src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "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_ElecCoul_VdwCSTab_GeomW3W3_VF_avx_256_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_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     nvdwtype         = fr->ntype;
 140     vdwparam         = fr->nbfp;
 141     vdwtype          = mdatoms->typeA;
 142
 143     vftab            = kernel_data->table_vdw->data;
 144     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 145
 146     /* Setup water-specific parameters */
 147     inr              = nlist->iinr[0];
 148     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 149     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 150     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 151     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 152
 153     jq0              = _mm256_set1_ps(charge[inr+0]);
 154     jq1              = _mm256_set1_ps(charge[inr+1]);
 155     jq2              = _mm256_set1_ps(charge[inr+2]);
 156     vdwjidx0A        = 2*vdwtype[inr+0];
 157     qq00             = _mm256_mul_ps(iq0,jq0);
 158     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 159     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 160     qq01             = _mm256_mul_ps(iq0,jq1);
 161     qq02             = _mm256_mul_ps(iq0,jq2);
 162     qq10             = _mm256_mul_ps(iq1,jq0);
 163     qq11             = _mm256_mul_ps(iq1,jq1);
 164     qq12             = _mm256_mul_ps(iq1,jq2);
 165     qq20             = _mm256_mul_ps(iq2,jq0);
 166     qq21             = _mm256_mul_ps(iq2,jq1);
 167     qq22             = _mm256_mul_ps(iq2,jq2);
 168
 169     /* Avoid stupid compiler warnings */
 170     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 171     j_coord_offsetA = 0;
 172     j_coord_offsetB = 0;
 173     j_coord_offsetC = 0;
 174     j_coord_offsetD = 0;
 175     j_coord_offsetE = 0;
 176     j_coord_offsetF = 0;
 177     j_coord_offsetG = 0;
 178     j_coord_offsetH = 0;
 179
 180     outeriter        = 0;
 181     inneriter        = 0;
 182
 183     for(iidx=0;iidx<4*DIM;iidx++)
 184     {
 185         scratch[iidx] = 0.0;
 186     }
 187
 188     /* Start outer loop over neighborlists */
 189     for(iidx=0; iidx<nri; iidx++)
 190     {
 191         /* Load shift vector for this list */
 192         i_shift_offset   = DIM*shiftidx[iidx];
 193
 194         /* Load limits for loop over neighbors */
 195         j_index_start    = jindex[iidx];
 196         j_index_end      = jindex[iidx+1];
 197
 198         /* Get outer coordinate index */
 199         inr              = iinr[iidx];
 200         i_coord_offset   = DIM*inr;
 201
 202         /* Load i particle coords and add shift vector */
 203         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 204                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 205
 206         fix0             = _mm256_setzero_ps();
 207         fiy0             = _mm256_setzero_ps();
 208         fiz0             = _mm256_setzero_ps();
 209         fix1             = _mm256_setzero_ps();
 210         fiy1             = _mm256_setzero_ps();
 211         fiz1             = _mm256_setzero_ps();
 212         fix2             = _mm256_setzero_ps();
 213         fiy2             = _mm256_setzero_ps();
 214         fiz2             = _mm256_setzero_ps();
 215
 216         /* Reset potential sums */
 217         velecsum         = _mm256_setzero_ps();
 218         vvdwsum          = _mm256_setzero_ps();
 219
 220         /* Start inner kernel loop */
 221         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 222         {
 223
 224             /* Get j neighbor index, and coordinate index */
 225             jnrA             = jjnr[jidx];
 226             jnrB             = jjnr[jidx+1];
 227             jnrC             = jjnr[jidx+2];
 228             jnrD             = jjnr[jidx+3];
 229             jnrE             = jjnr[jidx+4];
 230             jnrF             = jjnr[jidx+5];
 231             jnrG             = jjnr[jidx+6];
 232             jnrH             = jjnr[jidx+7];
 233             j_coord_offsetA  = DIM*jnrA;
 234             j_coord_offsetB  = DIM*jnrB;
 235             j_coord_offsetC  = DIM*jnrC;
 236             j_coord_offsetD  = DIM*jnrD;
 237             j_coord_offsetE  = DIM*jnrE;
 238             j_coord_offsetF  = DIM*jnrF;
 239             j_coord_offsetG  = DIM*jnrG;
 240             j_coord_offsetH  = DIM*jnrH;
 241
 242             /* load j atom coordinates */
 243             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 244                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 245                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 246                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 247                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 248
 249             /* Calculate displacement vector */
 250             dx00             = _mm256_sub_ps(ix0,jx0);
 251             dy00             = _mm256_sub_ps(iy0,jy0);
 252             dz00             = _mm256_sub_ps(iz0,jz0);
 253             dx01             = _mm256_sub_ps(ix0,jx1);
 254             dy01             = _mm256_sub_ps(iy0,jy1);
 255             dz01             = _mm256_sub_ps(iz0,jz1);
 256             dx02             = _mm256_sub_ps(ix0,jx2);
 257             dy02             = _mm256_sub_ps(iy0,jy2);
 258             dz02             = _mm256_sub_ps(iz0,jz2);
 259             dx10             = _mm256_sub_ps(ix1,jx0);
 260             dy10             = _mm256_sub_ps(iy1,jy0);
 261             dz10             = _mm256_sub_ps(iz1,jz0);
 262             dx11             = _mm256_sub_ps(ix1,jx1);
 263             dy11             = _mm256_sub_ps(iy1,jy1);
 264             dz11             = _mm256_sub_ps(iz1,jz1);
 265             dx12             = _mm256_sub_ps(ix1,jx2);
 266             dy12             = _mm256_sub_ps(iy1,jy2);
 267             dz12             = _mm256_sub_ps(iz1,jz2);
 268             dx20             = _mm256_sub_ps(ix2,jx0);
 269             dy20             = _mm256_sub_ps(iy2,jy0);
 270             dz20             = _mm256_sub_ps(iz2,jz0);
 271             dx21             = _mm256_sub_ps(ix2,jx1);
 272             dy21             = _mm256_sub_ps(iy2,jy1);
 273             dz21             = _mm256_sub_ps(iz2,jz1);
 274             dx22             = _mm256_sub_ps(ix2,jx2);
 275             dy22             = _mm256_sub_ps(iy2,jy2);
 276             dz22             = _mm256_sub_ps(iz2,jz2);
 277
 278             /* Calculate squared distance and things based on it */
 279             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 280             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 281             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 282             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 283             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 284             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 285             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 286             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 287             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 288
 289             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 290             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 291             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 292             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 293             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 294             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 295             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 296             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 297             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 298
 299             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 300             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 301             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 302             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 303             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 304             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 305             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 306             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 307             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 308
 309             fjx0             = _mm256_setzero_ps();
 310             fjy0             = _mm256_setzero_ps();
 311             fjz0             = _mm256_setzero_ps();
 312             fjx1             = _mm256_setzero_ps();
 313             fjy1             = _mm256_setzero_ps();
 314             fjz1             = _mm256_setzero_ps();
 315             fjx2             = _mm256_setzero_ps();
 316             fjy2             = _mm256_setzero_ps();
 317             fjz2             = _mm256_setzero_ps();
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             r00              = _mm256_mul_ps(rsq00,rinv00);
 324
 325             /* Calculate table index by multiplying r with table scale and truncate to integer */
 326             rt               = _mm256_mul_ps(r00,vftabscale);
 327             vfitab           = _mm256_cvttps_epi32(rt);
 328             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 329             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 330             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 331             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 332             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 333             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 334
 335             /* COULOMB ELECTROSTATICS */
 336             velec            = _mm256_mul_ps(qq00,rinv00);
 337             felec            = _mm256_mul_ps(velec,rinvsq00);
 338
 339             /* CUBIC SPLINE TABLE DISPERSION */
 340             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 341                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 342             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 343                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 344             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 345                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 346             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 347                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 348             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 349             Heps             = _mm256_mul_ps(vfeps,H);
 350             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 351             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 352             vvdw6            = _mm256_mul_ps(c6_00,VV);
 353             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 354             fvdw6            = _mm256_mul_ps(c6_00,FF);
 355
 356             /* CUBIC SPLINE TABLE REPULSION */
 357             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 358             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 359             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 360                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 361             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 362                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 363             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 364                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 365             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 366                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 367             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 368             Heps             = _mm256_mul_ps(vfeps,H);
 369             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 370             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 371             vvdw12           = _mm256_mul_ps(c12_00,VV);
 372             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 373             fvdw12           = _mm256_mul_ps(c12_00,FF);
 374             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 375             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velecsum         = _mm256_add_ps(velecsum,velec);
 379             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 380
 381             fscal            = _mm256_add_ps(felec,fvdw);
 382
 383             /* Calculate temporary vectorial force */
 384             tx               = _mm256_mul_ps(fscal,dx00);
 385             ty               = _mm256_mul_ps(fscal,dy00);
 386             tz               = _mm256_mul_ps(fscal,dz00);
 387
 388             /* Update vectorial force */
 389             fix0             = _mm256_add_ps(fix0,tx);
 390             fiy0             = _mm256_add_ps(fiy0,ty);
 391             fiz0             = _mm256_add_ps(fiz0,tz);
 392
 393             fjx0             = _mm256_add_ps(fjx0,tx);
 394             fjy0             = _mm256_add_ps(fjy0,ty);
 395             fjz0             = _mm256_add_ps(fjz0,tz);
 396
 397             /**************************
 398              * CALCULATE INTERACTIONS *
 399              **************************/
 400
 401             /* COULOMB ELECTROSTATICS */
 402             velec            = _mm256_mul_ps(qq01,rinv01);
 403             felec            = _mm256_mul_ps(velec,rinvsq01);
 404
 405             /* Update potential sum for this i atom from the interaction with this j atom. */
 406             velecsum         = _mm256_add_ps(velecsum,velec);
 407
 408             fscal            = felec;
 409
 410             /* Calculate temporary vectorial force */
 411             tx               = _mm256_mul_ps(fscal,dx01);
 412             ty               = _mm256_mul_ps(fscal,dy01);
 413             tz               = _mm256_mul_ps(fscal,dz01);
 414
 415             /* Update vectorial force */
 416             fix0             = _mm256_add_ps(fix0,tx);
 417             fiy0             = _mm256_add_ps(fiy0,ty);
 418             fiz0             = _mm256_add_ps(fiz0,tz);
 419
 420             fjx1             = _mm256_add_ps(fjx1,tx);
 421             fjy1             = _mm256_add_ps(fjy1,ty);
 422             fjz1             = _mm256_add_ps(fjz1,tz);
 423
 424             /**************************
 425              * CALCULATE INTERACTIONS *
 426              **************************/
 427
 428             /* COULOMB ELECTROSTATICS */
 429             velec            = _mm256_mul_ps(qq02,rinv02);
 430             felec            = _mm256_mul_ps(velec,rinvsq02);
 431
 432             /* Update potential sum for this i atom from the interaction with this j atom. */
 433             velecsum         = _mm256_add_ps(velecsum,velec);
 434
 435             fscal            = felec;
 436
 437             /* Calculate temporary vectorial force */
 438             tx               = _mm256_mul_ps(fscal,dx02);
 439             ty               = _mm256_mul_ps(fscal,dy02);
 440             tz               = _mm256_mul_ps(fscal,dz02);
 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             fjx2             = _mm256_add_ps(fjx2,tx);
 448             fjy2             = _mm256_add_ps(fjy2,ty);
 449             fjz2             = _mm256_add_ps(fjz2,tz);
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             /* COULOMB ELECTROSTATICS */
 456             velec            = _mm256_mul_ps(qq10,rinv10);
 457             felec            = _mm256_mul_ps(velec,rinvsq10);
 458
 459             /* Update potential sum for this i atom from the interaction with this j atom. */
 460             velecsum         = _mm256_add_ps(velecsum,velec);
 461
 462             fscal            = felec;
 463
 464             /* Calculate temporary vectorial force */
 465             tx               = _mm256_mul_ps(fscal,dx10);
 466             ty               = _mm256_mul_ps(fscal,dy10);
 467             tz               = _mm256_mul_ps(fscal,dz10);
 468
 469             /* Update vectorial force */
 470             fix1             = _mm256_add_ps(fix1,tx);
 471             fiy1             = _mm256_add_ps(fiy1,ty);
 472             fiz1             = _mm256_add_ps(fiz1,tz);
 473
 474             fjx0             = _mm256_add_ps(fjx0,tx);
 475             fjy0             = _mm256_add_ps(fjy0,ty);
 476             fjz0             = _mm256_add_ps(fjz0,tz);
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             /* COULOMB ELECTROSTATICS */
 483             velec            = _mm256_mul_ps(qq11,rinv11);
 484             felec            = _mm256_mul_ps(velec,rinvsq11);
 485
 486             /* Update potential sum for this i atom from the interaction with this j atom. */
 487             velecsum         = _mm256_add_ps(velecsum,velec);
 488
 489             fscal            = felec;
 490
 491             /* Calculate temporary vectorial force */
 492             tx               = _mm256_mul_ps(fscal,dx11);
 493             ty               = _mm256_mul_ps(fscal,dy11);
 494             tz               = _mm256_mul_ps(fscal,dz11);
 495
 496             /* Update vectorial force */
 497             fix1             = _mm256_add_ps(fix1,tx);
 498             fiy1             = _mm256_add_ps(fiy1,ty);
 499             fiz1             = _mm256_add_ps(fiz1,tz);
 500
 501             fjx1             = _mm256_add_ps(fjx1,tx);
 502             fjy1             = _mm256_add_ps(fjy1,ty);
 503             fjz1             = _mm256_add_ps(fjz1,tz);
 504
 505             /**************************
 506              * CALCULATE INTERACTIONS *
 507              **************************/
 508
 509             /* COULOMB ELECTROSTATICS */
 510             velec            = _mm256_mul_ps(qq12,rinv12);
 511             felec            = _mm256_mul_ps(velec,rinvsq12);
 512
 513             /* Update potential sum for this i atom from the interaction with this j atom. */
 514             velecsum         = _mm256_add_ps(velecsum,velec);
 515
 516             fscal            = felec;
 517
 518             /* Calculate temporary vectorial force */
 519             tx               = _mm256_mul_ps(fscal,dx12);
 520             ty               = _mm256_mul_ps(fscal,dy12);
 521             tz               = _mm256_mul_ps(fscal,dz12);
 522
 523             /* Update vectorial force */
 524             fix1             = _mm256_add_ps(fix1,tx);
 525             fiy1             = _mm256_add_ps(fiy1,ty);
 526             fiz1             = _mm256_add_ps(fiz1,tz);
 527
 528             fjx2             = _mm256_add_ps(fjx2,tx);
 529             fjy2             = _mm256_add_ps(fjy2,ty);
 530             fjz2             = _mm256_add_ps(fjz2,tz);
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             /* COULOMB ELECTROSTATICS */
 537             velec            = _mm256_mul_ps(qq20,rinv20);
 538             felec            = _mm256_mul_ps(velec,rinvsq20);
 539
 540             /* Update potential sum for this i atom from the interaction with this j atom. */
 541             velecsum         = _mm256_add_ps(velecsum,velec);
 542
 543             fscal            = felec;
 544
 545             /* Calculate temporary vectorial force */
 546             tx               = _mm256_mul_ps(fscal,dx20);
 547             ty               = _mm256_mul_ps(fscal,dy20);
 548             tz               = _mm256_mul_ps(fscal,dz20);
 549
 550             /* Update vectorial force */
 551             fix2             = _mm256_add_ps(fix2,tx);
 552             fiy2             = _mm256_add_ps(fiy2,ty);
 553             fiz2             = _mm256_add_ps(fiz2,tz);
 554
 555             fjx0             = _mm256_add_ps(fjx0,tx);
 556             fjy0             = _mm256_add_ps(fjy0,ty);
 557             fjz0             = _mm256_add_ps(fjz0,tz);
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             /* COULOMB ELECTROSTATICS */
 564             velec            = _mm256_mul_ps(qq21,rinv21);
 565             felec            = _mm256_mul_ps(velec,rinvsq21);
 566
 567             /* Update potential sum for this i atom from the interaction with this j atom. */
 568             velecsum         = _mm256_add_ps(velecsum,velec);
 569
 570             fscal            = felec;
 571
 572             /* Calculate temporary vectorial force */
 573             tx               = _mm256_mul_ps(fscal,dx21);
 574             ty               = _mm256_mul_ps(fscal,dy21);
 575             tz               = _mm256_mul_ps(fscal,dz21);
 576
 577             /* Update vectorial force */
 578             fix2             = _mm256_add_ps(fix2,tx);
 579             fiy2             = _mm256_add_ps(fiy2,ty);
 580             fiz2             = _mm256_add_ps(fiz2,tz);
 581
 582             fjx1             = _mm256_add_ps(fjx1,tx);
 583             fjy1             = _mm256_add_ps(fjy1,ty);
 584             fjz1             = _mm256_add_ps(fjz1,tz);
 585
 586             /**************************
 587              * CALCULATE INTERACTIONS *
 588              **************************/
 589
 590             /* COULOMB ELECTROSTATICS */
 591             velec            = _mm256_mul_ps(qq22,rinv22);
 592             felec            = _mm256_mul_ps(velec,rinvsq22);
 593
 594             /* Update potential sum for this i atom from the interaction with this j atom. */
 595             velecsum         = _mm256_add_ps(velecsum,velec);
 596
 597             fscal            = felec;
 598
 599             /* Calculate temporary vectorial force */
 600             tx               = _mm256_mul_ps(fscal,dx22);
 601             ty               = _mm256_mul_ps(fscal,dy22);
 602             tz               = _mm256_mul_ps(fscal,dz22);
 603
 604             /* Update vectorial force */
 605             fix2             = _mm256_add_ps(fix2,tx);
 606             fiy2             = _mm256_add_ps(fiy2,ty);
 607             fiz2             = _mm256_add_ps(fiz2,tz);
 608
 609             fjx2             = _mm256_add_ps(fjx2,tx);
 610             fjy2             = _mm256_add_ps(fjy2,ty);
 611             fjz2             = _mm256_add_ps(fjz2,tz);
 612
 613             fjptrA             = f+j_coord_offsetA;
 614             fjptrB             = f+j_coord_offsetB;
 615             fjptrC             = f+j_coord_offsetC;
 616             fjptrD             = f+j_coord_offsetD;
 617             fjptrE             = f+j_coord_offsetE;
 618             fjptrF             = f+j_coord_offsetF;
 619             fjptrG             = f+j_coord_offsetG;
 620             fjptrH             = f+j_coord_offsetH;
 621
 622             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 623                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 624
 625             /* Inner loop uses 278 flops */
 626         }
 627
 628         if(jidx<j_index_end)
 629         {
 630
 631             /* Get j neighbor index, and coordinate index */
 632             jnrlistA         = jjnr[jidx];
 633             jnrlistB         = jjnr[jidx+1];
 634             jnrlistC         = jjnr[jidx+2];
 635             jnrlistD         = jjnr[jidx+3];
 636             jnrlistE         = jjnr[jidx+4];
 637             jnrlistF         = jjnr[jidx+5];
 638             jnrlistG         = jjnr[jidx+6];
 639             jnrlistH         = jjnr[jidx+7];
 640             /* Sign of each element will be negative for non-real atoms.
 641              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 642              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 643              */
 644             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 645                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 646
 647             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 648             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 649             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 650             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 651             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 652             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 653             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 654             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 655             j_coord_offsetA  = DIM*jnrA;
 656             j_coord_offsetB  = DIM*jnrB;
 657             j_coord_offsetC  = DIM*jnrC;
 658             j_coord_offsetD  = DIM*jnrD;
 659             j_coord_offsetE  = DIM*jnrE;
 660             j_coord_offsetF  = DIM*jnrF;
 661             j_coord_offsetG  = DIM*jnrG;
 662             j_coord_offsetH  = DIM*jnrH;
 663
 664             /* load j atom coordinates */
 665             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 666                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 667                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 668                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 669                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 670
 671             /* Calculate displacement vector */
 672             dx00             = _mm256_sub_ps(ix0,jx0);
 673             dy00             = _mm256_sub_ps(iy0,jy0);
 674             dz00             = _mm256_sub_ps(iz0,jz0);
 675             dx01             = _mm256_sub_ps(ix0,jx1);
 676             dy01             = _mm256_sub_ps(iy0,jy1);
 677             dz01             = _mm256_sub_ps(iz0,jz1);
 678             dx02             = _mm256_sub_ps(ix0,jx2);
 679             dy02             = _mm256_sub_ps(iy0,jy2);
 680             dz02             = _mm256_sub_ps(iz0,jz2);
 681             dx10             = _mm256_sub_ps(ix1,jx0);
 682             dy10             = _mm256_sub_ps(iy1,jy0);
 683             dz10             = _mm256_sub_ps(iz1,jz0);
 684             dx11             = _mm256_sub_ps(ix1,jx1);
 685             dy11             = _mm256_sub_ps(iy1,jy1);
 686             dz11             = _mm256_sub_ps(iz1,jz1);
 687             dx12             = _mm256_sub_ps(ix1,jx2);
 688             dy12             = _mm256_sub_ps(iy1,jy2);
 689             dz12             = _mm256_sub_ps(iz1,jz2);
 690             dx20             = _mm256_sub_ps(ix2,jx0);
 691             dy20             = _mm256_sub_ps(iy2,jy0);
 692             dz20             = _mm256_sub_ps(iz2,jz0);
 693             dx21             = _mm256_sub_ps(ix2,jx1);
 694             dy21             = _mm256_sub_ps(iy2,jy1);
 695             dz21             = _mm256_sub_ps(iz2,jz1);
 696             dx22             = _mm256_sub_ps(ix2,jx2);
 697             dy22             = _mm256_sub_ps(iy2,jy2);
 698             dz22             = _mm256_sub_ps(iz2,jz2);
 699
 700             /* Calculate squared distance and things based on it */
 701             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 702             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 703             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 704             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 705             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 706             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 707             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 708             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 709             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 710
 711             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 712             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 713             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 714             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 715             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 716             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 717             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 718             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 719             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 720
 721             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 722             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 723             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 724             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 725             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 726             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 727             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 728             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 729             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 730
 731             fjx0             = _mm256_setzero_ps();
 732             fjy0             = _mm256_setzero_ps();
 733             fjz0             = _mm256_setzero_ps();
 734             fjx1             = _mm256_setzero_ps();
 735             fjy1             = _mm256_setzero_ps();
 736             fjz1             = _mm256_setzero_ps();
 737             fjx2             = _mm256_setzero_ps();
 738             fjy2             = _mm256_setzero_ps();
 739             fjz2             = _mm256_setzero_ps();
 740
 741             /**************************
 742              * CALCULATE INTERACTIONS *
 743              **************************/
 744
 745             r00              = _mm256_mul_ps(rsq00,rinv00);
 746             r00              = _mm256_andnot_ps(dummy_mask,r00);
 747
 748             /* Calculate table index by multiplying r with table scale and truncate to integer */
 749             rt               = _mm256_mul_ps(r00,vftabscale);
 750             vfitab           = _mm256_cvttps_epi32(rt);
 751             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 752             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 753             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 754             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 755             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 756             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 757
 758             /* COULOMB ELECTROSTATICS */
 759             velec            = _mm256_mul_ps(qq00,rinv00);
 760             felec            = _mm256_mul_ps(velec,rinvsq00);
 761
 762             /* CUBIC SPLINE TABLE DISPERSION */
 763             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 764                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 765             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 766                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 767             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 768                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 769             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 770                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 771             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 772             Heps             = _mm256_mul_ps(vfeps,H);
 773             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 774             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 775             vvdw6            = _mm256_mul_ps(c6_00,VV);
 776             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 777             fvdw6            = _mm256_mul_ps(c6_00,FF);
 778
 779             /* CUBIC SPLINE TABLE REPULSION */
 780             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 781             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 782             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 783                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 784             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 785                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 786             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 787                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 788             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 789                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 790             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 791             Heps             = _mm256_mul_ps(vfeps,H);
 792             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 793             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 794             vvdw12           = _mm256_mul_ps(c12_00,VV);
 795             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 796             fvdw12           = _mm256_mul_ps(c12_00,FF);
 797             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 798             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 799
 800             /* Update potential sum for this i atom from the interaction with this j atom. */
 801             velec            = _mm256_andnot_ps(dummy_mask,velec);
 802             velecsum         = _mm256_add_ps(velecsum,velec);
 803             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 804             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 805
 806             fscal            = _mm256_add_ps(felec,fvdw);
 807
 808             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 809
 810             /* Calculate temporary vectorial force */
 811             tx               = _mm256_mul_ps(fscal,dx00);
 812             ty               = _mm256_mul_ps(fscal,dy00);
 813             tz               = _mm256_mul_ps(fscal,dz00);
 814
 815             /* Update vectorial force */
 816             fix0             = _mm256_add_ps(fix0,tx);
 817             fiy0             = _mm256_add_ps(fiy0,ty);
 818             fiz0             = _mm256_add_ps(fiz0,tz);
 819
 820             fjx0             = _mm256_add_ps(fjx0,tx);
 821             fjy0             = _mm256_add_ps(fjy0,ty);
 822             fjz0             = _mm256_add_ps(fjz0,tz);
 823
 824             /**************************
 825              * CALCULATE INTERACTIONS *
 826              **************************/
 827
 828             /* COULOMB ELECTROSTATICS */
 829             velec            = _mm256_mul_ps(qq01,rinv01);
 830             felec            = _mm256_mul_ps(velec,rinvsq01);
 831
 832             /* Update potential sum for this i atom from the interaction with this j atom. */
 833             velec            = _mm256_andnot_ps(dummy_mask,velec);
 834             velecsum         = _mm256_add_ps(velecsum,velec);
 835
 836             fscal            = felec;
 837
 838             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 839
 840             /* Calculate temporary vectorial force */
 841             tx               = _mm256_mul_ps(fscal,dx01);
 842             ty               = _mm256_mul_ps(fscal,dy01);
 843             tz               = _mm256_mul_ps(fscal,dz01);
 844
 845             /* Update vectorial force */
 846             fix0             = _mm256_add_ps(fix0,tx);
 847             fiy0             = _mm256_add_ps(fiy0,ty);
 848             fiz0             = _mm256_add_ps(fiz0,tz);
 849
 850             fjx1             = _mm256_add_ps(fjx1,tx);
 851             fjy1             = _mm256_add_ps(fjy1,ty);
 852             fjz1             = _mm256_add_ps(fjz1,tz);
 853
 854             /**************************
 855              * CALCULATE INTERACTIONS *
 856              **************************/
 857
 858             /* COULOMB ELECTROSTATICS */
 859             velec            = _mm256_mul_ps(qq02,rinv02);
 860             felec            = _mm256_mul_ps(velec,rinvsq02);
 861
 862             /* Update potential sum for this i atom from the interaction with this j atom. */
 863             velec            = _mm256_andnot_ps(dummy_mask,velec);
 864             velecsum         = _mm256_add_ps(velecsum,velec);
 865
 866             fscal            = felec;
 867
 868             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 869
 870             /* Calculate temporary vectorial force */
 871             tx               = _mm256_mul_ps(fscal,dx02);
 872             ty               = _mm256_mul_ps(fscal,dy02);
 873             tz               = _mm256_mul_ps(fscal,dz02);
 874
 875             /* Update vectorial force */
 876             fix0             = _mm256_add_ps(fix0,tx);
 877             fiy0             = _mm256_add_ps(fiy0,ty);
 878             fiz0             = _mm256_add_ps(fiz0,tz);
 879
 880             fjx2             = _mm256_add_ps(fjx2,tx);
 881             fjy2             = _mm256_add_ps(fjy2,ty);
 882             fjz2             = _mm256_add_ps(fjz2,tz);
 883
 884             /**************************
 885              * CALCULATE INTERACTIONS *
 886              **************************/
 887
 888             /* COULOMB ELECTROSTATICS */
 889             velec            = _mm256_mul_ps(qq10,rinv10);
 890             felec            = _mm256_mul_ps(velec,rinvsq10);
 891
 892             /* Update potential sum for this i atom from the interaction with this j atom. */
 893             velec            = _mm256_andnot_ps(dummy_mask,velec);
 894             velecsum         = _mm256_add_ps(velecsum,velec);
 895
 896             fscal            = felec;
 897
 898             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 899
 900             /* Calculate temporary vectorial force */
 901             tx               = _mm256_mul_ps(fscal,dx10);
 902             ty               = _mm256_mul_ps(fscal,dy10);
 903             tz               = _mm256_mul_ps(fscal,dz10);
 904
 905             /* Update vectorial force */
 906             fix1             = _mm256_add_ps(fix1,tx);
 907             fiy1             = _mm256_add_ps(fiy1,ty);
 908             fiz1             = _mm256_add_ps(fiz1,tz);
 909
 910             fjx0             = _mm256_add_ps(fjx0,tx);
 911             fjy0             = _mm256_add_ps(fjy0,ty);
 912             fjz0             = _mm256_add_ps(fjz0,tz);
 913
 914             /**************************
 915              * CALCULATE INTERACTIONS *
 916              **************************/
 917
 918             /* COULOMB ELECTROSTATICS */
 919             velec            = _mm256_mul_ps(qq11,rinv11);
 920             felec            = _mm256_mul_ps(velec,rinvsq11);
 921
 922             /* Update potential sum for this i atom from the interaction with this j atom. */
 923             velec            = _mm256_andnot_ps(dummy_mask,velec);
 924             velecsum         = _mm256_add_ps(velecsum,velec);
 925
 926             fscal            = felec;
 927
 928             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 929
 930             /* Calculate temporary vectorial force */
 931             tx               = _mm256_mul_ps(fscal,dx11);
 932             ty               = _mm256_mul_ps(fscal,dy11);
 933             tz               = _mm256_mul_ps(fscal,dz11);
 934
 935             /* Update vectorial force */
 936             fix1             = _mm256_add_ps(fix1,tx);
 937             fiy1             = _mm256_add_ps(fiy1,ty);
 938             fiz1             = _mm256_add_ps(fiz1,tz);
 939
 940             fjx1             = _mm256_add_ps(fjx1,tx);
 941             fjy1             = _mm256_add_ps(fjy1,ty);
 942             fjz1             = _mm256_add_ps(fjz1,tz);
 943
 944             /**************************
 945              * CALCULATE INTERACTIONS *
 946              **************************/
 947
 948             /* COULOMB ELECTROSTATICS */
 949             velec            = _mm256_mul_ps(qq12,rinv12);
 950             felec            = _mm256_mul_ps(velec,rinvsq12);
 951
 952             /* Update potential sum for this i atom from the interaction with this j atom. */
 953             velec            = _mm256_andnot_ps(dummy_mask,velec);
 954             velecsum         = _mm256_add_ps(velecsum,velec);
 955
 956             fscal            = felec;
 957
 958             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 959
 960             /* Calculate temporary vectorial force */
 961             tx               = _mm256_mul_ps(fscal,dx12);
 962             ty               = _mm256_mul_ps(fscal,dy12);
 963             tz               = _mm256_mul_ps(fscal,dz12);
 964
 965             /* Update vectorial force */
 966             fix1             = _mm256_add_ps(fix1,tx);
 967             fiy1             = _mm256_add_ps(fiy1,ty);
 968             fiz1             = _mm256_add_ps(fiz1,tz);
 969
 970             fjx2             = _mm256_add_ps(fjx2,tx);
 971             fjy2             = _mm256_add_ps(fjy2,ty);
 972             fjz2             = _mm256_add_ps(fjz2,tz);
 973
 974             /**************************
 975              * CALCULATE INTERACTIONS *
 976              **************************/
 977
 978             /* COULOMB ELECTROSTATICS */
 979             velec            = _mm256_mul_ps(qq20,rinv20);
 980             felec            = _mm256_mul_ps(velec,rinvsq20);
 981
 982             /* Update potential sum for this i atom from the interaction with this j atom. */
 983             velec            = _mm256_andnot_ps(dummy_mask,velec);
 984             velecsum         = _mm256_add_ps(velecsum,velec);
 985
 986             fscal            = felec;
 987
 988             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 989
 990             /* Calculate temporary vectorial force */
 991             tx               = _mm256_mul_ps(fscal,dx20);
 992             ty               = _mm256_mul_ps(fscal,dy20);
 993             tz               = _mm256_mul_ps(fscal,dz20);
 994
 995             /* Update vectorial force */
 996             fix2             = _mm256_add_ps(fix2,tx);
 997             fiy2             = _mm256_add_ps(fiy2,ty);
 998             fiz2             = _mm256_add_ps(fiz2,tz);
 999
1000             fjx0             = _mm256_add_ps(fjx0,tx);
1001             fjy0             = _mm256_add_ps(fjy0,ty);
1002             fjz0             = _mm256_add_ps(fjz0,tz);
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             /* COULOMB ELECTROSTATICS */
1009             velec            = _mm256_mul_ps(qq21,rinv21);
1010             felec            = _mm256_mul_ps(velec,rinvsq21);
1011
1012             /* Update potential sum for this i atom from the interaction with this j atom. */
1013             velec            = _mm256_andnot_ps(dummy_mask,velec);
1014             velecsum         = _mm256_add_ps(velecsum,velec);
1015
1016             fscal            = felec;
1017
1018             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1019
1020             /* Calculate temporary vectorial force */
1021             tx               = _mm256_mul_ps(fscal,dx21);
1022             ty               = _mm256_mul_ps(fscal,dy21);
1023             tz               = _mm256_mul_ps(fscal,dz21);
1024
1025             /* Update vectorial force */
1026             fix2             = _mm256_add_ps(fix2,tx);
1027             fiy2             = _mm256_add_ps(fiy2,ty);
1028             fiz2             = _mm256_add_ps(fiz2,tz);
1029
1030             fjx1             = _mm256_add_ps(fjx1,tx);
1031             fjy1             = _mm256_add_ps(fjy1,ty);
1032             fjz1             = _mm256_add_ps(fjz1,tz);
1033
1034             /**************************
1035              * CALCULATE INTERACTIONS *
1036              **************************/
1037
1038             /* COULOMB ELECTROSTATICS */
1039             velec            = _mm256_mul_ps(qq22,rinv22);
1040             felec            = _mm256_mul_ps(velec,rinvsq22);
1041
1042             /* Update potential sum for this i atom from the interaction with this j atom. */
1043             velec            = _mm256_andnot_ps(dummy_mask,velec);
1044             velecsum         = _mm256_add_ps(velecsum,velec);
1045
1046             fscal            = felec;
1047
1048             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1049
1050             /* Calculate temporary vectorial force */
1051             tx               = _mm256_mul_ps(fscal,dx22);
1052             ty               = _mm256_mul_ps(fscal,dy22);
1053             tz               = _mm256_mul_ps(fscal,dz22);
1054
1055             /* Update vectorial force */
1056             fix2             = _mm256_add_ps(fix2,tx);
1057             fiy2             = _mm256_add_ps(fiy2,ty);
1058             fiz2             = _mm256_add_ps(fiz2,tz);
1059
1060             fjx2             = _mm256_add_ps(fjx2,tx);
1061             fjy2             = _mm256_add_ps(fjy2,ty);
1062             fjz2             = _mm256_add_ps(fjz2,tz);
1063
1064             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1065             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1066             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1067             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1068             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1069             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1070             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1071             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1072
1073             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1074                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1075
1076             /* Inner loop uses 279 flops */
1077         }
1078
1079         /* End of innermost loop */
1080
1081         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1082                                                  f+i_coord_offset,fshift+i_shift_offset);
1083
1084         ggid                        = gid[iidx];
1085         /* Update potential energies */
1086         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1087         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1088
1089         /* Increment number of inner iterations */
1090         inneriter                  += j_index_end - j_index_start;
1091
1092         /* Outer loop uses 20 flops */
1093     }
1094
1095     /* Increment number of outer iterations */
1096     outeriter        += nri;
1097
1098     /* Update outer/inner flops */
1099
1100     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*279);
1101 }
1102 /*
1103  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_F_avx_256_single
1104  * Electrostatics interaction: Coulomb
1105  * VdW interaction:            CubicSplineTable
1106  * Geometry:                   Water3-Water3
1107  * Calculate force/pot:        Force
1108  */
1109 void
1110 nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_F_avx_256_single
1111                     (t_nblist                    * gmx_restrict       nlist,
1112                      rvec                        * gmx_restrict          xx,
1113                      rvec                        * gmx_restrict          ff,
1114                      t_forcerec                  * gmx_restrict          fr,
1115                      t_mdatoms                   * gmx_restrict     mdatoms,
1116                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1117                      t_nrnb                      * gmx_restrict        nrnb)
1118 {
1119     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1120      * just 0 for non-waters.
1121      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1122      * jnr indices corresponding to data put in the four positions in the SIMD register.
1123      */
1124     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1125     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1126     int              jnrA,jnrB,jnrC,jnrD;
1127     int              jnrE,jnrF,jnrG,jnrH;
1128     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1129     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1130     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1131     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1132     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1133     real             rcutoff_scalar;
1134     real             *shiftvec,*fshift,*x,*f;
1135     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1136     real             scratch[4*DIM];
1137     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1138     real *           vdwioffsetptr0;
1139     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1140     real *           vdwioffsetptr1;
1141     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1142     real *           vdwioffsetptr2;
1143     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1144     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1145     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1146     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1147     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1148     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1149     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1150     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1151     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1152     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1153     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1154     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1155     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1156     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1157     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1158     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1159     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1160     real             *charge;
1161     int              nvdwtype;
1162     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1163     int              *vdwtype;
1164     real             *vdwparam;
1165     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1166     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1167     __m256i          vfitab;
1168     __m128i          vfitab_lo,vfitab_hi;
1169     __m128i          ifour       = _mm_set1_epi32(4);
1170     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1171     real             *vftab;
1172     __m256           dummy_mask,cutoff_mask;
1173     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1174     __m256           one     = _mm256_set1_ps(1.0);
1175     __m256           two     = _mm256_set1_ps(2.0);
1176     x                = xx[0];
1177     f                = ff[0];
1178
1179     nri              = nlist->nri;
1180     iinr             = nlist->iinr;
1181     jindex           = nlist->jindex;
1182     jjnr             = nlist->jjnr;
1183     shiftidx         = nlist->shift;
1184     gid              = nlist->gid;
1185     shiftvec         = fr->shift_vec[0];
1186     fshift           = fr->fshift[0];
1187     facel            = _mm256_set1_ps(fr->epsfac);
1188     charge           = mdatoms->chargeA;
1189     nvdwtype         = fr->ntype;
1190     vdwparam         = fr->nbfp;
1191     vdwtype          = mdatoms->typeA;
1192
1193     vftab            = kernel_data->table_vdw->data;
1194     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
1195
1196     /* Setup water-specific parameters */
1197     inr              = nlist->iinr[0];
1198     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1199     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1200     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1201     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1202
1203     jq0              = _mm256_set1_ps(charge[inr+0]);
1204     jq1              = _mm256_set1_ps(charge[inr+1]);
1205     jq2              = _mm256_set1_ps(charge[inr+2]);
1206     vdwjidx0A        = 2*vdwtype[inr+0];
1207     qq00             = _mm256_mul_ps(iq0,jq0);
1208     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1209     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1210     qq01             = _mm256_mul_ps(iq0,jq1);
1211     qq02             = _mm256_mul_ps(iq0,jq2);
1212     qq10             = _mm256_mul_ps(iq1,jq0);
1213     qq11             = _mm256_mul_ps(iq1,jq1);
1214     qq12             = _mm256_mul_ps(iq1,jq2);
1215     qq20             = _mm256_mul_ps(iq2,jq0);
1216     qq21             = _mm256_mul_ps(iq2,jq1);
1217     qq22             = _mm256_mul_ps(iq2,jq2);
1218
1219     /* Avoid stupid compiler warnings */
1220     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1221     j_coord_offsetA = 0;
1222     j_coord_offsetB = 0;
1223     j_coord_offsetC = 0;
1224     j_coord_offsetD = 0;
1225     j_coord_offsetE = 0;
1226     j_coord_offsetF = 0;
1227     j_coord_offsetG = 0;
1228     j_coord_offsetH = 0;
1229
1230     outeriter        = 0;
1231     inneriter        = 0;
1232
1233     for(iidx=0;iidx<4*DIM;iidx++)
1234     {
1235         scratch[iidx] = 0.0;
1236     }
1237
1238     /* Start outer loop over neighborlists */
1239     for(iidx=0; iidx<nri; iidx++)
1240     {
1241         /* Load shift vector for this list */
1242         i_shift_offset   = DIM*shiftidx[iidx];
1243
1244         /* Load limits for loop over neighbors */
1245         j_index_start    = jindex[iidx];
1246         j_index_end      = jindex[iidx+1];
1247
1248         /* Get outer coordinate index */
1249         inr              = iinr[iidx];
1250         i_coord_offset   = DIM*inr;
1251
1252         /* Load i particle coords and add shift vector */
1253         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1254                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1255
1256         fix0             = _mm256_setzero_ps();
1257         fiy0             = _mm256_setzero_ps();
1258         fiz0             = _mm256_setzero_ps();
1259         fix1             = _mm256_setzero_ps();
1260         fiy1             = _mm256_setzero_ps();
1261         fiz1             = _mm256_setzero_ps();
1262         fix2             = _mm256_setzero_ps();
1263         fiy2             = _mm256_setzero_ps();
1264         fiz2             = _mm256_setzero_ps();
1265
1266         /* Start inner kernel loop */
1267         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1268         {
1269
1270             /* Get j neighbor index, and coordinate index */
1271             jnrA             = jjnr[jidx];
1272             jnrB             = jjnr[jidx+1];
1273             jnrC             = jjnr[jidx+2];
1274             jnrD             = jjnr[jidx+3];
1275             jnrE             = jjnr[jidx+4];
1276             jnrF             = jjnr[jidx+5];
1277             jnrG             = jjnr[jidx+6];
1278             jnrH             = jjnr[jidx+7];
1279             j_coord_offsetA  = DIM*jnrA;
1280             j_coord_offsetB  = DIM*jnrB;
1281             j_coord_offsetC  = DIM*jnrC;
1282             j_coord_offsetD  = DIM*jnrD;
1283             j_coord_offsetE  = DIM*jnrE;
1284             j_coord_offsetF  = DIM*jnrF;
1285             j_coord_offsetG  = DIM*jnrG;
1286             j_coord_offsetH  = DIM*jnrH;
1287
1288             /* load j atom coordinates */
1289             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1290                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1291                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1292                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1293                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1294
1295             /* Calculate displacement vector */
1296             dx00             = _mm256_sub_ps(ix0,jx0);
1297             dy00             = _mm256_sub_ps(iy0,jy0);
1298             dz00             = _mm256_sub_ps(iz0,jz0);
1299             dx01             = _mm256_sub_ps(ix0,jx1);
1300             dy01             = _mm256_sub_ps(iy0,jy1);
1301             dz01             = _mm256_sub_ps(iz0,jz1);
1302             dx02             = _mm256_sub_ps(ix0,jx2);
1303             dy02             = _mm256_sub_ps(iy0,jy2);
1304             dz02             = _mm256_sub_ps(iz0,jz2);
1305             dx10             = _mm256_sub_ps(ix1,jx0);
1306             dy10             = _mm256_sub_ps(iy1,jy0);
1307             dz10             = _mm256_sub_ps(iz1,jz0);
1308             dx11             = _mm256_sub_ps(ix1,jx1);
1309             dy11             = _mm256_sub_ps(iy1,jy1);
1310             dz11             = _mm256_sub_ps(iz1,jz1);
1311             dx12             = _mm256_sub_ps(ix1,jx2);
1312             dy12             = _mm256_sub_ps(iy1,jy2);
1313             dz12             = _mm256_sub_ps(iz1,jz2);
1314             dx20             = _mm256_sub_ps(ix2,jx0);
1315             dy20             = _mm256_sub_ps(iy2,jy0);
1316             dz20             = _mm256_sub_ps(iz2,jz0);
1317             dx21             = _mm256_sub_ps(ix2,jx1);
1318             dy21             = _mm256_sub_ps(iy2,jy1);
1319             dz21             = _mm256_sub_ps(iz2,jz1);
1320             dx22             = _mm256_sub_ps(ix2,jx2);
1321             dy22             = _mm256_sub_ps(iy2,jy2);
1322             dz22             = _mm256_sub_ps(iz2,jz2);
1323
1324             /* Calculate squared distance and things based on it */
1325             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1326             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1327             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1328             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1329             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1330             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1331             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1332             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1333             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1334
1335             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1336             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1337             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1338             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1339             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1340             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1341             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1342             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1343             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1344
1345             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1346             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1347             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1348             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1349             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1350             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1351             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1352             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1353             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1354
1355             fjx0             = _mm256_setzero_ps();
1356             fjy0             = _mm256_setzero_ps();
1357             fjz0             = _mm256_setzero_ps();
1358             fjx1             = _mm256_setzero_ps();
1359             fjy1             = _mm256_setzero_ps();
1360             fjz1             = _mm256_setzero_ps();
1361             fjx2             = _mm256_setzero_ps();
1362             fjy2             = _mm256_setzero_ps();
1363             fjz2             = _mm256_setzero_ps();
1364
1365             /**************************
1366              * CALCULATE INTERACTIONS *
1367              **************************/
1368
1369             r00              = _mm256_mul_ps(rsq00,rinv00);
1370
1371             /* Calculate table index by multiplying r with table scale and truncate to integer */
1372             rt               = _mm256_mul_ps(r00,vftabscale);
1373             vfitab           = _mm256_cvttps_epi32(rt);
1374             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1375             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1376             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1377             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1378             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1379             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1380
1381             /* COULOMB ELECTROSTATICS */
1382             velec            = _mm256_mul_ps(qq00,rinv00);
1383             felec            = _mm256_mul_ps(velec,rinvsq00);
1384
1385             /* CUBIC SPLINE TABLE DISPERSION */
1386             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1387                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1388             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1389                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1390             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1391                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1392             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1393                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1394             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1395             Heps             = _mm256_mul_ps(vfeps,H);
1396             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1397             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1398             fvdw6            = _mm256_mul_ps(c6_00,FF);
1399
1400             /* CUBIC SPLINE TABLE REPULSION */
1401             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1402             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1403             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1404                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1405             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1406                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1407             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1408                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1409             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1410                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1411             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1412             Heps             = _mm256_mul_ps(vfeps,H);
1413             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1414             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1415             fvdw12           = _mm256_mul_ps(c12_00,FF);
1416             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1417
1418             fscal            = _mm256_add_ps(felec,fvdw);
1419
1420             /* Calculate temporary vectorial force */
1421             tx               = _mm256_mul_ps(fscal,dx00);
1422             ty               = _mm256_mul_ps(fscal,dy00);
1423             tz               = _mm256_mul_ps(fscal,dz00);
1424
1425             /* Update vectorial force */
1426             fix0             = _mm256_add_ps(fix0,tx);
1427             fiy0             = _mm256_add_ps(fiy0,ty);
1428             fiz0             = _mm256_add_ps(fiz0,tz);
1429
1430             fjx0             = _mm256_add_ps(fjx0,tx);
1431             fjy0             = _mm256_add_ps(fjy0,ty);
1432             fjz0             = _mm256_add_ps(fjz0,tz);
1433
1434             /**************************
1435              * CALCULATE INTERACTIONS *
1436              **************************/
1437
1438             /* COULOMB ELECTROSTATICS */
1439             velec            = _mm256_mul_ps(qq01,rinv01);
1440             felec            = _mm256_mul_ps(velec,rinvsq01);
1441
1442             fscal            = felec;
1443
1444             /* Calculate temporary vectorial force */
1445             tx               = _mm256_mul_ps(fscal,dx01);
1446             ty               = _mm256_mul_ps(fscal,dy01);
1447             tz               = _mm256_mul_ps(fscal,dz01);
1448
1449             /* Update vectorial force */
1450             fix0             = _mm256_add_ps(fix0,tx);
1451             fiy0             = _mm256_add_ps(fiy0,ty);
1452             fiz0             = _mm256_add_ps(fiz0,tz);
1453
1454             fjx1             = _mm256_add_ps(fjx1,tx);
1455             fjy1             = _mm256_add_ps(fjy1,ty);
1456             fjz1             = _mm256_add_ps(fjz1,tz);
1457
1458             /**************************
1459              * CALCULATE INTERACTIONS *
1460              **************************/
1461
1462             /* COULOMB ELECTROSTATICS */
1463             velec            = _mm256_mul_ps(qq02,rinv02);
1464             felec            = _mm256_mul_ps(velec,rinvsq02);
1465
1466             fscal            = felec;
1467
1468             /* Calculate temporary vectorial force */
1469             tx               = _mm256_mul_ps(fscal,dx02);
1470             ty               = _mm256_mul_ps(fscal,dy02);
1471             tz               = _mm256_mul_ps(fscal,dz02);
1472
1473             /* Update vectorial force */
1474             fix0             = _mm256_add_ps(fix0,tx);
1475             fiy0             = _mm256_add_ps(fiy0,ty);
1476             fiz0             = _mm256_add_ps(fiz0,tz);
1477
1478             fjx2             = _mm256_add_ps(fjx2,tx);
1479             fjy2             = _mm256_add_ps(fjy2,ty);
1480             fjz2             = _mm256_add_ps(fjz2,tz);
1481
1482             /**************************
1483              * CALCULATE INTERACTIONS *
1484              **************************/
1485
1486             /* COULOMB ELECTROSTATICS */
1487             velec            = _mm256_mul_ps(qq10,rinv10);
1488             felec            = _mm256_mul_ps(velec,rinvsq10);
1489
1490             fscal            = felec;
1491
1492             /* Calculate temporary vectorial force */
1493             tx               = _mm256_mul_ps(fscal,dx10);
1494             ty               = _mm256_mul_ps(fscal,dy10);
1495             tz               = _mm256_mul_ps(fscal,dz10);
1496
1497             /* Update vectorial force */
1498             fix1             = _mm256_add_ps(fix1,tx);
1499             fiy1             = _mm256_add_ps(fiy1,ty);
1500             fiz1             = _mm256_add_ps(fiz1,tz);
1501
1502             fjx0             = _mm256_add_ps(fjx0,tx);
1503             fjy0             = _mm256_add_ps(fjy0,ty);
1504             fjz0             = _mm256_add_ps(fjz0,tz);
1505
1506             /**************************
1507              * CALCULATE INTERACTIONS *
1508              **************************/
1509
1510             /* COULOMB ELECTROSTATICS */
1511             velec            = _mm256_mul_ps(qq11,rinv11);
1512             felec            = _mm256_mul_ps(velec,rinvsq11);
1513
1514             fscal            = felec;
1515
1516             /* Calculate temporary vectorial force */
1517             tx               = _mm256_mul_ps(fscal,dx11);
1518             ty               = _mm256_mul_ps(fscal,dy11);
1519             tz               = _mm256_mul_ps(fscal,dz11);
1520
1521             /* Update vectorial force */
1522             fix1             = _mm256_add_ps(fix1,tx);
1523             fiy1             = _mm256_add_ps(fiy1,ty);
1524             fiz1             = _mm256_add_ps(fiz1,tz);
1525
1526             fjx1             = _mm256_add_ps(fjx1,tx);
1527             fjy1             = _mm256_add_ps(fjy1,ty);
1528             fjz1             = _mm256_add_ps(fjz1,tz);
1529
1530             /**************************
1531              * CALCULATE INTERACTIONS *
1532              **************************/
1533
1534             /* COULOMB ELECTROSTATICS */
1535             velec            = _mm256_mul_ps(qq12,rinv12);
1536             felec            = _mm256_mul_ps(velec,rinvsq12);
1537
1538             fscal            = felec;
1539
1540             /* Calculate temporary vectorial force */
1541             tx               = _mm256_mul_ps(fscal,dx12);
1542             ty               = _mm256_mul_ps(fscal,dy12);
1543             tz               = _mm256_mul_ps(fscal,dz12);
1544
1545             /* Update vectorial force */
1546             fix1             = _mm256_add_ps(fix1,tx);
1547             fiy1             = _mm256_add_ps(fiy1,ty);
1548             fiz1             = _mm256_add_ps(fiz1,tz);
1549
1550             fjx2             = _mm256_add_ps(fjx2,tx);
1551             fjy2             = _mm256_add_ps(fjy2,ty);
1552             fjz2             = _mm256_add_ps(fjz2,tz);
1553
1554             /**************************
1555              * CALCULATE INTERACTIONS *
1556              **************************/
1557
1558             /* COULOMB ELECTROSTATICS */
1559             velec            = _mm256_mul_ps(qq20,rinv20);
1560             felec            = _mm256_mul_ps(velec,rinvsq20);
1561
1562             fscal            = felec;
1563
1564             /* Calculate temporary vectorial force */
1565             tx               = _mm256_mul_ps(fscal,dx20);
1566             ty               = _mm256_mul_ps(fscal,dy20);
1567             tz               = _mm256_mul_ps(fscal,dz20);
1568
1569             /* Update vectorial force */
1570             fix2             = _mm256_add_ps(fix2,tx);
1571             fiy2             = _mm256_add_ps(fiy2,ty);
1572             fiz2             = _mm256_add_ps(fiz2,tz);
1573
1574             fjx0             = _mm256_add_ps(fjx0,tx);
1575             fjy0             = _mm256_add_ps(fjy0,ty);
1576             fjz0             = _mm256_add_ps(fjz0,tz);
1577
1578             /**************************
1579              * CALCULATE INTERACTIONS *
1580              **************************/
1581
1582             /* COULOMB ELECTROSTATICS */
1583             velec            = _mm256_mul_ps(qq21,rinv21);
1584             felec            = _mm256_mul_ps(velec,rinvsq21);
1585
1586             fscal            = felec;
1587
1588             /* Calculate temporary vectorial force */
1589             tx               = _mm256_mul_ps(fscal,dx21);
1590             ty               = _mm256_mul_ps(fscal,dy21);
1591             tz               = _mm256_mul_ps(fscal,dz21);
1592
1593             /* Update vectorial force */
1594             fix2             = _mm256_add_ps(fix2,tx);
1595             fiy2             = _mm256_add_ps(fiy2,ty);
1596             fiz2             = _mm256_add_ps(fiz2,tz);
1597
1598             fjx1             = _mm256_add_ps(fjx1,tx);
1599             fjy1             = _mm256_add_ps(fjy1,ty);
1600             fjz1             = _mm256_add_ps(fjz1,tz);
1601
1602             /**************************
1603              * CALCULATE INTERACTIONS *
1604              **************************/
1605
1606             /* COULOMB ELECTROSTATICS */
1607             velec            = _mm256_mul_ps(qq22,rinv22);
1608             felec            = _mm256_mul_ps(velec,rinvsq22);
1609
1610             fscal            = felec;
1611
1612             /* Calculate temporary vectorial force */
1613             tx               = _mm256_mul_ps(fscal,dx22);
1614             ty               = _mm256_mul_ps(fscal,dy22);
1615             tz               = _mm256_mul_ps(fscal,dz22);
1616
1617             /* Update vectorial force */
1618             fix2             = _mm256_add_ps(fix2,tx);
1619             fiy2             = _mm256_add_ps(fiy2,ty);
1620             fiz2             = _mm256_add_ps(fiz2,tz);
1621
1622             fjx2             = _mm256_add_ps(fjx2,tx);
1623             fjy2             = _mm256_add_ps(fjy2,ty);
1624             fjz2             = _mm256_add_ps(fjz2,tz);
1625
1626             fjptrA             = f+j_coord_offsetA;
1627             fjptrB             = f+j_coord_offsetB;
1628             fjptrC             = f+j_coord_offsetC;
1629             fjptrD             = f+j_coord_offsetD;
1630             fjptrE             = f+j_coord_offsetE;
1631             fjptrF             = f+j_coord_offsetF;
1632             fjptrG             = f+j_coord_offsetG;
1633             fjptrH             = f+j_coord_offsetH;
1634
1635             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1636                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1637
1638             /* Inner loop uses 261 flops */
1639         }
1640
1641         if(jidx<j_index_end)
1642         {
1643
1644             /* Get j neighbor index, and coordinate index */
1645             jnrlistA         = jjnr[jidx];
1646             jnrlistB         = jjnr[jidx+1];
1647             jnrlistC         = jjnr[jidx+2];
1648             jnrlistD         = jjnr[jidx+3];
1649             jnrlistE         = jjnr[jidx+4];
1650             jnrlistF         = jjnr[jidx+5];
1651             jnrlistG         = jjnr[jidx+6];
1652             jnrlistH         = jjnr[jidx+7];
1653             /* Sign of each element will be negative for non-real atoms.
1654              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1655              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1656              */
1657             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1658                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1659
1660             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1661             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1662             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1663             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1664             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1665             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1666             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1667             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1668             j_coord_offsetA  = DIM*jnrA;
1669             j_coord_offsetB  = DIM*jnrB;
1670             j_coord_offsetC  = DIM*jnrC;
1671             j_coord_offsetD  = DIM*jnrD;
1672             j_coord_offsetE  = DIM*jnrE;
1673             j_coord_offsetF  = DIM*jnrF;
1674             j_coord_offsetG  = DIM*jnrG;
1675             j_coord_offsetH  = DIM*jnrH;
1676
1677             /* load j atom coordinates */
1678             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1679                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1680                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1681                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1682                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1683
1684             /* Calculate displacement vector */
1685             dx00             = _mm256_sub_ps(ix0,jx0);
1686             dy00             = _mm256_sub_ps(iy0,jy0);
1687             dz00             = _mm256_sub_ps(iz0,jz0);
1688             dx01             = _mm256_sub_ps(ix0,jx1);
1689             dy01             = _mm256_sub_ps(iy0,jy1);
1690             dz01             = _mm256_sub_ps(iz0,jz1);
1691             dx02             = _mm256_sub_ps(ix0,jx2);
1692             dy02             = _mm256_sub_ps(iy0,jy2);
1693             dz02             = _mm256_sub_ps(iz0,jz2);
1694             dx10             = _mm256_sub_ps(ix1,jx0);
1695             dy10             = _mm256_sub_ps(iy1,jy0);
1696             dz10             = _mm256_sub_ps(iz1,jz0);
1697             dx11             = _mm256_sub_ps(ix1,jx1);
1698             dy11             = _mm256_sub_ps(iy1,jy1);
1699             dz11             = _mm256_sub_ps(iz1,jz1);
1700             dx12             = _mm256_sub_ps(ix1,jx2);
1701             dy12             = _mm256_sub_ps(iy1,jy2);
1702             dz12             = _mm256_sub_ps(iz1,jz2);
1703             dx20             = _mm256_sub_ps(ix2,jx0);
1704             dy20             = _mm256_sub_ps(iy2,jy0);
1705             dz20             = _mm256_sub_ps(iz2,jz0);
1706             dx21             = _mm256_sub_ps(ix2,jx1);
1707             dy21             = _mm256_sub_ps(iy2,jy1);
1708             dz21             = _mm256_sub_ps(iz2,jz1);
1709             dx22             = _mm256_sub_ps(ix2,jx2);
1710             dy22             = _mm256_sub_ps(iy2,jy2);
1711             dz22             = _mm256_sub_ps(iz2,jz2);
1712
1713             /* Calculate squared distance and things based on it */
1714             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1715             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1716             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1717             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1718             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1719             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1720             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1721             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1722             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1723
1724             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1725             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1726             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1727             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1728             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1729             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1730             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1731             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1732             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1733
1734             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1735             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1736             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1737             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1738             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1739             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1740             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1741             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1742             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1743
1744             fjx0             = _mm256_setzero_ps();
1745             fjy0             = _mm256_setzero_ps();
1746             fjz0             = _mm256_setzero_ps();
1747             fjx1             = _mm256_setzero_ps();
1748             fjy1             = _mm256_setzero_ps();
1749             fjz1             = _mm256_setzero_ps();
1750             fjx2             = _mm256_setzero_ps();
1751             fjy2             = _mm256_setzero_ps();
1752             fjz2             = _mm256_setzero_ps();
1753
1754             /**************************
1755              * CALCULATE INTERACTIONS *
1756              **************************/
1757
1758             r00              = _mm256_mul_ps(rsq00,rinv00);
1759             r00              = _mm256_andnot_ps(dummy_mask,r00);
1760
1761             /* Calculate table index by multiplying r with table scale and truncate to integer */
1762             rt               = _mm256_mul_ps(r00,vftabscale);
1763             vfitab           = _mm256_cvttps_epi32(rt);
1764             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1765             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1766             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1767             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1768             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1769             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1770
1771             /* COULOMB ELECTROSTATICS */
1772             velec            = _mm256_mul_ps(qq00,rinv00);
1773             felec            = _mm256_mul_ps(velec,rinvsq00);
1774
1775             /* CUBIC SPLINE TABLE DISPERSION */
1776             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1777                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1778             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1779                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1780             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1781                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1782             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1783                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1784             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1785             Heps             = _mm256_mul_ps(vfeps,H);
1786             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1787             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1788             fvdw6            = _mm256_mul_ps(c6_00,FF);
1789
1790             /* CUBIC SPLINE TABLE REPULSION */
1791             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1792             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1793             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1794                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1795             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1796                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1797             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1798                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1799             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1800                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1801             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1802             Heps             = _mm256_mul_ps(vfeps,H);
1803             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1804             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1805             fvdw12           = _mm256_mul_ps(c12_00,FF);
1806             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1807
1808             fscal            = _mm256_add_ps(felec,fvdw);
1809
1810             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1811
1812             /* Calculate temporary vectorial force */
1813             tx               = _mm256_mul_ps(fscal,dx00);
1814             ty               = _mm256_mul_ps(fscal,dy00);
1815             tz               = _mm256_mul_ps(fscal,dz00);
1816
1817             /* Update vectorial force */
1818             fix0             = _mm256_add_ps(fix0,tx);
1819             fiy0             = _mm256_add_ps(fiy0,ty);
1820             fiz0             = _mm256_add_ps(fiz0,tz);
1821
1822             fjx0             = _mm256_add_ps(fjx0,tx);
1823             fjy0             = _mm256_add_ps(fjy0,ty);
1824             fjz0             = _mm256_add_ps(fjz0,tz);
1825
1826             /**************************
1827              * CALCULATE INTERACTIONS *
1828              **************************/
1829
1830             /* COULOMB ELECTROSTATICS */
1831             velec            = _mm256_mul_ps(qq01,rinv01);
1832             felec            = _mm256_mul_ps(velec,rinvsq01);
1833
1834             fscal            = felec;
1835
1836             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1837
1838             /* Calculate temporary vectorial force */
1839             tx               = _mm256_mul_ps(fscal,dx01);
1840             ty               = _mm256_mul_ps(fscal,dy01);
1841             tz               = _mm256_mul_ps(fscal,dz01);
1842
1843             /* Update vectorial force */
1844             fix0             = _mm256_add_ps(fix0,tx);
1845             fiy0             = _mm256_add_ps(fiy0,ty);
1846             fiz0             = _mm256_add_ps(fiz0,tz);
1847
1848             fjx1             = _mm256_add_ps(fjx1,tx);
1849             fjy1             = _mm256_add_ps(fjy1,ty);
1850             fjz1             = _mm256_add_ps(fjz1,tz);
1851
1852             /**************************
1853              * CALCULATE INTERACTIONS *
1854              **************************/
1855
1856             /* COULOMB ELECTROSTATICS */
1857             velec            = _mm256_mul_ps(qq02,rinv02);
1858             felec            = _mm256_mul_ps(velec,rinvsq02);
1859
1860             fscal            = felec;
1861
1862             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1863
1864             /* Calculate temporary vectorial force */
1865             tx               = _mm256_mul_ps(fscal,dx02);
1866             ty               = _mm256_mul_ps(fscal,dy02);
1867             tz               = _mm256_mul_ps(fscal,dz02);
1868
1869             /* Update vectorial force */
1870             fix0             = _mm256_add_ps(fix0,tx);
1871             fiy0             = _mm256_add_ps(fiy0,ty);
1872             fiz0             = _mm256_add_ps(fiz0,tz);
1873
1874             fjx2             = _mm256_add_ps(fjx2,tx);
1875             fjy2             = _mm256_add_ps(fjy2,ty);
1876             fjz2             = _mm256_add_ps(fjz2,tz);
1877
1878             /**************************
1879              * CALCULATE INTERACTIONS *
1880              **************************/
1881
1882             /* COULOMB ELECTROSTATICS */
1883             velec            = _mm256_mul_ps(qq10,rinv10);
1884             felec            = _mm256_mul_ps(velec,rinvsq10);
1885
1886             fscal            = felec;
1887
1888             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1889
1890             /* Calculate temporary vectorial force */
1891             tx               = _mm256_mul_ps(fscal,dx10);
1892             ty               = _mm256_mul_ps(fscal,dy10);
1893             tz               = _mm256_mul_ps(fscal,dz10);
1894
1895             /* Update vectorial force */
1896             fix1             = _mm256_add_ps(fix1,tx);
1897             fiy1             = _mm256_add_ps(fiy1,ty);
1898             fiz1             = _mm256_add_ps(fiz1,tz);
1899
1900             fjx0             = _mm256_add_ps(fjx0,tx);
1901             fjy0             = _mm256_add_ps(fjy0,ty);
1902             fjz0             = _mm256_add_ps(fjz0,tz);
1903
1904             /**************************
1905              * CALCULATE INTERACTIONS *
1906              **************************/
1907
1908             /* COULOMB ELECTROSTATICS */
1909             velec            = _mm256_mul_ps(qq11,rinv11);
1910             felec            = _mm256_mul_ps(velec,rinvsq11);
1911
1912             fscal            = felec;
1913
1914             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1915
1916             /* Calculate temporary vectorial force */
1917             tx               = _mm256_mul_ps(fscal,dx11);
1918             ty               = _mm256_mul_ps(fscal,dy11);
1919             tz               = _mm256_mul_ps(fscal,dz11);
1920
1921             /* Update vectorial force */
1922             fix1             = _mm256_add_ps(fix1,tx);
1923             fiy1             = _mm256_add_ps(fiy1,ty);
1924             fiz1             = _mm256_add_ps(fiz1,tz);
1925
1926             fjx1             = _mm256_add_ps(fjx1,tx);
1927             fjy1             = _mm256_add_ps(fjy1,ty);
1928             fjz1             = _mm256_add_ps(fjz1,tz);
1929
1930             /**************************
1931              * CALCULATE INTERACTIONS *
1932              **************************/
1933
1934             /* COULOMB ELECTROSTATICS */
1935             velec            = _mm256_mul_ps(qq12,rinv12);
1936             felec            = _mm256_mul_ps(velec,rinvsq12);
1937
1938             fscal            = felec;
1939
1940             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1941
1942             /* Calculate temporary vectorial force */
1943             tx               = _mm256_mul_ps(fscal,dx12);
1944             ty               = _mm256_mul_ps(fscal,dy12);
1945             tz               = _mm256_mul_ps(fscal,dz12);
1946
1947             /* Update vectorial force */
1948             fix1             = _mm256_add_ps(fix1,tx);
1949             fiy1             = _mm256_add_ps(fiy1,ty);
1950             fiz1             = _mm256_add_ps(fiz1,tz);
1951
1952             fjx2             = _mm256_add_ps(fjx2,tx);
1953             fjy2             = _mm256_add_ps(fjy2,ty);
1954             fjz2             = _mm256_add_ps(fjz2,tz);
1955
1956             /**************************
1957              * CALCULATE INTERACTIONS *
1958              **************************/
1959
1960             /* COULOMB ELECTROSTATICS */
1961             velec            = _mm256_mul_ps(qq20,rinv20);
1962             felec            = _mm256_mul_ps(velec,rinvsq20);
1963
1964             fscal            = felec;
1965
1966             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1967
1968             /* Calculate temporary vectorial force */
1969             tx               = _mm256_mul_ps(fscal,dx20);
1970             ty               = _mm256_mul_ps(fscal,dy20);
1971             tz               = _mm256_mul_ps(fscal,dz20);
1972
1973             /* Update vectorial force */
1974             fix2             = _mm256_add_ps(fix2,tx);
1975             fiy2             = _mm256_add_ps(fiy2,ty);
1976             fiz2             = _mm256_add_ps(fiz2,tz);
1977
1978             fjx0             = _mm256_add_ps(fjx0,tx);
1979             fjy0             = _mm256_add_ps(fjy0,ty);
1980             fjz0             = _mm256_add_ps(fjz0,tz);
1981
1982             /**************************
1983              * CALCULATE INTERACTIONS *
1984              **************************/
1985
1986             /* COULOMB ELECTROSTATICS */
1987             velec            = _mm256_mul_ps(qq21,rinv21);
1988             felec            = _mm256_mul_ps(velec,rinvsq21);
1989
1990             fscal            = felec;
1991
1992             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1993
1994             /* Calculate temporary vectorial force */
1995             tx               = _mm256_mul_ps(fscal,dx21);
1996             ty               = _mm256_mul_ps(fscal,dy21);
1997             tz               = _mm256_mul_ps(fscal,dz21);
1998
1999             /* Update vectorial force */
2000             fix2             = _mm256_add_ps(fix2,tx);
2001             fiy2             = _mm256_add_ps(fiy2,ty);
2002             fiz2             = _mm256_add_ps(fiz2,tz);
2003
2004             fjx1             = _mm256_add_ps(fjx1,tx);
2005             fjy1             = _mm256_add_ps(fjy1,ty);
2006             fjz1             = _mm256_add_ps(fjz1,tz);
2007
2008             /**************************
2009              * CALCULATE INTERACTIONS *
2010              **************************/
2011
2012             /* COULOMB ELECTROSTATICS */
2013             velec            = _mm256_mul_ps(qq22,rinv22);
2014             felec            = _mm256_mul_ps(velec,rinvsq22);
2015
2016             fscal            = felec;
2017
2018             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2019
2020             /* Calculate temporary vectorial force */
2021             tx               = _mm256_mul_ps(fscal,dx22);
2022             ty               = _mm256_mul_ps(fscal,dy22);
2023             tz               = _mm256_mul_ps(fscal,dz22);
2024
2025             /* Update vectorial force */
2026             fix2             = _mm256_add_ps(fix2,tx);
2027             fiy2             = _mm256_add_ps(fiy2,ty);
2028             fiz2             = _mm256_add_ps(fiz2,tz);
2029
2030             fjx2             = _mm256_add_ps(fjx2,tx);
2031             fjy2             = _mm256_add_ps(fjy2,ty);
2032             fjz2             = _mm256_add_ps(fjz2,tz);
2033
2034             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2035             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2036             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2037             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2038             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2039             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2040             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2041             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2042
2043             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2044                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2045
2046             /* Inner loop uses 262 flops */
2047         }
2048
2049         /* End of innermost loop */
2050
2051         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2052                                                  f+i_coord_offset,fshift+i_shift_offset);
2053
2054         /* Increment number of inner iterations */
2055         inneriter                  += j_index_end - j_index_start;
2056
2057         /* Outer loop uses 18 flops */
2058     }
2059
2060     /* Increment number of outer iterations */
2061     outeriter        += nri;
2062
2063     /* Update outer/inner flops */
2064
2065     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*262);
2066 }