offsetD  = DIM*jnrD;
1369             j_coord_offsetE  = DIM*jnrE;
1370             j_coord_offsetF  = DIM*jnrF;
1371             j_coord_offsetG  = DIM*jnrG;
1372             j_coord_offsetH  = DIM*jnrH;
1373
1374             /* load j atom coordinates */
1375             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1376                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1377                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
1378                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
1379                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1380
1381             /* Calculate displacement vector */
1382             dx11             = _mm256_sub_ps(ix1,jx1);
1383             dy11             = _mm256_sub_ps(iy1,jy1);
1384             dz11             = _mm256_sub_ps(iz1,jz1);
1385             dx12             = _mm256_sub_ps(ix1,jx2);
1386             dy12             = _mm256_sub_ps(iy1,jy2);
1387             dz12             = _mm256_sub_ps(iz1,jz2);
1388             dx13             = _mm256_sub_ps(ix1,jx3);
1389             dy13             = _mm256_sub_ps(iy1,jy3);
1390             dz13             = _mm256_sub_ps(iz1,jz3);
1391             dx21             = _mm256_sub_ps(ix2,jx1);
1392             dy21             = _mm256_sub_ps(iy2,jy1);
1393             dz21             = _mm256_sub_ps(iz2,jz1);
1394             dx22             = _mm256_sub_ps(ix2,jx2);
1395             dy22             = _mm256_sub_ps(iy2,jy2);
1396             dz22             = _mm256_sub_ps(iz2,jz2);
1397             dx23             = _mm256_sub_ps(ix2,jx3);
1398             dy23             = _mm256_sub_ps(iy2,jy3);
1399             dz23             = _mm256_sub_ps(iz2,jz3);
1400             dx31             = _mm256_sub_ps(ix3,jx1);
1401             dy31             = _mm256_sub_ps(iy3,jy1);
1402             dz31             = _mm256_sub_ps(iz3,jz1);
1403             dx32             = _mm256_sub_ps(ix3,jx2);
1404             dy32             = _mm256_sub_ps(iy3,jy2);
1405             dz32             = _mm256_sub_ps(iz3,jz2);
1406             dx33             = _mm256_sub_ps(ix3,jx3);
1407             dy33             = _mm256_sub_ps(iy3,jy3);
1408             dz33             = _mm256_sub_ps(iz3,jz3);
1409
1410             /* Calculate squared distance and things based on it */
1411             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1412             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1413             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1414             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1415             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1416             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1417             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1418             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1419             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1420
1421             rinv11           = avx256_invsqrt_f(rsq11);
1422             rinv12           = avx256_invsqrt_f(rsq12);
1423             rinv13           = avx256_invsqrt_f(rsq13);
1424             rinv21           = avx256_invsqrt_f(rsq21);
1425             rinv22           = avx256_invsqrt_f(rsq22);
1426             rinv23           = avx256_invsqrt_f(rsq23);
1427             rinv31           = avx256_invsqrt_f(rsq31);
1428             rinv32           = avx256_invsqrt_f(rsq32);
1429             rinv33           = avx256_invsqrt_f(rsq33);
1430
1431             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1432             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1433             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1434             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1435             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1436             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1437             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1438             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1439             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1440
1441             fjx1             = _mm256_setzero_ps();
1442             fjy1             = _mm256_setzero_ps();
1443             fjz1             = _mm256_setzero_ps();
1444             fjx2             = _mm256_setzero_ps();
1445             fjy2             = _mm256_setzero_ps();
1446             fjz2             = _mm256_setzero_ps();
1447             fjx3             = _mm256_setzero_ps();
1448             fjy3             = _mm256_setzero_ps();
1449             fjz3             = _mm256_setzero_ps();
1450
1451             /**************************
1452              * CALCULATE INTERACTIONS *
1453              **************************/
1454
1455             r11              = _mm256_mul_ps(rsq11,rinv11);
1456
1457             /* EWALD ELECTROSTATICS */
1458
1459             /* Analytical PME correction */
1460             zeta2            = _mm256_mul_ps(beta2,rsq11);
1461             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1462             pmecorrF         = avx256_pmecorrF_f(zeta2);
1463             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1464             felec            = _mm256_mul_ps(qq11,felec);
1465
1466             fscal            = felec;
1467
1468             /* Calculate temporary vectorial force */
1469             tx               = _mm256_mul_ps(fscal,dx11);
1470             ty               = _mm256_mul_ps(fscal,dy11);
1471             tz               = _mm256_mul_ps(fscal,dz11);
1472
1473             /* Update vectorial force */
1474             fix1             = _mm256_add_ps(fix1,tx);
1475             fiy1             = _mm256_add_ps(fiy1,ty);
1476             fiz1             = _mm256_add_ps(fiz1,tz);
1477
1478             fjx1             = _mm256_add_ps(fjx1,tx);
1479             fjy1             = _mm256_add_ps(fjy1,ty);
1480             fjz1             = _mm256_add_ps(fjz1,tz);
1481
1482             /**************************
1483              * CALCULATE INTERACTIONS *
1484              **************************/
1485
1486             r12              = _mm256_mul_ps(rsq12,rinv12);
1487
1488             /* EWALD ELECTROSTATICS */
1489
1490             /* Analytical PME correction */
1491             zeta2            = _mm256_mul_ps(beta2,rsq12);
1492             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1493             pmecorrF         = avx256_pmecorrF_f(zeta2);
1494             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1495             felec            = _mm256_mul_ps(qq12,felec);
1496
1497             fscal            = felec;
1498
1499             /* Calculate temporary vectorial force */
1500             tx               = _mm256_mul_ps(fscal,dx12);
1501             ty               = _mm256_mul_ps(fscal,dy12);
1502             tz               = _mm256_mul_ps(fscal,dz12);
1503
1504             /* Update vectorial force */
1505             fix1             = _mm256_add_ps(fix1,tx);
1506             fiy1             = _mm256_add_ps(fiy1,ty);
1507             fiz1             = _mm256_add_ps(fiz1,tz);
1508
1509             fjx2             = _mm256_add_ps(fjx2,tx);
1510             fjy2             = _mm256_add_ps(fjy2,ty);
1511             fjz2             = _mm256_add_ps(fjz2,tz);
1512
1513             /**************************
1514              * CALCULATE INTERACTIONS *
1515              **************************/
1516
1517             r13              = _mm256_mul_ps(rsq13,rinv13);
1518
1519             /* EWALD ELECTROSTATICS */
1520
1521             /* Analytical PME correction */
1522             zeta2            = _mm256_mul_ps(beta2,rsq13);
1523             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
1524             pmecorrF         = avx256_pmecorrF_f(zeta2);
1525             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1526             felec            = _mm256_mul_ps(qq13,felec);
1527
1528             fscal            = felec;
1529
1530             /* Calculate temporary vectorial force */
1531             tx               = _mm256_mul_ps(fscal,dx13);
1532             ty               = _mm256_mul_ps(fscal,dy13);
1533             tz               = _mm256_mul_ps(fscal,dz13);
1534
1535             /* Update vectorial force */
1536             fix1             = _mm256_add_ps(fix1,tx);
1537             fiy1             = _mm256_add_ps(fiy1,ty);
1538             fiz1             = _mm256_add_ps(fiz1,tz);
1539
1540             fjx3             = _mm256_add_ps(fjx3,tx);
1541             fjy3             = _mm256_add_ps(fjy3,ty);
1542             fjz3             = _mm256_add_ps(fjz3,tz);
1543
1544             /**************************
1545              * CALCULATE INTERACTIONS *
1546              **************************/
1547
1548             r21              = _mm256_mul_ps(rsq21,rinv21);
1549
1550             /* EWALD ELECTROSTATICS */
1551
1552             /* Analytical PME correction */
1553             zeta2            = _mm256_mul_ps(beta2,rsq21);
1554             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1555             pmecorrF         = avx256_pmecorrF_f(zeta2);
1556             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1557             felec            = _mm256_mul_ps(qq21,felec);
1558
1559             fscal            = felec;
1560
1561             /* Calculate temporary vectorial force */
1562             tx               = _mm256_mul_ps(fscal,dx21);
1563             ty               = _mm256_mul_ps(fscal,dy21);
1564             tz               = _mm256_mul_ps(fscal,dz21);
1565
1566             /* Update vectorial force */
1567             fix2             = _mm256_add_ps(fix2,tx);
1568             fiy2             = _mm256_add_ps(fiy2,ty);
1569             fiz2             = _mm256_add_ps(fiz2,tz);
1570
1571             fjx1             = _mm256_add_ps(fjx1,tx);
1572             fjy1             = _mm256_add_ps(fjy1,ty);
1573             fjz1             = _mm256_add_ps(fjz1,tz);
1574
1575             /**************************
1576              * CALCULATE INTERACTIONS *
1577              **************************/
1578
1579             r22              = _mm256_mul_ps(rsq22,rinv22);
1580
1581             /* EWALD ELECTROSTATICS */
1582
1583             /* Analytical PME correction */
1584             zeta2            = _mm256_mul_ps(beta2,rsq22);
1585             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1586             pmecorrF         = avx256_pmecorrF_f(zeta2);
1587             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1588             felec            = _mm256_mul_ps(qq22,felec);
1589
1590             fscal            = felec;
1591
1592             /* Calculate temporary vectorial force */
1593             tx               = _mm256_mul_ps(fscal,dx22);
1594             ty               = _mm256_mul_ps(fscal,dy22);
1595             tz               = _mm256_mul_ps(fscal,dz22);
1596
1597             /* Update vectorial force */
1598             fix2             = _mm256_add_ps(fix2,tx);
1599             fiy2             = _mm256_add_ps(fiy2,ty);
1600             fiz2             = _mm256_add_ps(fiz2,tz);
1601
1602             fjx2             = _mm256_add_ps(fjx2,tx);
1603             fjy2             = _mm256_add_ps(fjy2,ty);
1604             fjz2             = _mm256_add_ps(fjz2,tz);
1605
1606             /**************************
1607              * CALCULATE INTERACTIONS *
1608              **************************/
1609
1610             r23              = _mm256_mul_ps(rsq23,rinv23);
1611
1612             /* EWALD ELECTROSTATICS */
1613
1614             /* Analytical PME correction */
1615             zeta2            = _mm256_mul_ps(beta2,rsq23);
1616             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1617             pmecorrF         = avx256_pmecorrF_f(zeta2);
1618             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1619             felec            = _mm256_mul_ps(qq23,felec);
1620
1621             fscal            = felec;
1622
1623             /* Calculate temporary vectorial force */
1624             tx               = _mm256_mul_ps(fscal,dx23);
1625             ty               = _mm256_mul_ps(fscal,dy23);
1626             tz               = _mm256_mul_ps(fscal,dz23);
1627
1628             /* Update vectorial force */
1629             fix2             = _mm256_add_ps(fix2,tx);
1630             fiy2             = _mm256_add_ps(fiy2,ty);
1631             fiz2             = _mm256_add_ps(fiz2,tz);
1632
1633             fjx3             = _mm256_add_ps(fjx3,tx);
1634             fjy3             = _mm256_add_ps(fjy3,ty);
1635             fjz3             = _mm256_add_ps(fjz3,tz);
1636
1637             /**************************
1638              * CALCULATE INTERACTIONS *
1639              **************************/
1640
1641             r31              = _mm256_mul_ps(rsq31,rinv31);
1642
1643             /* EWALD ELECTROSTATICS */
1644
1645             /* Analytical PME correction */
1646             zeta2            = _mm256_mul_ps(beta2,rsq31);
1647             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1648             pmecorrF         = avx256_pmecorrF_f(zeta2);
1649             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1650             felec            = _mm256_mul_ps(qq31,felec);
1651
1652             fscal            = felec;
1653
1654             /* Calculate temporary vectorial force */
1655             tx               = _mm256_mul_ps(fscal,dx31);
1656             ty               = _mm256_mul_ps(fscal,dy31);
1657             tz               = _mm256_mul_ps(fscal,dz31);
1658
1659             /* Update vectorial force */
1660             fix3             = _mm256_add_ps(fix3,tx);
1661             fiy3             = _mm256_add_ps(fiy3,ty);
1662             fiz3             = _mm256_add_ps(fiz3,tz);
1663
1664             fjx1             = _mm256_add_ps(fjx1,tx);
1665             fjy1             = _mm256_add_ps(fjy1,ty);
1666             fjz1             = _mm256_add_ps(fjz1,tz);
1667
1668             /**************************
1669              * CALCULATE INTERACTIONS *
1670              **************************/
1671
1672             r32              = _mm256_mul_ps(rsq32,rinv32);
1673
1674             /* EWALD ELECTROSTATICS */
1675
1676             /* Analytical PME correction */
1677             zeta2            = _mm256_mul_ps(beta2,rsq32);
1678             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1679             pmecorrF         = avx256_pmecorrF_f(zeta2);
1680             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1681             felec            = _mm256_mul_ps(qq32,felec);
1682
1683             fscal            = felec;
1684
1685             /* Calculate temporary vectorial force */
1686             tx               = _mm256_mul_ps(fscal,dx32);
1687             ty               = _mm256_mul_ps(fscal,dy32);
1688             tz               = _mm256_mul_ps(fscal,dz32);
1689
1690             /* Update vectorial force */
1691             fix3             = _mm256_add_ps(fix3,tx);
1692             fiy3             = _mm256_add_ps(fiy3,ty);
1693             fiz3             = _mm256_add_ps(fiz3,tz);
1694
1695             fjx2             = _mm256_add_ps(fjx2,tx);
1696             fjy2             = _mm256_add_ps(fjy2,ty);
1697             fjz2             = _mm256_add_ps(fjz2,tz);
1698
1699             /**************************
1700              * CALCULATE INTERACTIONS *
1701              **************************/
1702
1703             r33              = _mm256_mul_ps(rsq33,rinv33);
1704
1705             /* EWALD ELECTROSTATICS */
1706
1707             /* Analytical PME correction */
1708             zeta2            = _mm256_mul_ps(beta2,rsq33);
1709             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
1710             pmecorrF         = avx256_pmecorrF_f(zeta2);
1711             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1712             felec            = _mm256_mul_ps(qq33,felec);
1713
1714             fscal            = felec;
1715
1716             /* Calculate temporary vectorial force */
1717             tx               = _mm256_mul_ps(fscal,dx33);
1718             ty               = _mm256_mul_ps(fscal,dy33);
1719             tz               = _mm256_mul_ps(fscal,dz33);
1720
1721             /* Update vectorial force */
1722             fix3             = _mm256_add_ps(fix3,tx);
1723             fiy3             = _mm256_add_ps(fiy3,ty);
1724             fiz3             = _mm256_add_ps(fiz3,tz);
1725
1726             fjx3             = _mm256_add_ps(fjx3,tx);
1727             fjy3             = _mm256_add_ps(fjy3,ty);
1728             fjz3             = _mm256_add_ps(fjz3,tz);
1729
1730             fjptrA             = f+j_coord_offsetA;
1731             fjptrB             = f+j_coord_offsetB;
1732             fjptrC             = f+j_coord_offsetC;
1733             fjptrD             = f+j_coord_offsetD;
1734             fjptrE             = f+j_coord_offsetE;
1735             fjptrF             = f+j_coord_offsetF;
1736             fjptrG             = f+j_coord_offsetG;
1737             fjptrH             = f+j_coord_offsetH;
1738
1739             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1740                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
1741                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1742
1743             /* Inner loop uses 504 flops */
1744         }
1745
1746         if(jidx<j_index_end)
1747         {
1748
1749             /* Get j neighbor index, and coordinate index */
1750             jnrlistA         = jjnr[jidx];
1751             jnrlistB         = jjnr[jidx+1];
1752             jnrlistC         = jjnr[jidx+2];
1753             jnrlistD         = jjnr[jidx+3];
1754             jnrlistE         = jjnr[jidx+4];
1755             jnrlistF         = jjnr[jidx+5];
1756             jnrlistG         = jjnr[jidx+6];
1757             jnrlistH         = jjnr[jidx+7];
1758             /* Sign of each element will be negative for non-real atoms.
1759              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1760              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1761              */
1762             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1763                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1764
1765             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1766             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1767             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1768             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1769             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1770             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1771             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1772             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1773             j_coord_offsetA  = DIM*jnrA;
1774             j_coord_offsetB  = DIM*jnrB;
1775             j_coord_offsetC  = DIM*jnrC;
1776             j_coord_offsetD  = DIM*jnrD;
1777             j_coord_offsetE  = DIM*jnrE;
1778             j_coord_offsetF  = DIM*jnrF;
1779             j_coord_offsetG  = DIM*jnrG;
1780             j_coord_offsetH  = DIM*jnrH;
1781
1782             /* load j atom coordinates */
1783             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1784                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1785                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
1786                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
1787                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1788
1789             /* Calculate displacement vector */
1790             dx11             = _mm256_sub_ps(ix1,jx1);
1791             dy11             = _mm256_sub_ps(iy1,jy1);
1792             dz11             = _mm256_sub_ps(iz1,jz1);
1793             dx12             = _mm256_sub_ps(ix1,jx2);
1794             dy12             = _mm256_sub_ps(iy1,jy2);
1795             dz12             = _mm256_sub_ps(iz1,jz2);
1796             dx13             = _mm256_sub_ps(ix1,jx3);
1797             dy13             = _mm256_sub_ps(iy1,jy3);
1798             dz13             = _mm256_sub_ps(iz1,jz3);
1799             dx21             = _mm256_sub_ps(ix2,jx1);
1800             dy21             = _mm256_sub_ps(iy2,jy1);
1801             dz21             = _mm256_sub_ps(iz2,jz1);
1802             dx22             = _mm256_sub_ps(ix2,jx2);
1803             dy22             = _mm256_sub_ps(iy2,jy2);
1804             dz22             = _mm256_sub_ps(iz2,jz2);
1805             dx23             = _mm256_sub_ps(ix2,jx3);
1806             dy23             = _mm256_sub_ps(iy2,jy3);
1807             dz23             = _mm256_sub_ps(iz2,jz3);
1808             dx31             = _mm256_sub_ps(ix3,jx1);
1809             dy31             = _mm256_sub_ps(iy3,jy1);
1810             dz31             = _mm256_sub_ps(iz3,jz1);
1811             dx32             = _mm256_sub_ps(ix3,jx2);
1812             dy32             = _mm256_sub_ps(iy3,jy2);
1813             dz32             = _mm256_sub_ps(iz3,jz2);
1814             dx33             = _mm256_sub_ps(ix3,jx3);
1815             dy33             = _mm256_sub_ps(iy3,jy3);
1816             dz33             = _mm256_sub_ps(iz3,jz3);
1817
1818             /* Calculate squared distance and things based on it */
1819             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1820             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1821             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1822             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1823             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1824             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1825             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1826             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1827             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1828
1829             rinv11           = avx256_invsqrt_f(rsq11);
1830             rinv12           = avx256_invsqrt_f(rsq12);
1831             rinv13           = avx256_invsqrt_f(rsq13);
1832             rinv21           = avx256_invsqrt_f(rsq21);
1833             rinv22           = avx256_invsqrt_f(rsq22);
1834             rinv23           = avx256_invsqrt_f(rsq23);
1835             rinv31           = avx256_invsqrt_f(rsq31);
1836             rinv32           = avx256_invsqrt_f(rsq32);
1837             rinv33           = avx256_invsqrt_f(rsq33);
1838
1839             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1840             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1841             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1842             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1843             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1844             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1845             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1846             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1847             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1848
1849             fjx1             = _mm256_setzero_ps();
1850             fjy1             = _mm256_setzero_ps();
1851             fjz1             = _mm256_setzero_ps();
1852             fjx2             = _mm256_setzero_ps();
1853             fjy2             = _mm256_setzero_ps();
1854             fjz2             = _mm256_setzero_ps();
1855             fjx3             = _mm256_setzero_ps();
1856             fjy3             = _mm256_setzero_ps();
1857             fjz3             = _mm256_setzero_ps();
1858
1859             /**************************
1860              * CALCULATE INTERACTIONS *
1861              **************************/
1862
1863             r11              = _mm256_mul_ps(rsq11,rinv11);
1864             r11              = _mm256_andnot_ps(dummy_mask,r11);
1865
1866             /* EWALD ELECTROSTATICS */
1867
1868             /* Analytical PME correction */
1869             zeta2            = _mm256_mul_ps(beta2,rsq11);
1870             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1871             pmecorrF         = avx256_pmecorrF_f(zeta2);
1872             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1873             felec            = _mm256_mul_ps(qq11,felec);
1874
1875             fscal            = felec;
1876
1877             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1878
1879             /* Calculate temporary vectorial force */
1880             tx               = _mm256_mul_ps(fscal,dx11);
1881             ty               = _mm256_mul_ps(fscal,dy11);
1882             tz               = _mm256_mul_ps(fscal,dz11);
1883
1884             /* Update vectorial force */
1885             fix1             = _mm256_add_ps(fix1,tx);
1886             fiy1             = _mm256_add_ps(fiy1,ty);
1887             fiz1             = _mm256_add_ps(fiz1,tz);
1888
1889             fjx1             = _mm256_add_ps(fjx1,tx);
1890             fjy1             = _mm256_add_ps(fjy1,ty);
1891             fjz1             = _mm256_add_ps(fjz1,tz);
1892
1893             /**************************
1894              * CALCULATE INTERACTIONS *
1895              **************************/
1896
1897             r12              = _mm256_mul_ps(rsq12,rinv12);
1898             r12              = _mm256_andnot_ps(dummy_mask,r12);
1899
1900             /* EWALD ELECTROSTATICS */
1901
1902             /* Analytical PME correction */
1903             zeta2            = _mm256_mul_ps(beta2,rsq12);
1904             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1905             pmecorrF         = avx256_pmecorrF_f(zeta2);
1906             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1907             felec            = _mm256_mul_ps(qq12,felec);
1908
1909             fscal            = felec;
1910
1911             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1912
1913             /* Calculate temporary vectorial force */
1914             tx               = _mm256_mul_ps(fscal,dx12);
1915             ty               = _mm256_mul_ps(fscal,dy12);
1916             tz               = _mm256_mul_ps(fscal,dz12);
1917
1918             /* Update vectorial force */
1919             fix1             = _mm256_add_ps(fix1,tx);
1920             fiy1             = _mm256_add_ps(fiy1,ty);
1921             fiz1             = _mm256_add_ps(fiz1,tz);
1922
1923             fjx2             = _mm256_add_ps(fjx2,tx);
1924             fjy2             = _mm256_add_ps(fjy2,ty);
1925             fjz2             = _mm256_add_ps(fjz2,tz);
1926
1927             /**************************
1928              * CALCULATE INTERACTIONS *
1929              **************************/
1930
1931             r13              = _mm256_mul_ps(rsq13,rinv13);
1932             r13              = _mm256_andnot_ps(dummy_mask,r13);
1933
1934             /* EWALD ELECTROSTATICS */
1935
1936             /* Analytical PME correction */
1937             zeta2            = _mm256_mul_ps(beta2,rsq13);
1938             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
1939             pmecorrF         = avx256_pmecorrF_f(zeta2);
1940             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1941             felec            = _mm256_mul_ps(qq13,felec);
1942
1943             fscal            = felec;
1944
1945             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1946
1947             /* Calculate temporary vectorial force */
1948             tx               = _mm256_mul_ps(fscal,dx13);
1949             ty               = _mm256_mul_ps(fscal,dy13);
1950             tz               = _mm256_mul_ps(fscal,dz13);
1951
1952             /* Update vectorial force */
1953             fix1             = _mm256_add_ps(fix1,tx);
1954             fiy1             = _mm256_add_ps(fiy1,ty);
1955             fiz1             = _mm256_add_ps(fiz1,tz);
1956
1957             fjx3             = _mm256_add_ps(fjx3,tx);
1958             fjy3             = _mm256_add_ps(fjy3,ty);
1959             fjz3             = _mm256_add_ps(fjz3,tz);
1960
1961             /**************************
1962              * CALCULATE INTERACTIONS *
1963              **************************/
1964
1965             r21              = _mm256_mul_ps(rsq21,rinv21);
1966             r21              = _mm256_andnot_ps(dummy_mask,r21);
1967
1968             /* EWALD ELECTROSTATICS */
1969
1970             /* Analytical PME correction */
1971             zeta2            = _mm256_mul_ps(beta2,rsq21);
1972             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1973             pmecorrF         = avx256_pmecorrF_f(zeta2);
1974             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1975             felec            = _mm256_mul_ps(qq21,felec);
1976
1977             fscal            = felec;
1978
1979             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1980
1981             /* Calculate temporary vectorial force */
1982             tx               = _mm256_mul_ps(fscal,dx21);
1983             ty               = _mm256_mul_ps(fscal,dy21);
1984             tz               = _mm256_mul_ps(fscal,dz21);
1985
1986             /* Update vectorial force */
1987             fix2             = _mm256_add_ps(fix2,tx);
1988             fiy2             = _mm256_add_ps(fiy2,ty);
1989             fiz2             = _mm256_add_ps(fiz2,tz);
1990
1991             fjx1             = _mm256_add_ps(fjx1,tx);
1992             fjy1             = _mm256_add_ps(fjy1,ty);
1993             fjz1             = _mm256_add_ps(fjz1,tz);
1994
1995             /**************************
1996              * CALCULATE INTERACTIONS *
1997              **************************/
1998
1999             r22              = _mm256_mul_ps(rsq22,rinv22);
2000             r22              = _mm256_andnot_ps(dummy_mask,r22);
2001
2002             /* EWALD ELECTROSTATICS */
2003
2004             /* Analytical PME correction */
2005             zeta2            = _mm256_mul_ps(beta2,rsq22);
2006             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2007             pmecorrF         = avx256_pmecorrF_f(zeta2);
2008             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2009             felec            = _mm256_mul_ps(qq22,felec);
2010
2011             fscal            = felec;
2012
2013             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2014
2015             /* Calculate temporary vectorial force */
2016             tx               = _mm256_mul_ps(fscal,dx22);
2017             ty               = _mm256_mul_ps(fscal,dy22);
2018             tz               = _mm256_mul_ps(fscal,dz22);
2019
2020             /* Update vectorial force */
2021             fix2             = _mm256_add_ps(fix2,tx);
2022             fiy2             = _mm256_add_ps(fiy2,ty);
2023             fiz2             = _mm256_add_ps(fiz2,tz);
2024
2025             fjx2             = _mm256_add_ps(fjx2,tx);
2026             fjy2             = _mm256_add_ps(fjy2,ty);
2027             fjz2             = _mm256_add_ps(fjz2,tz);
2028
2029             /**************************
2030              * CALCULATE INTERACTIONS *
2031              **************************/
2032
2033             r23              = _mm256_mul_ps(rsq23,rinv23);
2034             r23              = _mm256_andnot_ps(dummy_mask,r23);
2035
2036             /* EWALD ELECTROSTATICS */
2037
2038             /* Analytical PME correction */
2039             zeta2            = _mm256_mul_ps(beta2,rsq23);
2040             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
2041             pmecorrF         = avx256_pmecorrF_f(zeta2);
2042             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2043             felec            = _mm256_mul_ps(qq23,felec);
2044
2045             fscal            = felec;
2046
2047             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2048
2049             /* Calculate temporary vectorial force */
2050             tx               = _mm256_mul_ps(fscal,dx23);
2051             ty               = _mm256_mul_ps(fscal,dy23);
2052             tz               = _mm256_mul_ps(fscal,dz23);
2053
2054             /* Update vectorial force */
2055             fix2             = _mm256_add_ps(fix2,tx);
2056             fiy2             = _mm256_add_ps(fiy2,ty);
2057             fiz2             = _mm256_add_ps(fiz2,tz);
2058
2059             fjx3             = _mm256_add_ps(fjx3,tx);
2060             fjy3             = _mm256_add_ps(fjy3,ty);
2061             fjz3             = _mm256_add_ps(fjz3,tz);
2062
2063             /**************************
2064              * CALCULATE INTERACTIONS *
2065              **************************/
2066
2067             r31              = _mm256_mul_ps(rsq31,rinv31);
2068             r31              = _mm256_andnot_ps(dummy_mask,r31);
2069
2070             /* EWALD ELECTROSTATICS */
2071
2072             /* Analytical PME correction */
2073             zeta2            = _mm256_mul_ps(beta2,rsq31);
2074             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
2075             pmecorrF         = avx256_pmecorrF_f(zeta2);
2076             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2077             felec            = _mm256_mul_ps(qq31,felec);
2078
2079             fscal            = felec;
2080
2081             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2082
2083             /* Calculate temporary vectorial force */
2084             tx               = _mm256_mul_ps(fscal,dx31);
2085             ty               = _mm256_mul_ps(fscal,dy31);
2086             tz               = _mm256_mul_ps(fscal,dz31);
2087
2088             /* Update vectorial force */
2089             fix3             = _mm256_add_ps(fix3,tx);
2090             fiy3             = _mm256_add_ps(fiy3,ty);
2091             fiz3             = _mm256_add_ps(fiz3,tz);
2092
2093             fjx1             = _mm256_add_ps(fjx1,tx);
2094             fjy1             = _mm256_add_ps(fjy1,ty);
2095             fjz1             = _mm256_add_ps(fjz1,tz);
2096
2097             /**************************
2098              * CALCULATE INTERACTIONS *
2099              **************************/
2100
2101             r32              = _mm256_mul_ps(rsq32,rinv32);
2102             r32              = _mm256_andnot_ps(dummy_mask,r32);
2103
2104             /* EWALD ELECTROSTATICS */
2105
2106             /* Analytical PME correction */
2107             zeta2            = _mm256_mul_ps(beta2,rsq32);
2108             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
2109             pmecorrF         = avx256_pmecorrF_f(zeta2);
2110             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2111             felec            = _mm256_mul_ps(qq32,felec);
2112
2113             fscal            = felec;
2114
2115             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2116
2117             /* Calculate temporary vectorial force */
2118             tx               = _mm256_mul_ps(fscal,dx32);
2119             ty               = _mm256_mul_ps(fscal,dy32);
2120             tz               = _mm256_mul_ps(fscal,dz32);
2121
2122             /* Update vectorial force */
2123             fix3             = _mm256_add_ps(fix3,tx);
2124             fiy3             = _mm256_add_ps(fiy3,ty);
2125             fiz3             = _mm256_add_ps(fiz3,tz);
2126
2127             fjx2             = _mm256_add_ps(fjx2,tx);
2128             fjy2             = _mm256_add_ps(fjy2,ty);
2129             fjz2             = _mm256_add_ps(fjz2,tz);
2130
2131             /**************************
2132              * CALCULATE INTERACTIONS *
2133              **************************/
2134
2135             r33              = _mm256_mul_ps(rsq33,rinv33);
2136             r33              = _mm256_andnot_ps(dummy_mask,r33);
2137
2138             /* EWALD ELECTROSTATICS */
2139
2140             /* Analytical PME correction */
2141             zeta2            = _mm256_mul_ps(beta2,rsq33);
2142             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
2143             pmecorrF         = avx256_pmecorrF_f(zeta2);
2144             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2145             felec            = _mm256_mul_ps(qq33,felec);
2146
2147             fscal            = felec;
2148
2149             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2150
2151             /* Calculate temporary vectorial force */
2152             tx               = _mm256_mul_ps(fscal,dx33);
2153             ty               = _mm256_mul_ps(fscal,dy33);
2154             tz               = _mm256_mul_ps(fscal,dz33);
2155
2156             /* Update vectorial force */
2157             fix3             = _mm256_add_ps(fix3,tx);
2158             fiy3             = _mm256_add_ps(fiy3,ty);
2159             fiz3             = _mm256_add_ps(fiz3,tz);
2160
2161             fjx3             = _mm256_add_ps(fjx3,tx);
2162             fjy3             = _mm256_add_ps(fjy3,ty);
2163             fjz3             = _mm256_add_ps(fjz3,tz);
2164
2165             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2166             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2167             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2168             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2169             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2170             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2171             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2172             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2173
2174             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2175                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
2176                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2177
2178             /* Inner loop uses 513 flops */
2179         }
2180
2181         /* End of innermost loop */
2182
2183         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2184                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
2185
2186         /* Increment number of inner iterations */
2187         inneriter                  += j_index_end - j_index_start;
2188
2189         /* Outer loop uses 18 flops */
2190     }
2191
2192     /* Increment number of outer iterations */
2193     outeriter        += nri;
2194
2195     /* Update outer/inner flops */
2196
2197     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*513);
2198 }