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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4W4_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwNone_GeomW4W4_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  95     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  97     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
  99     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 100     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 101     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 102     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 103     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 104     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 105     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 106     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 107     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 108     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 109     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 110     real             *charge;
 111     __m256i          ewitab;
 112     __m128i          ewitab_lo,ewitab_hi;
 113     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 114     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 115     real             *ewtab;
 116     __m256           dummy_mask,cutoff_mask;
 117     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 118     __m256           one     = _mm256_set1_ps(1.0);
 119     __m256           two     = _mm256_set1_ps(2.0);
 120     x                = xx[0];
 121     f                = ff[0];
 122
 123     nri              = nlist->nri;
 124     iinr             = nlist->iinr;
 125     jindex           = nlist->jindex;
 126     jjnr             = nlist->jjnr;
 127     shiftidx         = nlist->shift;
 128     gid              = nlist->gid;
 129     shiftvec         = fr->shift_vec[0];
 130     fshift           = fr->fshift[0];
 131     facel            = _mm256_set1_ps(fr->epsfac);
 132     charge           = mdatoms->chargeA;
 133
 134     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 135     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 136     beta2            = _mm256_mul_ps(beta,beta);
 137     beta3            = _mm256_mul_ps(beta,beta2);
 138
 139     ewtab            = fr->ic->tabq_coul_FDV0;
 140     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 141     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 146     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 147     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 148
 149     jq1              = _mm256_set1_ps(charge[inr+1]);
 150     jq2              = _mm256_set1_ps(charge[inr+2]);
 151     jq3              = _mm256_set1_ps(charge[inr+3]);
 152     qq11             = _mm256_mul_ps(iq1,jq1);
 153     qq12             = _mm256_mul_ps(iq1,jq2);
 154     qq13             = _mm256_mul_ps(iq1,jq3);
 155     qq21             = _mm256_mul_ps(iq2,jq1);
 156     qq22             = _mm256_mul_ps(iq2,jq2);
 157     qq23             = _mm256_mul_ps(iq2,jq3);
 158     qq31             = _mm256_mul_ps(iq3,jq1);
 159     qq32             = _mm256_mul_ps(iq3,jq2);
 160     qq33             = _mm256_mul_ps(iq3,jq3);
 161
 162     /* Avoid stupid compiler warnings */
 163     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 164     j_coord_offsetA = 0;
 165     j_coord_offsetB = 0;
 166     j_coord_offsetC = 0;
 167     j_coord_offsetD = 0;
 168     j_coord_offsetE = 0;
 169     j_coord_offsetF = 0;
 170     j_coord_offsetG = 0;
 171     j_coord_offsetH = 0;
 172
 173     outeriter        = 0;
 174     inneriter        = 0;
 175
 176     for(iidx=0;iidx<4*DIM;iidx++)
 177     {
 178         scratch[iidx] = 0.0;
 179     }
 180
 181     /* Start outer loop over neighborlists */
 182     for(iidx=0; iidx<nri; iidx++)
 183     {
 184         /* Load shift vector for this list */
 185         i_shift_offset   = DIM*shiftidx[iidx];
 186
 187         /* Load limits for loop over neighbors */
 188         j_index_start    = jindex[iidx];
 189         j_index_end      = jindex[iidx+1];
 190
 191         /* Get outer coordinate index */
 192         inr              = iinr[iidx];
 193         i_coord_offset   = DIM*inr;
 194
 195         /* Load i particle coords and add shift vector */
 196         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 197                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 198
 199         fix1             = _mm256_setzero_ps();
 200         fiy1             = _mm256_setzero_ps();
 201         fiz1             = _mm256_setzero_ps();
 202         fix2             = _mm256_setzero_ps();
 203         fiy2             = _mm256_setzero_ps();
 204         fiz2             = _mm256_setzero_ps();
 205         fix3             = _mm256_setzero_ps();
 206         fiy3             = _mm256_setzero_ps();
 207         fiz3             = _mm256_setzero_ps();
 208
 209         /* Reset potential sums */
 210         velecsum         = _mm256_setzero_ps();
 211
 212         /* Start inner kernel loop */
 213         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 214         {
 215
 216             /* Get j neighbor index, and coordinate index */
 217             jnrA             = jjnr[jidx];
 218             jnrB             = jjnr[jidx+1];
 219             jnrC             = jjnr[jidx+2];
 220             jnrD             = jjnr[jidx+3];
 221             jnrE             = jjnr[jidx+4];
 222             jnrF             = jjnr[jidx+5];
 223             jnrG             = jjnr[jidx+6];
 224             jnrH             = jjnr[jidx+7];
 225             j_coord_offsetA  = DIM*jnrA;
 226             j_coord_offsetB  = DIM*jnrB;
 227             j_coord_offsetC  = DIM*jnrC;
 228             j_coord_offsetD  = DIM*jnrD;
 229             j_coord_offsetE  = DIM*jnrE;
 230             j_coord_offsetF  = DIM*jnrF;
 231             j_coord_offsetG  = DIM*jnrG;
 232             j_coord_offsetH  = DIM*jnrH;
 233
 234             /* load j atom coordinates */
 235             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 236                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 237                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
 238                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
 239                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 240
 241             /* Calculate displacement vector */
 242             dx11             = _mm256_sub_ps(ix1,jx1);
 243             dy11             = _mm256_sub_ps(iy1,jy1);
 244             dz11             = _mm256_sub_ps(iz1,jz1);
 245             dx12             = _mm256_sub_ps(ix1,jx2);
 246             dy12             = _mm256_sub_ps(iy1,jy2);
 247             dz12             = _mm256_sub_ps(iz1,jz2);
 248             dx13             = _mm256_sub_ps(ix1,jx3);
 249             dy13             = _mm256_sub_ps(iy1,jy3);
 250             dz13             = _mm256_sub_ps(iz1,jz3);
 251             dx21             = _mm256_sub_ps(ix2,jx1);
 252             dy21             = _mm256_sub_ps(iy2,jy1);
 253             dz21             = _mm256_sub_ps(iz2,jz1);
 254             dx22             = _mm256_sub_ps(ix2,jx2);
 255             dy22             = _mm256_sub_ps(iy2,jy2);
 256             dz22             = _mm256_sub_ps(iz2,jz2);
 257             dx23             = _mm256_sub_ps(ix2,jx3);
 258             dy23             = _mm256_sub_ps(iy2,jy3);
 259             dz23             = _mm256_sub_ps(iz2,jz3);
 260             dx31             = _mm256_sub_ps(ix3,jx1);
 261             dy31             = _mm256_sub_ps(iy3,jy1);
 262             dz31             = _mm256_sub_ps(iz3,jz1);
 263             dx32             = _mm256_sub_ps(ix3,jx2);
 264             dy32             = _mm256_sub_ps(iy3,jy2);
 265             dz32             = _mm256_sub_ps(iz3,jz2);
 266             dx33             = _mm256_sub_ps(ix3,jx3);
 267             dy33             = _mm256_sub_ps(iy3,jy3);
 268             dz33             = _mm256_sub_ps(iz3,jz3);
 269
 270             /* Calculate squared distance and things based on it */
 271             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 272             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 273             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 274             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 275             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 276             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 277             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 278             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 279             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 280
 281             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 282             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 283             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 284             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 285             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 286             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 287             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 288             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 289             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 290
 291             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 292             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 293             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 294             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 295             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 296             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 297             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 298             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 299             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 300
 301             fjx1             = _mm256_setzero_ps();
 302             fjy1             = _mm256_setzero_ps();
 303             fjz1             = _mm256_setzero_ps();
 304             fjx2             = _mm256_setzero_ps();
 305             fjy2             = _mm256_setzero_ps();
 306             fjz2             = _mm256_setzero_ps();
 307             fjx3             = _mm256_setzero_ps();
 308             fjy3             = _mm256_setzero_ps();
 309             fjz3             = _mm256_setzero_ps();
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             r11              = _mm256_mul_ps(rsq11,rinv11);
 316
 317             /* EWALD ELECTROSTATICS */
 318
 319             /* Analytical PME correction */
 320             zeta2            = _mm256_mul_ps(beta2,rsq11);
 321             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 322             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 323             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 324             felec            = _mm256_mul_ps(qq11,felec);
 325             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 326             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 327             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 328             velec            = _mm256_mul_ps(qq11,velec);
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velecsum         = _mm256_add_ps(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm256_mul_ps(fscal,dx11);
 337             ty               = _mm256_mul_ps(fscal,dy11);
 338             tz               = _mm256_mul_ps(fscal,dz11);
 339
 340             /* Update vectorial force */
 341             fix1             = _mm256_add_ps(fix1,tx);
 342             fiy1             = _mm256_add_ps(fiy1,ty);
 343             fiz1             = _mm256_add_ps(fiz1,tz);
 344
 345             fjx1             = _mm256_add_ps(fjx1,tx);
 346             fjy1             = _mm256_add_ps(fjy1,ty);
 347             fjz1             = _mm256_add_ps(fjz1,tz);
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             r12              = _mm256_mul_ps(rsq12,rinv12);
 354
 355             /* EWALD ELECTROSTATICS */
 356
 357             /* Analytical PME correction */
 358             zeta2            = _mm256_mul_ps(beta2,rsq12);
 359             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 360             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 361             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 362             felec            = _mm256_mul_ps(qq12,felec);
 363             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 364             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 365             velec            = _mm256_sub_ps(rinv12,pmecorrV);
 366             velec            = _mm256_mul_ps(qq12,velec);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velecsum         = _mm256_add_ps(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             /* Calculate temporary vectorial force */
 374             tx               = _mm256_mul_ps(fscal,dx12);
 375             ty               = _mm256_mul_ps(fscal,dy12);
 376             tz               = _mm256_mul_ps(fscal,dz12);
 377
 378             /* Update vectorial force */
 379             fix1             = _mm256_add_ps(fix1,tx);
 380             fiy1             = _mm256_add_ps(fiy1,ty);
 381             fiz1             = _mm256_add_ps(fiz1,tz);
 382
 383             fjx2             = _mm256_add_ps(fjx2,tx);
 384             fjy2             = _mm256_add_ps(fjy2,ty);
 385             fjz2             = _mm256_add_ps(fjz2,tz);
 386
 387             /**************************
 388              * CALCULATE INTERACTIONS *
 389              **************************/
 390
 391             r13              = _mm256_mul_ps(rsq13,rinv13);
 392
 393             /* EWALD ELECTROSTATICS */
 394
 395             /* Analytical PME correction */
 396             zeta2            = _mm256_mul_ps(beta2,rsq13);
 397             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
 398             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 399             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 400             felec            = _mm256_mul_ps(qq13,felec);
 401             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 402             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 403             velec            = _mm256_sub_ps(rinv13,pmecorrV);
 404             velec            = _mm256_mul_ps(qq13,velec);
 405
 406             /* Update potential sum for this i atom from the interaction with this j atom. */
 407             velecsum         = _mm256_add_ps(velecsum,velec);
 408
 409             fscal            = felec;
 410
 411             /* Calculate temporary vectorial force */
 412             tx               = _mm256_mul_ps(fscal,dx13);
 413             ty               = _mm256_mul_ps(fscal,dy13);
 414             tz               = _mm256_mul_ps(fscal,dz13);
 415
 416             /* Update vectorial force */
 417             fix1             = _mm256_add_ps(fix1,tx);
 418             fiy1             = _mm256_add_ps(fiy1,ty);
 419             fiz1             = _mm256_add_ps(fiz1,tz);
 420
 421             fjx3             = _mm256_add_ps(fjx3,tx);
 422             fjy3             = _mm256_add_ps(fjy3,ty);
 423             fjz3             = _mm256_add_ps(fjz3,tz);
 424
 425             /**************************
 426              * CALCULATE INTERACTIONS *
 427              **************************/
 428
 429             r21              = _mm256_mul_ps(rsq21,rinv21);
 430
 431             /* EWALD ELECTROSTATICS */
 432
 433             /* Analytical PME correction */
 434             zeta2            = _mm256_mul_ps(beta2,rsq21);
 435             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 436             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 437             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 438             felec            = _mm256_mul_ps(qq21,felec);
 439             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 440             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 441             velec            = _mm256_sub_ps(rinv21,pmecorrV);
 442             velec            = _mm256_mul_ps(qq21,velec);
 443
 444             /* Update potential sum for this i atom from the interaction with this j atom. */
 445             velecsum         = _mm256_add_ps(velecsum,velec);
 446
 447             fscal            = felec;
 448
 449             /* Calculate temporary vectorial force */
 450             tx               = _mm256_mul_ps(fscal,dx21);
 451             ty               = _mm256_mul_ps(fscal,dy21);
 452             tz               = _mm256_mul_ps(fscal,dz21);
 453
 454             /* Update vectorial force */
 455             fix2             = _mm256_add_ps(fix2,tx);
 456             fiy2             = _mm256_add_ps(fiy2,ty);
 457             fiz2             = _mm256_add_ps(fiz2,tz);
 458
 459             fjx1             = _mm256_add_ps(fjx1,tx);
 460             fjy1             = _mm256_add_ps(fjy1,ty);
 461             fjz1             = _mm256_add_ps(fjz1,tz);
 462
 463             /**************************
 464              * CALCULATE INTERACTIONS *
 465              **************************/
 466
 467             r22              = _mm256_mul_ps(rsq22,rinv22);
 468
 469             /* EWALD ELECTROSTATICS */
 470
 471             /* Analytical PME correction */
 472             zeta2            = _mm256_mul_ps(beta2,rsq22);
 473             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 474             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 475             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 476             felec            = _mm256_mul_ps(qq22,felec);
 477             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 478             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 479             velec            = _mm256_sub_ps(rinv22,pmecorrV);
 480             velec            = _mm256_mul_ps(qq22,velec);
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velecsum         = _mm256_add_ps(velecsum,velec);
 484
 485             fscal            = felec;
 486
 487             /* Calculate temporary vectorial force */
 488             tx               = _mm256_mul_ps(fscal,dx22);
 489             ty               = _mm256_mul_ps(fscal,dy22);
 490             tz               = _mm256_mul_ps(fscal,dz22);
 491
 492             /* Update vectorial force */
 493             fix2             = _mm256_add_ps(fix2,tx);
 494             fiy2             = _mm256_add_ps(fiy2,ty);
 495             fiz2             = _mm256_add_ps(fiz2,tz);
 496
 497             fjx2             = _mm256_add_ps(fjx2,tx);
 498             fjy2             = _mm256_add_ps(fjy2,ty);
 499             fjz2             = _mm256_add_ps(fjz2,tz);
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             r23              = _mm256_mul_ps(rsq23,rinv23);
 506
 507             /* EWALD ELECTROSTATICS */
 508
 509             /* Analytical PME correction */
 510             zeta2            = _mm256_mul_ps(beta2,rsq23);
 511             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
 512             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 513             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 514             felec            = _mm256_mul_ps(qq23,felec);
 515             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 516             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 517             velec            = _mm256_sub_ps(rinv23,pmecorrV);
 518             velec            = _mm256_mul_ps(qq23,velec);
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velecsum         = _mm256_add_ps(velecsum,velec);
 522
 523             fscal            = felec;
 524
 525             /* Calculate temporary vectorial force */
 526             tx               = _mm256_mul_ps(fscal,dx23);
 527             ty               = _mm256_mul_ps(fscal,dy23);
 528             tz               = _mm256_mul_ps(fscal,dz23);
 529
 530             /* Update vectorial force */
 531             fix2             = _mm256_add_ps(fix2,tx);
 532             fiy2             = _mm256_add_ps(fiy2,ty);
 533             fiz2             = _mm256_add_ps(fiz2,tz);
 534
 535             fjx3             = _mm256_add_ps(fjx3,tx);
 536             fjy3             = _mm256_add_ps(fjy3,ty);
 537             fjz3             = _mm256_add_ps(fjz3,tz);
 538
 539             /**************************
 540              * CALCULATE INTERACTIONS *
 541              **************************/
 542
 543             r31              = _mm256_mul_ps(rsq31,rinv31);
 544
 545             /* EWALD ELECTROSTATICS */
 546
 547             /* Analytical PME correction */
 548             zeta2            = _mm256_mul_ps(beta2,rsq31);
 549             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
 550             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 551             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 552             felec            = _mm256_mul_ps(qq31,felec);
 553             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 554             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 555             velec            = _mm256_sub_ps(rinv31,pmecorrV);
 556             velec            = _mm256_mul_ps(qq31,velec);
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             velecsum         = _mm256_add_ps(velecsum,velec);
 560
 561             fscal            = felec;
 562
 563             /* Calculate temporary vectorial force */
 564             tx               = _mm256_mul_ps(fscal,dx31);
 565             ty               = _mm256_mul_ps(fscal,dy31);
 566             tz               = _mm256_mul_ps(fscal,dz31);
 567
 568             /* Update vectorial force */
 569             fix3             = _mm256_add_ps(fix3,tx);
 570             fiy3             = _mm256_add_ps(fiy3,ty);
 571             fiz3             = _mm256_add_ps(fiz3,tz);
 572
 573             fjx1             = _mm256_add_ps(fjx1,tx);
 574             fjy1             = _mm256_add_ps(fjy1,ty);
 575             fjz1             = _mm256_add_ps(fjz1,tz);
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             r32              = _mm256_mul_ps(rsq32,rinv32);
 582
 583             /* EWALD ELECTROSTATICS */
 584
 585             /* Analytical PME correction */
 586             zeta2            = _mm256_mul_ps(beta2,rsq32);
 587             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
 588             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 589             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 590             felec            = _mm256_mul_ps(qq32,felec);
 591             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 592             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 593             velec            = _mm256_sub_ps(rinv32,pmecorrV);
 594             velec            = _mm256_mul_ps(qq32,velec);
 595
 596             /* Update potential sum for this i atom from the interaction with this j atom. */
 597             velecsum         = _mm256_add_ps(velecsum,velec);
 598
 599             fscal            = felec;
 600
 601             /* Calculate temporary vectorial force */
 602             tx               = _mm256_mul_ps(fscal,dx32);
 603             ty               = _mm256_mul_ps(fscal,dy32);
 604             tz               = _mm256_mul_ps(fscal,dz32);
 605
 606             /* Update vectorial force */
 607             fix3             = _mm256_add_ps(fix3,tx);
 608             fiy3             = _mm256_add_ps(fiy3,ty);
 609             fiz3             = _mm256_add_ps(fiz3,tz);
 610
 611             fjx2             = _mm256_add_ps(fjx2,tx);
 612             fjy2             = _mm256_add_ps(fjy2,ty);
 613             fjz2             = _mm256_add_ps(fjz2,tz);
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             r33              = _mm256_mul_ps(rsq33,rinv33);
 620
 621             /* EWALD ELECTROSTATICS */
 622
 623             /* Analytical PME correction */
 624             zeta2            = _mm256_mul_ps(beta2,rsq33);
 625             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
 626             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 627             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 628             felec            = _mm256_mul_ps(qq33,felec);
 629             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 630             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 631             velec            = _mm256_sub_ps(rinv33,pmecorrV);
 632             velec            = _mm256_mul_ps(qq33,velec);
 633
 634             /* Update potential sum for this i atom from the interaction with this j atom. */
 635             velecsum         = _mm256_add_ps(velecsum,velec);
 636
 637             fscal            = felec;
 638
 639             /* Calculate temporary vectorial force */
 640             tx               = _mm256_mul_ps(fscal,dx33);
 641             ty               = _mm256_mul_ps(fscal,dy33);
 642             tz               = _mm256_mul_ps(fscal,dz33);
 643
 644             /* Update vectorial force */
 645             fix3             = _mm256_add_ps(fix3,tx);
 646             fiy3             = _mm256_add_ps(fiy3,ty);
 647             fiz3             = _mm256_add_ps(fiz3,tz);
 648
 649             fjx3             = _mm256_add_ps(fjx3,tx);
 650             fjy3             = _mm256_add_ps(fjy3,ty);
 651             fjz3             = _mm256_add_ps(fjz3,tz);
 652
 653             fjptrA             = f+j_coord_offsetA;
 654             fjptrB             = f+j_coord_offsetB;
 655             fjptrC             = f+j_coord_offsetC;
 656             fjptrD             = f+j_coord_offsetD;
 657             fjptrE             = f+j_coord_offsetE;
 658             fjptrF             = f+j_coord_offsetF;
 659             fjptrG             = f+j_coord_offsetG;
 660             fjptrH             = f+j_coord_offsetH;
 661
 662             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 663                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
 664                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 665
 666             /* Inner loop uses 756 flops */
 667         }
 668
 669         if(jidx<j_index_end)
 670         {
 671
 672             /* Get j neighbor index, and coordinate index */
 673             jnrlistA         = jjnr[jidx];
 674             jnrlistB         = jjnr[jidx+1];
 675             jnrlistC         = jjnr[jidx+2];
 676             jnrlistD         = jjnr[jidx+3];
 677             jnrlistE         = jjnr[jidx+4];
 678             jnrlistF         = jjnr[jidx+5];
 679             jnrlistG         = jjnr[jidx+6];
 680             jnrlistH         = jjnr[jidx+7];
 681             /* Sign of each element will be negative for non-real atoms.
 682              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 683              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 684              */
 685             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 686                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 687
 688             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 689             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 690             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 691             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 692             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 693             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 694             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 695             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 696             j_coord_offsetA  = DIM*jnrA;
 697             j_coord_offsetB  = DIM*jnrB;
 698             j_coord_offsetC  = DIM*jnrC;
 699             j_coord_offsetD  = DIM*jnrD;
 700             j_coord_offsetE  = DIM*jnrE;
 701             j_coord_offsetF  = DIM*jnrF;
 702             j_coord_offsetG  = DIM*jnrG;
 703             j_coord_offsetH  = DIM*jnrH;
 704
 705             /* load j atom coordinates */
 706             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 707                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 708                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
 709                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
 710                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 711
 712             /* Calculate displacement vector */
 713             dx11             = _mm256_sub_ps(ix1,jx1);
 714             dy11             = _mm256_sub_ps(iy1,jy1);
 715             dz11             = _mm256_sub_ps(iz1,jz1);
 716             dx12             = _mm256_sub_ps(ix1,jx2);
 717             dy12             = _mm256_sub_ps(iy1,jy2);
 718             dz12             = _mm256_sub_ps(iz1,jz2);
 719             dx13             = _mm256_sub_ps(ix1,jx3);
 720             dy13             = _mm256_sub_ps(iy1,jy3);
 721             dz13             = _mm256_sub_ps(iz1,jz3);
 722             dx21             = _mm256_sub_ps(ix2,jx1);
 723             dy21             = _mm256_sub_ps(iy2,jy1);
 724             dz21             = _mm256_sub_ps(iz2,jz1);
 725             dx22             = _mm256_sub_ps(ix2,jx2);
 726             dy22             = _mm256_sub_ps(iy2,jy2);
 727             dz22             = _mm256_sub_ps(iz2,jz2);
 728             dx23             = _mm256_sub_ps(ix2,jx3);
 729             dy23             = _mm256_sub_ps(iy2,jy3);
 730             dz23             = _mm256_sub_ps(iz2,jz3);
 731             dx31             = _mm256_sub_ps(ix3,jx1);
 732             dy31             = _mm256_sub_ps(iy3,jy1);
 733             dz31             = _mm256_sub_ps(iz3,jz1);
 734             dx32             = _mm256_sub_ps(ix3,jx2);
 735             dy32             = _mm256_sub_ps(iy3,jy2);
 736             dz32             = _mm256_sub_ps(iz3,jz2);
 737             dx33             = _mm256_sub_ps(ix3,jx3);
 738             dy33             = _mm256_sub_ps(iy3,jy3);
 739             dz33             = _mm256_sub_ps(iz3,jz3);
 740
 741             /* Calculate squared distance and things based on it */
 742             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 743             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 744             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 745             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 746             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 747             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 748             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 749             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 750             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 751
 752             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 753             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 754             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 755             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 756             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 757             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 758             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 759             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 760             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 761
 762             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 763             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 764             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 765             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 766             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 767             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 768             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 769             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 770             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 771
 772             fjx1             = _mm256_setzero_ps();
 773             fjy1             = _mm256_setzero_ps();
 774             fjz1             = _mm256_setzero_ps();
 775             fjx2             = _mm256_setzero_ps();
 776             fjy2             = _mm256_setzero_ps();
 777             fjz2             = _mm256_setzero_ps();
 778             fjx3             = _mm256_setzero_ps();
 779             fjy3             = _mm256_setzero_ps();
 780             fjz3             = _mm256_setzero_ps();
 781
 782             /**************************
 783              * CALCULATE INTERACTIONS *
 784              **************************/
 785
 786             r11              = _mm256_mul_ps(rsq11,rinv11);
 787             r11              = _mm256_andnot_ps(dummy_mask,r11);
 788
 789             /* EWALD ELECTROSTATICS */
 790
 791             /* Analytical PME correction */
 792             zeta2            = _mm256_mul_ps(beta2,rsq11);
 793             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 794             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 795             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 796             felec            = _mm256_mul_ps(qq11,felec);
 797             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 798             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 799             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 800             velec            = _mm256_mul_ps(qq11,velec);
 801
 802             /* Update potential sum for this i atom from the interaction with this j atom. */
 803             velec            = _mm256_andnot_ps(dummy_mask,velec);
 804             velecsum         = _mm256_add_ps(velecsum,velec);
 805
 806             fscal            = felec;
 807
 808             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 809
 810             /* Calculate temporary vectorial force */
 811             tx               = _mm256_mul_ps(fscal,dx11);
 812             ty               = _mm256_mul_ps(fscal,dy11);
 813             tz               = _mm256_mul_ps(fscal,dz11);
 814
 815             /* Update vectorial force */
 816             fix1             = _mm256_add_ps(fix1,tx);
 817             fiy1             = _mm256_add_ps(fiy1,ty);
 818             fiz1             = _mm256_add_ps(fiz1,tz);
 819
 820             fjx1             = _mm256_add_ps(fjx1,tx);
 821             fjy1             = _mm256_add_ps(fjy1,ty);
 822             fjz1             = _mm256_add_ps(fjz1,tz);
 823
 824             /**************************
 825              * CALCULATE INTERACTIONS *
 826              **************************/
 827
 828             r12              = _mm256_mul_ps(rsq12,rinv12);
 829             r12              = _mm256_andnot_ps(dummy_mask,r12);
 830
 831             /* EWALD ELECTROSTATICS */
 832
 833             /* Analytical PME correction */
 834             zeta2            = _mm256_mul_ps(beta2,rsq12);
 835             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 836             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 837             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 838             felec            = _mm256_mul_ps(qq12,felec);
 839             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 840             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 841             velec            = _mm256_sub_ps(rinv12,pmecorrV);
 842             velec            = _mm256_mul_ps(qq12,velec);
 843
 844             /* Update potential sum for this i atom from the interaction with this j atom. */
 845             velec            = _mm256_andnot_ps(dummy_mask,velec);
 846             velecsum         = _mm256_add_ps(velecsum,velec);
 847
 848             fscal            = felec;
 849
 850             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 851
 852             /* Calculate temporary vectorial force */
 853             tx               = _mm256_mul_ps(fscal,dx12);
 854             ty               = _mm256_mul_ps(fscal,dy12);
 855             tz               = _mm256_mul_ps(fscal,dz12);
 856
 857             /* Update vectorial force */
 858             fix1             = _mm256_add_ps(fix1,tx);
 859             fiy1             = _mm256_add_ps(fiy1,ty);
 860             fiz1             = _mm256_add_ps(fiz1,tz);
 861
 862             fjx2             = _mm256_add_ps(fjx2,tx);
 863             fjy2             = _mm256_add_ps(fjy2,ty);
 864             fjz2             = _mm256_add_ps(fjz2,tz);
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             r13              = _mm256_mul_ps(rsq13,rinv13);
 871             r13              = _mm256_andnot_ps(dummy_mask,r13);
 872
 873             /* EWALD ELECTROSTATICS */
 874
 875             /* Analytical PME correction */
 876             zeta2            = _mm256_mul_ps(beta2,rsq13);
 877             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
 878             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 879             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 880             felec            = _mm256_mul_ps(qq13,felec);
 881             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 882             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 883             velec            = _mm256_sub_ps(rinv13,pmecorrV);
 884             velec            = _mm256_mul_ps(qq13,velec);
 885
 886             /* Update potential sum for this i atom from the interaction with this j atom. */
 887             velec            = _mm256_andnot_ps(dummy_mask,velec);
 888             velecsum         = _mm256_add_ps(velecsum,velec);
 889
 890             fscal            = felec;
 891
 892             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 893
 894             /* Calculate temporary vectorial force */
 895             tx               = _mm256_mul_ps(fscal,dx13);
 896             ty               = _mm256_mul_ps(fscal,dy13);
 897             tz               = _mm256_mul_ps(fscal,dz13);
 898
 899             /* Update vectorial force */
 900             fix1             = _mm256_add_ps(fix1,tx);
 901             fiy1             = _mm256_add_ps(fiy1,ty);
 902             fiz1             = _mm256_add_ps(fiz1,tz);
 903
 904             fjx3             = _mm256_add_ps(fjx3,tx);
 905             fjy3             = _mm256_add_ps(fjy3,ty);
 906             fjz3             = _mm256_add_ps(fjz3,tz);
 907
 908             /**************************
 909              * CALCULATE INTERACTIONS *
 910              **************************/
 911
 912             r21              = _mm256_mul_ps(rsq21,rinv21);
 913             r21              = _mm256_andnot_ps(dummy_mask,r21);
 914
 915             /* EWALD ELECTROSTATICS */
 916
 917             /* Analytical PME correction */
 918             zeta2            = _mm256_mul_ps(beta2,rsq21);
 919             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 920             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 921             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 922             felec            = _mm256_mul_ps(qq21,felec);
 923             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 924             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 925             velec            = _mm256_sub_ps(rinv21,pmecorrV);
 926             velec            = _mm256_mul_ps(qq21,velec);
 927
 928             /* Update potential sum for this i atom from the interaction with this j atom. */
 929             velec            = _mm256_andnot_ps(dummy_mask,velec);
 930             velecsum         = _mm256_add_ps(velecsum,velec);
 931
 932             fscal            = felec;
 933
 934             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 935
 936             /* Calculate temporary vectorial force */
 937             tx               = _mm256_mul_ps(fscal,dx21);
 938             ty               = _mm256_mul_ps(fscal,dy21);
 939             tz               = _mm256_mul_ps(fscal,dz21);
 940
 941             /* Update vectorial force */
 942             fix2             = _mm256_add_ps(fix2,tx);
 943             fiy2             = _mm256_add_ps(fiy2,ty);
 944             fiz2             = _mm256_add_ps(fiz2,tz);
 945
 946             fjx1             = _mm256_add_ps(fjx1,tx);
 947             fjy1             = _mm256_add_ps(fjy1,ty);
 948             fjz1             = _mm256_add_ps(fjz1,tz);
 949
 950             /**************************
 951              * CALCULATE INTERACTIONS *
 952              **************************/
 953
 954             r22              = _mm256_mul_ps(rsq22,rinv22);
 955             r22              = _mm256_andnot_ps(dummy_mask,r22);
 956
 957             /* EWALD ELECTROSTATICS */
 958
 959             /* Analytical PME correction */
 960             zeta2            = _mm256_mul_ps(beta2,rsq22);
 961             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 962             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 963             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 964             felec            = _mm256_mul_ps(qq22,felec);
 965             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 966             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 967             velec            = _mm256_sub_ps(rinv22,pmecorrV);
 968             velec            = _mm256_mul_ps(qq22,velec);
 969
 970             /* Update potential sum for this i atom from the interaction with this j atom. */
 971             velec            = _mm256_andnot_ps(dummy_mask,velec);
 972             velecsum         = _mm256_add_ps(velecsum,velec);
 973
 974             fscal            = felec;
 975
 976             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 977
 978             /* Calculate temporary vectorial force */
 979             tx               = _mm256_mul_ps(fscal,dx22);
 980             ty               = _mm256_mul_ps(fscal,dy22);
 981             tz               = _mm256_mul_ps(fscal,dz22);
 982
 983             /* Update vectorial force */
 984             fix2             = _mm256_add_ps(fix2,tx);
 985             fiy2             = _mm256_add_ps(fiy2,ty);
 986             fiz2             = _mm256_add_ps(fiz2,tz);
 987
 988             fjx2             = _mm256_add_ps(fjx2,tx);
 989             fjy2             = _mm256_add_ps(fjy2,ty);
 990             fjz2             = _mm256_add_ps(fjz2,tz);
 991
 992             /**************************
 993              * CALCULATE INTERACTIONS *
 994              **************************/
 995
 996             r23              = _mm256_mul_ps(rsq23,rinv23);
 997             r23              = _mm256_andnot_ps(dummy_mask,r23);
 998
 999             /* EWALD ELECTROSTATICS */
1000
1001             /* Analytical PME correction */
1002             zeta2            = _mm256_mul_ps(beta2,rsq23);
1003             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1004             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1005             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1006             felec            = _mm256_mul_ps(qq23,felec);
1007             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1008             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1009             velec            = _mm256_sub_ps(rinv23,pmecorrV);
1010             velec            = _mm256_mul_ps(qq23,velec);
1011
1012             /* Update potential sum for this i atom from the interaction with this j atom. */
1013             velec            = _mm256_andnot_ps(dummy_mask,velec);
1014             velecsum         = _mm256_add_ps(velecsum,velec);
1015
1016             fscal            = felec;
1017
1018             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1019
1020             /* Calculate temporary vectorial force */
1021             tx               = _mm256_mul_ps(fscal,dx23);
1022             ty               = _mm256_mul_ps(fscal,dy23);
1023             tz               = _mm256_mul_ps(fscal,dz23);
1024
1025             /* Update vectorial force */
1026             fix2             = _mm256_add_ps(fix2,tx);
1027             fiy2             = _mm256_add_ps(fiy2,ty);
1028             fiz2             = _mm256_add_ps(fiz2,tz);
1029
1030             fjx3             = _mm256_add_ps(fjx3,tx);
1031             fjy3             = _mm256_add_ps(fjy3,ty);
1032             fjz3             = _mm256_add_ps(fjz3,tz);
1033
1034             /**************************
1035              * CALCULATE INTERACTIONS *
1036              **************************/
1037
1038             r31              = _mm256_mul_ps(rsq31,rinv31);
1039             r31              = _mm256_andnot_ps(dummy_mask,r31);
1040
1041             /* EWALD ELECTROSTATICS */
1042
1043             /* Analytical PME correction */
1044             zeta2            = _mm256_mul_ps(beta2,rsq31);
1045             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1046             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1047             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1048             felec            = _mm256_mul_ps(qq31,felec);
1049             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1050             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1051             velec            = _mm256_sub_ps(rinv31,pmecorrV);
1052             velec            = _mm256_mul_ps(qq31,velec);
1053
1054             /* Update potential sum for this i atom from the interaction with this j atom. */
1055             velec            = _mm256_andnot_ps(dummy_mask,velec);
1056             velecsum         = _mm256_add_ps(velecsum,velec);
1057
1058             fscal            = felec;
1059
1060             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1061
1062             /* Calculate temporary vectorial force */
1063             tx               = _mm256_mul_ps(fscal,dx31);
1064             ty               = _mm256_mul_ps(fscal,dy31);
1065             tz               = _mm256_mul_ps(fscal,dz31);
1066
1067             /* Update vectorial force */
1068             fix3             = _mm256_add_ps(fix3,tx);
1069             fiy3             = _mm256_add_ps(fiy3,ty);
1070             fiz3             = _mm256_add_ps(fiz3,tz);
1071
1072             fjx1             = _mm256_add_ps(fjx1,tx);
1073             fjy1             = _mm256_add_ps(fjy1,ty);
1074             fjz1             = _mm256_add_ps(fjz1,tz);
1075
1076             /**************************
1077              * CALCULATE INTERACTIONS *
1078              **************************/
1079
1080             r32              = _mm256_mul_ps(rsq32,rinv32);
1081             r32              = _mm256_andnot_ps(dummy_mask,r32);
1082
1083             /* EWALD ELECTROSTATICS */
1084
1085             /* Analytical PME correction */
1086             zeta2            = _mm256_mul_ps(beta2,rsq32);
1087             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1088             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1089             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1090             felec            = _mm256_mul_ps(qq32,felec);
1091             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1092             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1093             velec            = _mm256_sub_ps(rinv32,pmecorrV);
1094             velec            = _mm256_mul_ps(qq32,velec);
1095
1096             /* Update potential sum for this i atom from the interaction with this j atom. */
1097             velec            = _mm256_andnot_ps(dummy_mask,velec);
1098             velecsum         = _mm256_add_ps(velecsum,velec);
1099
1100             fscal            = felec;
1101
1102             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1103
1104             /* Calculate temporary vectorial force */
1105             tx               = _mm256_mul_ps(fscal,dx32);
1106             ty               = _mm256_mul_ps(fscal,dy32);
1107             tz               = _mm256_mul_ps(fscal,dz32);
1108
1109             /* Update vectorial force */
1110             fix3             = _mm256_add_ps(fix3,tx);
1111             fiy3             = _mm256_add_ps(fiy3,ty);
1112             fiz3             = _mm256_add_ps(fiz3,tz);
1113
1114             fjx2             = _mm256_add_ps(fjx2,tx);
1115             fjy2             = _mm256_add_ps(fjy2,ty);
1116             fjz2             = _mm256_add_ps(fjz2,tz);
1117
1118             /**************************
1119              * CALCULATE INTERACTIONS *
1120              **************************/
1121
1122             r33              = _mm256_mul_ps(rsq33,rinv33);
1123             r33              = _mm256_andnot_ps(dummy_mask,r33);
1124
1125             /* EWALD ELECTROSTATICS */
1126
1127             /* Analytical PME correction */
1128             zeta2            = _mm256_mul_ps(beta2,rsq33);
1129             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
1130             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1131             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1132             felec            = _mm256_mul_ps(qq33,felec);
1133             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1134             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1135             velec            = _mm256_sub_ps(rinv33,pmecorrV);
1136             velec            = _mm256_mul_ps(qq33,velec);
1137
1138             /* Update potential sum for this i atom from the interaction with this j atom. */
1139             velec            = _mm256_andnot_ps(dummy_mask,velec);
1140             velecsum         = _mm256_add_ps(velecsum,velec);
1141
1142             fscal            = felec;
1143
1144             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1145
1146             /* Calculate temporary vectorial force */
1147             tx               = _mm256_mul_ps(fscal,dx33);
1148             ty               = _mm256_mul_ps(fscal,dy33);
1149             tz               = _mm256_mul_ps(fscal,dz33);
1150
1151             /* Update vectorial force */
1152             fix3             = _mm256_add_ps(fix3,tx);
1153             fiy3             = _mm256_add_ps(fiy3,ty);
1154             fiz3             = _mm256_add_ps(fiz3,tz);
1155
1156             fjx3             = _mm256_add_ps(fjx3,tx);
1157             fjy3             = _mm256_add_ps(fjy3,ty);
1158             fjz3             = _mm256_add_ps(fjz3,tz);
1159
1160             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1161             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1162             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1163             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1164             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1165             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1166             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1167             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1168
1169             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1170                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
1171                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1172
1173             /* Inner loop uses 765 flops */
1174         }
1175
1176         /* End of innermost loop */
1177
1178         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1179                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1180
1181         ggid                        = gid[iidx];
1182         /* Update potential energies */
1183         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1184
1185         /* Increment number of inner iterations */
1186         inneriter                  += j_index_end - j_index_start;
1187
1188         /* Outer loop uses 19 flops */
1189     }
1190
1191     /* Increment number of outer iterations */
1192     outeriter        += nri;
1193
1194     /* Update outer/inner flops */
1195
1196     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*765);
1197 }
1198 /*
1199  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4W4_F_avx_256_single
1200  * Electrostatics interaction: Ewald
1201  * VdW interaction:            None
1202  * Geometry:                   Water4-Water4
1203  * Calculate force/pot:        Force
1204  */
1205 void
1206 nb_kernel_ElecEw_VdwNone_GeomW4W4_F_avx_256_single
1207                     (t_nblist                    * gmx_restrict       nlist,
1208                      rvec                        * gmx_restrict          xx,
1209                      rvec                        * gmx_restrict          ff,
1210                      t_forcerec                  * gmx_restrict          fr,
1211                      t_mdatoms                   * gmx_restrict     mdatoms,
1212                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1213                      t_nrnb                      * gmx_restrict        nrnb)
1214 {
1215     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1216      * just 0 for non-waters.
1217      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1218      * jnr indices corresponding to data put in the four positions in the SIMD register.
1219      */
1220     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1221     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1222     int              jnrA,jnrB,jnrC,jnrD;
1223     int              jnrE,jnrF,jnrG,jnrH;
1224     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1225     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1226     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1227     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1228     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1229     real             rcutoff_scalar;
1230     real             *shiftvec,*fshift,*x,*f;
1231     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1232     real             scratch[4*DIM];
1233     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1234     real *           vdwioffsetptr1;
1235     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1236     real *           vdwioffsetptr2;
1237     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1238     real *           vdwioffsetptr3;
1239     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1240     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1241     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1242     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1243     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1244     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1245     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1246     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1247     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1248     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1249     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1250     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1251     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1252     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1253     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1254     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1255     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1256     real             *charge;
1257     __m256i          ewitab;
1258     __m128i          ewitab_lo,ewitab_hi;
1259     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1260     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1261     real             *ewtab;
1262     __m256           dummy_mask,cutoff_mask;
1263     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1264     __m256           one     = _mm256_set1_ps(1.0);
1265     __m256           two     = _mm256_set1_ps(2.0);
1266     x                = xx[0];
1267     f                = ff[0];
1268
1269     nri              = nlist->nri;
1270     iinr             = nlist->iinr;
1271     jindex           = nlist->jindex;
1272     jjnr             = nlist->jjnr;
1273     shiftidx         = nlist->shift;
1274     gid              = nlist->gid;
1275     shiftvec         = fr->shift_vec[0];
1276     fshift           = fr->fshift[0];
1277     facel            = _mm256_set1_ps(fr->epsfac);
1278     charge           = mdatoms->chargeA;
1279
1280     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1281     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
1282     beta2            = _mm256_mul_ps(beta,beta);
1283     beta3            = _mm256_mul_ps(beta,beta2);
1284
1285     ewtab            = fr->ic->tabq_coul_F;
1286     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1287     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1288
1289     /* Setup water-specific parameters */
1290     inr              = nlist->iinr[0];
1291     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1292     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1293     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1294
1295     jq1              = _mm256_set1_ps(charge[inr+1]);
1296     jq2              = _mm256_set1_ps(charge[inr+2]);
1297     jq3              = _mm256_set1_ps(charge[inr+3]);
1298     qq11             = _mm256_mul_ps(iq1,jq1);
1299     qq12             = _mm256_mul_ps(iq1,jq2);
1300     qq13             = _mm256_mul_ps(iq1,jq3);
1301     qq21             = _mm256_mul_ps(iq2,jq1);
1302     qq22             = _mm256_mul_ps(iq2,jq2);
1303     qq23             = _mm256_mul_ps(iq2,jq3);
1304     qq31             = _mm256_mul_ps(iq3,jq1);
1305     qq32             = _mm256_mul_ps(iq3,jq2);
1306     qq33             = _mm256_mul_ps(iq3,jq3);
1307
1308     /* Avoid stupid compiler warnings */
1309     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1310     j_coord_offsetA = 0;
1311     j_coord_offsetB = 0;
1312     j_coord_offsetC = 0;
1313     j_coord_offsetD = 0;
1314     j_coord_offsetE = 0;
1315     j_coord_offsetF = 0;
1316     j_coord_offsetG = 0;
1317     j_coord_offsetH = 0;
1318
1319     outeriter        = 0;
1320     inneriter        = 0;
1321
1322     for(iidx=0;iidx<4*DIM;iidx++)
1323     {
1324         scratch[iidx] = 0.0;
1325     }
1326
1327     /* Start outer loop over neighborlists */
1328     for(iidx=0; iidx<nri; iidx++)
1329     {
1330         /* Load shift vector for this list */
1331         i_shift_offset   = DIM*shiftidx[iidx];
1332
1333         /* Load limits for loop over neighbors */
1334         j_index_start    = jindex[iidx];
1335         j_index_end      = jindex[iidx+1];
1336
1337         /* Get outer coordinate index */
1338         inr              = iinr[iidx];
1339         i_coord_offset   = DIM*inr;
1340
1341         /* Load i particle coords and add shift vector */
1342         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1343                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1344
1345         fix1             = _mm256_setzero_ps();
1346         fiy1             = _mm256_setzero_ps();
1347         fiz1             = _mm256_setzero_ps();
1348         fix2             = _mm256_setzero_ps();
1349         fiy2             = _mm256_setzero_ps();
1350         fiz2             = _mm256_setzero_ps();
1351         fix3             = _mm256_setzero_ps();
1352         fiy3             = _mm256_setzero_ps();
1353         fiz3             = _mm256_setzero_ps();
1354
1355         /* Start inner kernel loop */
1356         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1357         {
1358
1359             /* Get j neighbor index, and coordinate index */
1360             jnrA             = jjnr[jidx];
1361             jnrB             = jjnr[jidx+1];
1362             jnrC             = jjnr[jidx+2];
1363             jnrD             = jjnr[jidx+3];
1364             jnrE             = jjnr[jidx+4];
1365             jnrF             = jjnr[jidx+5];
1366             jnrG             = jjnr[jidx+6];
1367             jnrH             = jjnr[jidx+7];
1368             j_coord_offsetA  = DIM*jnrA;
1369             j_coord_offsetB  = DIM*jnrB;
1370             j_coord_offsetC  = DIM*jnrC;
1371             j_coord_offsetD  = DIM*jnrD;
1372             j_coord_offsetE  = DIM*jnrE;
1373             j_coord_offsetF  = DIM*jnrF;
1374             j_coord_offsetG  = DIM*jnrG;
1375             j_coord_offsetH  = DIM*jnrH;
1376
1377             /* load j atom coordinates */
1378             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1379                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1380                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
1381                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
1382                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1383
1384             /* Calculate displacement vector */
1385             dx11             = _mm256_sub_ps(ix1,jx1);
1386             dy11             = _mm256_sub_ps(iy1,jy1);
1387             dz11             = _mm256_sub_ps(iz1,jz1);
1388             dx12             = _mm256_sub_ps(ix1,jx2);
1389             dy12             = _mm256_sub_ps(iy1,jy2);
1390             dz12             = _mm256_sub_ps(iz1,jz2);
1391             dx13             = _mm256_sub_ps(ix1,jx3);
1392             dy13             = _mm256_sub_ps(iy1,jy3);
1393             dz13             = _mm256_sub_ps(iz1,jz3);
1394             dx21             = _mm256_sub_ps(ix2,jx1);
1395             dy21             = _mm256_sub_ps(iy2,jy1);
1396             dz21             = _mm256_sub_ps(iz2,jz1);
1397             dx22             = _mm256_sub_ps(ix2,jx2);
1398             dy22             = _mm256_sub_ps(iy2,jy2);
1399             dz22             = _mm256_sub_ps(iz2,jz2);
1400             dx23             = _mm256_sub_ps(ix2,jx3);
1401             dy23             = _mm256_sub_ps(iy2,jy3);
1402             dz23             = _mm256_sub_ps(iz2,jz3);
1403             dx31             = _mm256_sub_ps(ix3,jx1);
1404             dy31             = _mm256_sub_ps(iy3,jy1);
1405             dz31             = _mm256_sub_ps(iz3,jz1);
1406             dx32             = _mm256_sub_ps(ix3,jx2);
1407             dy32             = _mm256_sub_ps(iy3,jy2);
1408             dz32             = _mm256_sub_ps(iz3,jz2);
1409             dx33             = _mm256_sub_ps(ix3,jx3);
1410             dy33             = _mm256_sub_ps(iy3,jy3);
1411             dz33             = _mm256_sub_ps(iz3,jz3);
1412
1413             /* Calculate squared distance and things based on it */
1414             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1415             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1416             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1417             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1418             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1419             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1420             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1421             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1422             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1423
1424             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1425             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1426             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
1427             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1428             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1429             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
1430             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
1431             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
1432             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
1433
1434             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1435             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1436             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1437             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1438             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1439             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1440             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1441             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1442             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1443
1444             fjx1             = _mm256_setzero_ps();
1445             fjy1             = _mm256_setzero_ps();
1446             fjz1             = _mm256_setzero_ps();
1447             fjx2             = _mm256_setzero_ps();
1448             fjy2             = _mm256_setzero_ps();
1449             fjz2             = _mm256_setzero_ps();
1450             fjx3             = _mm256_setzero_ps();
1451             fjy3             = _mm256_setzero_ps();
1452             fjz3             = _mm256_setzero_ps();
1453
1454             /**************************
1455              * CALCULATE INTERACTIONS *
1456              **************************/
1457
1458             r11              = _mm256_mul_ps(rsq11,rinv11);
1459
1460             /* EWALD ELECTROSTATICS */
1461
1462             /* Analytical PME correction */
1463             zeta2            = _mm256_mul_ps(beta2,rsq11);
1464             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1465             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1466             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1467             felec            = _mm256_mul_ps(qq11,felec);
1468
1469             fscal            = felec;
1470
1471             /* Calculate temporary vectorial force */
1472             tx               = _mm256_mul_ps(fscal,dx11);
1473             ty               = _mm256_mul_ps(fscal,dy11);
1474             tz               = _mm256_mul_ps(fscal,dz11);
1475
1476             /* Update vectorial force */
1477             fix1             = _mm256_add_ps(fix1,tx);
1478             fiy1             = _mm256_add_ps(fiy1,ty);
1479             fiz1             = _mm256_add_ps(fiz1,tz);
1480
1481             fjx1             = _mm256_add_ps(fjx1,tx);
1482             fjy1             = _mm256_add_ps(fjy1,ty);
1483             fjz1             = _mm256_add_ps(fjz1,tz);
1484
1485             /**************************
1486              * CALCULATE INTERACTIONS *
1487              **************************/
1488
1489             r12              = _mm256_mul_ps(rsq12,rinv12);
1490
1491             /* EWALD ELECTROSTATICS */
1492
1493             /* Analytical PME correction */
1494             zeta2            = _mm256_mul_ps(beta2,rsq12);
1495             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1496             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1497             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1498             felec            = _mm256_mul_ps(qq12,felec);
1499
1500             fscal            = felec;
1501
1502             /* Calculate temporary vectorial force */
1503             tx               = _mm256_mul_ps(fscal,dx12);
1504             ty               = _mm256_mul_ps(fscal,dy12);
1505             tz               = _mm256_mul_ps(fscal,dz12);
1506
1507             /* Update vectorial force */
1508             fix1             = _mm256_add_ps(fix1,tx);
1509             fiy1             = _mm256_add_ps(fiy1,ty);
1510             fiz1             = _mm256_add_ps(fiz1,tz);
1511
1512             fjx2             = _mm256_add_ps(fjx2,tx);
1513             fjy2             = _mm256_add_ps(fjy2,ty);
1514             fjz2             = _mm256_add_ps(fjz2,tz);
1515
1516             /**************************
1517              * CALCULATE INTERACTIONS *
1518              **************************/
1519
1520             r13              = _mm256_mul_ps(rsq13,rinv13);
1521
1522             /* EWALD ELECTROSTATICS */
1523
1524             /* Analytical PME correction */
1525             zeta2            = _mm256_mul_ps(beta2,rsq13);
1526             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
1527             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1528             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1529             felec            = _mm256_mul_ps(qq13,felec);
1530
1531             fscal            = felec;
1532
1533             /* Calculate temporary vectorial force */
1534             tx               = _mm256_mul_ps(fscal,dx13);
1535             ty               = _mm256_mul_ps(fscal,dy13);
1536             tz               = _mm256_mul_ps(fscal,dz13);
1537
1538             /* Update vectorial force */
1539             fix1             = _mm256_add_ps(fix1,tx);
1540             fiy1             = _mm256_add_ps(fiy1,ty);
1541             fiz1             = _mm256_add_ps(fiz1,tz);
1542
1543             fjx3             = _mm256_add_ps(fjx3,tx);
1544             fjy3             = _mm256_add_ps(fjy3,ty);
1545             fjz3             = _mm256_add_ps(fjz3,tz);
1546
1547             /**************************
1548              * CALCULATE INTERACTIONS *
1549              **************************/
1550
1551             r21              = _mm256_mul_ps(rsq21,rinv21);
1552
1553             /* EWALD ELECTROSTATICS */
1554
1555             /* Analytical PME correction */
1556             zeta2            = _mm256_mul_ps(beta2,rsq21);
1557             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1558             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1559             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1560             felec            = _mm256_mul_ps(qq21,felec);
1561
1562             fscal            = felec;
1563
1564             /* Calculate temporary vectorial force */
1565             tx               = _mm256_mul_ps(fscal,dx21);
1566             ty               = _mm256_mul_ps(fscal,dy21);
1567             tz               = _mm256_mul_ps(fscal,dz21);
1568
1569             /* Update vectorial force */
1570             fix2             = _mm256_add_ps(fix2,tx);
1571             fiy2             = _mm256_add_ps(fiy2,ty);
1572             fiz2             = _mm256_add_ps(fiz2,tz);
1573
1574             fjx1             = _mm256_add_ps(fjx1,tx);
1575             fjy1             = _mm256_add_ps(fjy1,ty);
1576             fjz1             = _mm256_add_ps(fjz1,tz);
1577
1578             /**************************
1579              * CALCULATE INTERACTIONS *
1580              **************************/
1581
1582             r22              = _mm256_mul_ps(rsq22,rinv22);
1583
1584             /* EWALD ELECTROSTATICS */
1585
1586             /* Analytical PME correction */
1587             zeta2            = _mm256_mul_ps(beta2,rsq22);
1588             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1589             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1590             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1591             felec            = _mm256_mul_ps(qq22,felec);
1592
1593             fscal            = felec;
1594
1595             /* Calculate temporary vectorial force */
1596             tx               = _mm256_mul_ps(fscal,dx22);
1597             ty               = _mm256_mul_ps(fscal,dy22);
1598             tz               = _mm256_mul_ps(fscal,dz22);
1599
1600             /* Update vectorial force */
1601             fix2             = _mm256_add_ps(fix2,tx);
1602             fiy2             = _mm256_add_ps(fiy2,ty);
1603             fiz2             = _mm256_add_ps(fiz2,tz);
1604
1605             fjx2             = _mm256_add_ps(fjx2,tx);
1606             fjy2             = _mm256_add_ps(fjy2,ty);
1607             fjz2             = _mm256_add_ps(fjz2,tz);
1608
1609             /**************************
1610              * CALCULATE INTERACTIONS *
1611              **************************/
1612
1613             r23              = _mm256_mul_ps(rsq23,rinv23);
1614
1615             /* EWALD ELECTROSTATICS */
1616
1617             /* Analytical PME correction */
1618             zeta2            = _mm256_mul_ps(beta2,rsq23);
1619             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1620             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1621             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1622             felec            = _mm256_mul_ps(qq23,felec);
1623
1624             fscal            = felec;
1625
1626             /* Calculate temporary vectorial force */
1627             tx               = _mm256_mul_ps(fscal,dx23);
1628             ty               = _mm256_mul_ps(fscal,dy23);
1629             tz               = _mm256_mul_ps(fscal,dz23);
1630
1631             /* Update vectorial force */
1632             fix2             = _mm256_add_ps(fix2,tx);
1633             fiy2             = _mm256_add_ps(fiy2,ty);
1634             fiz2             = _mm256_add_ps(fiz2,tz);
1635
1636             fjx3             = _mm256_add_ps(fjx3,tx);
1637             fjy3             = _mm256_add_ps(fjy3,ty);
1638             fjz3             = _mm256_add_ps(fjz3,tz);
1639
1640             /**************************
1641              * CALCULATE INTERACTIONS *
1642              **************************/
1643
1644             r31              = _mm256_mul_ps(rsq31,rinv31);
1645
1646             /* EWALD ELECTROSTATICS */
1647
1648             /* Analytical PME correction */
1649             zeta2            = _mm256_mul_ps(beta2,rsq31);
1650             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1651             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1652             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1653             felec            = _mm256_mul_ps(qq31,felec);
1654
1655             fscal            = felec;
1656
1657             /* Calculate temporary vectorial force */
1658             tx               = _mm256_mul_ps(fscal,dx31);
1659             ty               = _mm256_mul_ps(fscal,dy31);
1660             tz               = _mm256_mul_ps(fscal,dz31);
1661
1662             /* Update vectorial force */
1663             fix3             = _mm256_add_ps(fix3,tx);
1664             fiy3             = _mm256_add_ps(fiy3,ty);
1665             fiz3             = _mm256_add_ps(fiz3,tz);
1666
1667             fjx1             = _mm256_add_ps(fjx1,tx);
1668             fjy1             = _mm256_add_ps(fjy1,ty);
1669             fjz1             = _mm256_add_ps(fjz1,tz);
1670
1671             /**************************
1672              * CALCULATE INTERACTIONS *
1673              **************************/
1674
1675             r32              = _mm256_mul_ps(rsq32,rinv32);
1676
1677             /* EWALD ELECTROSTATICS */
1678
1679             /* Analytical PME correction */
1680             zeta2            = _mm256_mul_ps(beta2,rsq32);
1681             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1682             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1683             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1684             felec            = _mm256_mul_ps(qq32,felec);
1685
1686             fscal            = felec;
1687
1688             /* Calculate temporary vectorial force */
1689             tx               = _mm256_mul_ps(fscal,dx32);
1690             ty               = _mm256_mul_ps(fscal,dy32);
1691             tz               = _mm256_mul_ps(fscal,dz32);
1692
1693             /* Update vectorial force */
1694             fix3             = _mm256_add_ps(fix3,tx);
1695             fiy3             = _mm256_add_ps(fiy3,ty);
1696             fiz3             = _mm256_add_ps(fiz3,tz);
1697
1698             fjx2             = _mm256_add_ps(fjx2,tx);
1699             fjy2             = _mm256_add_ps(fjy2,ty);
1700             fjz2             = _mm256_add_ps(fjz2,tz);
1701
1702             /**************************
1703              * CALCULATE INTERACTIONS *
1704              **************************/
1705
1706             r33              = _mm256_mul_ps(rsq33,rinv33);
1707
1708             /* EWALD ELECTROSTATICS */
1709
1710             /* Analytical PME correction */
1711             zeta2            = _mm256_mul_ps(beta2,rsq33);
1712             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
1713             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1714             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1715             felec            = _mm256_mul_ps(qq33,felec);
1716
1717             fscal            = felec;
1718
1719             /* Calculate temporary vectorial force */
1720             tx               = _mm256_mul_ps(fscal,dx33);
1721             ty               = _mm256_mul_ps(fscal,dy33);
1722             tz               = _mm256_mul_ps(fscal,dz33);
1723
1724             /* Update vectorial force */
1725             fix3             = _mm256_add_ps(fix3,tx);
1726             fiy3             = _mm256_add_ps(fiy3,ty);
1727             fiz3             = _mm256_add_ps(fiz3,tz);
1728
1729             fjx3             = _mm256_add_ps(fjx3,tx);
1730             fjy3             = _mm256_add_ps(fjy3,ty);
1731             fjz3             = _mm256_add_ps(fjz3,tz);
1732
1733             fjptrA             = f+j_coord_offsetA;
1734             fjptrB             = f+j_coord_offsetB;
1735             fjptrC             = f+j_coord_offsetC;
1736             fjptrD             = f+j_coord_offsetD;
1737             fjptrE             = f+j_coord_offsetE;
1738             fjptrF             = f+j_coord_offsetF;
1739             fjptrG             = f+j_coord_offsetG;
1740             fjptrH             = f+j_coord_offsetH;
1741
1742             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1743                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
1744                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1745
1746             /* Inner loop uses 504 flops */
1747         }
1748
1749         if(jidx<j_index_end)
1750         {
1751
1752             /* Get j neighbor index, and coordinate index */
1753             jnrlistA         = jjnr[jidx];
1754             jnrlistB         = jjnr[jidx+1];
1755             jnrlistC         = jjnr[jidx+2];
1756             jnrlistD         = jjnr[jidx+3];
1757             jnrlistE         = jjnr[jidx+4];
1758             jnrlistF         = jjnr[jidx+5];
1759             jnrlistG         = jjnr[jidx+6];
1760             jnrlistH         = jjnr[jidx+7];
1761             /* Sign of each element will be negative for non-real atoms.
1762              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1763              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1764              */
1765             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1766                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1767
1768             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1769             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1770             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1771             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1772             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1773             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1774             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1775             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1776             j_coord_offsetA  = DIM*jnrA;
1777             j_coord_offsetB  = DIM*jnrB;
1778             j_coord_offsetC  = DIM*jnrC;
1779             j_coord_offsetD  = DIM*jnrD;
1780             j_coord_offsetE  = DIM*jnrE;
1781             j_coord_offsetF  = DIM*jnrF;
1782             j_coord_offsetG  = DIM*jnrG;
1783             j_coord_offsetH  = DIM*jnrH;
1784
1785             /* load j atom coordinates */
1786             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1787                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1788                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
1789                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
1790                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1791
1792             /* Calculate displacement vector */
1793             dx11             = _mm256_sub_ps(ix1,jx1);
1794             dy11             = _mm256_sub_ps(iy1,jy1);
1795             dz11             = _mm256_sub_ps(iz1,jz1);
1796             dx12             = _mm256_sub_ps(ix1,jx2);
1797             dy12             = _mm256_sub_ps(iy1,jy2);
1798             dz12             = _mm256_sub_ps(iz1,jz2);
1799             dx13             = _mm256_sub_ps(ix1,jx3);
1800             dy13             = _mm256_sub_ps(iy1,jy3);
1801             dz13             = _mm256_sub_ps(iz1,jz3);
1802             dx21             = _mm256_sub_ps(ix2,jx1);
1803             dy21             = _mm256_sub_ps(iy2,jy1);
1804             dz21             = _mm256_sub_ps(iz2,jz1);
1805             dx22             = _mm256_sub_ps(ix2,jx2);
1806             dy22             = _mm256_sub_ps(iy2,jy2);
1807             dz22             = _mm256_sub_ps(iz2,jz2);
1808             dx23             = _mm256_sub_ps(ix2,jx3);
1809             dy23             = _mm256_sub_ps(iy2,jy3);
1810             dz23             = _mm256_sub_ps(iz2,jz3);
1811             dx31             = _mm256_sub_ps(ix3,jx1);
1812             dy31             = _mm256_sub_ps(iy3,jy1);
1813             dz31             = _mm256_sub_ps(iz3,jz1);
1814             dx32             = _mm256_sub_ps(ix3,jx2);
1815             dy32             = _mm256_sub_ps(iy3,jy2);
1816             dz32             = _mm256_sub_ps(iz3,jz2);
1817             dx33             = _mm256_sub_ps(ix3,jx3);
1818             dy33             = _mm256_sub_ps(iy3,jy3);
1819             dz33             = _mm256_sub_ps(iz3,jz3);
1820
1821             /* Calculate squared distance and things based on it */
1822             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1823             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1824             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1825             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1826             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1827             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1828             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1829             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1830             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1831
1832             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1833             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1834             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
1835             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1836             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1837             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
1838             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
1839             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
1840             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
1841
1842             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1843             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1844             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1845             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1846             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1847             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1848             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1849             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1850             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1851
1852             fjx1             = _mm256_setzero_ps();
1853             fjy1             = _mm256_setzero_ps();
1854             fjz1             = _mm256_setzero_ps();
1855             fjx2             = _mm256_setzero_ps();
1856             fjy2             = _mm256_setzero_ps();
1857             fjz2             = _mm256_setzero_ps();
1858             fjx3             = _mm256_setzero_ps();
1859             fjy3             = _mm256_setzero_ps();
1860             fjz3             = _mm256_setzero_ps();
1861
1862             /**************************
1863              * CALCULATE INTERACTIONS *
1864              **************************/
1865
1866             r11              = _mm256_mul_ps(rsq11,rinv11);
1867             r11              = _mm256_andnot_ps(dummy_mask,r11);
1868
1869             /* EWALD ELECTROSTATICS */
1870
1871             /* Analytical PME correction */
1872             zeta2            = _mm256_mul_ps(beta2,rsq11);
1873             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1874             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1875             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1876             felec            = _mm256_mul_ps(qq11,felec);
1877
1878             fscal            = felec;
1879
1880             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1881
1882             /* Calculate temporary vectorial force */
1883             tx               = _mm256_mul_ps(fscal,dx11);
1884             ty               = _mm256_mul_ps(fscal,dy11);
1885             tz               = _mm256_mul_ps(fscal,dz11);
1886
1887             /* Update vectorial force */
1888             fix1             = _mm256_add_ps(fix1,tx);
1889             fiy1             = _mm256_add_ps(fiy1,ty);
1890             fiz1             = _mm256_add_ps(fiz1,tz);
1891
1892             fjx1             = _mm256_add_ps(fjx1,tx);
1893             fjy1             = _mm256_add_ps(fjy1,ty);
1894             fjz1             = _mm256_add_ps(fjz1,tz);
1895
1896             /**************************
1897              * CALCULATE INTERACTIONS *
1898              **************************/
1899
1900             r12              = _mm256_mul_ps(rsq12,rinv12);
1901             r12              = _mm256_andnot_ps(dummy_mask,r12);
1902
1903             /* EWALD ELECTROSTATICS */
1904
1905             /* Analytical PME correction */
1906             zeta2            = _mm256_mul_ps(beta2,rsq12);
1907             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1908             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1909             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1910             felec            = _mm256_mul_ps(qq12,felec);
1911
1912             fscal            = felec;
1913
1914             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1915
1916             /* Calculate temporary vectorial force */
1917             tx               = _mm256_mul_ps(fscal,dx12);
1918             ty               = _mm256_mul_ps(fscal,dy12);
1919             tz               = _mm256_mul_ps(fscal,dz12);
1920
1921             /* Update vectorial force */
1922             fix1             = _mm256_add_ps(fix1,tx);
1923             fiy1             = _mm256_add_ps(fiy1,ty);
1924             fiz1             = _mm256_add_ps(fiz1,tz);
1925
1926             fjx2             = _mm256_add_ps(fjx2,tx);
1927             fjy2             = _mm256_add_ps(fjy2,ty);
1928             fjz2             = _mm256_add_ps(fjz2,tz);
1929
1930             /**************************
1931              * CALCULATE INTERACTIONS *
1932              **************************/
1933
1934             r13              = _mm256_mul_ps(rsq13,rinv13);
1935             r13              = _mm256_andnot_ps(dummy_mask,r13);
1936
1937             /* EWALD ELECTROSTATICS */
1938
1939             /* Analytical PME correction */
1940             zeta2            = _mm256_mul_ps(beta2,rsq13);
1941             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
1942             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1943             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1944             felec            = _mm256_mul_ps(qq13,felec);
1945
1946             fscal            = felec;
1947
1948             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1949
1950             /* Calculate temporary vectorial force */
1951             tx               = _mm256_mul_ps(fscal,dx13);
1952             ty               = _mm256_mul_ps(fscal,dy13);
1953             tz               = _mm256_mul_ps(fscal,dz13);
1954
1955             /* Update vectorial force */
1956             fix1             = _mm256_add_ps(fix1,tx);
1957             fiy1             = _mm256_add_ps(fiy1,ty);
1958             fiz1             = _mm256_add_ps(fiz1,tz);
1959
1960             fjx3             = _mm256_add_ps(fjx3,tx);
1961             fjy3             = _mm256_add_ps(fjy3,ty);
1962             fjz3             = _mm256_add_ps(fjz3,tz);
1963
1964             /**************************
1965              * CALCULATE INTERACTIONS *
1966              **************************/
1967
1968             r21              = _mm256_mul_ps(rsq21,rinv21);
1969             r21              = _mm256_andnot_ps(dummy_mask,r21);
1970
1971             /* EWALD ELECTROSTATICS */
1972
1973             /* Analytical PME correction */
1974             zeta2            = _mm256_mul_ps(beta2,rsq21);
1975             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1976             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1977             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1978             felec            = _mm256_mul_ps(qq21,felec);
1979
1980             fscal            = felec;
1981
1982             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1983
1984             /* Calculate temporary vectorial force */
1985             tx               = _mm256_mul_ps(fscal,dx21);
1986             ty               = _mm256_mul_ps(fscal,dy21);
1987             tz               = _mm256_mul_ps(fscal,dz21);
1988
1989             /* Update vectorial force */
1990             fix2             = _mm256_add_ps(fix2,tx);
1991             fiy2             = _mm256_add_ps(fiy2,ty);
1992             fiz2             = _mm256_add_ps(fiz2,tz);
1993
1994             fjx1             = _mm256_add_ps(fjx1,tx);
1995             fjy1             = _mm256_add_ps(fjy1,ty);
1996             fjz1             = _mm256_add_ps(fjz1,tz);
1997
1998             /**************************
1999              * CALCULATE INTERACTIONS *
2000              **************************/
2001
2002             r22              = _mm256_mul_ps(rsq22,rinv22);
2003             r22              = _mm256_andnot_ps(dummy_mask,r22);
2004
2005             /* EWALD ELECTROSTATICS */
2006
2007             /* Analytical PME correction */
2008             zeta2            = _mm256_mul_ps(beta2,rsq22);
2009             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2010             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2011             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2012             felec            = _mm256_mul_ps(qq22,felec);
2013
2014             fscal            = felec;
2015
2016             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2017
2018             /* Calculate temporary vectorial force */
2019             tx               = _mm256_mul_ps(fscal,dx22);
2020             ty               = _mm256_mul_ps(fscal,dy22);
2021             tz               = _mm256_mul_ps(fscal,dz22);
2022
2023             /* Update vectorial force */
2024             fix2             = _mm256_add_ps(fix2,tx);
2025             fiy2             = _mm256_add_ps(fiy2,ty);
2026             fiz2             = _mm256_add_ps(fiz2,tz);
2027
2028             fjx2             = _mm256_add_ps(fjx2,tx);
2029             fjy2             = _mm256_add_ps(fjy2,ty);
2030             fjz2             = _mm256_add_ps(fjz2,tz);
2031
2032             /**************************
2033              * CALCULATE INTERACTIONS *
2034              **************************/
2035
2036             r23              = _mm256_mul_ps(rsq23,rinv23);
2037             r23              = _mm256_andnot_ps(dummy_mask,r23);
2038
2039             /* EWALD ELECTROSTATICS */
2040
2041             /* Analytical PME correction */
2042             zeta2            = _mm256_mul_ps(beta2,rsq23);
2043             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
2044             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2045             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2046             felec            = _mm256_mul_ps(qq23,felec);
2047
2048             fscal            = felec;
2049
2050             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2051
2052             /* Calculate temporary vectorial force */
2053             tx               = _mm256_mul_ps(fscal,dx23);
2054             ty               = _mm256_mul_ps(fscal,dy23);
2055             tz               = _mm256_mul_ps(fscal,dz23);
2056
2057             /* Update vectorial force */
2058             fix2             = _mm256_add_ps(fix2,tx);
2059             fiy2             = _mm256_add_ps(fiy2,ty);
2060             fiz2             = _mm256_add_ps(fiz2,tz);
2061
2062             fjx3             = _mm256_add_ps(fjx3,tx);
2063             fjy3             = _mm256_add_ps(fjy3,ty);
2064             fjz3             = _mm256_add_ps(fjz3,tz);
2065
2066             /**************************
2067              * CALCULATE INTERACTIONS *
2068              **************************/
2069
2070             r31              = _mm256_mul_ps(rsq31,rinv31);
2071             r31              = _mm256_andnot_ps(dummy_mask,r31);
2072
2073             /* EWALD ELECTROSTATICS */
2074
2075             /* Analytical PME correction */
2076             zeta2            = _mm256_mul_ps(beta2,rsq31);
2077             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
2078             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2079             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2080             felec            = _mm256_mul_ps(qq31,felec);
2081
2082             fscal            = felec;
2083
2084             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2085
2086             /* Calculate temporary vectorial force */
2087             tx               = _mm256_mul_ps(fscal,dx31);
2088             ty               = _mm256_mul_ps(fscal,dy31);
2089             tz               = _mm256_mul_ps(fscal,dz31);
2090
2091             /* Update vectorial force */
2092             fix3             = _mm256_add_ps(fix3,tx);
2093             fiy3             = _mm256_add_ps(fiy3,ty);
2094             fiz3             = _mm256_add_ps(fiz3,tz);
2095
2096             fjx1             = _mm256_add_ps(fjx1,tx);
2097             fjy1             = _mm256_add_ps(fjy1,ty);
2098             fjz1             = _mm256_add_ps(fjz1,tz);
2099
2100             /**************************
2101              * CALCULATE INTERACTIONS *
2102              **************************/
2103
2104             r32              = _mm256_mul_ps(rsq32,rinv32);
2105             r32              = _mm256_andnot_ps(dummy_mask,r32);
2106
2107             /* EWALD ELECTROSTATICS */
2108
2109             /* Analytical PME correction */
2110             zeta2            = _mm256_mul_ps(beta2,rsq32);
2111             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
2112             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2113             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2114             felec            = _mm256_mul_ps(qq32,felec);
2115
2116             fscal            = felec;
2117
2118             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2119
2120             /* Calculate temporary vectorial force */
2121             tx               = _mm256_mul_ps(fscal,dx32);
2122             ty               = _mm256_mul_ps(fscal,dy32);
2123             tz               = _mm256_mul_ps(fscal,dz32);
2124
2125             /* Update vectorial force */
2126             fix3             = _mm256_add_ps(fix3,tx);
2127             fiy3             = _mm256_add_ps(fiy3,ty);
2128             fiz3             = _mm256_add_ps(fiz3,tz);
2129
2130             fjx2             = _mm256_add_ps(fjx2,tx);
2131             fjy2             = _mm256_add_ps(fjy2,ty);
2132             fjz2             = _mm256_add_ps(fjz2,tz);
2133
2134             /**************************
2135              * CALCULATE INTERACTIONS *
2136              **************************/
2137
2138             r33              = _mm256_mul_ps(rsq33,rinv33);
2139             r33              = _mm256_andnot_ps(dummy_mask,r33);
2140
2141             /* EWALD ELECTROSTATICS */
2142
2143             /* Analytical PME correction */
2144             zeta2            = _mm256_mul_ps(beta2,rsq33);
2145             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
2146             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2147             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2148             felec            = _mm256_mul_ps(qq33,felec);
2149
2150             fscal            = felec;
2151
2152             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2153
2154             /* Calculate temporary vectorial force */
2155             tx               = _mm256_mul_ps(fscal,dx33);
2156             ty               = _mm256_mul_ps(fscal,dy33);
2157             tz               = _mm256_mul_ps(fscal,dz33);
2158
2159             /* Update vectorial force */
2160             fix3             = _mm256_add_ps(fix3,tx);
2161             fiy3             = _mm256_add_ps(fiy3,ty);
2162             fiz3             = _mm256_add_ps(fiz3,tz);
2163
2164             fjx3             = _mm256_add_ps(fjx3,tx);
2165             fjy3             = _mm256_add_ps(fjy3,ty);
2166             fjz3             = _mm256_add_ps(fjz3,tz);
2167
2168             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2169             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2170             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2171             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2172             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2173             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2174             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2175             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2176
2177             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2178                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
2179                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2180
2181             /* Inner loop uses 513 flops */
2182         }
2183
2184         /* End of innermost loop */
2185
2186         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2187                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
2188
2189         /* Increment number of inner iterations */
2190         inneriter                  += j_index_end - j_index_start;
2191
2192         /* Outer loop uses 18 flops */
2193     }
2194
2195     /* Increment number of outer iterations */
2196     outeriter        += nri;
2197
2198     /* Update outer/inner flops */
2199
2200     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*513);
2201 }