src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_VF_avx_256_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Water4
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     real *           vdwioffsetptr3;
  92     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  94     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  96     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  97     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  98     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  99     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
 100     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 101     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 102     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 105     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 108     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 109     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 110     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 111     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 112     real             *charge;
 113     int              nvdwtype;
 114     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 115     int              *vdwtype;
 116     real             *vdwparam;
 117     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 118     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 119     __m256i          vfitab;
 120     __m128i          vfitab_lo,vfitab_hi;
 121     __m128i          ifour       = _mm_set1_epi32(4);
 122     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 123     real             *vftab;
 124     __m256           dummy_mask,cutoff_mask;
 125     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 126     __m256           one     = _mm256_set1_ps(1.0);
 127     __m256           two     = _mm256_set1_ps(2.0);
 128     x                = xx[0];
 129     f                = ff[0];
 130
 131     nri              = nlist->nri;
 132     iinr             = nlist->iinr;
 133     jindex           = nlist->jindex;
 134     jjnr             = nlist->jjnr;
 135     shiftidx         = nlist->shift;
 136     gid              = nlist->gid;
 137     shiftvec         = fr->shift_vec[0];
 138     fshift           = fr->fshift[0];
 139     facel            = _mm256_set1_ps(fr->ic->epsfac);
 140     charge           = mdatoms->chargeA;
 141     krf              = _mm256_set1_ps(fr->ic->k_rf);
 142     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 143     crf              = _mm256_set1_ps(fr->ic->c_rf);
 144     nvdwtype         = fr->ntype;
 145     vdwparam         = fr->nbfp;
 146     vdwtype          = mdatoms->typeA;
 147
 148     vftab            = kernel_data->table_vdw->data;
 149     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 150
 151     /* Setup water-specific parameters */
 152     inr              = nlist->iinr[0];
 153     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 154     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 155     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 156     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 157
 158     jq1              = _mm256_set1_ps(charge[inr+1]);
 159     jq2              = _mm256_set1_ps(charge[inr+2]);
 160     jq3              = _mm256_set1_ps(charge[inr+3]);
 161     vdwjidx0A        = 2*vdwtype[inr+0];
 162     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 163     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 164     qq11             = _mm256_mul_ps(iq1,jq1);
 165     qq12             = _mm256_mul_ps(iq1,jq2);
 166     qq13             = _mm256_mul_ps(iq1,jq3);
 167     qq21             = _mm256_mul_ps(iq2,jq1);
 168     qq22             = _mm256_mul_ps(iq2,jq2);
 169     qq23             = _mm256_mul_ps(iq2,jq3);
 170     qq31             = _mm256_mul_ps(iq3,jq1);
 171     qq32             = _mm256_mul_ps(iq3,jq2);
 172     qq33             = _mm256_mul_ps(iq3,jq3);
 173
 174     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 175     rcutoff_scalar   = fr->ic->rcoulomb;
 176     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 177     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 178
 179     /* Avoid stupid compiler warnings */
 180     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 181     j_coord_offsetA = 0;
 182     j_coord_offsetB = 0;
 183     j_coord_offsetC = 0;
 184     j_coord_offsetD = 0;
 185     j_coord_offsetE = 0;
 186     j_coord_offsetF = 0;
 187     j_coord_offsetG = 0;
 188     j_coord_offsetH = 0;
 189
 190     outeriter        = 0;
 191     inneriter        = 0;
 192
 193     for(iidx=0;iidx<4*DIM;iidx++)
 194     {
 195         scratch[iidx] = 0.0;
 196     }
 197
 198     /* Start outer loop over neighborlists */
 199     for(iidx=0; iidx<nri; iidx++)
 200     {
 201         /* Load shift vector for this list */
 202         i_shift_offset   = DIM*shiftidx[iidx];
 203
 204         /* Load limits for loop over neighbors */
 205         j_index_start    = jindex[iidx];
 206         j_index_end      = jindex[iidx+1];
 207
 208         /* Get outer coordinate index */
 209         inr              = iinr[iidx];
 210         i_coord_offset   = DIM*inr;
 211
 212         /* Load i particle coords and add shift vector */
 213         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 214                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 215
 216         fix0             = _mm256_setzero_ps();
 217         fiy0             = _mm256_setzero_ps();
 218         fiz0             = _mm256_setzero_ps();
 219         fix1             = _mm256_setzero_ps();
 220         fiy1             = _mm256_setzero_ps();
 221         fiz1             = _mm256_setzero_ps();
 222         fix2             = _mm256_setzero_ps();
 223         fiy2             = _mm256_setzero_ps();
 224         fiz2             = _mm256_setzero_ps();
 225         fix3             = _mm256_setzero_ps();
 226         fiy3             = _mm256_setzero_ps();
 227         fiz3             = _mm256_setzero_ps();
 228
 229         /* Reset potential sums */
 230         velecsum         = _mm256_setzero_ps();
 231         vvdwsum          = _mm256_setzero_ps();
 232
 233         /* Start inner kernel loop */
 234         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 235         {
 236
 237             /* Get j neighbor index, and coordinate index */
 238             jnrA             = jjnr[jidx];
 239             jnrB             = jjnr[jidx+1];
 240             jnrC             = jjnr[jidx+2];
 241             jnrD             = jjnr[jidx+3];
 242             jnrE             = jjnr[jidx+4];
 243             jnrF             = jjnr[jidx+5];
 244             jnrG             = jjnr[jidx+6];
 245             jnrH             = jjnr[jidx+7];
 246             j_coord_offsetA  = DIM*jnrA;
 247             j_coord_offsetB  = DIM*jnrB;
 248             j_coord_offsetC  = DIM*jnrC;
 249             j_coord_offsetD  = DIM*jnrD;
 250             j_coord_offsetE  = DIM*jnrE;
 251             j_coord_offsetF  = DIM*jnrF;
 252             j_coord_offsetG  = DIM*jnrG;
 253             j_coord_offsetH  = DIM*jnrH;
 254
 255             /* load j atom coordinates */
 256             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 257                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 258                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 259                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 260                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 261                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 262
 263             /* Calculate displacement vector */
 264             dx00             = _mm256_sub_ps(ix0,jx0);
 265             dy00             = _mm256_sub_ps(iy0,jy0);
 266             dz00             = _mm256_sub_ps(iz0,jz0);
 267             dx11             = _mm256_sub_ps(ix1,jx1);
 268             dy11             = _mm256_sub_ps(iy1,jy1);
 269             dz11             = _mm256_sub_ps(iz1,jz1);
 270             dx12             = _mm256_sub_ps(ix1,jx2);
 271             dy12             = _mm256_sub_ps(iy1,jy2);
 272             dz12             = _mm256_sub_ps(iz1,jz2);
 273             dx13             = _mm256_sub_ps(ix1,jx3);
 274             dy13             = _mm256_sub_ps(iy1,jy3);
 275             dz13             = _mm256_sub_ps(iz1,jz3);
 276             dx21             = _mm256_sub_ps(ix2,jx1);
 277             dy21             = _mm256_sub_ps(iy2,jy1);
 278             dz21             = _mm256_sub_ps(iz2,jz1);
 279             dx22             = _mm256_sub_ps(ix2,jx2);
 280             dy22             = _mm256_sub_ps(iy2,jy2);
 281             dz22             = _mm256_sub_ps(iz2,jz2);
 282             dx23             = _mm256_sub_ps(ix2,jx3);
 283             dy23             = _mm256_sub_ps(iy2,jy3);
 284             dz23             = _mm256_sub_ps(iz2,jz3);
 285             dx31             = _mm256_sub_ps(ix3,jx1);
 286             dy31             = _mm256_sub_ps(iy3,jy1);
 287             dz31             = _mm256_sub_ps(iz3,jz1);
 288             dx32             = _mm256_sub_ps(ix3,jx2);
 289             dy32             = _mm256_sub_ps(iy3,jy2);
 290             dz32             = _mm256_sub_ps(iz3,jz2);
 291             dx33             = _mm256_sub_ps(ix3,jx3);
 292             dy33             = _mm256_sub_ps(iy3,jy3);
 293             dz33             = _mm256_sub_ps(iz3,jz3);
 294
 295             /* Calculate squared distance and things based on it */
 296             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 297             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 298             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 299             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 300             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 301             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 302             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 303             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 304             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 305             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 306
 307             rinv00           = avx256_invsqrt_f(rsq00);
 308             rinv11           = avx256_invsqrt_f(rsq11);
 309             rinv12           = avx256_invsqrt_f(rsq12);
 310             rinv13           = avx256_invsqrt_f(rsq13);
 311             rinv21           = avx256_invsqrt_f(rsq21);
 312             rinv22           = avx256_invsqrt_f(rsq22);
 313             rinv23           = avx256_invsqrt_f(rsq23);
 314             rinv31           = avx256_invsqrt_f(rsq31);
 315             rinv32           = avx256_invsqrt_f(rsq32);
 316             rinv33           = avx256_invsqrt_f(rsq33);
 317
 318             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 319             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 320             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 321             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 322             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 323             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 324             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 325             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 326             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 327
 328             fjx0             = _mm256_setzero_ps();
 329             fjy0             = _mm256_setzero_ps();
 330             fjz0             = _mm256_setzero_ps();
 331             fjx1             = _mm256_setzero_ps();
 332             fjy1             = _mm256_setzero_ps();
 333             fjz1             = _mm256_setzero_ps();
 334             fjx2             = _mm256_setzero_ps();
 335             fjy2             = _mm256_setzero_ps();
 336             fjz2             = _mm256_setzero_ps();
 337             fjx3             = _mm256_setzero_ps();
 338             fjy3             = _mm256_setzero_ps();
 339             fjz3             = _mm256_setzero_ps();
 340
 341             /**************************
 342              * CALCULATE INTERACTIONS *
 343              **************************/
 344
 345             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 346             {
 347
 348             r00              = _mm256_mul_ps(rsq00,rinv00);
 349
 350             /* Calculate table index by multiplying r with table scale and truncate to integer */
 351             rt               = _mm256_mul_ps(r00,vftabscale);
 352             vfitab           = _mm256_cvttps_epi32(rt);
 353             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 354             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 355             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 356             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 357             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 358             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 359
 360             /* CUBIC SPLINE TABLE DISPERSION */
 361             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 362                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 363             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 364                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 365             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 366                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 367             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 368                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 369             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 370             Heps             = _mm256_mul_ps(vfeps,H);
 371             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 372             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 373             vvdw6            = _mm256_mul_ps(c6_00,VV);
 374             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 375             fvdw6            = _mm256_mul_ps(c6_00,FF);
 376
 377             /* CUBIC SPLINE TABLE REPULSION */
 378             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 379             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 380             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 381                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 382             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 383                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 384             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 385                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 386             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 387                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 388             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 389             Heps             = _mm256_mul_ps(vfeps,H);
 390             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 391             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 392             vvdw12           = _mm256_mul_ps(c12_00,VV);
 393             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 394             fvdw12           = _mm256_mul_ps(c12_00,FF);
 395             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 396             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 397
 398             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 402             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 403
 404             fscal            = fvdw;
 405
 406             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 407
 408             /* Calculate temporary vectorial force */
 409             tx               = _mm256_mul_ps(fscal,dx00);
 410             ty               = _mm256_mul_ps(fscal,dy00);
 411             tz               = _mm256_mul_ps(fscal,dz00);
 412
 413             /* Update vectorial force */
 414             fix0             = _mm256_add_ps(fix0,tx);
 415             fiy0             = _mm256_add_ps(fiy0,ty);
 416             fiz0             = _mm256_add_ps(fiz0,tz);
 417
 418             fjx0             = _mm256_add_ps(fjx0,tx);
 419             fjy0             = _mm256_add_ps(fjy0,ty);
 420             fjz0             = _mm256_add_ps(fjz0,tz);
 421
 422             }
 423
 424             /**************************
 425              * CALCULATE INTERACTIONS *
 426              **************************/
 427
 428             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 429             {
 430
 431             /* REACTION-FIELD ELECTROSTATICS */
 432             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 433             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 434
 435             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 436
 437             /* Update potential sum for this i atom from the interaction with this j atom. */
 438             velec            = _mm256_and_ps(velec,cutoff_mask);
 439             velecsum         = _mm256_add_ps(velecsum,velec);
 440
 441             fscal            = felec;
 442
 443             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 444
 445             /* Calculate temporary vectorial force */
 446             tx               = _mm256_mul_ps(fscal,dx11);
 447             ty               = _mm256_mul_ps(fscal,dy11);
 448             tz               = _mm256_mul_ps(fscal,dz11);
 449
 450             /* Update vectorial force */
 451             fix1             = _mm256_add_ps(fix1,tx);
 452             fiy1             = _mm256_add_ps(fiy1,ty);
 453             fiz1             = _mm256_add_ps(fiz1,tz);
 454
 455             fjx1             = _mm256_add_ps(fjx1,tx);
 456             fjy1             = _mm256_add_ps(fjy1,ty);
 457             fjz1             = _mm256_add_ps(fjz1,tz);
 458
 459             }
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 466             {
 467
 468             /* REACTION-FIELD ELECTROSTATICS */
 469             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 470             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 471
 472             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 473
 474             /* Update potential sum for this i atom from the interaction with this j atom. */
 475             velec            = _mm256_and_ps(velec,cutoff_mask);
 476             velecsum         = _mm256_add_ps(velecsum,velec);
 477
 478             fscal            = felec;
 479
 480             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 481
 482             /* Calculate temporary vectorial force */
 483             tx               = _mm256_mul_ps(fscal,dx12);
 484             ty               = _mm256_mul_ps(fscal,dy12);
 485             tz               = _mm256_mul_ps(fscal,dz12);
 486
 487             /* Update vectorial force */
 488             fix1             = _mm256_add_ps(fix1,tx);
 489             fiy1             = _mm256_add_ps(fiy1,ty);
 490             fiz1             = _mm256_add_ps(fiz1,tz);
 491
 492             fjx2             = _mm256_add_ps(fjx2,tx);
 493             fjy2             = _mm256_add_ps(fjy2,ty);
 494             fjz2             = _mm256_add_ps(fjz2,tz);
 495
 496             }
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 503             {
 504
 505             /* REACTION-FIELD ELECTROSTATICS */
 506             velec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_add_ps(rinv13,_mm256_mul_ps(krf,rsq13)),crf));
 507             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
 508
 509             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
 510
 511             /* Update potential sum for this i atom from the interaction with this j atom. */
 512             velec            = _mm256_and_ps(velec,cutoff_mask);
 513             velecsum         = _mm256_add_ps(velecsum,velec);
 514
 515             fscal            = felec;
 516
 517             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 518
 519             /* Calculate temporary vectorial force */
 520             tx               = _mm256_mul_ps(fscal,dx13);
 521             ty               = _mm256_mul_ps(fscal,dy13);
 522             tz               = _mm256_mul_ps(fscal,dz13);
 523
 524             /* Update vectorial force */
 525             fix1             = _mm256_add_ps(fix1,tx);
 526             fiy1             = _mm256_add_ps(fiy1,ty);
 527             fiz1             = _mm256_add_ps(fiz1,tz);
 528
 529             fjx3             = _mm256_add_ps(fjx3,tx);
 530             fjy3             = _mm256_add_ps(fjy3,ty);
 531             fjz3             = _mm256_add_ps(fjz3,tz);
 532
 533             }
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 540             {
 541
 542             /* REACTION-FIELD ELECTROSTATICS */
 543             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
 544             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
 545
 546             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
 547
 548             /* Update potential sum for this i atom from the interaction with this j atom. */
 549             velec            = _mm256_and_ps(velec,cutoff_mask);
 550             velecsum         = _mm256_add_ps(velecsum,velec);
 551
 552             fscal            = felec;
 553
 554             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 555
 556             /* Calculate temporary vectorial force */
 557             tx               = _mm256_mul_ps(fscal,dx21);
 558             ty               = _mm256_mul_ps(fscal,dy21);
 559             tz               = _mm256_mul_ps(fscal,dz21);
 560
 561             /* Update vectorial force */
 562             fix2             = _mm256_add_ps(fix2,tx);
 563             fiy2             = _mm256_add_ps(fiy2,ty);
 564             fiz2             = _mm256_add_ps(fiz2,tz);
 565
 566             fjx1             = _mm256_add_ps(fjx1,tx);
 567             fjy1             = _mm256_add_ps(fjy1,ty);
 568             fjz1             = _mm256_add_ps(fjz1,tz);
 569
 570             }
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 577             {
 578
 579             /* REACTION-FIELD ELECTROSTATICS */
 580             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
 581             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
 582
 583             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
 584
 585             /* Update potential sum for this i atom from the interaction with this j atom. */
 586             velec            = _mm256_and_ps(velec,cutoff_mask);
 587             velecsum         = _mm256_add_ps(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 592
 593             /* Calculate temporary vectorial force */
 594             tx               = _mm256_mul_ps(fscal,dx22);
 595             ty               = _mm256_mul_ps(fscal,dy22);
 596             tz               = _mm256_mul_ps(fscal,dz22);
 597
 598             /* Update vectorial force */
 599             fix2             = _mm256_add_ps(fix2,tx);
 600             fiy2             = _mm256_add_ps(fiy2,ty);
 601             fiz2             = _mm256_add_ps(fiz2,tz);
 602
 603             fjx2             = _mm256_add_ps(fjx2,tx);
 604             fjy2             = _mm256_add_ps(fjy2,ty);
 605             fjz2             = _mm256_add_ps(fjz2,tz);
 606
 607             }
 608
 609             /**************************
 610              * CALCULATE INTERACTIONS *
 611              **************************/
 612
 613             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 614             {
 615
 616             /* REACTION-FIELD ELECTROSTATICS */
 617             velec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_add_ps(rinv23,_mm256_mul_ps(krf,rsq23)),crf));
 618             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
 619
 620             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
 621
 622             /* Update potential sum for this i atom from the interaction with this j atom. */
 623             velec            = _mm256_and_ps(velec,cutoff_mask);
 624             velecsum         = _mm256_add_ps(velecsum,velec);
 625
 626             fscal            = felec;
 627
 628             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 629
 630             /* Calculate temporary vectorial force */
 631             tx               = _mm256_mul_ps(fscal,dx23);
 632             ty               = _mm256_mul_ps(fscal,dy23);
 633             tz               = _mm256_mul_ps(fscal,dz23);
 634
 635             /* Update vectorial force */
 636             fix2             = _mm256_add_ps(fix2,tx);
 637             fiy2             = _mm256_add_ps(fiy2,ty);
 638             fiz2             = _mm256_add_ps(fiz2,tz);
 639
 640             fjx3             = _mm256_add_ps(fjx3,tx);
 641             fjy3             = _mm256_add_ps(fjy3,ty);
 642             fjz3             = _mm256_add_ps(fjz3,tz);
 643
 644             }
 645
 646             /**************************
 647              * CALCULATE INTERACTIONS *
 648              **************************/
 649
 650             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 651             {
 652
 653             /* REACTION-FIELD ELECTROSTATICS */
 654             velec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_add_ps(rinv31,_mm256_mul_ps(krf,rsq31)),crf));
 655             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
 656
 657             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
 658
 659             /* Update potential sum for this i atom from the interaction with this j atom. */
 660             velec            = _mm256_and_ps(velec,cutoff_mask);
 661             velecsum         = _mm256_add_ps(velecsum,velec);
 662
 663             fscal            = felec;
 664
 665             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 666
 667             /* Calculate temporary vectorial force */
 668             tx               = _mm256_mul_ps(fscal,dx31);
 669             ty               = _mm256_mul_ps(fscal,dy31);
 670             tz               = _mm256_mul_ps(fscal,dz31);
 671
 672             /* Update vectorial force */
 673             fix3             = _mm256_add_ps(fix3,tx);
 674             fiy3             = _mm256_add_ps(fiy3,ty);
 675             fiz3             = _mm256_add_ps(fiz3,tz);
 676
 677             fjx1             = _mm256_add_ps(fjx1,tx);
 678             fjy1             = _mm256_add_ps(fjy1,ty);
 679             fjz1             = _mm256_add_ps(fjz1,tz);
 680
 681             }
 682
 683             /**************************
 684              * CALCULATE INTERACTIONS *
 685              **************************/
 686
 687             if (gmx_mm256_any_lt(rsq32,rcutoff2))
 688             {
 689
 690             /* REACTION-FIELD ELECTROSTATICS */
 691             velec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_add_ps(rinv32,_mm256_mul_ps(krf,rsq32)),crf));
 692             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
 693
 694             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
 695
 696             /* Update potential sum for this i atom from the interaction with this j atom. */
 697             velec            = _mm256_and_ps(velec,cutoff_mask);
 698             velecsum         = _mm256_add_ps(velecsum,velec);
 699
 700             fscal            = felec;
 701
 702             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 703
 704             /* Calculate temporary vectorial force */
 705             tx               = _mm256_mul_ps(fscal,dx32);
 706             ty               = _mm256_mul_ps(fscal,dy32);
 707             tz               = _mm256_mul_ps(fscal,dz32);
 708
 709             /* Update vectorial force */
 710             fix3             = _mm256_add_ps(fix3,tx);
 711             fiy3             = _mm256_add_ps(fiy3,ty);
 712             fiz3             = _mm256_add_ps(fiz3,tz);
 713
 714             fjx2             = _mm256_add_ps(fjx2,tx);
 715             fjy2             = _mm256_add_ps(fjy2,ty);
 716             fjz2             = _mm256_add_ps(fjz2,tz);
 717
 718             }
 719
 720             /**************************
 721              * CALCULATE INTERACTIONS *
 722              **************************/
 723
 724             if (gmx_mm256_any_lt(rsq33,rcutoff2))
 725             {
 726
 727             /* REACTION-FIELD ELECTROSTATICS */
 728             velec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_add_ps(rinv33,_mm256_mul_ps(krf,rsq33)),crf));
 729             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
 730
 731             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
 732
 733             /* Update potential sum for this i atom from the interaction with this j atom. */
 734             velec            = _mm256_and_ps(velec,cutoff_mask);
 735             velecsum         = _mm256_add_ps(velecsum,velec);
 736
 737             fscal            = felec;
 738
 739             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 740
 741             /* Calculate temporary vectorial force */
 742             tx               = _mm256_mul_ps(fscal,dx33);
 743             ty               = _mm256_mul_ps(fscal,dy33);
 744             tz               = _mm256_mul_ps(fscal,dz33);
 745
 746             /* Update vectorial force */
 747             fix3             = _mm256_add_ps(fix3,tx);
 748             fiy3             = _mm256_add_ps(fiy3,ty);
 749             fiz3             = _mm256_add_ps(fiz3,tz);
 750
 751             fjx3             = _mm256_add_ps(fjx3,tx);
 752             fjy3             = _mm256_add_ps(fjy3,ty);
 753             fjz3             = _mm256_add_ps(fjz3,tz);
 754
 755             }
 756
 757             fjptrA             = f+j_coord_offsetA;
 758             fjptrB             = f+j_coord_offsetB;
 759             fjptrC             = f+j_coord_offsetC;
 760             fjptrD             = f+j_coord_offsetD;
 761             fjptrE             = f+j_coord_offsetE;
 762             fjptrF             = f+j_coord_offsetF;
 763             fjptrG             = f+j_coord_offsetG;
 764             fjptrH             = f+j_coord_offsetH;
 765
 766             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 767                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 768                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 769
 770             /* Inner loop uses 387 flops */
 771         }
 772
 773         if(jidx<j_index_end)
 774         {
 775
 776             /* Get j neighbor index, and coordinate index */
 777             jnrlistA         = jjnr[jidx];
 778             jnrlistB         = jjnr[jidx+1];
 779             jnrlistC         = jjnr[jidx+2];
 780             jnrlistD         = jjnr[jidx+3];
 781             jnrlistE         = jjnr[jidx+4];
 782             jnrlistF         = jjnr[jidx+5];
 783             jnrlistG         = jjnr[jidx+6];
 784             jnrlistH         = jjnr[jidx+7];
 785             /* Sign of each element will be negative for non-real atoms.
 786              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 787              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 788              */
 789             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 790                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 791
 792             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 793             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 794             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 795             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 796             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 797             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 798             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 799             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 800             j_coord_offsetA  = DIM*jnrA;
 801             j_coord_offsetB  = DIM*jnrB;
 802             j_coord_offsetC  = DIM*jnrC;
 803             j_coord_offsetD  = DIM*jnrD;
 804             j_coord_offsetE  = DIM*jnrE;
 805             j_coord_offsetF  = DIM*jnrF;
 806             j_coord_offsetG  = DIM*jnrG;
 807             j_coord_offsetH  = DIM*jnrH;
 808
 809             /* load j atom coordinates */
 810             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 811                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 812                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 813                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 814                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 815                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 816
 817             /* Calculate displacement vector */
 818             dx00             = _mm256_sub_ps(ix0,jx0);
 819             dy00             = _mm256_sub_ps(iy0,jy0);
 820             dz00             = _mm256_sub_ps(iz0,jz0);
 821             dx11             = _mm256_sub_ps(ix1,jx1);
 822             dy11             = _mm256_sub_ps(iy1,jy1);
 823             dz11             = _mm256_sub_ps(iz1,jz1);
 824             dx12             = _mm256_sub_ps(ix1,jx2);
 825             dy12             = _mm256_sub_ps(iy1,jy2);
 826             dz12             = _mm256_sub_ps(iz1,jz2);
 827             dx13             = _mm256_sub_ps(ix1,jx3);
 828             dy13             = _mm256_sub_ps(iy1,jy3);
 829             dz13             = _mm256_sub_ps(iz1,jz3);
 830             dx21             = _mm256_sub_ps(ix2,jx1);
 831             dy21             = _mm256_sub_ps(iy2,jy1);
 832             dz21             = _mm256_sub_ps(iz2,jz1);
 833             dx22             = _mm256_sub_ps(ix2,jx2);
 834             dy22             = _mm256_sub_ps(iy2,jy2);
 835             dz22             = _mm256_sub_ps(iz2,jz2);
 836             dx23             = _mm256_sub_ps(ix2,jx3);
 837             dy23             = _mm256_sub_ps(iy2,jy3);
 838             dz23             = _mm256_sub_ps(iz2,jz3);
 839             dx31             = _mm256_sub_ps(ix3,jx1);
 840             dy31             = _mm256_sub_ps(iy3,jy1);
 841             dz31             = _mm256_sub_ps(iz3,jz1);
 842             dx32             = _mm256_sub_ps(ix3,jx2);
 843             dy32             = _mm256_sub_ps(iy3,jy2);
 844             dz32             = _mm256_sub_ps(iz3,jz2);
 845             dx33             = _mm256_sub_ps(ix3,jx3);
 846             dy33             = _mm256_sub_ps(iy3,jy3);
 847             dz33             = _mm256_sub_ps(iz3,jz3);
 848
 849             /* Calculate squared distance and things based on it */
 850             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 851             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 852             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 853             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 854             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 855             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 856             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 857             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 858             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 859             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 860
 861             rinv00           = avx256_invsqrt_f(rsq00);
 862             rinv11           = avx256_invsqrt_f(rsq11);
 863             rinv12           = avx256_invsqrt_f(rsq12);
 864             rinv13           = avx256_invsqrt_f(rsq13);
 865             rinv21           = avx256_invsqrt_f(rsq21);
 866             rinv22           = avx256_invsqrt_f(rsq22);
 867             rinv23           = avx256_invsqrt_f(rsq23);
 868             rinv31           = avx256_invsqrt_f(rsq31);
 869             rinv32           = avx256_invsqrt_f(rsq32);
 870             rinv33           = avx256_invsqrt_f(rsq33);
 871
 872             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 873             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 874             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 875             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 876             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 877             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 878             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 879             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 880             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 881
 882             fjx0             = _mm256_setzero_ps();
 883             fjy0             = _mm256_setzero_ps();
 884             fjz0             = _mm256_setzero_ps();
 885             fjx1             = _mm256_setzero_ps();
 886             fjy1             = _mm256_setzero_ps();
 887             fjz1             = _mm256_setzero_ps();
 888             fjx2             = _mm256_setzero_ps();
 889             fjy2             = _mm256_setzero_ps();
 890             fjz2             = _mm256_setzero_ps();
 891             fjx3             = _mm256_setzero_ps();
 892             fjy3             = _mm256_setzero_ps();
 893             fjz3             = _mm256_setzero_ps();
 894
 895             /**************************
 896              * CALCULATE INTERACTIONS *
 897              **************************/
 898
 899             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 900             {
 901
 902             r00              = _mm256_mul_ps(rsq00,rinv00);
 903             r00              = _mm256_andnot_ps(dummy_mask,r00);
 904
 905             /* Calculate table index by multiplying r with table scale and truncate to integer */
 906             rt               = _mm256_mul_ps(r00,vftabscale);
 907             vfitab           = _mm256_cvttps_epi32(rt);
 908             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 909             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 910             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 911             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 912             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 913             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 914
 915             /* CUBIC SPLINE TABLE DISPERSION */
 916             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 917                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 918             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 919                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 920             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 921                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 922             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 923                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 924             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 925             Heps             = _mm256_mul_ps(vfeps,H);
 926             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 927             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 928             vvdw6            = _mm256_mul_ps(c6_00,VV);
 929             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 930             fvdw6            = _mm256_mul_ps(c6_00,FF);
 931
 932             /* CUBIC SPLINE TABLE REPULSION */
 933             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 934             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 935             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 936                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 937             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 938                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 939             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 940                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 941             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 942                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 943             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 944             Heps             = _mm256_mul_ps(vfeps,H);
 945             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 946             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 947             vvdw12           = _mm256_mul_ps(c12_00,VV);
 948             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 949             fvdw12           = _mm256_mul_ps(c12_00,FF);
 950             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 951             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 952
 953             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 954
 955             /* Update potential sum for this i atom from the interaction with this j atom. */
 956             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 957             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 958             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 959
 960             fscal            = fvdw;
 961
 962             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 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
 982             /**************************
 983              * CALCULATE INTERACTIONS *
 984              **************************/
 985
 986             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 987             {
 988
 989             /* REACTION-FIELD ELECTROSTATICS */
 990             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 991             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 992
 993             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 994
 995             /* Update potential sum for this i atom from the interaction with this j atom. */
 996             velec            = _mm256_and_ps(velec,cutoff_mask);
 997             velec            = _mm256_andnot_ps(dummy_mask,velec);
 998             velecsum         = _mm256_add_ps(velecsum,velec);
 999
1000             fscal            = felec;
1001
1002             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1003
1004             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1005
1006             /* Calculate temporary vectorial force */
1007             tx               = _mm256_mul_ps(fscal,dx11);
1008             ty               = _mm256_mul_ps(fscal,dy11);
1009             tz               = _mm256_mul_ps(fscal,dz11);
1010
1011             /* Update vectorial force */
1012             fix1             = _mm256_add_ps(fix1,tx);
1013             fiy1             = _mm256_add_ps(fiy1,ty);
1014             fiz1             = _mm256_add_ps(fiz1,tz);
1015
1016             fjx1             = _mm256_add_ps(fjx1,tx);
1017             fjy1             = _mm256_add_ps(fjy1,ty);
1018             fjz1             = _mm256_add_ps(fjz1,tz);
1019
1020             }
1021
1022             /**************************
1023              * CALCULATE INTERACTIONS *
1024              **************************/
1025
1026             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1027             {
1028
1029             /* REACTION-FIELD ELECTROSTATICS */
1030             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
1031             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1032
1033             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1034
1035             /* Update potential sum for this i atom from the interaction with this j atom. */
1036             velec            = _mm256_and_ps(velec,cutoff_mask);
1037             velec            = _mm256_andnot_ps(dummy_mask,velec);
1038             velecsum         = _mm256_add_ps(velecsum,velec);
1039
1040             fscal            = felec;
1041
1042             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1043
1044             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1045
1046             /* Calculate temporary vectorial force */
1047             tx               = _mm256_mul_ps(fscal,dx12);
1048             ty               = _mm256_mul_ps(fscal,dy12);
1049             tz               = _mm256_mul_ps(fscal,dz12);
1050
1051             /* Update vectorial force */
1052             fix1             = _mm256_add_ps(fix1,tx);
1053             fiy1             = _mm256_add_ps(fiy1,ty);
1054             fiz1             = _mm256_add_ps(fiz1,tz);
1055
1056             fjx2             = _mm256_add_ps(fjx2,tx);
1057             fjy2             = _mm256_add_ps(fjy2,ty);
1058             fjz2             = _mm256_add_ps(fjz2,tz);
1059
1060             }
1061
1062             /**************************
1063              * CALCULATE INTERACTIONS *
1064              **************************/
1065
1066             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1067             {
1068
1069             /* REACTION-FIELD ELECTROSTATICS */
1070             velec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_add_ps(rinv13,_mm256_mul_ps(krf,rsq13)),crf));
1071             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
1072
1073             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
1074
1075             /* Update potential sum for this i atom from the interaction with this j atom. */
1076             velec            = _mm256_and_ps(velec,cutoff_mask);
1077             velec            = _mm256_andnot_ps(dummy_mask,velec);
1078             velecsum         = _mm256_add_ps(velecsum,velec);
1079
1080             fscal            = felec;
1081
1082             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1083
1084             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1085
1086             /* Calculate temporary vectorial force */
1087             tx               = _mm256_mul_ps(fscal,dx13);
1088             ty               = _mm256_mul_ps(fscal,dy13);
1089             tz               = _mm256_mul_ps(fscal,dz13);
1090
1091             /* Update vectorial force */
1092             fix1             = _mm256_add_ps(fix1,tx);
1093             fiy1             = _mm256_add_ps(fiy1,ty);
1094             fiz1             = _mm256_add_ps(fiz1,tz);
1095
1096             fjx3             = _mm256_add_ps(fjx3,tx);
1097             fjy3             = _mm256_add_ps(fjy3,ty);
1098             fjz3             = _mm256_add_ps(fjz3,tz);
1099
1100             }
1101
1102             /**************************
1103              * CALCULATE INTERACTIONS *
1104              **************************/
1105
1106             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1107             {
1108
1109             /* REACTION-FIELD ELECTROSTATICS */
1110             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
1111             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1112
1113             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1114
1115             /* Update potential sum for this i atom from the interaction with this j atom. */
1116             velec            = _mm256_and_ps(velec,cutoff_mask);
1117             velec            = _mm256_andnot_ps(dummy_mask,velec);
1118             velecsum         = _mm256_add_ps(velecsum,velec);
1119
1120             fscal            = felec;
1121
1122             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1123
1124             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1125
1126             /* Calculate temporary vectorial force */
1127             tx               = _mm256_mul_ps(fscal,dx21);
1128             ty               = _mm256_mul_ps(fscal,dy21);
1129             tz               = _mm256_mul_ps(fscal,dz21);
1130
1131             /* Update vectorial force */
1132             fix2             = _mm256_add_ps(fix2,tx);
1133             fiy2             = _mm256_add_ps(fiy2,ty);
1134             fiz2             = _mm256_add_ps(fiz2,tz);
1135
1136             fjx1             = _mm256_add_ps(fjx1,tx);
1137             fjy1             = _mm256_add_ps(fjy1,ty);
1138             fjz1             = _mm256_add_ps(fjz1,tz);
1139
1140             }
1141
1142             /**************************
1143              * CALCULATE INTERACTIONS *
1144              **************************/
1145
1146             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1147             {
1148
1149             /* REACTION-FIELD ELECTROSTATICS */
1150             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
1151             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1152
1153             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1154
1155             /* Update potential sum for this i atom from the interaction with this j atom. */
1156             velec            = _mm256_and_ps(velec,cutoff_mask);
1157             velec            = _mm256_andnot_ps(dummy_mask,velec);
1158             velecsum         = _mm256_add_ps(velecsum,velec);
1159
1160             fscal            = felec;
1161
1162             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1163
1164             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1165
1166             /* Calculate temporary vectorial force */
1167             tx               = _mm256_mul_ps(fscal,dx22);
1168             ty               = _mm256_mul_ps(fscal,dy22);
1169             tz               = _mm256_mul_ps(fscal,dz22);
1170
1171             /* Update vectorial force */
1172             fix2             = _mm256_add_ps(fix2,tx);
1173             fiy2             = _mm256_add_ps(fiy2,ty);
1174             fiz2             = _mm256_add_ps(fiz2,tz);
1175
1176             fjx2             = _mm256_add_ps(fjx2,tx);
1177             fjy2             = _mm256_add_ps(fjy2,ty);
1178             fjz2             = _mm256_add_ps(fjz2,tz);
1179
1180             }
1181
1182             /**************************
1183              * CALCULATE INTERACTIONS *
1184              **************************/
1185
1186             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1187             {
1188
1189             /* REACTION-FIELD ELECTROSTATICS */
1190             velec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_add_ps(rinv23,_mm256_mul_ps(krf,rsq23)),crf));
1191             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
1192
1193             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
1194
1195             /* Update potential sum for this i atom from the interaction with this j atom. */
1196             velec            = _mm256_and_ps(velec,cutoff_mask);
1197             velec            = _mm256_andnot_ps(dummy_mask,velec);
1198             velecsum         = _mm256_add_ps(velecsum,velec);
1199
1200             fscal            = felec;
1201
1202             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1203
1204             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1205
1206             /* Calculate temporary vectorial force */
1207             tx               = _mm256_mul_ps(fscal,dx23);
1208             ty               = _mm256_mul_ps(fscal,dy23);
1209             tz               = _mm256_mul_ps(fscal,dz23);
1210
1211             /* Update vectorial force */
1212             fix2             = _mm256_add_ps(fix2,tx);
1213             fiy2             = _mm256_add_ps(fiy2,ty);
1214             fiz2             = _mm256_add_ps(fiz2,tz);
1215
1216             fjx3             = _mm256_add_ps(fjx3,tx);
1217             fjy3             = _mm256_add_ps(fjy3,ty);
1218             fjz3             = _mm256_add_ps(fjz3,tz);
1219
1220             }
1221
1222             /**************************
1223              * CALCULATE INTERACTIONS *
1224              **************************/
1225
1226             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1227             {
1228
1229             /* REACTION-FIELD ELECTROSTATICS */
1230             velec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_add_ps(rinv31,_mm256_mul_ps(krf,rsq31)),crf));
1231             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
1232
1233             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
1234
1235             /* Update potential sum for this i atom from the interaction with this j atom. */
1236             velec            = _mm256_and_ps(velec,cutoff_mask);
1237             velec            = _mm256_andnot_ps(dummy_mask,velec);
1238             velecsum         = _mm256_add_ps(velecsum,velec);
1239
1240             fscal            = felec;
1241
1242             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1243
1244             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1245
1246             /* Calculate temporary vectorial force */
1247             tx               = _mm256_mul_ps(fscal,dx31);
1248             ty               = _mm256_mul_ps(fscal,dy31);
1249             tz               = _mm256_mul_ps(fscal,dz31);
1250
1251             /* Update vectorial force */
1252             fix3             = _mm256_add_ps(fix3,tx);
1253             fiy3             = _mm256_add_ps(fiy3,ty);
1254             fiz3             = _mm256_add_ps(fiz3,tz);
1255
1256             fjx1             = _mm256_add_ps(fjx1,tx);
1257             fjy1             = _mm256_add_ps(fjy1,ty);
1258             fjz1             = _mm256_add_ps(fjz1,tz);
1259
1260             }
1261
1262             /**************************
1263              * CALCULATE INTERACTIONS *
1264              **************************/
1265
1266             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1267             {
1268
1269             /* REACTION-FIELD ELECTROSTATICS */
1270             velec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_add_ps(rinv32,_mm256_mul_ps(krf,rsq32)),crf));
1271             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
1272
1273             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
1274
1275             /* Update potential sum for this i atom from the interaction with this j atom. */
1276             velec            = _mm256_and_ps(velec,cutoff_mask);
1277             velec            = _mm256_andnot_ps(dummy_mask,velec);
1278             velecsum         = _mm256_add_ps(velecsum,velec);
1279
1280             fscal            = felec;
1281
1282             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1283
1284             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1285
1286             /* Calculate temporary vectorial force */
1287             tx               = _mm256_mul_ps(fscal,dx32);
1288             ty               = _mm256_mul_ps(fscal,dy32);
1289             tz               = _mm256_mul_ps(fscal,dz32);
1290
1291             /* Update vectorial force */
1292             fix3             = _mm256_add_ps(fix3,tx);
1293             fiy3             = _mm256_add_ps(fiy3,ty);
1294             fiz3             = _mm256_add_ps(fiz3,tz);
1295
1296             fjx2             = _mm256_add_ps(fjx2,tx);
1297             fjy2             = _mm256_add_ps(fjy2,ty);
1298             fjz2             = _mm256_add_ps(fjz2,tz);
1299
1300             }
1301
1302             /**************************
1303              * CALCULATE INTERACTIONS *
1304              **************************/
1305
1306             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1307             {
1308
1309             /* REACTION-FIELD ELECTROSTATICS */
1310             velec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_add_ps(rinv33,_mm256_mul_ps(krf,rsq33)),crf));
1311             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
1312
1313             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
1314
1315             /* Update potential sum for this i atom from the interaction with this j atom. */
1316             velec            = _mm256_and_ps(velec,cutoff_mask);
1317             velec            = _mm256_andnot_ps(dummy_mask,velec);
1318             velecsum         = _mm256_add_ps(velecsum,velec);
1319
1320             fscal            = felec;
1321
1322             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1323
1324             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1325
1326             /* Calculate temporary vectorial force */
1327             tx               = _mm256_mul_ps(fscal,dx33);
1328             ty               = _mm256_mul_ps(fscal,dy33);
1329             tz               = _mm256_mul_ps(fscal,dz33);
1330
1331             /* Update vectorial force */
1332             fix3             = _mm256_add_ps(fix3,tx);
1333             fiy3             = _mm256_add_ps(fiy3,ty);
1334             fiz3             = _mm256_add_ps(fiz3,tz);
1335
1336             fjx3             = _mm256_add_ps(fjx3,tx);
1337             fjy3             = _mm256_add_ps(fjy3,ty);
1338             fjz3             = _mm256_add_ps(fjz3,tz);
1339
1340             }
1341
1342             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1343             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1344             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1345             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1346             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1347             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1348             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1349             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1350
1351             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1352                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1353                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1354
1355             /* Inner loop uses 388 flops */
1356         }
1357
1358         /* End of innermost loop */
1359
1360         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1361                                                  f+i_coord_offset,fshift+i_shift_offset);
1362
1363         ggid                        = gid[iidx];
1364         /* Update potential energies */
1365         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1366         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1367
1368         /* Increment number of inner iterations */
1369         inneriter                  += j_index_end - j_index_start;
1370
1371         /* Outer loop uses 26 flops */
1372     }
1373
1374     /* Increment number of outer iterations */
1375     outeriter        += nri;
1376
1377     /* Update outer/inner flops */
1378
1379     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*388);
1380 }
1381 /*
1382  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_F_avx_256_single
1383  * Electrostatics interaction: ReactionField
1384  * VdW interaction:            CubicSplineTable
1385  * Geometry:                   Water4-Water4
1386  * Calculate force/pot:        Force
1387  */
1388 void
1389 nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_F_avx_256_single
1390                     (t_nblist                    * gmx_restrict       nlist,
1391                      rvec                        * gmx_restrict          xx,
1392                      rvec                        * gmx_restrict          ff,
1393                      struct t_forcerec           * gmx_restrict          fr,
1394                      t_mdatoms                   * gmx_restrict     mdatoms,
1395                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1396                      t_nrnb                      * gmx_restrict        nrnb)
1397 {
1398     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1399      * just 0 for non-waters.
1400      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1401      * jnr indices corresponding to data put in the four positions in the SIMD register.
1402      */
1403     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1404     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1405     int              jnrA,jnrB,jnrC,jnrD;
1406     int              jnrE,jnrF,jnrG,jnrH;
1407     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1408     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1409     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1410     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1411     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1412     real             rcutoff_scalar;
1413     real             *shiftvec,*fshift,*x,*f;
1414     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1415     real             scratch[4*DIM];
1416     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1417     real *           vdwioffsetptr0;
1418     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1419     real *           vdwioffsetptr1;
1420     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1421     real *           vdwioffsetptr2;
1422     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1423     real *           vdwioffsetptr3;
1424     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1425     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1426     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1427     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1428     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1429     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1430     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1431     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1432     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1433     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1434     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1435     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1436     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1437     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1438     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1439     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1440     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1441     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1442     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1443     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1444     real             *charge;
1445     int              nvdwtype;
1446     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1447     int              *vdwtype;
1448     real             *vdwparam;
1449     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1450     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1451     __m256i          vfitab;
1452     __m128i          vfitab_lo,vfitab_hi;
1453     __m128i          ifour       = _mm_set1_epi32(4);
1454     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1455     real             *vftab;
1456     __m256           dummy_mask,cutoff_mask;
1457     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1458     __m256           one     = _mm256_set1_ps(1.0);
1459     __m256           two     = _mm256_set1_ps(2.0);
1460     x                = xx[0];
1461     f                = ff[0];
1462
1463     nri              = nlist->nri;
1464     iinr             = nlist->iinr;
1465     jindex           = nlist->jindex;
1466     jjnr             = nlist->jjnr;
1467     shiftidx         = nlist->shift;
1468     gid              = nlist->gid;
1469     shiftvec         = fr->shift_vec[0];
1470     fshift           = fr->fshift[0];
1471     facel            = _mm256_set1_ps(fr->ic->epsfac);
1472     charge           = mdatoms->chargeA;
1473     krf              = _mm256_set1_ps(fr->ic->k_rf);
1474     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
1475     crf              = _mm256_set1_ps(fr->ic->c_rf);
1476     nvdwtype         = fr->ntype;
1477     vdwparam         = fr->nbfp;
1478     vdwtype          = mdatoms->typeA;
1479
1480     vftab            = kernel_data->table_vdw->data;
1481     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
1482
1483     /* Setup water-specific parameters */
1484     inr              = nlist->iinr[0];
1485     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1486     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1487     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1488     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1489
1490     jq1              = _mm256_set1_ps(charge[inr+1]);
1491     jq2              = _mm256_set1_ps(charge[inr+2]);
1492     jq3              = _mm256_set1_ps(charge[inr+3]);
1493     vdwjidx0A        = 2*vdwtype[inr+0];
1494     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1495     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1496     qq11             = _mm256_mul_ps(iq1,jq1);
1497     qq12             = _mm256_mul_ps(iq1,jq2);
1498     qq13             = _mm256_mul_ps(iq1,jq3);
1499     qq21             = _mm256_mul_ps(iq2,jq1);
1500     qq22             = _mm256_mul_ps(iq2,jq2);
1501     qq23             = _mm256_mul_ps(iq2,jq3);
1502     qq31             = _mm256_mul_ps(iq3,jq1);
1503     qq32             = _mm256_mul_ps(iq3,jq2);
1504     qq33             = _mm256_mul_ps(iq3,jq3);
1505
1506     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1507     rcutoff_scalar   = fr->ic->rcoulomb;
1508     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1509     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1510
1511     /* Avoid stupid compiler warnings */
1512     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1513     j_coord_offsetA = 0;
1514     j_coord_offsetB = 0;
1515     j_coord_offsetC = 0;
1516     j_coord_offsetD = 0;
1517     j_coord_offsetE = 0;
1518     j_coord_offsetF = 0;
1519     j_coord_offsetG = 0;
1520     j_coord_offsetH = 0;
1521
1522     outeriter        = 0;
1523     inneriter        = 0;
1524
1525     for(iidx=0;iidx<4*DIM;iidx++)
1526     {
1527         scratch[iidx] = 0.0;
1528     }
1529
1530     /* Start outer loop over neighborlists */
1531     for(iidx=0; iidx<nri; iidx++)
1532     {
1533         /* Load shift vector for this list */
1534         i_shift_offset   = DIM*shiftidx[iidx];
1535
1536         /* Load limits for loop over neighbors */
1537         j_index_start    = jindex[iidx];
1538         j_index_end      = jindex[iidx+1];
1539
1540         /* Get outer coordinate index */
1541         inr              = iinr[iidx];
1542         i_coord_offset   = DIM*inr;
1543
1544         /* Load i particle coords and add shift vector */
1545         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1546                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1547
1548         fix0             = _mm256_setzero_ps();
1549         fiy0             = _mm256_setzero_ps();
1550         fiz0             = _mm256_setzero_ps();
1551         fix1             = _mm256_setzero_ps();
1552         fiy1             = _mm256_setzero_ps();
1553         fiz1             = _mm256_setzero_ps();
1554         fix2             = _mm256_setzero_ps();
1555         fiy2             = _mm256_setzero_ps();
1556         fiz2             = _mm256_setzero_ps();
1557         fix3             = _mm256_setzero_ps();
1558         fiy3             = _mm256_setzero_ps();
1559         fiz3             = _mm256_setzero_ps();
1560
1561         /* Start inner kernel loop */
1562         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1563         {
1564
1565             /* Get j neighbor index, and coordinate index */
1566             jnrA             = jjnr[jidx];
1567             jnrB             = jjnr[jidx+1];
1568             jnrC             = jjnr[jidx+2];
1569             jnrD             = jjnr[jidx+3];
1570             jnrE             = jjnr[jidx+4];
1571             jnrF             = jjnr[jidx+5];
1572             jnrG             = jjnr[jidx+6];
1573             jnrH             = jjnr[jidx+7];
1574             j_coord_offsetA  = DIM*jnrA;
1575             j_coord_offsetB  = DIM*jnrB;
1576             j_coord_offsetC  = DIM*jnrC;
1577             j_coord_offsetD  = DIM*jnrD;
1578             j_coord_offsetE  = DIM*jnrE;
1579             j_coord_offsetF  = DIM*jnrF;
1580             j_coord_offsetG  = DIM*jnrG;
1581             j_coord_offsetH  = DIM*jnrH;
1582
1583             /* load j atom coordinates */
1584             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1585                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1586                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1587                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1588                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1589                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1590
1591             /* Calculate displacement vector */
1592             dx00             = _mm256_sub_ps(ix0,jx0);
1593             dy00             = _mm256_sub_ps(iy0,jy0);
1594             dz00             = _mm256_sub_ps(iz0,jz0);
1595             dx11             = _mm256_sub_ps(ix1,jx1);
1596             dy11             = _mm256_sub_ps(iy1,jy1);
1597             dz11             = _mm256_sub_ps(iz1,jz1);
1598             dx12             = _mm256_sub_ps(ix1,jx2);
1599             dy12             = _mm256_sub_ps(iy1,jy2);
1600             dz12             = _mm256_sub_ps(iz1,jz2);
1601             dx13             = _mm256_sub_ps(ix1,jx3);
1602             dy13             = _mm256_sub_ps(iy1,jy3);
1603             dz13             = _mm256_sub_ps(iz1,jz3);
1604             dx21             = _mm256_sub_ps(ix2,jx1);
1605             dy21             = _mm256_sub_ps(iy2,jy1);
1606             dz21             = _mm256_sub_ps(iz2,jz1);
1607             dx22             = _mm256_sub_ps(ix2,jx2);
1608             dy22             = _mm256_sub_ps(iy2,jy2);
1609             dz22             = _mm256_sub_ps(iz2,jz2);
1610             dx23             = _mm256_sub_ps(ix2,jx3);
1611             dy23             = _mm256_sub_ps(iy2,jy3);
1612             dz23             = _mm256_sub_ps(iz2,jz3);
1613             dx31             = _mm256_sub_ps(ix3,jx1);
1614             dy31             = _mm256_sub_ps(iy3,jy1);
1615             dz31             = _mm256_sub_ps(iz3,jz1);
1616             dx32             = _mm256_sub_ps(ix3,jx2);
1617             dy32             = _mm256_sub_ps(iy3,jy2);
1618             dz32             = _mm256_sub_ps(iz3,jz2);
1619             dx33             = _mm256_sub_ps(ix3,jx3);
1620             dy33             = _mm256_sub_ps(iy3,jy3);
1621             dz33             = _mm256_sub_ps(iz3,jz3);
1622
1623             /* Calculate squared distance and things based on it */
1624             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1625             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1626             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1627             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1628             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1629             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1630             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1631             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1632             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1633             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1634
1635             rinv00           = avx256_invsqrt_f(rsq00);
1636             rinv11           = avx256_invsqrt_f(rsq11);
1637             rinv12           = avx256_invsqrt_f(rsq12);
1638             rinv13           = avx256_invsqrt_f(rsq13);
1639             rinv21           = avx256_invsqrt_f(rsq21);
1640             rinv22           = avx256_invsqrt_f(rsq22);
1641             rinv23           = avx256_invsqrt_f(rsq23);
1642             rinv31           = avx256_invsqrt_f(rsq31);
1643             rinv32           = avx256_invsqrt_f(rsq32);
1644             rinv33           = avx256_invsqrt_f(rsq33);
1645
1646             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1647             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1648             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1649             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1650             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1651             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1652             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1653             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1654             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1655
1656             fjx0             = _mm256_setzero_ps();
1657             fjy0             = _mm256_setzero_ps();
1658             fjz0             = _mm256_setzero_ps();
1659             fjx1             = _mm256_setzero_ps();
1660             fjy1             = _mm256_setzero_ps();
1661             fjz1             = _mm256_setzero_ps();
1662             fjx2             = _mm256_setzero_ps();
1663             fjy2             = _mm256_setzero_ps();
1664             fjz2             = _mm256_setzero_ps();
1665             fjx3             = _mm256_setzero_ps();
1666             fjy3             = _mm256_setzero_ps();
1667             fjz3             = _mm256_setzero_ps();
1668
1669             /**************************
1670              * CALCULATE INTERACTIONS *
1671              **************************/
1672
1673             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1674             {
1675
1676             r00              = _mm256_mul_ps(rsq00,rinv00);
1677
1678             /* Calculate table index by multiplying r with table scale and truncate to integer */
1679             rt               = _mm256_mul_ps(r00,vftabscale);
1680             vfitab           = _mm256_cvttps_epi32(rt);
1681             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1682             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1683             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1684             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1685             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1686             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1687
1688             /* CUBIC SPLINE TABLE DISPERSION */
1689             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1690                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1691             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1692                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1693             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1694                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1695             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1696                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1697             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1698             Heps             = _mm256_mul_ps(vfeps,H);
1699             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1700             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1701             fvdw6            = _mm256_mul_ps(c6_00,FF);
1702
1703             /* CUBIC SPLINE TABLE REPULSION */
1704             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1705             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1706             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1707                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1708             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1709                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1710             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1711                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1712             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1713                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1714             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1715             Heps             = _mm256_mul_ps(vfeps,H);
1716             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1717             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1718             fvdw12           = _mm256_mul_ps(c12_00,FF);
1719             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1720
1721             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1722
1723             fscal            = fvdw;
1724
1725             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1726
1727             /* Calculate temporary vectorial force */
1728             tx               = _mm256_mul_ps(fscal,dx00);
1729             ty               = _mm256_mul_ps(fscal,dy00);
1730             tz               = _mm256_mul_ps(fscal,dz00);
1731
1732             /* Update vectorial force */
1733             fix0             = _mm256_add_ps(fix0,tx);
1734             fiy0             = _mm256_add_ps(fiy0,ty);
1735             fiz0             = _mm256_add_ps(fiz0,tz);
1736
1737             fjx0             = _mm256_add_ps(fjx0,tx);
1738             fjy0             = _mm256_add_ps(fjy0,ty);
1739             fjz0             = _mm256_add_ps(fjz0,tz);
1740
1741             }
1742
1743             /**************************
1744              * CALCULATE INTERACTIONS *
1745              **************************/
1746
1747             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1748             {
1749
1750             /* REACTION-FIELD ELECTROSTATICS */
1751             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1752
1753             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1754
1755             fscal            = felec;
1756
1757             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1758
1759             /* Calculate temporary vectorial force */
1760             tx               = _mm256_mul_ps(fscal,dx11);
1761             ty               = _mm256_mul_ps(fscal,dy11);
1762             tz               = _mm256_mul_ps(fscal,dz11);
1763
1764             /* Update vectorial force */
1765             fix1             = _mm256_add_ps(fix1,tx);
1766             fiy1             = _mm256_add_ps(fiy1,ty);
1767             fiz1             = _mm256_add_ps(fiz1,tz);
1768
1769             fjx1             = _mm256_add_ps(fjx1,tx);
1770             fjy1             = _mm256_add_ps(fjy1,ty);
1771             fjz1             = _mm256_add_ps(fjz1,tz);
1772
1773             }
1774
1775             /**************************
1776              * CALCULATE INTERACTIONS *
1777              **************************/
1778
1779             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1780             {
1781
1782             /* REACTION-FIELD ELECTROSTATICS */
1783             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1784
1785             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1786
1787             fscal            = felec;
1788
1789             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1790
1791             /* Calculate temporary vectorial force */
1792             tx               = _mm256_mul_ps(fscal,dx12);
1793             ty               = _mm256_mul_ps(fscal,dy12);
1794             tz               = _mm256_mul_ps(fscal,dz12);
1795
1796             /* Update vectorial force */
1797             fix1             = _mm256_add_ps(fix1,tx);
1798             fiy1             = _mm256_add_ps(fiy1,ty);
1799             fiz1             = _mm256_add_ps(fiz1,tz);
1800
1801             fjx2             = _mm256_add_ps(fjx2,tx);
1802             fjy2             = _mm256_add_ps(fjy2,ty);
1803             fjz2             = _mm256_add_ps(fjz2,tz);
1804
1805             }
1806
1807             /**************************
1808              * CALCULATE INTERACTIONS *
1809              **************************/
1810
1811             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1812             {
1813
1814             /* REACTION-FIELD ELECTROSTATICS */
1815             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
1816
1817             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
1818
1819             fscal            = felec;
1820
1821             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1822
1823             /* Calculate temporary vectorial force */
1824             tx               = _mm256_mul_ps(fscal,dx13);
1825             ty               = _mm256_mul_ps(fscal,dy13);
1826             tz               = _mm256_mul_ps(fscal,dz13);
1827
1828             /* Update vectorial force */
1829             fix1             = _mm256_add_ps(fix1,tx);
1830             fiy1             = _mm256_add_ps(fiy1,ty);
1831             fiz1             = _mm256_add_ps(fiz1,tz);
1832
1833             fjx3             = _mm256_add_ps(fjx3,tx);
1834             fjy3             = _mm256_add_ps(fjy3,ty);
1835             fjz3             = _mm256_add_ps(fjz3,tz);
1836
1837             }
1838
1839             /**************************
1840              * CALCULATE INTERACTIONS *
1841              **************************/
1842
1843             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1844             {
1845
1846             /* REACTION-FIELD ELECTROSTATICS */
1847             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1848
1849             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1850
1851             fscal            = felec;
1852
1853             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1854
1855             /* Calculate temporary vectorial force */
1856             tx               = _mm256_mul_ps(fscal,dx21);
1857             ty               = _mm256_mul_ps(fscal,dy21);
1858             tz               = _mm256_mul_ps(fscal,dz21);
1859
1860             /* Update vectorial force */
1861             fix2             = _mm256_add_ps(fix2,tx);
1862             fiy2             = _mm256_add_ps(fiy2,ty);
1863             fiz2             = _mm256_add_ps(fiz2,tz);
1864
1865             fjx1             = _mm256_add_ps(fjx1,tx);
1866             fjy1             = _mm256_add_ps(fjy1,ty);
1867             fjz1             = _mm256_add_ps(fjz1,tz);
1868
1869             }
1870
1871             /**************************
1872              * CALCULATE INTERACTIONS *
1873              **************************/
1874
1875             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1876             {
1877
1878             /* REACTION-FIELD ELECTROSTATICS */
1879             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1880
1881             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1882
1883             fscal            = felec;
1884
1885             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1886
1887             /* Calculate temporary vectorial force */
1888             tx               = _mm256_mul_ps(fscal,dx22);
1889             ty               = _mm256_mul_ps(fscal,dy22);
1890             tz               = _mm256_mul_ps(fscal,dz22);
1891
1892             /* Update vectorial force */
1893             fix2             = _mm256_add_ps(fix2,tx);
1894             fiy2             = _mm256_add_ps(fiy2,ty);
1895             fiz2             = _mm256_add_ps(fiz2,tz);
1896
1897             fjx2             = _mm256_add_ps(fjx2,tx);
1898             fjy2             = _mm256_add_ps(fjy2,ty);
1899             fjz2             = _mm256_add_ps(fjz2,tz);
1900
1901             }
1902
1903             /**************************
1904              * CALCULATE INTERACTIONS *
1905              **************************/
1906
1907             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1908             {
1909
1910             /* REACTION-FIELD ELECTROSTATICS */
1911             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
1912
1913             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
1914
1915             fscal            = felec;
1916
1917             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1918
1919             /* Calculate temporary vectorial force */
1920             tx               = _mm256_mul_ps(fscal,dx23);
1921             ty               = _mm256_mul_ps(fscal,dy23);
1922             tz               = _mm256_mul_ps(fscal,dz23);
1923
1924             /* Update vectorial force */
1925             fix2             = _mm256_add_ps(fix2,tx);
1926             fiy2             = _mm256_add_ps(fiy2,ty);
1927             fiz2             = _mm256_add_ps(fiz2,tz);
1928
1929             fjx3             = _mm256_add_ps(fjx3,tx);
1930             fjy3             = _mm256_add_ps(fjy3,ty);
1931             fjz3             = _mm256_add_ps(fjz3,tz);
1932
1933             }
1934
1935             /**************************
1936              * CALCULATE INTERACTIONS *
1937              **************************/
1938
1939             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1940             {
1941
1942             /* REACTION-FIELD ELECTROSTATICS */
1943             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
1944
1945             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
1946
1947             fscal            = felec;
1948
1949             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1950
1951             /* Calculate temporary vectorial force */
1952             tx               = _mm256_mul_ps(fscal,dx31);
1953             ty               = _mm256_mul_ps(fscal,dy31);
1954             tz               = _mm256_mul_ps(fscal,dz31);
1955
1956             /* Update vectorial force */
1957             fix3             = _mm256_add_ps(fix3,tx);
1958             fiy3             = _mm256_add_ps(fiy3,ty);
1959             fiz3             = _mm256_add_ps(fiz3,tz);
1960
1961             fjx1             = _mm256_add_ps(fjx1,tx);
1962             fjy1             = _mm256_add_ps(fjy1,ty);
1963             fjz1             = _mm256_add_ps(fjz1,tz);
1964
1965             }
1966
1967             /**************************
1968              * CALCULATE INTERACTIONS *
1969              **************************/
1970
1971             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1972             {
1973
1974             /* REACTION-FIELD ELECTROSTATICS */
1975             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
1976
1977             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
1978
1979             fscal            = felec;
1980
1981             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1982
1983             /* Calculate temporary vectorial force */
1984             tx               = _mm256_mul_ps(fscal,dx32);
1985             ty               = _mm256_mul_ps(fscal,dy32);
1986             tz               = _mm256_mul_ps(fscal,dz32);
1987
1988             /* Update vectorial force */
1989             fix3             = _mm256_add_ps(fix3,tx);
1990             fiy3             = _mm256_add_ps(fiy3,ty);
1991             fiz3             = _mm256_add_ps(fiz3,tz);
1992
1993             fjx2             = _mm256_add_ps(fjx2,tx);
1994             fjy2             = _mm256_add_ps(fjy2,ty);
1995             fjz2             = _mm256_add_ps(fjz2,tz);
1996
1997             }
1998
1999             /**************************
2000              * CALCULATE INTERACTIONS *
2001              **************************/
2002
2003             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2004             {
2005
2006             /* REACTION-FIELD ELECTROSTATICS */
2007             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
2008
2009             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
2010
2011             fscal            = felec;
2012
2013             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2014
2015             /* Calculate temporary vectorial force */
2016             tx               = _mm256_mul_ps(fscal,dx33);
2017             ty               = _mm256_mul_ps(fscal,dy33);
2018             tz               = _mm256_mul_ps(fscal,dz33);
2019
2020             /* Update vectorial force */
2021             fix3             = _mm256_add_ps(fix3,tx);
2022             fiy3             = _mm256_add_ps(fiy3,ty);
2023             fiz3             = _mm256_add_ps(fiz3,tz);
2024
2025             fjx3             = _mm256_add_ps(fjx3,tx);
2026             fjy3             = _mm256_add_ps(fjy3,ty);
2027             fjz3             = _mm256_add_ps(fjz3,tz);
2028
2029             }
2030
2031             fjptrA             = f+j_coord_offsetA;
2032             fjptrB             = f+j_coord_offsetB;
2033             fjptrC             = f+j_coord_offsetC;
2034             fjptrD             = f+j_coord_offsetD;
2035             fjptrE             = f+j_coord_offsetE;
2036             fjptrF             = f+j_coord_offsetF;
2037             fjptrG             = f+j_coord_offsetG;
2038             fjptrH             = f+j_coord_offsetH;
2039
2040             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2041                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2042                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2043
2044             /* Inner loop uses 324 flops */
2045         }
2046
2047         if(jidx<j_index_end)
2048         {
2049
2050             /* Get j neighbor index, and coordinate index */
2051             jnrlistA         = jjnr[jidx];
2052             jnrlistB         = jjnr[jidx+1];
2053             jnrlistC         = jjnr[jidx+2];
2054             jnrlistD         = jjnr[jidx+3];
2055             jnrlistE         = jjnr[jidx+4];
2056             jnrlistF         = jjnr[jidx+5];
2057             jnrlistG         = jjnr[jidx+6];
2058             jnrlistH         = jjnr[jidx+7];
2059             /* Sign of each element will be negative for non-real atoms.
2060              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2061              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2062              */
2063             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2064                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2065
2066             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2067             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2068             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2069             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2070             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2071             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2072             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2073             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2074             j_coord_offsetA  = DIM*jnrA;
2075             j_coord_offsetB  = DIM*jnrB;
2076             j_coord_offsetC  = DIM*jnrC;
2077             j_coord_offsetD  = DIM*jnrD;
2078             j_coord_offsetE  = DIM*jnrE;
2079             j_coord_offsetF  = DIM*jnrF;
2080             j_coord_offsetG  = DIM*jnrG;
2081             j_coord_offsetH  = DIM*jnrH;
2082
2083             /* load j atom coordinates */
2084             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2085                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2086                                                  x+j_coord_offsetE,x+j_coord_offsetF,
2087                                                  x+j_coord_offsetG,x+j_coord_offsetH,
2088                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2089                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
2090
2091             /* Calculate displacement vector */
2092             dx00             = _mm256_sub_ps(ix0,jx0);
2093             dy00             = _mm256_sub_ps(iy0,jy0);
2094             dz00             = _mm256_sub_ps(iz0,jz0);
2095             dx11             = _mm256_sub_ps(ix1,jx1);
2096             dy11             = _mm256_sub_ps(iy1,jy1);
2097             dz11             = _mm256_sub_ps(iz1,jz1);
2098             dx12             = _mm256_sub_ps(ix1,jx2);
2099             dy12             = _mm256_sub_ps(iy1,jy2);
2100             dz12             = _mm256_sub_ps(iz1,jz2);
2101             dx13             = _mm256_sub_ps(ix1,jx3);
2102             dy13             = _mm256_sub_ps(iy1,jy3);
2103             dz13             = _mm256_sub_ps(iz1,jz3);
2104             dx21             = _mm256_sub_ps(ix2,jx1);
2105             dy21             = _mm256_sub_ps(iy2,jy1);
2106             dz21             = _mm256_sub_ps(iz2,jz1);
2107             dx22             = _mm256_sub_ps(ix2,jx2);
2108             dy22             = _mm256_sub_ps(iy2,jy2);
2109             dz22             = _mm256_sub_ps(iz2,jz2);
2110             dx23             = _mm256_sub_ps(ix2,jx3);
2111             dy23             = _mm256_sub_ps(iy2,jy3);
2112             dz23             = _mm256_sub_ps(iz2,jz3);
2113             dx31             = _mm256_sub_ps(ix3,jx1);
2114             dy31             = _mm256_sub_ps(iy3,jy1);
2115             dz31             = _mm256_sub_ps(iz3,jz1);
2116             dx32             = _mm256_sub_ps(ix3,jx2);
2117             dy32             = _mm256_sub_ps(iy3,jy2);
2118             dz32             = _mm256_sub_ps(iz3,jz2);
2119             dx33             = _mm256_sub_ps(ix3,jx3);
2120             dy33             = _mm256_sub_ps(iy3,jy3);
2121             dz33             = _mm256_sub_ps(iz3,jz3);
2122
2123             /* Calculate squared distance and things based on it */
2124             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2125             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2126             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2127             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
2128             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2129             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2130             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
2131             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
2132             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
2133             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
2134
2135             rinv00           = avx256_invsqrt_f(rsq00);
2136             rinv11           = avx256_invsqrt_f(rsq11);
2137             rinv12           = avx256_invsqrt_f(rsq12);
2138             rinv13           = avx256_invsqrt_f(rsq13);
2139             rinv21           = avx256_invsqrt_f(rsq21);
2140             rinv22           = avx256_invsqrt_f(rsq22);
2141             rinv23           = avx256_invsqrt_f(rsq23);
2142             rinv31           = avx256_invsqrt_f(rsq31);
2143             rinv32           = avx256_invsqrt_f(rsq32);
2144             rinv33           = avx256_invsqrt_f(rsq33);
2145
2146             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2147             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2148             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
2149             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2150             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2151             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
2152             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
2153             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
2154             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
2155
2156             fjx0             = _mm256_setzero_ps();
2157             fjy0             = _mm256_setzero_ps();
2158             fjz0             = _mm256_setzero_ps();
2159             fjx1             = _mm256_setzero_ps();
2160             fjy1             = _mm256_setzero_ps();
2161             fjz1             = _mm256_setzero_ps();
2162             fjx2             = _mm256_setzero_ps();
2163             fjy2             = _mm256_setzero_ps();
2164             fjz2             = _mm256_setzero_ps();
2165             fjx3             = _mm256_setzero_ps();
2166             fjy3             = _mm256_setzero_ps();
2167             fjz3             = _mm256_setzero_ps();
2168
2169             /**************************
2170              * CALCULATE INTERACTIONS *
2171              **************************/
2172
2173             if (gmx_mm256_any_lt(rsq00,rcutoff2))
2174             {
2175
2176             r00              = _mm256_mul_ps(rsq00,rinv00);
2177             r00              = _mm256_andnot_ps(dummy_mask,r00);
2178
2179             /* Calculate table index by multiplying r with table scale and truncate to integer */
2180             rt               = _mm256_mul_ps(r00,vftabscale);
2181             vfitab           = _mm256_cvttps_epi32(rt);
2182             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2183             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2184             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2185             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2186             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
2187             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
2188
2189             /* CUBIC SPLINE TABLE DISPERSION */
2190             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2191                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2192             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2193                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2194             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2195                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2196             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2197                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2198             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2199             Heps             = _mm256_mul_ps(vfeps,H);
2200             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2201             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2202             fvdw6            = _mm256_mul_ps(c6_00,FF);
2203
2204             /* CUBIC SPLINE TABLE REPULSION */
2205             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
2206             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
2207             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2208                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2209             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2210                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2211             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2212                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2213             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2214                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2215             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2216             Heps             = _mm256_mul_ps(vfeps,H);
2217             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2218             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2219             fvdw12           = _mm256_mul_ps(c12_00,FF);
2220             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
2221
2222             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
2223
2224             fscal            = fvdw;
2225
2226             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2227
2228             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2229
2230             /* Calculate temporary vectorial force */
2231             tx               = _mm256_mul_ps(fscal,dx00);
2232             ty               = _mm256_mul_ps(fscal,dy00);
2233             tz               = _mm256_mul_ps(fscal,dz00);
2234
2235             /* Update vectorial force */
2236             fix0             = _mm256_add_ps(fix0,tx);
2237             fiy0             = _mm256_add_ps(fiy0,ty);
2238             fiz0             = _mm256_add_ps(fiz0,tz);
2239
2240             fjx0             = _mm256_add_ps(fjx0,tx);
2241             fjy0             = _mm256_add_ps(fjy0,ty);
2242             fjz0             = _mm256_add_ps(fjz0,tz);
2243
2244             }
2245
2246             /**************************
2247              * CALCULATE INTERACTIONS *
2248              **************************/
2249
2250             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2251             {
2252
2253             /* REACTION-FIELD ELECTROSTATICS */
2254             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
2255
2256             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2257
2258             fscal            = felec;
2259
2260             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2261
2262             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2263
2264             /* Calculate temporary vectorial force */
2265             tx               = _mm256_mul_ps(fscal,dx11);
2266             ty               = _mm256_mul_ps(fscal,dy11);
2267             tz               = _mm256_mul_ps(fscal,dz11);
2268
2269             /* Update vectorial force */
2270             fix1             = _mm256_add_ps(fix1,tx);
2271             fiy1             = _mm256_add_ps(fiy1,ty);
2272             fiz1             = _mm256_add_ps(fiz1,tz);
2273
2274             fjx1             = _mm256_add_ps(fjx1,tx);
2275             fjy1             = _mm256_add_ps(fjy1,ty);
2276             fjz1             = _mm256_add_ps(fjz1,tz);
2277
2278             }
2279
2280             /**************************
2281              * CALCULATE INTERACTIONS *
2282              **************************/
2283
2284             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2285             {
2286
2287             /* REACTION-FIELD ELECTROSTATICS */
2288             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
2289
2290             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2291
2292             fscal            = felec;
2293
2294             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2295
2296             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2297
2298             /* Calculate temporary vectorial force */
2299             tx               = _mm256_mul_ps(fscal,dx12);
2300             ty               = _mm256_mul_ps(fscal,dy12);
2301             tz               = _mm256_mul_ps(fscal,dz12);
2302
2303             /* Update vectorial force */
2304             fix1             = _mm256_add_ps(fix1,tx);
2305             fiy1             = _mm256_add_ps(fiy1,ty);
2306             fiz1             = _mm256_add_ps(fiz1,tz);
2307
2308             fjx2             = _mm256_add_ps(fjx2,tx);
2309             fjy2             = _mm256_add_ps(fjy2,ty);
2310             fjz2             = _mm256_add_ps(fjz2,tz);
2311
2312             }
2313
2314             /**************************
2315              * CALCULATE INTERACTIONS *
2316              **************************/
2317
2318             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2319             {
2320
2321             /* REACTION-FIELD ELECTROSTATICS */
2322             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
2323
2324             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
2325
2326             fscal            = felec;
2327
2328             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2329
2330             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2331
2332             /* Calculate temporary vectorial force */
2333             tx               = _mm256_mul_ps(fscal,dx13);
2334             ty               = _mm256_mul_ps(fscal,dy13);
2335             tz               = _mm256_mul_ps(fscal,dz13);
2336
2337             /* Update vectorial force */
2338             fix1             = _mm256_add_ps(fix1,tx);
2339             fiy1             = _mm256_add_ps(fiy1,ty);
2340             fiz1             = _mm256_add_ps(fiz1,tz);
2341
2342             fjx3             = _mm256_add_ps(fjx3,tx);
2343             fjy3             = _mm256_add_ps(fjy3,ty);
2344             fjz3             = _mm256_add_ps(fjz3,tz);
2345
2346             }
2347
2348             /**************************
2349              * CALCULATE INTERACTIONS *
2350              **************************/
2351
2352             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2353             {
2354
2355             /* REACTION-FIELD ELECTROSTATICS */
2356             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
2357
2358             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2359
2360             fscal            = felec;
2361
2362             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2363
2364             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2365
2366             /* Calculate temporary vectorial force */
2367             tx               = _mm256_mul_ps(fscal,dx21);
2368             ty               = _mm256_mul_ps(fscal,dy21);
2369             tz               = _mm256_mul_ps(fscal,dz21);
2370
2371             /* Update vectorial force */
2372             fix2             = _mm256_add_ps(fix2,tx);
2373             fiy2             = _mm256_add_ps(fiy2,ty);
2374             fiz2             = _mm256_add_ps(fiz2,tz);
2375
2376             fjx1             = _mm256_add_ps(fjx1,tx);
2377             fjy1             = _mm256_add_ps(fjy1,ty);
2378             fjz1             = _mm256_add_ps(fjz1,tz);
2379
2380             }
2381
2382             /**************************
2383              * CALCULATE INTERACTIONS *
2384              **************************/
2385
2386             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2387             {
2388
2389             /* REACTION-FIELD ELECTROSTATICS */
2390             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
2391
2392             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2393
2394             fscal            = felec;
2395
2396             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2397
2398             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2399
2400             /* Calculate temporary vectorial force */
2401             tx               = _mm256_mul_ps(fscal,dx22);
2402             ty               = _mm256_mul_ps(fscal,dy22);
2403             tz               = _mm256_mul_ps(fscal,dz22);
2404
2405             /* Update vectorial force */
2406             fix2             = _mm256_add_ps(fix2,tx);
2407             fiy2             = _mm256_add_ps(fiy2,ty);
2408             fiz2             = _mm256_add_ps(fiz2,tz);
2409
2410             fjx2             = _mm256_add_ps(fjx2,tx);
2411             fjy2             = _mm256_add_ps(fjy2,ty);
2412             fjz2             = _mm256_add_ps(fjz2,tz);
2413
2414             }
2415
2416             /**************************
2417              * CALCULATE INTERACTIONS *
2418              **************************/
2419
2420             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2421             {
2422
2423             /* REACTION-FIELD ELECTROSTATICS */
2424             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
2425
2426             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
2427
2428             fscal            = felec;
2429
2430             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2431
2432             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2433
2434             /* Calculate temporary vectorial force */
2435             tx               = _mm256_mul_ps(fscal,dx23);
2436             ty               = _mm256_mul_ps(fscal,dy23);
2437             tz               = _mm256_mul_ps(fscal,dz23);
2438
2439             /* Update vectorial force */
2440             fix2             = _mm256_add_ps(fix2,tx);
2441             fiy2             = _mm256_add_ps(fiy2,ty);
2442             fiz2             = _mm256_add_ps(fiz2,tz);
2443
2444             fjx3             = _mm256_add_ps(fjx3,tx);
2445             fjy3             = _mm256_add_ps(fjy3,ty);
2446             fjz3             = _mm256_add_ps(fjz3,tz);
2447
2448             }
2449
2450             /**************************
2451              * CALCULATE INTERACTIONS *
2452              **************************/
2453
2454             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2455             {
2456
2457             /* REACTION-FIELD ELECTROSTATICS */
2458             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
2459
2460             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
2461
2462             fscal            = felec;
2463
2464             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2465
2466             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2467
2468             /* Calculate temporary vectorial force */
2469             tx               = _mm256_mul_ps(fscal,dx31);
2470             ty               = _mm256_mul_ps(fscal,dy31);
2471             tz               = _mm256_mul_ps(fscal,dz31);
2472
2473             /* Update vectorial force */
2474             fix3             = _mm256_add_ps(fix3,tx);
2475             fiy3             = _mm256_add_ps(fiy3,ty);
2476             fiz3             = _mm256_add_ps(fiz3,tz);
2477
2478             fjx1             = _mm256_add_ps(fjx1,tx);
2479             fjy1             = _mm256_add_ps(fjy1,ty);
2480             fjz1             = _mm256_add_ps(fjz1,tz);
2481
2482             }
2483
2484             /**************************
2485              * CALCULATE INTERACTIONS *
2486              **************************/
2487
2488             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2489             {
2490
2491             /* REACTION-FIELD ELECTROSTATICS */
2492             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
2493
2494             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
2495
2496             fscal            = felec;
2497
2498             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2499
2500             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2501
2502             /* Calculate temporary vectorial force */
2503             tx               = _mm256_mul_ps(fscal,dx32);
2504             ty               = _mm256_mul_ps(fscal,dy32);
2505             tz               = _mm256_mul_ps(fscal,dz32);
2506
2507             /* Update vectorial force */
2508             fix3             = _mm256_add_ps(fix3,tx);
2509             fiy3             = _mm256_add_ps(fiy3,ty);
2510             fiz3             = _mm256_add_ps(fiz3,tz);
2511
2512             fjx2             = _mm256_add_ps(fjx2,tx);
2513             fjy2             = _mm256_add_ps(fjy2,ty);
2514             fjz2             = _mm256_add_ps(fjz2,tz);
2515
2516             }
2517
2518             /**************************
2519              * CALCULATE INTERACTIONS *
2520              **************************/
2521
2522             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2523             {
2524
2525             /* REACTION-FIELD ELECTROSTATICS */
2526             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
2527
2528             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
2529
2530             fscal            = felec;
2531
2532             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2533
2534             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2535
2536             /* Calculate temporary vectorial force */
2537             tx               = _mm256_mul_ps(fscal,dx33);
2538             ty               = _mm256_mul_ps(fscal,dy33);
2539             tz               = _mm256_mul_ps(fscal,dz33);
2540
2541             /* Update vectorial force */
2542             fix3             = _mm256_add_ps(fix3,tx);
2543             fiy3             = _mm256_add_ps(fiy3,ty);
2544             fiz3             = _mm256_add_ps(fiz3,tz);
2545
2546             fjx3             = _mm256_add_ps(fjx3,tx);
2547             fjy3             = _mm256_add_ps(fjy3,ty);
2548             fjz3             = _mm256_add_ps(fjz3,tz);
2549
2550             }
2551
2552             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2553             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2554             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2555             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2556             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2557             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2558             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2559             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2560
2561             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2562                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2563                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2564
2565             /* Inner loop uses 325 flops */
2566         }
2567
2568         /* End of innermost loop */
2569
2570         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2571                                                  f+i_coord_offset,fshift+i_shift_offset);
2572
2573         /* Increment number of inner iterations */
2574         inneriter                  += j_index_end - j_index_start;
2575
2576         /* Outer loop uses 24 flops */
2577     }
2578
2579     /* Increment number of outer iterations */
2580     outeriter        += nri;
2581
2582     /* Update outer/inner flops */
2583
2584     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*325);
2585 }