src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_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_ElecRF_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_ElecRF_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     /* Avoid stupid compiler warnings */
 176     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 177     j_coord_offsetA = 0;
 178     j_coord_offsetB = 0;
 179     j_coord_offsetC = 0;
 180     j_coord_offsetD = 0;
 181     j_coord_offsetE = 0;
 182     j_coord_offsetF = 0;
 183     j_coord_offsetG = 0;
 184     j_coord_offsetH = 0;
 185
 186     outeriter        = 0;
 187     inneriter        = 0;
 188
 189     for(iidx=0;iidx<4*DIM;iidx++)
 190     {
 191         scratch[iidx] = 0.0;
 192     }
 193
 194     /* Start outer loop over neighborlists */
 195     for(iidx=0; iidx<nri; iidx++)
 196     {
 197         /* Load shift vector for this list */
 198         i_shift_offset   = DIM*shiftidx[iidx];
 199
 200         /* Load limits for loop over neighbors */
 201         j_index_start    = jindex[iidx];
 202         j_index_end      = jindex[iidx+1];
 203
 204         /* Get outer coordinate index */
 205         inr              = iinr[iidx];
 206         i_coord_offset   = DIM*inr;
 207
 208         /* Load i particle coords and add shift vector */
 209         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 210                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 211
 212         fix0             = _mm256_setzero_ps();
 213         fiy0             = _mm256_setzero_ps();
 214         fiz0             = _mm256_setzero_ps();
 215         fix1             = _mm256_setzero_ps();
 216         fiy1             = _mm256_setzero_ps();
 217         fiz1             = _mm256_setzero_ps();
 218         fix2             = _mm256_setzero_ps();
 219         fiy2             = _mm256_setzero_ps();
 220         fiz2             = _mm256_setzero_ps();
 221         fix3             = _mm256_setzero_ps();
 222         fiy3             = _mm256_setzero_ps();
 223         fiz3             = _mm256_setzero_ps();
 224
 225         /* Reset potential sums */
 226         velecsum         = _mm256_setzero_ps();
 227         vvdwsum          = _mm256_setzero_ps();
 228
 229         /* Start inner kernel loop */
 230         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 231         {
 232
 233             /* Get j neighbor index, and coordinate index */
 234             jnrA             = jjnr[jidx];
 235             jnrB             = jjnr[jidx+1];
 236             jnrC             = jjnr[jidx+2];
 237             jnrD             = jjnr[jidx+3];
 238             jnrE             = jjnr[jidx+4];
 239             jnrF             = jjnr[jidx+5];
 240             jnrG             = jjnr[jidx+6];
 241             jnrH             = jjnr[jidx+7];
 242             j_coord_offsetA  = DIM*jnrA;
 243             j_coord_offsetB  = DIM*jnrB;
 244             j_coord_offsetC  = DIM*jnrC;
 245             j_coord_offsetD  = DIM*jnrD;
 246             j_coord_offsetE  = DIM*jnrE;
 247             j_coord_offsetF  = DIM*jnrF;
 248             j_coord_offsetG  = DIM*jnrG;
 249             j_coord_offsetH  = DIM*jnrH;
 250
 251             /* load j atom coordinates */
 252             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 253                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 254                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 255                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 256                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 257                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 258
 259             /* Calculate displacement vector */
 260             dx00             = _mm256_sub_ps(ix0,jx0);
 261             dy00             = _mm256_sub_ps(iy0,jy0);
 262             dz00             = _mm256_sub_ps(iz0,jz0);
 263             dx11             = _mm256_sub_ps(ix1,jx1);
 264             dy11             = _mm256_sub_ps(iy1,jy1);
 265             dz11             = _mm256_sub_ps(iz1,jz1);
 266             dx12             = _mm256_sub_ps(ix1,jx2);
 267             dy12             = _mm256_sub_ps(iy1,jy2);
 268             dz12             = _mm256_sub_ps(iz1,jz2);
 269             dx13             = _mm256_sub_ps(ix1,jx3);
 270             dy13             = _mm256_sub_ps(iy1,jy3);
 271             dz13             = _mm256_sub_ps(iz1,jz3);
 272             dx21             = _mm256_sub_ps(ix2,jx1);
 273             dy21             = _mm256_sub_ps(iy2,jy1);
 274             dz21             = _mm256_sub_ps(iz2,jz1);
 275             dx22             = _mm256_sub_ps(ix2,jx2);
 276             dy22             = _mm256_sub_ps(iy2,jy2);
 277             dz22             = _mm256_sub_ps(iz2,jz2);
 278             dx23             = _mm256_sub_ps(ix2,jx3);
 279             dy23             = _mm256_sub_ps(iy2,jy3);
 280             dz23             = _mm256_sub_ps(iz2,jz3);
 281             dx31             = _mm256_sub_ps(ix3,jx1);
 282             dy31             = _mm256_sub_ps(iy3,jy1);
 283             dz31             = _mm256_sub_ps(iz3,jz1);
 284             dx32             = _mm256_sub_ps(ix3,jx2);
 285             dy32             = _mm256_sub_ps(iy3,jy2);
 286             dz32             = _mm256_sub_ps(iz3,jz2);
 287             dx33             = _mm256_sub_ps(ix3,jx3);
 288             dy33             = _mm256_sub_ps(iy3,jy3);
 289             dz33             = _mm256_sub_ps(iz3,jz3);
 290
 291             /* Calculate squared distance and things based on it */
 292             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 293             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 294             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 295             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 296             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 297             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 298             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 299             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 300             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 301             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 302
 303             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 304             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 305             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 306             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 307             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 308             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 309             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 310             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 311             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 312             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 313
 314             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 315             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 316             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 317             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 318             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 319             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 320             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 321             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 322             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 323
 324             fjx0             = _mm256_setzero_ps();
 325             fjy0             = _mm256_setzero_ps();
 326             fjz0             = _mm256_setzero_ps();
 327             fjx1             = _mm256_setzero_ps();
 328             fjy1             = _mm256_setzero_ps();
 329             fjz1             = _mm256_setzero_ps();
 330             fjx2             = _mm256_setzero_ps();
 331             fjy2             = _mm256_setzero_ps();
 332             fjz2             = _mm256_setzero_ps();
 333             fjx3             = _mm256_setzero_ps();
 334             fjy3             = _mm256_setzero_ps();
 335             fjz3             = _mm256_setzero_ps();
 336
 337             /**************************
 338              * CALCULATE INTERACTIONS *
 339              **************************/
 340
 341             r00              = _mm256_mul_ps(rsq00,rinv00);
 342
 343             /* Calculate table index by multiplying r with table scale and truncate to integer */
 344             rt               = _mm256_mul_ps(r00,vftabscale);
 345             vfitab           = _mm256_cvttps_epi32(rt);
 346             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 347             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 348             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 349             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 350             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 351             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 352
 353             /* CUBIC SPLINE TABLE DISPERSION */
 354             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 355                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 356             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 357                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 358             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 359                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 360             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 361                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 362             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 363             Heps             = _mm256_mul_ps(vfeps,H);
 364             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 365             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 366             vvdw6            = _mm256_mul_ps(c6_00,VV);
 367             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 368             fvdw6            = _mm256_mul_ps(c6_00,FF);
 369
 370             /* CUBIC SPLINE TABLE REPULSION */
 371             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 372             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 373             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 374                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 375             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 376                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 377             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 378                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 379             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 380                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 381             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 382             Heps             = _mm256_mul_ps(vfeps,H);
 383             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 384             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 385             vvdw12           = _mm256_mul_ps(c12_00,VV);
 386             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 387             fvdw12           = _mm256_mul_ps(c12_00,FF);
 388             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 389             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 393
 394             fscal            = fvdw;
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm256_mul_ps(fscal,dx00);
 398             ty               = _mm256_mul_ps(fscal,dy00);
 399             tz               = _mm256_mul_ps(fscal,dz00);
 400
 401             /* Update vectorial force */
 402             fix0             = _mm256_add_ps(fix0,tx);
 403             fiy0             = _mm256_add_ps(fiy0,ty);
 404             fiz0             = _mm256_add_ps(fiz0,tz);
 405
 406             fjx0             = _mm256_add_ps(fjx0,tx);
 407             fjy0             = _mm256_add_ps(fjy0,ty);
 408             fjz0             = _mm256_add_ps(fjz0,tz);
 409
 410             /**************************
 411              * CALCULATE INTERACTIONS *
 412              **************************/
 413
 414             /* REACTION-FIELD ELECTROSTATICS */
 415             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 416             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 417
 418             /* Update potential sum for this i atom from the interaction with this j atom. */
 419             velecsum         = _mm256_add_ps(velecsum,velec);
 420
 421             fscal            = felec;
 422
 423             /* Calculate temporary vectorial force */
 424             tx               = _mm256_mul_ps(fscal,dx11);
 425             ty               = _mm256_mul_ps(fscal,dy11);
 426             tz               = _mm256_mul_ps(fscal,dz11);
 427
 428             /* Update vectorial force */
 429             fix1             = _mm256_add_ps(fix1,tx);
 430             fiy1             = _mm256_add_ps(fiy1,ty);
 431             fiz1             = _mm256_add_ps(fiz1,tz);
 432
 433             fjx1             = _mm256_add_ps(fjx1,tx);
 434             fjy1             = _mm256_add_ps(fjy1,ty);
 435             fjz1             = _mm256_add_ps(fjz1,tz);
 436
 437             /**************************
 438              * CALCULATE INTERACTIONS *
 439              **************************/
 440
 441             /* REACTION-FIELD ELECTROSTATICS */
 442             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 443             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 444
 445             /* Update potential sum for this i atom from the interaction with this j atom. */
 446             velecsum         = _mm256_add_ps(velecsum,velec);
 447
 448             fscal            = felec;
 449
 450             /* Calculate temporary vectorial force */
 451             tx               = _mm256_mul_ps(fscal,dx12);
 452             ty               = _mm256_mul_ps(fscal,dy12);
 453             tz               = _mm256_mul_ps(fscal,dz12);
 454
 455             /* Update vectorial force */
 456             fix1             = _mm256_add_ps(fix1,tx);
 457             fiy1             = _mm256_add_ps(fiy1,ty);
 458             fiz1             = _mm256_add_ps(fiz1,tz);
 459
 460             fjx2             = _mm256_add_ps(fjx2,tx);
 461             fjy2             = _mm256_add_ps(fjy2,ty);
 462             fjz2             = _mm256_add_ps(fjz2,tz);
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             /* REACTION-FIELD ELECTROSTATICS */
 469             velec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_add_ps(rinv13,_mm256_mul_ps(krf,rsq13)),crf));
 470             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
 471
 472             /* Update potential sum for this i atom from the interaction with this j atom. */
 473             velecsum         = _mm256_add_ps(velecsum,velec);
 474
 475             fscal            = felec;
 476
 477             /* Calculate temporary vectorial force */
 478             tx               = _mm256_mul_ps(fscal,dx13);
 479             ty               = _mm256_mul_ps(fscal,dy13);
 480             tz               = _mm256_mul_ps(fscal,dz13);
 481
 482             /* Update vectorial force */
 483             fix1             = _mm256_add_ps(fix1,tx);
 484             fiy1             = _mm256_add_ps(fiy1,ty);
 485             fiz1             = _mm256_add_ps(fiz1,tz);
 486
 487             fjx3             = _mm256_add_ps(fjx3,tx);
 488             fjy3             = _mm256_add_ps(fjy3,ty);
 489             fjz3             = _mm256_add_ps(fjz3,tz);
 490
 491             /**************************
 492              * CALCULATE INTERACTIONS *
 493              **************************/
 494
 495             /* REACTION-FIELD ELECTROSTATICS */
 496             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
 497             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
 498
 499             /* Update potential sum for this i atom from the interaction with this j atom. */
 500             velecsum         = _mm256_add_ps(velecsum,velec);
 501
 502             fscal            = felec;
 503
 504             /* Calculate temporary vectorial force */
 505             tx               = _mm256_mul_ps(fscal,dx21);
 506             ty               = _mm256_mul_ps(fscal,dy21);
 507             tz               = _mm256_mul_ps(fscal,dz21);
 508
 509             /* Update vectorial force */
 510             fix2             = _mm256_add_ps(fix2,tx);
 511             fiy2             = _mm256_add_ps(fiy2,ty);
 512             fiz2             = _mm256_add_ps(fiz2,tz);
 513
 514             fjx1             = _mm256_add_ps(fjx1,tx);
 515             fjy1             = _mm256_add_ps(fjy1,ty);
 516             fjz1             = _mm256_add_ps(fjz1,tz);
 517
 518             /**************************
 519              * CALCULATE INTERACTIONS *
 520              **************************/
 521
 522             /* REACTION-FIELD ELECTROSTATICS */
 523             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
 524             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
 525
 526             /* Update potential sum for this i atom from the interaction with this j atom. */
 527             velecsum         = _mm256_add_ps(velecsum,velec);
 528
 529             fscal            = felec;
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm256_mul_ps(fscal,dx22);
 533             ty               = _mm256_mul_ps(fscal,dy22);
 534             tz               = _mm256_mul_ps(fscal,dz22);
 535
 536             /* Update vectorial force */
 537             fix2             = _mm256_add_ps(fix2,tx);
 538             fiy2             = _mm256_add_ps(fiy2,ty);
 539             fiz2             = _mm256_add_ps(fiz2,tz);
 540
 541             fjx2             = _mm256_add_ps(fjx2,tx);
 542             fjy2             = _mm256_add_ps(fjy2,ty);
 543             fjz2             = _mm256_add_ps(fjz2,tz);
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             /* REACTION-FIELD ELECTROSTATICS */
 550             velec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_add_ps(rinv23,_mm256_mul_ps(krf,rsq23)),crf));
 551             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
 552
 553             /* Update potential sum for this i atom from the interaction with this j atom. */
 554             velecsum         = _mm256_add_ps(velecsum,velec);
 555
 556             fscal            = felec;
 557
 558             /* Calculate temporary vectorial force */
 559             tx               = _mm256_mul_ps(fscal,dx23);
 560             ty               = _mm256_mul_ps(fscal,dy23);
 561             tz               = _mm256_mul_ps(fscal,dz23);
 562
 563             /* Update vectorial force */
 564             fix2             = _mm256_add_ps(fix2,tx);
 565             fiy2             = _mm256_add_ps(fiy2,ty);
 566             fiz2             = _mm256_add_ps(fiz2,tz);
 567
 568             fjx3             = _mm256_add_ps(fjx3,tx);
 569             fjy3             = _mm256_add_ps(fjy3,ty);
 570             fjz3             = _mm256_add_ps(fjz3,tz);
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             /* REACTION-FIELD ELECTROSTATICS */
 577             velec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_add_ps(rinv31,_mm256_mul_ps(krf,rsq31)),crf));
 578             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
 579
 580             /* Update potential sum for this i atom from the interaction with this j atom. */
 581             velecsum         = _mm256_add_ps(velecsum,velec);
 582
 583             fscal            = felec;
 584
 585             /* Calculate temporary vectorial force */
 586             tx               = _mm256_mul_ps(fscal,dx31);
 587             ty               = _mm256_mul_ps(fscal,dy31);
 588             tz               = _mm256_mul_ps(fscal,dz31);
 589
 590             /* Update vectorial force */
 591             fix3             = _mm256_add_ps(fix3,tx);
 592             fiy3             = _mm256_add_ps(fiy3,ty);
 593             fiz3             = _mm256_add_ps(fiz3,tz);
 594
 595             fjx1             = _mm256_add_ps(fjx1,tx);
 596             fjy1             = _mm256_add_ps(fjy1,ty);
 597             fjz1             = _mm256_add_ps(fjz1,tz);
 598
 599             /**************************
 600              * CALCULATE INTERACTIONS *
 601              **************************/
 602
 603             /* REACTION-FIELD ELECTROSTATICS */
 604             velec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_add_ps(rinv32,_mm256_mul_ps(krf,rsq32)),crf));
 605             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
 606
 607             /* Update potential sum for this i atom from the interaction with this j atom. */
 608             velecsum         = _mm256_add_ps(velecsum,velec);
 609
 610             fscal            = felec;
 611
 612             /* Calculate temporary vectorial force */
 613             tx               = _mm256_mul_ps(fscal,dx32);
 614             ty               = _mm256_mul_ps(fscal,dy32);
 615             tz               = _mm256_mul_ps(fscal,dz32);
 616
 617             /* Update vectorial force */
 618             fix3             = _mm256_add_ps(fix3,tx);
 619             fiy3             = _mm256_add_ps(fiy3,ty);
 620             fiz3             = _mm256_add_ps(fiz3,tz);
 621
 622             fjx2             = _mm256_add_ps(fjx2,tx);
 623             fjy2             = _mm256_add_ps(fjy2,ty);
 624             fjz2             = _mm256_add_ps(fjz2,tz);
 625
 626             /**************************
 627              * CALCULATE INTERACTIONS *
 628              **************************/
 629
 630             /* REACTION-FIELD ELECTROSTATICS */
 631             velec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_add_ps(rinv33,_mm256_mul_ps(krf,rsq33)),crf));
 632             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
 633
 634             /* Update potential sum for this i atom from the interaction with this j atom. */
 635             velecsum         = _mm256_add_ps(velecsum,velec);
 636
 637             fscal            = felec;
 638
 639             /* Calculate temporary vectorial force */
 640             tx               = _mm256_mul_ps(fscal,dx33);
 641             ty               = _mm256_mul_ps(fscal,dy33);
 642             tz               = _mm256_mul_ps(fscal,dz33);
 643
 644             /* Update vectorial force */
 645             fix3             = _mm256_add_ps(fix3,tx);
 646             fiy3             = _mm256_add_ps(fiy3,ty);
 647             fiz3             = _mm256_add_ps(fiz3,tz);
 648
 649             fjx3             = _mm256_add_ps(fjx3,tx);
 650             fjy3             = _mm256_add_ps(fjy3,ty);
 651             fjz3             = _mm256_add_ps(fjz3,tz);
 652
 653             fjptrA             = f+j_coord_offsetA;
 654             fjptrB             = f+j_coord_offsetB;
 655             fjptrC             = f+j_coord_offsetC;
 656             fjptrD             = f+j_coord_offsetD;
 657             fjptrE             = f+j_coord_offsetE;
 658             fjptrF             = f+j_coord_offsetF;
 659             fjptrG             = f+j_coord_offsetG;
 660             fjptrH             = f+j_coord_offsetH;
 661
 662             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 663                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 664                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 665
 666             /* Inner loop uses 347 flops */
 667         }
 668
 669         if(jidx<j_index_end)
 670         {
 671
 672             /* Get j neighbor index, and coordinate index */
 673             jnrlistA         = jjnr[jidx];
 674             jnrlistB         = jjnr[jidx+1];
 675             jnrlistC         = jjnr[jidx+2];
 676             jnrlistD         = jjnr[jidx+3];
 677             jnrlistE         = jjnr[jidx+4];
 678             jnrlistF         = jjnr[jidx+5];
 679             jnrlistG         = jjnr[jidx+6];
 680             jnrlistH         = jjnr[jidx+7];
 681             /* Sign of each element will be negative for non-real atoms.
 682              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 683              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 684              */
 685             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 686                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 687
 688             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 689             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 690             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 691             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 692             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 693             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 694             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 695             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 696             j_coord_offsetA  = DIM*jnrA;
 697             j_coord_offsetB  = DIM*jnrB;
 698             j_coord_offsetC  = DIM*jnrC;
 699             j_coord_offsetD  = DIM*jnrD;
 700             j_coord_offsetE  = DIM*jnrE;
 701             j_coord_offsetF  = DIM*jnrF;
 702             j_coord_offsetG  = DIM*jnrG;
 703             j_coord_offsetH  = DIM*jnrH;
 704
 705             /* load j atom coordinates */
 706             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 707                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 708                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 709                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 710                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 711                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 712
 713             /* Calculate displacement vector */
 714             dx00             = _mm256_sub_ps(ix0,jx0);
 715             dy00             = _mm256_sub_ps(iy0,jy0);
 716             dz00             = _mm256_sub_ps(iz0,jz0);
 717             dx11             = _mm256_sub_ps(ix1,jx1);
 718             dy11             = _mm256_sub_ps(iy1,jy1);
 719             dz11             = _mm256_sub_ps(iz1,jz1);
 720             dx12             = _mm256_sub_ps(ix1,jx2);
 721             dy12             = _mm256_sub_ps(iy1,jy2);
 722             dz12             = _mm256_sub_ps(iz1,jz2);
 723             dx13             = _mm256_sub_ps(ix1,jx3);
 724             dy13             = _mm256_sub_ps(iy1,jy3);
 725             dz13             = _mm256_sub_ps(iz1,jz3);
 726             dx21             = _mm256_sub_ps(ix2,jx1);
 727             dy21             = _mm256_sub_ps(iy2,jy1);
 728             dz21             = _mm256_sub_ps(iz2,jz1);
 729             dx22             = _mm256_sub_ps(ix2,jx2);
 730             dy22             = _mm256_sub_ps(iy2,jy2);
 731             dz22             = _mm256_sub_ps(iz2,jz2);
 732             dx23             = _mm256_sub_ps(ix2,jx3);
 733             dy23             = _mm256_sub_ps(iy2,jy3);
 734             dz23             = _mm256_sub_ps(iz2,jz3);
 735             dx31             = _mm256_sub_ps(ix3,jx1);
 736             dy31             = _mm256_sub_ps(iy3,jy1);
 737             dz31             = _mm256_sub_ps(iz3,jz1);
 738             dx32             = _mm256_sub_ps(ix3,jx2);
 739             dy32             = _mm256_sub_ps(iy3,jy2);
 740             dz32             = _mm256_sub_ps(iz3,jz2);
 741             dx33             = _mm256_sub_ps(ix3,jx3);
 742             dy33             = _mm256_sub_ps(iy3,jy3);
 743             dz33             = _mm256_sub_ps(iz3,jz3);
 744
 745             /* Calculate squared distance and things based on it */
 746             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 747             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 748             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 749             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 750             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 751             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 752             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 753             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 754             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 755             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 756
 757             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 758             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 759             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 760             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 761             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 762             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 763             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 764             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 765             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 766             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 767
 768             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 769             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 770             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 771             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 772             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 773             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 774             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 775             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 776             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 777
 778             fjx0             = _mm256_setzero_ps();
 779             fjy0             = _mm256_setzero_ps();
 780             fjz0             = _mm256_setzero_ps();
 781             fjx1             = _mm256_setzero_ps();
 782             fjy1             = _mm256_setzero_ps();
 783             fjz1             = _mm256_setzero_ps();
 784             fjx2             = _mm256_setzero_ps();
 785             fjy2             = _mm256_setzero_ps();
 786             fjz2             = _mm256_setzero_ps();
 787             fjx3             = _mm256_setzero_ps();
 788             fjy3             = _mm256_setzero_ps();
 789             fjz3             = _mm256_setzero_ps();
 790
 791             /**************************
 792              * CALCULATE INTERACTIONS *
 793              **************************/
 794
 795             r00              = _mm256_mul_ps(rsq00,rinv00);
 796             r00              = _mm256_andnot_ps(dummy_mask,r00);
 797
 798             /* Calculate table index by multiplying r with table scale and truncate to integer */
 799             rt               = _mm256_mul_ps(r00,vftabscale);
 800             vfitab           = _mm256_cvttps_epi32(rt);
 801             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 802             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 803             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 804             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 805             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 806             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 807
 808             /* CUBIC SPLINE TABLE DISPERSION */
 809             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 810                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 811             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 812                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 813             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 814                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 815             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 816                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 817             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 818             Heps             = _mm256_mul_ps(vfeps,H);
 819             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 820             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 821             vvdw6            = _mm256_mul_ps(c6_00,VV);
 822             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 823             fvdw6            = _mm256_mul_ps(c6_00,FF);
 824
 825             /* CUBIC SPLINE TABLE REPULSION */
 826             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 827             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 828             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 829                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 830             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 831                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 832             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 833                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 834             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 835                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 836             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 837             Heps             = _mm256_mul_ps(vfeps,H);
 838             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 839             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 840             vvdw12           = _mm256_mul_ps(c12_00,VV);
 841             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 842             fvdw12           = _mm256_mul_ps(c12_00,FF);
 843             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 844             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 845
 846             /* Update potential sum for this i atom from the interaction with this j atom. */
 847             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 848             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 849
 850             fscal            = fvdw;
 851
 852             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 853
 854             /* Calculate temporary vectorial force */
 855             tx               = _mm256_mul_ps(fscal,dx00);
 856             ty               = _mm256_mul_ps(fscal,dy00);
 857             tz               = _mm256_mul_ps(fscal,dz00);
 858
 859             /* Update vectorial force */
 860             fix0             = _mm256_add_ps(fix0,tx);
 861             fiy0             = _mm256_add_ps(fiy0,ty);
 862             fiz0             = _mm256_add_ps(fiz0,tz);
 863
 864             fjx0             = _mm256_add_ps(fjx0,tx);
 865             fjy0             = _mm256_add_ps(fjy0,ty);
 866             fjz0             = _mm256_add_ps(fjz0,tz);
 867
 868             /**************************
 869              * CALCULATE INTERACTIONS *
 870              **************************/
 871
 872             /* REACTION-FIELD ELECTROSTATICS */
 873             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 874             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 875
 876             /* Update potential sum for this i atom from the interaction with this j atom. */
 877             velec            = _mm256_andnot_ps(dummy_mask,velec);
 878             velecsum         = _mm256_add_ps(velecsum,velec);
 879
 880             fscal            = felec;
 881
 882             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 883
 884             /* Calculate temporary vectorial force */
 885             tx               = _mm256_mul_ps(fscal,dx11);
 886             ty               = _mm256_mul_ps(fscal,dy11);
 887             tz               = _mm256_mul_ps(fscal,dz11);
 888
 889             /* Update vectorial force */
 890             fix1             = _mm256_add_ps(fix1,tx);
 891             fiy1             = _mm256_add_ps(fiy1,ty);
 892             fiz1             = _mm256_add_ps(fiz1,tz);
 893
 894             fjx1             = _mm256_add_ps(fjx1,tx);
 895             fjy1             = _mm256_add_ps(fjy1,ty);
 896             fjz1             = _mm256_add_ps(fjz1,tz);
 897
 898             /**************************
 899              * CALCULATE INTERACTIONS *
 900              **************************/
 901
 902             /* REACTION-FIELD ELECTROSTATICS */
 903             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 904             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 905
 906             /* Update potential sum for this i atom from the interaction with this j atom. */
 907             velec            = _mm256_andnot_ps(dummy_mask,velec);
 908             velecsum         = _mm256_add_ps(velecsum,velec);
 909
 910             fscal            = felec;
 911
 912             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 913
 914             /* Calculate temporary vectorial force */
 915             tx               = _mm256_mul_ps(fscal,dx12);
 916             ty               = _mm256_mul_ps(fscal,dy12);
 917             tz               = _mm256_mul_ps(fscal,dz12);
 918
 919             /* Update vectorial force */
 920             fix1             = _mm256_add_ps(fix1,tx);
 921             fiy1             = _mm256_add_ps(fiy1,ty);
 922             fiz1             = _mm256_add_ps(fiz1,tz);
 923
 924             fjx2             = _mm256_add_ps(fjx2,tx);
 925             fjy2             = _mm256_add_ps(fjy2,ty);
 926             fjz2             = _mm256_add_ps(fjz2,tz);
 927
 928             /**************************
 929              * CALCULATE INTERACTIONS *
 930              **************************/
 931
 932             /* REACTION-FIELD ELECTROSTATICS */
 933             velec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_add_ps(rinv13,_mm256_mul_ps(krf,rsq13)),crf));
 934             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
 935
 936             /* Update potential sum for this i atom from the interaction with this j atom. */
 937             velec            = _mm256_andnot_ps(dummy_mask,velec);
 938             velecsum         = _mm256_add_ps(velecsum,velec);
 939
 940             fscal            = felec;
 941
 942             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 943
 944             /* Calculate temporary vectorial force */
 945             tx               = _mm256_mul_ps(fscal,dx13);
 946             ty               = _mm256_mul_ps(fscal,dy13);
 947             tz               = _mm256_mul_ps(fscal,dz13);
 948
 949             /* Update vectorial force */
 950             fix1             = _mm256_add_ps(fix1,tx);
 951             fiy1             = _mm256_add_ps(fiy1,ty);
 952             fiz1             = _mm256_add_ps(fiz1,tz);
 953
 954             fjx3             = _mm256_add_ps(fjx3,tx);
 955             fjy3             = _mm256_add_ps(fjy3,ty);
 956             fjz3             = _mm256_add_ps(fjz3,tz);
 957
 958             /**************************
 959              * CALCULATE INTERACTIONS *
 960              **************************/
 961
 962             /* REACTION-FIELD ELECTROSTATICS */
 963             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
 964             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
 965
 966             /* Update potential sum for this i atom from the interaction with this j atom. */
 967             velec            = _mm256_andnot_ps(dummy_mask,velec);
 968             velecsum         = _mm256_add_ps(velecsum,velec);
 969
 970             fscal            = felec;
 971
 972             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 973
 974             /* Calculate temporary vectorial force */
 975             tx               = _mm256_mul_ps(fscal,dx21);
 976             ty               = _mm256_mul_ps(fscal,dy21);
 977             tz               = _mm256_mul_ps(fscal,dz21);
 978
 979             /* Update vectorial force */
 980             fix2             = _mm256_add_ps(fix2,tx);
 981             fiy2             = _mm256_add_ps(fiy2,ty);
 982             fiz2             = _mm256_add_ps(fiz2,tz);
 983
 984             fjx1             = _mm256_add_ps(fjx1,tx);
 985             fjy1             = _mm256_add_ps(fjy1,ty);
 986             fjz1             = _mm256_add_ps(fjz1,tz);
 987
 988             /**************************
 989              * CALCULATE INTERACTIONS *
 990              **************************/
 991
 992             /* REACTION-FIELD ELECTROSTATICS */
 993             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
 994             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
 995
 996             /* Update potential sum for this i atom from the interaction with this j atom. */
 997             velec            = _mm256_andnot_ps(dummy_mask,velec);
 998             velecsum         = _mm256_add_ps(velecsum,velec);
 999
1000             fscal            = felec;
1001
1002             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1003
1004             /* Calculate temporary vectorial force */
1005             tx               = _mm256_mul_ps(fscal,dx22);
1006             ty               = _mm256_mul_ps(fscal,dy22);
1007             tz               = _mm256_mul_ps(fscal,dz22);
1008
1009             /* Update vectorial force */
1010             fix2             = _mm256_add_ps(fix2,tx);
1011             fiy2             = _mm256_add_ps(fiy2,ty);
1012             fiz2             = _mm256_add_ps(fiz2,tz);
1013
1014             fjx2             = _mm256_add_ps(fjx2,tx);
1015             fjy2             = _mm256_add_ps(fjy2,ty);
1016             fjz2             = _mm256_add_ps(fjz2,tz);
1017
1018             /**************************
1019              * CALCULATE INTERACTIONS *
1020              **************************/
1021
1022             /* REACTION-FIELD ELECTROSTATICS */
1023             velec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_add_ps(rinv23,_mm256_mul_ps(krf,rsq23)),crf));
1024             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
1025
1026             /* Update potential sum for this i atom from the interaction with this j atom. */
1027             velec            = _mm256_andnot_ps(dummy_mask,velec);
1028             velecsum         = _mm256_add_ps(velecsum,velec);
1029
1030             fscal            = felec;
1031
1032             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1033
1034             /* Calculate temporary vectorial force */
1035             tx               = _mm256_mul_ps(fscal,dx23);
1036             ty               = _mm256_mul_ps(fscal,dy23);
1037             tz               = _mm256_mul_ps(fscal,dz23);
1038
1039             /* Update vectorial force */
1040             fix2             = _mm256_add_ps(fix2,tx);
1041             fiy2             = _mm256_add_ps(fiy2,ty);
1042             fiz2             = _mm256_add_ps(fiz2,tz);
1043
1044             fjx3             = _mm256_add_ps(fjx3,tx);
1045             fjy3             = _mm256_add_ps(fjy3,ty);
1046             fjz3             = _mm256_add_ps(fjz3,tz);
1047
1048             /**************************
1049              * CALCULATE INTERACTIONS *
1050              **************************/
1051
1052             /* REACTION-FIELD ELECTROSTATICS */
1053             velec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_add_ps(rinv31,_mm256_mul_ps(krf,rsq31)),crf));
1054             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
1055
1056             /* Update potential sum for this i atom from the interaction with this j atom. */
1057             velec            = _mm256_andnot_ps(dummy_mask,velec);
1058             velecsum         = _mm256_add_ps(velecsum,velec);
1059
1060             fscal            = felec;
1061
1062             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1063
1064             /* Calculate temporary vectorial force */
1065             tx               = _mm256_mul_ps(fscal,dx31);
1066             ty               = _mm256_mul_ps(fscal,dy31);
1067             tz               = _mm256_mul_ps(fscal,dz31);
1068
1069             /* Update vectorial force */
1070             fix3             = _mm256_add_ps(fix3,tx);
1071             fiy3             = _mm256_add_ps(fiy3,ty);
1072             fiz3             = _mm256_add_ps(fiz3,tz);
1073
1074             fjx1             = _mm256_add_ps(fjx1,tx);
1075             fjy1             = _mm256_add_ps(fjy1,ty);
1076             fjz1             = _mm256_add_ps(fjz1,tz);
1077
1078             /**************************
1079              * CALCULATE INTERACTIONS *
1080              **************************/
1081
1082             /* REACTION-FIELD ELECTROSTATICS */
1083             velec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_add_ps(rinv32,_mm256_mul_ps(krf,rsq32)),crf));
1084             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
1085
1086             /* Update potential sum for this i atom from the interaction with this j atom. */
1087             velec            = _mm256_andnot_ps(dummy_mask,velec);
1088             velecsum         = _mm256_add_ps(velecsum,velec);
1089
1090             fscal            = felec;
1091
1092             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1093
1094             /* Calculate temporary vectorial force */
1095             tx               = _mm256_mul_ps(fscal,dx32);
1096             ty               = _mm256_mul_ps(fscal,dy32);
1097             tz               = _mm256_mul_ps(fscal,dz32);
1098
1099             /* Update vectorial force */
1100             fix3             = _mm256_add_ps(fix3,tx);
1101             fiy3             = _mm256_add_ps(fiy3,ty);
1102             fiz3             = _mm256_add_ps(fiz3,tz);
1103
1104             fjx2             = _mm256_add_ps(fjx2,tx);
1105             fjy2             = _mm256_add_ps(fjy2,ty);
1106             fjz2             = _mm256_add_ps(fjz2,tz);
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             /* REACTION-FIELD ELECTROSTATICS */
1113             velec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_add_ps(rinv33,_mm256_mul_ps(krf,rsq33)),crf));
1114             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
1115
1116             /* Update potential sum for this i atom from the interaction with this j atom. */
1117             velec            = _mm256_andnot_ps(dummy_mask,velec);
1118             velecsum         = _mm256_add_ps(velecsum,velec);
1119
1120             fscal            = felec;
1121
1122             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1123
1124             /* Calculate temporary vectorial force */
1125             tx               = _mm256_mul_ps(fscal,dx33);
1126             ty               = _mm256_mul_ps(fscal,dy33);
1127             tz               = _mm256_mul_ps(fscal,dz33);
1128
1129             /* Update vectorial force */
1130             fix3             = _mm256_add_ps(fix3,tx);
1131             fiy3             = _mm256_add_ps(fiy3,ty);
1132             fiz3             = _mm256_add_ps(fiz3,tz);
1133
1134             fjx3             = _mm256_add_ps(fjx3,tx);
1135             fjy3             = _mm256_add_ps(fjy3,ty);
1136             fjz3             = _mm256_add_ps(fjz3,tz);
1137
1138             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1139             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1140             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1141             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1142             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1143             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1144             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1145             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1146
1147             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1148                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1149                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1150
1151             /* Inner loop uses 348 flops */
1152         }
1153
1154         /* End of innermost loop */
1155
1156         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1157                                                  f+i_coord_offset,fshift+i_shift_offset);
1158
1159         ggid                        = gid[iidx];
1160         /* Update potential energies */
1161         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1162         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1163
1164         /* Increment number of inner iterations */
1165         inneriter                  += j_index_end - j_index_start;
1166
1167         /* Outer loop uses 26 flops */
1168     }
1169
1170     /* Increment number of outer iterations */
1171     outeriter        += nri;
1172
1173     /* Update outer/inner flops */
1174
1175     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*348);
1176 }
1177 /*
1178  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4W4_F_avx_256_single
1179  * Electrostatics interaction: ReactionField
1180  * VdW interaction:            CubicSplineTable
1181  * Geometry:                   Water4-Water4
1182  * Calculate force/pot:        Force
1183  */
1184 void
1185 nb_kernel_ElecRF_VdwCSTab_GeomW4W4_F_avx_256_single
1186                     (t_nblist                    * gmx_restrict       nlist,
1187                      rvec                        * gmx_restrict          xx,
1188                      rvec                        * gmx_restrict          ff,
1189                      t_forcerec                  * gmx_restrict          fr,
1190                      t_mdatoms                   * gmx_restrict     mdatoms,
1191                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1192                      t_nrnb                      * gmx_restrict        nrnb)
1193 {
1194     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1195      * just 0 for non-waters.
1196      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1197      * jnr indices corresponding to data put in the four positions in the SIMD register.
1198      */
1199     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1200     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1201     int              jnrA,jnrB,jnrC,jnrD;
1202     int              jnrE,jnrF,jnrG,jnrH;
1203     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1204     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1205     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1206     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1207     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1208     real             rcutoff_scalar;
1209     real             *shiftvec,*fshift,*x,*f;
1210     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1211     real             scratch[4*DIM];
1212     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1213     real *           vdwioffsetptr0;
1214     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1215     real *           vdwioffsetptr1;
1216     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1217     real *           vdwioffsetptr2;
1218     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1219     real *           vdwioffsetptr3;
1220     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1221     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1222     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1223     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1224     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1225     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1226     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1227     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1228     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1229     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1230     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1231     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1232     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1233     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1234     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1235     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1236     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1237     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1238     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1239     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1240     real             *charge;
1241     int              nvdwtype;
1242     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1243     int              *vdwtype;
1244     real             *vdwparam;
1245     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1246     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1247     __m256i          vfitab;
1248     __m128i          vfitab_lo,vfitab_hi;
1249     __m128i          ifour       = _mm_set1_epi32(4);
1250     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1251     real             *vftab;
1252     __m256           dummy_mask,cutoff_mask;
1253     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1254     __m256           one     = _mm256_set1_ps(1.0);
1255     __m256           two     = _mm256_set1_ps(2.0);
1256     x                = xx[0];
1257     f                = ff[0];
1258
1259     nri              = nlist->nri;
1260     iinr             = nlist->iinr;
1261     jindex           = nlist->jindex;
1262     jjnr             = nlist->jjnr;
1263     shiftidx         = nlist->shift;
1264     gid              = nlist->gid;
1265     shiftvec         = fr->shift_vec[0];
1266     fshift           = fr->fshift[0];
1267     facel            = _mm256_set1_ps(fr->epsfac);
1268     charge           = mdatoms->chargeA;
1269     krf              = _mm256_set1_ps(fr->ic->k_rf);
1270     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
1271     crf              = _mm256_set1_ps(fr->ic->c_rf);
1272     nvdwtype         = fr->ntype;
1273     vdwparam         = fr->nbfp;
1274     vdwtype          = mdatoms->typeA;
1275
1276     vftab            = kernel_data->table_vdw->data;
1277     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
1278
1279     /* Setup water-specific parameters */
1280     inr              = nlist->iinr[0];
1281     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1282     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1283     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1284     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1285
1286     jq1              = _mm256_set1_ps(charge[inr+1]);
1287     jq2              = _mm256_set1_ps(charge[inr+2]);
1288     jq3              = _mm256_set1_ps(charge[inr+3]);
1289     vdwjidx0A        = 2*vdwtype[inr+0];
1290     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1291     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1292     qq11             = _mm256_mul_ps(iq1,jq1);
1293     qq12             = _mm256_mul_ps(iq1,jq2);
1294     qq13             = _mm256_mul_ps(iq1,jq3);
1295     qq21             = _mm256_mul_ps(iq2,jq1);
1296     qq22             = _mm256_mul_ps(iq2,jq2);
1297     qq23             = _mm256_mul_ps(iq2,jq3);
1298     qq31             = _mm256_mul_ps(iq3,jq1);
1299     qq32             = _mm256_mul_ps(iq3,jq2);
1300     qq33             = _mm256_mul_ps(iq3,jq3);
1301
1302     /* Avoid stupid compiler warnings */
1303     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1304     j_coord_offsetA = 0;
1305     j_coord_offsetB = 0;
1306     j_coord_offsetC = 0;
1307     j_coord_offsetD = 0;
1308     j_coord_offsetE = 0;
1309     j_coord_offsetF = 0;
1310     j_coord_offsetG = 0;
1311     j_coord_offsetH = 0;
1312
1313     outeriter        = 0;
1314     inneriter        = 0;
1315
1316     for(iidx=0;iidx<4*DIM;iidx++)
1317     {
1318         scratch[iidx] = 0.0;
1319     }
1320
1321     /* Start outer loop over neighborlists */
1322     for(iidx=0; iidx<nri; iidx++)
1323     {
1324         /* Load shift vector for this list */
1325         i_shift_offset   = DIM*shiftidx[iidx];
1326
1327         /* Load limits for loop over neighbors */
1328         j_index_start    = jindex[iidx];
1329         j_index_end      = jindex[iidx+1];
1330
1331         /* Get outer coordinate index */
1332         inr              = iinr[iidx];
1333         i_coord_offset   = DIM*inr;
1334
1335         /* Load i particle coords and add shift vector */
1336         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1337                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1338
1339         fix0             = _mm256_setzero_ps();
1340         fiy0             = _mm256_setzero_ps();
1341         fiz0             = _mm256_setzero_ps();
1342         fix1             = _mm256_setzero_ps();
1343         fiy1             = _mm256_setzero_ps();
1344         fiz1             = _mm256_setzero_ps();
1345         fix2             = _mm256_setzero_ps();
1346         fiy2             = _mm256_setzero_ps();
1347         fiz2             = _mm256_setzero_ps();
1348         fix3             = _mm256_setzero_ps();
1349         fiy3             = _mm256_setzero_ps();
1350         fiz3             = _mm256_setzero_ps();
1351
1352         /* Start inner kernel loop */
1353         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1354         {
1355
1356             /* Get j neighbor index, and coordinate index */
1357             jnrA             = jjnr[jidx];
1358             jnrB             = jjnr[jidx+1];
1359             jnrC             = jjnr[jidx+2];
1360             jnrD             = jjnr[jidx+3];
1361             jnrE             = jjnr[jidx+4];
1362             jnrF             = jjnr[jidx+5];
1363             jnrG             = jjnr[jidx+6];
1364             jnrH             = jjnr[jidx+7];
1365             j_coord_offsetA  = DIM*jnrA;
1366             j_coord_offsetB  = DIM*jnrB;
1367             j_coord_offsetC  = DIM*jnrC;
1368             j_coord_offsetD  = DIM*jnrD;
1369             j_coord_offsetE  = DIM*jnrE;
1370             j_coord_offsetF  = DIM*jnrF;
1371             j_coord_offsetG  = DIM*jnrG;
1372             j_coord_offsetH  = DIM*jnrH;
1373
1374             /* load j atom coordinates */
1375             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1376                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1377                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1378                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1379                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1380                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1381
1382             /* Calculate displacement vector */
1383             dx00             = _mm256_sub_ps(ix0,jx0);
1384             dy00             = _mm256_sub_ps(iy0,jy0);
1385             dz00             = _mm256_sub_ps(iz0,jz0);
1386             dx11             = _mm256_sub_ps(ix1,jx1);
1387             dy11             = _mm256_sub_ps(iy1,jy1);
1388             dz11             = _mm256_sub_ps(iz1,jz1);
1389             dx12             = _mm256_sub_ps(ix1,jx2);
1390             dy12             = _mm256_sub_ps(iy1,jy2);
1391             dz12             = _mm256_sub_ps(iz1,jz2);
1392             dx13             = _mm256_sub_ps(ix1,jx3);
1393             dy13             = _mm256_sub_ps(iy1,jy3);
1394             dz13             = _mm256_sub_ps(iz1,jz3);
1395             dx21             = _mm256_sub_ps(ix2,jx1);
1396             dy21             = _mm256_sub_ps(iy2,jy1);
1397             dz21             = _mm256_sub_ps(iz2,jz1);
1398             dx22             = _mm256_sub_ps(ix2,jx2);
1399             dy22             = _mm256_sub_ps(iy2,jy2);
1400             dz22             = _mm256_sub_ps(iz2,jz2);
1401             dx23             = _mm256_sub_ps(ix2,jx3);
1402             dy23             = _mm256_sub_ps(iy2,jy3);
1403             dz23             = _mm256_sub_ps(iz2,jz3);
1404             dx31             = _mm256_sub_ps(ix3,jx1);
1405             dy31             = _mm256_sub_ps(iy3,jy1);
1406             dz31             = _mm256_sub_ps(iz3,jz1);
1407             dx32             = _mm256_sub_ps(ix3,jx2);
1408             dy32             = _mm256_sub_ps(iy3,jy2);
1409             dz32             = _mm256_sub_ps(iz3,jz2);
1410             dx33             = _mm256_sub_ps(ix3,jx3);
1411             dy33             = _mm256_sub_ps(iy3,jy3);
1412             dz33             = _mm256_sub_ps(iz3,jz3);
1413
1414             /* Calculate squared distance and things based on it */
1415             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1416             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1417             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1418             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1419             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1420             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1421             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1422             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1423             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1424             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1425
1426             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1427             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1428             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1429             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
1430             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1431             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1432             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
1433             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
1434             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
1435             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
1436
1437             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1438             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1439             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1440             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1441             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1442             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1443             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1444             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1445             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1446
1447             fjx0             = _mm256_setzero_ps();
1448             fjy0             = _mm256_setzero_ps();
1449             fjz0             = _mm256_setzero_ps();
1450             fjx1             = _mm256_setzero_ps();
1451             fjy1             = _mm256_setzero_ps();
1452             fjz1             = _mm256_setzero_ps();
1453             fjx2             = _mm256_setzero_ps();
1454             fjy2             = _mm256_setzero_ps();
1455             fjz2             = _mm256_setzero_ps();
1456             fjx3             = _mm256_setzero_ps();
1457             fjy3             = _mm256_setzero_ps();
1458             fjz3             = _mm256_setzero_ps();
1459
1460             /**************************
1461              * CALCULATE INTERACTIONS *
1462              **************************/
1463
1464             r00              = _mm256_mul_ps(rsq00,rinv00);
1465
1466             /* Calculate table index by multiplying r with table scale and truncate to integer */
1467             rt               = _mm256_mul_ps(r00,vftabscale);
1468             vfitab           = _mm256_cvttps_epi32(rt);
1469             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1470             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1471             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1472             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1473             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1474             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1475
1476             /* CUBIC SPLINE TABLE DISPERSION */
1477             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1478                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1479             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1480                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1481             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1482                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1483             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1484                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1485             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1486             Heps             = _mm256_mul_ps(vfeps,H);
1487             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1488             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1489             fvdw6            = _mm256_mul_ps(c6_00,FF);
1490
1491             /* CUBIC SPLINE TABLE REPULSION */
1492             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1493             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1494             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1495                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1496             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1497                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1498             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1499                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1500             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1501                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1502             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1503             Heps             = _mm256_mul_ps(vfeps,H);
1504             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1505             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1506             fvdw12           = _mm256_mul_ps(c12_00,FF);
1507             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1508
1509             fscal            = fvdw;
1510
1511             /* Calculate temporary vectorial force */
1512             tx               = _mm256_mul_ps(fscal,dx00);
1513             ty               = _mm256_mul_ps(fscal,dy00);
1514             tz               = _mm256_mul_ps(fscal,dz00);
1515
1516             /* Update vectorial force */
1517             fix0             = _mm256_add_ps(fix0,tx);
1518             fiy0             = _mm256_add_ps(fiy0,ty);
1519             fiz0             = _mm256_add_ps(fiz0,tz);
1520
1521             fjx0             = _mm256_add_ps(fjx0,tx);
1522             fjy0             = _mm256_add_ps(fjy0,ty);
1523             fjz0             = _mm256_add_ps(fjz0,tz);
1524
1525             /**************************
1526              * CALCULATE INTERACTIONS *
1527              **************************/
1528
1529             /* REACTION-FIELD ELECTROSTATICS */
1530             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1531
1532             fscal            = felec;
1533
1534             /* Calculate temporary vectorial force */
1535             tx               = _mm256_mul_ps(fscal,dx11);
1536             ty               = _mm256_mul_ps(fscal,dy11);
1537             tz               = _mm256_mul_ps(fscal,dz11);
1538
1539             /* Update vectorial force */
1540             fix1             = _mm256_add_ps(fix1,tx);
1541             fiy1             = _mm256_add_ps(fiy1,ty);
1542             fiz1             = _mm256_add_ps(fiz1,tz);
1543
1544             fjx1             = _mm256_add_ps(fjx1,tx);
1545             fjy1             = _mm256_add_ps(fjy1,ty);
1546             fjz1             = _mm256_add_ps(fjz1,tz);
1547
1548             /**************************
1549              * CALCULATE INTERACTIONS *
1550              **************************/
1551
1552             /* REACTION-FIELD ELECTROSTATICS */
1553             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1554
1555             fscal            = felec;
1556
1557             /* Calculate temporary vectorial force */
1558             tx               = _mm256_mul_ps(fscal,dx12);
1559             ty               = _mm256_mul_ps(fscal,dy12);
1560             tz               = _mm256_mul_ps(fscal,dz12);
1561
1562             /* Update vectorial force */
1563             fix1             = _mm256_add_ps(fix1,tx);
1564             fiy1             = _mm256_add_ps(fiy1,ty);
1565             fiz1             = _mm256_add_ps(fiz1,tz);
1566
1567             fjx2             = _mm256_add_ps(fjx2,tx);
1568             fjy2             = _mm256_add_ps(fjy2,ty);
1569             fjz2             = _mm256_add_ps(fjz2,tz);
1570
1571             /**************************
1572              * CALCULATE INTERACTIONS *
1573              **************************/
1574
1575             /* REACTION-FIELD ELECTROSTATICS */
1576             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
1577
1578             fscal            = felec;
1579
1580             /* Calculate temporary vectorial force */
1581             tx               = _mm256_mul_ps(fscal,dx13);
1582             ty               = _mm256_mul_ps(fscal,dy13);
1583             tz               = _mm256_mul_ps(fscal,dz13);
1584
1585             /* Update vectorial force */
1586             fix1             = _mm256_add_ps(fix1,tx);
1587             fiy1             = _mm256_add_ps(fiy1,ty);
1588             fiz1             = _mm256_add_ps(fiz1,tz);
1589
1590             fjx3             = _mm256_add_ps(fjx3,tx);
1591             fjy3             = _mm256_add_ps(fjy3,ty);
1592             fjz3             = _mm256_add_ps(fjz3,tz);
1593
1594             /**************************
1595              * CALCULATE INTERACTIONS *
1596              **************************/
1597
1598             /* REACTION-FIELD ELECTROSTATICS */
1599             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1600
1601             fscal            = felec;
1602
1603             /* Calculate temporary vectorial force */
1604             tx               = _mm256_mul_ps(fscal,dx21);
1605             ty               = _mm256_mul_ps(fscal,dy21);
1606             tz               = _mm256_mul_ps(fscal,dz21);
1607
1608             /* Update vectorial force */
1609             fix2             = _mm256_add_ps(fix2,tx);
1610             fiy2             = _mm256_add_ps(fiy2,ty);
1611             fiz2             = _mm256_add_ps(fiz2,tz);
1612
1613             fjx1             = _mm256_add_ps(fjx1,tx);
1614             fjy1             = _mm256_add_ps(fjy1,ty);
1615             fjz1             = _mm256_add_ps(fjz1,tz);
1616
1617             /**************************
1618              * CALCULATE INTERACTIONS *
1619              **************************/
1620
1621             /* REACTION-FIELD ELECTROSTATICS */
1622             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1623
1624             fscal            = felec;
1625
1626             /* Calculate temporary vectorial force */
1627             tx               = _mm256_mul_ps(fscal,dx22);
1628             ty               = _mm256_mul_ps(fscal,dy22);
1629             tz               = _mm256_mul_ps(fscal,dz22);
1630
1631             /* Update vectorial force */
1632             fix2             = _mm256_add_ps(fix2,tx);
1633             fiy2             = _mm256_add_ps(fiy2,ty);
1634             fiz2             = _mm256_add_ps(fiz2,tz);
1635
1636             fjx2             = _mm256_add_ps(fjx2,tx);
1637             fjy2             = _mm256_add_ps(fjy2,ty);
1638             fjz2             = _mm256_add_ps(fjz2,tz);
1639
1640             /**************************
1641              * CALCULATE INTERACTIONS *
1642              **************************/
1643
1644             /* REACTION-FIELD ELECTROSTATICS */
1645             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
1646
1647             fscal            = felec;
1648
1649             /* Calculate temporary vectorial force */
1650             tx               = _mm256_mul_ps(fscal,dx23);
1651             ty               = _mm256_mul_ps(fscal,dy23);
1652             tz               = _mm256_mul_ps(fscal,dz23);
1653
1654             /* Update vectorial force */
1655             fix2             = _mm256_add_ps(fix2,tx);
1656             fiy2             = _mm256_add_ps(fiy2,ty);
1657             fiz2             = _mm256_add_ps(fiz2,tz);
1658
1659             fjx3             = _mm256_add_ps(fjx3,tx);
1660             fjy3             = _mm256_add_ps(fjy3,ty);
1661             fjz3             = _mm256_add_ps(fjz3,tz);
1662
1663             /**************************
1664              * CALCULATE INTERACTIONS *
1665              **************************/
1666
1667             /* REACTION-FIELD ELECTROSTATICS */
1668             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
1669
1670             fscal            = felec;
1671
1672             /* Calculate temporary vectorial force */
1673             tx               = _mm256_mul_ps(fscal,dx31);
1674             ty               = _mm256_mul_ps(fscal,dy31);
1675             tz               = _mm256_mul_ps(fscal,dz31);
1676
1677             /* Update vectorial force */
1678             fix3             = _mm256_add_ps(fix3,tx);
1679             fiy3             = _mm256_add_ps(fiy3,ty);
1680             fiz3             = _mm256_add_ps(fiz3,tz);
1681
1682             fjx1             = _mm256_add_ps(fjx1,tx);
1683             fjy1             = _mm256_add_ps(fjy1,ty);
1684             fjz1             = _mm256_add_ps(fjz1,tz);
1685
1686             /**************************
1687              * CALCULATE INTERACTIONS *
1688              **************************/
1689
1690             /* REACTION-FIELD ELECTROSTATICS */
1691             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
1692
1693             fscal            = felec;
1694
1695             /* Calculate temporary vectorial force */
1696             tx               = _mm256_mul_ps(fscal,dx32);
1697             ty               = _mm256_mul_ps(fscal,dy32);
1698             tz               = _mm256_mul_ps(fscal,dz32);
1699
1700             /* Update vectorial force */
1701             fix3             = _mm256_add_ps(fix3,tx);
1702             fiy3             = _mm256_add_ps(fiy3,ty);
1703             fiz3             = _mm256_add_ps(fiz3,tz);
1704
1705             fjx2             = _mm256_add_ps(fjx2,tx);
1706             fjy2             = _mm256_add_ps(fjy2,ty);
1707             fjz2             = _mm256_add_ps(fjz2,tz);
1708
1709             /**************************
1710              * CALCULATE INTERACTIONS *
1711              **************************/
1712
1713             /* REACTION-FIELD ELECTROSTATICS */
1714             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
1715
1716             fscal            = felec;
1717
1718             /* Calculate temporary vectorial force */
1719             tx               = _mm256_mul_ps(fscal,dx33);
1720             ty               = _mm256_mul_ps(fscal,dy33);
1721             tz               = _mm256_mul_ps(fscal,dz33);
1722
1723             /* Update vectorial force */
1724             fix3             = _mm256_add_ps(fix3,tx);
1725             fiy3             = _mm256_add_ps(fiy3,ty);
1726             fiz3             = _mm256_add_ps(fiz3,tz);
1727
1728             fjx3             = _mm256_add_ps(fjx3,tx);
1729             fjy3             = _mm256_add_ps(fjy3,ty);
1730             fjz3             = _mm256_add_ps(fjz3,tz);
1731
1732             fjptrA             = f+j_coord_offsetA;
1733             fjptrB             = f+j_coord_offsetB;
1734             fjptrC             = f+j_coord_offsetC;
1735             fjptrD             = f+j_coord_offsetD;
1736             fjptrE             = f+j_coord_offsetE;
1737             fjptrF             = f+j_coord_offsetF;
1738             fjptrG             = f+j_coord_offsetG;
1739             fjptrH             = f+j_coord_offsetH;
1740
1741             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1742                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1743                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1744
1745             /* Inner loop uses 294 flops */
1746         }
1747
1748         if(jidx<j_index_end)
1749         {
1750
1751             /* Get j neighbor index, and coordinate index */
1752             jnrlistA         = jjnr[jidx];
1753             jnrlistB         = jjnr[jidx+1];
1754             jnrlistC         = jjnr[jidx+2];
1755             jnrlistD         = jjnr[jidx+3];
1756             jnrlistE         = jjnr[jidx+4];
1757             jnrlistF         = jjnr[jidx+5];
1758             jnrlistG         = jjnr[jidx+6];
1759             jnrlistH         = jjnr[jidx+7];
1760             /* Sign of each element will be negative for non-real atoms.
1761              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1762              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1763              */
1764             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1765                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1766
1767             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1768             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1769             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1770             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1771             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1772             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1773             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1774             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1775             j_coord_offsetA  = DIM*jnrA;
1776             j_coord_offsetB  = DIM*jnrB;
1777             j_coord_offsetC  = DIM*jnrC;
1778             j_coord_offsetD  = DIM*jnrD;
1779             j_coord_offsetE  = DIM*jnrE;
1780             j_coord_offsetF  = DIM*jnrF;
1781             j_coord_offsetG  = DIM*jnrG;
1782             j_coord_offsetH  = DIM*jnrH;
1783
1784             /* load j atom coordinates */
1785             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1786                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1787                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1788                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1789                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1790                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1791
1792             /* Calculate displacement vector */
1793             dx00             = _mm256_sub_ps(ix0,jx0);
1794             dy00             = _mm256_sub_ps(iy0,jy0);
1795             dz00             = _mm256_sub_ps(iz0,jz0);
1796             dx11             = _mm256_sub_ps(ix1,jx1);
1797             dy11             = _mm256_sub_ps(iy1,jy1);
1798             dz11             = _mm256_sub_ps(iz1,jz1);
1799             dx12             = _mm256_sub_ps(ix1,jx2);
1800             dy12             = _mm256_sub_ps(iy1,jy2);
1801             dz12             = _mm256_sub_ps(iz1,jz2);
1802             dx13             = _mm256_sub_ps(ix1,jx3);
1803             dy13             = _mm256_sub_ps(iy1,jy3);
1804             dz13             = _mm256_sub_ps(iz1,jz3);
1805             dx21             = _mm256_sub_ps(ix2,jx1);
1806             dy21             = _mm256_sub_ps(iy2,jy1);
1807             dz21             = _mm256_sub_ps(iz2,jz1);
1808             dx22             = _mm256_sub_ps(ix2,jx2);
1809             dy22             = _mm256_sub_ps(iy2,jy2);
1810             dz22             = _mm256_sub_ps(iz2,jz2);
1811             dx23             = _mm256_sub_ps(ix2,jx3);
1812             dy23             = _mm256_sub_ps(iy2,jy3);
1813             dz23             = _mm256_sub_ps(iz2,jz3);
1814             dx31             = _mm256_sub_ps(ix3,jx1);
1815             dy31             = _mm256_sub_ps(iy3,jy1);
1816             dz31             = _mm256_sub_ps(iz3,jz1);
1817             dx32             = _mm256_sub_ps(ix3,jx2);
1818             dy32             = _mm256_sub_ps(iy3,jy2);
1819             dz32             = _mm256_sub_ps(iz3,jz2);
1820             dx33             = _mm256_sub_ps(ix3,jx3);
1821             dy33             = _mm256_sub_ps(iy3,jy3);
1822             dz33             = _mm256_sub_ps(iz3,jz3);
1823
1824             /* Calculate squared distance and things based on it */
1825             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1826             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1827             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1828             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1829             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1830             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1831             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1832             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1833             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1834             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1835
1836             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1837             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1838             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1839             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
1840             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1841             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1842             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
1843             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
1844             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
1845             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
1846
1847             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1848             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1849             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1850             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1851             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1852             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1853             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1854             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1855             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1856
1857             fjx0             = _mm256_setzero_ps();
1858             fjy0             = _mm256_setzero_ps();
1859             fjz0             = _mm256_setzero_ps();
1860             fjx1             = _mm256_setzero_ps();
1861             fjy1             = _mm256_setzero_ps();
1862             fjz1             = _mm256_setzero_ps();
1863             fjx2             = _mm256_setzero_ps();
1864             fjy2             = _mm256_setzero_ps();
1865             fjz2             = _mm256_setzero_ps();
1866             fjx3             = _mm256_setzero_ps();
1867             fjy3             = _mm256_setzero_ps();
1868             fjz3             = _mm256_setzero_ps();
1869
1870             /**************************
1871              * CALCULATE INTERACTIONS *
1872              **************************/
1873
1874             r00              = _mm256_mul_ps(rsq00,rinv00);
1875             r00              = _mm256_andnot_ps(dummy_mask,r00);
1876
1877             /* Calculate table index by multiplying r with table scale and truncate to integer */
1878             rt               = _mm256_mul_ps(r00,vftabscale);
1879             vfitab           = _mm256_cvttps_epi32(rt);
1880             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1881             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1882             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1883             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1884             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1885             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1886
1887             /* CUBIC SPLINE TABLE DISPERSION */
1888             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1889                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1890             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1891                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1892             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1893                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1894             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1895                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1896             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1897             Heps             = _mm256_mul_ps(vfeps,H);
1898             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1899             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1900             fvdw6            = _mm256_mul_ps(c6_00,FF);
1901
1902             /* CUBIC SPLINE TABLE REPULSION */
1903             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1904             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1905             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1906                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1907             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1908                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1909             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1910                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1911             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1912                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1913             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1914             Heps             = _mm256_mul_ps(vfeps,H);
1915             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1916             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1917             fvdw12           = _mm256_mul_ps(c12_00,FF);
1918             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1919
1920             fscal            = fvdw;
1921
1922             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1923
1924             /* Calculate temporary vectorial force */
1925             tx               = _mm256_mul_ps(fscal,dx00);
1926             ty               = _mm256_mul_ps(fscal,dy00);
1927             tz               = _mm256_mul_ps(fscal,dz00);
1928
1929             /* Update vectorial force */
1930             fix0             = _mm256_add_ps(fix0,tx);
1931             fiy0             = _mm256_add_ps(fiy0,ty);
1932             fiz0             = _mm256_add_ps(fiz0,tz);
1933
1934             fjx0             = _mm256_add_ps(fjx0,tx);
1935             fjy0             = _mm256_add_ps(fjy0,ty);
1936             fjz0             = _mm256_add_ps(fjz0,tz);
1937
1938             /**************************
1939              * CALCULATE INTERACTIONS *
1940              **************************/
1941
1942             /* REACTION-FIELD ELECTROSTATICS */
1943             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1944
1945             fscal            = felec;
1946
1947             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1948
1949             /* Calculate temporary vectorial force */
1950             tx               = _mm256_mul_ps(fscal,dx11);
1951             ty               = _mm256_mul_ps(fscal,dy11);
1952             tz               = _mm256_mul_ps(fscal,dz11);
1953
1954             /* Update vectorial force */
1955             fix1             = _mm256_add_ps(fix1,tx);
1956             fiy1             = _mm256_add_ps(fiy1,ty);
1957             fiz1             = _mm256_add_ps(fiz1,tz);
1958
1959             fjx1             = _mm256_add_ps(fjx1,tx);
1960             fjy1             = _mm256_add_ps(fjy1,ty);
1961             fjz1             = _mm256_add_ps(fjz1,tz);
1962
1963             /**************************
1964              * CALCULATE INTERACTIONS *
1965              **************************/
1966
1967             /* REACTION-FIELD ELECTROSTATICS */
1968             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1969
1970             fscal            = felec;
1971
1972             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1973
1974             /* Calculate temporary vectorial force */
1975             tx               = _mm256_mul_ps(fscal,dx12);
1976             ty               = _mm256_mul_ps(fscal,dy12);
1977             tz               = _mm256_mul_ps(fscal,dz12);
1978
1979             /* Update vectorial force */
1980             fix1             = _mm256_add_ps(fix1,tx);
1981             fiy1             = _mm256_add_ps(fiy1,ty);
1982             fiz1             = _mm256_add_ps(fiz1,tz);
1983
1984             fjx2             = _mm256_add_ps(fjx2,tx);
1985             fjy2             = _mm256_add_ps(fjy2,ty);
1986             fjz2             = _mm256_add_ps(fjz2,tz);
1987
1988             /**************************
1989              * CALCULATE INTERACTIONS *
1990              **************************/
1991
1992             /* REACTION-FIELD ELECTROSTATICS */
1993             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
1994
1995             fscal            = felec;
1996
1997             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1998
1999             /* Calculate temporary vectorial force */
2000             tx               = _mm256_mul_ps(fscal,dx13);
2001             ty               = _mm256_mul_ps(fscal,dy13);
2002             tz               = _mm256_mul_ps(fscal,dz13);
2003
2004             /* Update vectorial force */
2005             fix1             = _mm256_add_ps(fix1,tx);
2006             fiy1             = _mm256_add_ps(fiy1,ty);
2007             fiz1             = _mm256_add_ps(fiz1,tz);
2008
2009             fjx3             = _mm256_add_ps(fjx3,tx);
2010             fjy3             = _mm256_add_ps(fjy3,ty);
2011             fjz3             = _mm256_add_ps(fjz3,tz);
2012
2013             /**************************
2014              * CALCULATE INTERACTIONS *
2015              **************************/
2016
2017             /* REACTION-FIELD ELECTROSTATICS */
2018             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
2019
2020             fscal            = felec;
2021
2022             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2023
2024             /* Calculate temporary vectorial force */
2025             tx               = _mm256_mul_ps(fscal,dx21);
2026             ty               = _mm256_mul_ps(fscal,dy21);
2027             tz               = _mm256_mul_ps(fscal,dz21);
2028
2029             /* Update vectorial force */
2030             fix2             = _mm256_add_ps(fix2,tx);
2031             fiy2             = _mm256_add_ps(fiy2,ty);
2032             fiz2             = _mm256_add_ps(fiz2,tz);
2033
2034             fjx1             = _mm256_add_ps(fjx1,tx);
2035             fjy1             = _mm256_add_ps(fjy1,ty);
2036             fjz1             = _mm256_add_ps(fjz1,tz);
2037
2038             /**************************
2039              * CALCULATE INTERACTIONS *
2040              **************************/
2041
2042             /* REACTION-FIELD ELECTROSTATICS */
2043             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
2044
2045             fscal            = felec;
2046
2047             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2048
2049             /* Calculate temporary vectorial force */
2050             tx               = _mm256_mul_ps(fscal,dx22);
2051             ty               = _mm256_mul_ps(fscal,dy22);
2052             tz               = _mm256_mul_ps(fscal,dz22);
2053
2054             /* Update vectorial force */
2055             fix2             = _mm256_add_ps(fix2,tx);
2056             fiy2             = _mm256_add_ps(fiy2,ty);
2057             fiz2             = _mm256_add_ps(fiz2,tz);
2058
2059             fjx2             = _mm256_add_ps(fjx2,tx);
2060             fjy2             = _mm256_add_ps(fjy2,ty);
2061             fjz2             = _mm256_add_ps(fjz2,tz);
2062
2063             /**************************
2064              * CALCULATE INTERACTIONS *
2065              **************************/
2066
2067             /* REACTION-FIELD ELECTROSTATICS */
2068             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
2069
2070             fscal            = felec;
2071
2072             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2073
2074             /* Calculate temporary vectorial force */
2075             tx               = _mm256_mul_ps(fscal,dx23);
2076             ty               = _mm256_mul_ps(fscal,dy23);
2077             tz               = _mm256_mul_ps(fscal,dz23);
2078
2079             /* Update vectorial force */
2080             fix2             = _mm256_add_ps(fix2,tx);
2081             fiy2             = _mm256_add_ps(fiy2,ty);
2082             fiz2             = _mm256_add_ps(fiz2,tz);
2083
2084             fjx3             = _mm256_add_ps(fjx3,tx);
2085             fjy3             = _mm256_add_ps(fjy3,ty);
2086             fjz3             = _mm256_add_ps(fjz3,tz);
2087
2088             /**************************
2089              * CALCULATE INTERACTIONS *
2090              **************************/
2091
2092             /* REACTION-FIELD ELECTROSTATICS */
2093             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
2094
2095             fscal            = felec;
2096
2097             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2098
2099             /* Calculate temporary vectorial force */
2100             tx               = _mm256_mul_ps(fscal,dx31);
2101             ty               = _mm256_mul_ps(fscal,dy31);
2102             tz               = _mm256_mul_ps(fscal,dz31);
2103
2104             /* Update vectorial force */
2105             fix3             = _mm256_add_ps(fix3,tx);
2106             fiy3             = _mm256_add_ps(fiy3,ty);
2107             fiz3             = _mm256_add_ps(fiz3,tz);
2108
2109             fjx1             = _mm256_add_ps(fjx1,tx);
2110             fjy1             = _mm256_add_ps(fjy1,ty);
2111             fjz1             = _mm256_add_ps(fjz1,tz);
2112
2113             /**************************
2114              * CALCULATE INTERACTIONS *
2115              **************************/
2116
2117             /* REACTION-FIELD ELECTROSTATICS */
2118             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
2119
2120             fscal            = felec;
2121
2122             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2123
2124             /* Calculate temporary vectorial force */
2125             tx               = _mm256_mul_ps(fscal,dx32);
2126             ty               = _mm256_mul_ps(fscal,dy32);
2127             tz               = _mm256_mul_ps(fscal,dz32);
2128
2129             /* Update vectorial force */
2130             fix3             = _mm256_add_ps(fix3,tx);
2131             fiy3             = _mm256_add_ps(fiy3,ty);
2132             fiz3             = _mm256_add_ps(fiz3,tz);
2133
2134             fjx2             = _mm256_add_ps(fjx2,tx);
2135             fjy2             = _mm256_add_ps(fjy2,ty);
2136             fjz2             = _mm256_add_ps(fjz2,tz);
2137
2138             /**************************
2139              * CALCULATE INTERACTIONS *
2140              **************************/
2141
2142             /* REACTION-FIELD ELECTROSTATICS */
2143             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
2144
2145             fscal            = felec;
2146
2147             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2148
2149             /* Calculate temporary vectorial force */
2150             tx               = _mm256_mul_ps(fscal,dx33);
2151             ty               = _mm256_mul_ps(fscal,dy33);
2152             tz               = _mm256_mul_ps(fscal,dz33);
2153
2154             /* Update vectorial force */
2155             fix3             = _mm256_add_ps(fix3,tx);
2156             fiy3             = _mm256_add_ps(fiy3,ty);
2157             fiz3             = _mm256_add_ps(fiz3,tz);
2158
2159             fjx3             = _mm256_add_ps(fjx3,tx);
2160             fjy3             = _mm256_add_ps(fjy3,ty);
2161             fjz3             = _mm256_add_ps(fjz3,tz);
2162
2163             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2164             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2165             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2166             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2167             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2168             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2169             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2170             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2171
2172             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2173                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2174                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2175
2176             /* Inner loop uses 295 flops */
2177         }
2178
2179         /* End of innermost loop */
2180
2181         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2182                                                  f+i_coord_offset,fshift+i_shift_offset);
2183
2184         /* Increment number of inner iterations */
2185         inneriter                  += j_index_end - j_index_start;
2186
2187         /* Outer loop uses 24 flops */
2188     }
2189
2190     /* Increment number of outer iterations */
2191     outeriter        += nri;
2192
2193     /* Update outer/inner flops */
2194
2195     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*295);
2196 }