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, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_VF_avx_256_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Water4
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  97     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  98     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  99     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 100     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
 101     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 102     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 103     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 104     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 105     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 106     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 107     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 108     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 109     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 110     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 111     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 112     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 113     real             *charge;
 114     int              nvdwtype;
 115     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 116     int              *vdwtype;
 117     real             *vdwparam;
 118     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 119     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 120     __m256i          vfitab;
 121     __m128i          vfitab_lo,vfitab_hi;
 122     __m128i          ifour       = _mm_set1_epi32(4);
 123     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 124     real             *vftab;
 125     __m256           dummy_mask,cutoff_mask;
 126     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 127     __m256           one     = _mm256_set1_ps(1.0);
 128     __m256           two     = _mm256_set1_ps(2.0);
 129     x                = xx[0];
 130     f                = ff[0];
 131
 132     nri              = nlist->nri;
 133     iinr             = nlist->iinr;
 134     jindex           = nlist->jindex;
 135     jjnr             = nlist->jjnr;
 136     shiftidx         = nlist->shift;
 137     gid              = nlist->gid;
 138     shiftvec         = fr->shift_vec[0];
 139     fshift           = fr->fshift[0];
 140     facel            = _mm256_set1_ps(fr->epsfac);
 141     charge           = mdatoms->chargeA;
 142     krf              = _mm256_set1_ps(fr->ic->k_rf);
 143     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 144     crf              = _mm256_set1_ps(fr->ic->c_rf);
 145     nvdwtype         = fr->ntype;
 146     vdwparam         = fr->nbfp;
 147     vdwtype          = mdatoms->typeA;
 148
 149     vftab            = kernel_data->table_vdw->data;
 150     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 151
 152     /* Setup water-specific parameters */
 153     inr              = nlist->iinr[0];
 154     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 155     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 156     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 157     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 158
 159     jq1              = _mm256_set1_ps(charge[inr+1]);
 160     jq2              = _mm256_set1_ps(charge[inr+2]);
 161     jq3              = _mm256_set1_ps(charge[inr+3]);
 162     vdwjidx0A        = 2*vdwtype[inr+0];
 163     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 164     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 165     qq11             = _mm256_mul_ps(iq1,jq1);
 166     qq12             = _mm256_mul_ps(iq1,jq2);
 167     qq13             = _mm256_mul_ps(iq1,jq3);
 168     qq21             = _mm256_mul_ps(iq2,jq1);
 169     qq22             = _mm256_mul_ps(iq2,jq2);
 170     qq23             = _mm256_mul_ps(iq2,jq3);
 171     qq31             = _mm256_mul_ps(iq3,jq1);
 172     qq32             = _mm256_mul_ps(iq3,jq2);
 173     qq33             = _mm256_mul_ps(iq3,jq3);
 174
 175     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 176     rcutoff_scalar   = fr->rcoulomb;
 177     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 178     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 179
 180     /* Avoid stupid compiler warnings */
 181     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 182     j_coord_offsetA = 0;
 183     j_coord_offsetB = 0;
 184     j_coord_offsetC = 0;
 185     j_coord_offsetD = 0;
 186     j_coord_offsetE = 0;
 187     j_coord_offsetF = 0;
 188     j_coord_offsetG = 0;
 189     j_coord_offsetH = 0;
 190
 191     outeriter        = 0;
 192     inneriter        = 0;
 193
 194     for(iidx=0;iidx<4*DIM;iidx++)
 195     {
 196         scratch[iidx] = 0.0;
 197     }
 198
 199     /* Start outer loop over neighborlists */
 200     for(iidx=0; iidx<nri; iidx++)
 201     {
 202         /* Load shift vector for this list */
 203         i_shift_offset   = DIM*shiftidx[iidx];
 204
 205         /* Load limits for loop over neighbors */
 206         j_index_start    = jindex[iidx];
 207         j_index_end      = jindex[iidx+1];
 208
 209         /* Get outer coordinate index */
 210         inr              = iinr[iidx];
 211         i_coord_offset   = DIM*inr;
 212
 213         /* Load i particle coords and add shift vector */
 214         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 215                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 216
 217         fix0             = _mm256_setzero_ps();
 218         fiy0             = _mm256_setzero_ps();
 219         fiz0             = _mm256_setzero_ps();
 220         fix1             = _mm256_setzero_ps();
 221         fiy1             = _mm256_setzero_ps();
 222         fiz1             = _mm256_setzero_ps();
 223         fix2             = _mm256_setzero_ps();
 224         fiy2             = _mm256_setzero_ps();
 225         fiz2             = _mm256_setzero_ps();
 226         fix3             = _mm256_setzero_ps();
 227         fiy3             = _mm256_setzero_ps();
 228         fiz3             = _mm256_setzero_ps();
 229
 230         /* Reset potential sums */
 231         velecsum         = _mm256_setzero_ps();
 232         vvdwsum          = _mm256_setzero_ps();
 233
 234         /* Start inner kernel loop */
 235         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 236         {
 237
 238             /* Get j neighbor index, and coordinate index */
 239             jnrA             = jjnr[jidx];
 240             jnrB             = jjnr[jidx+1];
 241             jnrC             = jjnr[jidx+2];
 242             jnrD             = jjnr[jidx+3];
 243             jnrE             = jjnr[jidx+4];
 244             jnrF             = jjnr[jidx+5];
 245             jnrG             = jjnr[jidx+6];
 246             jnrH             = jjnr[jidx+7];
 247             j_coord_offsetA  = DIM*jnrA;
 248             j_coord_offsetB  = DIM*jnrB;
 249             j_coord_offsetC  = DIM*jnrC;
 250             j_coord_offsetD  = DIM*jnrD;
 251             j_coord_offsetE  = DIM*jnrE;
 252             j_coord_offsetF  = DIM*jnrF;
 253             j_coord_offsetG  = DIM*jnrG;
 254             j_coord_offsetH  = DIM*jnrH;
 255
 256             /* load j atom coordinates */
 257             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 258                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 259                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 260                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 261                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 262                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 263
 264             /* Calculate displacement vector */
 265             dx00             = _mm256_sub_ps(ix0,jx0);
 266             dy00             = _mm256_sub_ps(iy0,jy0);
 267             dz00             = _mm256_sub_ps(iz0,jz0);
 268             dx11             = _mm256_sub_ps(ix1,jx1);
 269             dy11             = _mm256_sub_ps(iy1,jy1);
 270             dz11             = _mm256_sub_ps(iz1,jz1);
 271             dx12             = _mm256_sub_ps(ix1,jx2);
 272             dy12             = _mm256_sub_ps(iy1,jy2);
 273             dz12             = _mm256_sub_ps(iz1,jz2);
 274             dx13             = _mm256_sub_ps(ix1,jx3);
 275             dy13             = _mm256_sub_ps(iy1,jy3);
 276             dz13             = _mm256_sub_ps(iz1,jz3);
 277             dx21             = _mm256_sub_ps(ix2,jx1);
 278             dy21             = _mm256_sub_ps(iy2,jy1);
 279             dz21             = _mm256_sub_ps(iz2,jz1);
 280             dx22             = _mm256_sub_ps(ix2,jx2);
 281             dy22             = _mm256_sub_ps(iy2,jy2);
 282             dz22             = _mm256_sub_ps(iz2,jz2);
 283             dx23             = _mm256_sub_ps(ix2,jx3);
 284             dy23             = _mm256_sub_ps(iy2,jy3);
 285             dz23             = _mm256_sub_ps(iz2,jz3);
 286             dx31             = _mm256_sub_ps(ix3,jx1);
 287             dy31             = _mm256_sub_ps(iy3,jy1);
 288             dz31             = _mm256_sub_ps(iz3,jz1);
 289             dx32             = _mm256_sub_ps(ix3,jx2);
 290             dy32             = _mm256_sub_ps(iy3,jy2);
 291             dz32             = _mm256_sub_ps(iz3,jz2);
 292             dx33             = _mm256_sub_ps(ix3,jx3);
 293             dy33             = _mm256_sub_ps(iy3,jy3);
 294             dz33             = _mm256_sub_ps(iz3,jz3);
 295
 296             /* Calculate squared distance and things based on it */
 297             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 298             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 299             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 300             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 301             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 302             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 303             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 304             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 305             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 306             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 307
 308             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 309             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 310             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 311             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 312             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 313             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 314             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 315             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 316             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 317             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 318
 319             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 320             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 321             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 322             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 323             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 324             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 325             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 326             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 327             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 328
 329             fjx0             = _mm256_setzero_ps();
 330             fjy0             = _mm256_setzero_ps();
 331             fjz0             = _mm256_setzero_ps();
 332             fjx1             = _mm256_setzero_ps();
 333             fjy1             = _mm256_setzero_ps();
 334             fjz1             = _mm256_setzero_ps();
 335             fjx2             = _mm256_setzero_ps();
 336             fjy2             = _mm256_setzero_ps();
 337             fjz2             = _mm256_setzero_ps();
 338             fjx3             = _mm256_setzero_ps();
 339             fjy3             = _mm256_setzero_ps();
 340             fjz3             = _mm256_setzero_ps();
 341
 342             /**************************
 343              * CALCULATE INTERACTIONS *
 344              **************************/
 345
 346             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 347             {
 348
 349             r00              = _mm256_mul_ps(rsq00,rinv00);
 350
 351             /* Calculate table index by multiplying r with table scale and truncate to integer */
 352             rt               = _mm256_mul_ps(r00,vftabscale);
 353             vfitab           = _mm256_cvttps_epi32(rt);
 354             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 355             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 356             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 357             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 358             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 359             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 360
 361             /* CUBIC SPLINE TABLE DISPERSION */
 362             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 363                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 364             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 365                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 366             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 367                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 368             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 369                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 370             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 371             Heps             = _mm256_mul_ps(vfeps,H);
 372             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 373             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 374             vvdw6            = _mm256_mul_ps(c6_00,VV);
 375             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 376             fvdw6            = _mm256_mul_ps(c6_00,FF);
 377
 378             /* CUBIC SPLINE TABLE REPULSION */
 379             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 380             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 381             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 382                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 383             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 384                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 385             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 386                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 387             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 388                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 389             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 390             Heps             = _mm256_mul_ps(vfeps,H);
 391             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 392             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 393             vvdw12           = _mm256_mul_ps(c12_00,VV);
 394             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 395             fvdw12           = _mm256_mul_ps(c12_00,FF);
 396             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 397             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 398
 399             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 400
 401             /* Update potential sum for this i atom from the interaction with this j atom. */
 402             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 403             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 404
 405             fscal            = fvdw;
 406
 407             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm256_mul_ps(fscal,dx00);
 411             ty               = _mm256_mul_ps(fscal,dy00);
 412             tz               = _mm256_mul_ps(fscal,dz00);
 413
 414             /* Update vectorial force */
 415             fix0             = _mm256_add_ps(fix0,tx);
 416             fiy0             = _mm256_add_ps(fiy0,ty);
 417             fiz0             = _mm256_add_ps(fiz0,tz);
 418
 419             fjx0             = _mm256_add_ps(fjx0,tx);
 420             fjy0             = _mm256_add_ps(fjy0,ty);
 421             fjz0             = _mm256_add_ps(fjz0,tz);
 422
 423             }
 424
 425             /**************************
 426              * CALCULATE INTERACTIONS *
 427              **************************/
 428
 429             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 430             {
 431
 432             /* REACTION-FIELD ELECTROSTATICS */
 433             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 434             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 435
 436             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 437
 438             /* Update potential sum for this i atom from the interaction with this j atom. */
 439             velec            = _mm256_and_ps(velec,cutoff_mask);
 440             velecsum         = _mm256_add_ps(velecsum,velec);
 441
 442             fscal            = felec;
 443
 444             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 445
 446             /* Calculate temporary vectorial force */
 447             tx               = _mm256_mul_ps(fscal,dx11);
 448             ty               = _mm256_mul_ps(fscal,dy11);
 449             tz               = _mm256_mul_ps(fscal,dz11);
 450
 451             /* Update vectorial force */
 452             fix1             = _mm256_add_ps(fix1,tx);
 453             fiy1             = _mm256_add_ps(fiy1,ty);
 454             fiz1             = _mm256_add_ps(fiz1,tz);
 455
 456             fjx1             = _mm256_add_ps(fjx1,tx);
 457             fjy1             = _mm256_add_ps(fjy1,ty);
 458             fjz1             = _mm256_add_ps(fjz1,tz);
 459
 460             }
 461
 462             /**************************
 463              * CALCULATE INTERACTIONS *
 464              **************************/
 465
 466             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 467             {
 468
 469             /* REACTION-FIELD ELECTROSTATICS */
 470             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 471             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 472
 473             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 474
 475             /* Update potential sum for this i atom from the interaction with this j atom. */
 476             velec            = _mm256_and_ps(velec,cutoff_mask);
 477             velecsum         = _mm256_add_ps(velecsum,velec);
 478
 479             fscal            = felec;
 480
 481             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 482
 483             /* Calculate temporary vectorial force */
 484             tx               = _mm256_mul_ps(fscal,dx12);
 485             ty               = _mm256_mul_ps(fscal,dy12);
 486             tz               = _mm256_mul_ps(fscal,dz12);
 487
 488             /* Update vectorial force */
 489             fix1             = _mm256_add_ps(fix1,tx);
 490             fiy1             = _mm256_add_ps(fiy1,ty);
 491             fiz1             = _mm256_add_ps(fiz1,tz);
 492
 493             fjx2             = _mm256_add_ps(fjx2,tx);
 494             fjy2             = _mm256_add_ps(fjy2,ty);
 495             fjz2             = _mm256_add_ps(fjz2,tz);
 496
 497             }
 498
 499             /**************************
 500              * CALCULATE INTERACTIONS *
 501              **************************/
 502
 503             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 504             {
 505
 506             /* REACTION-FIELD ELECTROSTATICS */
 507             velec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_add_ps(rinv13,_mm256_mul_ps(krf,rsq13)),crf));
 508             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
 509
 510             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
 511
 512             /* Update potential sum for this i atom from the interaction with this j atom. */
 513             velec            = _mm256_and_ps(velec,cutoff_mask);
 514             velecsum         = _mm256_add_ps(velecsum,velec);
 515
 516             fscal            = felec;
 517
 518             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 519
 520             /* Calculate temporary vectorial force */
 521             tx               = _mm256_mul_ps(fscal,dx13);
 522             ty               = _mm256_mul_ps(fscal,dy13);
 523             tz               = _mm256_mul_ps(fscal,dz13);
 524
 525             /* Update vectorial force */
 526             fix1             = _mm256_add_ps(fix1,tx);
 527             fiy1             = _mm256_add_ps(fiy1,ty);
 528             fiz1             = _mm256_add_ps(fiz1,tz);
 529
 530             fjx3             = _mm256_add_ps(fjx3,tx);
 531             fjy3             = _mm256_add_ps(fjy3,ty);
 532             fjz3             = _mm256_add_ps(fjz3,tz);
 533
 534             }
 535
 536             /**************************
 537              * CALCULATE INTERACTIONS *
 538              **************************/
 539
 540             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 541             {
 542
 543             /* REACTION-FIELD ELECTROSTATICS */
 544             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
 545             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
 546
 547             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
 548
 549             /* Update potential sum for this i atom from the interaction with this j atom. */
 550             velec            = _mm256_and_ps(velec,cutoff_mask);
 551             velecsum         = _mm256_add_ps(velecsum,velec);
 552
 553             fscal            = felec;
 554
 555             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 556
 557             /* Calculate temporary vectorial force */
 558             tx               = _mm256_mul_ps(fscal,dx21);
 559             ty               = _mm256_mul_ps(fscal,dy21);
 560             tz               = _mm256_mul_ps(fscal,dz21);
 561
 562             /* Update vectorial force */
 563             fix2             = _mm256_add_ps(fix2,tx);
 564             fiy2             = _mm256_add_ps(fiy2,ty);
 565             fiz2             = _mm256_add_ps(fiz2,tz);
 566
 567             fjx1             = _mm256_add_ps(fjx1,tx);
 568             fjy1             = _mm256_add_ps(fjy1,ty);
 569             fjz1             = _mm256_add_ps(fjz1,tz);
 570
 571             }
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 578             {
 579
 580             /* REACTION-FIELD ELECTROSTATICS */
 581             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
 582             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
 583
 584             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             velec            = _mm256_and_ps(velec,cutoff_mask);
 588             velecsum         = _mm256_add_ps(velecsum,velec);
 589
 590             fscal            = felec;
 591
 592             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 593
 594             /* Calculate temporary vectorial force */
 595             tx               = _mm256_mul_ps(fscal,dx22);
 596             ty               = _mm256_mul_ps(fscal,dy22);
 597             tz               = _mm256_mul_ps(fscal,dz22);
 598
 599             /* Update vectorial force */
 600             fix2             = _mm256_add_ps(fix2,tx);
 601             fiy2             = _mm256_add_ps(fiy2,ty);
 602             fiz2             = _mm256_add_ps(fiz2,tz);
 603
 604             fjx2             = _mm256_add_ps(fjx2,tx);
 605             fjy2             = _mm256_add_ps(fjy2,ty);
 606             fjz2             = _mm256_add_ps(fjz2,tz);
 607
 608             }
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 615             {
 616
 617             /* REACTION-FIELD ELECTROSTATICS */
 618             velec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_add_ps(rinv23,_mm256_mul_ps(krf,rsq23)),crf));
 619             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
 620
 621             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
 622
 623             /* Update potential sum for this i atom from the interaction with this j atom. */
 624             velec            = _mm256_and_ps(velec,cutoff_mask);
 625             velecsum         = _mm256_add_ps(velecsum,velec);
 626
 627             fscal            = felec;
 628
 629             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 630
 631             /* Calculate temporary vectorial force */
 632             tx               = _mm256_mul_ps(fscal,dx23);
 633             ty               = _mm256_mul_ps(fscal,dy23);
 634             tz               = _mm256_mul_ps(fscal,dz23);
 635
 636             /* Update vectorial force */
 637             fix2             = _mm256_add_ps(fix2,tx);
 638             fiy2             = _mm256_add_ps(fiy2,ty);
 639             fiz2             = _mm256_add_ps(fiz2,tz);
 640
 641             fjx3             = _mm256_add_ps(fjx3,tx);
 642             fjy3             = _mm256_add_ps(fjy3,ty);
 643             fjz3             = _mm256_add_ps(fjz3,tz);
 644
 645             }
 646
 647             /**************************
 648              * CALCULATE INTERACTIONS *
 649              **************************/
 650
 651             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 652             {
 653
 654             /* REACTION-FIELD ELECTROSTATICS */
 655             velec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_add_ps(rinv31,_mm256_mul_ps(krf,rsq31)),crf));
 656             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
 657
 658             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
 659
 660             /* Update potential sum for this i atom from the interaction with this j atom. */
 661             velec            = _mm256_and_ps(velec,cutoff_mask);
 662             velecsum         = _mm256_add_ps(velecsum,velec);
 663
 664             fscal            = felec;
 665
 666             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 667
 668             /* Calculate temporary vectorial force */
 669             tx               = _mm256_mul_ps(fscal,dx31);
 670             ty               = _mm256_mul_ps(fscal,dy31);
 671             tz               = _mm256_mul_ps(fscal,dz31);
 672
 673             /* Update vectorial force */
 674             fix3             = _mm256_add_ps(fix3,tx);
 675             fiy3             = _mm256_add_ps(fiy3,ty);
 676             fiz3             = _mm256_add_ps(fiz3,tz);
 677
 678             fjx1             = _mm256_add_ps(fjx1,tx);
 679             fjy1             = _mm256_add_ps(fjy1,ty);
 680             fjz1             = _mm256_add_ps(fjz1,tz);
 681
 682             }
 683
 684             /**************************
 685              * CALCULATE INTERACTIONS *
 686              **************************/
 687
 688             if (gmx_mm256_any_lt(rsq32,rcutoff2))
 689             {
 690
 691             /* REACTION-FIELD ELECTROSTATICS */
 692             velec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_add_ps(rinv32,_mm256_mul_ps(krf,rsq32)),crf));
 693             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
 694
 695             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
 696
 697             /* Update potential sum for this i atom from the interaction with this j atom. */
 698             velec            = _mm256_and_ps(velec,cutoff_mask);
 699             velecsum         = _mm256_add_ps(velecsum,velec);
 700
 701             fscal            = felec;
 702
 703             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 704
 705             /* Calculate temporary vectorial force */
 706             tx               = _mm256_mul_ps(fscal,dx32);
 707             ty               = _mm256_mul_ps(fscal,dy32);
 708             tz               = _mm256_mul_ps(fscal,dz32);
 709
 710             /* Update vectorial force */
 711             fix3             = _mm256_add_ps(fix3,tx);
 712             fiy3             = _mm256_add_ps(fiy3,ty);
 713             fiz3             = _mm256_add_ps(fiz3,tz);
 714
 715             fjx2             = _mm256_add_ps(fjx2,tx);
 716             fjy2             = _mm256_add_ps(fjy2,ty);
 717             fjz2             = _mm256_add_ps(fjz2,tz);
 718
 719             }
 720
 721             /**************************
 722              * CALCULATE INTERACTIONS *
 723              **************************/
 724
 725             if (gmx_mm256_any_lt(rsq33,rcutoff2))
 726             {
 727
 728             /* REACTION-FIELD ELECTROSTATICS */
 729             velec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_add_ps(rinv33,_mm256_mul_ps(krf,rsq33)),crf));
 730             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
 731
 732             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
 733
 734             /* Update potential sum for this i atom from the interaction with this j atom. */
 735             velec            = _mm256_and_ps(velec,cutoff_mask);
 736             velecsum         = _mm256_add_ps(velecsum,velec);
 737
 738             fscal            = felec;
 739
 740             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 741
 742             /* Calculate temporary vectorial force */
 743             tx               = _mm256_mul_ps(fscal,dx33);
 744             ty               = _mm256_mul_ps(fscal,dy33);
 745             tz               = _mm256_mul_ps(fscal,dz33);
 746
 747             /* Update vectorial force */
 748             fix3             = _mm256_add_ps(fix3,tx);
 749             fiy3             = _mm256_add_ps(fiy3,ty);
 750             fiz3             = _mm256_add_ps(fiz3,tz);
 751
 752             fjx3             = _mm256_add_ps(fjx3,tx);
 753             fjy3             = _mm256_add_ps(fjy3,ty);
 754             fjz3             = _mm256_add_ps(fjz3,tz);
 755
 756             }
 757
 758             fjptrA             = f+j_coord_offsetA;
 759             fjptrB             = f+j_coord_offsetB;
 760             fjptrC             = f+j_coord_offsetC;
 761             fjptrD             = f+j_coord_offsetD;
 762             fjptrE             = f+j_coord_offsetE;
 763             fjptrF             = f+j_coord_offsetF;
 764             fjptrG             = f+j_coord_offsetG;
 765             fjptrH             = f+j_coord_offsetH;
 766
 767             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 768                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 769                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 770
 771             /* Inner loop uses 387 flops */
 772         }
 773
 774         if(jidx<j_index_end)
 775         {
 776
 777             /* Get j neighbor index, and coordinate index */
 778             jnrlistA         = jjnr[jidx];
 779             jnrlistB         = jjnr[jidx+1];
 780             jnrlistC         = jjnr[jidx+2];
 781             jnrlistD         = jjnr[jidx+3];
 782             jnrlistE         = jjnr[jidx+4];
 783             jnrlistF         = jjnr[jidx+5];
 784             jnrlistG         = jjnr[jidx+6];
 785             jnrlistH         = jjnr[jidx+7];
 786             /* Sign of each element will be negative for non-real atoms.
 787              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 788              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 789              */
 790             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 791                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 792
 793             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 794             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 795             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 796             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 797             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 798             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 799             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 800             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 801             j_coord_offsetA  = DIM*jnrA;
 802             j_coord_offsetB  = DIM*jnrB;
 803             j_coord_offsetC  = DIM*jnrC;
 804             j_coord_offsetD  = DIM*jnrD;
 805             j_coord_offsetE  = DIM*jnrE;
 806             j_coord_offsetF  = DIM*jnrF;
 807             j_coord_offsetG  = DIM*jnrG;
 808             j_coord_offsetH  = DIM*jnrH;
 809
 810             /* load j atom coordinates */
 811             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 812                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 813                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 814                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 815                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 816                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 817
 818             /* Calculate displacement vector */
 819             dx00             = _mm256_sub_ps(ix0,jx0);
 820             dy00             = _mm256_sub_ps(iy0,jy0);
 821             dz00             = _mm256_sub_ps(iz0,jz0);
 822             dx11             = _mm256_sub_ps(ix1,jx1);
 823             dy11             = _mm256_sub_ps(iy1,jy1);
 824             dz11             = _mm256_sub_ps(iz1,jz1);
 825             dx12             = _mm256_sub_ps(ix1,jx2);
 826             dy12             = _mm256_sub_ps(iy1,jy2);
 827             dz12             = _mm256_sub_ps(iz1,jz2);
 828             dx13             = _mm256_sub_ps(ix1,jx3);
 829             dy13             = _mm256_sub_ps(iy1,jy3);
 830             dz13             = _mm256_sub_ps(iz1,jz3);
 831             dx21             = _mm256_sub_ps(ix2,jx1);
 832             dy21             = _mm256_sub_ps(iy2,jy1);
 833             dz21             = _mm256_sub_ps(iz2,jz1);
 834             dx22             = _mm256_sub_ps(ix2,jx2);
 835             dy22             = _mm256_sub_ps(iy2,jy2);
 836             dz22             = _mm256_sub_ps(iz2,jz2);
 837             dx23             = _mm256_sub_ps(ix2,jx3);
 838             dy23             = _mm256_sub_ps(iy2,jy3);
 839             dz23             = _mm256_sub_ps(iz2,jz3);
 840             dx31             = _mm256_sub_ps(ix3,jx1);
 841             dy31             = _mm256_sub_ps(iy3,jy1);
 842             dz31             = _mm256_sub_ps(iz3,jz1);
 843             dx32             = _mm256_sub_ps(ix3,jx2);
 844             dy32             = _mm256_sub_ps(iy3,jy2);
 845             dz32             = _mm256_sub_ps(iz3,jz2);
 846             dx33             = _mm256_sub_ps(ix3,jx3);
 847             dy33             = _mm256_sub_ps(iy3,jy3);
 848             dz33             = _mm256_sub_ps(iz3,jz3);
 849
 850             /* Calculate squared distance and things based on it */
 851             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 852             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 853             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 854             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 855             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 856             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 857             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 858             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 859             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 860             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 861
 862             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 863             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 864             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 865             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 866             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 867             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 868             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 869             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 870             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 871             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 872
 873             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 874             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 875             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 876             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 877             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 878             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 879             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 880             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 881             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 882
 883             fjx0             = _mm256_setzero_ps();
 884             fjy0             = _mm256_setzero_ps();
 885             fjz0             = _mm256_setzero_ps();
 886             fjx1             = _mm256_setzero_ps();
 887             fjy1             = _mm256_setzero_ps();
 888             fjz1             = _mm256_setzero_ps();
 889             fjx2             = _mm256_setzero_ps();
 890             fjy2             = _mm256_setzero_ps();
 891             fjz2             = _mm256_setzero_ps();
 892             fjx3             = _mm256_setzero_ps();
 893             fjy3             = _mm256_setzero_ps();
 894             fjz3             = _mm256_setzero_ps();
 895
 896             /**************************
 897              * CALCULATE INTERACTIONS *
 898              **************************/
 899
 900             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 901             {
 902
 903             r00              = _mm256_mul_ps(rsq00,rinv00);
 904             r00              = _mm256_andnot_ps(dummy_mask,r00);
 905
 906             /* Calculate table index by multiplying r with table scale and truncate to integer */
 907             rt               = _mm256_mul_ps(r00,vftabscale);
 908             vfitab           = _mm256_cvttps_epi32(rt);
 909             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 910             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 911             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 912             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 913             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 914             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 915
 916             /* CUBIC SPLINE TABLE DISPERSION */
 917             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 918                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 919             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 920                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 921             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 922                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 923             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 924                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 925             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 926             Heps             = _mm256_mul_ps(vfeps,H);
 927             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 928             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 929             vvdw6            = _mm256_mul_ps(c6_00,VV);
 930             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 931             fvdw6            = _mm256_mul_ps(c6_00,FF);
 932
 933             /* CUBIC SPLINE TABLE REPULSION */
 934             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 935             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 936             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 937                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 938             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 939                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 940             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 941                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 942             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 943                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 944             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 945             Heps             = _mm256_mul_ps(vfeps,H);
 946             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 947             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 948             vvdw12           = _mm256_mul_ps(c12_00,VV);
 949             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 950             fvdw12           = _mm256_mul_ps(c12_00,FF);
 951             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 952             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 953
 954             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 955
 956             /* Update potential sum for this i atom from the interaction with this j atom. */
 957             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 958             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 959             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 960
 961             fscal            = fvdw;
 962
 963             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 964
 965             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 966
 967             /* Calculate temporary vectorial force */
 968             tx               = _mm256_mul_ps(fscal,dx00);
 969             ty               = _mm256_mul_ps(fscal,dy00);
 970             tz               = _mm256_mul_ps(fscal,dz00);
 971
 972             /* Update vectorial force */
 973             fix0             = _mm256_add_ps(fix0,tx);
 974             fiy0             = _mm256_add_ps(fiy0,ty);
 975             fiz0             = _mm256_add_ps(fiz0,tz);
 976
 977             fjx0             = _mm256_add_ps(fjx0,tx);
 978             fjy0             = _mm256_add_ps(fjy0,ty);
 979             fjz0             = _mm256_add_ps(fjz0,tz);
 980
 981             }
 982
 983             /**************************
 984              * CALCULATE INTERACTIONS *
 985              **************************/
 986
 987             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 988             {
 989
 990             /* REACTION-FIELD ELECTROSTATICS */
 991             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 992             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 993
 994             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 995
 996             /* Update potential sum for this i atom from the interaction with this j atom. */
 997             velec            = _mm256_and_ps(velec,cutoff_mask);
 998             velec            = _mm256_andnot_ps(dummy_mask,velec);
 999             velecsum         = _mm256_add_ps(velecsum,velec);
1000
1001             fscal            = felec;
1002
1003             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1004
1005             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1006
1007             /* Calculate temporary vectorial force */
1008             tx               = _mm256_mul_ps(fscal,dx11);
1009             ty               = _mm256_mul_ps(fscal,dy11);
1010             tz               = _mm256_mul_ps(fscal,dz11);
1011
1012             /* Update vectorial force */
1013             fix1             = _mm256_add_ps(fix1,tx);
1014             fiy1             = _mm256_add_ps(fiy1,ty);
1015             fiz1             = _mm256_add_ps(fiz1,tz);
1016
1017             fjx1             = _mm256_add_ps(fjx1,tx);
1018             fjy1             = _mm256_add_ps(fjy1,ty);
1019             fjz1             = _mm256_add_ps(fjz1,tz);
1020
1021             }
1022
1023             /**************************
1024              * CALCULATE INTERACTIONS *
1025              **************************/
1026
1027             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1028             {
1029
1030             /* REACTION-FIELD ELECTROSTATICS */
1031             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
1032             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1033
1034             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1035
1036             /* Update potential sum for this i atom from the interaction with this j atom. */
1037             velec            = _mm256_and_ps(velec,cutoff_mask);
1038             velec            = _mm256_andnot_ps(dummy_mask,velec);
1039             velecsum         = _mm256_add_ps(velecsum,velec);
1040
1041             fscal            = felec;
1042
1043             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1044
1045             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1046
1047             /* Calculate temporary vectorial force */
1048             tx               = _mm256_mul_ps(fscal,dx12);
1049             ty               = _mm256_mul_ps(fscal,dy12);
1050             tz               = _mm256_mul_ps(fscal,dz12);
1051
1052             /* Update vectorial force */
1053             fix1             = _mm256_add_ps(fix1,tx);
1054             fiy1             = _mm256_add_ps(fiy1,ty);
1055             fiz1             = _mm256_add_ps(fiz1,tz);
1056
1057             fjx2             = _mm256_add_ps(fjx2,tx);
1058             fjy2             = _mm256_add_ps(fjy2,ty);
1059             fjz2             = _mm256_add_ps(fjz2,tz);
1060
1061             }
1062
1063             /**************************
1064              * CALCULATE INTERACTIONS *
1065              **************************/
1066
1067             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1068             {
1069
1070             /* REACTION-FIELD ELECTROSTATICS */
1071             velec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_add_ps(rinv13,_mm256_mul_ps(krf,rsq13)),crf));
1072             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
1073
1074             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
1075
1076             /* Update potential sum for this i atom from the interaction with this j atom. */
1077             velec            = _mm256_and_ps(velec,cutoff_mask);
1078             velec            = _mm256_andnot_ps(dummy_mask,velec);
1079             velecsum         = _mm256_add_ps(velecsum,velec);
1080
1081             fscal            = felec;
1082
1083             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1084
1085             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1086
1087             /* Calculate temporary vectorial force */
1088             tx               = _mm256_mul_ps(fscal,dx13);
1089             ty               = _mm256_mul_ps(fscal,dy13);
1090             tz               = _mm256_mul_ps(fscal,dz13);
1091
1092             /* Update vectorial force */
1093             fix1             = _mm256_add_ps(fix1,tx);
1094             fiy1             = _mm256_add_ps(fiy1,ty);
1095             fiz1             = _mm256_add_ps(fiz1,tz);
1096
1097             fjx3             = _mm256_add_ps(fjx3,tx);
1098             fjy3             = _mm256_add_ps(fjy3,ty);
1099             fjz3             = _mm256_add_ps(fjz3,tz);
1100
1101             }
1102
1103             /**************************
1104              * CALCULATE INTERACTIONS *
1105              **************************/
1106
1107             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1108             {
1109
1110             /* REACTION-FIELD ELECTROSTATICS */
1111             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
1112             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1113
1114             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1115
1116             /* Update potential sum for this i atom from the interaction with this j atom. */
1117             velec            = _mm256_and_ps(velec,cutoff_mask);
1118             velec            = _mm256_andnot_ps(dummy_mask,velec);
1119             velecsum         = _mm256_add_ps(velecsum,velec);
1120
1121             fscal            = felec;
1122
1123             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1124
1125             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1126
1127             /* Calculate temporary vectorial force */
1128             tx               = _mm256_mul_ps(fscal,dx21);
1129             ty               = _mm256_mul_ps(fscal,dy21);
1130             tz               = _mm256_mul_ps(fscal,dz21);
1131
1132             /* Update vectorial force */
1133             fix2             = _mm256_add_ps(fix2,tx);
1134             fiy2             = _mm256_add_ps(fiy2,ty);
1135             fiz2             = _mm256_add_ps(fiz2,tz);
1136
1137             fjx1             = _mm256_add_ps(fjx1,tx);
1138             fjy1             = _mm256_add_ps(fjy1,ty);
1139             fjz1             = _mm256_add_ps(fjz1,tz);
1140
1141             }
1142
1143             /**************************
1144              * CALCULATE INTERACTIONS *
1145              **************************/
1146
1147             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1148             {
1149
1150             /* REACTION-FIELD ELECTROSTATICS */
1151             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
1152             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1153
1154             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1155
1156             /* Update potential sum for this i atom from the interaction with this j atom. */
1157             velec            = _mm256_and_ps(velec,cutoff_mask);
1158             velec            = _mm256_andnot_ps(dummy_mask,velec);
1159             velecsum         = _mm256_add_ps(velecsum,velec);
1160
1161             fscal            = felec;
1162
1163             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1164
1165             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1166
1167             /* Calculate temporary vectorial force */
1168             tx               = _mm256_mul_ps(fscal,dx22);
1169             ty               = _mm256_mul_ps(fscal,dy22);
1170             tz               = _mm256_mul_ps(fscal,dz22);
1171
1172             /* Update vectorial force */
1173             fix2             = _mm256_add_ps(fix2,tx);
1174             fiy2             = _mm256_add_ps(fiy2,ty);
1175             fiz2             = _mm256_add_ps(fiz2,tz);
1176
1177             fjx2             = _mm256_add_ps(fjx2,tx);
1178             fjy2             = _mm256_add_ps(fjy2,ty);
1179             fjz2             = _mm256_add_ps(fjz2,tz);
1180
1181             }
1182
1183             /**************************
1184              * CALCULATE INTERACTIONS *
1185              **************************/
1186
1187             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1188             {
1189
1190             /* REACTION-FIELD ELECTROSTATICS */
1191             velec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_add_ps(rinv23,_mm256_mul_ps(krf,rsq23)),crf));
1192             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
1193
1194             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
1195
1196             /* Update potential sum for this i atom from the interaction with this j atom. */
1197             velec            = _mm256_and_ps(velec,cutoff_mask);
1198             velec            = _mm256_andnot_ps(dummy_mask,velec);
1199             velecsum         = _mm256_add_ps(velecsum,velec);
1200
1201             fscal            = felec;
1202
1203             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1204
1205             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1206
1207             /* Calculate temporary vectorial force */
1208             tx               = _mm256_mul_ps(fscal,dx23);
1209             ty               = _mm256_mul_ps(fscal,dy23);
1210             tz               = _mm256_mul_ps(fscal,dz23);
1211
1212             /* Update vectorial force */
1213             fix2             = _mm256_add_ps(fix2,tx);
1214             fiy2             = _mm256_add_ps(fiy2,ty);
1215             fiz2             = _mm256_add_ps(fiz2,tz);
1216
1217             fjx3             = _mm256_add_ps(fjx3,tx);
1218             fjy3             = _mm256_add_ps(fjy3,ty);
1219             fjz3             = _mm256_add_ps(fjz3,tz);
1220
1221             }
1222
1223             /**************************
1224              * CALCULATE INTERACTIONS *
1225              **************************/
1226
1227             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1228             {
1229
1230             /* REACTION-FIELD ELECTROSTATICS */
1231             velec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_add_ps(rinv31,_mm256_mul_ps(krf,rsq31)),crf));
1232             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
1233
1234             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
1235
1236             /* Update potential sum for this i atom from the interaction with this j atom. */
1237             velec            = _mm256_and_ps(velec,cutoff_mask);
1238             velec            = _mm256_andnot_ps(dummy_mask,velec);
1239             velecsum         = _mm256_add_ps(velecsum,velec);
1240
1241             fscal            = felec;
1242
1243             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1244
1245             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1246
1247             /* Calculate temporary vectorial force */
1248             tx               = _mm256_mul_ps(fscal,dx31);
1249             ty               = _mm256_mul_ps(fscal,dy31);
1250             tz               = _mm256_mul_ps(fscal,dz31);
1251
1252             /* Update vectorial force */
1253             fix3             = _mm256_add_ps(fix3,tx);
1254             fiy3             = _mm256_add_ps(fiy3,ty);
1255             fiz3             = _mm256_add_ps(fiz3,tz);
1256
1257             fjx1             = _mm256_add_ps(fjx1,tx);
1258             fjy1             = _mm256_add_ps(fjy1,ty);
1259             fjz1             = _mm256_add_ps(fjz1,tz);
1260
1261             }
1262
1263             /**************************
1264              * CALCULATE INTERACTIONS *
1265              **************************/
1266
1267             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1268             {
1269
1270             /* REACTION-FIELD ELECTROSTATICS */
1271             velec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_add_ps(rinv32,_mm256_mul_ps(krf,rsq32)),crf));
1272             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
1273
1274             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
1275
1276             /* Update potential sum for this i atom from the interaction with this j atom. */
1277             velec            = _mm256_and_ps(velec,cutoff_mask);
1278             velec            = _mm256_andnot_ps(dummy_mask,velec);
1279             velecsum         = _mm256_add_ps(velecsum,velec);
1280
1281             fscal            = felec;
1282
1283             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1284
1285             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1286
1287             /* Calculate temporary vectorial force */
1288             tx               = _mm256_mul_ps(fscal,dx32);
1289             ty               = _mm256_mul_ps(fscal,dy32);
1290             tz               = _mm256_mul_ps(fscal,dz32);
1291
1292             /* Update vectorial force */
1293             fix3             = _mm256_add_ps(fix3,tx);
1294             fiy3             = _mm256_add_ps(fiy3,ty);
1295             fiz3             = _mm256_add_ps(fiz3,tz);
1296
1297             fjx2             = _mm256_add_ps(fjx2,tx);
1298             fjy2             = _mm256_add_ps(fjy2,ty);
1299             fjz2             = _mm256_add_ps(fjz2,tz);
1300
1301             }
1302
1303             /**************************
1304              * CALCULATE INTERACTIONS *
1305              **************************/
1306
1307             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1308             {
1309
1310             /* REACTION-FIELD ELECTROSTATICS */
1311             velec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_add_ps(rinv33,_mm256_mul_ps(krf,rsq33)),crf));
1312             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
1313
1314             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
1315
1316             /* Update potential sum for this i atom from the interaction with this j atom. */
1317             velec            = _mm256_and_ps(velec,cutoff_mask);
1318             velec            = _mm256_andnot_ps(dummy_mask,velec);
1319             velecsum         = _mm256_add_ps(velecsum,velec);
1320
1321             fscal            = felec;
1322
1323             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1324
1325             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1326
1327             /* Calculate temporary vectorial force */
1328             tx               = _mm256_mul_ps(fscal,dx33);
1329             ty               = _mm256_mul_ps(fscal,dy33);
1330             tz               = _mm256_mul_ps(fscal,dz33);
1331
1332             /* Update vectorial force */
1333             fix3             = _mm256_add_ps(fix3,tx);
1334             fiy3             = _mm256_add_ps(fiy3,ty);
1335             fiz3             = _mm256_add_ps(fiz3,tz);
1336
1337             fjx3             = _mm256_add_ps(fjx3,tx);
1338             fjy3             = _mm256_add_ps(fjy3,ty);
1339             fjz3             = _mm256_add_ps(fjz3,tz);
1340
1341             }
1342
1343             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1344             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1345             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1346             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1347             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1348             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1349             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1350             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1351
1352             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1353                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1354                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1355
1356             /* Inner loop uses 388 flops */
1357         }
1358
1359         /* End of innermost loop */
1360
1361         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1362                                                  f+i_coord_offset,fshift+i_shift_offset);
1363
1364         ggid                        = gid[iidx];
1365         /* Update potential energies */
1366         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1367         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1368
1369         /* Increment number of inner iterations */
1370         inneriter                  += j_index_end - j_index_start;
1371
1372         /* Outer loop uses 26 flops */
1373     }
1374
1375     /* Increment number of outer iterations */
1376     outeriter        += nri;
1377
1378     /* Update outer/inner flops */
1379
1380     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*388);
1381 }
1382 /*
1383  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_F_avx_256_single
1384  * Electrostatics interaction: ReactionField
1385  * VdW interaction:            CubicSplineTable
1386  * Geometry:                   Water4-Water4
1387  * Calculate force/pot:        Force
1388  */
1389 void
1390 nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_F_avx_256_single
1391                     (t_nblist                    * gmx_restrict       nlist,
1392                      rvec                        * gmx_restrict          xx,
1393                      rvec                        * gmx_restrict          ff,
1394                      t_forcerec                  * gmx_restrict          fr,
1395                      t_mdatoms                   * gmx_restrict     mdatoms,
1396                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1397                      t_nrnb                      * gmx_restrict        nrnb)
1398 {
1399     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1400      * just 0 for non-waters.
1401      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1402      * jnr indices corresponding to data put in the four positions in the SIMD register.
1403      */
1404     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1405     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1406     int              jnrA,jnrB,jnrC,jnrD;
1407     int              jnrE,jnrF,jnrG,jnrH;
1408     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1409     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1410     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1411     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1412     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1413     real             rcutoff_scalar;
1414     real             *shiftvec,*fshift,*x,*f;
1415     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1416     real             scratch[4*DIM];
1417     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1418     real *           vdwioffsetptr0;
1419     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1420     real *           vdwioffsetptr1;
1421     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1422     real *           vdwioffsetptr2;
1423     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1424     real *           vdwioffsetptr3;
1425     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1426     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1427     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1428     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1429     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1430     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1431     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1432     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1433     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1434     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1435     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1436     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1437     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1438     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1439     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1440     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1441     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1442     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1443     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1444     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1445     real             *charge;
1446     int              nvdwtype;
1447     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1448     int              *vdwtype;
1449     real             *vdwparam;
1450     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1451     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1452     __m256i          vfitab;
1453     __m128i          vfitab_lo,vfitab_hi;
1454     __m128i          ifour       = _mm_set1_epi32(4);
1455     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1456     real             *vftab;
1457     __m256           dummy_mask,cutoff_mask;
1458     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1459     __m256           one     = _mm256_set1_ps(1.0);
1460     __m256           two     = _mm256_set1_ps(2.0);
1461     x                = xx[0];
1462     f                = ff[0];
1463
1464     nri              = nlist->nri;
1465     iinr             = nlist->iinr;
1466     jindex           = nlist->jindex;
1467     jjnr             = nlist->jjnr;
1468     shiftidx         = nlist->shift;
1469     gid              = nlist->gid;
1470     shiftvec         = fr->shift_vec[0];
1471     fshift           = fr->fshift[0];
1472     facel            = _mm256_set1_ps(fr->epsfac);
1473     charge           = mdatoms->chargeA;
1474     krf              = _mm256_set1_ps(fr->ic->k_rf);
1475     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
1476     crf              = _mm256_set1_ps(fr->ic->c_rf);
1477     nvdwtype         = fr->ntype;
1478     vdwparam         = fr->nbfp;
1479     vdwtype          = mdatoms->typeA;
1480
1481     vftab            = kernel_data->table_vdw->data;
1482     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
1483
1484     /* Setup water-specific parameters */
1485     inr              = nlist->iinr[0];
1486     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1487     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1488     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1489     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1490
1491     jq1              = _mm256_set1_ps(charge[inr+1]);
1492     jq2              = _mm256_set1_ps(charge[inr+2]);
1493     jq3              = _mm256_set1_ps(charge[inr+3]);
1494     vdwjidx0A        = 2*vdwtype[inr+0];
1495     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1496     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1497     qq11             = _mm256_mul_ps(iq1,jq1);
1498     qq12             = _mm256_mul_ps(iq1,jq2);
1499     qq13             = _mm256_mul_ps(iq1,jq3);
1500     qq21             = _mm256_mul_ps(iq2,jq1);
1501     qq22             = _mm256_mul_ps(iq2,jq2);
1502     qq23             = _mm256_mul_ps(iq2,jq3);
1503     qq31             = _mm256_mul_ps(iq3,jq1);
1504     qq32             = _mm256_mul_ps(iq3,jq2);
1505     qq33             = _mm256_mul_ps(iq3,jq3);
1506
1507     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1508     rcutoff_scalar   = fr->rcoulomb;
1509     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1510     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1511
1512     /* Avoid stupid compiler warnings */
1513     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1514     j_coord_offsetA = 0;
1515     j_coord_offsetB = 0;
1516     j_coord_offsetC = 0;
1517     j_coord_offsetD = 0;
1518     j_coord_offsetE = 0;
1519     j_coord_offsetF = 0;
1520     j_coord_offsetG = 0;
1521     j_coord_offsetH = 0;
1522
1523     outeriter        = 0;
1524     inneriter        = 0;
1525
1526     for(iidx=0;iidx<4*DIM;iidx++)
1527     {
1528         scratch[iidx] = 0.0;
1529     }
1530
1531     /* Start outer loop over neighborlists */
1532     for(iidx=0; iidx<nri; iidx++)
1533     {
1534         /* Load shift vector for this list */
1535         i_shift_offset   = DIM*shiftidx[iidx];
1536
1537         /* Load limits for loop over neighbors */
1538         j_index_start    = jindex[iidx];
1539         j_index_end      = jindex[iidx+1];
1540
1541         /* Get outer coordinate index */
1542         inr              = iinr[iidx];
1543         i_coord_offset   = DIM*inr;
1544
1545         /* Load i particle coords and add shift vector */
1546         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1547                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1548
1549         fix0             = _mm256_setzero_ps();
1550         fiy0             = _mm256_setzero_ps();
1551         fiz0             = _mm256_setzero_ps();
1552         fix1             = _mm256_setzero_ps();
1553         fiy1             = _mm256_setzero_ps();
1554         fiz1             = _mm256_setzero_ps();
1555         fix2             = _mm256_setzero_ps();
1556         fiy2             = _mm256_setzero_ps();
1557         fiz2             = _mm256_setzero_ps();
1558         fix3             = _mm256_setzero_ps();
1559         fiy3             = _mm256_setzero_ps();
1560         fiz3             = _mm256_setzero_ps();
1561
1562         /* Start inner kernel loop */
1563         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1564         {
1565
1566             /* Get j neighbor index, and coordinate index */
1567             jnrA             = jjnr[jidx];
1568             jnrB             = jjnr[jidx+1];
1569             jnrC             = jjnr[jidx+2];
1570             jnrD             = jjnr[jidx+3];
1571             jnrE             = jjnr[jidx+4];
1572             jnrF             = jjnr[jidx+5];
1573             jnrG             = jjnr[jidx+6];
1574             jnrH             = jjnr[jidx+7];
1575             j_coord_offsetA  = DIM*jnrA;
1576             j_coord_offsetB  = DIM*jnrB;
1577             j_coord_offsetC  = DIM*jnrC;
1578             j_coord_offsetD  = DIM*jnrD;
1579             j_coord_offsetE  = DIM*jnrE;
1580             j_coord_offsetF  = DIM*jnrF;
1581             j_coord_offsetG  = DIM*jnrG;
1582             j_coord_offsetH  = DIM*jnrH;
1583
1584             /* load j atom coordinates */
1585             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1586                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1587                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1588                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1589                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1590                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1591
1592             /* Calculate displacement vector */
1593             dx00             = _mm256_sub_ps(ix0,jx0);
1594             dy00             = _mm256_sub_ps(iy0,jy0);
1595             dz00             = _mm256_sub_ps(iz0,jz0);
1596             dx11             = _mm256_sub_ps(ix1,jx1);
1597             dy11             = _mm256_sub_ps(iy1,jy1);
1598             dz11             = _mm256_sub_ps(iz1,jz1);
1599             dx12             = _mm256_sub_ps(ix1,jx2);
1600             dy12             = _mm256_sub_ps(iy1,jy2);
1601             dz12             = _mm256_sub_ps(iz1,jz2);
1602             dx13             = _mm256_sub_ps(ix1,jx3);
1603             dy13             = _mm256_sub_ps(iy1,jy3);
1604             dz13             = _mm256_sub_ps(iz1,jz3);
1605             dx21             = _mm256_sub_ps(ix2,jx1);
1606             dy21             = _mm256_sub_ps(iy2,jy1);
1607             dz21             = _mm256_sub_ps(iz2,jz1);
1608             dx22             = _mm256_sub_ps(ix2,jx2);
1609             dy22             = _mm256_sub_ps(iy2,jy2);
1610             dz22             = _mm256_sub_ps(iz2,jz2);
1611             dx23             = _mm256_sub_ps(ix2,jx3);
1612             dy23             = _mm256_sub_ps(iy2,jy3);
1613             dz23             = _mm256_sub_ps(iz2,jz3);
1614             dx31             = _mm256_sub_ps(ix3,jx1);
1615             dy31             = _mm256_sub_ps(iy3,jy1);
1616             dz31             = _mm256_sub_ps(iz3,jz1);
1617             dx32             = _mm256_sub_ps(ix3,jx2);
1618             dy32             = _mm256_sub_ps(iy3,jy2);
1619             dz32             = _mm256_sub_ps(iz3,jz2);
1620             dx33             = _mm256_sub_ps(ix3,jx3);
1621             dy33             = _mm256_sub_ps(iy3,jy3);
1622             dz33             = _mm256_sub_ps(iz3,jz3);
1623
1624             /* Calculate squared distance and things based on it */
1625             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1626             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1627             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1628             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1629             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1630             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1631             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1632             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1633             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1634             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1635
1636             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1637             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1638             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1639             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
1640             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1641             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1642             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
1643             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
1644             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
1645             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
1646
1647             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1648             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1649             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1650             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1651             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1652             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1653             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1654             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1655             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1656
1657             fjx0             = _mm256_setzero_ps();
1658             fjy0             = _mm256_setzero_ps();
1659             fjz0             = _mm256_setzero_ps();
1660             fjx1             = _mm256_setzero_ps();
1661             fjy1             = _mm256_setzero_ps();
1662             fjz1             = _mm256_setzero_ps();
1663             fjx2             = _mm256_setzero_ps();
1664             fjy2             = _mm256_setzero_ps();
1665             fjz2             = _mm256_setzero_ps();
1666             fjx3             = _mm256_setzero_ps();
1667             fjy3             = _mm256_setzero_ps();
1668             fjz3             = _mm256_setzero_ps();
1669
1670             /**************************
1671              * CALCULATE INTERACTIONS *
1672              **************************/
1673
1674             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1675             {
1676
1677             r00              = _mm256_mul_ps(rsq00,rinv00);
1678
1679             /* Calculate table index by multiplying r with table scale and truncate to integer */
1680             rt               = _mm256_mul_ps(r00,vftabscale);
1681             vfitab           = _mm256_cvttps_epi32(rt);
1682             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1683             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1684             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1685             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1686             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1687             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1688
1689             /* CUBIC SPLINE TABLE DISPERSION */
1690             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1691                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1692             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1693                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1694             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1695                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1696             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1697                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1698             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1699             Heps             = _mm256_mul_ps(vfeps,H);
1700             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1701             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1702             fvdw6            = _mm256_mul_ps(c6_00,FF);
1703
1704             /* CUBIC SPLINE TABLE REPULSION */
1705             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1706             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1707             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1708                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1709             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1710                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1711             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1712                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1713             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1714                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1715             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1716             Heps             = _mm256_mul_ps(vfeps,H);
1717             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1718             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1719             fvdw12           = _mm256_mul_ps(c12_00,FF);
1720             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1721
1722             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1723
1724             fscal            = fvdw;
1725
1726             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1727
1728             /* Calculate temporary vectorial force */
1729             tx               = _mm256_mul_ps(fscal,dx00);
1730             ty               = _mm256_mul_ps(fscal,dy00);
1731             tz               = _mm256_mul_ps(fscal,dz00);
1732
1733             /* Update vectorial force */
1734             fix0             = _mm256_add_ps(fix0,tx);
1735             fiy0             = _mm256_add_ps(fiy0,ty);
1736             fiz0             = _mm256_add_ps(fiz0,tz);
1737
1738             fjx0             = _mm256_add_ps(fjx0,tx);
1739             fjy0             = _mm256_add_ps(fjy0,ty);
1740             fjz0             = _mm256_add_ps(fjz0,tz);
1741
1742             }
1743
1744             /**************************
1745              * CALCULATE INTERACTIONS *
1746              **************************/
1747
1748             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1749             {
1750
1751             /* REACTION-FIELD ELECTROSTATICS */
1752             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1753
1754             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1755
1756             fscal            = felec;
1757
1758             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1759
1760             /* Calculate temporary vectorial force */
1761             tx               = _mm256_mul_ps(fscal,dx11);
1762             ty               = _mm256_mul_ps(fscal,dy11);
1763             tz               = _mm256_mul_ps(fscal,dz11);
1764
1765             /* Update vectorial force */
1766             fix1             = _mm256_add_ps(fix1,tx);
1767             fiy1             = _mm256_add_ps(fiy1,ty);
1768             fiz1             = _mm256_add_ps(fiz1,tz);
1769
1770             fjx1             = _mm256_add_ps(fjx1,tx);
1771             fjy1             = _mm256_add_ps(fjy1,ty);
1772             fjz1             = _mm256_add_ps(fjz1,tz);
1773
1774             }
1775
1776             /**************************
1777              * CALCULATE INTERACTIONS *
1778              **************************/
1779
1780             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1781             {
1782
1783             /* REACTION-FIELD ELECTROSTATICS */
1784             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1785
1786             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1787
1788             fscal            = felec;
1789
1790             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1791
1792             /* Calculate temporary vectorial force */
1793             tx               = _mm256_mul_ps(fscal,dx12);
1794             ty               = _mm256_mul_ps(fscal,dy12);
1795             tz               = _mm256_mul_ps(fscal,dz12);
1796
1797             /* Update vectorial force */
1798             fix1             = _mm256_add_ps(fix1,tx);
1799             fiy1             = _mm256_add_ps(fiy1,ty);
1800             fiz1             = _mm256_add_ps(fiz1,tz);
1801
1802             fjx2             = _mm256_add_ps(fjx2,tx);
1803             fjy2             = _mm256_add_ps(fjy2,ty);
1804             fjz2             = _mm256_add_ps(fjz2,tz);
1805
1806             }
1807
1808             /**************************
1809              * CALCULATE INTERACTIONS *
1810              **************************/
1811
1812             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1813             {
1814
1815             /* REACTION-FIELD ELECTROSTATICS */
1816             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
1817
1818             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
1819
1820             fscal            = felec;
1821
1822             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1823
1824             /* Calculate temporary vectorial force */
1825             tx               = _mm256_mul_ps(fscal,dx13);
1826             ty               = _mm256_mul_ps(fscal,dy13);
1827             tz               = _mm256_mul_ps(fscal,dz13);
1828
1829             /* Update vectorial force */
1830             fix1             = _mm256_add_ps(fix1,tx);
1831             fiy1             = _mm256_add_ps(fiy1,ty);
1832             fiz1             = _mm256_add_ps(fiz1,tz);
1833
1834             fjx3             = _mm256_add_ps(fjx3,tx);
1835             fjy3             = _mm256_add_ps(fjy3,ty);
1836             fjz3             = _mm256_add_ps(fjz3,tz);
1837
1838             }
1839
1840             /**************************
1841              * CALCULATE INTERACTIONS *
1842              **************************/
1843
1844             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1845             {
1846
1847             /* REACTION-FIELD ELECTROSTATICS */
1848             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1849
1850             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1851
1852             fscal            = felec;
1853
1854             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1855
1856             /* Calculate temporary vectorial force */
1857             tx               = _mm256_mul_ps(fscal,dx21);
1858             ty               = _mm256_mul_ps(fscal,dy21);
1859             tz               = _mm256_mul_ps(fscal,dz21);
1860
1861             /* Update vectorial force */
1862             fix2             = _mm256_add_ps(fix2,tx);
1863             fiy2             = _mm256_add_ps(fiy2,ty);
1864             fiz2             = _mm256_add_ps(fiz2,tz);
1865
1866             fjx1             = _mm256_add_ps(fjx1,tx);
1867             fjy1             = _mm256_add_ps(fjy1,ty);
1868             fjz1             = _mm256_add_ps(fjz1,tz);
1869
1870             }
1871
1872             /**************************
1873              * CALCULATE INTERACTIONS *
1874              **************************/
1875
1876             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1877             {
1878
1879             /* REACTION-FIELD ELECTROSTATICS */
1880             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1881
1882             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1883
1884             fscal            = felec;
1885
1886             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1887
1888             /* Calculate temporary vectorial force */
1889             tx               = _mm256_mul_ps(fscal,dx22);
1890             ty               = _mm256_mul_ps(fscal,dy22);
1891             tz               = _mm256_mul_ps(fscal,dz22);
1892
1893             /* Update vectorial force */
1894             fix2             = _mm256_add_ps(fix2,tx);
1895             fiy2             = _mm256_add_ps(fiy2,ty);
1896             fiz2             = _mm256_add_ps(fiz2,tz);
1897
1898             fjx2             = _mm256_add_ps(fjx2,tx);
1899             fjy2             = _mm256_add_ps(fjy2,ty);
1900             fjz2             = _mm256_add_ps(fjz2,tz);
1901
1902             }
1903
1904             /**************************
1905              * CALCULATE INTERACTIONS *
1906              **************************/
1907
1908             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1909             {
1910
1911             /* REACTION-FIELD ELECTROSTATICS */
1912             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
1913
1914             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
1915
1916             fscal            = felec;
1917
1918             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1919
1920             /* Calculate temporary vectorial force */
1921             tx               = _mm256_mul_ps(fscal,dx23);
1922             ty               = _mm256_mul_ps(fscal,dy23);
1923             tz               = _mm256_mul_ps(fscal,dz23);
1924
1925             /* Update vectorial force */
1926             fix2             = _mm256_add_ps(fix2,tx);
1927             fiy2             = _mm256_add_ps(fiy2,ty);
1928             fiz2             = _mm256_add_ps(fiz2,tz);
1929
1930             fjx3             = _mm256_add_ps(fjx3,tx);
1931             fjy3             = _mm256_add_ps(fjy3,ty);
1932             fjz3             = _mm256_add_ps(fjz3,tz);
1933
1934             }
1935
1936             /**************************
1937              * CALCULATE INTERACTIONS *
1938              **************************/
1939
1940             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1941             {
1942
1943             /* REACTION-FIELD ELECTROSTATICS */
1944             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
1945
1946             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
1947
1948             fscal            = felec;
1949
1950             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1951
1952             /* Calculate temporary vectorial force */
1953             tx               = _mm256_mul_ps(fscal,dx31);
1954             ty               = _mm256_mul_ps(fscal,dy31);
1955             tz               = _mm256_mul_ps(fscal,dz31);
1956
1957             /* Update vectorial force */
1958             fix3             = _mm256_add_ps(fix3,tx);
1959             fiy3             = _mm256_add_ps(fiy3,ty);
1960             fiz3             = _mm256_add_ps(fiz3,tz);
1961
1962             fjx1             = _mm256_add_ps(fjx1,tx);
1963             fjy1             = _mm256_add_ps(fjy1,ty);
1964             fjz1             = _mm256_add_ps(fjz1,tz);
1965
1966             }
1967
1968             /**************************
1969              * CALCULATE INTERACTIONS *
1970              **************************/
1971
1972             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1973             {
1974
1975             /* REACTION-FIELD ELECTROSTATICS */
1976             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
1977
1978             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
1979
1980             fscal            = felec;
1981
1982             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1983
1984             /* Calculate temporary vectorial force */
1985             tx               = _mm256_mul_ps(fscal,dx32);
1986             ty               = _mm256_mul_ps(fscal,dy32);
1987             tz               = _mm256_mul_ps(fscal,dz32);
1988
1989             /* Update vectorial force */
1990             fix3             = _mm256_add_ps(fix3,tx);
1991             fiy3             = _mm256_add_ps(fiy3,ty);
1992             fiz3             = _mm256_add_ps(fiz3,tz);
1993
1994             fjx2             = _mm256_add_ps(fjx2,tx);
1995             fjy2             = _mm256_add_ps(fjy2,ty);
1996             fjz2             = _mm256_add_ps(fjz2,tz);
1997
1998             }
1999
2000             /**************************
2001              * CALCULATE INTERACTIONS *
2002              **************************/
2003
2004             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2005             {
2006
2007             /* REACTION-FIELD ELECTROSTATICS */
2008             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
2009
2010             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
2011
2012             fscal            = felec;
2013
2014             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2015
2016             /* Calculate temporary vectorial force */
2017             tx               = _mm256_mul_ps(fscal,dx33);
2018             ty               = _mm256_mul_ps(fscal,dy33);
2019             tz               = _mm256_mul_ps(fscal,dz33);
2020
2021             /* Update vectorial force */
2022             fix3             = _mm256_add_ps(fix3,tx);
2023             fiy3             = _mm256_add_ps(fiy3,ty);
2024             fiz3             = _mm256_add_ps(fiz3,tz);
2025
2026             fjx3             = _mm256_add_ps(fjx3,tx);
2027             fjy3             = _mm256_add_ps(fjy3,ty);
2028             fjz3             = _mm256_add_ps(fjz3,tz);
2029
2030             }
2031
2032             fjptrA             = f+j_coord_offsetA;
2033             fjptrB             = f+j_coord_offsetB;
2034             fjptrC             = f+j_coord_offsetC;
2035             fjptrD             = f+j_coord_offsetD;
2036             fjptrE             = f+j_coord_offsetE;
2037             fjptrF             = f+j_coord_offsetF;
2038             fjptrG             = f+j_coord_offsetG;
2039             fjptrH             = f+j_coord_offsetH;
2040
2041             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2042                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2043                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2044
2045             /* Inner loop uses 324 flops */
2046         }
2047
2048         if(jidx<j_index_end)
2049         {
2050
2051             /* Get j neighbor index, and coordinate index */
2052             jnrlistA         = jjnr[jidx];
2053             jnrlistB         = jjnr[jidx+1];
2054             jnrlistC         = jjnr[jidx+2];
2055             jnrlistD         = jjnr[jidx+3];
2056             jnrlistE         = jjnr[jidx+4];
2057             jnrlistF         = jjnr[jidx+5];
2058             jnrlistG         = jjnr[jidx+6];
2059             jnrlistH         = jjnr[jidx+7];
2060             /* Sign of each element will be negative for non-real atoms.
2061              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2062              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2063              */
2064             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2065                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2066
2067             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2068             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2069             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2070             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2071             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2072             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2073             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2074             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2075             j_coord_offsetA  = DIM*jnrA;
2076             j_coord_offsetB  = DIM*jnrB;
2077             j_coord_offsetC  = DIM*jnrC;
2078             j_coord_offsetD  = DIM*jnrD;
2079             j_coord_offsetE  = DIM*jnrE;
2080             j_coord_offsetF  = DIM*jnrF;
2081             j_coord_offsetG  = DIM*jnrG;
2082             j_coord_offsetH  = DIM*jnrH;
2083
2084             /* load j atom coordinates */
2085             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2086                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2087                                                  x+j_coord_offsetE,x+j_coord_offsetF,
2088                                                  x+j_coord_offsetG,x+j_coord_offsetH,
2089                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2090                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
2091
2092             /* Calculate displacement vector */
2093             dx00             = _mm256_sub_ps(ix0,jx0);
2094             dy00             = _mm256_sub_ps(iy0,jy0);
2095             dz00             = _mm256_sub_ps(iz0,jz0);
2096             dx11             = _mm256_sub_ps(ix1,jx1);
2097             dy11             = _mm256_sub_ps(iy1,jy1);
2098             dz11             = _mm256_sub_ps(iz1,jz1);
2099             dx12             = _mm256_sub_ps(ix1,jx2);
2100             dy12             = _mm256_sub_ps(iy1,jy2);
2101             dz12             = _mm256_sub_ps(iz1,jz2);
2102             dx13             = _mm256_sub_ps(ix1,jx3);
2103             dy13             = _mm256_sub_ps(iy1,jy3);
2104             dz13             = _mm256_sub_ps(iz1,jz3);
2105             dx21             = _mm256_sub_ps(ix2,jx1);
2106             dy21             = _mm256_sub_ps(iy2,jy1);
2107             dz21             = _mm256_sub_ps(iz2,jz1);
2108             dx22             = _mm256_sub_ps(ix2,jx2);
2109             dy22             = _mm256_sub_ps(iy2,jy2);
2110             dz22             = _mm256_sub_ps(iz2,jz2);
2111             dx23             = _mm256_sub_ps(ix2,jx3);
2112             dy23             = _mm256_sub_ps(iy2,jy3);
2113             dz23             = _mm256_sub_ps(iz2,jz3);
2114             dx31             = _mm256_sub_ps(ix3,jx1);
2115             dy31             = _mm256_sub_ps(iy3,jy1);
2116             dz31             = _mm256_sub_ps(iz3,jz1);
2117             dx32             = _mm256_sub_ps(ix3,jx2);
2118             dy32             = _mm256_sub_ps(iy3,jy2);
2119             dz32             = _mm256_sub_ps(iz3,jz2);
2120             dx33             = _mm256_sub_ps(ix3,jx3);
2121             dy33             = _mm256_sub_ps(iy3,jy3);
2122             dz33             = _mm256_sub_ps(iz3,jz3);
2123
2124             /* Calculate squared distance and things based on it */
2125             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2126             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2127             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2128             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
2129             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2130             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2131             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
2132             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
2133             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
2134             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
2135
2136             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
2137             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2138             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2139             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
2140             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2141             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2142             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
2143             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
2144             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
2145             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
2146
2147             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2148             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2149             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
2150             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2151             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2152             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
2153             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
2154             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
2155             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
2156
2157             fjx0             = _mm256_setzero_ps();
2158             fjy0             = _mm256_setzero_ps();
2159             fjz0             = _mm256_setzero_ps();
2160             fjx1             = _mm256_setzero_ps();
2161             fjy1             = _mm256_setzero_ps();
2162             fjz1             = _mm256_setzero_ps();
2163             fjx2             = _mm256_setzero_ps();
2164             fjy2             = _mm256_setzero_ps();
2165             fjz2             = _mm256_setzero_ps();
2166             fjx3             = _mm256_setzero_ps();
2167             fjy3             = _mm256_setzero_ps();
2168             fjz3             = _mm256_setzero_ps();
2169
2170             /**************************
2171              * CALCULATE INTERACTIONS *
2172              **************************/
2173
2174             if (gmx_mm256_any_lt(rsq00,rcutoff2))
2175             {
2176
2177             r00              = _mm256_mul_ps(rsq00,rinv00);
2178             r00              = _mm256_andnot_ps(dummy_mask,r00);
2179
2180             /* Calculate table index by multiplying r with table scale and truncate to integer */
2181             rt               = _mm256_mul_ps(r00,vftabscale);
2182             vfitab           = _mm256_cvttps_epi32(rt);
2183             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2184             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2185             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2186             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2187             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
2188             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
2189
2190             /* CUBIC SPLINE TABLE DISPERSION */
2191             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2192                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2193             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2194                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2195             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2196                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2197             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2198                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2199             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2200             Heps             = _mm256_mul_ps(vfeps,H);
2201             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2202             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2203             fvdw6            = _mm256_mul_ps(c6_00,FF);
2204
2205             /* CUBIC SPLINE TABLE REPULSION */
2206             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
2207             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
2208             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2209                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2210             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2211                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2212             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2213                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2214             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2215                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2216             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2217             Heps             = _mm256_mul_ps(vfeps,H);
2218             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2219             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2220             fvdw12           = _mm256_mul_ps(c12_00,FF);
2221             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
2222
2223             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
2224
2225             fscal            = fvdw;
2226
2227             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2228
2229             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2230
2231             /* Calculate temporary vectorial force */
2232             tx               = _mm256_mul_ps(fscal,dx00);
2233             ty               = _mm256_mul_ps(fscal,dy00);
2234             tz               = _mm256_mul_ps(fscal,dz00);
2235
2236             /* Update vectorial force */
2237             fix0             = _mm256_add_ps(fix0,tx);
2238             fiy0             = _mm256_add_ps(fiy0,ty);
2239             fiz0             = _mm256_add_ps(fiz0,tz);
2240
2241             fjx0             = _mm256_add_ps(fjx0,tx);
2242             fjy0             = _mm256_add_ps(fjy0,ty);
2243             fjz0             = _mm256_add_ps(fjz0,tz);
2244
2245             }
2246
2247             /**************************
2248              * CALCULATE INTERACTIONS *
2249              **************************/
2250
2251             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2252             {
2253
2254             /* REACTION-FIELD ELECTROSTATICS */
2255             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
2256
2257             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2258
2259             fscal            = felec;
2260
2261             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2262
2263             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2264
2265             /* Calculate temporary vectorial force */
2266             tx               = _mm256_mul_ps(fscal,dx11);
2267             ty               = _mm256_mul_ps(fscal,dy11);
2268             tz               = _mm256_mul_ps(fscal,dz11);
2269
2270             /* Update vectorial force */
2271             fix1             = _mm256_add_ps(fix1,tx);
2272             fiy1             = _mm256_add_ps(fiy1,ty);
2273             fiz1             = _mm256_add_ps(fiz1,tz);
2274
2275             fjx1             = _mm256_add_ps(fjx1,tx);
2276             fjy1             = _mm256_add_ps(fjy1,ty);
2277             fjz1             = _mm256_add_ps(fjz1,tz);
2278
2279             }
2280
2281             /**************************
2282              * CALCULATE INTERACTIONS *
2283              **************************/
2284
2285             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2286             {
2287
2288             /* REACTION-FIELD ELECTROSTATICS */
2289             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
2290
2291             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2292
2293             fscal            = felec;
2294
2295             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2296
2297             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2298
2299             /* Calculate temporary vectorial force */
2300             tx               = _mm256_mul_ps(fscal,dx12);
2301             ty               = _mm256_mul_ps(fscal,dy12);
2302             tz               = _mm256_mul_ps(fscal,dz12);
2303
2304             /* Update vectorial force */
2305             fix1             = _mm256_add_ps(fix1,tx);
2306             fiy1             = _mm256_add_ps(fiy1,ty);
2307             fiz1             = _mm256_add_ps(fiz1,tz);
2308
2309             fjx2             = _mm256_add_ps(fjx2,tx);
2310             fjy2             = _mm256_add_ps(fjy2,ty);
2311             fjz2             = _mm256_add_ps(fjz2,tz);
2312
2313             }
2314
2315             /**************************
2316              * CALCULATE INTERACTIONS *
2317              **************************/
2318
2319             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2320             {
2321
2322             /* REACTION-FIELD ELECTROSTATICS */
2323             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
2324
2325             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
2326
2327             fscal            = felec;
2328
2329             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2330
2331             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2332
2333             /* Calculate temporary vectorial force */
2334             tx               = _mm256_mul_ps(fscal,dx13);
2335             ty               = _mm256_mul_ps(fscal,dy13);
2336             tz               = _mm256_mul_ps(fscal,dz13);
2337
2338             /* Update vectorial force */
2339             fix1             = _mm256_add_ps(fix1,tx);
2340             fiy1             = _mm256_add_ps(fiy1,ty);
2341             fiz1             = _mm256_add_ps(fiz1,tz);
2342
2343             fjx3             = _mm256_add_ps(fjx3,tx);
2344             fjy3             = _mm256_add_ps(fjy3,ty);
2345             fjz3             = _mm256_add_ps(fjz3,tz);
2346
2347             }
2348
2349             /**************************
2350              * CALCULATE INTERACTIONS *
2351              **************************/
2352
2353             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2354             {
2355
2356             /* REACTION-FIELD ELECTROSTATICS */
2357             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
2358
2359             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2360
2361             fscal            = felec;
2362
2363             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2364
2365             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2366
2367             /* Calculate temporary vectorial force */
2368             tx               = _mm256_mul_ps(fscal,dx21);
2369             ty               = _mm256_mul_ps(fscal,dy21);
2370             tz               = _mm256_mul_ps(fscal,dz21);
2371
2372             /* Update vectorial force */
2373             fix2             = _mm256_add_ps(fix2,tx);
2374             fiy2             = _mm256_add_ps(fiy2,ty);
2375             fiz2             = _mm256_add_ps(fiz2,tz);
2376
2377             fjx1             = _mm256_add_ps(fjx1,tx);
2378             fjy1             = _mm256_add_ps(fjy1,ty);
2379             fjz1             = _mm256_add_ps(fjz1,tz);
2380
2381             }
2382
2383             /**************************
2384              * CALCULATE INTERACTIONS *
2385              **************************/
2386
2387             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2388             {
2389
2390             /* REACTION-FIELD ELECTROSTATICS */
2391             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
2392
2393             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2394
2395             fscal            = felec;
2396
2397             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2398
2399             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2400
2401             /* Calculate temporary vectorial force */
2402             tx               = _mm256_mul_ps(fscal,dx22);
2403             ty               = _mm256_mul_ps(fscal,dy22);
2404             tz               = _mm256_mul_ps(fscal,dz22);
2405
2406             /* Update vectorial force */
2407             fix2             = _mm256_add_ps(fix2,tx);
2408             fiy2             = _mm256_add_ps(fiy2,ty);
2409             fiz2             = _mm256_add_ps(fiz2,tz);
2410
2411             fjx2             = _mm256_add_ps(fjx2,tx);
2412             fjy2             = _mm256_add_ps(fjy2,ty);
2413             fjz2             = _mm256_add_ps(fjz2,tz);
2414
2415             }
2416
2417             /**************************
2418              * CALCULATE INTERACTIONS *
2419              **************************/
2420
2421             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2422             {
2423
2424             /* REACTION-FIELD ELECTROSTATICS */
2425             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
2426
2427             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
2428
2429             fscal            = felec;
2430
2431             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2432
2433             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2434
2435             /* Calculate temporary vectorial force */
2436             tx               = _mm256_mul_ps(fscal,dx23);
2437             ty               = _mm256_mul_ps(fscal,dy23);
2438             tz               = _mm256_mul_ps(fscal,dz23);
2439
2440             /* Update vectorial force */
2441             fix2             = _mm256_add_ps(fix2,tx);
2442             fiy2             = _mm256_add_ps(fiy2,ty);
2443             fiz2             = _mm256_add_ps(fiz2,tz);
2444
2445             fjx3             = _mm256_add_ps(fjx3,tx);
2446             fjy3             = _mm256_add_ps(fjy3,ty);
2447             fjz3             = _mm256_add_ps(fjz3,tz);
2448
2449             }
2450
2451             /**************************
2452              * CALCULATE INTERACTIONS *
2453              **************************/
2454
2455             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2456             {
2457
2458             /* REACTION-FIELD ELECTROSTATICS */
2459             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
2460
2461             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
2462
2463             fscal            = felec;
2464
2465             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2466
2467             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2468
2469             /* Calculate temporary vectorial force */
2470             tx               = _mm256_mul_ps(fscal,dx31);
2471             ty               = _mm256_mul_ps(fscal,dy31);
2472             tz               = _mm256_mul_ps(fscal,dz31);
2473
2474             /* Update vectorial force */
2475             fix3             = _mm256_add_ps(fix3,tx);
2476             fiy3             = _mm256_add_ps(fiy3,ty);
2477             fiz3             = _mm256_add_ps(fiz3,tz);
2478
2479             fjx1             = _mm256_add_ps(fjx1,tx);
2480             fjy1             = _mm256_add_ps(fjy1,ty);
2481             fjz1             = _mm256_add_ps(fjz1,tz);
2482
2483             }
2484
2485             /**************************
2486              * CALCULATE INTERACTIONS *
2487              **************************/
2488
2489             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2490             {
2491
2492             /* REACTION-FIELD ELECTROSTATICS */
2493             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
2494
2495             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
2496
2497             fscal            = felec;
2498
2499             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2500
2501             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2502
2503             /* Calculate temporary vectorial force */
2504             tx               = _mm256_mul_ps(fscal,dx32);
2505             ty               = _mm256_mul_ps(fscal,dy32);
2506             tz               = _mm256_mul_ps(fscal,dz32);
2507
2508             /* Update vectorial force */
2509             fix3             = _mm256_add_ps(fix3,tx);
2510             fiy3             = _mm256_add_ps(fiy3,ty);
2511             fiz3             = _mm256_add_ps(fiz3,tz);
2512
2513             fjx2             = _mm256_add_ps(fjx2,tx);
2514             fjy2             = _mm256_add_ps(fjy2,ty);
2515             fjz2             = _mm256_add_ps(fjz2,tz);
2516
2517             }
2518
2519             /**************************
2520              * CALCULATE INTERACTIONS *
2521              **************************/
2522
2523             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2524             {
2525
2526             /* REACTION-FIELD ELECTROSTATICS */
2527             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
2528
2529             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
2530
2531             fscal            = felec;
2532
2533             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2534
2535             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2536
2537             /* Calculate temporary vectorial force */
2538             tx               = _mm256_mul_ps(fscal,dx33);
2539             ty               = _mm256_mul_ps(fscal,dy33);
2540             tz               = _mm256_mul_ps(fscal,dz33);
2541
2542             /* Update vectorial force */
2543             fix3             = _mm256_add_ps(fix3,tx);
2544             fiy3             = _mm256_add_ps(fiy3,ty);
2545             fiz3             = _mm256_add_ps(fiz3,tz);
2546
2547             fjx3             = _mm256_add_ps(fjx3,tx);
2548             fjy3             = _mm256_add_ps(fjy3,ty);
2549             fjz3             = _mm256_add_ps(fjz3,tz);
2550
2551             }
2552
2553             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2554             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2555             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2556             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2557             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2558             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2559             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2560             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2561
2562             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2563                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2564                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2565
2566             /* Inner loop uses 325 flops */
2567         }
2568
2569         /* End of innermost loop */
2570
2571         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2572                                                  f+i_coord_offset,fshift+i_shift_offset);
2573
2574         /* Increment number of inner iterations */
2575         inneriter                  += j_index_end - j_index_start;
2576
2577         /* Outer loop uses 24 flops */
2578     }
2579
2580     /* Increment number of outer iterations */
2581     outeriter        += nri;
2582
2583     /* Update outer/inner flops */
2584
2585     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*325);
2586 }