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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4W4_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwCSTab_GeomW4W4_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     real *           vdwioffsetptr3;
  95     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  96     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  97     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  98     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  99     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 100     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
 101     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 102     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
 103     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 104     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 105     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 106     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 107     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 108     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 109     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 110     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 111     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 112     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 113     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 114     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 115     real             *charge;
 116     int              nvdwtype;
 117     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 118     int              *vdwtype;
 119     real             *vdwparam;
 120     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 121     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 122     __m256i          vfitab;
 123     __m128i          vfitab_lo,vfitab_hi;
 124     __m128i          ifour       = _mm_set1_epi32(4);
 125     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 126     real             *vftab;
 127     __m256i          ewitab;
 128     __m128i          ewitab_lo,ewitab_hi;
 129     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 130     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 131     real             *ewtab;
 132     __m256           dummy_mask,cutoff_mask;
 133     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 134     __m256           one     = _mm256_set1_ps(1.0);
 135     __m256           two     = _mm256_set1_ps(2.0);
 136     x                = xx[0];
 137     f                = ff[0];
 138
 139     nri              = nlist->nri;
 140     iinr             = nlist->iinr;
 141     jindex           = nlist->jindex;
 142     jjnr             = nlist->jjnr;
 143     shiftidx         = nlist->shift;
 144     gid              = nlist->gid;
 145     shiftvec         = fr->shift_vec[0];
 146     fshift           = fr->fshift[0];
 147     facel            = _mm256_set1_ps(fr->epsfac);
 148     charge           = mdatoms->chargeA;
 149     nvdwtype         = fr->ntype;
 150     vdwparam         = fr->nbfp;
 151     vdwtype          = mdatoms->typeA;
 152
 153     vftab            = kernel_data->table_vdw->data;
 154     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 155
 156     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 157     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 158     beta2            = _mm256_mul_ps(beta,beta);
 159     beta3            = _mm256_mul_ps(beta,beta2);
 160
 161     ewtab            = fr->ic->tabq_coul_FDV0;
 162     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 163     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 164
 165     /* Setup water-specific parameters */
 166     inr              = nlist->iinr[0];
 167     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 168     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 169     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 170     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 171
 172     jq1              = _mm256_set1_ps(charge[inr+1]);
 173     jq2              = _mm256_set1_ps(charge[inr+2]);
 174     jq3              = _mm256_set1_ps(charge[inr+3]);
 175     vdwjidx0A        = 2*vdwtype[inr+0];
 176     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 177     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 178     qq11             = _mm256_mul_ps(iq1,jq1);
 179     qq12             = _mm256_mul_ps(iq1,jq2);
 180     qq13             = _mm256_mul_ps(iq1,jq3);
 181     qq21             = _mm256_mul_ps(iq2,jq1);
 182     qq22             = _mm256_mul_ps(iq2,jq2);
 183     qq23             = _mm256_mul_ps(iq2,jq3);
 184     qq31             = _mm256_mul_ps(iq3,jq1);
 185     qq32             = _mm256_mul_ps(iq3,jq2);
 186     qq33             = _mm256_mul_ps(iq3,jq3);
 187
 188     /* Avoid stupid compiler warnings */
 189     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 190     j_coord_offsetA = 0;
 191     j_coord_offsetB = 0;
 192     j_coord_offsetC = 0;
 193     j_coord_offsetD = 0;
 194     j_coord_offsetE = 0;
 195     j_coord_offsetF = 0;
 196     j_coord_offsetG = 0;
 197     j_coord_offsetH = 0;
 198
 199     outeriter        = 0;
 200     inneriter        = 0;
 201
 202     for(iidx=0;iidx<4*DIM;iidx++)
 203     {
 204         scratch[iidx] = 0.0;
 205     }
 206
 207     /* Start outer loop over neighborlists */
 208     for(iidx=0; iidx<nri; iidx++)
 209     {
 210         /* Load shift vector for this list */
 211         i_shift_offset   = DIM*shiftidx[iidx];
 212
 213         /* Load limits for loop over neighbors */
 214         j_index_start    = jindex[iidx];
 215         j_index_end      = jindex[iidx+1];
 216
 217         /* Get outer coordinate index */
 218         inr              = iinr[iidx];
 219         i_coord_offset   = DIM*inr;
 220
 221         /* Load i particle coords and add shift vector */
 222         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 223                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 224
 225         fix0             = _mm256_setzero_ps();
 226         fiy0             = _mm256_setzero_ps();
 227         fiz0             = _mm256_setzero_ps();
 228         fix1             = _mm256_setzero_ps();
 229         fiy1             = _mm256_setzero_ps();
 230         fiz1             = _mm256_setzero_ps();
 231         fix2             = _mm256_setzero_ps();
 232         fiy2             = _mm256_setzero_ps();
 233         fiz2             = _mm256_setzero_ps();
 234         fix3             = _mm256_setzero_ps();
 235         fiy3             = _mm256_setzero_ps();
 236         fiz3             = _mm256_setzero_ps();
 237
 238         /* Reset potential sums */
 239         velecsum         = _mm256_setzero_ps();
 240         vvdwsum          = _mm256_setzero_ps();
 241
 242         /* Start inner kernel loop */
 243         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 244         {
 245
 246             /* Get j neighbor index, and coordinate index */
 247             jnrA             = jjnr[jidx];
 248             jnrB             = jjnr[jidx+1];
 249             jnrC             = jjnr[jidx+2];
 250             jnrD             = jjnr[jidx+3];
 251             jnrE             = jjnr[jidx+4];
 252             jnrF             = jjnr[jidx+5];
 253             jnrG             = jjnr[jidx+6];
 254             jnrH             = jjnr[jidx+7];
 255             j_coord_offsetA  = DIM*jnrA;
 256             j_coord_offsetB  = DIM*jnrB;
 257             j_coord_offsetC  = DIM*jnrC;
 258             j_coord_offsetD  = DIM*jnrD;
 259             j_coord_offsetE  = DIM*jnrE;
 260             j_coord_offsetF  = DIM*jnrF;
 261             j_coord_offsetG  = DIM*jnrG;
 262             j_coord_offsetH  = DIM*jnrH;
 263
 264             /* load j atom coordinates */
 265             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 266                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 267                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 268                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 269                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 270                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 271
 272             /* Calculate displacement vector */
 273             dx00             = _mm256_sub_ps(ix0,jx0);
 274             dy00             = _mm256_sub_ps(iy0,jy0);
 275             dz00             = _mm256_sub_ps(iz0,jz0);
 276             dx11             = _mm256_sub_ps(ix1,jx1);
 277             dy11             = _mm256_sub_ps(iy1,jy1);
 278             dz11             = _mm256_sub_ps(iz1,jz1);
 279             dx12             = _mm256_sub_ps(ix1,jx2);
 280             dy12             = _mm256_sub_ps(iy1,jy2);
 281             dz12             = _mm256_sub_ps(iz1,jz2);
 282             dx13             = _mm256_sub_ps(ix1,jx3);
 283             dy13             = _mm256_sub_ps(iy1,jy3);
 284             dz13             = _mm256_sub_ps(iz1,jz3);
 285             dx21             = _mm256_sub_ps(ix2,jx1);
 286             dy21             = _mm256_sub_ps(iy2,jy1);
 287             dz21             = _mm256_sub_ps(iz2,jz1);
 288             dx22             = _mm256_sub_ps(ix2,jx2);
 289             dy22             = _mm256_sub_ps(iy2,jy2);
 290             dz22             = _mm256_sub_ps(iz2,jz2);
 291             dx23             = _mm256_sub_ps(ix2,jx3);
 292             dy23             = _mm256_sub_ps(iy2,jy3);
 293             dz23             = _mm256_sub_ps(iz2,jz3);
 294             dx31             = _mm256_sub_ps(ix3,jx1);
 295             dy31             = _mm256_sub_ps(iy3,jy1);
 296             dz31             = _mm256_sub_ps(iz3,jz1);
 297             dx32             = _mm256_sub_ps(ix3,jx2);
 298             dy32             = _mm256_sub_ps(iy3,jy2);
 299             dz32             = _mm256_sub_ps(iz3,jz2);
 300             dx33             = _mm256_sub_ps(ix3,jx3);
 301             dy33             = _mm256_sub_ps(iy3,jy3);
 302             dz33             = _mm256_sub_ps(iz3,jz3);
 303
 304             /* Calculate squared distance and things based on it */
 305             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 306             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 307             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 308             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 309             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 310             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 311             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 312             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 313             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 314             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 315
 316             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 317             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 318             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 319             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 320             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 321             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 322             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 323             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 324             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 325             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 326
 327             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 328             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 329             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 330             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 331             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 332             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 333             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 334             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 335             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 336
 337             fjx0             = _mm256_setzero_ps();
 338             fjy0             = _mm256_setzero_ps();
 339             fjz0             = _mm256_setzero_ps();
 340             fjx1             = _mm256_setzero_ps();
 341             fjy1             = _mm256_setzero_ps();
 342             fjz1             = _mm256_setzero_ps();
 343             fjx2             = _mm256_setzero_ps();
 344             fjy2             = _mm256_setzero_ps();
 345             fjz2             = _mm256_setzero_ps();
 346             fjx3             = _mm256_setzero_ps();
 347             fjy3             = _mm256_setzero_ps();
 348             fjz3             = _mm256_setzero_ps();
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             r00              = _mm256_mul_ps(rsq00,rinv00);
 355
 356             /* Calculate table index by multiplying r with table scale and truncate to integer */
 357             rt               = _mm256_mul_ps(r00,vftabscale);
 358             vfitab           = _mm256_cvttps_epi32(rt);
 359             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 360             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 361             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 362             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 363             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 364             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 365
 366             /* CUBIC SPLINE TABLE DISPERSION */
 367             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 368                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 369             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 370                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 371             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 372                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 373             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 374                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 375             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 376             Heps             = _mm256_mul_ps(vfeps,H);
 377             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 378             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 379             vvdw6            = _mm256_mul_ps(c6_00,VV);
 380             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 381             fvdw6            = _mm256_mul_ps(c6_00,FF);
 382
 383             /* CUBIC SPLINE TABLE REPULSION */
 384             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 385             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 386             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 387                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 388             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 389                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 390             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 391                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 392             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 393                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 394             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 395             Heps             = _mm256_mul_ps(vfeps,H);
 396             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 397             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 398             vvdw12           = _mm256_mul_ps(c12_00,VV);
 399             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 400             fvdw12           = _mm256_mul_ps(c12_00,FF);
 401             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 402             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 406
 407             fscal            = fvdw;
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm256_mul_ps(fscal,dx00);
 411             ty               = _mm256_mul_ps(fscal,dy00);
 412             tz               = _mm256_mul_ps(fscal,dz00);
 413
 414             /* Update vectorial force */
 415             fix0             = _mm256_add_ps(fix0,tx);
 416             fiy0             = _mm256_add_ps(fiy0,ty);
 417             fiz0             = _mm256_add_ps(fiz0,tz);
 418
 419             fjx0             = _mm256_add_ps(fjx0,tx);
 420             fjy0             = _mm256_add_ps(fjy0,ty);
 421             fjz0             = _mm256_add_ps(fjz0,tz);
 422
 423             /**************************
 424              * CALCULATE INTERACTIONS *
 425              **************************/
 426
 427             r11              = _mm256_mul_ps(rsq11,rinv11);
 428
 429             /* EWALD ELECTROSTATICS */
 430
 431             /* Analytical PME correction */
 432             zeta2            = _mm256_mul_ps(beta2,rsq11);
 433             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 434             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 435             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 436             felec            = _mm256_mul_ps(qq11,felec);
 437             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 438             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 439             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 440             velec            = _mm256_mul_ps(qq11,velec);
 441
 442             /* Update potential sum for this i atom from the interaction with this j atom. */
 443             velecsum         = _mm256_add_ps(velecsum,velec);
 444
 445             fscal            = felec;
 446
 447             /* Calculate temporary vectorial force */
 448             tx               = _mm256_mul_ps(fscal,dx11);
 449             ty               = _mm256_mul_ps(fscal,dy11);
 450             tz               = _mm256_mul_ps(fscal,dz11);
 451
 452             /* Update vectorial force */
 453             fix1             = _mm256_add_ps(fix1,tx);
 454             fiy1             = _mm256_add_ps(fiy1,ty);
 455             fiz1             = _mm256_add_ps(fiz1,tz);
 456
 457             fjx1             = _mm256_add_ps(fjx1,tx);
 458             fjy1             = _mm256_add_ps(fjy1,ty);
 459             fjz1             = _mm256_add_ps(fjz1,tz);
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             r12              = _mm256_mul_ps(rsq12,rinv12);
 466
 467             /* EWALD ELECTROSTATICS */
 468
 469             /* Analytical PME correction */
 470             zeta2            = _mm256_mul_ps(beta2,rsq12);
 471             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 472             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 473             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 474             felec            = _mm256_mul_ps(qq12,felec);
 475             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 476             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 477             velec            = _mm256_sub_ps(rinv12,pmecorrV);
 478             velec            = _mm256_mul_ps(qq12,velec);
 479
 480             /* Update potential sum for this i atom from the interaction with this j atom. */
 481             velecsum         = _mm256_add_ps(velecsum,velec);
 482
 483             fscal            = felec;
 484
 485             /* Calculate temporary vectorial force */
 486             tx               = _mm256_mul_ps(fscal,dx12);
 487             ty               = _mm256_mul_ps(fscal,dy12);
 488             tz               = _mm256_mul_ps(fscal,dz12);
 489
 490             /* Update vectorial force */
 491             fix1             = _mm256_add_ps(fix1,tx);
 492             fiy1             = _mm256_add_ps(fiy1,ty);
 493             fiz1             = _mm256_add_ps(fiz1,tz);
 494
 495             fjx2             = _mm256_add_ps(fjx2,tx);
 496             fjy2             = _mm256_add_ps(fjy2,ty);
 497             fjz2             = _mm256_add_ps(fjz2,tz);
 498
 499             /**************************
 500              * CALCULATE INTERACTIONS *
 501              **************************/
 502
 503             r13              = _mm256_mul_ps(rsq13,rinv13);
 504
 505             /* EWALD ELECTROSTATICS */
 506
 507             /* Analytical PME correction */
 508             zeta2            = _mm256_mul_ps(beta2,rsq13);
 509             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
 510             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 511             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 512             felec            = _mm256_mul_ps(qq13,felec);
 513             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 514             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 515             velec            = _mm256_sub_ps(rinv13,pmecorrV);
 516             velec            = _mm256_mul_ps(qq13,velec);
 517
 518             /* Update potential sum for this i atom from the interaction with this j atom. */
 519             velecsum         = _mm256_add_ps(velecsum,velec);
 520
 521             fscal            = felec;
 522
 523             /* Calculate temporary vectorial force */
 524             tx               = _mm256_mul_ps(fscal,dx13);
 525             ty               = _mm256_mul_ps(fscal,dy13);
 526             tz               = _mm256_mul_ps(fscal,dz13);
 527
 528             /* Update vectorial force */
 529             fix1             = _mm256_add_ps(fix1,tx);
 530             fiy1             = _mm256_add_ps(fiy1,ty);
 531             fiz1             = _mm256_add_ps(fiz1,tz);
 532
 533             fjx3             = _mm256_add_ps(fjx3,tx);
 534             fjy3             = _mm256_add_ps(fjy3,ty);
 535             fjz3             = _mm256_add_ps(fjz3,tz);
 536
 537             /**************************
 538              * CALCULATE INTERACTIONS *
 539              **************************/
 540
 541             r21              = _mm256_mul_ps(rsq21,rinv21);
 542
 543             /* EWALD ELECTROSTATICS */
 544
 545             /* Analytical PME correction */
 546             zeta2            = _mm256_mul_ps(beta2,rsq21);
 547             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 548             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 549             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 550             felec            = _mm256_mul_ps(qq21,felec);
 551             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 552             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 553             velec            = _mm256_sub_ps(rinv21,pmecorrV);
 554             velec            = _mm256_mul_ps(qq21,velec);
 555
 556             /* Update potential sum for this i atom from the interaction with this j atom. */
 557             velecsum         = _mm256_add_ps(velecsum,velec);
 558
 559             fscal            = felec;
 560
 561             /* Calculate temporary vectorial force */
 562             tx               = _mm256_mul_ps(fscal,dx21);
 563             ty               = _mm256_mul_ps(fscal,dy21);
 564             tz               = _mm256_mul_ps(fscal,dz21);
 565
 566             /* Update vectorial force */
 567             fix2             = _mm256_add_ps(fix2,tx);
 568             fiy2             = _mm256_add_ps(fiy2,ty);
 569             fiz2             = _mm256_add_ps(fiz2,tz);
 570
 571             fjx1             = _mm256_add_ps(fjx1,tx);
 572             fjy1             = _mm256_add_ps(fjy1,ty);
 573             fjz1             = _mm256_add_ps(fjz1,tz);
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             r22              = _mm256_mul_ps(rsq22,rinv22);
 580
 581             /* EWALD ELECTROSTATICS */
 582
 583             /* Analytical PME correction */
 584             zeta2            = _mm256_mul_ps(beta2,rsq22);
 585             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 586             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 587             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 588             felec            = _mm256_mul_ps(qq22,felec);
 589             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 590             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 591             velec            = _mm256_sub_ps(rinv22,pmecorrV);
 592             velec            = _mm256_mul_ps(qq22,velec);
 593
 594             /* Update potential sum for this i atom from the interaction with this j atom. */
 595             velecsum         = _mm256_add_ps(velecsum,velec);
 596
 597             fscal            = felec;
 598
 599             /* Calculate temporary vectorial force */
 600             tx               = _mm256_mul_ps(fscal,dx22);
 601             ty               = _mm256_mul_ps(fscal,dy22);
 602             tz               = _mm256_mul_ps(fscal,dz22);
 603
 604             /* Update vectorial force */
 605             fix2             = _mm256_add_ps(fix2,tx);
 606             fiy2             = _mm256_add_ps(fiy2,ty);
 607             fiz2             = _mm256_add_ps(fiz2,tz);
 608
 609             fjx2             = _mm256_add_ps(fjx2,tx);
 610             fjy2             = _mm256_add_ps(fjy2,ty);
 611             fjz2             = _mm256_add_ps(fjz2,tz);
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             r23              = _mm256_mul_ps(rsq23,rinv23);
 618
 619             /* EWALD ELECTROSTATICS */
 620
 621             /* Analytical PME correction */
 622             zeta2            = _mm256_mul_ps(beta2,rsq23);
 623             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
 624             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 625             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 626             felec            = _mm256_mul_ps(qq23,felec);
 627             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 628             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 629             velec            = _mm256_sub_ps(rinv23,pmecorrV);
 630             velec            = _mm256_mul_ps(qq23,velec);
 631
 632             /* Update potential sum for this i atom from the interaction with this j atom. */
 633             velecsum         = _mm256_add_ps(velecsum,velec);
 634
 635             fscal            = felec;
 636
 637             /* Calculate temporary vectorial force */
 638             tx               = _mm256_mul_ps(fscal,dx23);
 639             ty               = _mm256_mul_ps(fscal,dy23);
 640             tz               = _mm256_mul_ps(fscal,dz23);
 641
 642             /* Update vectorial force */
 643             fix2             = _mm256_add_ps(fix2,tx);
 644             fiy2             = _mm256_add_ps(fiy2,ty);
 645             fiz2             = _mm256_add_ps(fiz2,tz);
 646
 647             fjx3             = _mm256_add_ps(fjx3,tx);
 648             fjy3             = _mm256_add_ps(fjy3,ty);
 649             fjz3             = _mm256_add_ps(fjz3,tz);
 650
 651             /**************************
 652              * CALCULATE INTERACTIONS *
 653              **************************/
 654
 655             r31              = _mm256_mul_ps(rsq31,rinv31);
 656
 657             /* EWALD ELECTROSTATICS */
 658
 659             /* Analytical PME correction */
 660             zeta2            = _mm256_mul_ps(beta2,rsq31);
 661             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
 662             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 663             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 664             felec            = _mm256_mul_ps(qq31,felec);
 665             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 666             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 667             velec            = _mm256_sub_ps(rinv31,pmecorrV);
 668             velec            = _mm256_mul_ps(qq31,velec);
 669
 670             /* Update potential sum for this i atom from the interaction with this j atom. */
 671             velecsum         = _mm256_add_ps(velecsum,velec);
 672
 673             fscal            = felec;
 674
 675             /* Calculate temporary vectorial force */
 676             tx               = _mm256_mul_ps(fscal,dx31);
 677             ty               = _mm256_mul_ps(fscal,dy31);
 678             tz               = _mm256_mul_ps(fscal,dz31);
 679
 680             /* Update vectorial force */
 681             fix3             = _mm256_add_ps(fix3,tx);
 682             fiy3             = _mm256_add_ps(fiy3,ty);
 683             fiz3             = _mm256_add_ps(fiz3,tz);
 684
 685             fjx1             = _mm256_add_ps(fjx1,tx);
 686             fjy1             = _mm256_add_ps(fjy1,ty);
 687             fjz1             = _mm256_add_ps(fjz1,tz);
 688
 689             /**************************
 690              * CALCULATE INTERACTIONS *
 691              **************************/
 692
 693             r32              = _mm256_mul_ps(rsq32,rinv32);
 694
 695             /* EWALD ELECTROSTATICS */
 696
 697             /* Analytical PME correction */
 698             zeta2            = _mm256_mul_ps(beta2,rsq32);
 699             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
 700             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 701             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 702             felec            = _mm256_mul_ps(qq32,felec);
 703             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 704             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 705             velec            = _mm256_sub_ps(rinv32,pmecorrV);
 706             velec            = _mm256_mul_ps(qq32,velec);
 707
 708             /* Update potential sum for this i atom from the interaction with this j atom. */
 709             velecsum         = _mm256_add_ps(velecsum,velec);
 710
 711             fscal            = felec;
 712
 713             /* Calculate temporary vectorial force */
 714             tx               = _mm256_mul_ps(fscal,dx32);
 715             ty               = _mm256_mul_ps(fscal,dy32);
 716             tz               = _mm256_mul_ps(fscal,dz32);
 717
 718             /* Update vectorial force */
 719             fix3             = _mm256_add_ps(fix3,tx);
 720             fiy3             = _mm256_add_ps(fiy3,ty);
 721             fiz3             = _mm256_add_ps(fiz3,tz);
 722
 723             fjx2             = _mm256_add_ps(fjx2,tx);
 724             fjy2             = _mm256_add_ps(fjy2,ty);
 725             fjz2             = _mm256_add_ps(fjz2,tz);
 726
 727             /**************************
 728              * CALCULATE INTERACTIONS *
 729              **************************/
 730
 731             r33              = _mm256_mul_ps(rsq33,rinv33);
 732
 733             /* EWALD ELECTROSTATICS */
 734
 735             /* Analytical PME correction */
 736             zeta2            = _mm256_mul_ps(beta2,rsq33);
 737             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
 738             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 739             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 740             felec            = _mm256_mul_ps(qq33,felec);
 741             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 742             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 743             velec            = _mm256_sub_ps(rinv33,pmecorrV);
 744             velec            = _mm256_mul_ps(qq33,velec);
 745
 746             /* Update potential sum for this i atom from the interaction with this j atom. */
 747             velecsum         = _mm256_add_ps(velecsum,velec);
 748
 749             fscal            = felec;
 750
 751             /* Calculate temporary vectorial force */
 752             tx               = _mm256_mul_ps(fscal,dx33);
 753             ty               = _mm256_mul_ps(fscal,dy33);
 754             tz               = _mm256_mul_ps(fscal,dz33);
 755
 756             /* Update vectorial force */
 757             fix3             = _mm256_add_ps(fix3,tx);
 758             fiy3             = _mm256_add_ps(fiy3,ty);
 759             fiz3             = _mm256_add_ps(fiz3,tz);
 760
 761             fjx3             = _mm256_add_ps(fjx3,tx);
 762             fjy3             = _mm256_add_ps(fjy3,ty);
 763             fjz3             = _mm256_add_ps(fjz3,tz);
 764
 765             fjptrA             = f+j_coord_offsetA;
 766             fjptrB             = f+j_coord_offsetB;
 767             fjptrC             = f+j_coord_offsetC;
 768             fjptrD             = f+j_coord_offsetD;
 769             fjptrE             = f+j_coord_offsetE;
 770             fjptrF             = f+j_coord_offsetF;
 771             fjptrG             = f+j_coord_offsetG;
 772             fjptrH             = f+j_coord_offsetH;
 773
 774             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 775                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 776                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 777
 778             /* Inner loop uses 815 flops */
 779         }
 780
 781         if(jidx<j_index_end)
 782         {
 783
 784             /* Get j neighbor index, and coordinate index */
 785             jnrlistA         = jjnr[jidx];
 786             jnrlistB         = jjnr[jidx+1];
 787             jnrlistC         = jjnr[jidx+2];
 788             jnrlistD         = jjnr[jidx+3];
 789             jnrlistE         = jjnr[jidx+4];
 790             jnrlistF         = jjnr[jidx+5];
 791             jnrlistG         = jjnr[jidx+6];
 792             jnrlistH         = jjnr[jidx+7];
 793             /* Sign of each element will be negative for non-real atoms.
 794              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 795              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 796              */
 797             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 798                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 799
 800             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 801             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 802             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 803             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 804             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 805             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 806             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 807             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 808             j_coord_offsetA  = DIM*jnrA;
 809             j_coord_offsetB  = DIM*jnrB;
 810             j_coord_offsetC  = DIM*jnrC;
 811             j_coord_offsetD  = DIM*jnrD;
 812             j_coord_offsetE  = DIM*jnrE;
 813             j_coord_offsetF  = DIM*jnrF;
 814             j_coord_offsetG  = DIM*jnrG;
 815             j_coord_offsetH  = DIM*jnrH;
 816
 817             /* load j atom coordinates */
 818             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 819                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 820                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 821                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 822                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 823                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 824
 825             /* Calculate displacement vector */
 826             dx00             = _mm256_sub_ps(ix0,jx0);
 827             dy00             = _mm256_sub_ps(iy0,jy0);
 828             dz00             = _mm256_sub_ps(iz0,jz0);
 829             dx11             = _mm256_sub_ps(ix1,jx1);
 830             dy11             = _mm256_sub_ps(iy1,jy1);
 831             dz11             = _mm256_sub_ps(iz1,jz1);
 832             dx12             = _mm256_sub_ps(ix1,jx2);
 833             dy12             = _mm256_sub_ps(iy1,jy2);
 834             dz12             = _mm256_sub_ps(iz1,jz2);
 835             dx13             = _mm256_sub_ps(ix1,jx3);
 836             dy13             = _mm256_sub_ps(iy1,jy3);
 837             dz13             = _mm256_sub_ps(iz1,jz3);
 838             dx21             = _mm256_sub_ps(ix2,jx1);
 839             dy21             = _mm256_sub_ps(iy2,jy1);
 840             dz21             = _mm256_sub_ps(iz2,jz1);
 841             dx22             = _mm256_sub_ps(ix2,jx2);
 842             dy22             = _mm256_sub_ps(iy2,jy2);
 843             dz22             = _mm256_sub_ps(iz2,jz2);
 844             dx23             = _mm256_sub_ps(ix2,jx3);
 845             dy23             = _mm256_sub_ps(iy2,jy3);
 846             dz23             = _mm256_sub_ps(iz2,jz3);
 847             dx31             = _mm256_sub_ps(ix3,jx1);
 848             dy31             = _mm256_sub_ps(iy3,jy1);
 849             dz31             = _mm256_sub_ps(iz3,jz1);
 850             dx32             = _mm256_sub_ps(ix3,jx2);
 851             dy32             = _mm256_sub_ps(iy3,jy2);
 852             dz32             = _mm256_sub_ps(iz3,jz2);
 853             dx33             = _mm256_sub_ps(ix3,jx3);
 854             dy33             = _mm256_sub_ps(iy3,jy3);
 855             dz33             = _mm256_sub_ps(iz3,jz3);
 856
 857             /* Calculate squared distance and things based on it */
 858             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 859             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 860             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 861             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 862             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 863             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 864             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 865             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 866             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 867             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 868
 869             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 870             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 871             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 872             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 873             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 874             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 875             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 876             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 877             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 878             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 879
 880             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 881             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 882             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 883             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 884             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 885             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 886             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 887             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 888             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 889
 890             fjx0             = _mm256_setzero_ps();
 891             fjy0             = _mm256_setzero_ps();
 892             fjz0             = _mm256_setzero_ps();
 893             fjx1             = _mm256_setzero_ps();
 894             fjy1             = _mm256_setzero_ps();
 895             fjz1             = _mm256_setzero_ps();
 896             fjx2             = _mm256_setzero_ps();
 897             fjy2             = _mm256_setzero_ps();
 898             fjz2             = _mm256_setzero_ps();
 899             fjx3             = _mm256_setzero_ps();
 900             fjy3             = _mm256_setzero_ps();
 901             fjz3             = _mm256_setzero_ps();
 902
 903             /**************************
 904              * CALCULATE INTERACTIONS *
 905              **************************/
 906
 907             r00              = _mm256_mul_ps(rsq00,rinv00);
 908             r00              = _mm256_andnot_ps(dummy_mask,r00);
 909
 910             /* Calculate table index by multiplying r with table scale and truncate to integer */
 911             rt               = _mm256_mul_ps(r00,vftabscale);
 912             vfitab           = _mm256_cvttps_epi32(rt);
 913             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 914             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 915             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 916             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 917             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 918             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 919
 920             /* CUBIC SPLINE TABLE DISPERSION */
 921             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 922                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 923             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 924                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 925             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 926                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 927             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 928                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 929             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 930             Heps             = _mm256_mul_ps(vfeps,H);
 931             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 932             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 933             vvdw6            = _mm256_mul_ps(c6_00,VV);
 934             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 935             fvdw6            = _mm256_mul_ps(c6_00,FF);
 936
 937             /* CUBIC SPLINE TABLE REPULSION */
 938             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 939             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 940             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 941                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 942             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 943                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 944             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 945                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 946             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 947                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 948             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 949             Heps             = _mm256_mul_ps(vfeps,H);
 950             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 951             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 952             vvdw12           = _mm256_mul_ps(c12_00,VV);
 953             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 954             fvdw12           = _mm256_mul_ps(c12_00,FF);
 955             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 956             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 957
 958             /* Update potential sum for this i atom from the interaction with this j atom. */
 959             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 960             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 961
 962             fscal            = fvdw;
 963
 964             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 965
 966             /* Calculate temporary vectorial force */
 967             tx               = _mm256_mul_ps(fscal,dx00);
 968             ty               = _mm256_mul_ps(fscal,dy00);
 969             tz               = _mm256_mul_ps(fscal,dz00);
 970
 971             /* Update vectorial force */
 972             fix0             = _mm256_add_ps(fix0,tx);
 973             fiy0             = _mm256_add_ps(fiy0,ty);
 974             fiz0             = _mm256_add_ps(fiz0,tz);
 975
 976             fjx0             = _mm256_add_ps(fjx0,tx);
 977             fjy0             = _mm256_add_ps(fjy0,ty);
 978             fjz0             = _mm256_add_ps(fjz0,tz);
 979
 980             /**************************
 981              * CALCULATE INTERACTIONS *
 982              **************************/
 983
 984             r11              = _mm256_mul_ps(rsq11,rinv11);
 985             r11              = _mm256_andnot_ps(dummy_mask,r11);
 986
 987             /* EWALD ELECTROSTATICS */
 988
 989             /* Analytical PME correction */
 990             zeta2            = _mm256_mul_ps(beta2,rsq11);
 991             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 992             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 993             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 994             felec            = _mm256_mul_ps(qq11,felec);
 995             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 996             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 997             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 998             velec            = _mm256_mul_ps(qq11,velec);
 999
1000             /* Update potential sum for this i atom from the interaction with this j atom. */
1001             velec            = _mm256_andnot_ps(dummy_mask,velec);
1002             velecsum         = _mm256_add_ps(velecsum,velec);
1003
1004             fscal            = felec;
1005
1006             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1007
1008             /* Calculate temporary vectorial force */
1009             tx               = _mm256_mul_ps(fscal,dx11);
1010             ty               = _mm256_mul_ps(fscal,dy11);
1011             tz               = _mm256_mul_ps(fscal,dz11);
1012
1013             /* Update vectorial force */
1014             fix1             = _mm256_add_ps(fix1,tx);
1015             fiy1             = _mm256_add_ps(fiy1,ty);
1016             fiz1             = _mm256_add_ps(fiz1,tz);
1017
1018             fjx1             = _mm256_add_ps(fjx1,tx);
1019             fjy1             = _mm256_add_ps(fjy1,ty);
1020             fjz1             = _mm256_add_ps(fjz1,tz);
1021
1022             /**************************
1023              * CALCULATE INTERACTIONS *
1024              **************************/
1025
1026             r12              = _mm256_mul_ps(rsq12,rinv12);
1027             r12              = _mm256_andnot_ps(dummy_mask,r12);
1028
1029             /* EWALD ELECTROSTATICS */
1030
1031             /* Analytical PME correction */
1032             zeta2            = _mm256_mul_ps(beta2,rsq12);
1033             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1034             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1035             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1036             felec            = _mm256_mul_ps(qq12,felec);
1037             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1038             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1039             velec            = _mm256_sub_ps(rinv12,pmecorrV);
1040             velec            = _mm256_mul_ps(qq12,velec);
1041
1042             /* Update potential sum for this i atom from the interaction with this j atom. */
1043             velec            = _mm256_andnot_ps(dummy_mask,velec);
1044             velecsum         = _mm256_add_ps(velecsum,velec);
1045
1046             fscal            = felec;
1047
1048             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1049
1050             /* Calculate temporary vectorial force */
1051             tx               = _mm256_mul_ps(fscal,dx12);
1052             ty               = _mm256_mul_ps(fscal,dy12);
1053             tz               = _mm256_mul_ps(fscal,dz12);
1054
1055             /* Update vectorial force */
1056             fix1             = _mm256_add_ps(fix1,tx);
1057             fiy1             = _mm256_add_ps(fiy1,ty);
1058             fiz1             = _mm256_add_ps(fiz1,tz);
1059
1060             fjx2             = _mm256_add_ps(fjx2,tx);
1061             fjy2             = _mm256_add_ps(fjy2,ty);
1062             fjz2             = _mm256_add_ps(fjz2,tz);
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             r13              = _mm256_mul_ps(rsq13,rinv13);
1069             r13              = _mm256_andnot_ps(dummy_mask,r13);
1070
1071             /* EWALD ELECTROSTATICS */
1072
1073             /* Analytical PME correction */
1074             zeta2            = _mm256_mul_ps(beta2,rsq13);
1075             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
1076             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1077             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1078             felec            = _mm256_mul_ps(qq13,felec);
1079             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1080             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1081             velec            = _mm256_sub_ps(rinv13,pmecorrV);
1082             velec            = _mm256_mul_ps(qq13,velec);
1083
1084             /* Update potential sum for this i atom from the interaction with this j atom. */
1085             velec            = _mm256_andnot_ps(dummy_mask,velec);
1086             velecsum         = _mm256_add_ps(velecsum,velec);
1087
1088             fscal            = felec;
1089
1090             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1091
1092             /* Calculate temporary vectorial force */
1093             tx               = _mm256_mul_ps(fscal,dx13);
1094             ty               = _mm256_mul_ps(fscal,dy13);
1095             tz               = _mm256_mul_ps(fscal,dz13);
1096
1097             /* Update vectorial force */
1098             fix1             = _mm256_add_ps(fix1,tx);
1099             fiy1             = _mm256_add_ps(fiy1,ty);
1100             fiz1             = _mm256_add_ps(fiz1,tz);
1101
1102             fjx3             = _mm256_add_ps(fjx3,tx);
1103             fjy3             = _mm256_add_ps(fjy3,ty);
1104             fjz3             = _mm256_add_ps(fjz3,tz);
1105
1106             /**************************
1107              * CALCULATE INTERACTIONS *
1108              **************************/
1109
1110             r21              = _mm256_mul_ps(rsq21,rinv21);
1111             r21              = _mm256_andnot_ps(dummy_mask,r21);
1112
1113             /* EWALD ELECTROSTATICS */
1114
1115             /* Analytical PME correction */
1116             zeta2            = _mm256_mul_ps(beta2,rsq21);
1117             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1118             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1119             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1120             felec            = _mm256_mul_ps(qq21,felec);
1121             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1122             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1123             velec            = _mm256_sub_ps(rinv21,pmecorrV);
1124             velec            = _mm256_mul_ps(qq21,velec);
1125
1126             /* Update potential sum for this i atom from the interaction with this j atom. */
1127             velec            = _mm256_andnot_ps(dummy_mask,velec);
1128             velecsum         = _mm256_add_ps(velecsum,velec);
1129
1130             fscal            = felec;
1131
1132             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1133
1134             /* Calculate temporary vectorial force */
1135             tx               = _mm256_mul_ps(fscal,dx21);
1136             ty               = _mm256_mul_ps(fscal,dy21);
1137             tz               = _mm256_mul_ps(fscal,dz21);
1138
1139             /* Update vectorial force */
1140             fix2             = _mm256_add_ps(fix2,tx);
1141             fiy2             = _mm256_add_ps(fiy2,ty);
1142             fiz2             = _mm256_add_ps(fiz2,tz);
1143
1144             fjx1             = _mm256_add_ps(fjx1,tx);
1145             fjy1             = _mm256_add_ps(fjy1,ty);
1146             fjz1             = _mm256_add_ps(fjz1,tz);
1147
1148             /**************************
1149              * CALCULATE INTERACTIONS *
1150              **************************/
1151
1152             r22              = _mm256_mul_ps(rsq22,rinv22);
1153             r22              = _mm256_andnot_ps(dummy_mask,r22);
1154
1155             /* EWALD ELECTROSTATICS */
1156
1157             /* Analytical PME correction */
1158             zeta2            = _mm256_mul_ps(beta2,rsq22);
1159             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1160             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1161             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1162             felec            = _mm256_mul_ps(qq22,felec);
1163             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1164             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1165             velec            = _mm256_sub_ps(rinv22,pmecorrV);
1166             velec            = _mm256_mul_ps(qq22,velec);
1167
1168             /* Update potential sum for this i atom from the interaction with this j atom. */
1169             velec            = _mm256_andnot_ps(dummy_mask,velec);
1170             velecsum         = _mm256_add_ps(velecsum,velec);
1171
1172             fscal            = felec;
1173
1174             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1175
1176             /* Calculate temporary vectorial force */
1177             tx               = _mm256_mul_ps(fscal,dx22);
1178             ty               = _mm256_mul_ps(fscal,dy22);
1179             tz               = _mm256_mul_ps(fscal,dz22);
1180
1181             /* Update vectorial force */
1182             fix2             = _mm256_add_ps(fix2,tx);
1183             fiy2             = _mm256_add_ps(fiy2,ty);
1184             fiz2             = _mm256_add_ps(fiz2,tz);
1185
1186             fjx2             = _mm256_add_ps(fjx2,tx);
1187             fjy2             = _mm256_add_ps(fjy2,ty);
1188             fjz2             = _mm256_add_ps(fjz2,tz);
1189
1190             /**************************
1191              * CALCULATE INTERACTIONS *
1192              **************************/
1193
1194             r23              = _mm256_mul_ps(rsq23,rinv23);
1195             r23              = _mm256_andnot_ps(dummy_mask,r23);
1196
1197             /* EWALD ELECTROSTATICS */
1198
1199             /* Analytical PME correction */
1200             zeta2            = _mm256_mul_ps(beta2,rsq23);
1201             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1202             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1203             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1204             felec            = _mm256_mul_ps(qq23,felec);
1205             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1206             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1207             velec            = _mm256_sub_ps(rinv23,pmecorrV);
1208             velec            = _mm256_mul_ps(qq23,velec);
1209
1210             /* Update potential sum for this i atom from the interaction with this j atom. */
1211             velec            = _mm256_andnot_ps(dummy_mask,velec);
1212             velecsum         = _mm256_add_ps(velecsum,velec);
1213
1214             fscal            = felec;
1215
1216             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1217
1218             /* Calculate temporary vectorial force */
1219             tx               = _mm256_mul_ps(fscal,dx23);
1220             ty               = _mm256_mul_ps(fscal,dy23);
1221             tz               = _mm256_mul_ps(fscal,dz23);
1222
1223             /* Update vectorial force */
1224             fix2             = _mm256_add_ps(fix2,tx);
1225             fiy2             = _mm256_add_ps(fiy2,ty);
1226             fiz2             = _mm256_add_ps(fiz2,tz);
1227
1228             fjx3             = _mm256_add_ps(fjx3,tx);
1229             fjy3             = _mm256_add_ps(fjy3,ty);
1230             fjz3             = _mm256_add_ps(fjz3,tz);
1231
1232             /**************************
1233              * CALCULATE INTERACTIONS *
1234              **************************/
1235
1236             r31              = _mm256_mul_ps(rsq31,rinv31);
1237             r31              = _mm256_andnot_ps(dummy_mask,r31);
1238
1239             /* EWALD ELECTROSTATICS */
1240
1241             /* Analytical PME correction */
1242             zeta2            = _mm256_mul_ps(beta2,rsq31);
1243             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1244             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1245             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1246             felec            = _mm256_mul_ps(qq31,felec);
1247             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1248             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1249             velec            = _mm256_sub_ps(rinv31,pmecorrV);
1250             velec            = _mm256_mul_ps(qq31,velec);
1251
1252             /* Update potential sum for this i atom from the interaction with this j atom. */
1253             velec            = _mm256_andnot_ps(dummy_mask,velec);
1254             velecsum         = _mm256_add_ps(velecsum,velec);
1255
1256             fscal            = felec;
1257
1258             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1259
1260             /* Calculate temporary vectorial force */
1261             tx               = _mm256_mul_ps(fscal,dx31);
1262             ty               = _mm256_mul_ps(fscal,dy31);
1263             tz               = _mm256_mul_ps(fscal,dz31);
1264
1265             /* Update vectorial force */
1266             fix3             = _mm256_add_ps(fix3,tx);
1267             fiy3             = _mm256_add_ps(fiy3,ty);
1268             fiz3             = _mm256_add_ps(fiz3,tz);
1269
1270             fjx1             = _mm256_add_ps(fjx1,tx);
1271             fjy1             = _mm256_add_ps(fjy1,ty);
1272             fjz1             = _mm256_add_ps(fjz1,tz);
1273
1274             /**************************
1275              * CALCULATE INTERACTIONS *
1276              **************************/
1277
1278             r32              = _mm256_mul_ps(rsq32,rinv32);
1279             r32              = _mm256_andnot_ps(dummy_mask,r32);
1280
1281             /* EWALD ELECTROSTATICS */
1282
1283             /* Analytical PME correction */
1284             zeta2            = _mm256_mul_ps(beta2,rsq32);
1285             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1286             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1287             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1288             felec            = _mm256_mul_ps(qq32,felec);
1289             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1290             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1291             velec            = _mm256_sub_ps(rinv32,pmecorrV);
1292             velec            = _mm256_mul_ps(qq32,velec);
1293
1294             /* Update potential sum for this i atom from the interaction with this j atom. */
1295             velec            = _mm256_andnot_ps(dummy_mask,velec);
1296             velecsum         = _mm256_add_ps(velecsum,velec);
1297
1298             fscal            = felec;
1299
1300             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1301
1302             /* Calculate temporary vectorial force */
1303             tx               = _mm256_mul_ps(fscal,dx32);
1304             ty               = _mm256_mul_ps(fscal,dy32);
1305             tz               = _mm256_mul_ps(fscal,dz32);
1306
1307             /* Update vectorial force */
1308             fix3             = _mm256_add_ps(fix3,tx);
1309             fiy3             = _mm256_add_ps(fiy3,ty);
1310             fiz3             = _mm256_add_ps(fiz3,tz);
1311
1312             fjx2             = _mm256_add_ps(fjx2,tx);
1313             fjy2             = _mm256_add_ps(fjy2,ty);
1314             fjz2             = _mm256_add_ps(fjz2,tz);
1315
1316             /**************************
1317              * CALCULATE INTERACTIONS *
1318              **************************/
1319
1320             r33              = _mm256_mul_ps(rsq33,rinv33);
1321             r33              = _mm256_andnot_ps(dummy_mask,r33);
1322
1323             /* EWALD ELECTROSTATICS */
1324
1325             /* Analytical PME correction */
1326             zeta2            = _mm256_mul_ps(beta2,rsq33);
1327             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
1328             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1329             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1330             felec            = _mm256_mul_ps(qq33,felec);
1331             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1332             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1333             velec            = _mm256_sub_ps(rinv33,pmecorrV);
1334             velec            = _mm256_mul_ps(qq33,velec);
1335
1336             /* Update potential sum for this i atom from the interaction with this j atom. */
1337             velec            = _mm256_andnot_ps(dummy_mask,velec);
1338             velecsum         = _mm256_add_ps(velecsum,velec);
1339
1340             fscal            = felec;
1341
1342             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1343
1344             /* Calculate temporary vectorial force */
1345             tx               = _mm256_mul_ps(fscal,dx33);
1346             ty               = _mm256_mul_ps(fscal,dy33);
1347             tz               = _mm256_mul_ps(fscal,dz33);
1348
1349             /* Update vectorial force */
1350             fix3             = _mm256_add_ps(fix3,tx);
1351             fiy3             = _mm256_add_ps(fiy3,ty);
1352             fiz3             = _mm256_add_ps(fiz3,tz);
1353
1354             fjx3             = _mm256_add_ps(fjx3,tx);
1355             fjy3             = _mm256_add_ps(fjy3,ty);
1356             fjz3             = _mm256_add_ps(fjz3,tz);
1357
1358             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1359             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1360             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1361             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1362             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1363             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1364             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1365             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1366
1367             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1368                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1369                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1370
1371             /* Inner loop uses 825 flops */
1372         }
1373
1374         /* End of innermost loop */
1375
1376         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1377                                                  f+i_coord_offset,fshift+i_shift_offset);
1378
1379         ggid                        = gid[iidx];
1380         /* Update potential energies */
1381         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1382         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1383
1384         /* Increment number of inner iterations */
1385         inneriter                  += j_index_end - j_index_start;
1386
1387         /* Outer loop uses 26 flops */
1388     }
1389
1390     /* Increment number of outer iterations */
1391     outeriter        += nri;
1392
1393     /* Update outer/inner flops */
1394
1395     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*825);
1396 }
1397 /*
1398  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4W4_F_avx_256_single
1399  * Electrostatics interaction: Ewald
1400  * VdW interaction:            CubicSplineTable
1401  * Geometry:                   Water4-Water4
1402  * Calculate force/pot:        Force
1403  */
1404 void
1405 nb_kernel_ElecEw_VdwCSTab_GeomW4W4_F_avx_256_single
1406                     (t_nblist                    * gmx_restrict       nlist,
1407                      rvec                        * gmx_restrict          xx,
1408                      rvec                        * gmx_restrict          ff,
1409                      t_forcerec                  * gmx_restrict          fr,
1410                      t_mdatoms                   * gmx_restrict     mdatoms,
1411                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1412                      t_nrnb                      * gmx_restrict        nrnb)
1413 {
1414     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1415      * just 0 for non-waters.
1416      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1417      * jnr indices corresponding to data put in the four positions in the SIMD register.
1418      */
1419     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1420     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1421     int              jnrA,jnrB,jnrC,jnrD;
1422     int              jnrE,jnrF,jnrG,jnrH;
1423     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1424     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1425     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1426     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1427     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1428     real             rcutoff_scalar;
1429     real             *shiftvec,*fshift,*x,*f;
1430     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1431     real             scratch[4*DIM];
1432     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1433     real *           vdwioffsetptr0;
1434     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1435     real *           vdwioffsetptr1;
1436     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1437     real *           vdwioffsetptr2;
1438     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1439     real *           vdwioffsetptr3;
1440     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1441     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1442     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1443     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1444     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1445     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1446     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1447     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1448     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1449     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1450     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1451     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1452     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1453     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1454     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1455     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1456     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1457     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1458     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1459     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1460     real             *charge;
1461     int              nvdwtype;
1462     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1463     int              *vdwtype;
1464     real             *vdwparam;
1465     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1466     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1467     __m256i          vfitab;
1468     __m128i          vfitab_lo,vfitab_hi;
1469     __m128i          ifour       = _mm_set1_epi32(4);
1470     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1471     real             *vftab;
1472     __m256i          ewitab;
1473     __m128i          ewitab_lo,ewitab_hi;
1474     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1475     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1476     real             *ewtab;
1477     __m256           dummy_mask,cutoff_mask;
1478     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1479     __m256           one     = _mm256_set1_ps(1.0);
1480     __m256           two     = _mm256_set1_ps(2.0);
1481     x                = xx[0];
1482     f                = ff[0];
1483
1484     nri              = nlist->nri;
1485     iinr             = nlist->iinr;
1486     jindex           = nlist->jindex;
1487     jjnr             = nlist->jjnr;
1488     shiftidx         = nlist->shift;
1489     gid              = nlist->gid;
1490     shiftvec         = fr->shift_vec[0];
1491     fshift           = fr->fshift[0];
1492     facel            = _mm256_set1_ps(fr->epsfac);
1493     charge           = mdatoms->chargeA;
1494     nvdwtype         = fr->ntype;
1495     vdwparam         = fr->nbfp;
1496     vdwtype          = mdatoms->typeA;
1497
1498     vftab            = kernel_data->table_vdw->data;
1499     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
1500
1501     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1502     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1503     beta2            = _mm256_mul_ps(beta,beta);
1504     beta3            = _mm256_mul_ps(beta,beta2);
1505
1506     ewtab            = fr->ic->tabq_coul_F;
1507     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1508     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1509
1510     /* Setup water-specific parameters */
1511     inr              = nlist->iinr[0];
1512     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1513     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1514     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1515     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1516
1517     jq1              = _mm256_set1_ps(charge[inr+1]);
1518     jq2              = _mm256_set1_ps(charge[inr+2]);
1519     jq3              = _mm256_set1_ps(charge[inr+3]);
1520     vdwjidx0A        = 2*vdwtype[inr+0];
1521     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1522     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1523     qq11             = _mm256_mul_ps(iq1,jq1);
1524     qq12             = _mm256_mul_ps(iq1,jq2);
1525     qq13             = _mm256_mul_ps(iq1,jq3);
1526     qq21             = _mm256_mul_ps(iq2,jq1);
1527     qq22             = _mm256_mul_ps(iq2,jq2);
1528     qq23             = _mm256_mul_ps(iq2,jq3);
1529     qq31             = _mm256_mul_ps(iq3,jq1);
1530     qq32             = _mm256_mul_ps(iq3,jq2);
1531     qq33             = _mm256_mul_ps(iq3,jq3);
1532
1533     /* Avoid stupid compiler warnings */
1534     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1535     j_coord_offsetA = 0;
1536     j_coord_offsetB = 0;
1537     j_coord_offsetC = 0;
1538     j_coord_offsetD = 0;
1539     j_coord_offsetE = 0;
1540     j_coord_offsetF = 0;
1541     j_coord_offsetG = 0;
1542     j_coord_offsetH = 0;
1543
1544     outeriter        = 0;
1545     inneriter        = 0;
1546
1547     for(iidx=0;iidx<4*DIM;iidx++)
1548     {
1549         scratch[iidx] = 0.0;
1550     }
1551
1552     /* Start outer loop over neighborlists */
1553     for(iidx=0; iidx<nri; iidx++)
1554     {
1555         /* Load shift vector for this list */
1556         i_shift_offset   = DIM*shiftidx[iidx];
1557
1558         /* Load limits for loop over neighbors */
1559         j_index_start    = jindex[iidx];
1560         j_index_end      = jindex[iidx+1];
1561
1562         /* Get outer coordinate index */
1563         inr              = iinr[iidx];
1564         i_coord_offset   = DIM*inr;
1565
1566         /* Load i particle coords and add shift vector */
1567         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1568                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1569
1570         fix0             = _mm256_setzero_ps();
1571         fiy0             = _mm256_setzero_ps();
1572         fiz0             = _mm256_setzero_ps();
1573         fix1             = _mm256_setzero_ps();
1574         fiy1             = _mm256_setzero_ps();
1575         fiz1             = _mm256_setzero_ps();
1576         fix2             = _mm256_setzero_ps();
1577         fiy2             = _mm256_setzero_ps();
1578         fiz2             = _mm256_setzero_ps();
1579         fix3             = _mm256_setzero_ps();
1580         fiy3             = _mm256_setzero_ps();
1581         fiz3             = _mm256_setzero_ps();
1582
1583         /* Start inner kernel loop */
1584         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1585         {
1586
1587             /* Get j neighbor index, and coordinate index */
1588             jnrA             = jjnr[jidx];
1589             jnrB             = jjnr[jidx+1];
1590             jnrC             = jjnr[jidx+2];
1591             jnrD             = jjnr[jidx+3];
1592             jnrE             = jjnr[jidx+4];
1593             jnrF             = jjnr[jidx+5];
1594             jnrG             = jjnr[jidx+6];
1595             jnrH             = jjnr[jidx+7];
1596             j_coord_offsetA  = DIM*jnrA;
1597             j_coord_offsetB  = DIM*jnrB;
1598             j_coord_offsetC  = DIM*jnrC;
1599             j_coord_offsetD  = DIM*jnrD;
1600             j_coord_offsetE  = DIM*jnrE;
1601             j_coord_offsetF  = DIM*jnrF;
1602             j_coord_offsetG  = DIM*jnrG;
1603             j_coord_offsetH  = DIM*jnrH;
1604
1605             /* load j atom coordinates */
1606             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1607                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1608                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1609                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1610                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1611                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1612
1613             /* Calculate displacement vector */
1614             dx00             = _mm256_sub_ps(ix0,jx0);
1615             dy00             = _mm256_sub_ps(iy0,jy0);
1616             dz00             = _mm256_sub_ps(iz0,jz0);
1617             dx11             = _mm256_sub_ps(ix1,jx1);
1618             dy11             = _mm256_sub_ps(iy1,jy1);
1619             dz11             = _mm256_sub_ps(iz1,jz1);
1620             dx12             = _mm256_sub_ps(ix1,jx2);
1621             dy12             = _mm256_sub_ps(iy1,jy2);
1622             dz12             = _mm256_sub_ps(iz1,jz2);
1623             dx13             = _mm256_sub_ps(ix1,jx3);
1624             dy13             = _mm256_sub_ps(iy1,jy3);
1625             dz13             = _mm256_sub_ps(iz1,jz3);
1626             dx21             = _mm256_sub_ps(ix2,jx1);
1627             dy21             = _mm256_sub_ps(iy2,jy1);
1628             dz21             = _mm256_sub_ps(iz2,jz1);
1629             dx22             = _mm256_sub_ps(ix2,jx2);
1630             dy22             = _mm256_sub_ps(iy2,jy2);
1631             dz22             = _mm256_sub_ps(iz2,jz2);
1632             dx23             = _mm256_sub_ps(ix2,jx3);
1633             dy23             = _mm256_sub_ps(iy2,jy3);
1634             dz23             = _mm256_sub_ps(iz2,jz3);
1635             dx31             = _mm256_sub_ps(ix3,jx1);
1636             dy31             = _mm256_sub_ps(iy3,jy1);
1637             dz31             = _mm256_sub_ps(iz3,jz1);
1638             dx32             = _mm256_sub_ps(ix3,jx2);
1639             dy32             = _mm256_sub_ps(iy3,jy2);
1640             dz32             = _mm256_sub_ps(iz3,jz2);
1641             dx33             = _mm256_sub_ps(ix3,jx3);
1642             dy33             = _mm256_sub_ps(iy3,jy3);
1643             dz33             = _mm256_sub_ps(iz3,jz3);
1644
1645             /* Calculate squared distance and things based on it */
1646             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1647             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1648             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1649             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1650             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1651             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1652             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1653             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1654             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1655             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1656
1657             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1658             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1659             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1660             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
1661             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1662             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1663             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
1664             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
1665             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
1666             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
1667
1668             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1669             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1670             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1671             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1672             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1673             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1674             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1675             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1676             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1677
1678             fjx0             = _mm256_setzero_ps();
1679             fjy0             = _mm256_setzero_ps();
1680             fjz0             = _mm256_setzero_ps();
1681             fjx1             = _mm256_setzero_ps();
1682             fjy1             = _mm256_setzero_ps();
1683             fjz1             = _mm256_setzero_ps();
1684             fjx2             = _mm256_setzero_ps();
1685             fjy2             = _mm256_setzero_ps();
1686             fjz2             = _mm256_setzero_ps();
1687             fjx3             = _mm256_setzero_ps();
1688             fjy3             = _mm256_setzero_ps();
1689             fjz3             = _mm256_setzero_ps();
1690
1691             /**************************
1692              * CALCULATE INTERACTIONS *
1693              **************************/
1694
1695             r00              = _mm256_mul_ps(rsq00,rinv00);
1696
1697             /* Calculate table index by multiplying r with table scale and truncate to integer */
1698             rt               = _mm256_mul_ps(r00,vftabscale);
1699             vfitab           = _mm256_cvttps_epi32(rt);
1700             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1701             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1702             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1703             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1704             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1705             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1706
1707             /* CUBIC SPLINE TABLE DISPERSION */
1708             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1709                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1710             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1711                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1712             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1713                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1714             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1715                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1716             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1717             Heps             = _mm256_mul_ps(vfeps,H);
1718             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1719             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1720             fvdw6            = _mm256_mul_ps(c6_00,FF);
1721
1722             /* CUBIC SPLINE TABLE REPULSION */
1723             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1724             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1725             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1726                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1727             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1728                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1729             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1730                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1731             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1732                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1733             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1734             Heps             = _mm256_mul_ps(vfeps,H);
1735             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1736             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1737             fvdw12           = _mm256_mul_ps(c12_00,FF);
1738             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1739
1740             fscal            = fvdw;
1741
1742             /* Calculate temporary vectorial force */
1743             tx               = _mm256_mul_ps(fscal,dx00);
1744             ty               = _mm256_mul_ps(fscal,dy00);
1745             tz               = _mm256_mul_ps(fscal,dz00);
1746
1747             /* Update vectorial force */
1748             fix0             = _mm256_add_ps(fix0,tx);
1749             fiy0             = _mm256_add_ps(fiy0,ty);
1750             fiz0             = _mm256_add_ps(fiz0,tz);
1751
1752             fjx0             = _mm256_add_ps(fjx0,tx);
1753             fjy0             = _mm256_add_ps(fjy0,ty);
1754             fjz0             = _mm256_add_ps(fjz0,tz);
1755
1756             /**************************
1757              * CALCULATE INTERACTIONS *
1758              **************************/
1759
1760             r11              = _mm256_mul_ps(rsq11,rinv11);
1761
1762             /* EWALD ELECTROSTATICS */
1763
1764             /* Analytical PME correction */
1765             zeta2            = _mm256_mul_ps(beta2,rsq11);
1766             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1767             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1768             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1769             felec            = _mm256_mul_ps(qq11,felec);
1770
1771             fscal            = felec;
1772
1773             /* Calculate temporary vectorial force */
1774             tx               = _mm256_mul_ps(fscal,dx11);
1775             ty               = _mm256_mul_ps(fscal,dy11);
1776             tz               = _mm256_mul_ps(fscal,dz11);
1777
1778             /* Update vectorial force */
1779             fix1             = _mm256_add_ps(fix1,tx);
1780             fiy1             = _mm256_add_ps(fiy1,ty);
1781             fiz1             = _mm256_add_ps(fiz1,tz);
1782
1783             fjx1             = _mm256_add_ps(fjx1,tx);
1784             fjy1             = _mm256_add_ps(fjy1,ty);
1785             fjz1             = _mm256_add_ps(fjz1,tz);
1786
1787             /**************************
1788              * CALCULATE INTERACTIONS *
1789              **************************/
1790
1791             r12              = _mm256_mul_ps(rsq12,rinv12);
1792
1793             /* EWALD ELECTROSTATICS */
1794
1795             /* Analytical PME correction */
1796             zeta2            = _mm256_mul_ps(beta2,rsq12);
1797             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1798             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1799             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1800             felec            = _mm256_mul_ps(qq12,felec);
1801
1802             fscal            = felec;
1803
1804             /* Calculate temporary vectorial force */
1805             tx               = _mm256_mul_ps(fscal,dx12);
1806             ty               = _mm256_mul_ps(fscal,dy12);
1807             tz               = _mm256_mul_ps(fscal,dz12);
1808
1809             /* Update vectorial force */
1810             fix1             = _mm256_add_ps(fix1,tx);
1811             fiy1             = _mm256_add_ps(fiy1,ty);
1812             fiz1             = _mm256_add_ps(fiz1,tz);
1813
1814             fjx2             = _mm256_add_ps(fjx2,tx);
1815             fjy2             = _mm256_add_ps(fjy2,ty);
1816             fjz2             = _mm256_add_ps(fjz2,tz);
1817
1818             /**************************
1819              * CALCULATE INTERACTIONS *
1820              **************************/
1821
1822             r13              = _mm256_mul_ps(rsq13,rinv13);
1823
1824             /* EWALD ELECTROSTATICS */
1825
1826             /* Analytical PME correction */
1827             zeta2            = _mm256_mul_ps(beta2,rsq13);
1828             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
1829             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1830             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1831             felec            = _mm256_mul_ps(qq13,felec);
1832
1833             fscal            = felec;
1834
1835             /* Calculate temporary vectorial force */
1836             tx               = _mm256_mul_ps(fscal,dx13);
1837             ty               = _mm256_mul_ps(fscal,dy13);
1838             tz               = _mm256_mul_ps(fscal,dz13);
1839
1840             /* Update vectorial force */
1841             fix1             = _mm256_add_ps(fix1,tx);
1842             fiy1             = _mm256_add_ps(fiy1,ty);
1843             fiz1             = _mm256_add_ps(fiz1,tz);
1844
1845             fjx3             = _mm256_add_ps(fjx3,tx);
1846             fjy3             = _mm256_add_ps(fjy3,ty);
1847             fjz3             = _mm256_add_ps(fjz3,tz);
1848
1849             /**************************
1850              * CALCULATE INTERACTIONS *
1851              **************************/
1852
1853             r21              = _mm256_mul_ps(rsq21,rinv21);
1854
1855             /* EWALD ELECTROSTATICS */
1856
1857             /* Analytical PME correction */
1858             zeta2            = _mm256_mul_ps(beta2,rsq21);
1859             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1860             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1861             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1862             felec            = _mm256_mul_ps(qq21,felec);
1863
1864             fscal            = felec;
1865
1866             /* Calculate temporary vectorial force */
1867             tx               = _mm256_mul_ps(fscal,dx21);
1868             ty               = _mm256_mul_ps(fscal,dy21);
1869             tz               = _mm256_mul_ps(fscal,dz21);
1870
1871             /* Update vectorial force */
1872             fix2             = _mm256_add_ps(fix2,tx);
1873             fiy2             = _mm256_add_ps(fiy2,ty);
1874             fiz2             = _mm256_add_ps(fiz2,tz);
1875
1876             fjx1             = _mm256_add_ps(fjx1,tx);
1877             fjy1             = _mm256_add_ps(fjy1,ty);
1878             fjz1             = _mm256_add_ps(fjz1,tz);
1879
1880             /**************************
1881              * CALCULATE INTERACTIONS *
1882              **************************/
1883
1884             r22              = _mm256_mul_ps(rsq22,rinv22);
1885
1886             /* EWALD ELECTROSTATICS */
1887
1888             /* Analytical PME correction */
1889             zeta2            = _mm256_mul_ps(beta2,rsq22);
1890             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1891             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1892             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1893             felec            = _mm256_mul_ps(qq22,felec);
1894
1895             fscal            = felec;
1896
1897             /* Calculate temporary vectorial force */
1898             tx               = _mm256_mul_ps(fscal,dx22);
1899             ty               = _mm256_mul_ps(fscal,dy22);
1900             tz               = _mm256_mul_ps(fscal,dz22);
1901
1902             /* Update vectorial force */
1903             fix2             = _mm256_add_ps(fix2,tx);
1904             fiy2             = _mm256_add_ps(fiy2,ty);
1905             fiz2             = _mm256_add_ps(fiz2,tz);
1906
1907             fjx2             = _mm256_add_ps(fjx2,tx);
1908             fjy2             = _mm256_add_ps(fjy2,ty);
1909             fjz2             = _mm256_add_ps(fjz2,tz);
1910
1911             /**************************
1912              * CALCULATE INTERACTIONS *
1913              **************************/
1914
1915             r23              = _mm256_mul_ps(rsq23,rinv23);
1916
1917             /* EWALD ELECTROSTATICS */
1918
1919             /* Analytical PME correction */
1920             zeta2            = _mm256_mul_ps(beta2,rsq23);
1921             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
1922             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1923             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1924             felec            = _mm256_mul_ps(qq23,felec);
1925
1926             fscal            = felec;
1927
1928             /* Calculate temporary vectorial force */
1929             tx               = _mm256_mul_ps(fscal,dx23);
1930             ty               = _mm256_mul_ps(fscal,dy23);
1931             tz               = _mm256_mul_ps(fscal,dz23);
1932
1933             /* Update vectorial force */
1934             fix2             = _mm256_add_ps(fix2,tx);
1935             fiy2             = _mm256_add_ps(fiy2,ty);
1936             fiz2             = _mm256_add_ps(fiz2,tz);
1937
1938             fjx3             = _mm256_add_ps(fjx3,tx);
1939             fjy3             = _mm256_add_ps(fjy3,ty);
1940             fjz3             = _mm256_add_ps(fjz3,tz);
1941
1942             /**************************
1943              * CALCULATE INTERACTIONS *
1944              **************************/
1945
1946             r31              = _mm256_mul_ps(rsq31,rinv31);
1947
1948             /* EWALD ELECTROSTATICS */
1949
1950             /* Analytical PME correction */
1951             zeta2            = _mm256_mul_ps(beta2,rsq31);
1952             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
1953             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1954             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1955             felec            = _mm256_mul_ps(qq31,felec);
1956
1957             fscal            = felec;
1958
1959             /* Calculate temporary vectorial force */
1960             tx               = _mm256_mul_ps(fscal,dx31);
1961             ty               = _mm256_mul_ps(fscal,dy31);
1962             tz               = _mm256_mul_ps(fscal,dz31);
1963
1964             /* Update vectorial force */
1965             fix3             = _mm256_add_ps(fix3,tx);
1966             fiy3             = _mm256_add_ps(fiy3,ty);
1967             fiz3             = _mm256_add_ps(fiz3,tz);
1968
1969             fjx1             = _mm256_add_ps(fjx1,tx);
1970             fjy1             = _mm256_add_ps(fjy1,ty);
1971             fjz1             = _mm256_add_ps(fjz1,tz);
1972
1973             /**************************
1974              * CALCULATE INTERACTIONS *
1975              **************************/
1976
1977             r32              = _mm256_mul_ps(rsq32,rinv32);
1978
1979             /* EWALD ELECTROSTATICS */
1980
1981             /* Analytical PME correction */
1982             zeta2            = _mm256_mul_ps(beta2,rsq32);
1983             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
1984             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1985             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1986             felec            = _mm256_mul_ps(qq32,felec);
1987
1988             fscal            = felec;
1989
1990             /* Calculate temporary vectorial force */
1991             tx               = _mm256_mul_ps(fscal,dx32);
1992             ty               = _mm256_mul_ps(fscal,dy32);
1993             tz               = _mm256_mul_ps(fscal,dz32);
1994
1995             /* Update vectorial force */
1996             fix3             = _mm256_add_ps(fix3,tx);
1997             fiy3             = _mm256_add_ps(fiy3,ty);
1998             fiz3             = _mm256_add_ps(fiz3,tz);
1999
2000             fjx2             = _mm256_add_ps(fjx2,tx);
2001             fjy2             = _mm256_add_ps(fjy2,ty);
2002             fjz2             = _mm256_add_ps(fjz2,tz);
2003
2004             /**************************
2005              * CALCULATE INTERACTIONS *
2006              **************************/
2007
2008             r33              = _mm256_mul_ps(rsq33,rinv33);
2009
2010             /* EWALD ELECTROSTATICS */
2011
2012             /* Analytical PME correction */
2013             zeta2            = _mm256_mul_ps(beta2,rsq33);
2014             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
2015             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2016             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2017             felec            = _mm256_mul_ps(qq33,felec);
2018
2019             fscal            = felec;
2020
2021             /* Calculate temporary vectorial force */
2022             tx               = _mm256_mul_ps(fscal,dx33);
2023             ty               = _mm256_mul_ps(fscal,dy33);
2024             tz               = _mm256_mul_ps(fscal,dz33);
2025
2026             /* Update vectorial force */
2027             fix3             = _mm256_add_ps(fix3,tx);
2028             fiy3             = _mm256_add_ps(fiy3,ty);
2029             fiz3             = _mm256_add_ps(fiz3,tz);
2030
2031             fjx3             = _mm256_add_ps(fjx3,tx);
2032             fjy3             = _mm256_add_ps(fjy3,ty);
2033             fjz3             = _mm256_add_ps(fjz3,tz);
2034
2035             fjptrA             = f+j_coord_offsetA;
2036             fjptrB             = f+j_coord_offsetB;
2037             fjptrC             = f+j_coord_offsetC;
2038             fjptrD             = f+j_coord_offsetD;
2039             fjptrE             = f+j_coord_offsetE;
2040             fjptrF             = f+j_coord_offsetF;
2041             fjptrG             = f+j_coord_offsetG;
2042             fjptrH             = f+j_coord_offsetH;
2043
2044             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2045                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2046                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2047
2048             /* Inner loop uses 555 flops */
2049         }
2050
2051         if(jidx<j_index_end)
2052         {
2053
2054             /* Get j neighbor index, and coordinate index */
2055             jnrlistA         = jjnr[jidx];
2056             jnrlistB         = jjnr[jidx+1];
2057             jnrlistC         = jjnr[jidx+2];
2058             jnrlistD         = jjnr[jidx+3];
2059             jnrlistE         = jjnr[jidx+4];
2060             jnrlistF         = jjnr[jidx+5];
2061             jnrlistG         = jjnr[jidx+6];
2062             jnrlistH         = jjnr[jidx+7];
2063             /* Sign of each element will be negative for non-real atoms.
2064              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2065              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2066              */
2067             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2068                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2069
2070             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2071             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2072             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2073             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2074             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2075             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2076             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2077             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2078             j_coord_offsetA  = DIM*jnrA;
2079             j_coord_offsetB  = DIM*jnrB;
2080             j_coord_offsetC  = DIM*jnrC;
2081             j_coord_offsetD  = DIM*jnrD;
2082             j_coord_offsetE  = DIM*jnrE;
2083             j_coord_offsetF  = DIM*jnrF;
2084             j_coord_offsetG  = DIM*jnrG;
2085             j_coord_offsetH  = DIM*jnrH;
2086
2087             /* load j atom coordinates */
2088             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2089                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2090                                                  x+j_coord_offsetE,x+j_coord_offsetF,
2091                                                  x+j_coord_offsetG,x+j_coord_offsetH,
2092                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2093                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
2094
2095             /* Calculate displacement vector */
2096             dx00             = _mm256_sub_ps(ix0,jx0);
2097             dy00             = _mm256_sub_ps(iy0,jy0);
2098             dz00             = _mm256_sub_ps(iz0,jz0);
2099             dx11             = _mm256_sub_ps(ix1,jx1);
2100             dy11             = _mm256_sub_ps(iy1,jy1);
2101             dz11             = _mm256_sub_ps(iz1,jz1);
2102             dx12             = _mm256_sub_ps(ix1,jx2);
2103             dy12             = _mm256_sub_ps(iy1,jy2);
2104             dz12             = _mm256_sub_ps(iz1,jz2);
2105             dx13             = _mm256_sub_ps(ix1,jx3);
2106             dy13             = _mm256_sub_ps(iy1,jy3);
2107             dz13             = _mm256_sub_ps(iz1,jz3);
2108             dx21             = _mm256_sub_ps(ix2,jx1);
2109             dy21             = _mm256_sub_ps(iy2,jy1);
2110             dz21             = _mm256_sub_ps(iz2,jz1);
2111             dx22             = _mm256_sub_ps(ix2,jx2);
2112             dy22             = _mm256_sub_ps(iy2,jy2);
2113             dz22             = _mm256_sub_ps(iz2,jz2);
2114             dx23             = _mm256_sub_ps(ix2,jx3);
2115             dy23             = _mm256_sub_ps(iy2,jy3);
2116             dz23             = _mm256_sub_ps(iz2,jz3);
2117             dx31             = _mm256_sub_ps(ix3,jx1);
2118             dy31             = _mm256_sub_ps(iy3,jy1);
2119             dz31             = _mm256_sub_ps(iz3,jz1);
2120             dx32             = _mm256_sub_ps(ix3,jx2);
2121             dy32             = _mm256_sub_ps(iy3,jy2);
2122             dz32             = _mm256_sub_ps(iz3,jz2);
2123             dx33             = _mm256_sub_ps(ix3,jx3);
2124             dy33             = _mm256_sub_ps(iy3,jy3);
2125             dz33             = _mm256_sub_ps(iz3,jz3);
2126
2127             /* Calculate squared distance and things based on it */
2128             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2129             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2130             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2131             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
2132             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2133             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2134             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
2135             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
2136             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
2137             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
2138
2139             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
2140             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2141             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2142             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
2143             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2144             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2145             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
2146             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
2147             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
2148             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
2149
2150             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2151             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2152             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
2153             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2154             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2155             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
2156             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
2157             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
2158             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
2159
2160             fjx0             = _mm256_setzero_ps();
2161             fjy0             = _mm256_setzero_ps();
2162             fjz0             = _mm256_setzero_ps();
2163             fjx1             = _mm256_setzero_ps();
2164             fjy1             = _mm256_setzero_ps();
2165             fjz1             = _mm256_setzero_ps();
2166             fjx2             = _mm256_setzero_ps();
2167             fjy2             = _mm256_setzero_ps();
2168             fjz2             = _mm256_setzero_ps();
2169             fjx3             = _mm256_setzero_ps();
2170             fjy3             = _mm256_setzero_ps();
2171             fjz3             = _mm256_setzero_ps();
2172
2173             /**************************
2174              * CALCULATE INTERACTIONS *
2175              **************************/
2176
2177             r00              = _mm256_mul_ps(rsq00,rinv00);
2178             r00              = _mm256_andnot_ps(dummy_mask,r00);
2179
2180             /* Calculate table index by multiplying r with table scale and truncate to integer */
2181             rt               = _mm256_mul_ps(r00,vftabscale);
2182             vfitab           = _mm256_cvttps_epi32(rt);
2183             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2184             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2185             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2186             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2187             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
2188             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
2189
2190             /* CUBIC SPLINE TABLE DISPERSION */
2191             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2192                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2193             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2194                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2195             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2196                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2197             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2198                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2199             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2200             Heps             = _mm256_mul_ps(vfeps,H);
2201             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2202             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2203             fvdw6            = _mm256_mul_ps(c6_00,FF);
2204
2205             /* CUBIC SPLINE TABLE REPULSION */
2206             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
2207             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
2208             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2209                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2210             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2211                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2212             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2213                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2214             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2215                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2216             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2217             Heps             = _mm256_mul_ps(vfeps,H);
2218             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2219             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2220             fvdw12           = _mm256_mul_ps(c12_00,FF);
2221             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
2222
2223             fscal            = fvdw;
2224
2225             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2226
2227             /* Calculate temporary vectorial force */
2228             tx               = _mm256_mul_ps(fscal,dx00);
2229             ty               = _mm256_mul_ps(fscal,dy00);
2230             tz               = _mm256_mul_ps(fscal,dz00);
2231
2232             /* Update vectorial force */
2233             fix0             = _mm256_add_ps(fix0,tx);
2234             fiy0             = _mm256_add_ps(fiy0,ty);
2235             fiz0             = _mm256_add_ps(fiz0,tz);
2236
2237             fjx0             = _mm256_add_ps(fjx0,tx);
2238             fjy0             = _mm256_add_ps(fjy0,ty);
2239             fjz0             = _mm256_add_ps(fjz0,tz);
2240
2241             /**************************
2242              * CALCULATE INTERACTIONS *
2243              **************************/
2244
2245             r11              = _mm256_mul_ps(rsq11,rinv11);
2246             r11              = _mm256_andnot_ps(dummy_mask,r11);
2247
2248             /* EWALD ELECTROSTATICS */
2249
2250             /* Analytical PME correction */
2251             zeta2            = _mm256_mul_ps(beta2,rsq11);
2252             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
2253             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2254             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2255             felec            = _mm256_mul_ps(qq11,felec);
2256
2257             fscal            = felec;
2258
2259             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2260
2261             /* Calculate temporary vectorial force */
2262             tx               = _mm256_mul_ps(fscal,dx11);
2263             ty               = _mm256_mul_ps(fscal,dy11);
2264             tz               = _mm256_mul_ps(fscal,dz11);
2265
2266             /* Update vectorial force */
2267             fix1             = _mm256_add_ps(fix1,tx);
2268             fiy1             = _mm256_add_ps(fiy1,ty);
2269             fiz1             = _mm256_add_ps(fiz1,tz);
2270
2271             fjx1             = _mm256_add_ps(fjx1,tx);
2272             fjy1             = _mm256_add_ps(fjy1,ty);
2273             fjz1             = _mm256_add_ps(fjz1,tz);
2274
2275             /**************************
2276              * CALCULATE INTERACTIONS *
2277              **************************/
2278
2279             r12              = _mm256_mul_ps(rsq12,rinv12);
2280             r12              = _mm256_andnot_ps(dummy_mask,r12);
2281
2282             /* EWALD ELECTROSTATICS */
2283
2284             /* Analytical PME correction */
2285             zeta2            = _mm256_mul_ps(beta2,rsq12);
2286             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2287             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2288             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2289             felec            = _mm256_mul_ps(qq12,felec);
2290
2291             fscal            = felec;
2292
2293             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2294
2295             /* Calculate temporary vectorial force */
2296             tx               = _mm256_mul_ps(fscal,dx12);
2297             ty               = _mm256_mul_ps(fscal,dy12);
2298             tz               = _mm256_mul_ps(fscal,dz12);
2299
2300             /* Update vectorial force */
2301             fix1             = _mm256_add_ps(fix1,tx);
2302             fiy1             = _mm256_add_ps(fiy1,ty);
2303             fiz1             = _mm256_add_ps(fiz1,tz);
2304
2305             fjx2             = _mm256_add_ps(fjx2,tx);
2306             fjy2             = _mm256_add_ps(fjy2,ty);
2307             fjz2             = _mm256_add_ps(fjz2,tz);
2308
2309             /**************************
2310              * CALCULATE INTERACTIONS *
2311              **************************/
2312
2313             r13              = _mm256_mul_ps(rsq13,rinv13);
2314             r13              = _mm256_andnot_ps(dummy_mask,r13);
2315
2316             /* EWALD ELECTROSTATICS */
2317
2318             /* Analytical PME correction */
2319             zeta2            = _mm256_mul_ps(beta2,rsq13);
2320             rinv3            = _mm256_mul_ps(rinvsq13,rinv13);
2321             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2322             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2323             felec            = _mm256_mul_ps(qq13,felec);
2324
2325             fscal            = felec;
2326
2327             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2328
2329             /* Calculate temporary vectorial force */
2330             tx               = _mm256_mul_ps(fscal,dx13);
2331             ty               = _mm256_mul_ps(fscal,dy13);
2332             tz               = _mm256_mul_ps(fscal,dz13);
2333
2334             /* Update vectorial force */
2335             fix1             = _mm256_add_ps(fix1,tx);
2336             fiy1             = _mm256_add_ps(fiy1,ty);
2337             fiz1             = _mm256_add_ps(fiz1,tz);
2338
2339             fjx3             = _mm256_add_ps(fjx3,tx);
2340             fjy3             = _mm256_add_ps(fjy3,ty);
2341             fjz3             = _mm256_add_ps(fjz3,tz);
2342
2343             /**************************
2344              * CALCULATE INTERACTIONS *
2345              **************************/
2346
2347             r21              = _mm256_mul_ps(rsq21,rinv21);
2348             r21              = _mm256_andnot_ps(dummy_mask,r21);
2349
2350             /* EWALD ELECTROSTATICS */
2351
2352             /* Analytical PME correction */
2353             zeta2            = _mm256_mul_ps(beta2,rsq21);
2354             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2355             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2356             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2357             felec            = _mm256_mul_ps(qq21,felec);
2358
2359             fscal            = felec;
2360
2361             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2362
2363             /* Calculate temporary vectorial force */
2364             tx               = _mm256_mul_ps(fscal,dx21);
2365             ty               = _mm256_mul_ps(fscal,dy21);
2366             tz               = _mm256_mul_ps(fscal,dz21);
2367
2368             /* Update vectorial force */
2369             fix2             = _mm256_add_ps(fix2,tx);
2370             fiy2             = _mm256_add_ps(fiy2,ty);
2371             fiz2             = _mm256_add_ps(fiz2,tz);
2372
2373             fjx1             = _mm256_add_ps(fjx1,tx);
2374             fjy1             = _mm256_add_ps(fjy1,ty);
2375             fjz1             = _mm256_add_ps(fjz1,tz);
2376
2377             /**************************
2378              * CALCULATE INTERACTIONS *
2379              **************************/
2380
2381             r22              = _mm256_mul_ps(rsq22,rinv22);
2382             r22              = _mm256_andnot_ps(dummy_mask,r22);
2383
2384             /* EWALD ELECTROSTATICS */
2385
2386             /* Analytical PME correction */
2387             zeta2            = _mm256_mul_ps(beta2,rsq22);
2388             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2389             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2390             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2391             felec            = _mm256_mul_ps(qq22,felec);
2392
2393             fscal            = felec;
2394
2395             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2396
2397             /* Calculate temporary vectorial force */
2398             tx               = _mm256_mul_ps(fscal,dx22);
2399             ty               = _mm256_mul_ps(fscal,dy22);
2400             tz               = _mm256_mul_ps(fscal,dz22);
2401
2402             /* Update vectorial force */
2403             fix2             = _mm256_add_ps(fix2,tx);
2404             fiy2             = _mm256_add_ps(fiy2,ty);
2405             fiz2             = _mm256_add_ps(fiz2,tz);
2406
2407             fjx2             = _mm256_add_ps(fjx2,tx);
2408             fjy2             = _mm256_add_ps(fjy2,ty);
2409             fjz2             = _mm256_add_ps(fjz2,tz);
2410
2411             /**************************
2412              * CALCULATE INTERACTIONS *
2413              **************************/
2414
2415             r23              = _mm256_mul_ps(rsq23,rinv23);
2416             r23              = _mm256_andnot_ps(dummy_mask,r23);
2417
2418             /* EWALD ELECTROSTATICS */
2419
2420             /* Analytical PME correction */
2421             zeta2            = _mm256_mul_ps(beta2,rsq23);
2422             rinv3            = _mm256_mul_ps(rinvsq23,rinv23);
2423             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2424             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2425             felec            = _mm256_mul_ps(qq23,felec);
2426
2427             fscal            = felec;
2428
2429             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2430
2431             /* Calculate temporary vectorial force */
2432             tx               = _mm256_mul_ps(fscal,dx23);
2433             ty               = _mm256_mul_ps(fscal,dy23);
2434             tz               = _mm256_mul_ps(fscal,dz23);
2435
2436             /* Update vectorial force */
2437             fix2             = _mm256_add_ps(fix2,tx);
2438             fiy2             = _mm256_add_ps(fiy2,ty);
2439             fiz2             = _mm256_add_ps(fiz2,tz);
2440
2441             fjx3             = _mm256_add_ps(fjx3,tx);
2442             fjy3             = _mm256_add_ps(fjy3,ty);
2443             fjz3             = _mm256_add_ps(fjz3,tz);
2444
2445             /**************************
2446              * CALCULATE INTERACTIONS *
2447              **************************/
2448
2449             r31              = _mm256_mul_ps(rsq31,rinv31);
2450             r31              = _mm256_andnot_ps(dummy_mask,r31);
2451
2452             /* EWALD ELECTROSTATICS */
2453
2454             /* Analytical PME correction */
2455             zeta2            = _mm256_mul_ps(beta2,rsq31);
2456             rinv3            = _mm256_mul_ps(rinvsq31,rinv31);
2457             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2458             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2459             felec            = _mm256_mul_ps(qq31,felec);
2460
2461             fscal            = felec;
2462
2463             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2464
2465             /* Calculate temporary vectorial force */
2466             tx               = _mm256_mul_ps(fscal,dx31);
2467             ty               = _mm256_mul_ps(fscal,dy31);
2468             tz               = _mm256_mul_ps(fscal,dz31);
2469
2470             /* Update vectorial force */
2471             fix3             = _mm256_add_ps(fix3,tx);
2472             fiy3             = _mm256_add_ps(fiy3,ty);
2473             fiz3             = _mm256_add_ps(fiz3,tz);
2474
2475             fjx1             = _mm256_add_ps(fjx1,tx);
2476             fjy1             = _mm256_add_ps(fjy1,ty);
2477             fjz1             = _mm256_add_ps(fjz1,tz);
2478
2479             /**************************
2480              * CALCULATE INTERACTIONS *
2481              **************************/
2482
2483             r32              = _mm256_mul_ps(rsq32,rinv32);
2484             r32              = _mm256_andnot_ps(dummy_mask,r32);
2485
2486             /* EWALD ELECTROSTATICS */
2487
2488             /* Analytical PME correction */
2489             zeta2            = _mm256_mul_ps(beta2,rsq32);
2490             rinv3            = _mm256_mul_ps(rinvsq32,rinv32);
2491             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2492             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2493             felec            = _mm256_mul_ps(qq32,felec);
2494
2495             fscal            = felec;
2496
2497             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2498
2499             /* Calculate temporary vectorial force */
2500             tx               = _mm256_mul_ps(fscal,dx32);
2501             ty               = _mm256_mul_ps(fscal,dy32);
2502             tz               = _mm256_mul_ps(fscal,dz32);
2503
2504             /* Update vectorial force */
2505             fix3             = _mm256_add_ps(fix3,tx);
2506             fiy3             = _mm256_add_ps(fiy3,ty);
2507             fiz3             = _mm256_add_ps(fiz3,tz);
2508
2509             fjx2             = _mm256_add_ps(fjx2,tx);
2510             fjy2             = _mm256_add_ps(fjy2,ty);
2511             fjz2             = _mm256_add_ps(fjz2,tz);
2512
2513             /**************************
2514              * CALCULATE INTERACTIONS *
2515              **************************/
2516
2517             r33              = _mm256_mul_ps(rsq33,rinv33);
2518             r33              = _mm256_andnot_ps(dummy_mask,r33);
2519
2520             /* EWALD ELECTROSTATICS */
2521
2522             /* Analytical PME correction */
2523             zeta2            = _mm256_mul_ps(beta2,rsq33);
2524             rinv3            = _mm256_mul_ps(rinvsq33,rinv33);
2525             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2526             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2527             felec            = _mm256_mul_ps(qq33,felec);
2528
2529             fscal            = felec;
2530
2531             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2532
2533             /* Calculate temporary vectorial force */
2534             tx               = _mm256_mul_ps(fscal,dx33);
2535             ty               = _mm256_mul_ps(fscal,dy33);
2536             tz               = _mm256_mul_ps(fscal,dz33);
2537
2538             /* Update vectorial force */
2539             fix3             = _mm256_add_ps(fix3,tx);
2540             fiy3             = _mm256_add_ps(fiy3,ty);
2541             fiz3             = _mm256_add_ps(fiz3,tz);
2542
2543             fjx3             = _mm256_add_ps(fjx3,tx);
2544             fjy3             = _mm256_add_ps(fjy3,ty);
2545             fjz3             = _mm256_add_ps(fjz3,tz);
2546
2547             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2548             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2549             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2550             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2551             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2552             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2553             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2554             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2555
2556             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2557                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2558                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2559
2560             /* Inner loop uses 565 flops */
2561         }
2562
2563         /* End of innermost loop */
2564
2565         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2566                                                  f+i_coord_offset,fshift+i_shift_offset);
2567
2568         /* Increment number of inner iterations */
2569         inneriter                  += j_index_end - j_index_start;
2570
2571         /* Outer loop uses 24 flops */
2572     }
2573
2574     /* Increment number of outer iterations */
2575     outeriter        += nri;
2576
2577     /* Update outer/inner flops */
2578
2579     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*565);
2580 }