src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_VF_avx_256_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     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     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 101     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 102     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 103     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 104     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 105     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 106     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 107     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 108     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 109     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 110     real             *charge;
 111     int              nvdwtype;
 112     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 113     int              *vdwtype;
 114     real             *vdwparam;
 115     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 116     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 117     __m256i          vfitab;
 118     __m128i          vfitab_lo,vfitab_hi;
 119     __m128i          ifour       = _mm_set1_epi32(4);
 120     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 121     real             *vftab;
 122     __m256           dummy_mask,cutoff_mask;
 123     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 124     __m256           one     = _mm256_set1_ps(1.0);
 125     __m256           two     = _mm256_set1_ps(2.0);
 126     x                = xx[0];
 127     f                = ff[0];
 128
 129     nri              = nlist->nri;
 130     iinr             = nlist->iinr;
 131     jindex           = nlist->jindex;
 132     jjnr             = nlist->jjnr;
 133     shiftidx         = nlist->shift;
 134     gid              = nlist->gid;
 135     shiftvec         = fr->shift_vec[0];
 136     fshift           = fr->fshift[0];
 137     facel            = _mm256_set1_ps(fr->epsfac);
 138     charge           = mdatoms->chargeA;
 139     krf              = _mm256_set1_ps(fr->ic->k_rf);
 140     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 141     crf              = _mm256_set1_ps(fr->ic->c_rf);
 142     nvdwtype         = fr->ntype;
 143     vdwparam         = fr->nbfp;
 144     vdwtype          = mdatoms->typeA;
 145
 146     vftab            = kernel_data->table_vdw->data;
 147     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 148
 149     /* Setup water-specific parameters */
 150     inr              = nlist->iinr[0];
 151     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 152     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 153     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 154     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 155
 156     jq0              = _mm256_set1_ps(charge[inr+0]);
 157     jq1              = _mm256_set1_ps(charge[inr+1]);
 158     jq2              = _mm256_set1_ps(charge[inr+2]);
 159     vdwjidx0A        = 2*vdwtype[inr+0];
 160     qq00             = _mm256_mul_ps(iq0,jq0);
 161     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 162     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 163     qq01             = _mm256_mul_ps(iq0,jq1);
 164     qq02             = _mm256_mul_ps(iq0,jq2);
 165     qq10             = _mm256_mul_ps(iq1,jq0);
 166     qq11             = _mm256_mul_ps(iq1,jq1);
 167     qq12             = _mm256_mul_ps(iq1,jq2);
 168     qq20             = _mm256_mul_ps(iq2,jq0);
 169     qq21             = _mm256_mul_ps(iq2,jq1);
 170     qq22             = _mm256_mul_ps(iq2,jq2);
 171
 172     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 173     rcutoff_scalar   = fr->rcoulomb;
 174     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 175     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 176
 177     /* Avoid stupid compiler warnings */
 178     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 179     j_coord_offsetA = 0;
 180     j_coord_offsetB = 0;
 181     j_coord_offsetC = 0;
 182     j_coord_offsetD = 0;
 183     j_coord_offsetE = 0;
 184     j_coord_offsetF = 0;
 185     j_coord_offsetG = 0;
 186     j_coord_offsetH = 0;
 187
 188     outeriter        = 0;
 189     inneriter        = 0;
 190
 191     for(iidx=0;iidx<4*DIM;iidx++)
 192     {
 193         scratch[iidx] = 0.0;
 194     }
 195
 196     /* Start outer loop over neighborlists */
 197     for(iidx=0; iidx<nri; iidx++)
 198     {
 199         /* Load shift vector for this list */
 200         i_shift_offset   = DIM*shiftidx[iidx];
 201
 202         /* Load limits for loop over neighbors */
 203         j_index_start    = jindex[iidx];
 204         j_index_end      = jindex[iidx+1];
 205
 206         /* Get outer coordinate index */
 207         inr              = iinr[iidx];
 208         i_coord_offset   = DIM*inr;
 209
 210         /* Load i particle coords and add shift vector */
 211         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 212                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 213
 214         fix0             = _mm256_setzero_ps();
 215         fiy0             = _mm256_setzero_ps();
 216         fiz0             = _mm256_setzero_ps();
 217         fix1             = _mm256_setzero_ps();
 218         fiy1             = _mm256_setzero_ps();
 219         fiz1             = _mm256_setzero_ps();
 220         fix2             = _mm256_setzero_ps();
 221         fiy2             = _mm256_setzero_ps();
 222         fiz2             = _mm256_setzero_ps();
 223
 224         /* Reset potential sums */
 225         velecsum         = _mm256_setzero_ps();
 226         vvdwsum          = _mm256_setzero_ps();
 227
 228         /* Start inner kernel loop */
 229         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 230         {
 231
 232             /* Get j neighbor index, and coordinate index */
 233             jnrA             = jjnr[jidx];
 234             jnrB             = jjnr[jidx+1];
 235             jnrC             = jjnr[jidx+2];
 236             jnrD             = jjnr[jidx+3];
 237             jnrE             = jjnr[jidx+4];
 238             jnrF             = jjnr[jidx+5];
 239             jnrG             = jjnr[jidx+6];
 240             jnrH             = jjnr[jidx+7];
 241             j_coord_offsetA  = DIM*jnrA;
 242             j_coord_offsetB  = DIM*jnrB;
 243             j_coord_offsetC  = DIM*jnrC;
 244             j_coord_offsetD  = DIM*jnrD;
 245             j_coord_offsetE  = DIM*jnrE;
 246             j_coord_offsetF  = DIM*jnrF;
 247             j_coord_offsetG  = DIM*jnrG;
 248             j_coord_offsetH  = DIM*jnrH;
 249
 250             /* load j atom coordinates */
 251             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 252                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 253                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 254                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 255                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 256
 257             /* Calculate displacement vector */
 258             dx00             = _mm256_sub_ps(ix0,jx0);
 259             dy00             = _mm256_sub_ps(iy0,jy0);
 260             dz00             = _mm256_sub_ps(iz0,jz0);
 261             dx01             = _mm256_sub_ps(ix0,jx1);
 262             dy01             = _mm256_sub_ps(iy0,jy1);
 263             dz01             = _mm256_sub_ps(iz0,jz1);
 264             dx02             = _mm256_sub_ps(ix0,jx2);
 265             dy02             = _mm256_sub_ps(iy0,jy2);
 266             dz02             = _mm256_sub_ps(iz0,jz2);
 267             dx10             = _mm256_sub_ps(ix1,jx0);
 268             dy10             = _mm256_sub_ps(iy1,jy0);
 269             dz10             = _mm256_sub_ps(iz1,jz0);
 270             dx11             = _mm256_sub_ps(ix1,jx1);
 271             dy11             = _mm256_sub_ps(iy1,jy1);
 272             dz11             = _mm256_sub_ps(iz1,jz1);
 273             dx12             = _mm256_sub_ps(ix1,jx2);
 274             dy12             = _mm256_sub_ps(iy1,jy2);
 275             dz12             = _mm256_sub_ps(iz1,jz2);
 276             dx20             = _mm256_sub_ps(ix2,jx0);
 277             dy20             = _mm256_sub_ps(iy2,jy0);
 278             dz20             = _mm256_sub_ps(iz2,jz0);
 279             dx21             = _mm256_sub_ps(ix2,jx1);
 280             dy21             = _mm256_sub_ps(iy2,jy1);
 281             dz21             = _mm256_sub_ps(iz2,jz1);
 282             dx22             = _mm256_sub_ps(ix2,jx2);
 283             dy22             = _mm256_sub_ps(iy2,jy2);
 284             dz22             = _mm256_sub_ps(iz2,jz2);
 285
 286             /* Calculate squared distance and things based on it */
 287             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 288             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 289             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 290             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 291             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 292             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 293             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 294             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 295             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 296
 297             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 298             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 299             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 300             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 301             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 302             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 303             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 304             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 305             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 306
 307             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 308             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 309             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 310             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 311             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 312             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 313             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 314             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 315             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 316
 317             fjx0             = _mm256_setzero_ps();
 318             fjy0             = _mm256_setzero_ps();
 319             fjz0             = _mm256_setzero_ps();
 320             fjx1             = _mm256_setzero_ps();
 321             fjy1             = _mm256_setzero_ps();
 322             fjz1             = _mm256_setzero_ps();
 323             fjx2             = _mm256_setzero_ps();
 324             fjy2             = _mm256_setzero_ps();
 325             fjz2             = _mm256_setzero_ps();
 326
 327             /**************************
 328              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 332             {
 333
 334             r00              = _mm256_mul_ps(rsq00,rinv00);
 335
 336             /* Calculate table index by multiplying r with table scale and truncate to integer */
 337             rt               = _mm256_mul_ps(r00,vftabscale);
 338             vfitab           = _mm256_cvttps_epi32(rt);
 339             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 340             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 341             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 342             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 343             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 344             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 345
 346             /* REACTION-FIELD ELECTROSTATICS */
 347             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 348             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 349
 350             /* CUBIC SPLINE TABLE DISPERSION */
 351             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 352                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 353             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 354                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 355             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 356                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 357             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 358                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 359             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 360             Heps             = _mm256_mul_ps(vfeps,H);
 361             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 362             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 363             vvdw6            = _mm256_mul_ps(c6_00,VV);
 364             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 365             fvdw6            = _mm256_mul_ps(c6_00,FF);
 366
 367             /* CUBIC SPLINE TABLE REPULSION */
 368             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 369             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 370             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 371                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 372             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 373                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 374             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 375                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 376             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 377                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 378             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 379             Heps             = _mm256_mul_ps(vfeps,H);
 380             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 381             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 382             vvdw12           = _mm256_mul_ps(c12_00,VV);
 383             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 384             fvdw12           = _mm256_mul_ps(c12_00,FF);
 385             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 386             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 387
 388             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velec            = _mm256_and_ps(velec,cutoff_mask);
 392             velecsum         = _mm256_add_ps(velecsum,velec);
 393             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 394             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 395
 396             fscal            = _mm256_add_ps(felec,fvdw);
 397
 398             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 399
 400             /* Calculate temporary vectorial force */
 401             tx               = _mm256_mul_ps(fscal,dx00);
 402             ty               = _mm256_mul_ps(fscal,dy00);
 403             tz               = _mm256_mul_ps(fscal,dz00);
 404
 405             /* Update vectorial force */
 406             fix0             = _mm256_add_ps(fix0,tx);
 407             fiy0             = _mm256_add_ps(fiy0,ty);
 408             fiz0             = _mm256_add_ps(fiz0,tz);
 409
 410             fjx0             = _mm256_add_ps(fjx0,tx);
 411             fjy0             = _mm256_add_ps(fjy0,ty);
 412             fjz0             = _mm256_add_ps(fjz0,tz);
 413
 414             }
 415
 416             /**************************
 417              * CALCULATE INTERACTIONS *
 418              **************************/
 419
 420             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 421             {
 422
 423             /* REACTION-FIELD ELECTROSTATICS */
 424             velec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
 425             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
 426
 427             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 428
 429             /* Update potential sum for this i atom from the interaction with this j atom. */
 430             velec            = _mm256_and_ps(velec,cutoff_mask);
 431             velecsum         = _mm256_add_ps(velecsum,velec);
 432
 433             fscal            = felec;
 434
 435             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 436
 437             /* Calculate temporary vectorial force */
 438             tx               = _mm256_mul_ps(fscal,dx01);
 439             ty               = _mm256_mul_ps(fscal,dy01);
 440             tz               = _mm256_mul_ps(fscal,dz01);
 441
 442             /* Update vectorial force */
 443             fix0             = _mm256_add_ps(fix0,tx);
 444             fiy0             = _mm256_add_ps(fiy0,ty);
 445             fiz0             = _mm256_add_ps(fiz0,tz);
 446
 447             fjx1             = _mm256_add_ps(fjx1,tx);
 448             fjy1             = _mm256_add_ps(fjy1,ty);
 449             fjz1             = _mm256_add_ps(fjz1,tz);
 450
 451             }
 452
 453             /**************************
 454              * CALCULATE INTERACTIONS *
 455              **************************/
 456
 457             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 458             {
 459
 460             /* REACTION-FIELD ELECTROSTATICS */
 461             velec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
 462             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
 463
 464             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
 465
 466             /* Update potential sum for this i atom from the interaction with this j atom. */
 467             velec            = _mm256_and_ps(velec,cutoff_mask);
 468             velecsum         = _mm256_add_ps(velecsum,velec);
 469
 470             fscal            = felec;
 471
 472             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 473
 474             /* Calculate temporary vectorial force */
 475             tx               = _mm256_mul_ps(fscal,dx02);
 476             ty               = _mm256_mul_ps(fscal,dy02);
 477             tz               = _mm256_mul_ps(fscal,dz02);
 478
 479             /* Update vectorial force */
 480             fix0             = _mm256_add_ps(fix0,tx);
 481             fiy0             = _mm256_add_ps(fiy0,ty);
 482             fiz0             = _mm256_add_ps(fiz0,tz);
 483
 484             fjx2             = _mm256_add_ps(fjx2,tx);
 485             fjy2             = _mm256_add_ps(fjy2,ty);
 486             fjz2             = _mm256_add_ps(fjz2,tz);
 487
 488             }
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 495             {
 496
 497             /* REACTION-FIELD ELECTROSTATICS */
 498             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 499             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 500
 501             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 502
 503             /* Update potential sum for this i atom from the interaction with this j atom. */
 504             velec            = _mm256_and_ps(velec,cutoff_mask);
 505             velecsum         = _mm256_add_ps(velecsum,velec);
 506
 507             fscal            = felec;
 508
 509             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 510
 511             /* Calculate temporary vectorial force */
 512             tx               = _mm256_mul_ps(fscal,dx10);
 513             ty               = _mm256_mul_ps(fscal,dy10);
 514             tz               = _mm256_mul_ps(fscal,dz10);
 515
 516             /* Update vectorial force */
 517             fix1             = _mm256_add_ps(fix1,tx);
 518             fiy1             = _mm256_add_ps(fiy1,ty);
 519             fiz1             = _mm256_add_ps(fiz1,tz);
 520
 521             fjx0             = _mm256_add_ps(fjx0,tx);
 522             fjy0             = _mm256_add_ps(fjy0,ty);
 523             fjz0             = _mm256_add_ps(fjz0,tz);
 524
 525             }
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 532             {
 533
 534             /* REACTION-FIELD ELECTROSTATICS */
 535             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 536             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 537
 538             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 539
 540             /* Update potential sum for this i atom from the interaction with this j atom. */
 541             velec            = _mm256_and_ps(velec,cutoff_mask);
 542             velecsum         = _mm256_add_ps(velecsum,velec);
 543
 544             fscal            = felec;
 545
 546             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 547
 548             /* Calculate temporary vectorial force */
 549             tx               = _mm256_mul_ps(fscal,dx11);
 550             ty               = _mm256_mul_ps(fscal,dy11);
 551             tz               = _mm256_mul_ps(fscal,dz11);
 552
 553             /* Update vectorial force */
 554             fix1             = _mm256_add_ps(fix1,tx);
 555             fiy1             = _mm256_add_ps(fiy1,ty);
 556             fiz1             = _mm256_add_ps(fiz1,tz);
 557
 558             fjx1             = _mm256_add_ps(fjx1,tx);
 559             fjy1             = _mm256_add_ps(fjy1,ty);
 560             fjz1             = _mm256_add_ps(fjz1,tz);
 561
 562             }
 563
 564             /**************************
 565              * CALCULATE INTERACTIONS *
 566              **************************/
 567
 568             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 569             {
 570
 571             /* REACTION-FIELD ELECTROSTATICS */
 572             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 573             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 574
 575             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm256_and_ps(velec,cutoff_mask);
 579             velecsum         = _mm256_add_ps(velecsum,velec);
 580
 581             fscal            = felec;
 582
 583             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 584
 585             /* Calculate temporary vectorial force */
 586             tx               = _mm256_mul_ps(fscal,dx12);
 587             ty               = _mm256_mul_ps(fscal,dy12);
 588             tz               = _mm256_mul_ps(fscal,dz12);
 589
 590             /* Update vectorial force */
 591             fix1             = _mm256_add_ps(fix1,tx);
 592             fiy1             = _mm256_add_ps(fiy1,ty);
 593             fiz1             = _mm256_add_ps(fiz1,tz);
 594
 595             fjx2             = _mm256_add_ps(fjx2,tx);
 596             fjy2             = _mm256_add_ps(fjy2,ty);
 597             fjz2             = _mm256_add_ps(fjz2,tz);
 598
 599             }
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 606             {
 607
 608             /* REACTION-FIELD ELECTROSTATICS */
 609             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 610             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 611
 612             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 613
 614             /* Update potential sum for this i atom from the interaction with this j atom. */
 615             velec            = _mm256_and_ps(velec,cutoff_mask);
 616             velecsum         = _mm256_add_ps(velecsum,velec);
 617
 618             fscal            = felec;
 619
 620             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 621
 622             /* Calculate temporary vectorial force */
 623             tx               = _mm256_mul_ps(fscal,dx20);
 624             ty               = _mm256_mul_ps(fscal,dy20);
 625             tz               = _mm256_mul_ps(fscal,dz20);
 626
 627             /* Update vectorial force */
 628             fix2             = _mm256_add_ps(fix2,tx);
 629             fiy2             = _mm256_add_ps(fiy2,ty);
 630             fiz2             = _mm256_add_ps(fiz2,tz);
 631
 632             fjx0             = _mm256_add_ps(fjx0,tx);
 633             fjy0             = _mm256_add_ps(fjy0,ty);
 634             fjz0             = _mm256_add_ps(fjz0,tz);
 635
 636             }
 637
 638             /**************************
 639              * CALCULATE INTERACTIONS *
 640              **************************/
 641
 642             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 643             {
 644
 645             /* REACTION-FIELD ELECTROSTATICS */
 646             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
 647             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
 648
 649             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
 650
 651             /* Update potential sum for this i atom from the interaction with this j atom. */
 652             velec            = _mm256_and_ps(velec,cutoff_mask);
 653             velecsum         = _mm256_add_ps(velecsum,velec);
 654
 655             fscal            = felec;
 656
 657             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 658
 659             /* Calculate temporary vectorial force */
 660             tx               = _mm256_mul_ps(fscal,dx21);
 661             ty               = _mm256_mul_ps(fscal,dy21);
 662             tz               = _mm256_mul_ps(fscal,dz21);
 663
 664             /* Update vectorial force */
 665             fix2             = _mm256_add_ps(fix2,tx);
 666             fiy2             = _mm256_add_ps(fiy2,ty);
 667             fiz2             = _mm256_add_ps(fiz2,tz);
 668
 669             fjx1             = _mm256_add_ps(fjx1,tx);
 670             fjy1             = _mm256_add_ps(fjy1,ty);
 671             fjz1             = _mm256_add_ps(fjz1,tz);
 672
 673             }
 674
 675             /**************************
 676              * CALCULATE INTERACTIONS *
 677              **************************/
 678
 679             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 680             {
 681
 682             /* REACTION-FIELD ELECTROSTATICS */
 683             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
 684             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
 685
 686             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
 687
 688             /* Update potential sum for this i atom from the interaction with this j atom. */
 689             velec            = _mm256_and_ps(velec,cutoff_mask);
 690             velecsum         = _mm256_add_ps(velecsum,velec);
 691
 692             fscal            = felec;
 693
 694             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 695
 696             /* Calculate temporary vectorial force */
 697             tx               = _mm256_mul_ps(fscal,dx22);
 698             ty               = _mm256_mul_ps(fscal,dy22);
 699             tz               = _mm256_mul_ps(fscal,dz22);
 700
 701             /* Update vectorial force */
 702             fix2             = _mm256_add_ps(fix2,tx);
 703             fiy2             = _mm256_add_ps(fiy2,ty);
 704             fiz2             = _mm256_add_ps(fiz2,tz);
 705
 706             fjx2             = _mm256_add_ps(fjx2,tx);
 707             fjy2             = _mm256_add_ps(fjy2,ty);
 708             fjz2             = _mm256_add_ps(fjz2,tz);
 709
 710             }
 711
 712             fjptrA             = f+j_coord_offsetA;
 713             fjptrB             = f+j_coord_offsetB;
 714             fjptrC             = f+j_coord_offsetC;
 715             fjptrD             = f+j_coord_offsetD;
 716             fjptrE             = f+j_coord_offsetE;
 717             fjptrF             = f+j_coord_offsetF;
 718             fjptrG             = f+j_coord_offsetG;
 719             fjptrH             = f+j_coord_offsetH;
 720
 721             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 722                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 723
 724             /* Inner loop uses 360 flops */
 725         }
 726
 727         if(jidx<j_index_end)
 728         {
 729
 730             /* Get j neighbor index, and coordinate index */
 731             jnrlistA         = jjnr[jidx];
 732             jnrlistB         = jjnr[jidx+1];
 733             jnrlistC         = jjnr[jidx+2];
 734             jnrlistD         = jjnr[jidx+3];
 735             jnrlistE         = jjnr[jidx+4];
 736             jnrlistF         = jjnr[jidx+5];
 737             jnrlistG         = jjnr[jidx+6];
 738             jnrlistH         = jjnr[jidx+7];
 739             /* Sign of each element will be negative for non-real atoms.
 740              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 741              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 742              */
 743             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 744                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 745
 746             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 747             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 748             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 749             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 750             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 751             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 752             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 753             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 754             j_coord_offsetA  = DIM*jnrA;
 755             j_coord_offsetB  = DIM*jnrB;
 756             j_coord_offsetC  = DIM*jnrC;
 757             j_coord_offsetD  = DIM*jnrD;
 758             j_coord_offsetE  = DIM*jnrE;
 759             j_coord_offsetF  = DIM*jnrF;
 760             j_coord_offsetG  = DIM*jnrG;
 761             j_coord_offsetH  = DIM*jnrH;
 762
 763             /* load j atom coordinates */
 764             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 765                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 766                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 767                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 768                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 769
 770             /* Calculate displacement vector */
 771             dx00             = _mm256_sub_ps(ix0,jx0);
 772             dy00             = _mm256_sub_ps(iy0,jy0);
 773             dz00             = _mm256_sub_ps(iz0,jz0);
 774             dx01             = _mm256_sub_ps(ix0,jx1);
 775             dy01             = _mm256_sub_ps(iy0,jy1);
 776             dz01             = _mm256_sub_ps(iz0,jz1);
 777             dx02             = _mm256_sub_ps(ix0,jx2);
 778             dy02             = _mm256_sub_ps(iy0,jy2);
 779             dz02             = _mm256_sub_ps(iz0,jz2);
 780             dx10             = _mm256_sub_ps(ix1,jx0);
 781             dy10             = _mm256_sub_ps(iy1,jy0);
 782             dz10             = _mm256_sub_ps(iz1,jz0);
 783             dx11             = _mm256_sub_ps(ix1,jx1);
 784             dy11             = _mm256_sub_ps(iy1,jy1);
 785             dz11             = _mm256_sub_ps(iz1,jz1);
 786             dx12             = _mm256_sub_ps(ix1,jx2);
 787             dy12             = _mm256_sub_ps(iy1,jy2);
 788             dz12             = _mm256_sub_ps(iz1,jz2);
 789             dx20             = _mm256_sub_ps(ix2,jx0);
 790             dy20             = _mm256_sub_ps(iy2,jy0);
 791             dz20             = _mm256_sub_ps(iz2,jz0);
 792             dx21             = _mm256_sub_ps(ix2,jx1);
 793             dy21             = _mm256_sub_ps(iy2,jy1);
 794             dz21             = _mm256_sub_ps(iz2,jz1);
 795             dx22             = _mm256_sub_ps(ix2,jx2);
 796             dy22             = _mm256_sub_ps(iy2,jy2);
 797             dz22             = _mm256_sub_ps(iz2,jz2);
 798
 799             /* Calculate squared distance and things based on it */
 800             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 801             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 802             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 803             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 804             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 805             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 806             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 807             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 808             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 809
 810             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 811             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 812             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 813             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 814             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 815             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 816             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 817             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 818             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 819
 820             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 821             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 822             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 823             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 824             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 825             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 826             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 827             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 828             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 829
 830             fjx0             = _mm256_setzero_ps();
 831             fjy0             = _mm256_setzero_ps();
 832             fjz0             = _mm256_setzero_ps();
 833             fjx1             = _mm256_setzero_ps();
 834             fjy1             = _mm256_setzero_ps();
 835             fjz1             = _mm256_setzero_ps();
 836             fjx2             = _mm256_setzero_ps();
 837             fjy2             = _mm256_setzero_ps();
 838             fjz2             = _mm256_setzero_ps();
 839
 840             /**************************
 841              * CALCULATE INTERACTIONS *
 842              **************************/
 843
 844             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 845             {
 846
 847             r00              = _mm256_mul_ps(rsq00,rinv00);
 848             r00              = _mm256_andnot_ps(dummy_mask,r00);
 849
 850             /* Calculate table index by multiplying r with table scale and truncate to integer */
 851             rt               = _mm256_mul_ps(r00,vftabscale);
 852             vfitab           = _mm256_cvttps_epi32(rt);
 853             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 854             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 855             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 856             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 857             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 858             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 859
 860             /* REACTION-FIELD ELECTROSTATICS */
 861             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 862             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 863
 864             /* CUBIC SPLINE TABLE DISPERSION */
 865             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 866                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 867             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 868                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 869             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 870                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 871             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 872                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 873             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 874             Heps             = _mm256_mul_ps(vfeps,H);
 875             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 876             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 877             vvdw6            = _mm256_mul_ps(c6_00,VV);
 878             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 879             fvdw6            = _mm256_mul_ps(c6_00,FF);
 880
 881             /* CUBIC SPLINE TABLE REPULSION */
 882             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 883             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 884             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 885                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 886             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 887                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 888             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 889                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 890             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 891                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 892             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 893             Heps             = _mm256_mul_ps(vfeps,H);
 894             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 895             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 896             vvdw12           = _mm256_mul_ps(c12_00,VV);
 897             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 898             fvdw12           = _mm256_mul_ps(c12_00,FF);
 899             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 900             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 901
 902             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 903
 904             /* Update potential sum for this i atom from the interaction with this j atom. */
 905             velec            = _mm256_and_ps(velec,cutoff_mask);
 906             velec            = _mm256_andnot_ps(dummy_mask,velec);
 907             velecsum         = _mm256_add_ps(velecsum,velec);
 908             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 909             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 910             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 911
 912             fscal            = _mm256_add_ps(felec,fvdw);
 913
 914             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 915
 916             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 917
 918             /* Calculate temporary vectorial force */
 919             tx               = _mm256_mul_ps(fscal,dx00);
 920             ty               = _mm256_mul_ps(fscal,dy00);
 921             tz               = _mm256_mul_ps(fscal,dz00);
 922
 923             /* Update vectorial force */
 924             fix0             = _mm256_add_ps(fix0,tx);
 925             fiy0             = _mm256_add_ps(fiy0,ty);
 926             fiz0             = _mm256_add_ps(fiz0,tz);
 927
 928             fjx0             = _mm256_add_ps(fjx0,tx);
 929             fjy0             = _mm256_add_ps(fjy0,ty);
 930             fjz0             = _mm256_add_ps(fjz0,tz);
 931
 932             }
 933
 934             /**************************
 935              * CALCULATE INTERACTIONS *
 936              **************************/
 937
 938             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 939             {
 940
 941             /* REACTION-FIELD ELECTROSTATICS */
 942             velec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
 943             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
 944
 945             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 946
 947             /* Update potential sum for this i atom from the interaction with this j atom. */
 948             velec            = _mm256_and_ps(velec,cutoff_mask);
 949             velec            = _mm256_andnot_ps(dummy_mask,velec);
 950             velecsum         = _mm256_add_ps(velecsum,velec);
 951
 952             fscal            = felec;
 953
 954             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 955
 956             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 957
 958             /* Calculate temporary vectorial force */
 959             tx               = _mm256_mul_ps(fscal,dx01);
 960             ty               = _mm256_mul_ps(fscal,dy01);
 961             tz               = _mm256_mul_ps(fscal,dz01);
 962
 963             /* Update vectorial force */
 964             fix0             = _mm256_add_ps(fix0,tx);
 965             fiy0             = _mm256_add_ps(fiy0,ty);
 966             fiz0             = _mm256_add_ps(fiz0,tz);
 967
 968             fjx1             = _mm256_add_ps(fjx1,tx);
 969             fjy1             = _mm256_add_ps(fjy1,ty);
 970             fjz1             = _mm256_add_ps(fjz1,tz);
 971
 972             }
 973
 974             /**************************
 975              * CALCULATE INTERACTIONS *
 976              **************************/
 977
 978             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 979             {
 980
 981             /* REACTION-FIELD ELECTROSTATICS */
 982             velec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
 983             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
 984
 985             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
 986
 987             /* Update potential sum for this i atom from the interaction with this j atom. */
 988             velec            = _mm256_and_ps(velec,cutoff_mask);
 989             velec            = _mm256_andnot_ps(dummy_mask,velec);
 990             velecsum         = _mm256_add_ps(velecsum,velec);
 991
 992             fscal            = felec;
 993
 994             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 995
 996             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 997
 998             /* Calculate temporary vectorial force */
 999             tx               = _mm256_mul_ps(fscal,dx02);
1000             ty               = _mm256_mul_ps(fscal,dy02);
1001             tz               = _mm256_mul_ps(fscal,dz02);
1002
1003             /* Update vectorial force */
1004             fix0             = _mm256_add_ps(fix0,tx);
1005             fiy0             = _mm256_add_ps(fiy0,ty);
1006             fiz0             = _mm256_add_ps(fiz0,tz);
1007
1008             fjx2             = _mm256_add_ps(fjx2,tx);
1009             fjy2             = _mm256_add_ps(fjy2,ty);
1010             fjz2             = _mm256_add_ps(fjz2,tz);
1011
1012             }
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1019             {
1020
1021             /* REACTION-FIELD ELECTROSTATICS */
1022             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
1023             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1024
1025             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1026
1027             /* Update potential sum for this i atom from the interaction with this j atom. */
1028             velec            = _mm256_and_ps(velec,cutoff_mask);
1029             velec            = _mm256_andnot_ps(dummy_mask,velec);
1030             velecsum         = _mm256_add_ps(velecsum,velec);
1031
1032             fscal            = felec;
1033
1034             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1035
1036             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1037
1038             /* Calculate temporary vectorial force */
1039             tx               = _mm256_mul_ps(fscal,dx10);
1040             ty               = _mm256_mul_ps(fscal,dy10);
1041             tz               = _mm256_mul_ps(fscal,dz10);
1042
1043             /* Update vectorial force */
1044             fix1             = _mm256_add_ps(fix1,tx);
1045             fiy1             = _mm256_add_ps(fiy1,ty);
1046             fiz1             = _mm256_add_ps(fiz1,tz);
1047
1048             fjx0             = _mm256_add_ps(fjx0,tx);
1049             fjy0             = _mm256_add_ps(fjy0,ty);
1050             fjz0             = _mm256_add_ps(fjz0,tz);
1051
1052             }
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1059             {
1060
1061             /* REACTION-FIELD ELECTROSTATICS */
1062             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
1063             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1064
1065             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1066
1067             /* Update potential sum for this i atom from the interaction with this j atom. */
1068             velec            = _mm256_and_ps(velec,cutoff_mask);
1069             velec            = _mm256_andnot_ps(dummy_mask,velec);
1070             velecsum         = _mm256_add_ps(velecsum,velec);
1071
1072             fscal            = felec;
1073
1074             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1075
1076             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1077
1078             /* Calculate temporary vectorial force */
1079             tx               = _mm256_mul_ps(fscal,dx11);
1080             ty               = _mm256_mul_ps(fscal,dy11);
1081             tz               = _mm256_mul_ps(fscal,dz11);
1082
1083             /* Update vectorial force */
1084             fix1             = _mm256_add_ps(fix1,tx);
1085             fiy1             = _mm256_add_ps(fiy1,ty);
1086             fiz1             = _mm256_add_ps(fiz1,tz);
1087
1088             fjx1             = _mm256_add_ps(fjx1,tx);
1089             fjy1             = _mm256_add_ps(fjy1,ty);
1090             fjz1             = _mm256_add_ps(fjz1,tz);
1091
1092             }
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1099             {
1100
1101             /* REACTION-FIELD ELECTROSTATICS */
1102             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
1103             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1104
1105             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1106
1107             /* Update potential sum for this i atom from the interaction with this j atom. */
1108             velec            = _mm256_and_ps(velec,cutoff_mask);
1109             velec            = _mm256_andnot_ps(dummy_mask,velec);
1110             velecsum         = _mm256_add_ps(velecsum,velec);
1111
1112             fscal            = felec;
1113
1114             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1115
1116             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1117
1118             /* Calculate temporary vectorial force */
1119             tx               = _mm256_mul_ps(fscal,dx12);
1120             ty               = _mm256_mul_ps(fscal,dy12);
1121             tz               = _mm256_mul_ps(fscal,dz12);
1122
1123             /* Update vectorial force */
1124             fix1             = _mm256_add_ps(fix1,tx);
1125             fiy1             = _mm256_add_ps(fiy1,ty);
1126             fiz1             = _mm256_add_ps(fiz1,tz);
1127
1128             fjx2             = _mm256_add_ps(fjx2,tx);
1129             fjy2             = _mm256_add_ps(fjy2,ty);
1130             fjz2             = _mm256_add_ps(fjz2,tz);
1131
1132             }
1133
1134             /**************************
1135              * CALCULATE INTERACTIONS *
1136              **************************/
1137
1138             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1139             {
1140
1141             /* REACTION-FIELD ELECTROSTATICS */
1142             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
1143             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1144
1145             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1146
1147             /* Update potential sum for this i atom from the interaction with this j atom. */
1148             velec            = _mm256_and_ps(velec,cutoff_mask);
1149             velec            = _mm256_andnot_ps(dummy_mask,velec);
1150             velecsum         = _mm256_add_ps(velecsum,velec);
1151
1152             fscal            = felec;
1153
1154             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1155
1156             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1157
1158             /* Calculate temporary vectorial force */
1159             tx               = _mm256_mul_ps(fscal,dx20);
1160             ty               = _mm256_mul_ps(fscal,dy20);
1161             tz               = _mm256_mul_ps(fscal,dz20);
1162
1163             /* Update vectorial force */
1164             fix2             = _mm256_add_ps(fix2,tx);
1165             fiy2             = _mm256_add_ps(fiy2,ty);
1166             fiz2             = _mm256_add_ps(fiz2,tz);
1167
1168             fjx0             = _mm256_add_ps(fjx0,tx);
1169             fjy0             = _mm256_add_ps(fjy0,ty);
1170             fjz0             = _mm256_add_ps(fjz0,tz);
1171
1172             }
1173
1174             /**************************
1175              * CALCULATE INTERACTIONS *
1176              **************************/
1177
1178             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1179             {
1180
1181             /* REACTION-FIELD ELECTROSTATICS */
1182             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
1183             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1184
1185             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1186
1187             /* Update potential sum for this i atom from the interaction with this j atom. */
1188             velec            = _mm256_and_ps(velec,cutoff_mask);
1189             velec            = _mm256_andnot_ps(dummy_mask,velec);
1190             velecsum         = _mm256_add_ps(velecsum,velec);
1191
1192             fscal            = felec;
1193
1194             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1195
1196             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1197
1198             /* Calculate temporary vectorial force */
1199             tx               = _mm256_mul_ps(fscal,dx21);
1200             ty               = _mm256_mul_ps(fscal,dy21);
1201             tz               = _mm256_mul_ps(fscal,dz21);
1202
1203             /* Update vectorial force */
1204             fix2             = _mm256_add_ps(fix2,tx);
1205             fiy2             = _mm256_add_ps(fiy2,ty);
1206             fiz2             = _mm256_add_ps(fiz2,tz);
1207
1208             fjx1             = _mm256_add_ps(fjx1,tx);
1209             fjy1             = _mm256_add_ps(fjy1,ty);
1210             fjz1             = _mm256_add_ps(fjz1,tz);
1211
1212             }
1213
1214             /**************************
1215              * CALCULATE INTERACTIONS *
1216              **************************/
1217
1218             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1219             {
1220
1221             /* REACTION-FIELD ELECTROSTATICS */
1222             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
1223             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1224
1225             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1226
1227             /* Update potential sum for this i atom from the interaction with this j atom. */
1228             velec            = _mm256_and_ps(velec,cutoff_mask);
1229             velec            = _mm256_andnot_ps(dummy_mask,velec);
1230             velecsum         = _mm256_add_ps(velecsum,velec);
1231
1232             fscal            = felec;
1233
1234             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1235
1236             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1237
1238             /* Calculate temporary vectorial force */
1239             tx               = _mm256_mul_ps(fscal,dx22);
1240             ty               = _mm256_mul_ps(fscal,dy22);
1241             tz               = _mm256_mul_ps(fscal,dz22);
1242
1243             /* Update vectorial force */
1244             fix2             = _mm256_add_ps(fix2,tx);
1245             fiy2             = _mm256_add_ps(fiy2,ty);
1246             fiz2             = _mm256_add_ps(fiz2,tz);
1247
1248             fjx2             = _mm256_add_ps(fjx2,tx);
1249             fjy2             = _mm256_add_ps(fjy2,ty);
1250             fjz2             = _mm256_add_ps(fjz2,tz);
1251
1252             }
1253
1254             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1255             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1256             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1257             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1258             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1259             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1260             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1261             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1262
1263             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1264                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1265
1266             /* Inner loop uses 361 flops */
1267         }
1268
1269         /* End of innermost loop */
1270
1271         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1272                                                  f+i_coord_offset,fshift+i_shift_offset);
1273
1274         ggid                        = gid[iidx];
1275         /* Update potential energies */
1276         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1277         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1278
1279         /* Increment number of inner iterations */
1280         inneriter                  += j_index_end - j_index_start;
1281
1282         /* Outer loop uses 20 flops */
1283     }
1284
1285     /* Increment number of outer iterations */
1286     outeriter        += nri;
1287
1288     /* Update outer/inner flops */
1289
1290     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*361);
1291 }
1292 /*
1293  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_F_avx_256_single
1294  * Electrostatics interaction: ReactionField
1295  * VdW interaction:            CubicSplineTable
1296  * Geometry:                   Water3-Water3
1297  * Calculate force/pot:        Force
1298  */
1299 void
1300 nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_F_avx_256_single
1301                     (t_nblist                    * gmx_restrict       nlist,
1302                      rvec                        * gmx_restrict          xx,
1303                      rvec                        * gmx_restrict          ff,
1304                      t_forcerec                  * gmx_restrict          fr,
1305                      t_mdatoms                   * gmx_restrict     mdatoms,
1306                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1307                      t_nrnb                      * gmx_restrict        nrnb)
1308 {
1309     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1310      * just 0 for non-waters.
1311      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1312      * jnr indices corresponding to data put in the four positions in the SIMD register.
1313      */
1314     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1315     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1316     int              jnrA,jnrB,jnrC,jnrD;
1317     int              jnrE,jnrF,jnrG,jnrH;
1318     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1319     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1320     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1321     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1322     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1323     real             rcutoff_scalar;
1324     real             *shiftvec,*fshift,*x,*f;
1325     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1326     real             scratch[4*DIM];
1327     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1328     real *           vdwioffsetptr0;
1329     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1330     real *           vdwioffsetptr1;
1331     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1332     real *           vdwioffsetptr2;
1333     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1334     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1335     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1336     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1337     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1338     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1339     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1340     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1341     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1342     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1343     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1344     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1345     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1346     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1347     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1348     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1349     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1350     real             *charge;
1351     int              nvdwtype;
1352     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1353     int              *vdwtype;
1354     real             *vdwparam;
1355     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1356     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1357     __m256i          vfitab;
1358     __m128i          vfitab_lo,vfitab_hi;
1359     __m128i          ifour       = _mm_set1_epi32(4);
1360     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1361     real             *vftab;
1362     __m256           dummy_mask,cutoff_mask;
1363     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1364     __m256           one     = _mm256_set1_ps(1.0);
1365     __m256           two     = _mm256_set1_ps(2.0);
1366     x                = xx[0];
1367     f                = ff[0];
1368
1369     nri              = nlist->nri;
1370     iinr             = nlist->iinr;
1371     jindex           = nlist->jindex;
1372     jjnr             = nlist->jjnr;
1373     shiftidx         = nlist->shift;
1374     gid              = nlist->gid;
1375     shiftvec         = fr->shift_vec[0];
1376     fshift           = fr->fshift[0];
1377     facel            = _mm256_set1_ps(fr->epsfac);
1378     charge           = mdatoms->chargeA;
1379     krf              = _mm256_set1_ps(fr->ic->k_rf);
1380     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
1381     crf              = _mm256_set1_ps(fr->ic->c_rf);
1382     nvdwtype         = fr->ntype;
1383     vdwparam         = fr->nbfp;
1384     vdwtype          = mdatoms->typeA;
1385
1386     vftab            = kernel_data->table_vdw->data;
1387     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
1388
1389     /* Setup water-specific parameters */
1390     inr              = nlist->iinr[0];
1391     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1392     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1393     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1394     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1395
1396     jq0              = _mm256_set1_ps(charge[inr+0]);
1397     jq1              = _mm256_set1_ps(charge[inr+1]);
1398     jq2              = _mm256_set1_ps(charge[inr+2]);
1399     vdwjidx0A        = 2*vdwtype[inr+0];
1400     qq00             = _mm256_mul_ps(iq0,jq0);
1401     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1402     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1403     qq01             = _mm256_mul_ps(iq0,jq1);
1404     qq02             = _mm256_mul_ps(iq0,jq2);
1405     qq10             = _mm256_mul_ps(iq1,jq0);
1406     qq11             = _mm256_mul_ps(iq1,jq1);
1407     qq12             = _mm256_mul_ps(iq1,jq2);
1408     qq20             = _mm256_mul_ps(iq2,jq0);
1409     qq21             = _mm256_mul_ps(iq2,jq1);
1410     qq22             = _mm256_mul_ps(iq2,jq2);
1411
1412     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1413     rcutoff_scalar   = fr->rcoulomb;
1414     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1415     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1416
1417     /* Avoid stupid compiler warnings */
1418     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1419     j_coord_offsetA = 0;
1420     j_coord_offsetB = 0;
1421     j_coord_offsetC = 0;
1422     j_coord_offsetD = 0;
1423     j_coord_offsetE = 0;
1424     j_coord_offsetF = 0;
1425     j_coord_offsetG = 0;
1426     j_coord_offsetH = 0;
1427
1428     outeriter        = 0;
1429     inneriter        = 0;
1430
1431     for(iidx=0;iidx<4*DIM;iidx++)
1432     {
1433         scratch[iidx] = 0.0;
1434     }
1435
1436     /* Start outer loop over neighborlists */
1437     for(iidx=0; iidx<nri; iidx++)
1438     {
1439         /* Load shift vector for this list */
1440         i_shift_offset   = DIM*shiftidx[iidx];
1441
1442         /* Load limits for loop over neighbors */
1443         j_index_start    = jindex[iidx];
1444         j_index_end      = jindex[iidx+1];
1445
1446         /* Get outer coordinate index */
1447         inr              = iinr[iidx];
1448         i_coord_offset   = DIM*inr;
1449
1450         /* Load i particle coords and add shift vector */
1451         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1452                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1453
1454         fix0             = _mm256_setzero_ps();
1455         fiy0             = _mm256_setzero_ps();
1456         fiz0             = _mm256_setzero_ps();
1457         fix1             = _mm256_setzero_ps();
1458         fiy1             = _mm256_setzero_ps();
1459         fiz1             = _mm256_setzero_ps();
1460         fix2             = _mm256_setzero_ps();
1461         fiy2             = _mm256_setzero_ps();
1462         fiz2             = _mm256_setzero_ps();
1463
1464         /* Start inner kernel loop */
1465         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1466         {
1467
1468             /* Get j neighbor index, and coordinate index */
1469             jnrA             = jjnr[jidx];
1470             jnrB             = jjnr[jidx+1];
1471             jnrC             = jjnr[jidx+2];
1472             jnrD             = jjnr[jidx+3];
1473             jnrE             = jjnr[jidx+4];
1474             jnrF             = jjnr[jidx+5];
1475             jnrG             = jjnr[jidx+6];
1476             jnrH             = jjnr[jidx+7];
1477             j_coord_offsetA  = DIM*jnrA;
1478             j_coord_offsetB  = DIM*jnrB;
1479             j_coord_offsetC  = DIM*jnrC;
1480             j_coord_offsetD  = DIM*jnrD;
1481             j_coord_offsetE  = DIM*jnrE;
1482             j_coord_offsetF  = DIM*jnrF;
1483             j_coord_offsetG  = DIM*jnrG;
1484             j_coord_offsetH  = DIM*jnrH;
1485
1486             /* load j atom coordinates */
1487             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1488                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1489                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1490                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1491                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1492
1493             /* Calculate displacement vector */
1494             dx00             = _mm256_sub_ps(ix0,jx0);
1495             dy00             = _mm256_sub_ps(iy0,jy0);
1496             dz00             = _mm256_sub_ps(iz0,jz0);
1497             dx01             = _mm256_sub_ps(ix0,jx1);
1498             dy01             = _mm256_sub_ps(iy0,jy1);
1499             dz01             = _mm256_sub_ps(iz0,jz1);
1500             dx02             = _mm256_sub_ps(ix0,jx2);
1501             dy02             = _mm256_sub_ps(iy0,jy2);
1502             dz02             = _mm256_sub_ps(iz0,jz2);
1503             dx10             = _mm256_sub_ps(ix1,jx0);
1504             dy10             = _mm256_sub_ps(iy1,jy0);
1505             dz10             = _mm256_sub_ps(iz1,jz0);
1506             dx11             = _mm256_sub_ps(ix1,jx1);
1507             dy11             = _mm256_sub_ps(iy1,jy1);
1508             dz11             = _mm256_sub_ps(iz1,jz1);
1509             dx12             = _mm256_sub_ps(ix1,jx2);
1510             dy12             = _mm256_sub_ps(iy1,jy2);
1511             dz12             = _mm256_sub_ps(iz1,jz2);
1512             dx20             = _mm256_sub_ps(ix2,jx0);
1513             dy20             = _mm256_sub_ps(iy2,jy0);
1514             dz20             = _mm256_sub_ps(iz2,jz0);
1515             dx21             = _mm256_sub_ps(ix2,jx1);
1516             dy21             = _mm256_sub_ps(iy2,jy1);
1517             dz21             = _mm256_sub_ps(iz2,jz1);
1518             dx22             = _mm256_sub_ps(ix2,jx2);
1519             dy22             = _mm256_sub_ps(iy2,jy2);
1520             dz22             = _mm256_sub_ps(iz2,jz2);
1521
1522             /* Calculate squared distance and things based on it */
1523             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1524             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1525             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1526             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1527             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1528             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1529             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1530             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1531             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1532
1533             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1534             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1535             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1536             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1537             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1538             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1539             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1540             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1541             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1542
1543             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1544             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1545             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1546             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1547             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1548             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1549             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1550             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1551             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1552
1553             fjx0             = _mm256_setzero_ps();
1554             fjy0             = _mm256_setzero_ps();
1555             fjz0             = _mm256_setzero_ps();
1556             fjx1             = _mm256_setzero_ps();
1557             fjy1             = _mm256_setzero_ps();
1558             fjz1             = _mm256_setzero_ps();
1559             fjx2             = _mm256_setzero_ps();
1560             fjy2             = _mm256_setzero_ps();
1561             fjz2             = _mm256_setzero_ps();
1562
1563             /**************************
1564              * CALCULATE INTERACTIONS *
1565              **************************/
1566
1567             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1568             {
1569
1570             r00              = _mm256_mul_ps(rsq00,rinv00);
1571
1572             /* Calculate table index by multiplying r with table scale and truncate to integer */
1573             rt               = _mm256_mul_ps(r00,vftabscale);
1574             vfitab           = _mm256_cvttps_epi32(rt);
1575             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1576             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1577             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1578             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1579             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1580             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1581
1582             /* REACTION-FIELD ELECTROSTATICS */
1583             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1584
1585             /* CUBIC SPLINE TABLE DISPERSION */
1586             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1587                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1588             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1589                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1590             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1591                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1592             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1593                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1594             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1595             Heps             = _mm256_mul_ps(vfeps,H);
1596             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1597             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1598             fvdw6            = _mm256_mul_ps(c6_00,FF);
1599
1600             /* CUBIC SPLINE TABLE REPULSION */
1601             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1602             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1603             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1604                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1605             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1606                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1607             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1608                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1609             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1610                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1611             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1612             Heps             = _mm256_mul_ps(vfeps,H);
1613             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1614             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1615             fvdw12           = _mm256_mul_ps(c12_00,FF);
1616             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1617
1618             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1619
1620             fscal            = _mm256_add_ps(felec,fvdw);
1621
1622             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1623
1624             /* Calculate temporary vectorial force */
1625             tx               = _mm256_mul_ps(fscal,dx00);
1626             ty               = _mm256_mul_ps(fscal,dy00);
1627             tz               = _mm256_mul_ps(fscal,dz00);
1628
1629             /* Update vectorial force */
1630             fix0             = _mm256_add_ps(fix0,tx);
1631             fiy0             = _mm256_add_ps(fiy0,ty);
1632             fiz0             = _mm256_add_ps(fiz0,tz);
1633
1634             fjx0             = _mm256_add_ps(fjx0,tx);
1635             fjy0             = _mm256_add_ps(fjy0,ty);
1636             fjz0             = _mm256_add_ps(fjz0,tz);
1637
1638             }
1639
1640             /**************************
1641              * CALCULATE INTERACTIONS *
1642              **************************/
1643
1644             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1645             {
1646
1647             /* REACTION-FIELD ELECTROSTATICS */
1648             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1649
1650             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
1651
1652             fscal            = felec;
1653
1654             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1655
1656             /* Calculate temporary vectorial force */
1657             tx               = _mm256_mul_ps(fscal,dx01);
1658             ty               = _mm256_mul_ps(fscal,dy01);
1659             tz               = _mm256_mul_ps(fscal,dz01);
1660
1661             /* Update vectorial force */
1662             fix0             = _mm256_add_ps(fix0,tx);
1663             fiy0             = _mm256_add_ps(fiy0,ty);
1664             fiz0             = _mm256_add_ps(fiz0,tz);
1665
1666             fjx1             = _mm256_add_ps(fjx1,tx);
1667             fjy1             = _mm256_add_ps(fjy1,ty);
1668             fjz1             = _mm256_add_ps(fjz1,tz);
1669
1670             }
1671
1672             /**************************
1673              * CALCULATE INTERACTIONS *
1674              **************************/
1675
1676             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1677             {
1678
1679             /* REACTION-FIELD ELECTROSTATICS */
1680             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1681
1682             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1683
1684             fscal            = felec;
1685
1686             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1687
1688             /* Calculate temporary vectorial force */
1689             tx               = _mm256_mul_ps(fscal,dx02);
1690             ty               = _mm256_mul_ps(fscal,dy02);
1691             tz               = _mm256_mul_ps(fscal,dz02);
1692
1693             /* Update vectorial force */
1694             fix0             = _mm256_add_ps(fix0,tx);
1695             fiy0             = _mm256_add_ps(fiy0,ty);
1696             fiz0             = _mm256_add_ps(fiz0,tz);
1697
1698             fjx2             = _mm256_add_ps(fjx2,tx);
1699             fjy2             = _mm256_add_ps(fjy2,ty);
1700             fjz2             = _mm256_add_ps(fjz2,tz);
1701
1702             }
1703
1704             /**************************
1705              * CALCULATE INTERACTIONS *
1706              **************************/
1707
1708             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1709             {
1710
1711             /* REACTION-FIELD ELECTROSTATICS */
1712             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1713
1714             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1715
1716             fscal            = felec;
1717
1718             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1719
1720             /* Calculate temporary vectorial force */
1721             tx               = _mm256_mul_ps(fscal,dx10);
1722             ty               = _mm256_mul_ps(fscal,dy10);
1723             tz               = _mm256_mul_ps(fscal,dz10);
1724
1725             /* Update vectorial force */
1726             fix1             = _mm256_add_ps(fix1,tx);
1727             fiy1             = _mm256_add_ps(fiy1,ty);
1728             fiz1             = _mm256_add_ps(fiz1,tz);
1729
1730             fjx0             = _mm256_add_ps(fjx0,tx);
1731             fjy0             = _mm256_add_ps(fjy0,ty);
1732             fjz0             = _mm256_add_ps(fjz0,tz);
1733
1734             }
1735
1736             /**************************
1737              * CALCULATE INTERACTIONS *
1738              **************************/
1739
1740             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1741             {
1742
1743             /* REACTION-FIELD ELECTROSTATICS */
1744             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1745
1746             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1747
1748             fscal            = felec;
1749
1750             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1751
1752             /* Calculate temporary vectorial force */
1753             tx               = _mm256_mul_ps(fscal,dx11);
1754             ty               = _mm256_mul_ps(fscal,dy11);
1755             tz               = _mm256_mul_ps(fscal,dz11);
1756
1757             /* Update vectorial force */
1758             fix1             = _mm256_add_ps(fix1,tx);
1759             fiy1             = _mm256_add_ps(fiy1,ty);
1760             fiz1             = _mm256_add_ps(fiz1,tz);
1761
1762             fjx1             = _mm256_add_ps(fjx1,tx);
1763             fjy1             = _mm256_add_ps(fjy1,ty);
1764             fjz1             = _mm256_add_ps(fjz1,tz);
1765
1766             }
1767
1768             /**************************
1769              * CALCULATE INTERACTIONS *
1770              **************************/
1771
1772             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1773             {
1774
1775             /* REACTION-FIELD ELECTROSTATICS */
1776             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1777
1778             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1779
1780             fscal            = felec;
1781
1782             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1783
1784             /* Calculate temporary vectorial force */
1785             tx               = _mm256_mul_ps(fscal,dx12);
1786             ty               = _mm256_mul_ps(fscal,dy12);
1787             tz               = _mm256_mul_ps(fscal,dz12);
1788
1789             /* Update vectorial force */
1790             fix1             = _mm256_add_ps(fix1,tx);
1791             fiy1             = _mm256_add_ps(fiy1,ty);
1792             fiz1             = _mm256_add_ps(fiz1,tz);
1793
1794             fjx2             = _mm256_add_ps(fjx2,tx);
1795             fjy2             = _mm256_add_ps(fjy2,ty);
1796             fjz2             = _mm256_add_ps(fjz2,tz);
1797
1798             }
1799
1800             /**************************
1801              * CALCULATE INTERACTIONS *
1802              **************************/
1803
1804             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1805             {
1806
1807             /* REACTION-FIELD ELECTROSTATICS */
1808             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1809
1810             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1811
1812             fscal            = felec;
1813
1814             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1815
1816             /* Calculate temporary vectorial force */
1817             tx               = _mm256_mul_ps(fscal,dx20);
1818             ty               = _mm256_mul_ps(fscal,dy20);
1819             tz               = _mm256_mul_ps(fscal,dz20);
1820
1821             /* Update vectorial force */
1822             fix2             = _mm256_add_ps(fix2,tx);
1823             fiy2             = _mm256_add_ps(fiy2,ty);
1824             fiz2             = _mm256_add_ps(fiz2,tz);
1825
1826             fjx0             = _mm256_add_ps(fjx0,tx);
1827             fjy0             = _mm256_add_ps(fjy0,ty);
1828             fjz0             = _mm256_add_ps(fjz0,tz);
1829
1830             }
1831
1832             /**************************
1833              * CALCULATE INTERACTIONS *
1834              **************************/
1835
1836             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1837             {
1838
1839             /* REACTION-FIELD ELECTROSTATICS */
1840             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1841
1842             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1843
1844             fscal            = felec;
1845
1846             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1847
1848             /* Calculate temporary vectorial force */
1849             tx               = _mm256_mul_ps(fscal,dx21);
1850             ty               = _mm256_mul_ps(fscal,dy21);
1851             tz               = _mm256_mul_ps(fscal,dz21);
1852
1853             /* Update vectorial force */
1854             fix2             = _mm256_add_ps(fix2,tx);
1855             fiy2             = _mm256_add_ps(fiy2,ty);
1856             fiz2             = _mm256_add_ps(fiz2,tz);
1857
1858             fjx1             = _mm256_add_ps(fjx1,tx);
1859             fjy1             = _mm256_add_ps(fjy1,ty);
1860             fjz1             = _mm256_add_ps(fjz1,tz);
1861
1862             }
1863
1864             /**************************
1865              * CALCULATE INTERACTIONS *
1866              **************************/
1867
1868             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1869             {
1870
1871             /* REACTION-FIELD ELECTROSTATICS */
1872             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1873
1874             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1875
1876             fscal            = felec;
1877
1878             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1879
1880             /* Calculate temporary vectorial force */
1881             tx               = _mm256_mul_ps(fscal,dx22);
1882             ty               = _mm256_mul_ps(fscal,dy22);
1883             tz               = _mm256_mul_ps(fscal,dz22);
1884
1885             /* Update vectorial force */
1886             fix2             = _mm256_add_ps(fix2,tx);
1887             fiy2             = _mm256_add_ps(fiy2,ty);
1888             fiz2             = _mm256_add_ps(fiz2,tz);
1889
1890             fjx2             = _mm256_add_ps(fjx2,tx);
1891             fjy2             = _mm256_add_ps(fjy2,ty);
1892             fjz2             = _mm256_add_ps(fjz2,tz);
1893
1894             }
1895
1896             fjptrA             = f+j_coord_offsetA;
1897             fjptrB             = f+j_coord_offsetB;
1898             fjptrC             = f+j_coord_offsetC;
1899             fjptrD             = f+j_coord_offsetD;
1900             fjptrE             = f+j_coord_offsetE;
1901             fjptrF             = f+j_coord_offsetF;
1902             fjptrG             = f+j_coord_offsetG;
1903             fjptrH             = f+j_coord_offsetH;
1904
1905             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1906                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1907
1908             /* Inner loop uses 297 flops */
1909         }
1910
1911         if(jidx<j_index_end)
1912         {
1913
1914             /* Get j neighbor index, and coordinate index */
1915             jnrlistA         = jjnr[jidx];
1916             jnrlistB         = jjnr[jidx+1];
1917             jnrlistC         = jjnr[jidx+2];
1918             jnrlistD         = jjnr[jidx+3];
1919             jnrlistE         = jjnr[jidx+4];
1920             jnrlistF         = jjnr[jidx+5];
1921             jnrlistG         = jjnr[jidx+6];
1922             jnrlistH         = jjnr[jidx+7];
1923             /* Sign of each element will be negative for non-real atoms.
1924              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1925              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1926              */
1927             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1928                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1929
1930             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1931             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1932             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1933             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1934             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1935             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1936             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1937             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1938             j_coord_offsetA  = DIM*jnrA;
1939             j_coord_offsetB  = DIM*jnrB;
1940             j_coord_offsetC  = DIM*jnrC;
1941             j_coord_offsetD  = DIM*jnrD;
1942             j_coord_offsetE  = DIM*jnrE;
1943             j_coord_offsetF  = DIM*jnrF;
1944             j_coord_offsetG  = DIM*jnrG;
1945             j_coord_offsetH  = DIM*jnrH;
1946
1947             /* load j atom coordinates */
1948             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1949                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1950                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1951                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1952                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1953
1954             /* Calculate displacement vector */
1955             dx00             = _mm256_sub_ps(ix0,jx0);
1956             dy00             = _mm256_sub_ps(iy0,jy0);
1957             dz00             = _mm256_sub_ps(iz0,jz0);
1958             dx01             = _mm256_sub_ps(ix0,jx1);
1959             dy01             = _mm256_sub_ps(iy0,jy1);
1960             dz01             = _mm256_sub_ps(iz0,jz1);
1961             dx02             = _mm256_sub_ps(ix0,jx2);
1962             dy02             = _mm256_sub_ps(iy0,jy2);
1963             dz02             = _mm256_sub_ps(iz0,jz2);
1964             dx10             = _mm256_sub_ps(ix1,jx0);
1965             dy10             = _mm256_sub_ps(iy1,jy0);
1966             dz10             = _mm256_sub_ps(iz1,jz0);
1967             dx11             = _mm256_sub_ps(ix1,jx1);
1968             dy11             = _mm256_sub_ps(iy1,jy1);
1969             dz11             = _mm256_sub_ps(iz1,jz1);
1970             dx12             = _mm256_sub_ps(ix1,jx2);
1971             dy12             = _mm256_sub_ps(iy1,jy2);
1972             dz12             = _mm256_sub_ps(iz1,jz2);
1973             dx20             = _mm256_sub_ps(ix2,jx0);
1974             dy20             = _mm256_sub_ps(iy2,jy0);
1975             dz20             = _mm256_sub_ps(iz2,jz0);
1976             dx21             = _mm256_sub_ps(ix2,jx1);
1977             dy21             = _mm256_sub_ps(iy2,jy1);
1978             dz21             = _mm256_sub_ps(iz2,jz1);
1979             dx22             = _mm256_sub_ps(ix2,jx2);
1980             dy22             = _mm256_sub_ps(iy2,jy2);
1981             dz22             = _mm256_sub_ps(iz2,jz2);
1982
1983             /* Calculate squared distance and things based on it */
1984             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1985             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1986             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1987             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1988             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1989             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1990             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1991             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1992             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1993
1994             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1995             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1996             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1997             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1998             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1999             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2000             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
2001             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2002             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2003
2004             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
2005             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
2006             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
2007             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
2008             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2009             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2010             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
2011             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2012             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2013
2014             fjx0             = _mm256_setzero_ps();
2015             fjy0             = _mm256_setzero_ps();
2016             fjz0             = _mm256_setzero_ps();
2017             fjx1             = _mm256_setzero_ps();
2018             fjy1             = _mm256_setzero_ps();
2019             fjz1             = _mm256_setzero_ps();
2020             fjx2             = _mm256_setzero_ps();
2021             fjy2             = _mm256_setzero_ps();
2022             fjz2             = _mm256_setzero_ps();
2023
2024             /**************************
2025              * CALCULATE INTERACTIONS *
2026              **************************/
2027
2028             if (gmx_mm256_any_lt(rsq00,rcutoff2))
2029             {
2030
2031             r00              = _mm256_mul_ps(rsq00,rinv00);
2032             r00              = _mm256_andnot_ps(dummy_mask,r00);
2033
2034             /* Calculate table index by multiplying r with table scale and truncate to integer */
2035             rt               = _mm256_mul_ps(r00,vftabscale);
2036             vfitab           = _mm256_cvttps_epi32(rt);
2037             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2038             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2039             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2040             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2041             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
2042             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
2043
2044             /* REACTION-FIELD ELECTROSTATICS */
2045             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
2046
2047             /* CUBIC SPLINE TABLE DISPERSION */
2048             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2049                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2050             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2051                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2052             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2053                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2054             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2055                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2056             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2057             Heps             = _mm256_mul_ps(vfeps,H);
2058             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2059             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2060             fvdw6            = _mm256_mul_ps(c6_00,FF);
2061
2062             /* CUBIC SPLINE TABLE REPULSION */
2063             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
2064             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
2065             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2066                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2067             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2068                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2069             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2070                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2071             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2072                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2073             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2074             Heps             = _mm256_mul_ps(vfeps,H);
2075             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2076             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2077             fvdw12           = _mm256_mul_ps(c12_00,FF);
2078             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
2079
2080             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
2081
2082             fscal            = _mm256_add_ps(felec,fvdw);
2083
2084             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2085
2086             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2087
2088             /* Calculate temporary vectorial force */
2089             tx               = _mm256_mul_ps(fscal,dx00);
2090             ty               = _mm256_mul_ps(fscal,dy00);
2091             tz               = _mm256_mul_ps(fscal,dz00);
2092
2093             /* Update vectorial force */
2094             fix0             = _mm256_add_ps(fix0,tx);
2095             fiy0             = _mm256_add_ps(fiy0,ty);
2096             fiz0             = _mm256_add_ps(fiz0,tz);
2097
2098             fjx0             = _mm256_add_ps(fjx0,tx);
2099             fjy0             = _mm256_add_ps(fjy0,ty);
2100             fjz0             = _mm256_add_ps(fjz0,tz);
2101
2102             }
2103
2104             /**************************
2105              * CALCULATE INTERACTIONS *
2106              **************************/
2107
2108             if (gmx_mm256_any_lt(rsq01,rcutoff2))
2109             {
2110
2111             /* REACTION-FIELD ELECTROSTATICS */
2112             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
2113
2114             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
2115
2116             fscal            = felec;
2117
2118             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2119
2120             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2121
2122             /* Calculate temporary vectorial force */
2123             tx               = _mm256_mul_ps(fscal,dx01);
2124             ty               = _mm256_mul_ps(fscal,dy01);
2125             tz               = _mm256_mul_ps(fscal,dz01);
2126
2127             /* Update vectorial force */
2128             fix0             = _mm256_add_ps(fix0,tx);
2129             fiy0             = _mm256_add_ps(fiy0,ty);
2130             fiz0             = _mm256_add_ps(fiz0,tz);
2131
2132             fjx1             = _mm256_add_ps(fjx1,tx);
2133             fjy1             = _mm256_add_ps(fjy1,ty);
2134             fjz1             = _mm256_add_ps(fjz1,tz);
2135
2136             }
2137
2138             /**************************
2139              * CALCULATE INTERACTIONS *
2140              **************************/
2141
2142             if (gmx_mm256_any_lt(rsq02,rcutoff2))
2143             {
2144
2145             /* REACTION-FIELD ELECTROSTATICS */
2146             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
2147
2148             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
2149
2150             fscal            = felec;
2151
2152             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2153
2154             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2155
2156             /* Calculate temporary vectorial force */
2157             tx               = _mm256_mul_ps(fscal,dx02);
2158             ty               = _mm256_mul_ps(fscal,dy02);
2159             tz               = _mm256_mul_ps(fscal,dz02);
2160
2161             /* Update vectorial force */
2162             fix0             = _mm256_add_ps(fix0,tx);
2163             fiy0             = _mm256_add_ps(fiy0,ty);
2164             fiz0             = _mm256_add_ps(fiz0,tz);
2165
2166             fjx2             = _mm256_add_ps(fjx2,tx);
2167             fjy2             = _mm256_add_ps(fjy2,ty);
2168             fjz2             = _mm256_add_ps(fjz2,tz);
2169
2170             }
2171
2172             /**************************
2173              * CALCULATE INTERACTIONS *
2174              **************************/
2175
2176             if (gmx_mm256_any_lt(rsq10,rcutoff2))
2177             {
2178
2179             /* REACTION-FIELD ELECTROSTATICS */
2180             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
2181
2182             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
2183
2184             fscal            = felec;
2185
2186             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2187
2188             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2189
2190             /* Calculate temporary vectorial force */
2191             tx               = _mm256_mul_ps(fscal,dx10);
2192             ty               = _mm256_mul_ps(fscal,dy10);
2193             tz               = _mm256_mul_ps(fscal,dz10);
2194
2195             /* Update vectorial force */
2196             fix1             = _mm256_add_ps(fix1,tx);
2197             fiy1             = _mm256_add_ps(fiy1,ty);
2198             fiz1             = _mm256_add_ps(fiz1,tz);
2199
2200             fjx0             = _mm256_add_ps(fjx0,tx);
2201             fjy0             = _mm256_add_ps(fjy0,ty);
2202             fjz0             = _mm256_add_ps(fjz0,tz);
2203
2204             }
2205
2206             /**************************
2207              * CALCULATE INTERACTIONS *
2208              **************************/
2209
2210             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2211             {
2212
2213             /* REACTION-FIELD ELECTROSTATICS */
2214             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
2215
2216             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2217
2218             fscal            = felec;
2219
2220             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2221
2222             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2223
2224             /* Calculate temporary vectorial force */
2225             tx               = _mm256_mul_ps(fscal,dx11);
2226             ty               = _mm256_mul_ps(fscal,dy11);
2227             tz               = _mm256_mul_ps(fscal,dz11);
2228
2229             /* Update vectorial force */
2230             fix1             = _mm256_add_ps(fix1,tx);
2231             fiy1             = _mm256_add_ps(fiy1,ty);
2232             fiz1             = _mm256_add_ps(fiz1,tz);
2233
2234             fjx1             = _mm256_add_ps(fjx1,tx);
2235             fjy1             = _mm256_add_ps(fjy1,ty);
2236             fjz1             = _mm256_add_ps(fjz1,tz);
2237
2238             }
2239
2240             /**************************
2241              * CALCULATE INTERACTIONS *
2242              **************************/
2243
2244             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2245             {
2246
2247             /* REACTION-FIELD ELECTROSTATICS */
2248             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
2249
2250             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2251
2252             fscal            = felec;
2253
2254             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2255
2256             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2257
2258             /* Calculate temporary vectorial force */
2259             tx               = _mm256_mul_ps(fscal,dx12);
2260             ty               = _mm256_mul_ps(fscal,dy12);
2261             tz               = _mm256_mul_ps(fscal,dz12);
2262
2263             /* Update vectorial force */
2264             fix1             = _mm256_add_ps(fix1,tx);
2265             fiy1             = _mm256_add_ps(fiy1,ty);
2266             fiz1             = _mm256_add_ps(fiz1,tz);
2267
2268             fjx2             = _mm256_add_ps(fjx2,tx);
2269             fjy2             = _mm256_add_ps(fjy2,ty);
2270             fjz2             = _mm256_add_ps(fjz2,tz);
2271
2272             }
2273
2274             /**************************
2275              * CALCULATE INTERACTIONS *
2276              **************************/
2277
2278             if (gmx_mm256_any_lt(rsq20,rcutoff2))
2279             {
2280
2281             /* REACTION-FIELD ELECTROSTATICS */
2282             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
2283
2284             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
2285
2286             fscal            = felec;
2287
2288             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2289
2290             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2291
2292             /* Calculate temporary vectorial force */
2293             tx               = _mm256_mul_ps(fscal,dx20);
2294             ty               = _mm256_mul_ps(fscal,dy20);
2295             tz               = _mm256_mul_ps(fscal,dz20);
2296
2297             /* Update vectorial force */
2298             fix2             = _mm256_add_ps(fix2,tx);
2299             fiy2             = _mm256_add_ps(fiy2,ty);
2300             fiz2             = _mm256_add_ps(fiz2,tz);
2301
2302             fjx0             = _mm256_add_ps(fjx0,tx);
2303             fjy0             = _mm256_add_ps(fjy0,ty);
2304             fjz0             = _mm256_add_ps(fjz0,tz);
2305
2306             }
2307
2308             /**************************
2309              * CALCULATE INTERACTIONS *
2310              **************************/
2311
2312             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2313             {
2314
2315             /* REACTION-FIELD ELECTROSTATICS */
2316             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
2317
2318             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2319
2320             fscal            = felec;
2321
2322             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2323
2324             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2325
2326             /* Calculate temporary vectorial force */
2327             tx               = _mm256_mul_ps(fscal,dx21);
2328             ty               = _mm256_mul_ps(fscal,dy21);
2329             tz               = _mm256_mul_ps(fscal,dz21);
2330
2331             /* Update vectorial force */
2332             fix2             = _mm256_add_ps(fix2,tx);
2333             fiy2             = _mm256_add_ps(fiy2,ty);
2334             fiz2             = _mm256_add_ps(fiz2,tz);
2335
2336             fjx1             = _mm256_add_ps(fjx1,tx);
2337             fjy1             = _mm256_add_ps(fjy1,ty);
2338             fjz1             = _mm256_add_ps(fjz1,tz);
2339
2340             }
2341
2342             /**************************
2343              * CALCULATE INTERACTIONS *
2344              **************************/
2345
2346             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2347             {
2348
2349             /* REACTION-FIELD ELECTROSTATICS */
2350             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
2351
2352             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2353
2354             fscal            = felec;
2355
2356             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2357
2358             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2359
2360             /* Calculate temporary vectorial force */
2361             tx               = _mm256_mul_ps(fscal,dx22);
2362             ty               = _mm256_mul_ps(fscal,dy22);
2363             tz               = _mm256_mul_ps(fscal,dz22);
2364
2365             /* Update vectorial force */
2366             fix2             = _mm256_add_ps(fix2,tx);
2367             fiy2             = _mm256_add_ps(fiy2,ty);
2368             fiz2             = _mm256_add_ps(fiz2,tz);
2369
2370             fjx2             = _mm256_add_ps(fjx2,tx);
2371             fjy2             = _mm256_add_ps(fjy2,ty);
2372             fjz2             = _mm256_add_ps(fjz2,tz);
2373
2374             }
2375
2376             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2377             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2378             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2379             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2380             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2381             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2382             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2383             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2384
2385             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2386                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2387
2388             /* Inner loop uses 298 flops */
2389         }
2390
2391         /* End of innermost loop */
2392
2393         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2394                                                  f+i_coord_offset,fshift+i_shift_offset);
2395
2396         /* Increment number of inner iterations */
2397         inneriter                  += j_index_end - j_index_start;
2398
2399         /* Outer loop uses 18 flops */
2400     }
2401
2402     /* Increment number of outer iterations */
2403     outeriter        += nri;
2404
2405     /* Update outer/inner flops */
2406
2407     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*298);
2408 }