src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwCSTab_GeomW4W4_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4W4_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Water4
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwCSTab_GeomW4W4_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  97     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  98     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  99     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 100     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
 101     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 102     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 103     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 104     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 105     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 106     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 107     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 108     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 109     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 110     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 111     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 112     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 113     real             *charge;
 114     int              nvdwtype;
 115     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 116     int              *vdwtype;
 117     real             *vdwparam;
 118     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 119     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 120     __m256i          vfitab;
 121     __m128i          vfitab_lo,vfitab_hi;
 122     __m128i          ifour       = _mm_set1_epi32(4);
 123     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 124     real             *vftab;
 125     __m256i          ewitab;
 126     __m128i          ewitab_lo,ewitab_hi;
 127     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 128     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 129     real             *ewtab;
 130     __m256           dummy_mask,cutoff_mask;
 131     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 132     __m256           one     = _mm256_set1_ps(1.0);
 133     __m256           two     = _mm256_set1_ps(2.0);
 134     x                = xx[0];
 135     f                = ff[0];
 136
 137     nri              = nlist->nri;
 138     iinr             = nlist->iinr;
 139     jindex           = nlist->jindex;
 140     jjnr             = nlist->jjnr;
 141     shiftidx         = nlist->shift;
 142     gid              = nlist->gid;
 143     shiftvec         = fr->shift_vec[0];
 144     fshift           = fr->fshift[0];
 145     facel            = _mm256_set1_ps(fr->epsfac);
 146     charge           = mdatoms->chargeA;
 147     nvdwtype         = fr->ntype;
 148     vdwparam         = fr->nbfp;
 149     vdwtype          = mdatoms->typeA;
 150
 151     vftab            = kernel_data->table_vdw->data;
 152     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 153
 154     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 155     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 156     beta2            = _mm256_mul_ps(beta,beta);
 157     beta3            = _mm256_mul_ps(beta,beta2);
 158
 159     ewtab            = fr->ic->tabq_coul_FDV0;
 160     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 161     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 162
 163     /* Setup water-specific parameters */
 164     inr              = nlist->iinr[0];
 165     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 166     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 167     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 168     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 169
 170     jq1              = _mm256_set1_ps(charge[inr+1]);
 171     jq2              = _mm256_set1_ps(charge[inr+2]);
 172     jq3              = _mm256_set1_ps(charge[inr+3]);
 173     vdwjidx0A        = 2*vdwtype[inr+0];
 174     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 175     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 176     qq11             = _mm256_mul_ps(iq1,jq1);
 177     qq12             = _mm256_mul_ps(iq1,jq2);
 178     qq13             = _mm256_mul_ps(iq1,jq3);
 179     qq21             = _mm256_mul_ps(iq2,jq1);
 180     qq22             = _mm256_mul_ps(iq2,jq2);
 181     qq23             = _mm256_mul_ps(iq2,jq3);
 182     qq31             = _mm256_mul_ps(iq3,jq1);
 183     qq32             = _mm256_mul_ps(iq3,jq2);
 184     qq33             = _mm256_mul_ps(iq3,jq3);
 185
 186     /* Avoid stupid compiler warnings */
 187     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 188     j_coord_offsetA = 0;
 189     j_coord_offsetB = 0;
 190     j_coord_offsetC = 0;
 191     j_coord_offsetD = 0;
 192     j_coord_offsetE = 0;
 193     j_coord_offsetF = 0;
 194     j_coord_offsetG = 0;
 195     j_coord_offsetH = 0;
 196
 197     outeriter        = 0;
 198     inneriter        = 0;
 199
 200     for(iidx=0;iidx<4*DIM;iidx++)
 201     {
 202         scratch[iidx] = 0.0;
 203     }
 204
 205     /* Start outer loop over neighborlists */
 206     for(iidx=0; iidx<nri; iidx++)
 207     {
 208         /* Load shift vector for this list */
 209         i_shift_offset   = DIM*shiftidx[iidx];
 210
 211         /* Load limits for loop over neighbors */
 212         j_index_start    = jindex[iidx];
 213         j_index_end      = jindex[iidx+1];
 214
 215         /* Get outer coordinate index */
 216         inr              = iinr[iidx];
 217         i_coord_offset   = DIM*inr;
 218
 219         /* Load i particle coords and add shift vector */
 220         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 221                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 222
 223         fix0             = _mm256_setzero_ps();
 224         fiy0             = _mm256_setzero_ps();
 225         fiz0             = _mm256_setzero_ps();
 226         fix1             = _mm256_setzero_ps();
 227         fiy1             = _mm256_setzero_ps();
 228         fiz1             = _mm256_setzero_ps();
 229         fix2             = _mm256_setzero_ps();
 230         fiy2             = _mm256_setzero_ps();
 231         fiz2             = _mm256_setzero_ps();
 232         fix3             = _mm256_setzero_ps();
 233         fiy3             = _mm256_setzero_ps();
 234         fiz3             = _mm256_setzero_ps();
 235
 236         /* Reset potential sums */
 237         velecsum         = _mm256_setzero_ps();
 238         vvdwsum          = _mm256_setzero_ps();
 239
 240         /* Start inner kernel loop */
 241         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 242         {
 243
 244             /* Get j neighbor index, and coordinate index */
 245             jnrA             = jjnr[jidx];
 246             jnrB             = jjnr[jidx+1];
 247             jnrC             = jjnr[jidx+2];
 248             jnrD             = jjnr[jidx+3];
 249             jnrE             = jjnr[jidx+4];
 250             jnrF             = jjnr[jidx+5];
 251             jnrG             = jjnr[jidx+6];
 252             jnrH             = jjnr[jidx+7];
 253             j_coord_offsetA  = DIM*jnrA;
 254             j_coord_offsetB  = DIM*jnrB;
 255             j_coord_offsetC  = DIM*jnrC;
 256             j_coord_offsetD  = DIM*jnrD;
 257             j_coord_offsetE  = DIM*jnrE;
 258             j_coord_offsetF  = DIM*jnrF;
 259             j_coord_offsetG  = DIM*jnrG;
 260             j_coord_offsetH  = DIM*jnrH;
 261
 262             /* load j atom coordinates */
 263             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 264                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 265                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 266                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 267                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 268                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 269
 270             /* Calculate displacement vector */
 271             dx00             = _mm256_sub_ps(ix0,jx0);
 272             dy00             = _mm256_sub_ps(iy0,jy0);
 273             dz00             = _mm256_sub_ps(iz0,jz0);
 274             dx11             = _mm256_sub_ps(ix1,jx1);
 275             dy11             = _mm256_sub_ps(iy1,jy1);
 276             dz11             = _mm256_sub_ps(iz1,jz1);
 277             dx12             = _mm256_sub_ps(ix1,jx2);
 278             dy12             = _mm256_sub_ps(iy1,jy2);
 279             dz12             = _mm256_sub_ps(iz1,jz2);
 280             dx13             = _mm256_sub_ps(ix1,jx3);
 281             dy13             = _mm256_sub_ps(iy1,jy3);
 282             dz13             = _mm256_sub_ps(iz1,jz3);
 283             dx21             = _mm256_sub_ps(ix2,jx1);
 284             dy21             = _mm256_sub_ps(iy2,jy1);
 285             dz21             = _mm256_sub_ps(iz2,jz1);
 286             dx22             = _mm256_sub_ps(ix2,jx2);
 287             dy22             = _mm256_sub_ps(iy2,jy2);
 288             dz22             = _mm256_sub_ps(iz2,jz2);
 289             dx23             = _mm256_sub_ps(ix2,jx3);
 290             dy23             = _mm256_sub_ps(iy2,jy3);
 291             dz23             = _mm256_sub_ps(iz2,jz3);
 292             dx31             = _mm256_sub_ps(ix3,jx1);
 293             dy31             = _mm256_sub_ps(iy3,jy1);
 294             dz31             = _mm256_sub_ps(iz3,jz1);
 295             dx32             = _mm256_sub_ps(ix3,jx2);
 296             dy32             = _mm256_sub_ps(iy3,jy2);
 297             dz32             = _mm256_sub_ps(iz3,jz2);
 298             dx33             = _mm256_sub_ps(ix3,jx3);
 299             dy33             = _mm256_sub_ps(iy3,jy3);
 300             dz33             = _mm256_sub_ps(iz3,jz3);
 301
 302             /* Calculate squared distance and things based on it */
 303             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 304             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 305             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 306             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 307             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 308             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 309             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 310             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 311             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 312             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 313
 314             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 315             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 316             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 317             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 318             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 319             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 320             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 321             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 322             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 323             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 324
 325             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 326             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 327             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 328             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 329             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 330             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 331             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 332             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 333             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 334
 335             fjx0             = _mm256_setzero_ps();
 336             fjy0             = _mm256_setzero_ps();
 337             fjz0             = _mm256_setzero_ps();
 338             fjx1             = _mm256_setzero_ps();
 339             fjy1             = _mm256_setzero_ps();
 340             fjz1             = _mm256_setzero_ps();
 341             fjx2             = _mm256_setzero_ps();
 342             fjy2             = _mm256_setzero_ps();
 343             fjz2             = _mm256_setzero_ps();
 344             fjx3             = _mm256_setzero_ps();
 345             fjy3             = _mm256_setzero_ps();
 346             fjz3             = _mm256_setzero_ps();
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             r00              = _mm256_mul_ps(rsq00,rinv00);
 353
 354             /* Calculate table index by multiplying r with table scale and truncate to integer */
 355             rt               = _mm256_mul_ps(r00,vftabscale);
 356             vfitab           = _mm256_cvttps_epi32(rt);
 357             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 358             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 359             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 360             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 361             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 362             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 363
 364             /* CUBIC SPLINE TABLE DISPERSION */
 365             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 366                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 367             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 368                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 369             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 370                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 371             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 372                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 373             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 374             Heps             = _mm256_mul_ps(vfeps,H);
 375             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 376             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 377             vvdw6            = _mm256_mul_ps(c6_00,VV);
 378             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 379             fvdw6            = _mm256_mul_ps(c6_00,FF);
 380
 381             /* CUBIC SPLINE TABLE REPULSION */
 382             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 383             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 384             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 385                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 386             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 387                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 388             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 389                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 390             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 391                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 392             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 393             Heps             = _mm256_mul_ps(vfeps,H);
 394             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 395             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 396             vvdw12           = _mm256_mul_ps(c12_00,VV);
 397             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 398             fvdw12           = _mm256_mul_ps(c12_00,FF);
 399             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 400             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 401
 402             /* Update potential sum for this i atom from the interaction with this j atom. */
 403             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 404
 405             fscal            = fvdw;
 406
 407             /* Calculate temporary vectorial force */
 408             tx               = _mm256_mul_ps(fscal,dx00);
 409             ty               = _mm256_mul_ps(fscal,dy00);
 410             tz               = _mm256_mul_ps(fscal,dz00);
 411
 412             /* Update vectorial force */
 413             fix0             = _mm256_add_ps(fix0,tx);
 414             fiy0             = _mm256_add_ps(fiy0,ty);
 415             fiz0             = _mm256_add_ps(fiz0,tz);
 416
 417             fjx0             = _mm256_add_ps(fjx0,tx);
 418             fjy0             = _mm256_add_ps(fjy0,ty);
 419             fjz0             = _mm256_add_ps(fjz0,tz);
 420
 421             /**************************
 422              * CALCULATE INTERACTIONS *
 423              **************************/
 424
 425             r11              = _mm256_mul_ps(rsq11,rinv11);
 426
 427             /* EWALD ELECTROSTATICS */
 428
 429             /* Analytical PME correction */
 430             zeta2            = _mm256_mul_ps(beta2,rsq11);
 431             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 432             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 433             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 434             felec            = _mm256_mul_ps(qq11,felec);
 435             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 436             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 437             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 438             velec            = _mm256_mul_ps(qq11,velec);
 439
 440             /* Update potential sum for this i atom from the interaction with this j atom. */
 441             velecsum         = _mm256_add_ps(velecsum,velec);
 442
 443             fscal            = felec;
 444
 445             /* Calculate temporary vectorial force */
 446             tx               = _mm256_mul_ps(fscal,dx11);
 447             ty               = _mm256_mul_ps(fscal,dy11);
 448             tz               = _mm256_mul_ps(fscal,dz11);
 449
 450             /* Update vectorial force */
 451             fix1             = _mm256_add_ps(fix1,tx);
 452             fiy1             = _mm256_add_ps(fiy1,ty);
 453             fiz1             = _mm256_add_ps(fiz1,tz);
 454
 455             fjx1             = _mm256_add_ps(fjx1,tx);
 456             fjy1             = _mm256_add_ps(fjy1,ty);
 457             fjz1             = _mm256_add_ps(fjz1,tz);
 458
 459             /**************************
 460              * CALCULATE INTERACTIONS *
 461              **************************/
 462
 463             r12              = _mm256_mul_ps(rsq12,rinv12);
 464
 465             /* EWALD ELECTROSTATICS */
 466
 467             /* Analytical PME correction */
 468             zeta2            = _mm256_mul_ps(beta2,rsq12);
 469             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 470             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 471             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 472             felec            = _mm256_mul_ps(qq12,felec);
 473             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 474             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 475             velec            = _mm256_sub_ps(rinv12,pmecorrV);
 476             velec            = _mm256_mul_ps(qq12,velec);
 477
 478             /* Update potential sum for this i atom from the interaction with this j atom. */
 479             velecsum         = _mm256_add_ps(velecsum,velec);
 480
 481             fscal            = felec;
 482
 483             /* Calculate temporary vectorial force */
 484             tx               = _mm256_mul_ps(fscal,dx12);
 485             ty               = _mm256_mul_ps(fscal,dy12);
 486             tz               = _mm256_mul_ps(fscal,dz12);
 487
 488             /* Update vectorial force */
 489             fix1             = _mm256_add_ps(fix1,tx);
 490             fiy1             = _mm256_add_ps(fiy1,ty);
 491             fiz1             = _mm256_add_ps(fiz1,tz);
 492
 493             fjx2             = _mm256_add_ps(fjx2,tx);
 494             fjy2             = _mm256_add_ps(fjy2,ty);
 495             fjz2             = _mm256_add_ps(fjz2,tz);
 496
 497             /**************************
 498              * CALCULATE INTERACTIONS *
 499              **************************/
 500
 501             r13              = _mm256_mul_ps(rsq13,rinv13);
 502
 503             /* EWALD ELECTROSTATICS */
 504
 505             /* Analytical PME correction */
 506             zeta2            = _mm256_mul_ps(beta2,rsq13);
 507             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
 508             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 509             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 510             felec            = _mm256_mul_ps(qq13,felec);
 511             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 512             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 513             velec            = _mm256_sub_ps(rinv13,pmecorrV);
 514             velec            = _mm256_mul_ps(qq13,velec);
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velecsum         = _mm256_add_ps(velecsum,velec);
 518
 519             fscal            = felec;
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm256_mul_ps(fscal,dx13);
 523             ty               = _mm256_mul_ps(fscal,dy13);
 524             tz               = _mm256_mul_ps(fscal,dz13);
 525
 526             /* Update vectorial force */
 527             fix1             = _mm256_add_ps(fix1,tx);
 528             fiy1             = _mm256_add_ps(fiy1,ty);
 529             fiz1             = _mm256_add_ps(fiz1,tz);
 530
 531             fjx3             = _mm256_add_ps(fjx3,tx);
 532             fjy3             = _mm256_add_ps(fjy3,ty);
 533             fjz3             = _mm256_add_ps(fjz3,tz);
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             r21              = _mm256_mul_ps(rsq21,rinv21);
 540
 541             /* EWALD ELECTROSTATICS */
 542
 543             /* Analytical PME correction */
 544             zeta2            = _mm256_mul_ps(beta2,rsq21);
 545             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 546             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 547             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 548             felec            = _mm256_mul_ps(qq21,felec);
 549             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 550             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 551             velec            = _mm256_sub_ps(rinv21,pmecorrV);
 552             velec            = _mm256_mul_ps(qq21,velec);
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velecsum         = _mm256_add_ps(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm256_mul_ps(fscal,dx21);
 561             ty               = _mm256_mul_ps(fscal,dy21);
 562             tz               = _mm256_mul_ps(fscal,dz21);
 563
 564             /* Update vectorial force */
 565             fix2             = _mm256_add_ps(fix2,tx);
 566             fiy2             = _mm256_add_ps(fiy2,ty);
 567             fiz2             = _mm256_add_ps(fiz2,tz);
 568
 569             fjx1             = _mm256_add_ps(fjx1,tx);
 570             fjy1             = _mm256_add_ps(fjy1,ty);
 571             fjz1             = _mm256_add_ps(fjz1,tz);
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             r22              = _mm256_mul_ps(rsq22,rinv22);
 578
 579             /* EWALD ELECTROSTATICS */
 580
 581             /* Analytical PME correction */
 582             zeta2            = _mm256_mul_ps(beta2,rsq22);
 583             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 584             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 585             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 586             felec            = _mm256_mul_ps(qq22,felec);
 587             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 588             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 589             velec            = _mm256_sub_ps(rinv22,pmecorrV);
 590             velec            = _mm256_mul_ps(qq22,velec);
 591
 592             /* Update potential sum for this i atom from the interaction with this j atom. */
 593             velecsum         = _mm256_add_ps(velecsum,velec);
 594
 595             fscal            = felec;
 596
 597             /* Calculate temporary vectorial force */
 598             tx               = _mm256_mul_ps(fscal,dx22);
 599             ty               = _mm256_mul_ps(fscal,dy22);
 600             tz               = _mm256_mul_ps(fscal,dz22);
 601
 602             /* Update vectorial force */
 603             fix2             = _mm256_add_ps(fix2,tx);
 604             fiy2             = _mm256_add_ps(fiy2,ty);
 605             fiz2             = _mm256_add_ps(fiz2,tz);
 606
 607             fjx2             = _mm256_add_ps(fjx2,tx);
 608             fjy2             = _mm256_add_ps(fjy2,ty);
 609             fjz2             = _mm256_add_ps(fjz2,tz);
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             r23              = _mm256_mul_ps(rsq23,rinv23);
 616
 617             /* EWALD ELECTROSTATICS */
 618
 619             /* Analytical PME correction */
 620             zeta2            = _mm256_mul_ps(beta2,rsq23);
 621             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
 622             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 623             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 624             felec            = _mm256_mul_ps(qq23,felec);
 625             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 626             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 627             velec            = _mm256_sub_ps(rinv23,pmecorrV);
 628             velec            = _mm256_mul_ps(qq23,velec);
 629
 630             /* Update potential sum for this i atom from the interaction with this j atom. */
 631             velecsum         = _mm256_add_ps(velecsum,velec);
 632
 633             fscal            = felec;
 634
 635             /* Calculate temporary vectorial force */
 636             tx               = _mm256_mul_ps(fscal,dx23);
 637             ty               = _mm256_mul_ps(fscal,dy23);
 638             tz               = _mm256_mul_ps(fscal,dz23);
 639
 640             /* Update vectorial force */
 641             fix2             = _mm256_add_ps(fix2,tx);
 642             fiy2             = _mm256_add_ps(fiy2,ty);
 643             fiz2             = _mm256_add_ps(fiz2,tz);
 644
 645             fjx3             = _mm256_add_ps(fjx3,tx);
 646             fjy3             = _mm256_add_ps(fjy3,ty);
 647             fjz3             = _mm256_add_ps(fjz3,tz);
 648
 649             /**************************
 650              * CALCULATE INTERACTIONS *
 651              **************************/
 652
 653             r31              = _mm256_mul_ps(rsq31,rinv31);
 654
 655             /* EWALD ELECTROSTATICS */
 656
 657             /* Analytical PME correction */
 658             zeta2            = _mm256_mul_ps(beta2,rsq31);
 659             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
 660             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 661             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 662             felec            = _mm256_mul_ps(qq31,felec);
 663             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 664             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 665             velec            = _mm256_sub_ps(rinv31,pmecorrV);
 666             velec            = _mm256_mul_ps(qq31,velec);
 667
 668             /* Update potential sum for this i atom from the interaction with this j atom. */
 669             velecsum         = _mm256_add_ps(velecsum,velec);
 670
 671             fscal            = felec;
 672
 673             /* Calculate temporary vectorial force */
 674             tx               = _mm256_mul_ps(fscal,dx31);
 675             ty               = _mm256_mul_ps(fscal,dy31);
 676             tz               = _mm256_mul_ps(fscal,dz31);
 677
 678             /* Update vectorial force */
 679             fix3             = _mm256_add_ps(fix3,tx);
 680             fiy3             = _mm256_add_ps(fiy3,ty);
 681             fiz3             = _mm256_add_ps(fiz3,tz);
 682
 683             fjx1             = _mm256_add_ps(fjx1,tx);
 684             fjy1             = _mm256_add_ps(fjy1,ty);
 685             fjz1             = _mm256_add_ps(fjz1,tz);
 686
 687             /**************************
 688              * CALCULATE INTERACTIONS *
 689              **************************/
 690
 691             r32              = _mm256_mul_ps(rsq32,rinv32);
 692
 693             /* EWALD ELECTROSTATICS */
 694
 695             /* Analytical PME correction */
 696             zeta2            = _mm256_mul_ps(beta2,rsq32);
 697             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
 698             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 699             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 700             felec            = _mm256_mul_ps(qq32,felec);
 701             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 702             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 703             velec            = _mm256_sub_ps(rinv32,pmecorrV);
 704             velec            = _mm256_mul_ps(qq32,velec);
 705
 706             /* Update potential sum for this i atom from the interaction with this j atom. */
 707             velecsum         = _mm256_add_ps(velecsum,velec);
 708
 709             fscal            = felec;
 710
 711             /* Calculate temporary vectorial force */
 712             tx               = _mm256_mul_ps(fscal,dx32);
 713             ty               = _mm256_mul_ps(fscal,dy32);
 714             tz               = _mm256_mul_ps(fscal,dz32);
 715
 716             /* Update vectorial force */
 717             fix3             = _mm256_add_ps(fix3,tx);
 718             fiy3             = _mm256_add_ps(fiy3,ty);
 719             fiz3             = _mm256_add_ps(fiz3,tz);
 720
 721             fjx2             = _mm256_add_ps(fjx2,tx);
 722             fjy2             = _mm256_add_ps(fjy2,ty);
 723             fjz2             = _mm256_add_ps(fjz2,tz);
 724
 725             /**************************
 726              * CALCULATE INTERACTIONS *
 727              **************************/
 728
 729             r33              = _mm256_mul_ps(rsq33,rinv33);
 730
 731             /* EWALD ELECTROSTATICS */
 732
 733             /* Analytical PME correction */
 734             zeta2            = _mm256_mul_ps(beta2,rsq33);
 735             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
 736             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 737             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 738             felec            = _mm256_mul_ps(qq33,felec);
 739             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 740             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 741             velec            = _mm256_sub_ps(rinv33,pmecorrV);
 742             velec            = _mm256_mul_ps(qq33,velec);
 743
 744             /* Update potential sum for this i atom from the interaction with this j atom. */
 745             velecsum         = _mm256_add_ps(velecsum,velec);
 746
 747             fscal            = felec;
 748
 749             /* Calculate temporary vectorial force */
 750             tx               = _mm256_mul_ps(fscal,dx33);
 751             ty               = _mm256_mul_ps(fscal,dy33);
 752             tz               = _mm256_mul_ps(fscal,dz33);
 753
 754             /* Update vectorial force */
 755             fix3             = _mm256_add_ps(fix3,tx);
 756             fiy3             = _mm256_add_ps(fiy3,ty);
 757             fiz3             = _mm256_add_ps(fiz3,tz);
 758
 759             fjx3             = _mm256_add_ps(fjx3,tx);
 760             fjy3             = _mm256_add_ps(fjy3,ty);
 761             fjz3             = _mm256_add_ps(fjz3,tz);
 762
 763             fjptrA             = f+j_coord_offsetA;
 764             fjptrB             = f+j_coord_offsetB;
 765             fjptrC             = f+j_coord_offsetC;
 766             fjptrD             = f+j_coord_offsetD;
 767             fjptrE             = f+j_coord_offsetE;
 768             fjptrF             = f+j_coord_offsetF;
 769             fjptrG             = f+j_coord_offsetG;
 770             fjptrH             = f+j_coord_offsetH;
 771
 772             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 773                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 774                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 775
 776             /* Inner loop uses 815 flops */
 777         }
 778
 779         if(jidx<j_index_end)
 780         {
 781
 782             /* Get j neighbor index, and coordinate index */
 783             jnrlistA         = jjnr[jidx];
 784             jnrlistB         = jjnr[jidx+1];
 785             jnrlistC         = jjnr[jidx+2];
 786             jnrlistD         = jjnr[jidx+3];
 787             jnrlistE         = jjnr[jidx+4];
 788             jnrlistF         = jjnr[jidx+5];
 789             jnrlistG         = jjnr[jidx+6];
 790             jnrlistH         = jjnr[jidx+7];
 791             /* Sign of each element will be negative for non-real atoms.
 792              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 793              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 794              */
 795             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 796                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 797
 798             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 799             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 800             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 801             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 802             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 803             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 804             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 805             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 806             j_coord_offsetA  = DIM*jnrA;
 807             j_coord_offsetB  = DIM*jnrB;
 808             j_coord_offsetC  = DIM*jnrC;
 809             j_coord_offsetD  = DIM*jnrD;
 810             j_coord_offsetE  = DIM*jnrE;
 811             j_coord_offsetF  = DIM*jnrF;
 812             j_coord_offsetG  = DIM*jnrG;
 813             j_coord_offsetH  = DIM*jnrH;
 814
 815             /* load j atom coordinates */
 816             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 817                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 818                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 819                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 820                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 821                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 822
 823             /* Calculate displacement vector */
 824             dx00             = _mm256_sub_ps(ix0,jx0);
 825             dy00             = _mm256_sub_ps(iy0,jy0);
 826             dz00             = _mm256_sub_ps(iz0,jz0);
 827             dx11             = _mm256_sub_ps(ix1,jx1);
 828             dy11             = _mm256_sub_ps(iy1,jy1);
 829             dz11             = _mm256_sub_ps(iz1,jz1);
 830             dx12             = _mm256_sub_ps(ix1,jx2);
 831             dy12             = _mm256_sub_ps(iy1,jy2);
 832             dz12             = _mm256_sub_ps(iz1,jz2);
 833             dx13             = _mm256_sub_ps(ix1,jx3);
 834             dy13             = _mm256_sub_ps(iy1,jy3);
 835             dz13             = _mm256_sub_ps(iz1,jz3);
 836             dx21             = _mm256_sub_ps(ix2,jx1);
 837             dy21             = _mm256_sub_ps(iy2,jy1);
 838             dz21             = _mm256_sub_ps(iz2,jz1);
 839             dx22             = _mm256_sub_ps(ix2,jx2);
 840             dy22             = _mm256_sub_ps(iy2,jy2);
 841             dz22             = _mm256_sub_ps(iz2,jz2);
 842             dx23             = _mm256_sub_ps(ix2,jx3);
 843             dy23             = _mm256_sub_ps(iy2,jy3);
 844             dz23             = _mm256_sub_ps(iz2,jz3);
 845             dx31             = _mm256_sub_ps(ix3,jx1);
 846             dy31             = _mm256_sub_ps(iy3,jy1);
 847             dz31             = _mm256_sub_ps(iz3,jz1);
 848             dx32             = _mm256_sub_ps(ix3,jx2);
 849             dy32             = _mm256_sub_ps(iy3,jy2);
 850             dz32             = _mm256_sub_ps(iz3,jz2);
 851             dx33             = _mm256_sub_ps(ix3,jx3);
 852             dy33             = _mm256_sub_ps(iy3,jy3);
 853             dz33             = _mm256_sub_ps(iz3,jz3);
 854
 855             /* Calculate squared distance and things based on it */
 856             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 857             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 858             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 859             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 860             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 861             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 862             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 863             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 864             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 865             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 866
 867             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 868             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 869             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 870             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 871             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 872             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 873             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 874             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 875             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 876             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 877
 878             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 879             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 880             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 881             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 882             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 883             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 884             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 885             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 886             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 887
 888             fjx0             = _mm256_setzero_ps();
 889             fjy0             = _mm256_setzero_ps();
 890             fjz0             = _mm256_setzero_ps();
 891             fjx1             = _mm256_setzero_ps();
 892             fjy1             = _mm256_setzero_ps();
 893             fjz1             = _mm256_setzero_ps();
 894             fjx2             = _mm256_setzero_ps();
 895             fjy2             = _mm256_setzero_ps();
 896             fjz2             = _mm256_setzero_ps();
 897             fjx3             = _mm256_setzero_ps();
 898             fjy3             = _mm256_setzero_ps();
 899             fjz3             = _mm256_setzero_ps();
 900
 901             /**************************
 902              * CALCULATE INTERACTIONS *
 903              **************************/
 904
 905             r00              = _mm256_mul_ps(rsq00,rinv00);
 906             r00              = _mm256_andnot_ps(dummy_mask,r00);
 907
 908             /* Calculate table index by multiplying r with table scale and truncate to integer */
 909             rt               = _mm256_mul_ps(r00,vftabscale);
 910             vfitab           = _mm256_cvttps_epi32(rt);
 911             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 912             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 913             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 914             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 915             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 916             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 917
 918             /* CUBIC SPLINE TABLE DISPERSION */
 919             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 920                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 921             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 922                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 923             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 924                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 925             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 926                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 927             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 928             Heps             = _mm256_mul_ps(vfeps,H);
 929             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 930             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 931             vvdw6            = _mm256_mul_ps(c6_00,VV);
 932             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 933             fvdw6            = _mm256_mul_ps(c6_00,FF);
 934
 935             /* CUBIC SPLINE TABLE REPULSION */
 936             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 937             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 938             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 939                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 940             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 941                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 942             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 943                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 944             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 945                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 946             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 947             Heps             = _mm256_mul_ps(vfeps,H);
 948             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 949             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 950             vvdw12           = _mm256_mul_ps(c12_00,VV);
 951             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 952             fvdw12           = _mm256_mul_ps(c12_00,FF);
 953             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 954             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 955
 956             /* Update potential sum for this i atom from the interaction with this j atom. */
 957             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 958             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 959
 960             fscal            = fvdw;
 961
 962             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 963
 964             /* Calculate temporary vectorial force */
 965             tx               = _mm256_mul_ps(fscal,dx00);
 966             ty               = _mm256_mul_ps(fscal,dy00);
 967             tz               = _mm256_mul_ps(fscal,dz00);
 968
 969             /* Update vectorial force */
 970             fix0             = _mm256_add_ps(fix0,tx);
 971             fiy0             = _mm256_add_ps(fiy0,ty);
 972             fiz0             = _mm256_add_ps(fiz0,tz);
 973
 974             fjx0             = _mm256_add_ps(fjx0,tx);
 975             fjy0             = _mm256_add_ps(fjy0,ty);
 976             fjz0             = _mm256_add_ps(fjz0,tz);
 977
 978             /**************************
 979              * CALCULATE INTERACTIONS *
 980              **************************/
 981
 982             r11              = _mm256_mul_ps(rsq11,rinv11);
 983             r11              = _mm256_andnot_ps(dummy_mask,r11);
 984
 985             /* EWALD ELECTROSTATICS */
 986
 987             /* Analytical PME correction */
 988             zeta2            = _mm256_mul_ps(beta2,rsq11);
 989             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 990             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 991             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 992             felec            = _mm256_mul_ps(qq11,felec);
 993             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 994             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 995             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 996             velec            = _mm256_mul_ps(qq11,velec);
 997
 998             /* Update potential sum for this i atom from the interaction with this j atom. */
 999             velec            = _mm256_andnot_ps(dummy_mask,velec);
1000             velecsum         = _mm256_add_ps(velecsum,velec);
1001
1002             fscal            = felec;
1003
1004             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1005
1006             /* Calculate temporary vectorial force */
1007             tx               = _mm256_mul_ps(fscal,dx11);
1008             ty               = _mm256_mul_ps(fscal,dy11);
1009             tz               = _mm256_mul_ps(fscal,dz11);
1010
1011             /* Update vectorial force */
1012             fix1             = _mm256_add_ps(fix1,tx);
1013             fiy1             = _mm256_add_ps(fiy1,ty);
1014             fiz1             = _mm256_add_ps(fiz1,tz);
1015
1016             fjx1             = _mm256_add_ps(fjx1,tx);
1017             fjy1             = _mm256_add_ps(fjy1,ty);
1018             fjz1             = _mm256_add_ps(fjz1,tz);
1019
1020             /**************************
1021              * CALCULATE INTERACTIONS *
1022              **************************/
1023
1024             r12              = _mm256_mul_ps(rsq12,rinv12);
1025             r12              = _mm256_andnot_ps(dummy_mask,r12);
1026
1027             /* EWALD ELECTROSTATICS */
1028
1029             /* Analytical PME correction */
1030             zeta2            = _mm256_mul_ps(beta2,rsq12);
1031             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1032             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1033             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1034             felec            = _mm256_mul_ps(qq12,felec);
1035             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1036             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1037             velec            = _mm256_sub_ps(rinv12,pmecorrV);
1038             velec            = _mm256_mul_ps(qq12,velec);
1039
1040             /* Update potential sum for this i atom from the interaction with this j atom. */
1041             velec            = _mm256_andnot_ps(dummy_mask,velec);
1042             velecsum         = _mm256_add_ps(velecsum,velec);
1043
1044             fscal            = felec;
1045
1046             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1047
1048             /* Calculate temporary vectorial force */
1049             tx               = _mm256_mul_ps(fscal,dx12);
1050             ty               = _mm256_mul_ps(fscal,dy12);
1051             tz               = _mm256_mul_ps(fscal,dz12);
1052
1053             /* Update vectorial force */
1054             fix1             = _mm256_add_ps(fix1,tx);
1055             fiy1             = _mm256_add_ps(fiy1,ty);
1056             fiz1             = _mm256_add_ps(fiz1,tz);
1057
1058             fjx2             = _mm256_add_ps(fjx2,tx);
1059             fjy2             = _mm256_add_ps(fjy2,ty);
1060             fjz2             = _mm256_add_ps(fjz2,tz);
1061
1062             /**************************
1063              * CALCULATE INTERACTIONS *
1064              **************************/
1065
1066             r13              = _mm256_mul_ps(rsq13,rinv13);
1067             r13              = _mm256_andnot_ps(dummy_mask,r13);
1068
1069             /* EWALD ELECTROSTATICS */
1070
1071             /* Analytical PME correction */
1072             zeta2            = _mm256_mul_ps(beta2,rsq13);
1073             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
1074             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1075             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1076             felec            = _mm256_mul_ps(qq13,felec);
1077             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1078             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1079             velec            = _mm256_sub_ps(rinv13,pmecorrV);
1080             velec            = _mm256_mul_ps(qq13,velec);
1081
1082             /* Update potential sum for this i atom from the interaction with this j atom. */
1083             velec            = _mm256_andnot_ps(dummy_mask,velec);
1084             velecsum         = _mm256_add_ps(velecsum,velec);
1085
1086             fscal            = felec;
1087
1088             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1089
1090             /* Calculate temporary vectorial force */
1091             tx               = _mm256_mul_ps(fscal,dx13);
1092             ty               = _mm256_mul_ps(fscal,dy13);
1093             tz               = _mm256_mul_ps(fscal,dz13);
1094
1095             /* Update vectorial force */
1096             fix1             = _mm256_add_ps(fix1,tx);
1097             fiy1             = _mm256_add_ps(fiy1,ty);
1098             fiz1             = _mm256_add_ps(fiz1,tz);
1099
1100             fjx3             = _mm256_add_ps(fjx3,tx);
1101             fjy3             = _mm256_add_ps(fjy3,ty);
1102             fjz3             = _mm256_add_ps(fjz3,tz);
1103
1104             /**************************
1105              * CALCULATE INTERACTIONS *
1106              **************************/
1107
1108             r21              = _mm256_mul_ps(rsq21,rinv21);
1109             r21              = _mm256_andnot_ps(dummy_mask,r21);
1110
1111             /* EWALD ELECTROSTATICS */
1112
1113             /* Analytical PME correction */
1114             zeta2            = _mm256_mul_ps(beta2,rsq21);
1115             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1116             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1117             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1118             felec            = _mm256_mul_ps(qq21,felec);
1119             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1120             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1121             velec            = _mm256_sub_ps(rinv21,pmecorrV);
1122             velec            = _mm256_mul_ps(qq21,velec);
1123
1124             /* Update potential sum for this i atom from the interaction with this j atom. */
1125             velec            = _mm256_andnot_ps(dummy_mask,velec);
1126             velecsum         = _mm256_add_ps(velecsum,velec);
1127
1128             fscal            = felec;
1129
1130             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1131
1132             /* Calculate temporary vectorial force */
1133             tx               = _mm256_mul_ps(fscal,dx21);
1134             ty               = _mm256_mul_ps(fscal,dy21);
1135             tz               = _mm256_mul_ps(fscal,dz21);
1136
1137             /* Update vectorial force */
1138             fix2             = _mm256_add_ps(fix2,tx);
1139             fiy2             = _mm256_add_ps(fiy2,ty);
1140             fiz2             = _mm256_add_ps(fiz2,tz);
1141
1142             fjx1             = _mm256_add_ps(fjx1,tx);
1143             fjy1             = _mm256_add_ps(fjy1,ty);
1144             fjz1             = _mm256_add_ps(fjz1,tz);
1145
1146             /**************************
1147              * CALCULATE INTERACTIONS *
1148              **************************/
1149
1150             r22              = _mm256_mul_ps(rsq22,rinv22);
1151             r22              = _mm256_andnot_ps(dummy_mask,r22);
1152
1153             /* EWALD ELECTROSTATICS */
1154
1155             /* Analytical PME correction */
1156             zeta2            = _mm256_mul_ps(beta2,rsq22);
1157             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1158             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1159             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1160             felec            = _mm256_mul_ps(qq22,felec);
1161             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1162             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1163             velec            = _mm256_sub_ps(rinv22,pmecorrV);
1164             velec            = _mm256_mul_ps(qq22,velec);
1165
1166             /* Update potential sum for this i atom from the interaction with this j atom. */
1167             velec            = _mm256_andnot_ps(dummy_mask,velec);
1168             velecsum         = _mm256_add_ps(velecsum,velec);
1169
1170             fscal            = felec;
1171
1172             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1173
1174             /* Calculate temporary vectorial force */
1175             tx               = _mm256_mul_ps(fscal,dx22);
1176             ty               = _mm256_mul_ps(fscal,dy22);
1177             tz               = _mm256_mul_ps(fscal,dz22);
1178
1179             /* Update vectorial force */
1180             fix2             = _mm256_add_ps(fix2,tx);
1181             fiy2             = _mm256_add_ps(fiy2,ty);
1182             fiz2             = _mm256_add_ps(fiz2,tz);
1183
1184             fjx2             = _mm256_add_ps(fjx2,tx);
1185             fjy2             = _mm256_add_ps(fjy2,ty);
1186             fjz2             = _mm256_add_ps(fjz2,tz);
1187
1188             /**************************
1189              * CALCULATE INTERACTIONS *
1190              **************************/
1191
1192             r23              = _mm256_mul_ps(rsq23,rinv23);
1193             r23              = _mm256_andnot_ps(dummy_mask,r23);
1194
1195             /* EWALD ELECTROSTATICS */
1196
1197             /* Analytical PME correction */
1198             zeta2            = _mm256_mul_ps(beta2,rsq23);
1199             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1200             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1201             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1202             felec            = _mm256_mul_ps(qq23,felec);
1203             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1204             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1205             velec            = _mm256_sub_ps(rinv23,pmecorrV);
1206             velec            = _mm256_mul_ps(qq23,velec);
1207
1208             /* Update potential sum for this i atom from the interaction with this j atom. */
1209             velec            = _mm256_andnot_ps(dummy_mask,velec);
1210             velecsum         = _mm256_add_ps(velecsum,velec);
1211
1212             fscal            = felec;
1213
1214             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1215
1216             /* Calculate temporary vectorial force */
1217             tx               = _mm256_mul_ps(fscal,dx23);
1218             ty               = _mm256_mul_ps(fscal,dy23);
1219             tz               = _mm256_mul_ps(fscal,dz23);
1220
1221             /* Update vectorial force */
1222             fix2             = _mm256_add_ps(fix2,tx);
1223             fiy2             = _mm256_add_ps(fiy2,ty);
1224             fiz2             = _mm256_add_ps(fiz2,tz);
1225
1226             fjx3             = _mm256_add_ps(fjx3,tx);
1227             fjy3             = _mm256_add_ps(fjy3,ty);
1228             fjz3             = _mm256_add_ps(fjz3,tz);
1229
1230             /**************************
1231              * CALCULATE INTERACTIONS *
1232              **************************/
1233
1234             r31              = _mm256_mul_ps(rsq31,rinv31);
1235             r31              = _mm256_andnot_ps(dummy_mask,r31);
1236
1237             /* EWALD ELECTROSTATICS */
1238
1239             /* Analytical PME correction */
1240             zeta2            = _mm256_mul_ps(beta2,rsq31);
1241             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1242             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1243             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1244             felec            = _mm256_mul_ps(qq31,felec);
1245             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1246             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1247             velec            = _mm256_sub_ps(rinv31,pmecorrV);
1248             velec            = _mm256_mul_ps(qq31,velec);
1249
1250             /* Update potential sum for this i atom from the interaction with this j atom. */
1251             velec            = _mm256_andnot_ps(dummy_mask,velec);
1252             velecsum         = _mm256_add_ps(velecsum,velec);
1253
1254             fscal            = felec;
1255
1256             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1257
1258             /* Calculate temporary vectorial force */
1259             tx               = _mm256_mul_ps(fscal,dx31);
1260             ty               = _mm256_mul_ps(fscal,dy31);
1261             tz               = _mm256_mul_ps(fscal,dz31);
1262
1263             /* Update vectorial force */
1264             fix3             = _mm256_add_ps(fix3,tx);
1265             fiy3             = _mm256_add_ps(fiy3,ty);
1266             fiz3             = _mm256_add_ps(fiz3,tz);
1267
1268             fjx1             = _mm256_add_ps(fjx1,tx);
1269             fjy1             = _mm256_add_ps(fjy1,ty);
1270             fjz1             = _mm256_add_ps(fjz1,tz);
1271
1272             /**************************
1273              * CALCULATE INTERACTIONS *
1274              **************************/
1275
1276             r32              = _mm256_mul_ps(rsq32,rinv32);
1277             r32              = _mm256_andnot_ps(dummy_mask,r32);
1278
1279             /* EWALD ELECTROSTATICS */
1280
1281             /* Analytical PME correction */
1282             zeta2            = _mm256_mul_ps(beta2,rsq32);
1283             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1284             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1285             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1286             felec            = _mm256_mul_ps(qq32,felec);
1287             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1288             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1289             velec            = _mm256_sub_ps(rinv32,pmecorrV);
1290             velec            = _mm256_mul_ps(qq32,velec);
1291
1292             /* Update potential sum for this i atom from the interaction with this j atom. */
1293             velec            = _mm256_andnot_ps(dummy_mask,velec);
1294             velecsum         = _mm256_add_ps(velecsum,velec);
1295
1296             fscal            = felec;
1297
1298             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1299
1300             /* Calculate temporary vectorial force */
1301             tx               = _mm256_mul_ps(fscal,dx32);
1302             ty               = _mm256_mul_ps(fscal,dy32);
1303             tz               = _mm256_mul_ps(fscal,dz32);
1304
1305             /* Update vectorial force */
1306             fix3             = _mm256_add_ps(fix3,tx);
1307             fiy3             = _mm256_add_ps(fiy3,ty);
1308             fiz3             = _mm256_add_ps(fiz3,tz);
1309
1310             fjx2             = _mm256_add_ps(fjx2,tx);
1311             fjy2             = _mm256_add_ps(fjy2,ty);
1312             fjz2             = _mm256_add_ps(fjz2,tz);
1313
1314             /**************************
1315              * CALCULATE INTERACTIONS *
1316              **************************/
1317
1318             r33              = _mm256_mul_ps(rsq33,rinv33);
1319             r33              = _mm256_andnot_ps(dummy_mask,r33);
1320
1321             /* EWALD ELECTROSTATICS */
1322
1323             /* Analytical PME correction */
1324             zeta2            = _mm256_mul_ps(beta2,rsq33);
1325             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
1326             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1327             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1328             felec            = _mm256_mul_ps(qq33,felec);
1329             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1330             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1331             velec            = _mm256_sub_ps(rinv33,pmecorrV);
1332             velec            = _mm256_mul_ps(qq33,velec);
1333
1334             /* Update potential sum for this i atom from the interaction with this j atom. */
1335             velec            = _mm256_andnot_ps(dummy_mask,velec);
1336             velecsum         = _mm256_add_ps(velecsum,velec);
1337
1338             fscal            = felec;
1339
1340             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1341
1342             /* Calculate temporary vectorial force */
1343             tx               = _mm256_mul_ps(fscal,dx33);
1344             ty               = _mm256_mul_ps(fscal,dy33);
1345             tz               = _mm256_mul_ps(fscal,dz33);
1346
1347             /* Update vectorial force */
1348             fix3             = _mm256_add_ps(fix3,tx);
1349             fiy3             = _mm256_add_ps(fiy3,ty);
1350             fiz3             = _mm256_add_ps(fiz3,tz);
1351
1352             fjx3             = _mm256_add_ps(fjx3,tx);
1353             fjy3             = _mm256_add_ps(fjy3,ty);
1354             fjz3             = _mm256_add_ps(fjz3,tz);
1355
1356             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1357             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1358             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1359             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1360             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1361             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1362             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1363             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1364
1365             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1366                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1367                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1368
1369             /* Inner loop uses 825 flops */
1370         }
1371
1372         /* End of innermost loop */
1373
1374         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1375                                                  f+i_coord_offset,fshift+i_shift_offset);
1376
1377         ggid                        = gid[iidx];
1378         /* Update potential energies */
1379         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1380         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1381
1382         /* Increment number of inner iterations */
1383         inneriter                  += j_index_end - j_index_start;
1384
1385         /* Outer loop uses 26 flops */
1386     }
1387
1388     /* Increment number of outer iterations */
1389     outeriter        += nri;
1390
1391     /* Update outer/inner flops */
1392
1393     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*825);
1394 }
1395 /*
1396  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4W4_F_avx_256_single
1397  * Electrostatics interaction: Ewald
1398  * VdW interaction:            CubicSplineTable
1399  * Geometry:                   Water4-Water4
1400  * Calculate force/pot:        Force
1401  */
1402 void
1403 nb_kernel_ElecEw_VdwCSTab_GeomW4W4_F_avx_256_single
1404                     (t_nblist                    * gmx_restrict       nlist,
1405                      rvec                        * gmx_restrict          xx,
1406                      rvec                        * gmx_restrict          ff,
1407                      t_forcerec                  * gmx_restrict          fr,
1408                      t_mdatoms                   * gmx_restrict     mdatoms,
1409                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1410                      t_nrnb                      * gmx_restrict        nrnb)
1411 {
1412     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1413      * just 0 for non-waters.
1414      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1415      * jnr indices corresponding to data put in the four positions in the SIMD register.
1416      */
1417     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1418     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1419     int              jnrA,jnrB,jnrC,jnrD;
1420     int              jnrE,jnrF,jnrG,jnrH;
1421     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1422     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1423     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1424     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1425     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1426     real             rcutoff_scalar;
1427     real             *shiftvec,*fshift,*x,*f;
1428     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1429     real             scratch[4*DIM];
1430     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1431     real *           vdwioffsetptr0;
1432     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1433     real *           vdwioffsetptr1;
1434     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1435     real *           vdwioffsetptr2;
1436     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1437     real *           vdwioffsetptr3;
1438     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1439     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1440     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1441     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1442     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1443     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1444     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1445     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1446     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1447     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1448     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1449     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1450     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1451     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1452     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1453     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1454     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1455     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1456     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1457     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1458     real             *charge;
1459     int              nvdwtype;
1460     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1461     int              *vdwtype;
1462     real             *vdwparam;
1463     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1464     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1465     __m256i          vfitab;
1466     __m128i          vfitab_lo,vfitab_hi;
1467     __m128i          ifour       = _mm_set1_epi32(4);
1468     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1469     real             *vftab;
1470     __m256i          ewitab;
1471     __m128i          ewitab_lo,ewitab_hi;
1472     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1473     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1474     real             *ewtab;
1475     __m256           dummy_mask,cutoff_mask;
1476     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1477     __m256           one     = _mm256_set1_ps(1.0);
1478     __m256           two     = _mm256_set1_ps(2.0);
1479     x                = xx[0];
1480     f                = ff[0];
1481
1482     nri              = nlist->nri;
1483     iinr             = nlist->iinr;
1484     jindex           = nlist->jindex;
1485     jjnr             = nlist->jjnr;
1486     shiftidx         = nlist->shift;
1487     gid              = nlist->gid;
1488     shiftvec         = fr->shift_vec[0];
1489     fshift           = fr->fshift[0];
1490     facel            = _mm256_set1_ps(fr->epsfac);
1491     charge           = mdatoms->chargeA;
1492     nvdwtype         = fr->ntype;
1493     vdwparam         = fr->nbfp;
1494     vdwtype          = mdatoms->typeA;
1495
1496     vftab            = kernel_data->table_vdw->data;
1497     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
1498
1499     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1500     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1501     beta2            = _mm256_mul_ps(beta,beta);
1502     beta3            = _mm256_mul_ps(beta,beta2);
1503
1504     ewtab            = fr->ic->tabq_coul_F;
1505     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1506     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1507
1508     /* Setup water-specific parameters */
1509     inr              = nlist->iinr[0];
1510     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1511     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1512     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1513     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1514
1515     jq1              = _mm256_set1_ps(charge[inr+1]);
1516     jq2              = _mm256_set1_ps(charge[inr+2]);
1517     jq3              = _mm256_set1_ps(charge[inr+3]);
1518     vdwjidx0A        = 2*vdwtype[inr+0];
1519     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1520     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1521     qq11             = _mm256_mul_ps(iq1,jq1);
1522     qq12             = _mm256_mul_ps(iq1,jq2);
1523     qq13             = _mm256_mul_ps(iq1,jq3);
1524     qq21             = _mm256_mul_ps(iq2,jq1);
1525     qq22             = _mm256_mul_ps(iq2,jq2);
1526     qq23             = _mm256_mul_ps(iq2,jq3);
1527     qq31             = _mm256_mul_ps(iq3,jq1);
1528     qq32             = _mm256_mul_ps(iq3,jq2);
1529     qq33             = _mm256_mul_ps(iq3,jq3);
1530
1531     /* Avoid stupid compiler warnings */
1532     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1533     j_coord_offsetA = 0;
1534     j_coord_offsetB = 0;
1535     j_coord_offsetC = 0;
1536     j_coord_offsetD = 0;
1537     j_coord_offsetE = 0;
1538     j_coord_offsetF = 0;
1539     j_coord_offsetG = 0;
1540     j_coord_offsetH = 0;
1541
1542     outeriter        = 0;
1543     inneriter        = 0;
1544
1545     for(iidx=0;iidx<4*DIM;iidx++)
1546     {
1547         scratch[iidx] = 0.0;
1548     }
1549
1550     /* Start outer loop over neighborlists */
1551     for(iidx=0; iidx<nri; iidx++)
1552     {
1553         /* Load shift vector for this list */
1554         i_shift_offset   = DIM*shiftidx[iidx];
1555
1556         /* Load limits for loop over neighbors */
1557         j_index_start    = jindex[iidx];
1558         j_index_end      = jindex[iidx+1];
1559
1560         /* Get outer coordinate index */
1561         inr              = iinr[iidx];
1562         i_coord_offset   = DIM*inr;
1563
1564         /* Load i particle coords and add shift vector */
1565         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1566                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1567
1568         fix0             = _mm256_setzero_ps();
1569         fiy0             = _mm256_setzero_ps();
1570         fiz0             = _mm256_setzero_ps();
1571         fix1             = _mm256_setzero_ps();
1572         fiy1             = _mm256_setzero_ps();
1573         fiz1             = _mm256_setzero_ps();
1574         fix2             = _mm256_setzero_ps();
1575         fiy2             = _mm256_setzero_ps();
1576         fiz2             = _mm256_setzero_ps();
1577         fix3             = _mm256_setzero_ps();
1578         fiy3             = _mm256_setzero_ps();
1579         fiz3             = _mm256_setzero_ps();
1580
1581         /* Start inner kernel loop */
1582         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1583         {
1584
1585             /* Get j neighbor index, and coordinate index */
1586             jnrA             = jjnr[jidx];
1587             jnrB             = jjnr[jidx+1];
1588             jnrC             = jjnr[jidx+2];
1589             jnrD             = jjnr[jidx+3];
1590             jnrE             = jjnr[jidx+4];
1591             jnrF             = jjnr[jidx+5];
1592             jnrG             = jjnr[jidx+6];
1593             jnrH             = jjnr[jidx+7];
1594             j_coord_offsetA  = DIM*jnrA;
1595             j_coord_offsetB  = DIM*jnrB;
1596             j_coord_offsetC  = DIM*jnrC;
1597             j_coord_offsetD  = DIM*jnrD;
1598             j_coord_offsetE  = DIM*jnrE;
1599             j_coord_offsetF  = DIM*jnrF;
1600             j_coord_offsetG  = DIM*jnrG;
1601             j_coord_offsetH  = DIM*jnrH;
1602
1603             /* load j atom coordinates */
1604             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1605                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1606                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1607                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1608                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1609                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1610
1611             /* Calculate displacement vector */
1612             dx00             = _mm256_sub_ps(ix0,jx0);
1613             dy00             = _mm256_sub_ps(iy0,jy0);
1614             dz00             = _mm256_sub_ps(iz0,jz0);
1615             dx11             = _mm256_sub_ps(ix1,jx1);
1616             dy11             = _mm256_sub_ps(iy1,jy1);
1617             dz11             = _mm256_sub_ps(iz1,jz1);
1618             dx12             = _mm256_sub_ps(ix1,jx2);
1619             dy12             = _mm256_sub_ps(iy1,jy2);
1620             dz12             = _mm256_sub_ps(iz1,jz2);
1621             dx13             = _mm256_sub_ps(ix1,jx3);
1622             dy13             = _mm256_sub_ps(iy1,jy3);
1623             dz13             = _mm256_sub_ps(iz1,jz3);
1624             dx21             = _mm256_sub_ps(ix2,jx1);
1625             dy21             = _mm256_sub_ps(iy2,jy1);
1626             dz21             = _mm256_sub_ps(iz2,jz1);
1627             dx22             = _mm256_sub_ps(ix2,jx2);
1628             dy22             = _mm256_sub_ps(iy2,jy2);
1629             dz22             = _mm256_sub_ps(iz2,jz2);
1630             dx23             = _mm256_sub_ps(ix2,jx3);
1631             dy23             = _mm256_sub_ps(iy2,jy3);
1632             dz23             = _mm256_sub_ps(iz2,jz3);
1633             dx31             = _mm256_sub_ps(ix3,jx1);
1634             dy31             = _mm256_sub_ps(iy3,jy1);
1635             dz31             = _mm256_sub_ps(iz3,jz1);
1636             dx32             = _mm256_sub_ps(ix3,jx2);
1637             dy32             = _mm256_sub_ps(iy3,jy2);
1638             dz32             = _mm256_sub_ps(iz3,jz2);
1639             dx33             = _mm256_sub_ps(ix3,jx3);
1640             dy33             = _mm256_sub_ps(iy3,jy3);
1641             dz33             = _mm256_sub_ps(iz3,jz3);
1642
1643             /* Calculate squared distance and things based on it */
1644             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1645             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1646             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1647             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1648             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1649             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1650             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1651             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1652             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1653             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1654
1655             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1656             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1657             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1658             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
1659             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1660             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1661             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
1662             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
1663             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
1664             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
1665
1666             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1667             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1668             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1669             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1670             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1671             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1672             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1673             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1674             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1675
1676             fjx0             = _mm256_setzero_ps();
1677             fjy0             = _mm256_setzero_ps();
1678             fjz0             = _mm256_setzero_ps();
1679             fjx1             = _mm256_setzero_ps();
1680             fjy1             = _mm256_setzero_ps();
1681             fjz1             = _mm256_setzero_ps();
1682             fjx2             = _mm256_setzero_ps();
1683             fjy2             = _mm256_setzero_ps();
1684             fjz2             = _mm256_setzero_ps();
1685             fjx3             = _mm256_setzero_ps();
1686             fjy3             = _mm256_setzero_ps();
1687             fjz3             = _mm256_setzero_ps();
1688
1689             /**************************
1690              * CALCULATE INTERACTIONS *
1691              **************************/
1692
1693             r00              = _mm256_mul_ps(rsq00,rinv00);
1694
1695             /* Calculate table index by multiplying r with table scale and truncate to integer */
1696             rt               = _mm256_mul_ps(r00,vftabscale);
1697             vfitab           = _mm256_cvttps_epi32(rt);
1698             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1699             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1700             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1701             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1702             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1703             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1704
1705             /* CUBIC SPLINE TABLE DISPERSION */
1706             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1707                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1708             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1709                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1710             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1711                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1712             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1713                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1714             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1715             Heps             = _mm256_mul_ps(vfeps,H);
1716             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1717             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1718             fvdw6            = _mm256_mul_ps(c6_00,FF);
1719
1720             /* CUBIC SPLINE TABLE REPULSION */
1721             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1722             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1723             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1724                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1725             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1726                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1727             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1728                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1729             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1730                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1731             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1732             Heps             = _mm256_mul_ps(vfeps,H);
1733             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1734             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1735             fvdw12           = _mm256_mul_ps(c12_00,FF);
1736             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1737
1738             fscal            = fvdw;
1739
1740             /* Calculate temporary vectorial force */
1741             tx               = _mm256_mul_ps(fscal,dx00);
1742             ty               = _mm256_mul_ps(fscal,dy00);
1743             tz               = _mm256_mul_ps(fscal,dz00);
1744
1745             /* Update vectorial force */
1746             fix0             = _mm256_add_ps(fix0,tx);
1747             fiy0             = _mm256_add_ps(fiy0,ty);
1748             fiz0             = _mm256_add_ps(fiz0,tz);
1749
1750             fjx0             = _mm256_add_ps(fjx0,tx);
1751             fjy0             = _mm256_add_ps(fjy0,ty);
1752             fjz0             = _mm256_add_ps(fjz0,tz);
1753
1754             /**************************
1755              * CALCULATE INTERACTIONS *
1756              **************************/
1757
1758             r11              = _mm256_mul_ps(rsq11,rinv11);
1759
1760             /* EWALD ELECTROSTATICS */
1761
1762             /* Analytical PME correction */
1763             zeta2            = _mm256_mul_ps(beta2,rsq11);
1764             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1765             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1766             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1767             felec            = _mm256_mul_ps(qq11,felec);
1768
1769             fscal            = felec;
1770
1771             /* Calculate temporary vectorial force */
1772             tx               = _mm256_mul_ps(fscal,dx11);
1773             ty               = _mm256_mul_ps(fscal,dy11);
1774             tz               = _mm256_mul_ps(fscal,dz11);
1775
1776             /* Update vectorial force */
1777             fix1             = _mm256_add_ps(fix1,tx);
1778             fiy1             = _mm256_add_ps(fiy1,ty);
1779             fiz1             = _mm256_add_ps(fiz1,tz);
1780
1781             fjx1             = _mm256_add_ps(fjx1,tx);
1782             fjy1             = _mm256_add_ps(fjy1,ty);
1783             fjz1             = _mm256_add_ps(fjz1,tz);
1784
1785             /**************************
1786              * CALCULATE INTERACTIONS *
1787              **************************/
1788
1789             r12              = _mm256_mul_ps(rsq12,rinv12);
1790
1791             /* EWALD ELECTROSTATICS */
1792
1793             /* Analytical PME correction */
1794             zeta2            = _mm256_mul_ps(beta2,rsq12);
1795             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1796             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1797             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1798             felec            = _mm256_mul_ps(qq12,felec);
1799
1800             fscal            = felec;
1801
1802             /* Calculate temporary vectorial force */
1803             tx               = _mm256_mul_ps(fscal,dx12);
1804             ty               = _mm256_mul_ps(fscal,dy12);
1805             tz               = _mm256_mul_ps(fscal,dz12);
1806
1807             /* Update vectorial force */
1808             fix1             = _mm256_add_ps(fix1,tx);
1809             fiy1             = _mm256_add_ps(fiy1,ty);
1810             fiz1             = _mm256_add_ps(fiz1,tz);
1811
1812             fjx2             = _mm256_add_ps(fjx2,tx);
1813             fjy2             = _mm256_add_ps(fjy2,ty);
1814             fjz2             = _mm256_add_ps(fjz2,tz);
1815
1816             /**************************
1817              * CALCULATE INTERACTIONS *
1818              **************************/
1819
1820             r13              = _mm256_mul_ps(rsq13,rinv13);
1821
1822             /* EWALD ELECTROSTATICS */
1823
1824             /* Analytical PME correction */
1825             zeta2            = _mm256_mul_ps(beta2,rsq13);
1826             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
1827             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1828             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1829             felec            = _mm256_mul_ps(qq13,felec);
1830
1831             fscal            = felec;
1832
1833             /* Calculate temporary vectorial force */
1834             tx               = _mm256_mul_ps(fscal,dx13);
1835             ty               = _mm256_mul_ps(fscal,dy13);
1836             tz               = _mm256_mul_ps(fscal,dz13);
1837
1838             /* Update vectorial force */
1839             fix1             = _mm256_add_ps(fix1,tx);
1840             fiy1             = _mm256_add_ps(fiy1,ty);
1841             fiz1             = _mm256_add_ps(fiz1,tz);
1842
1843             fjx3             = _mm256_add_ps(fjx3,tx);
1844             fjy3             = _mm256_add_ps(fjy3,ty);
1845             fjz3             = _mm256_add_ps(fjz3,tz);
1846
1847             /**************************
1848              * CALCULATE INTERACTIONS *
1849              **************************/
1850
1851             r21              = _mm256_mul_ps(rsq21,rinv21);
1852
1853             /* EWALD ELECTROSTATICS */
1854
1855             /* Analytical PME correction */
1856             zeta2            = _mm256_mul_ps(beta2,rsq21);
1857             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1858             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1859             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1860             felec            = _mm256_mul_ps(qq21,felec);
1861
1862             fscal            = felec;
1863
1864             /* Calculate temporary vectorial force */
1865             tx               = _mm256_mul_ps(fscal,dx21);
1866             ty               = _mm256_mul_ps(fscal,dy21);
1867             tz               = _mm256_mul_ps(fscal,dz21);
1868
1869             /* Update vectorial force */
1870             fix2             = _mm256_add_ps(fix2,tx);
1871             fiy2             = _mm256_add_ps(fiy2,ty);
1872             fiz2             = _mm256_add_ps(fiz2,tz);
1873
1874             fjx1             = _mm256_add_ps(fjx1,tx);
1875             fjy1             = _mm256_add_ps(fjy1,ty);
1876             fjz1             = _mm256_add_ps(fjz1,tz);
1877
1878             /**************************
1879              * CALCULATE INTERACTIONS *
1880              **************************/
1881
1882             r22              = _mm256_mul_ps(rsq22,rinv22);
1883
1884             /* EWALD ELECTROSTATICS */
1885
1886             /* Analytical PME correction */
1887             zeta2            = _mm256_mul_ps(beta2,rsq22);
1888             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1889             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1890             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1891             felec            = _mm256_mul_ps(qq22,felec);
1892
1893             fscal            = felec;
1894
1895             /* Calculate temporary vectorial force */
1896             tx               = _mm256_mul_ps(fscal,dx22);
1897             ty               = _mm256_mul_ps(fscal,dy22);
1898             tz               = _mm256_mul_ps(fscal,dz22);
1899
1900             /* Update vectorial force */
1901             fix2             = _mm256_add_ps(fix2,tx);
1902             fiy2             = _mm256_add_ps(fiy2,ty);
1903             fiz2             = _mm256_add_ps(fiz2,tz);
1904
1905             fjx2             = _mm256_add_ps(fjx2,tx);
1906             fjy2             = _mm256_add_ps(fjy2,ty);
1907             fjz2             = _mm256_add_ps(fjz2,tz);
1908
1909             /**************************
1910              * CALCULATE INTERACTIONS *
1911              **************************/
1912
1913             r23              = _mm256_mul_ps(rsq23,rinv23);
1914
1915             /* EWALD ELECTROSTATICS */
1916
1917             /* Analytical PME correction */
1918             zeta2            = _mm256_mul_ps(beta2,rsq23);
1919             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1920             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1921             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1922             felec            = _mm256_mul_ps(qq23,felec);
1923
1924             fscal            = felec;
1925
1926             /* Calculate temporary vectorial force */
1927             tx               = _mm256_mul_ps(fscal,dx23);
1928             ty               = _mm256_mul_ps(fscal,dy23);
1929             tz               = _mm256_mul_ps(fscal,dz23);
1930
1931             /* Update vectorial force */
1932             fix2             = _mm256_add_ps(fix2,tx);
1933             fiy2             = _mm256_add_ps(fiy2,ty);
1934             fiz2             = _mm256_add_ps(fiz2,tz);
1935
1936             fjx3             = _mm256_add_ps(fjx3,tx);
1937             fjy3             = _mm256_add_ps(fjy3,ty);
1938             fjz3             = _mm256_add_ps(fjz3,tz);
1939
1940             /**************************
1941              * CALCULATE INTERACTIONS *
1942              **************************/
1943
1944             r31              = _mm256_mul_ps(rsq31,rinv31);
1945
1946             /* EWALD ELECTROSTATICS */
1947
1948             /* Analytical PME correction */
1949             zeta2            = _mm256_mul_ps(beta2,rsq31);
1950             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1951             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1952             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1953             felec            = _mm256_mul_ps(qq31,felec);
1954
1955             fscal            = felec;
1956
1957             /* Calculate temporary vectorial force */
1958             tx               = _mm256_mul_ps(fscal,dx31);
1959             ty               = _mm256_mul_ps(fscal,dy31);
1960             tz               = _mm256_mul_ps(fscal,dz31);
1961
1962             /* Update vectorial force */
1963             fix3             = _mm256_add_ps(fix3,tx);
1964             fiy3             = _mm256_add_ps(fiy3,ty);
1965             fiz3             = _mm256_add_ps(fiz3,tz);
1966
1967             fjx1             = _mm256_add_ps(fjx1,tx);
1968             fjy1             = _mm256_add_ps(fjy1,ty);
1969             fjz1             = _mm256_add_ps(fjz1,tz);
1970
1971             /**************************
1972              * CALCULATE INTERACTIONS *
1973              **************************/
1974
1975             r32              = _mm256_mul_ps(rsq32,rinv32);
1976
1977             /* EWALD ELECTROSTATICS */
1978
1979             /* Analytical PME correction */
1980             zeta2            = _mm256_mul_ps(beta2,rsq32);
1981             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1982             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1983             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1984             felec            = _mm256_mul_ps(qq32,felec);
1985
1986             fscal            = felec;
1987
1988             /* Calculate temporary vectorial force */
1989             tx               = _mm256_mul_ps(fscal,dx32);
1990             ty               = _mm256_mul_ps(fscal,dy32);
1991             tz               = _mm256_mul_ps(fscal,dz32);
1992
1993             /* Update vectorial force */
1994             fix3             = _mm256_add_ps(fix3,tx);
1995             fiy3             = _mm256_add_ps(fiy3,ty);
1996             fiz3             = _mm256_add_ps(fiz3,tz);
1997
1998             fjx2             = _mm256_add_ps(fjx2,tx);
1999             fjy2             = _mm256_add_ps(fjy2,ty);
2000             fjz2             = _mm256_add_ps(fjz2,tz);
2001
2002             /**************************
2003              * CALCULATE INTERACTIONS *
2004              **************************/
2005
2006             r33              = _mm256_mul_ps(rsq33,rinv33);
2007
2008             /* EWALD ELECTROSTATICS */
2009
2010             /* Analytical PME correction */
2011             zeta2            = _mm256_mul_ps(beta2,rsq33);
2012             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
2013             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2014             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2015             felec            = _mm256_mul_ps(qq33,felec);
2016
2017             fscal            = felec;
2018
2019             /* Calculate temporary vectorial force */
2020             tx               = _mm256_mul_ps(fscal,dx33);
2021             ty               = _mm256_mul_ps(fscal,dy33);
2022             tz               = _mm256_mul_ps(fscal,dz33);
2023
2024             /* Update vectorial force */
2025             fix3             = _mm256_add_ps(fix3,tx);
2026             fiy3             = _mm256_add_ps(fiy3,ty);
2027             fiz3             = _mm256_add_ps(fiz3,tz);
2028
2029             fjx3             = _mm256_add_ps(fjx3,tx);
2030             fjy3             = _mm256_add_ps(fjy3,ty);
2031             fjz3             = _mm256_add_ps(fjz3,tz);
2032
2033             fjptrA             = f+j_coord_offsetA;
2034             fjptrB             = f+j_coord_offsetB;
2035             fjptrC             = f+j_coord_offsetC;
2036             fjptrD             = f+j_coord_offsetD;
2037             fjptrE             = f+j_coord_offsetE;
2038             fjptrF             = f+j_coord_offsetF;
2039             fjptrG             = f+j_coord_offsetG;
2040             fjptrH             = f+j_coord_offsetH;
2041
2042             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2043                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2044                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2045
2046             /* Inner loop uses 555 flops */
2047         }
2048
2049         if(jidx<j_index_end)
2050         {
2051
2052             /* Get j neighbor index, and coordinate index */
2053             jnrlistA         = jjnr[jidx];
2054             jnrlistB         = jjnr[jidx+1];
2055             jnrlistC         = jjnr[jidx+2];
2056             jnrlistD         = jjnr[jidx+3];
2057             jnrlistE         = jjnr[jidx+4];
2058             jnrlistF         = jjnr[jidx+5];
2059             jnrlistG         = jjnr[jidx+6];
2060             jnrlistH         = jjnr[jidx+7];
2061             /* Sign of each element will be negative for non-real atoms.
2062              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2063              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2064              */
2065             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2066                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2067
2068             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2069             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2070             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2071             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2072             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2073             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2074             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2075             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2076             j_coord_offsetA  = DIM*jnrA;
2077             j_coord_offsetB  = DIM*jnrB;
2078             j_coord_offsetC  = DIM*jnrC;
2079             j_coord_offsetD  = DIM*jnrD;
2080             j_coord_offsetE  = DIM*jnrE;
2081             j_coord_offsetF  = DIM*jnrF;
2082             j_coord_offsetG  = DIM*jnrG;
2083             j_coord_offsetH  = DIM*jnrH;
2084
2085             /* load j atom coordinates */
2086             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2087                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2088                                                  x+j_coord_offsetE,x+j_coord_offsetF,
2089                                                  x+j_coord_offsetG,x+j_coord_offsetH,
2090                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2091                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
2092
2093             /* Calculate displacement vector */
2094             dx00             = _mm256_sub_ps(ix0,jx0);
2095             dy00             = _mm256_sub_ps(iy0,jy0);
2096             dz00             = _mm256_sub_ps(iz0,jz0);
2097             dx11             = _mm256_sub_ps(ix1,jx1);
2098             dy11             = _mm256_sub_ps(iy1,jy1);
2099             dz11             = _mm256_sub_ps(iz1,jz1);
2100             dx12             = _mm256_sub_ps(ix1,jx2);
2101             dy12             = _mm256_sub_ps(iy1,jy2);
2102             dz12             = _mm256_sub_ps(iz1,jz2);
2103             dx13             = _mm256_sub_ps(ix1,jx3);
2104             dy13             = _mm256_sub_ps(iy1,jy3);
2105             dz13             = _mm256_sub_ps(iz1,jz3);
2106             dx21             = _mm256_sub_ps(ix2,jx1);
2107             dy21             = _mm256_sub_ps(iy2,jy1);
2108             dz21             = _mm256_sub_ps(iz2,jz1);
2109             dx22             = _mm256_sub_ps(ix2,jx2);
2110             dy22             = _mm256_sub_ps(iy2,jy2);
2111             dz22             = _mm256_sub_ps(iz2,jz2);
2112             dx23             = _mm256_sub_ps(ix2,jx3);
2113             dy23             = _mm256_sub_ps(iy2,jy3);
2114             dz23             = _mm256_sub_ps(iz2,jz3);
2115             dx31             = _mm256_sub_ps(ix3,jx1);
2116             dy31             = _mm256_sub_ps(iy3,jy1);
2117             dz31             = _mm256_sub_ps(iz3,jz1);
2118             dx32             = _mm256_sub_ps(ix3,jx2);
2119             dy32             = _mm256_sub_ps(iy3,jy2);
2120             dz32             = _mm256_sub_ps(iz3,jz2);
2121             dx33             = _mm256_sub_ps(ix3,jx3);
2122             dy33             = _mm256_sub_ps(iy3,jy3);
2123             dz33             = _mm256_sub_ps(iz3,jz3);
2124
2125             /* Calculate squared distance and things based on it */
2126             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2127             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2128             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2129             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
2130             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2131             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2132             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
2133             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
2134             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
2135             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
2136
2137             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
2138             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2139             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2140             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
2141             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2142             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2143             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
2144             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
2145             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
2146             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
2147
2148             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2149             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2150             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
2151             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2152             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2153             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
2154             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
2155             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
2156             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
2157
2158             fjx0             = _mm256_setzero_ps();
2159             fjy0             = _mm256_setzero_ps();
2160             fjz0             = _mm256_setzero_ps();
2161             fjx1             = _mm256_setzero_ps();
2162             fjy1             = _mm256_setzero_ps();
2163             fjz1             = _mm256_setzero_ps();
2164             fjx2             = _mm256_setzero_ps();
2165             fjy2             = _mm256_setzero_ps();
2166             fjz2             = _mm256_setzero_ps();
2167             fjx3             = _mm256_setzero_ps();
2168             fjy3             = _mm256_setzero_ps();
2169             fjz3             = _mm256_setzero_ps();
2170
2171             /**************************
2172              * CALCULATE INTERACTIONS *
2173              **************************/
2174
2175             r00              = _mm256_mul_ps(rsq00,rinv00);
2176             r00              = _mm256_andnot_ps(dummy_mask,r00);
2177
2178             /* Calculate table index by multiplying r with table scale and truncate to integer */
2179             rt               = _mm256_mul_ps(r00,vftabscale);
2180             vfitab           = _mm256_cvttps_epi32(rt);
2181             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2182             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2183             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2184             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2185             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
2186             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
2187
2188             /* CUBIC SPLINE TABLE DISPERSION */
2189             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2190                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2191             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2192                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2193             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2194                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2195             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2196                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2197             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2198             Heps             = _mm256_mul_ps(vfeps,H);
2199             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2200             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2201             fvdw6            = _mm256_mul_ps(c6_00,FF);
2202
2203             /* CUBIC SPLINE TABLE REPULSION */
2204             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
2205             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
2206             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2207                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2208             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2209                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2210             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2211                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2212             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2213                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2214             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2215             Heps             = _mm256_mul_ps(vfeps,H);
2216             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2217             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2218             fvdw12           = _mm256_mul_ps(c12_00,FF);
2219             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
2220
2221             fscal            = fvdw;
2222
2223             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2224
2225             /* Calculate temporary vectorial force */
2226             tx               = _mm256_mul_ps(fscal,dx00);
2227             ty               = _mm256_mul_ps(fscal,dy00);
2228             tz               = _mm256_mul_ps(fscal,dz00);
2229
2230             /* Update vectorial force */
2231             fix0             = _mm256_add_ps(fix0,tx);
2232             fiy0             = _mm256_add_ps(fiy0,ty);
2233             fiz0             = _mm256_add_ps(fiz0,tz);
2234
2235             fjx0             = _mm256_add_ps(fjx0,tx);
2236             fjy0             = _mm256_add_ps(fjy0,ty);
2237             fjz0             = _mm256_add_ps(fjz0,tz);
2238
2239             /**************************
2240              * CALCULATE INTERACTIONS *
2241              **************************/
2242
2243             r11              = _mm256_mul_ps(rsq11,rinv11);
2244             r11              = _mm256_andnot_ps(dummy_mask,r11);
2245
2246             /* EWALD ELECTROSTATICS */
2247
2248             /* Analytical PME correction */
2249             zeta2            = _mm256_mul_ps(beta2,rsq11);
2250             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
2251             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2252             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2253             felec            = _mm256_mul_ps(qq11,felec);
2254
2255             fscal            = felec;
2256
2257             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2258
2259             /* Calculate temporary vectorial force */
2260             tx               = _mm256_mul_ps(fscal,dx11);
2261             ty               = _mm256_mul_ps(fscal,dy11);
2262             tz               = _mm256_mul_ps(fscal,dz11);
2263
2264             /* Update vectorial force */
2265             fix1             = _mm256_add_ps(fix1,tx);
2266             fiy1             = _mm256_add_ps(fiy1,ty);
2267             fiz1             = _mm256_add_ps(fiz1,tz);
2268
2269             fjx1             = _mm256_add_ps(fjx1,tx);
2270             fjy1             = _mm256_add_ps(fjy1,ty);
2271             fjz1             = _mm256_add_ps(fjz1,tz);
2272
2273             /**************************
2274              * CALCULATE INTERACTIONS *
2275              **************************/
2276
2277             r12              = _mm256_mul_ps(rsq12,rinv12);
2278             r12              = _mm256_andnot_ps(dummy_mask,r12);
2279
2280             /* EWALD ELECTROSTATICS */
2281
2282             /* Analytical PME correction */
2283             zeta2            = _mm256_mul_ps(beta2,rsq12);
2284             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2285             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2286             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2287             felec            = _mm256_mul_ps(qq12,felec);
2288
2289             fscal            = felec;
2290
2291             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2292
2293             /* Calculate temporary vectorial force */
2294             tx               = _mm256_mul_ps(fscal,dx12);
2295             ty               = _mm256_mul_ps(fscal,dy12);
2296             tz               = _mm256_mul_ps(fscal,dz12);
2297
2298             /* Update vectorial force */
2299             fix1             = _mm256_add_ps(fix1,tx);
2300             fiy1             = _mm256_add_ps(fiy1,ty);
2301             fiz1             = _mm256_add_ps(fiz1,tz);
2302
2303             fjx2             = _mm256_add_ps(fjx2,tx);
2304             fjy2             = _mm256_add_ps(fjy2,ty);
2305             fjz2             = _mm256_add_ps(fjz2,tz);
2306
2307             /**************************
2308              * CALCULATE INTERACTIONS *
2309              **************************/
2310
2311             r13              = _mm256_mul_ps(rsq13,rinv13);
2312             r13              = _mm256_andnot_ps(dummy_mask,r13);
2313
2314             /* EWALD ELECTROSTATICS */
2315
2316             /* Analytical PME correction */
2317             zeta2            = _mm256_mul_ps(beta2,rsq13);
2318             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
2319             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2320             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2321             felec            = _mm256_mul_ps(qq13,felec);
2322
2323             fscal            = felec;
2324
2325             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2326
2327             /* Calculate temporary vectorial force */
2328             tx               = _mm256_mul_ps(fscal,dx13);
2329             ty               = _mm256_mul_ps(fscal,dy13);
2330             tz               = _mm256_mul_ps(fscal,dz13);
2331
2332             /* Update vectorial force */
2333             fix1             = _mm256_add_ps(fix1,tx);
2334             fiy1             = _mm256_add_ps(fiy1,ty);
2335             fiz1             = _mm256_add_ps(fiz1,tz);
2336
2337             fjx3             = _mm256_add_ps(fjx3,tx);
2338             fjy3             = _mm256_add_ps(fjy3,ty);
2339             fjz3             = _mm256_add_ps(fjz3,tz);
2340
2341             /**************************
2342              * CALCULATE INTERACTIONS *
2343              **************************/
2344
2345             r21              = _mm256_mul_ps(rsq21,rinv21);
2346             r21              = _mm256_andnot_ps(dummy_mask,r21);
2347
2348             /* EWALD ELECTROSTATICS */
2349
2350             /* Analytical PME correction */
2351             zeta2            = _mm256_mul_ps(beta2,rsq21);
2352             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2353             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2354             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2355             felec            = _mm256_mul_ps(qq21,felec);
2356
2357             fscal            = felec;
2358
2359             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2360
2361             /* Calculate temporary vectorial force */
2362             tx               = _mm256_mul_ps(fscal,dx21);
2363             ty               = _mm256_mul_ps(fscal,dy21);
2364             tz               = _mm256_mul_ps(fscal,dz21);
2365
2366             /* Update vectorial force */
2367             fix2             = _mm256_add_ps(fix2,tx);
2368             fiy2             = _mm256_add_ps(fiy2,ty);
2369             fiz2             = _mm256_add_ps(fiz2,tz);
2370
2371             fjx1             = _mm256_add_ps(fjx1,tx);
2372             fjy1             = _mm256_add_ps(fjy1,ty);
2373             fjz1             = _mm256_add_ps(fjz1,tz);
2374
2375             /**************************
2376              * CALCULATE INTERACTIONS *
2377              **************************/
2378
2379             r22              = _mm256_mul_ps(rsq22,rinv22);
2380             r22              = _mm256_andnot_ps(dummy_mask,r22);
2381
2382             /* EWALD ELECTROSTATICS */
2383
2384             /* Analytical PME correction */
2385             zeta2            = _mm256_mul_ps(beta2,rsq22);
2386             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2387             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2388             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2389             felec            = _mm256_mul_ps(qq22,felec);
2390
2391             fscal            = felec;
2392
2393             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2394
2395             /* Calculate temporary vectorial force */
2396             tx               = _mm256_mul_ps(fscal,dx22);
2397             ty               = _mm256_mul_ps(fscal,dy22);
2398             tz               = _mm256_mul_ps(fscal,dz22);
2399
2400             /* Update vectorial force */
2401             fix2             = _mm256_add_ps(fix2,tx);
2402             fiy2             = _mm256_add_ps(fiy2,ty);
2403             fiz2             = _mm256_add_ps(fiz2,tz);
2404
2405             fjx2             = _mm256_add_ps(fjx2,tx);
2406             fjy2             = _mm256_add_ps(fjy2,ty);
2407             fjz2             = _mm256_add_ps(fjz2,tz);
2408
2409             /**************************
2410              * CALCULATE INTERACTIONS *
2411              **************************/
2412
2413             r23              = _mm256_mul_ps(rsq23,rinv23);
2414             r23              = _mm256_andnot_ps(dummy_mask,r23);
2415
2416             /* EWALD ELECTROSTATICS */
2417
2418             /* Analytical PME correction */
2419             zeta2            = _mm256_mul_ps(beta2,rsq23);
2420             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
2421             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2422             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2423             felec            = _mm256_mul_ps(qq23,felec);
2424
2425             fscal            = felec;
2426
2427             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2428
2429             /* Calculate temporary vectorial force */
2430             tx               = _mm256_mul_ps(fscal,dx23);
2431             ty               = _mm256_mul_ps(fscal,dy23);
2432             tz               = _mm256_mul_ps(fscal,dz23);
2433
2434             /* Update vectorial force */
2435             fix2             = _mm256_add_ps(fix2,tx);
2436             fiy2             = _mm256_add_ps(fiy2,ty);
2437             fiz2             = _mm256_add_ps(fiz2,tz);
2438
2439             fjx3             = _mm256_add_ps(fjx3,tx);
2440             fjy3             = _mm256_add_ps(fjy3,ty);
2441             fjz3             = _mm256_add_ps(fjz3,tz);
2442
2443             /**************************
2444              * CALCULATE INTERACTIONS *
2445              **************************/
2446
2447             r31              = _mm256_mul_ps(rsq31,rinv31);
2448             r31              = _mm256_andnot_ps(dummy_mask,r31);
2449
2450             /* EWALD ELECTROSTATICS */
2451
2452             /* Analytical PME correction */
2453             zeta2            = _mm256_mul_ps(beta2,rsq31);
2454             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
2455             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2456             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2457             felec            = _mm256_mul_ps(qq31,felec);
2458
2459             fscal            = felec;
2460
2461             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2462
2463             /* Calculate temporary vectorial force */
2464             tx               = _mm256_mul_ps(fscal,dx31);
2465             ty               = _mm256_mul_ps(fscal,dy31);
2466             tz               = _mm256_mul_ps(fscal,dz31);
2467
2468             /* Update vectorial force */
2469             fix3             = _mm256_add_ps(fix3,tx);
2470             fiy3             = _mm256_add_ps(fiy3,ty);
2471             fiz3             = _mm256_add_ps(fiz3,tz);
2472
2473             fjx1             = _mm256_add_ps(fjx1,tx);
2474             fjy1             = _mm256_add_ps(fjy1,ty);
2475             fjz1             = _mm256_add_ps(fjz1,tz);
2476
2477             /**************************
2478              * CALCULATE INTERACTIONS *
2479              **************************/
2480
2481             r32              = _mm256_mul_ps(rsq32,rinv32);
2482             r32              = _mm256_andnot_ps(dummy_mask,r32);
2483
2484             /* EWALD ELECTROSTATICS */
2485
2486             /* Analytical PME correction */
2487             zeta2            = _mm256_mul_ps(beta2,rsq32);
2488             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
2489             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2490             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2491             felec            = _mm256_mul_ps(qq32,felec);
2492
2493             fscal            = felec;
2494
2495             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2496
2497             /* Calculate temporary vectorial force */
2498             tx               = _mm256_mul_ps(fscal,dx32);
2499             ty               = _mm256_mul_ps(fscal,dy32);
2500             tz               = _mm256_mul_ps(fscal,dz32);
2501
2502             /* Update vectorial force */
2503             fix3             = _mm256_add_ps(fix3,tx);
2504             fiy3             = _mm256_add_ps(fiy3,ty);
2505             fiz3             = _mm256_add_ps(fiz3,tz);
2506
2507             fjx2             = _mm256_add_ps(fjx2,tx);
2508             fjy2             = _mm256_add_ps(fjy2,ty);
2509             fjz2             = _mm256_add_ps(fjz2,tz);
2510
2511             /**************************
2512              * CALCULATE INTERACTIONS *
2513              **************************/
2514
2515             r33              = _mm256_mul_ps(rsq33,rinv33);
2516             r33              = _mm256_andnot_ps(dummy_mask,r33);
2517
2518             /* EWALD ELECTROSTATICS */
2519
2520             /* Analytical PME correction */
2521             zeta2            = _mm256_mul_ps(beta2,rsq33);
2522             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
2523             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2524             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2525             felec            = _mm256_mul_ps(qq33,felec);
2526
2527             fscal            = felec;
2528
2529             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2530
2531             /* Calculate temporary vectorial force */
2532             tx               = _mm256_mul_ps(fscal,dx33);
2533             ty               = _mm256_mul_ps(fscal,dy33);
2534             tz               = _mm256_mul_ps(fscal,dz33);
2535
2536             /* Update vectorial force */
2537             fix3             = _mm256_add_ps(fix3,tx);
2538             fiy3             = _mm256_add_ps(fiy3,ty);
2539             fiz3             = _mm256_add_ps(fiz3,tz);
2540
2541             fjx3             = _mm256_add_ps(fjx3,tx);
2542             fjy3             = _mm256_add_ps(fjy3,ty);
2543             fjz3             = _mm256_add_ps(fjz3,tz);
2544
2545             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2546             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2547             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2548             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2549             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2550             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2551             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2552             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2553
2554             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2555                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2556                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2557
2558             /* Inner loop uses 565 flops */
2559         }
2560
2561         /* End of innermost loop */
2562
2563         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2564                                                  f+i_coord_offset,fshift+i_shift_offset);
2565
2566         /* Increment number of inner iterations */
2567         inneriter                  += j_index_end - j_index_start;
2568
2569         /* Outer loop uses 24 flops */
2570     }
2571
2572     /* Increment number of outer iterations */
2573     outeriter        += nri;
2574
2575     /* Update outer/inner flops */
2576
2577     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*565);
2578 }