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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4W4_VF_avx_256_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Water4
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwCSTab_GeomW4W4_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     real *           vdwioffsetptr3;
  92     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  94     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  96     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  97     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  98     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  99     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
 100     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 101     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 102     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 105     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 108     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 109     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 110     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 111     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 112     real             *charge;
 113     int              nvdwtype;
 114     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 115     int              *vdwtype;
 116     real             *vdwparam;
 117     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 118     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 119     __m256i          vfitab;
 120     __m128i          vfitab_lo,vfitab_hi;
 121     __m128i          ifour       = _mm_set1_epi32(4);
 122     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 123     real             *vftab;
 124     __m256i          ewitab;
 125     __m128i          ewitab_lo,ewitab_hi;
 126     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 127     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 128     real             *ewtab;
 129     __m256           dummy_mask,cutoff_mask;
 130     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 131     __m256           one     = _mm256_set1_ps(1.0);
 132     __m256           two     = _mm256_set1_ps(2.0);
 133     x                = xx[0];
 134     f                = ff[0];
 135
 136     nri              = nlist->nri;
 137     iinr             = nlist->iinr;
 138     jindex           = nlist->jindex;
 139     jjnr             = nlist->jjnr;
 140     shiftidx         = nlist->shift;
 141     gid              = nlist->gid;
 142     shiftvec         = fr->shift_vec[0];
 143     fshift           = fr->fshift[0];
 144     facel            = _mm256_set1_ps(fr->ic->epsfac);
 145     charge           = mdatoms->chargeA;
 146     nvdwtype         = fr->ntype;
 147     vdwparam         = fr->nbfp;
 148     vdwtype          = mdatoms->typeA;
 149
 150     vftab            = kernel_data->table_vdw->data;
 151     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 152
 153     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 154     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 155     beta2            = _mm256_mul_ps(beta,beta);
 156     beta3            = _mm256_mul_ps(beta,beta2);
 157
 158     ewtab            = fr->ic->tabq_coul_FDV0;
 159     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 160     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 161
 162     /* Setup water-specific parameters */
 163     inr              = nlist->iinr[0];
 164     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 165     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 166     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 167     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 168
 169     jq1              = _mm256_set1_ps(charge[inr+1]);
 170     jq2              = _mm256_set1_ps(charge[inr+2]);
 171     jq3              = _mm256_set1_ps(charge[inr+3]);
 172     vdwjidx0A        = 2*vdwtype[inr+0];
 173     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 174     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 175     qq11             = _mm256_mul_ps(iq1,jq1);
 176     qq12             = _mm256_mul_ps(iq1,jq2);
 177     qq13             = _mm256_mul_ps(iq1,jq3);
 178     qq21             = _mm256_mul_ps(iq2,jq1);
 179     qq22             = _mm256_mul_ps(iq2,jq2);
 180     qq23             = _mm256_mul_ps(iq2,jq3);
 181     qq31             = _mm256_mul_ps(iq3,jq1);
 182     qq32             = _mm256_mul_ps(iq3,jq2);
 183     qq33             = _mm256_mul_ps(iq3,jq3);
 184
 185     /* Avoid stupid compiler warnings */
 186     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 187     j_coord_offsetA = 0;
 188     j_coord_offsetB = 0;
 189     j_coord_offsetC = 0;
 190     j_coord_offsetD = 0;
 191     j_coord_offsetE = 0;
 192     j_coord_offsetF = 0;
 193     j_coord_offsetG = 0;
 194     j_coord_offsetH = 0;
 195
 196     outeriter        = 0;
 197     inneriter        = 0;
 198
 199     for(iidx=0;iidx<4*DIM;iidx++)
 200     {
 201         scratch[iidx] = 0.0;
 202     }
 203
 204     /* Start outer loop over neighborlists */
 205     for(iidx=0; iidx<nri; iidx++)
 206     {
 207         /* Load shift vector for this list */
 208         i_shift_offset   = DIM*shiftidx[iidx];
 209
 210         /* Load limits for loop over neighbors */
 211         j_index_start    = jindex[iidx];
 212         j_index_end      = jindex[iidx+1];
 213
 214         /* Get outer coordinate index */
 215         inr              = iinr[iidx];
 216         i_coord_offset   = DIM*inr;
 217
 218         /* Load i particle coords and add shift vector */
 219         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 220                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 221
 222         fix0             = _mm256_setzero_ps();
 223         fiy0             = _mm256_setzero_ps();
 224         fiz0             = _mm256_setzero_ps();
 225         fix1             = _mm256_setzero_ps();
 226         fiy1             = _mm256_setzero_ps();
 227         fiz1             = _mm256_setzero_ps();
 228         fix2             = _mm256_setzero_ps();
 229         fiy2             = _mm256_setzero_ps();
 230         fiz2             = _mm256_setzero_ps();
 231         fix3             = _mm256_setzero_ps();
 232         fiy3             = _mm256_setzero_ps();
 233         fiz3             = _mm256_setzero_ps();
 234
 235         /* Reset potential sums */
 236         velecsum         = _mm256_setzero_ps();
 237         vvdwsum          = _mm256_setzero_ps();
 238
 239         /* Start inner kernel loop */
 240         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 241         {
 242
 243             /* Get j neighbor index, and coordinate index */
 244             jnrA             = jjnr[jidx];
 245             jnrB             = jjnr[jidx+1];
 246             jnrC             = jjnr[jidx+2];
 247             jnrD             = jjnr[jidx+3];
 248             jnrE             = jjnr[jidx+4];
 249             jnrF             = jjnr[jidx+5];
 250             jnrG             = jjnr[jidx+6];
 251             jnrH             = jjnr[jidx+7];
 252             j_coord_offsetA  = DIM*jnrA;
 253             j_coord_offsetB  = DIM*jnrB;
 254             j_coord_offsetC  = DIM*jnrC;
 255             j_coord_offsetD  = DIM*jnrD;
 256             j_coord_offsetE  = DIM*jnrE;
 257             j_coord_offsetF  = DIM*jnrF;
 258             j_coord_offsetG  = DIM*jnrG;
 259             j_coord_offsetH  = DIM*jnrH;
 260
 261             /* load j atom coordinates */
 262             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 263                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 264                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 265                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 266                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 267                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 268
 269             /* Calculate displacement vector */
 270             dx00             = _mm256_sub_ps(ix0,jx0);
 271             dy00             = _mm256_sub_ps(iy0,jy0);
 272             dz00             = _mm256_sub_ps(iz0,jz0);
 273             dx11             = _mm256_sub_ps(ix1,jx1);
 274             dy11             = _mm256_sub_ps(iy1,jy1);
 275             dz11             = _mm256_sub_ps(iz1,jz1);
 276             dx12             = _mm256_sub_ps(ix1,jx2);
 277             dy12             = _mm256_sub_ps(iy1,jy2);
 278             dz12             = _mm256_sub_ps(iz1,jz2);
 279             dx13             = _mm256_sub_ps(ix1,jx3);
 280             dy13             = _mm256_sub_ps(iy1,jy3);
 281             dz13             = _mm256_sub_ps(iz1,jz3);
 282             dx21             = _mm256_sub_ps(ix2,jx1);
 283             dy21             = _mm256_sub_ps(iy2,jy1);
 284             dz21             = _mm256_sub_ps(iz2,jz1);
 285             dx22             = _mm256_sub_ps(ix2,jx2);
 286             dy22             = _mm256_sub_ps(iy2,jy2);
 287             dz22             = _mm256_sub_ps(iz2,jz2);
 288             dx23             = _mm256_sub_ps(ix2,jx3);
 289             dy23             = _mm256_sub_ps(iy2,jy3);
 290             dz23             = _mm256_sub_ps(iz2,jz3);
 291             dx31             = _mm256_sub_ps(ix3,jx1);
 292             dy31             = _mm256_sub_ps(iy3,jy1);
 293             dz31             = _mm256_sub_ps(iz3,jz1);
 294             dx32             = _mm256_sub_ps(ix3,jx2);
 295             dy32             = _mm256_sub_ps(iy3,jy2);
 296             dz32             = _mm256_sub_ps(iz3,jz2);
 297             dx33             = _mm256_sub_ps(ix3,jx3);
 298             dy33             = _mm256_sub_ps(iy3,jy3);
 299             dz33             = _mm256_sub_ps(iz3,jz3);
 300
 301             /* Calculate squared distance and things based on it */
 302             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 303             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 304             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 305             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 306             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 307             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 308             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 309             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 310             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 311             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 312
 313             rinv00           = avx256_invsqrt_f(rsq00);
 314             rinv11           = avx256_invsqrt_f(rsq11);
 315             rinv12           = avx256_invsqrt_f(rsq12);
 316             rinv13           = avx256_invsqrt_f(rsq13);
 317             rinv21           = avx256_invsqrt_f(rsq21);
 318             rinv22           = avx256_invsqrt_f(rsq22);
 319             rinv23           = avx256_invsqrt_f(rsq23);
 320             rinv31           = avx256_invsqrt_f(rsq31);
 321             rinv32           = avx256_invsqrt_f(rsq32);
 322             rinv33           = avx256_invsqrt_f(rsq33);
 323
 324             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 325             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 326             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 327             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 328             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 329             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 330             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 331             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 332             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 333
 334             fjx0             = _mm256_setzero_ps();
 335             fjy0             = _mm256_setzero_ps();
 336             fjz0             = _mm256_setzero_ps();
 337             fjx1             = _mm256_setzero_ps();
 338             fjy1             = _mm256_setzero_ps();
 339             fjz1             = _mm256_setzero_ps();
 340             fjx2             = _mm256_setzero_ps();
 341             fjy2             = _mm256_setzero_ps();
 342             fjz2             = _mm256_setzero_ps();
 343             fjx3             = _mm256_setzero_ps();
 344             fjy3             = _mm256_setzero_ps();
 345             fjz3             = _mm256_setzero_ps();
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             r00              = _mm256_mul_ps(rsq00,rinv00);
 352
 353             /* Calculate table index by multiplying r with table scale and truncate to integer */
 354             rt               = _mm256_mul_ps(r00,vftabscale);
 355             vfitab           = _mm256_cvttps_epi32(rt);
 356             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 357             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 358             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 359             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 360             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 361             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 362
 363             /* CUBIC SPLINE TABLE DISPERSION */
 364             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 365                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 366             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 367                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 368             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 369                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 370             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 371                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 372             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 373             Heps             = _mm256_mul_ps(vfeps,H);
 374             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 375             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 376             vvdw6            = _mm256_mul_ps(c6_00,VV);
 377             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 378             fvdw6            = _mm256_mul_ps(c6_00,FF);
 379
 380             /* CUBIC SPLINE TABLE REPULSION */
 381             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 382             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 383             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 384                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 385             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 386                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 387             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 388                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 389             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 390                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 391             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 392             Heps             = _mm256_mul_ps(vfeps,H);
 393             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 394             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 395             vvdw12           = _mm256_mul_ps(c12_00,VV);
 396             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 397             fvdw12           = _mm256_mul_ps(c12_00,FF);
 398             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 399             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 400
 401             /* Update potential sum for this i atom from the interaction with this j atom. */
 402             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 403
 404             fscal            = fvdw;
 405
 406             /* Calculate temporary vectorial force */
 407             tx               = _mm256_mul_ps(fscal,dx00);
 408             ty               = _mm256_mul_ps(fscal,dy00);
 409             tz               = _mm256_mul_ps(fscal,dz00);
 410
 411             /* Update vectorial force */
 412             fix0             = _mm256_add_ps(fix0,tx);
 413             fiy0             = _mm256_add_ps(fiy0,ty);
 414             fiz0             = _mm256_add_ps(fiz0,tz);
 415
 416             fjx0             = _mm256_add_ps(fjx0,tx);
 417             fjy0             = _mm256_add_ps(fjy0,ty);
 418             fjz0             = _mm256_add_ps(fjz0,tz);
 419
 420             /**************************
 421              * CALCULATE INTERACTIONS *
 422              **************************/
 423
 424             r11              = _mm256_mul_ps(rsq11,rinv11);
 425
 426             /* EWALD ELECTROSTATICS */
 427
 428             /* Analytical PME correction */
 429             zeta2            = _mm256_mul_ps(beta2,rsq11);
 430             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 431             pmecorrF         = avx256_pmecorrF_f(zeta2);
 432             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 433             felec            = _mm256_mul_ps(qq11,felec);
 434             pmecorrV         = avx256_pmecorrV_f(zeta2);
 435             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 436             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 437             velec            = _mm256_mul_ps(qq11,velec);
 438
 439             /* Update potential sum for this i atom from the interaction with this j atom. */
 440             velecsum         = _mm256_add_ps(velecsum,velec);
 441
 442             fscal            = felec;
 443
 444             /* Calculate temporary vectorial force */
 445             tx               = _mm256_mul_ps(fscal,dx11);
 446             ty               = _mm256_mul_ps(fscal,dy11);
 447             tz               = _mm256_mul_ps(fscal,dz11);
 448
 449             /* Update vectorial force */
 450             fix1             = _mm256_add_ps(fix1,tx);
 451             fiy1             = _mm256_add_ps(fiy1,ty);
 452             fiz1             = _mm256_add_ps(fiz1,tz);
 453
 454             fjx1             = _mm256_add_ps(fjx1,tx);
 455             fjy1             = _mm256_add_ps(fjy1,ty);
 456             fjz1             = _mm256_add_ps(fjz1,tz);
 457
 458             /**************************
 459              * CALCULATE INTERACTIONS *
 460              **************************/
 461
 462             r12              = _mm256_mul_ps(rsq12,rinv12);
 463
 464             /* EWALD ELECTROSTATICS */
 465
 466             /* Analytical PME correction */
 467             zeta2            = _mm256_mul_ps(beta2,rsq12);
 468             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 469             pmecorrF         = avx256_pmecorrF_f(zeta2);
 470             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 471             felec            = _mm256_mul_ps(qq12,felec);
 472             pmecorrV         = avx256_pmecorrV_f(zeta2);
 473             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 474             velec            = _mm256_sub_ps(rinv12,pmecorrV);
 475             velec            = _mm256_mul_ps(qq12,velec);
 476
 477             /* Update potential sum for this i atom from the interaction with this j atom. */
 478             velecsum         = _mm256_add_ps(velecsum,velec);
 479
 480             fscal            = felec;
 481
 482             /* Calculate temporary vectorial force */
 483             tx               = _mm256_mul_ps(fscal,dx12);
 484             ty               = _mm256_mul_ps(fscal,dy12);
 485             tz               = _mm256_mul_ps(fscal,dz12);
 486
 487             /* Update vectorial force */
 488             fix1             = _mm256_add_ps(fix1,tx);
 489             fiy1             = _mm256_add_ps(fiy1,ty);
 490             fiz1             = _mm256_add_ps(fiz1,tz);
 491
 492             fjx2             = _mm256_add_ps(fjx2,tx);
 493             fjy2             = _mm256_add_ps(fjy2,ty);
 494             fjz2             = _mm256_add_ps(fjz2,tz);
 495
 496             /**************************
 497              * CALCULATE INTERACTIONS *
 498              **************************/
 499
 500             r13              = _mm256_mul_ps(rsq13,rinv13);
 501
 502             /* EWALD ELECTROSTATICS */
 503
 504             /* Analytical PME correction */
 505             zeta2            = _mm256_mul_ps(beta2,rsq13);
 506             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
 507             pmecorrF         = avx256_pmecorrF_f(zeta2);
 508             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 509             felec            = _mm256_mul_ps(qq13,felec);
 510             pmecorrV         = avx256_pmecorrV_f(zeta2);
 511             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 512             velec            = _mm256_sub_ps(rinv13,pmecorrV);
 513             velec            = _mm256_mul_ps(qq13,velec);
 514
 515             /* Update potential sum for this i atom from the interaction with this j atom. */
 516             velecsum         = _mm256_add_ps(velecsum,velec);
 517
 518             fscal            = felec;
 519
 520             /* Calculate temporary vectorial force */
 521             tx               = _mm256_mul_ps(fscal,dx13);
 522             ty               = _mm256_mul_ps(fscal,dy13);
 523             tz               = _mm256_mul_ps(fscal,dz13);
 524
 525             /* Update vectorial force */
 526             fix1             = _mm256_add_ps(fix1,tx);
 527             fiy1             = _mm256_add_ps(fiy1,ty);
 528             fiz1             = _mm256_add_ps(fiz1,tz);
 529
 530             fjx3             = _mm256_add_ps(fjx3,tx);
 531             fjy3             = _mm256_add_ps(fjy3,ty);
 532             fjz3             = _mm256_add_ps(fjz3,tz);
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             r21              = _mm256_mul_ps(rsq21,rinv21);
 539
 540             /* EWALD ELECTROSTATICS */
 541
 542             /* Analytical PME correction */
 543             zeta2            = _mm256_mul_ps(beta2,rsq21);
 544             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 545             pmecorrF         = avx256_pmecorrF_f(zeta2);
 546             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 547             felec            = _mm256_mul_ps(qq21,felec);
 548             pmecorrV         = avx256_pmecorrV_f(zeta2);
 549             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 550             velec            = _mm256_sub_ps(rinv21,pmecorrV);
 551             velec            = _mm256_mul_ps(qq21,velec);
 552
 553             /* Update potential sum for this i atom from the interaction with this j atom. */
 554             velecsum         = _mm256_add_ps(velecsum,velec);
 555
 556             fscal            = felec;
 557
 558             /* Calculate temporary vectorial force */
 559             tx               = _mm256_mul_ps(fscal,dx21);
 560             ty               = _mm256_mul_ps(fscal,dy21);
 561             tz               = _mm256_mul_ps(fscal,dz21);
 562
 563             /* Update vectorial force */
 564             fix2             = _mm256_add_ps(fix2,tx);
 565             fiy2             = _mm256_add_ps(fiy2,ty);
 566             fiz2             = _mm256_add_ps(fiz2,tz);
 567
 568             fjx1             = _mm256_add_ps(fjx1,tx);
 569             fjy1             = _mm256_add_ps(fjy1,ty);
 570             fjz1             = _mm256_add_ps(fjz1,tz);
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             r22              = _mm256_mul_ps(rsq22,rinv22);
 577
 578             /* EWALD ELECTROSTATICS */
 579
 580             /* Analytical PME correction */
 581             zeta2            = _mm256_mul_ps(beta2,rsq22);
 582             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 583             pmecorrF         = avx256_pmecorrF_f(zeta2);
 584             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 585             felec            = _mm256_mul_ps(qq22,felec);
 586             pmecorrV         = avx256_pmecorrV_f(zeta2);
 587             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 588             velec            = _mm256_sub_ps(rinv22,pmecorrV);
 589             velec            = _mm256_mul_ps(qq22,velec);
 590
 591             /* Update potential sum for this i atom from the interaction with this j atom. */
 592             velecsum         = _mm256_add_ps(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596             /* Calculate temporary vectorial force */
 597             tx               = _mm256_mul_ps(fscal,dx22);
 598             ty               = _mm256_mul_ps(fscal,dy22);
 599             tz               = _mm256_mul_ps(fscal,dz22);
 600
 601             /* Update vectorial force */
 602             fix2             = _mm256_add_ps(fix2,tx);
 603             fiy2             = _mm256_add_ps(fiy2,ty);
 604             fiz2             = _mm256_add_ps(fiz2,tz);
 605
 606             fjx2             = _mm256_add_ps(fjx2,tx);
 607             fjy2             = _mm256_add_ps(fjy2,ty);
 608             fjz2             = _mm256_add_ps(fjz2,tz);
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             r23              = _mm256_mul_ps(rsq23,rinv23);
 615
 616             /* EWALD ELECTROSTATICS */
 617
 618             /* Analytical PME correction */
 619             zeta2            = _mm256_mul_ps(beta2,rsq23);
 620             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
 621             pmecorrF         = avx256_pmecorrF_f(zeta2);
 622             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 623             felec            = _mm256_mul_ps(qq23,felec);
 624             pmecorrV         = avx256_pmecorrV_f(zeta2);
 625             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 626             velec            = _mm256_sub_ps(rinv23,pmecorrV);
 627             velec            = _mm256_mul_ps(qq23,velec);
 628
 629             /* Update potential sum for this i atom from the interaction with this j atom. */
 630             velecsum         = _mm256_add_ps(velecsum,velec);
 631
 632             fscal            = felec;
 633
 634             /* Calculate temporary vectorial force */
 635             tx               = _mm256_mul_ps(fscal,dx23);
 636             ty               = _mm256_mul_ps(fscal,dy23);
 637             tz               = _mm256_mul_ps(fscal,dz23);
 638
 639             /* Update vectorial force */
 640             fix2             = _mm256_add_ps(fix2,tx);
 641             fiy2             = _mm256_add_ps(fiy2,ty);
 642             fiz2             = _mm256_add_ps(fiz2,tz);
 643
 644             fjx3             = _mm256_add_ps(fjx3,tx);
 645             fjy3             = _mm256_add_ps(fjy3,ty);
 646             fjz3             = _mm256_add_ps(fjz3,tz);
 647
 648             /**************************
 649              * CALCULATE INTERACTIONS *
 650              **************************/
 651
 652             r31              = _mm256_mul_ps(rsq31,rinv31);
 653
 654             /* EWALD ELECTROSTATICS */
 655
 656             /* Analytical PME correction */
 657             zeta2            = _mm256_mul_ps(beta2,rsq31);
 658             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
 659             pmecorrF         = avx256_pmecorrF_f(zeta2);
 660             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 661             felec            = _mm256_mul_ps(qq31,felec);
 662             pmecorrV         = avx256_pmecorrV_f(zeta2);
 663             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 664             velec            = _mm256_sub_ps(rinv31,pmecorrV);
 665             velec            = _mm256_mul_ps(qq31,velec);
 666
 667             /* Update potential sum for this i atom from the interaction with this j atom. */
 668             velecsum         = _mm256_add_ps(velecsum,velec);
 669
 670             fscal            = felec;
 671
 672             /* Calculate temporary vectorial force */
 673             tx               = _mm256_mul_ps(fscal,dx31);
 674             ty               = _mm256_mul_ps(fscal,dy31);
 675             tz               = _mm256_mul_ps(fscal,dz31);
 676
 677             /* Update vectorial force */
 678             fix3             = _mm256_add_ps(fix3,tx);
 679             fiy3             = _mm256_add_ps(fiy3,ty);
 680             fiz3             = _mm256_add_ps(fiz3,tz);
 681
 682             fjx1             = _mm256_add_ps(fjx1,tx);
 683             fjy1             = _mm256_add_ps(fjy1,ty);
 684             fjz1             = _mm256_add_ps(fjz1,tz);
 685
 686             /**************************
 687              * CALCULATE INTERACTIONS *
 688              **************************/
 689
 690             r32              = _mm256_mul_ps(rsq32,rinv32);
 691
 692             /* EWALD ELECTROSTATICS */
 693
 694             /* Analytical PME correction */
 695             zeta2            = _mm256_mul_ps(beta2,rsq32);
 696             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
 697             pmecorrF         = avx256_pmecorrF_f(zeta2);
 698             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 699             felec            = _mm256_mul_ps(qq32,felec);
 700             pmecorrV         = avx256_pmecorrV_f(zeta2);
 701             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 702             velec            = _mm256_sub_ps(rinv32,pmecorrV);
 703             velec            = _mm256_mul_ps(qq32,velec);
 704
 705             /* Update potential sum for this i atom from the interaction with this j atom. */
 706             velecsum         = _mm256_add_ps(velecsum,velec);
 707
 708             fscal            = felec;
 709
 710             /* Calculate temporary vectorial force */
 711             tx               = _mm256_mul_ps(fscal,dx32);
 712             ty               = _mm256_mul_ps(fscal,dy32);
 713             tz               = _mm256_mul_ps(fscal,dz32);
 714
 715             /* Update vectorial force */
 716             fix3             = _mm256_add_ps(fix3,tx);
 717             fiy3             = _mm256_add_ps(fiy3,ty);
 718             fiz3             = _mm256_add_ps(fiz3,tz);
 719
 720             fjx2             = _mm256_add_ps(fjx2,tx);
 721             fjy2             = _mm256_add_ps(fjy2,ty);
 722             fjz2             = _mm256_add_ps(fjz2,tz);
 723
 724             /**************************
 725              * CALCULATE INTERACTIONS *
 726              **************************/
 727
 728             r33              = _mm256_mul_ps(rsq33,rinv33);
 729
 730             /* EWALD ELECTROSTATICS */
 731
 732             /* Analytical PME correction */
 733             zeta2            = _mm256_mul_ps(beta2,rsq33);
 734             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
 735             pmecorrF         = avx256_pmecorrF_f(zeta2);
 736             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 737             felec            = _mm256_mul_ps(qq33,felec);
 738             pmecorrV         = avx256_pmecorrV_f(zeta2);
 739             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 740             velec            = _mm256_sub_ps(rinv33,pmecorrV);
 741             velec            = _mm256_mul_ps(qq33,velec);
 742
 743             /* Update potential sum for this i atom from the interaction with this j atom. */
 744             velecsum         = _mm256_add_ps(velecsum,velec);
 745
 746             fscal            = felec;
 747
 748             /* Calculate temporary vectorial force */
 749             tx               = _mm256_mul_ps(fscal,dx33);
 750             ty               = _mm256_mul_ps(fscal,dy33);
 751             tz               = _mm256_mul_ps(fscal,dz33);
 752
 753             /* Update vectorial force */
 754             fix3             = _mm256_add_ps(fix3,tx);
 755             fiy3             = _mm256_add_ps(fiy3,ty);
 756             fiz3             = _mm256_add_ps(fiz3,tz);
 757
 758             fjx3             = _mm256_add_ps(fjx3,tx);
 759             fjy3             = _mm256_add_ps(fjy3,ty);
 760             fjz3             = _mm256_add_ps(fjz3,tz);
 761
 762             fjptrA             = f+j_coord_offsetA;
 763             fjptrB             = f+j_coord_offsetB;
 764             fjptrC             = f+j_coord_offsetC;
 765             fjptrD             = f+j_coord_offsetD;
 766             fjptrE             = f+j_coord_offsetE;
 767             fjptrF             = f+j_coord_offsetF;
 768             fjptrG             = f+j_coord_offsetG;
 769             fjptrH             = f+j_coord_offsetH;
 770
 771             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 772                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 773                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 774
 775             /* Inner loop uses 815 flops */
 776         }
 777
 778         if(jidx<j_index_end)
 779         {
 780
 781             /* Get j neighbor index, and coordinate index */
 782             jnrlistA         = jjnr[jidx];
 783             jnrlistB         = jjnr[jidx+1];
 784             jnrlistC         = jjnr[jidx+2];
 785             jnrlistD         = jjnr[jidx+3];
 786             jnrlistE         = jjnr[jidx+4];
 787             jnrlistF         = jjnr[jidx+5];
 788             jnrlistG         = jjnr[jidx+6];
 789             jnrlistH         = jjnr[jidx+7];
 790             /* Sign of each element will be negative for non-real atoms.
 791              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 792              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 793              */
 794             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 795                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 796
 797             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 798             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 799             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 800             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 801             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 802             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 803             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 804             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 805             j_coord_offsetA  = DIM*jnrA;
 806             j_coord_offsetB  = DIM*jnrB;
 807             j_coord_offsetC  = DIM*jnrC;
 808             j_coord_offsetD  = DIM*jnrD;
 809             j_coord_offsetE  = DIM*jnrE;
 810             j_coord_offsetF  = DIM*jnrF;
 811             j_coord_offsetG  = DIM*jnrG;
 812             j_coord_offsetH  = DIM*jnrH;
 813
 814             /* load j atom coordinates */
 815             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 816                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 817                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 818                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 819                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 820                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 821
 822             /* Calculate displacement vector */
 823             dx00             = _mm256_sub_ps(ix0,jx0);
 824             dy00             = _mm256_sub_ps(iy0,jy0);
 825             dz00             = _mm256_sub_ps(iz0,jz0);
 826             dx11             = _mm256_sub_ps(ix1,jx1);
 827             dy11             = _mm256_sub_ps(iy1,jy1);
 828             dz11             = _mm256_sub_ps(iz1,jz1);
 829             dx12             = _mm256_sub_ps(ix1,jx2);
 830             dy12             = _mm256_sub_ps(iy1,jy2);
 831             dz12             = _mm256_sub_ps(iz1,jz2);
 832             dx13             = _mm256_sub_ps(ix1,jx3);
 833             dy13             = _mm256_sub_ps(iy1,jy3);
 834             dz13             = _mm256_sub_ps(iz1,jz3);
 835             dx21             = _mm256_sub_ps(ix2,jx1);
 836             dy21             = _mm256_sub_ps(iy2,jy1);
 837             dz21             = _mm256_sub_ps(iz2,jz1);
 838             dx22             = _mm256_sub_ps(ix2,jx2);
 839             dy22             = _mm256_sub_ps(iy2,jy2);
 840             dz22             = _mm256_sub_ps(iz2,jz2);
 841             dx23             = _mm256_sub_ps(ix2,jx3);
 842             dy23             = _mm256_sub_ps(iy2,jy3);
 843             dz23             = _mm256_sub_ps(iz2,jz3);
 844             dx31             = _mm256_sub_ps(ix3,jx1);
 845             dy31             = _mm256_sub_ps(iy3,jy1);
 846             dz31             = _mm256_sub_ps(iz3,jz1);
 847             dx32             = _mm256_sub_ps(ix3,jx2);
 848             dy32             = _mm256_sub_ps(iy3,jy2);
 849             dz32             = _mm256_sub_ps(iz3,jz2);
 850             dx33             = _mm256_sub_ps(ix3,jx3);
 851             dy33             = _mm256_sub_ps(iy3,jy3);
 852             dz33             = _mm256_sub_ps(iz3,jz3);
 853
 854             /* Calculate squared distance and things based on it */
 855             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 856             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 857             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 858             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 859             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 860             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 861             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 862             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 863             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 864             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 865
 866             rinv00           = avx256_invsqrt_f(rsq00);
 867             rinv11           = avx256_invsqrt_f(rsq11);
 868             rinv12           = avx256_invsqrt_f(rsq12);
 869             rinv13           = avx256_invsqrt_f(rsq13);
 870             rinv21           = avx256_invsqrt_f(rsq21);
 871             rinv22           = avx256_invsqrt_f(rsq22);
 872             rinv23           = avx256_invsqrt_f(rsq23);
 873             rinv31           = avx256_invsqrt_f(rsq31);
 874             rinv32           = avx256_invsqrt_f(rsq32);
 875             rinv33           = avx256_invsqrt_f(rsq33);
 876
 877             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 878             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 879             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 880             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 881             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 882             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 883             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 884             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 885             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 886
 887             fjx0             = _mm256_setzero_ps();
 888             fjy0             = _mm256_setzero_ps();
 889             fjz0             = _mm256_setzero_ps();
 890             fjx1             = _mm256_setzero_ps();
 891             fjy1             = _mm256_setzero_ps();
 892             fjz1             = _mm256_setzero_ps();
 893             fjx2             = _mm256_setzero_ps();
 894             fjy2             = _mm256_setzero_ps();
 895             fjz2             = _mm256_setzero_ps();
 896             fjx3             = _mm256_setzero_ps();
 897             fjy3             = _mm256_setzero_ps();
 898             fjz3             = _mm256_setzero_ps();
 899
 900             /**************************
 901              * CALCULATE INTERACTIONS *
 902              **************************/
 903
 904             r00              = _mm256_mul_ps(rsq00,rinv00);
 905             r00              = _mm256_andnot_ps(dummy_mask,r00);
 906
 907             /* Calculate table index by multiplying r with table scale and truncate to integer */
 908             rt               = _mm256_mul_ps(r00,vftabscale);
 909             vfitab           = _mm256_cvttps_epi32(rt);
 910             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 911             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 912             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 913             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 914             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 915             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 916
 917             /* CUBIC SPLINE TABLE DISPERSION */
 918             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 919                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 920             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 921                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 922             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 923                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 924             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 925                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 926             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 927             Heps             = _mm256_mul_ps(vfeps,H);
 928             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 929             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 930             vvdw6            = _mm256_mul_ps(c6_00,VV);
 931             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 932             fvdw6            = _mm256_mul_ps(c6_00,FF);
 933
 934             /* CUBIC SPLINE TABLE REPULSION */
 935             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 936             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 937             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 938                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 939             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 940                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 941             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 942                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 943             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 944                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 945             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 946             Heps             = _mm256_mul_ps(vfeps,H);
 947             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 948             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 949             vvdw12           = _mm256_mul_ps(c12_00,VV);
 950             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 951             fvdw12           = _mm256_mul_ps(c12_00,FF);
 952             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 953             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 954
 955             /* Update potential sum for this i atom from the interaction with this j atom. */
 956             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 957             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 958
 959             fscal            = fvdw;
 960
 961             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 962
 963             /* Calculate temporary vectorial force */
 964             tx               = _mm256_mul_ps(fscal,dx00);
 965             ty               = _mm256_mul_ps(fscal,dy00);
 966             tz               = _mm256_mul_ps(fscal,dz00);
 967
 968             /* Update vectorial force */
 969             fix0             = _mm256_add_ps(fix0,tx);
 970             fiy0             = _mm256_add_ps(fiy0,ty);
 971             fiz0             = _mm256_add_ps(fiz0,tz);
 972
 973             fjx0             = _mm256_add_ps(fjx0,tx);
 974             fjy0             = _mm256_add_ps(fjy0,ty);
 975             fjz0             = _mm256_add_ps(fjz0,tz);
 976
 977             /**************************
 978              * CALCULATE INTERACTIONS *
 979              **************************/
 980
 981             r11              = _mm256_mul_ps(rsq11,rinv11);
 982             r11              = _mm256_andnot_ps(dummy_mask,r11);
 983
 984             /* EWALD ELECTROSTATICS */
 985
 986             /* Analytical PME correction */
 987             zeta2            = _mm256_mul_ps(beta2,rsq11);
 988             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 989             pmecorrF         = avx256_pmecorrF_f(zeta2);
 990             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 991             felec            = _mm256_mul_ps(qq11,felec);
 992             pmecorrV         = avx256_pmecorrV_f(zeta2);
 993             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 994             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 995             velec            = _mm256_mul_ps(qq11,velec);
 996
 997             /* Update potential sum for this i atom from the interaction with this j atom. */
 998             velec            = _mm256_andnot_ps(dummy_mask,velec);
 999             velecsum         = _mm256_add_ps(velecsum,velec);
1000
1001             fscal            = felec;
1002
1003             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1004
1005             /* Calculate temporary vectorial force */
1006             tx               = _mm256_mul_ps(fscal,dx11);
1007             ty               = _mm256_mul_ps(fscal,dy11);
1008             tz               = _mm256_mul_ps(fscal,dz11);
1009
1010             /* Update vectorial force */
1011             fix1             = _mm256_add_ps(fix1,tx);
1012             fiy1             = _mm256_add_ps(fiy1,ty);
1013             fiz1             = _mm256_add_ps(fiz1,tz);
1014
1015             fjx1             = _mm256_add_ps(fjx1,tx);
1016             fjy1             = _mm256_add_ps(fjy1,ty);
1017             fjz1             = _mm256_add_ps(fjz1,tz);
1018
1019             /**************************
1020              * CALCULATE INTERACTIONS *
1021              **************************/
1022
1023             r12              = _mm256_mul_ps(rsq12,rinv12);
1024             r12              = _mm256_andnot_ps(dummy_mask,r12);
1025
1026             /* EWALD ELECTROSTATICS */
1027
1028             /* Analytical PME correction */
1029             zeta2            = _mm256_mul_ps(beta2,rsq12);
1030             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1031             pmecorrF         = avx256_pmecorrF_f(zeta2);
1032             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1033             felec            = _mm256_mul_ps(qq12,felec);
1034             pmecorrV         = avx256_pmecorrV_f(zeta2);
1035             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1036             velec            = _mm256_sub_ps(rinv12,pmecorrV);
1037             velec            = _mm256_mul_ps(qq12,velec);
1038
1039             /* Update potential sum for this i atom from the interaction with this j atom. */
1040             velec            = _mm256_andnot_ps(dummy_mask,velec);
1041             velecsum         = _mm256_add_ps(velecsum,velec);
1042
1043             fscal            = felec;
1044
1045             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1046
1047             /* Calculate temporary vectorial force */
1048             tx               = _mm256_mul_ps(fscal,dx12);
1049             ty               = _mm256_mul_ps(fscal,dy12);
1050             tz               = _mm256_mul_ps(fscal,dz12);
1051
1052             /* Update vectorial force */
1053             fix1             = _mm256_add_ps(fix1,tx);
1054             fiy1             = _mm256_add_ps(fiy1,ty);
1055             fiz1             = _mm256_add_ps(fiz1,tz);
1056
1057             fjx2             = _mm256_add_ps(fjx2,tx);
1058             fjy2             = _mm256_add_ps(fjy2,ty);
1059             fjz2             = _mm256_add_ps(fjz2,tz);
1060
1061             /**************************
1062              * CALCULATE INTERACTIONS *
1063              **************************/
1064
1065             r13              = _mm256_mul_ps(rsq13,rinv13);
1066             r13              = _mm256_andnot_ps(dummy_mask,r13);
1067
1068             /* EWALD ELECTROSTATICS */
1069
1070             /* Analytical PME correction */
1071             zeta2            = _mm256_mul_ps(beta2,rsq13);
1072             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
1073             pmecorrF         = avx256_pmecorrF_f(zeta2);
1074             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1075             felec            = _mm256_mul_ps(qq13,felec);
1076             pmecorrV         = avx256_pmecorrV_f(zeta2);
1077             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1078             velec            = _mm256_sub_ps(rinv13,pmecorrV);
1079             velec            = _mm256_mul_ps(qq13,velec);
1080
1081             /* Update potential sum for this i atom from the interaction with this j atom. */
1082             velec            = _mm256_andnot_ps(dummy_mask,velec);
1083             velecsum         = _mm256_add_ps(velecsum,velec);
1084
1085             fscal            = felec;
1086
1087             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1088
1089             /* Calculate temporary vectorial force */
1090             tx               = _mm256_mul_ps(fscal,dx13);
1091             ty               = _mm256_mul_ps(fscal,dy13);
1092             tz               = _mm256_mul_ps(fscal,dz13);
1093
1094             /* Update vectorial force */
1095             fix1             = _mm256_add_ps(fix1,tx);
1096             fiy1             = _mm256_add_ps(fiy1,ty);
1097             fiz1             = _mm256_add_ps(fiz1,tz);
1098
1099             fjx3             = _mm256_add_ps(fjx3,tx);
1100             fjy3             = _mm256_add_ps(fjy3,ty);
1101             fjz3             = _mm256_add_ps(fjz3,tz);
1102
1103             /**************************
1104              * CALCULATE INTERACTIONS *
1105              **************************/
1106
1107             r21              = _mm256_mul_ps(rsq21,rinv21);
1108             r21              = _mm256_andnot_ps(dummy_mask,r21);
1109
1110             /* EWALD ELECTROSTATICS */
1111
1112             /* Analytical PME correction */
1113             zeta2            = _mm256_mul_ps(beta2,rsq21);
1114             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1115             pmecorrF         = avx256_pmecorrF_f(zeta2);
1116             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1117             felec            = _mm256_mul_ps(qq21,felec);
1118             pmecorrV         = avx256_pmecorrV_f(zeta2);
1119             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1120             velec            = _mm256_sub_ps(rinv21,pmecorrV);
1121             velec            = _mm256_mul_ps(qq21,velec);
1122
1123             /* Update potential sum for this i atom from the interaction with this j atom. */
1124             velec            = _mm256_andnot_ps(dummy_mask,velec);
1125             velecsum         = _mm256_add_ps(velecsum,velec);
1126
1127             fscal            = felec;
1128
1129             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1130
1131             /* Calculate temporary vectorial force */
1132             tx               = _mm256_mul_ps(fscal,dx21);
1133             ty               = _mm256_mul_ps(fscal,dy21);
1134             tz               = _mm256_mul_ps(fscal,dz21);
1135
1136             /* Update vectorial force */
1137             fix2             = _mm256_add_ps(fix2,tx);
1138             fiy2             = _mm256_add_ps(fiy2,ty);
1139             fiz2             = _mm256_add_ps(fiz2,tz);
1140
1141             fjx1             = _mm256_add_ps(fjx1,tx);
1142             fjy1             = _mm256_add_ps(fjy1,ty);
1143             fjz1             = _mm256_add_ps(fjz1,tz);
1144
1145             /**************************
1146              * CALCULATE INTERACTIONS *
1147              **************************/
1148
1149             r22              = _mm256_mul_ps(rsq22,rinv22);
1150             r22              = _mm256_andnot_ps(dummy_mask,r22);
1151
1152             /* EWALD ELECTROSTATICS */
1153
1154             /* Analytical PME correction */
1155             zeta2            = _mm256_mul_ps(beta2,rsq22);
1156             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1157             pmecorrF         = avx256_pmecorrF_f(zeta2);
1158             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1159             felec            = _mm256_mul_ps(qq22,felec);
1160             pmecorrV         = avx256_pmecorrV_f(zeta2);
1161             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1162             velec            = _mm256_sub_ps(rinv22,pmecorrV);
1163             velec            = _mm256_mul_ps(qq22,velec);
1164
1165             /* Update potential sum for this i atom from the interaction with this j atom. */
1166             velec            = _mm256_andnot_ps(dummy_mask,velec);
1167             velecsum         = _mm256_add_ps(velecsum,velec);
1168
1169             fscal            = felec;
1170
1171             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1172
1173             /* Calculate temporary vectorial force */
1174             tx               = _mm256_mul_ps(fscal,dx22);
1175             ty               = _mm256_mul_ps(fscal,dy22);
1176             tz               = _mm256_mul_ps(fscal,dz22);
1177
1178             /* Update vectorial force */
1179             fix2             = _mm256_add_ps(fix2,tx);
1180             fiy2             = _mm256_add_ps(fiy2,ty);
1181             fiz2             = _mm256_add_ps(fiz2,tz);
1182
1183             fjx2             = _mm256_add_ps(fjx2,tx);
1184             fjy2             = _mm256_add_ps(fjy2,ty);
1185             fjz2             = _mm256_add_ps(fjz2,tz);
1186
1187             /**************************
1188              * CALCULATE INTERACTIONS *
1189              **************************/
1190
1191             r23              = _mm256_mul_ps(rsq23,rinv23);
1192             r23              = _mm256_andnot_ps(dummy_mask,r23);
1193
1194             /* EWALD ELECTROSTATICS */
1195
1196             /* Analytical PME correction */
1197             zeta2            = _mm256_mul_ps(beta2,rsq23);
1198             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1199             pmecorrF         = avx256_pmecorrF_f(zeta2);
1200             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1201             felec            = _mm256_mul_ps(qq23,felec);
1202             pmecorrV         = avx256_pmecorrV_f(zeta2);
1203             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1204             velec            = _mm256_sub_ps(rinv23,pmecorrV);
1205             velec            = _mm256_mul_ps(qq23,velec);
1206
1207             /* Update potential sum for this i atom from the interaction with this j atom. */
1208             velec            = _mm256_andnot_ps(dummy_mask,velec);
1209             velecsum         = _mm256_add_ps(velecsum,velec);
1210
1211             fscal            = felec;
1212
1213             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1214
1215             /* Calculate temporary vectorial force */
1216             tx               = _mm256_mul_ps(fscal,dx23);
1217             ty               = _mm256_mul_ps(fscal,dy23);
1218             tz               = _mm256_mul_ps(fscal,dz23);
1219
1220             /* Update vectorial force */
1221             fix2             = _mm256_add_ps(fix2,tx);
1222             fiy2             = _mm256_add_ps(fiy2,ty);
1223             fiz2             = _mm256_add_ps(fiz2,tz);
1224
1225             fjx3             = _mm256_add_ps(fjx3,tx);
1226             fjy3             = _mm256_add_ps(fjy3,ty);
1227             fjz3             = _mm256_add_ps(fjz3,tz);
1228
1229             /**************************
1230              * CALCULATE INTERACTIONS *
1231              **************************/
1232
1233             r31              = _mm256_mul_ps(rsq31,rinv31);
1234             r31              = _mm256_andnot_ps(dummy_mask,r31);
1235
1236             /* EWALD ELECTROSTATICS */
1237
1238             /* Analytical PME correction */
1239             zeta2            = _mm256_mul_ps(beta2,rsq31);
1240             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1241             pmecorrF         = avx256_pmecorrF_f(zeta2);
1242             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1243             felec            = _mm256_mul_ps(qq31,felec);
1244             pmecorrV         = avx256_pmecorrV_f(zeta2);
1245             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1246             velec            = _mm256_sub_ps(rinv31,pmecorrV);
1247             velec            = _mm256_mul_ps(qq31,velec);
1248
1249             /* Update potential sum for this i atom from the interaction with this j atom. */
1250             velec            = _mm256_andnot_ps(dummy_mask,velec);
1251             velecsum         = _mm256_add_ps(velecsum,velec);
1252
1253             fscal            = felec;
1254
1255             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1256
1257             /* Calculate temporary vectorial force */
1258             tx               = _mm256_mul_ps(fscal,dx31);
1259             ty               = _mm256_mul_ps(fscal,dy31);
1260             tz               = _mm256_mul_ps(fscal,dz31);
1261
1262             /* Update vectorial force */
1263             fix3             = _mm256_add_ps(fix3,tx);
1264             fiy3             = _mm256_add_ps(fiy3,ty);
1265             fiz3             = _mm256_add_ps(fiz3,tz);
1266
1267             fjx1             = _mm256_add_ps(fjx1,tx);
1268             fjy1             = _mm256_add_ps(fjy1,ty);
1269             fjz1             = _mm256_add_ps(fjz1,tz);
1270
1271             /**************************
1272              * CALCULATE INTERACTIONS *
1273              **************************/
1274
1275             r32              = _mm256_mul_ps(rsq32,rinv32);
1276             r32              = _mm256_andnot_ps(dummy_mask,r32);
1277
1278             /* EWALD ELECTROSTATICS */
1279
1280             /* Analytical PME correction */
1281             zeta2            = _mm256_mul_ps(beta2,rsq32);
1282             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1283             pmecorrF         = avx256_pmecorrF_f(zeta2);
1284             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1285             felec            = _mm256_mul_ps(qq32,felec);
1286             pmecorrV         = avx256_pmecorrV_f(zeta2);
1287             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1288             velec            = _mm256_sub_ps(rinv32,pmecorrV);
1289             velec            = _mm256_mul_ps(qq32,velec);
1290
1291             /* Update potential sum for this i atom from the interaction with this j atom. */
1292             velec            = _mm256_andnot_ps(dummy_mask,velec);
1293             velecsum         = _mm256_add_ps(velecsum,velec);
1294
1295             fscal            = felec;
1296
1297             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1298
1299             /* Calculate temporary vectorial force */
1300             tx               = _mm256_mul_ps(fscal,dx32);
1301             ty               = _mm256_mul_ps(fscal,dy32);
1302             tz               = _mm256_mul_ps(fscal,dz32);
1303
1304             /* Update vectorial force */
1305             fix3             = _mm256_add_ps(fix3,tx);
1306             fiy3             = _mm256_add_ps(fiy3,ty);
1307             fiz3             = _mm256_add_ps(fiz3,tz);
1308
1309             fjx2             = _mm256_add_ps(fjx2,tx);
1310             fjy2             = _mm256_add_ps(fjy2,ty);
1311             fjz2             = _mm256_add_ps(fjz2,tz);
1312
1313             /**************************
1314              * CALCULATE INTERACTIONS *
1315              **************************/
1316
1317             r33              = _mm256_mul_ps(rsq33,rinv33);
1318             r33              = _mm256_andnot_ps(dummy_mask,r33);
1319
1320             /* EWALD ELECTROSTATICS */
1321
1322             /* Analytical PME correction */
1323             zeta2            = _mm256_mul_ps(beta2,rsq33);
1324             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
1325             pmecorrF         = avx256_pmecorrF_f(zeta2);
1326             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1327             felec            = _mm256_mul_ps(qq33,felec);
1328             pmecorrV         = avx256_pmecorrV_f(zeta2);
1329             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1330             velec            = _mm256_sub_ps(rinv33,pmecorrV);
1331             velec            = _mm256_mul_ps(qq33,velec);
1332
1333             /* Update potential sum for this i atom from the interaction with this j atom. */
1334             velec            = _mm256_andnot_ps(dummy_mask,velec);
1335             velecsum         = _mm256_add_ps(velecsum,velec);
1336
1337             fscal            = felec;
1338
1339             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1340
1341             /* Calculate temporary vectorial force */
1342             tx               = _mm256_mul_ps(fscal,dx33);
1343             ty               = _mm256_mul_ps(fscal,dy33);
1344             tz               = _mm256_mul_ps(fscal,dz33);
1345
1346             /* Update vectorial force */
1347             fix3             = _mm256_add_ps(fix3,tx);
1348             fiy3             = _mm256_add_ps(fiy3,ty);
1349             fiz3             = _mm256_add_ps(fiz3,tz);
1350
1351             fjx3             = _mm256_add_ps(fjx3,tx);
1352             fjy3             = _mm256_add_ps(fjy3,ty);
1353             fjz3             = _mm256_add_ps(fjz3,tz);
1354
1355             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1356             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1357             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1358             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1359             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1360             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1361             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1362             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1363
1364             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1365                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1366                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1367
1368             /* Inner loop uses 825 flops */
1369         }
1370
1371         /* End of innermost loop */
1372
1373         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1374                                                  f+i_coord_offset,fshift+i_shift_offset);
1375
1376         ggid                        = gid[iidx];
1377         /* Update potential energies */
1378         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1379         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1380
1381         /* Increment number of inner iterations */
1382         inneriter                  += j_index_end - j_index_start;
1383
1384         /* Outer loop uses 26 flops */
1385     }
1386
1387     /* Increment number of outer iterations */
1388     outeriter        += nri;
1389
1390     /* Update outer/inner flops */
1391
1392     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*825);
1393 }
1394 /*
1395  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4W4_F_avx_256_single
1396  * Electrostatics interaction: Ewald
1397  * VdW interaction:            CubicSplineTable
1398  * Geometry:                   Water4-Water4
1399  * Calculate force/pot:        Force
1400  */
1401 void
1402 nb_kernel_ElecEw_VdwCSTab_GeomW4W4_F_avx_256_single
1403                     (t_nblist                    * gmx_restrict       nlist,
1404                      rvec                        * gmx_restrict          xx,
1405                      rvec                        * gmx_restrict          ff,
1406                      struct t_forcerec           * gmx_restrict          fr,
1407                      t_mdatoms                   * gmx_restrict     mdatoms,
1408                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1409                      t_nrnb                      * gmx_restrict        nrnb)
1410 {
1411     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1412      * just 0 for non-waters.
1413      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1414      * jnr indices corresponding to data put in the four positions in the SIMD register.
1415      */
1416     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1417     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1418     int              jnrA,jnrB,jnrC,jnrD;
1419     int              jnrE,jnrF,jnrG,jnrH;
1420     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1421     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1422     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1423     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1424     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1425     real             rcutoff_scalar;
1426     real             *shiftvec,*fshift,*x,*f;
1427     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1428     real             scratch[4*DIM];
1429     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1430     real *           vdwioffsetptr0;
1431     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1432     real *           vdwioffsetptr1;
1433     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1434     real *           vdwioffsetptr2;
1435     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1436     real *           vdwioffsetptr3;
1437     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1438     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1439     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1440     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1441     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1442     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1443     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1444     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1445     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1446     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1447     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1448     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1449     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1450     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1451     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1452     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1453     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1454     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1455     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1456     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1457     real             *charge;
1458     int              nvdwtype;
1459     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1460     int              *vdwtype;
1461     real             *vdwparam;
1462     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1463     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1464     __m256i          vfitab;
1465     __m128i          vfitab_lo,vfitab_hi;
1466     __m128i          ifour       = _mm_set1_epi32(4);
1467     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1468     real             *vftab;
1469     __m256i          ewitab;
1470     __m128i          ewitab_lo,ewitab_hi;
1471     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1472     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1473     real             *ewtab;
1474     __m256           dummy_mask,cutoff_mask;
1475     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1476     __m256           one     = _mm256_set1_ps(1.0);
1477     __m256           two     = _mm256_set1_ps(2.0);
1478     x                = xx[0];
1479     f                = ff[0];
1480
1481     nri              = nlist->nri;
1482     iinr             = nlist->iinr;
1483     jindex           = nlist->jindex;
1484     jjnr             = nlist->jjnr;
1485     shiftidx         = nlist->shift;
1486     gid              = nlist->gid;
1487     shiftvec         = fr->shift_vec[0];
1488     fshift           = fr->fshift[0];
1489     facel            = _mm256_set1_ps(fr->ic->epsfac);
1490     charge           = mdatoms->chargeA;
1491     nvdwtype         = fr->ntype;
1492     vdwparam         = fr->nbfp;
1493     vdwtype          = mdatoms->typeA;
1494
1495     vftab            = kernel_data->table_vdw->data;
1496     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
1497
1498     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1499     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1500     beta2            = _mm256_mul_ps(beta,beta);
1501     beta3            = _mm256_mul_ps(beta,beta2);
1502
1503     ewtab            = fr->ic->tabq_coul_F;
1504     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1505     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1506
1507     /* Setup water-specific parameters */
1508     inr              = nlist->iinr[0];
1509     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1510     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1511     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1512     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1513
1514     jq1              = _mm256_set1_ps(charge[inr+1]);
1515     jq2              = _mm256_set1_ps(charge[inr+2]);
1516     jq3              = _mm256_set1_ps(charge[inr+3]);
1517     vdwjidx0A        = 2*vdwtype[inr+0];
1518     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1519     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1520     qq11             = _mm256_mul_ps(iq1,jq1);
1521     qq12             = _mm256_mul_ps(iq1,jq2);
1522     qq13             = _mm256_mul_ps(iq1,jq3);
1523     qq21             = _mm256_mul_ps(iq2,jq1);
1524     qq22             = _mm256_mul_ps(iq2,jq2);
1525     qq23             = _mm256_mul_ps(iq2,jq3);
1526     qq31             = _mm256_mul_ps(iq3,jq1);
1527     qq32             = _mm256_mul_ps(iq3,jq2);
1528     qq33             = _mm256_mul_ps(iq3,jq3);
1529
1530     /* Avoid stupid compiler warnings */
1531     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1532     j_coord_offsetA = 0;
1533     j_coord_offsetB = 0;
1534     j_coord_offsetC = 0;
1535     j_coord_offsetD = 0;
1536     j_coord_offsetE = 0;
1537     j_coord_offsetF = 0;
1538     j_coord_offsetG = 0;
1539     j_coord_offsetH = 0;
1540
1541     outeriter        = 0;
1542     inneriter        = 0;
1543
1544     for(iidx=0;iidx<4*DIM;iidx++)
1545     {
1546         scratch[iidx] = 0.0;
1547     }
1548
1549     /* Start outer loop over neighborlists */
1550     for(iidx=0; iidx<nri; iidx++)
1551     {
1552         /* Load shift vector for this list */
1553         i_shift_offset   = DIM*shiftidx[iidx];
1554
1555         /* Load limits for loop over neighbors */
1556         j_index_start    = jindex[iidx];
1557         j_index_end      = jindex[iidx+1];
1558
1559         /* Get outer coordinate index */
1560         inr              = iinr[iidx];
1561         i_coord_offset   = DIM*inr;
1562
1563         /* Load i particle coords and add shift vector */
1564         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1565                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1566
1567         fix0             = _mm256_setzero_ps();
1568         fiy0             = _mm256_setzero_ps();
1569         fiz0             = _mm256_setzero_ps();
1570         fix1             = _mm256_setzero_ps();
1571         fiy1             = _mm256_setzero_ps();
1572         fiz1             = _mm256_setzero_ps();
1573         fix2             = _mm256_setzero_ps();
1574         fiy2             = _mm256_setzero_ps();
1575         fiz2             = _mm256_setzero_ps();
1576         fix3             = _mm256_setzero_ps();
1577         fiy3             = _mm256_setzero_ps();
1578         fiz3             = _mm256_setzero_ps();
1579
1580         /* Start inner kernel loop */
1581         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1582         {
1583
1584             /* Get j neighbor index, and coordinate index */
1585             jnrA             = jjnr[jidx];
1586             jnrB             = jjnr[jidx+1];
1587             jnrC             = jjnr[jidx+2];
1588             jnrD             = jjnr[jidx+3];
1589             jnrE             = jjnr[jidx+4];
1590             jnrF             = jjnr[jidx+5];
1591             jnrG             = jjnr[jidx+6];
1592             jnrH             = jjnr[jidx+7];
1593             j_coord_offsetA  = DIM*jnrA;
1594             j_coord_offsetB  = DIM*jnrB;
1595             j_coord_offsetC  = DIM*jnrC;
1596             j_coord_offsetD  = DIM*jnrD;
1597             j_coord_offsetE  = DIM*jnrE;
1598             j_coord_offsetF  = DIM*jnrF;
1599             j_coord_offsetG  = DIM*jnrG;
1600             j_coord_offsetH  = DIM*jnrH;
1601
1602             /* load j atom coordinates */
1603             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1604                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1605                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1606                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1607                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1608                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1609
1610             /* Calculate displacement vector */
1611             dx00             = _mm256_sub_ps(ix0,jx0);
1612             dy00             = _mm256_sub_ps(iy0,jy0);
1613             dz00             = _mm256_sub_ps(iz0,jz0);
1614             dx11             = _mm256_sub_ps(ix1,jx1);
1615             dy11             = _mm256_sub_ps(iy1,jy1);
1616             dz11             = _mm256_sub_ps(iz1,jz1);
1617             dx12             = _mm256_sub_ps(ix1,jx2);
1618             dy12             = _mm256_sub_ps(iy1,jy2);
1619             dz12             = _mm256_sub_ps(iz1,jz2);
1620             dx13             = _mm256_sub_ps(ix1,jx3);
1621             dy13             = _mm256_sub_ps(iy1,jy3);
1622             dz13             = _mm256_sub_ps(iz1,jz3);
1623             dx21             = _mm256_sub_ps(ix2,jx1);
1624             dy21             = _mm256_sub_ps(iy2,jy1);
1625             dz21             = _mm256_sub_ps(iz2,jz1);
1626             dx22             = _mm256_sub_ps(ix2,jx2);
1627             dy22             = _mm256_sub_ps(iy2,jy2);
1628             dz22             = _mm256_sub_ps(iz2,jz2);
1629             dx23             = _mm256_sub_ps(ix2,jx3);
1630             dy23             = _mm256_sub_ps(iy2,jy3);
1631             dz23             = _mm256_sub_ps(iz2,jz3);
1632             dx31             = _mm256_sub_ps(ix3,jx1);
1633             dy31             = _mm256_sub_ps(iy3,jy1);
1634             dz31             = _mm256_sub_ps(iz3,jz1);
1635             dx32             = _mm256_sub_ps(ix3,jx2);
1636             dy32             = _mm256_sub_ps(iy3,jy2);
1637             dz32             = _mm256_sub_ps(iz3,jz2);
1638             dx33             = _mm256_sub_ps(ix3,jx3);
1639             dy33             = _mm256_sub_ps(iy3,jy3);
1640             dz33             = _mm256_sub_ps(iz3,jz3);
1641
1642             /* Calculate squared distance and things based on it */
1643             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1644             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1645             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1646             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1647             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1648             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1649             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1650             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1651             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1652             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1653
1654             rinv00           = avx256_invsqrt_f(rsq00);
1655             rinv11           = avx256_invsqrt_f(rsq11);
1656             rinv12           = avx256_invsqrt_f(rsq12);
1657             rinv13           = avx256_invsqrt_f(rsq13);
1658             rinv21           = avx256_invsqrt_f(rsq21);
1659             rinv22           = avx256_invsqrt_f(rsq22);
1660             rinv23           = avx256_invsqrt_f(rsq23);
1661             rinv31           = avx256_invsqrt_f(rsq31);
1662             rinv32           = avx256_invsqrt_f(rsq32);
1663             rinv33           = avx256_invsqrt_f(rsq33);
1664
1665             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1666             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1667             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1668             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1669             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1670             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1671             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1672             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1673             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1674
1675             fjx0             = _mm256_setzero_ps();
1676             fjy0             = _mm256_setzero_ps();
1677             fjz0             = _mm256_setzero_ps();
1678             fjx1             = _mm256_setzero_ps();
1679             fjy1             = _mm256_setzero_ps();
1680             fjz1             = _mm256_setzero_ps();
1681             fjx2             = _mm256_setzero_ps();
1682             fjy2             = _mm256_setzero_ps();
1683             fjz2             = _mm256_setzero_ps();
1684             fjx3             = _mm256_setzero_ps();
1685             fjy3             = _mm256_setzero_ps();
1686             fjz3             = _mm256_setzero_ps();
1687
1688             /**************************
1689              * CALCULATE INTERACTIONS *
1690              **************************/
1691
1692             r00              = _mm256_mul_ps(rsq00,rinv00);
1693
1694             /* Calculate table index by multiplying r with table scale and truncate to integer */
1695             rt               = _mm256_mul_ps(r00,vftabscale);
1696             vfitab           = _mm256_cvttps_epi32(rt);
1697             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1698             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1699             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1700             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1701             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1702             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1703
1704             /* CUBIC SPLINE TABLE DISPERSION */
1705             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1706                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1707             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1708                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1709             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1710                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1711             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1712                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1713             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1714             Heps             = _mm256_mul_ps(vfeps,H);
1715             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1716             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1717             fvdw6            = _mm256_mul_ps(c6_00,FF);
1718
1719             /* CUBIC SPLINE TABLE REPULSION */
1720             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1721             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1722             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1723                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1724             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1725                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1726             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1727                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1728             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1729                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1730             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1731             Heps             = _mm256_mul_ps(vfeps,H);
1732             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1733             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1734             fvdw12           = _mm256_mul_ps(c12_00,FF);
1735             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1736
1737             fscal            = fvdw;
1738
1739             /* Calculate temporary vectorial force */
1740             tx               = _mm256_mul_ps(fscal,dx00);
1741             ty               = _mm256_mul_ps(fscal,dy00);
1742             tz               = _mm256_mul_ps(fscal,dz00);
1743
1744             /* Update vectorial force */
1745             fix0             = _mm256_add_ps(fix0,tx);
1746             fiy0             = _mm256_add_ps(fiy0,ty);
1747             fiz0             = _mm256_add_ps(fiz0,tz);
1748
1749             fjx0             = _mm256_add_ps(fjx0,tx);
1750             fjy0             = _mm256_add_ps(fjy0,ty);
1751             fjz0             = _mm256_add_ps(fjz0,tz);
1752
1753             /**************************
1754              * CALCULATE INTERACTIONS *
1755              **************************/
1756
1757             r11              = _mm256_mul_ps(rsq11,rinv11);
1758
1759             /* EWALD ELECTROSTATICS */
1760
1761             /* Analytical PME correction */
1762             zeta2            = _mm256_mul_ps(beta2,rsq11);
1763             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1764             pmecorrF         = avx256_pmecorrF_f(zeta2);
1765             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1766             felec            = _mm256_mul_ps(qq11,felec);
1767
1768             fscal            = felec;
1769
1770             /* Calculate temporary vectorial force */
1771             tx               = _mm256_mul_ps(fscal,dx11);
1772             ty               = _mm256_mul_ps(fscal,dy11);
1773             tz               = _mm256_mul_ps(fscal,dz11);
1774
1775             /* Update vectorial force */
1776             fix1             = _mm256_add_ps(fix1,tx);
1777             fiy1             = _mm256_add_ps(fiy1,ty);
1778             fiz1             = _mm256_add_ps(fiz1,tz);
1779
1780             fjx1             = _mm256_add_ps(fjx1,tx);
1781             fjy1             = _mm256_add_ps(fjy1,ty);
1782             fjz1             = _mm256_add_ps(fjz1,tz);
1783
1784             /**************************
1785              * CALCULATE INTERACTIONS *
1786              **************************/
1787
1788             r12              = _mm256_mul_ps(rsq12,rinv12);
1789
1790             /* EWALD ELECTROSTATICS */
1791
1792             /* Analytical PME correction */
1793             zeta2            = _mm256_mul_ps(beta2,rsq12);
1794             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1795             pmecorrF         = avx256_pmecorrF_f(zeta2);
1796             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1797             felec            = _mm256_mul_ps(qq12,felec);
1798
1799             fscal            = felec;
1800
1801             /* Calculate temporary vectorial force */
1802             tx               = _mm256_mul_ps(fscal,dx12);
1803             ty               = _mm256_mul_ps(fscal,dy12);
1804             tz               = _mm256_mul_ps(fscal,dz12);
1805
1806             /* Update vectorial force */
1807             fix1             = _mm256_add_ps(fix1,tx);
1808             fiy1             = _mm256_add_ps(fiy1,ty);
1809             fiz1             = _mm256_add_ps(fiz1,tz);
1810
1811             fjx2             = _mm256_add_ps(fjx2,tx);
1812             fjy2             = _mm256_add_ps(fjy2,ty);
1813             fjz2             = _mm256_add_ps(fjz2,tz);
1814
1815             /**************************
1816              * CALCULATE INTERACTIONS *
1817              **************************/
1818
1819             r13              = _mm256_mul_ps(rsq13,rinv13);
1820
1821             /* EWALD ELECTROSTATICS */
1822
1823             /* Analytical PME correction */
1824             zeta2            = _mm256_mul_ps(beta2,rsq13);
1825             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
1826             pmecorrF         = avx256_pmecorrF_f(zeta2);
1827             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1828             felec            = _mm256_mul_ps(qq13,felec);
1829
1830             fscal            = felec;
1831
1832             /* Calculate temporary vectorial force */
1833             tx               = _mm256_mul_ps(fscal,dx13);
1834             ty               = _mm256_mul_ps(fscal,dy13);
1835             tz               = _mm256_mul_ps(fscal,dz13);
1836
1837             /* Update vectorial force */
1838             fix1             = _mm256_add_ps(fix1,tx);
1839             fiy1             = _mm256_add_ps(fiy1,ty);
1840             fiz1             = _mm256_add_ps(fiz1,tz);
1841
1842             fjx3             = _mm256_add_ps(fjx3,tx);
1843             fjy3             = _mm256_add_ps(fjy3,ty);
1844             fjz3             = _mm256_add_ps(fjz3,tz);
1845
1846             /**************************
1847              * CALCULATE INTERACTIONS *
1848              **************************/
1849
1850             r21              = _mm256_mul_ps(rsq21,rinv21);
1851
1852             /* EWALD ELECTROSTATICS */
1853
1854             /* Analytical PME correction */
1855             zeta2            = _mm256_mul_ps(beta2,rsq21);
1856             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1857             pmecorrF         = avx256_pmecorrF_f(zeta2);
1858             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1859             felec            = _mm256_mul_ps(qq21,felec);
1860
1861             fscal            = felec;
1862
1863             /* Calculate temporary vectorial force */
1864             tx               = _mm256_mul_ps(fscal,dx21);
1865             ty               = _mm256_mul_ps(fscal,dy21);
1866             tz               = _mm256_mul_ps(fscal,dz21);
1867
1868             /* Update vectorial force */
1869             fix2             = _mm256_add_ps(fix2,tx);
1870             fiy2             = _mm256_add_ps(fiy2,ty);
1871             fiz2             = _mm256_add_ps(fiz2,tz);
1872
1873             fjx1             = _mm256_add_ps(fjx1,tx);
1874             fjy1             = _mm256_add_ps(fjy1,ty);
1875             fjz1             = _mm256_add_ps(fjz1,tz);
1876
1877             /**************************
1878              * CALCULATE INTERACTIONS *
1879              **************************/
1880
1881             r22              = _mm256_mul_ps(rsq22,rinv22);
1882
1883             /* EWALD ELECTROSTATICS */
1884
1885             /* Analytical PME correction */
1886             zeta2            = _mm256_mul_ps(beta2,rsq22);
1887             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1888             pmecorrF         = avx256_pmecorrF_f(zeta2);
1889             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1890             felec            = _mm256_mul_ps(qq22,felec);
1891
1892             fscal            = felec;
1893
1894             /* Calculate temporary vectorial force */
1895             tx               = _mm256_mul_ps(fscal,dx22);
1896             ty               = _mm256_mul_ps(fscal,dy22);
1897             tz               = _mm256_mul_ps(fscal,dz22);
1898
1899             /* Update vectorial force */
1900             fix2             = _mm256_add_ps(fix2,tx);
1901             fiy2             = _mm256_add_ps(fiy2,ty);
1902             fiz2             = _mm256_add_ps(fiz2,tz);
1903
1904             fjx2             = _mm256_add_ps(fjx2,tx);
1905             fjy2             = _mm256_add_ps(fjy2,ty);
1906             fjz2             = _mm256_add_ps(fjz2,tz);
1907
1908             /**************************
1909              * CALCULATE INTERACTIONS *
1910              **************************/
1911
1912             r23              = _mm256_mul_ps(rsq23,rinv23);
1913
1914             /* EWALD ELECTROSTATICS */
1915
1916             /* Analytical PME correction */
1917             zeta2            = _mm256_mul_ps(beta2,rsq23);
1918             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1919             pmecorrF         = avx256_pmecorrF_f(zeta2);
1920             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1921             felec            = _mm256_mul_ps(qq23,felec);
1922
1923             fscal            = felec;
1924
1925             /* Calculate temporary vectorial force */
1926             tx               = _mm256_mul_ps(fscal,dx23);
1927             ty               = _mm256_mul_ps(fscal,dy23);
1928             tz               = _mm256_mul_ps(fscal,dz23);
1929
1930             /* Update vectorial force */
1931             fix2             = _mm256_add_ps(fix2,tx);
1932             fiy2             = _mm256_add_ps(fiy2,ty);
1933             fiz2             = _mm256_add_ps(fiz2,tz);
1934
1935             fjx3             = _mm256_add_ps(fjx3,tx);
1936             fjy3             = _mm256_add_ps(fjy3,ty);
1937             fjz3             = _mm256_add_ps(fjz3,tz);
1938
1939             /**************************
1940              * CALCULATE INTERACTIONS *
1941              **************************/
1942
1943             r31              = _mm256_mul_ps(rsq31,rinv31);
1944
1945             /* EWALD ELECTROSTATICS */
1946
1947             /* Analytical PME correction */
1948             zeta2            = _mm256_mul_ps(beta2,rsq31);
1949             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1950             pmecorrF         = avx256_pmecorrF_f(zeta2);
1951             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1952             felec            = _mm256_mul_ps(qq31,felec);
1953
1954             fscal            = felec;
1955
1956             /* Calculate temporary vectorial force */
1957             tx               = _mm256_mul_ps(fscal,dx31);
1958             ty               = _mm256_mul_ps(fscal,dy31);
1959             tz               = _mm256_mul_ps(fscal,dz31);
1960
1961             /* Update vectorial force */
1962             fix3             = _mm256_add_ps(fix3,tx);
1963             fiy3             = _mm256_add_ps(fiy3,ty);
1964             fiz3             = _mm256_add_ps(fiz3,tz);
1965
1966             fjx1             = _mm256_add_ps(fjx1,tx);
1967             fjy1             = _mm256_add_ps(fjy1,ty);
1968             fjz1             = _mm256_add_ps(fjz1,tz);
1969
1970             /**************************
1971              * CALCULATE INTERACTIONS *
1972              **************************/
1973
1974             r32              = _mm256_mul_ps(rsq32,rinv32);
1975
1976             /* EWALD ELECTROSTATICS */
1977
1978             /* Analytical PME correction */
1979             zeta2            = _mm256_mul_ps(beta2,rsq32);
1980             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1981             pmecorrF         = avx256_pmecorrF_f(zeta2);
1982             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1983             felec            = _mm256_mul_ps(qq32,felec);
1984
1985             fscal            = felec;
1986
1987             /* Calculate temporary vectorial force */
1988             tx               = _mm256_mul_ps(fscal,dx32);
1989             ty               = _mm256_mul_ps(fscal,dy32);
1990             tz               = _mm256_mul_ps(fscal,dz32);
1991
1992             /* Update vectorial force */
1993             fix3             = _mm256_add_ps(fix3,tx);
1994             fiy3             = _mm256_add_ps(fiy3,ty);
1995             fiz3             = _mm256_add_ps(fiz3,tz);
1996
1997             fjx2             = _mm256_add_ps(fjx2,tx);
1998             fjy2             = _mm256_add_ps(fjy2,ty);
1999             fjz2             = _mm256_add_ps(fjz2,tz);
2000
2001             /**************************
2002              * CALCULATE INTERACTIONS *
2003              **************************/
2004
2005             r33              = _mm256_mul_ps(rsq33,rinv33);
2006
2007             /* EWALD ELECTROSTATICS */
2008
2009             /* Analytical PME correction */
2010             zeta2            = _mm256_mul_ps(beta2,rsq33);
2011             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
2012             pmecorrF         = avx256_pmecorrF_f(zeta2);
2013             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2014             felec            = _mm256_mul_ps(qq33,felec);
2015
2016             fscal            = felec;
2017
2018             /* Calculate temporary vectorial force */
2019             tx               = _mm256_mul_ps(fscal,dx33);
2020             ty               = _mm256_mul_ps(fscal,dy33);
2021             tz               = _mm256_mul_ps(fscal,dz33);
2022
2023             /* Update vectorial force */
2024             fix3             = _mm256_add_ps(fix3,tx);
2025             fiy3             = _mm256_add_ps(fiy3,ty);
2026             fiz3             = _mm256_add_ps(fiz3,tz);
2027
2028             fjx3             = _mm256_add_ps(fjx3,tx);
2029             fjy3             = _mm256_add_ps(fjy3,ty);
2030             fjz3             = _mm256_add_ps(fjz3,tz);
2031
2032             fjptrA             = f+j_coord_offsetA;
2033             fjptrB             = f+j_coord_offsetB;
2034             fjptrC             = f+j_coord_offsetC;
2035             fjptrD             = f+j_coord_offsetD;
2036             fjptrE             = f+j_coord_offsetE;
2037             fjptrF             = f+j_coord_offsetF;
2038             fjptrG             = f+j_coord_offsetG;
2039             fjptrH             = f+j_coord_offsetH;
2040
2041             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2042                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2043                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2044
2045             /* Inner loop uses 555 flops */
2046         }
2047
2048         if(jidx<j_index_end)
2049         {
2050
2051             /* Get j neighbor index, and coordinate index */
2052             jnrlistA         = jjnr[jidx];
2053             jnrlistB         = jjnr[jidx+1];
2054             jnrlistC         = jjnr[jidx+2];
2055             jnrlistD         = jjnr[jidx+3];
2056             jnrlistE         = jjnr[jidx+4];
2057             jnrlistF         = jjnr[jidx+5];
2058             jnrlistG         = jjnr[jidx+6];
2059             jnrlistH         = jjnr[jidx+7];
2060             /* Sign of each element will be negative for non-real atoms.
2061              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2062              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2063              */
2064             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2065                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2066
2067             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2068             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2069             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2070             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2071             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2072             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2073             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2074             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2075             j_coord_offsetA  = DIM*jnrA;
2076             j_coord_offsetB  = DIM*jnrB;
2077             j_coord_offsetC  = DIM*jnrC;
2078             j_coord_offsetD  = DIM*jnrD;
2079             j_coord_offsetE  = DIM*jnrE;
2080             j_coord_offsetF  = DIM*jnrF;
2081             j_coord_offsetG  = DIM*jnrG;
2082             j_coord_offsetH  = DIM*jnrH;
2083
2084             /* load j atom coordinates */
2085             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2086                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2087                                                  x+j_coord_offsetE,x+j_coord_offsetF,
2088                                                  x+j_coord_offsetG,x+j_coord_offsetH,
2089                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2090                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
2091
2092             /* Calculate displacement vector */
2093             dx00             = _mm256_sub_ps(ix0,jx0);
2094             dy00             = _mm256_sub_ps(iy0,jy0);
2095             dz00             = _mm256_sub_ps(iz0,jz0);
2096             dx11             = _mm256_sub_ps(ix1,jx1);
2097             dy11             = _mm256_sub_ps(iy1,jy1);
2098             dz11             = _mm256_sub_ps(iz1,jz1);
2099             dx12             = _mm256_sub_ps(ix1,jx2);
2100             dy12             = _mm256_sub_ps(iy1,jy2);
2101             dz12             = _mm256_sub_ps(iz1,jz2);
2102             dx13             = _mm256_sub_ps(ix1,jx3);
2103             dy13             = _mm256_sub_ps(iy1,jy3);
2104             dz13             = _mm256_sub_ps(iz1,jz3);
2105             dx21             = _mm256_sub_ps(ix2,jx1);
2106             dy21             = _mm256_sub_ps(iy2,jy1);
2107             dz21             = _mm256_sub_ps(iz2,jz1);
2108             dx22             = _mm256_sub_ps(ix2,jx2);
2109             dy22             = _mm256_sub_ps(iy2,jy2);
2110             dz22             = _mm256_sub_ps(iz2,jz2);
2111             dx23             = _mm256_sub_ps(ix2,jx3);
2112             dy23             = _mm256_sub_ps(iy2,jy3);
2113             dz23             = _mm256_sub_ps(iz2,jz3);
2114             dx31             = _mm256_sub_ps(ix3,jx1);
2115             dy31             = _mm256_sub_ps(iy3,jy1);
2116             dz31             = _mm256_sub_ps(iz3,jz1);
2117             dx32             = _mm256_sub_ps(ix3,jx2);
2118             dy32             = _mm256_sub_ps(iy3,jy2);
2119             dz32             = _mm256_sub_ps(iz3,jz2);
2120             dx33             = _mm256_sub_ps(ix3,jx3);
2121             dy33             = _mm256_sub_ps(iy3,jy3);
2122             dz33             = _mm256_sub_ps(iz3,jz3);
2123
2124             /* Calculate squared distance and things based on it */
2125             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2126             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2127             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2128             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
2129             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2130             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2131             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
2132             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
2133             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
2134             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
2135
2136             rinv00           = avx256_invsqrt_f(rsq00);
2137             rinv11           = avx256_invsqrt_f(rsq11);
2138             rinv12           = avx256_invsqrt_f(rsq12);
2139             rinv13           = avx256_invsqrt_f(rsq13);
2140             rinv21           = avx256_invsqrt_f(rsq21);
2141             rinv22           = avx256_invsqrt_f(rsq22);
2142             rinv23           = avx256_invsqrt_f(rsq23);
2143             rinv31           = avx256_invsqrt_f(rsq31);
2144             rinv32           = avx256_invsqrt_f(rsq32);
2145             rinv33           = avx256_invsqrt_f(rsq33);
2146
2147             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2148             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2149             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
2150             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2151             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2152             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
2153             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
2154             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
2155             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
2156
2157             fjx0             = _mm256_setzero_ps();
2158             fjy0             = _mm256_setzero_ps();
2159             fjz0             = _mm256_setzero_ps();
2160             fjx1             = _mm256_setzero_ps();
2161             fjy1             = _mm256_setzero_ps();
2162             fjz1             = _mm256_setzero_ps();
2163             fjx2             = _mm256_setzero_ps();
2164             fjy2             = _mm256_setzero_ps();
2165             fjz2             = _mm256_setzero_ps();
2166             fjx3             = _mm256_setzero_ps();
2167             fjy3             = _mm256_setzero_ps();
2168             fjz3             = _mm256_setzero_ps();
2169
2170             /**************************
2171              * CALCULATE INTERACTIONS *
2172              **************************/
2173
2174             r00              = _mm256_mul_ps(rsq00,rinv00);
2175             r00              = _mm256_andnot_ps(dummy_mask,r00);
2176
2177             /* Calculate table index by multiplying r with table scale and truncate to integer */
2178             rt               = _mm256_mul_ps(r00,vftabscale);
2179             vfitab           = _mm256_cvttps_epi32(rt);
2180             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2181             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2182             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2183             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2184             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
2185             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
2186
2187             /* CUBIC SPLINE TABLE DISPERSION */
2188             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2189                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2190             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2191                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2192             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2193                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2194             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2195                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2196             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2197             Heps             = _mm256_mul_ps(vfeps,H);
2198             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2199             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2200             fvdw6            = _mm256_mul_ps(c6_00,FF);
2201
2202             /* CUBIC SPLINE TABLE REPULSION */
2203             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
2204             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
2205             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2206                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2207             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2208                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2209             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2210                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2211             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2212                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2213             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2214             Heps             = _mm256_mul_ps(vfeps,H);
2215             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2216             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2217             fvdw12           = _mm256_mul_ps(c12_00,FF);
2218             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
2219
2220             fscal            = fvdw;
2221
2222             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2223
2224             /* Calculate temporary vectorial force */
2225             tx               = _mm256_mul_ps(fscal,dx00);
2226             ty               = _mm256_mul_ps(fscal,dy00);
2227             tz               = _mm256_mul_ps(fscal,dz00);
2228
2229             /* Update vectorial force */
2230             fix0             = _mm256_add_ps(fix0,tx);
2231             fiy0             = _mm256_add_ps(fiy0,ty);
2232             fiz0             = _mm256_add_ps(fiz0,tz);
2233
2234             fjx0             = _mm256_add_ps(fjx0,tx);
2235             fjy0             = _mm256_add_ps(fjy0,ty);
2236             fjz0             = _mm256_add_ps(fjz0,tz);
2237
2238             /**************************
2239              * CALCULATE INTERACTIONS *
2240              **************************/
2241
2242             r11              = _mm256_mul_ps(rsq11,rinv11);
2243             r11              = _mm256_andnot_ps(dummy_mask,r11);
2244
2245             /* EWALD ELECTROSTATICS */
2246
2247             /* Analytical PME correction */
2248             zeta2            = _mm256_mul_ps(beta2,rsq11);
2249             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
2250             pmecorrF         = avx256_pmecorrF_f(zeta2);
2251             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2252             felec            = _mm256_mul_ps(qq11,felec);
2253
2254             fscal            = felec;
2255
2256             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2257
2258             /* Calculate temporary vectorial force */
2259             tx               = _mm256_mul_ps(fscal,dx11);
2260             ty               = _mm256_mul_ps(fscal,dy11);
2261             tz               = _mm256_mul_ps(fscal,dz11);
2262
2263             /* Update vectorial force */
2264             fix1             = _mm256_add_ps(fix1,tx);
2265             fiy1             = _mm256_add_ps(fiy1,ty);
2266             fiz1             = _mm256_add_ps(fiz1,tz);
2267
2268             fjx1             = _mm256_add_ps(fjx1,tx);
2269             fjy1             = _mm256_add_ps(fjy1,ty);
2270             fjz1             = _mm256_add_ps(fjz1,tz);
2271
2272             /**************************
2273              * CALCULATE INTERACTIONS *
2274              **************************/
2275
2276             r12              = _mm256_mul_ps(rsq12,rinv12);
2277             r12              = _mm256_andnot_ps(dummy_mask,r12);
2278
2279             /* EWALD ELECTROSTATICS */
2280
2281             /* Analytical PME correction */
2282             zeta2            = _mm256_mul_ps(beta2,rsq12);
2283             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2284             pmecorrF         = avx256_pmecorrF_f(zeta2);
2285             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2286             felec            = _mm256_mul_ps(qq12,felec);
2287
2288             fscal            = felec;
2289
2290             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2291
2292             /* Calculate temporary vectorial force */
2293             tx               = _mm256_mul_ps(fscal,dx12);
2294             ty               = _mm256_mul_ps(fscal,dy12);
2295             tz               = _mm256_mul_ps(fscal,dz12);
2296
2297             /* Update vectorial force */
2298             fix1             = _mm256_add_ps(fix1,tx);
2299             fiy1             = _mm256_add_ps(fiy1,ty);
2300             fiz1             = _mm256_add_ps(fiz1,tz);
2301
2302             fjx2             = _mm256_add_ps(fjx2,tx);
2303             fjy2             = _mm256_add_ps(fjy2,ty);
2304             fjz2             = _mm256_add_ps(fjz2,tz);
2305
2306             /**************************
2307              * CALCULATE INTERACTIONS *
2308              **************************/
2309
2310             r13              = _mm256_mul_ps(rsq13,rinv13);
2311             r13              = _mm256_andnot_ps(dummy_mask,r13);
2312
2313             /* EWALD ELECTROSTATICS */
2314
2315             /* Analytical PME correction */
2316             zeta2            = _mm256_mul_ps(beta2,rsq13);
2317             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
2318             pmecorrF         = avx256_pmecorrF_f(zeta2);
2319             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2320             felec            = _mm256_mul_ps(qq13,felec);
2321
2322             fscal            = felec;
2323
2324             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2325
2326             /* Calculate temporary vectorial force */
2327             tx               = _mm256_mul_ps(fscal,dx13);
2328             ty               = _mm256_mul_ps(fscal,dy13);
2329             tz               = _mm256_mul_ps(fscal,dz13);
2330
2331             /* Update vectorial force */
2332             fix1             = _mm256_add_ps(fix1,tx);
2333             fiy1             = _mm256_add_ps(fiy1,ty);
2334             fiz1             = _mm256_add_ps(fiz1,tz);
2335
2336             fjx3             = _mm256_add_ps(fjx3,tx);
2337             fjy3             = _mm256_add_ps(fjy3,ty);
2338             fjz3             = _mm256_add_ps(fjz3,tz);
2339
2340             /**************************
2341              * CALCULATE INTERACTIONS *
2342              **************************/
2343
2344             r21              = _mm256_mul_ps(rsq21,rinv21);
2345             r21              = _mm256_andnot_ps(dummy_mask,r21);
2346
2347             /* EWALD ELECTROSTATICS */
2348
2349             /* Analytical PME correction */
2350             zeta2            = _mm256_mul_ps(beta2,rsq21);
2351             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2352             pmecorrF         = avx256_pmecorrF_f(zeta2);
2353             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2354             felec            = _mm256_mul_ps(qq21,felec);
2355
2356             fscal            = felec;
2357
2358             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2359
2360             /* Calculate temporary vectorial force */
2361             tx               = _mm256_mul_ps(fscal,dx21);
2362             ty               = _mm256_mul_ps(fscal,dy21);
2363             tz               = _mm256_mul_ps(fscal,dz21);
2364
2365             /* Update vectorial force */
2366             fix2             = _mm256_add_ps(fix2,tx);
2367             fiy2             = _mm256_add_ps(fiy2,ty);
2368             fiz2             = _mm256_add_ps(fiz2,tz);
2369
2370             fjx1             = _mm256_add_ps(fjx1,tx);
2371             fjy1             = _mm256_add_ps(fjy1,ty);
2372             fjz1             = _mm256_add_ps(fjz1,tz);
2373
2374             /**************************
2375              * CALCULATE INTERACTIONS *
2376              **************************/
2377
2378             r22              = _mm256_mul_ps(rsq22,rinv22);
2379             r22              = _mm256_andnot_ps(dummy_mask,r22);
2380
2381             /* EWALD ELECTROSTATICS */
2382
2383             /* Analytical PME correction */
2384             zeta2            = _mm256_mul_ps(beta2,rsq22);
2385             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2386             pmecorrF         = avx256_pmecorrF_f(zeta2);
2387             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2388             felec            = _mm256_mul_ps(qq22,felec);
2389
2390             fscal            = felec;
2391
2392             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2393
2394             /* Calculate temporary vectorial force */
2395             tx               = _mm256_mul_ps(fscal,dx22);
2396             ty               = _mm256_mul_ps(fscal,dy22);
2397             tz               = _mm256_mul_ps(fscal,dz22);
2398
2399             /* Update vectorial force */
2400             fix2             = _mm256_add_ps(fix2,tx);
2401             fiy2             = _mm256_add_ps(fiy2,ty);
2402             fiz2             = _mm256_add_ps(fiz2,tz);
2403
2404             fjx2             = _mm256_add_ps(fjx2,tx);
2405             fjy2             = _mm256_add_ps(fjy2,ty);
2406             fjz2             = _mm256_add_ps(fjz2,tz);
2407
2408             /**************************
2409              * CALCULATE INTERACTIONS *
2410              **************************/
2411
2412             r23              = _mm256_mul_ps(rsq23,rinv23);
2413             r23              = _mm256_andnot_ps(dummy_mask,r23);
2414
2415             /* EWALD ELECTROSTATICS */
2416
2417             /* Analytical PME correction */
2418             zeta2            = _mm256_mul_ps(beta2,rsq23);
2419             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
2420             pmecorrF         = avx256_pmecorrF_f(zeta2);
2421             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2422             felec            = _mm256_mul_ps(qq23,felec);
2423
2424             fscal            = felec;
2425
2426             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2427
2428             /* Calculate temporary vectorial force */
2429             tx               = _mm256_mul_ps(fscal,dx23);
2430             ty               = _mm256_mul_ps(fscal,dy23);
2431             tz               = _mm256_mul_ps(fscal,dz23);
2432
2433             /* Update vectorial force */
2434             fix2             = _mm256_add_ps(fix2,tx);
2435             fiy2             = _mm256_add_ps(fiy2,ty);
2436             fiz2             = _mm256_add_ps(fiz2,tz);
2437
2438             fjx3             = _mm256_add_ps(fjx3,tx);
2439             fjy3             = _mm256_add_ps(fjy3,ty);
2440             fjz3             = _mm256_add_ps(fjz3,tz);
2441
2442             /**************************
2443              * CALCULATE INTERACTIONS *
2444              **************************/
2445
2446             r31              = _mm256_mul_ps(rsq31,rinv31);
2447             r31              = _mm256_andnot_ps(dummy_mask,r31);
2448
2449             /* EWALD ELECTROSTATICS */
2450
2451             /* Analytical PME correction */
2452             zeta2            = _mm256_mul_ps(beta2,rsq31);
2453             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
2454             pmecorrF         = avx256_pmecorrF_f(zeta2);
2455             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2456             felec            = _mm256_mul_ps(qq31,felec);
2457
2458             fscal            = felec;
2459
2460             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2461
2462             /* Calculate temporary vectorial force */
2463             tx               = _mm256_mul_ps(fscal,dx31);
2464             ty               = _mm256_mul_ps(fscal,dy31);
2465             tz               = _mm256_mul_ps(fscal,dz31);
2466
2467             /* Update vectorial force */
2468             fix3             = _mm256_add_ps(fix3,tx);
2469             fiy3             = _mm256_add_ps(fiy3,ty);
2470             fiz3             = _mm256_add_ps(fiz3,tz);
2471
2472             fjx1             = _mm256_add_ps(fjx1,tx);
2473             fjy1             = _mm256_add_ps(fjy1,ty);
2474             fjz1             = _mm256_add_ps(fjz1,tz);
2475
2476             /**************************
2477              * CALCULATE INTERACTIONS *
2478              **************************/
2479
2480             r32              = _mm256_mul_ps(rsq32,rinv32);
2481             r32              = _mm256_andnot_ps(dummy_mask,r32);
2482
2483             /* EWALD ELECTROSTATICS */
2484
2485             /* Analytical PME correction */
2486             zeta2            = _mm256_mul_ps(beta2,rsq32);
2487             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
2488             pmecorrF         = avx256_pmecorrF_f(zeta2);
2489             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2490             felec            = _mm256_mul_ps(qq32,felec);
2491
2492             fscal            = felec;
2493
2494             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2495
2496             /* Calculate temporary vectorial force */
2497             tx               = _mm256_mul_ps(fscal,dx32);
2498             ty               = _mm256_mul_ps(fscal,dy32);
2499             tz               = _mm256_mul_ps(fscal,dz32);
2500
2501             /* Update vectorial force */
2502             fix3             = _mm256_add_ps(fix3,tx);
2503             fiy3             = _mm256_add_ps(fiy3,ty);
2504             fiz3             = _mm256_add_ps(fiz3,tz);
2505
2506             fjx2             = _mm256_add_ps(fjx2,tx);
2507             fjy2             = _mm256_add_ps(fjy2,ty);
2508             fjz2             = _mm256_add_ps(fjz2,tz);
2509
2510             /**************************
2511              * CALCULATE INTERACTIONS *
2512              **************************/
2513
2514             r33              = _mm256_mul_ps(rsq33,rinv33);
2515             r33              = _mm256_andnot_ps(dummy_mask,r33);
2516
2517             /* EWALD ELECTROSTATICS */
2518
2519             /* Analytical PME correction */
2520             zeta2            = _mm256_mul_ps(beta2,rsq33);
2521             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
2522             pmecorrF         = avx256_pmecorrF_f(zeta2);
2523             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2524             felec            = _mm256_mul_ps(qq33,felec);
2525
2526             fscal            = felec;
2527
2528             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2529
2530             /* Calculate temporary vectorial force */
2531             tx               = _mm256_mul_ps(fscal,dx33);
2532             ty               = _mm256_mul_ps(fscal,dy33);
2533             tz               = _mm256_mul_ps(fscal,dz33);
2534
2535             /* Update vectorial force */
2536             fix3             = _mm256_add_ps(fix3,tx);
2537             fiy3             = _mm256_add_ps(fiy3,ty);
2538             fiz3             = _mm256_add_ps(fiz3,tz);
2539
2540             fjx3             = _mm256_add_ps(fjx3,tx);
2541             fjy3             = _mm256_add_ps(fjy3,ty);
2542             fjz3             = _mm256_add_ps(fjz3,tz);
2543
2544             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2545             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2546             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2547             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2548             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2549             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2550             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2551             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2552
2553             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2554                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2555                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2556
2557             /* Inner loop uses 565 flops */
2558         }
2559
2560         /* End of innermost loop */
2561
2562         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2563                                                  f+i_coord_offset,fshift+i_shift_offset);
2564
2565         /* Increment number of inner iterations */
2566         inneriter                  += j_index_end - j_index_start;
2567
2568         /* Outer loop uses 24 flops */
2569     }
2570
2571     /* Increment number of outer iterations */
2572     outeriter        += nri;
2573
2574     /* Update outer/inner flops */
2575
2576     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*565);
2577 }