src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwCSTab_GeomW3W3_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3W3_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwCSTab_GeomW3W3_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  95     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  97     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 100     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 101     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 102     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 105     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     int              nvdwtype;
 110     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 111     int              *vdwtype;
 112     real             *vdwparam;
 113     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 114     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 115     __m256i          vfitab;
 116     __m128i          vfitab_lo,vfitab_hi;
 117     __m128i          ifour       = _mm_set1_epi32(4);
 118     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 119     real             *vftab;
 120     __m256i          ewitab;
 121     __m128i          ewitab_lo,ewitab_hi;
 122     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 123     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 124     real             *ewtab;
 125     __m256           dummy_mask,cutoff_mask;
 126     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 127     __m256           one     = _mm256_set1_ps(1.0);
 128     __m256           two     = _mm256_set1_ps(2.0);
 129     x                = xx[0];
 130     f                = ff[0];
 131
 132     nri              = nlist->nri;
 133     iinr             = nlist->iinr;
 134     jindex           = nlist->jindex;
 135     jjnr             = nlist->jjnr;
 136     shiftidx         = nlist->shift;
 137     gid              = nlist->gid;
 138     shiftvec         = fr->shift_vec[0];
 139     fshift           = fr->fshift[0];
 140     facel            = _mm256_set1_ps(fr->epsfac);
 141     charge           = mdatoms->chargeA;
 142     nvdwtype         = fr->ntype;
 143     vdwparam         = fr->nbfp;
 144     vdwtype          = mdatoms->typeA;
 145
 146     vftab            = kernel_data->table_vdw->data;
 147     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 148
 149     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 150     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 151     beta2            = _mm256_mul_ps(beta,beta);
 152     beta3            = _mm256_mul_ps(beta,beta2);
 153
 154     ewtab            = fr->ic->tabq_coul_FDV0;
 155     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 156     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 157
 158     /* Setup water-specific parameters */
 159     inr              = nlist->iinr[0];
 160     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 161     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 162     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 163     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 164
 165     jq0              = _mm256_set1_ps(charge[inr+0]);
 166     jq1              = _mm256_set1_ps(charge[inr+1]);
 167     jq2              = _mm256_set1_ps(charge[inr+2]);
 168     vdwjidx0A        = 2*vdwtype[inr+0];
 169     qq00             = _mm256_mul_ps(iq0,jq0);
 170     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 171     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 172     qq01             = _mm256_mul_ps(iq0,jq1);
 173     qq02             = _mm256_mul_ps(iq0,jq2);
 174     qq10             = _mm256_mul_ps(iq1,jq0);
 175     qq11             = _mm256_mul_ps(iq1,jq1);
 176     qq12             = _mm256_mul_ps(iq1,jq2);
 177     qq20             = _mm256_mul_ps(iq2,jq0);
 178     qq21             = _mm256_mul_ps(iq2,jq1);
 179     qq22             = _mm256_mul_ps(iq2,jq2);
 180
 181     /* Avoid stupid compiler warnings */
 182     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 183     j_coord_offsetA = 0;
 184     j_coord_offsetB = 0;
 185     j_coord_offsetC = 0;
 186     j_coord_offsetD = 0;
 187     j_coord_offsetE = 0;
 188     j_coord_offsetF = 0;
 189     j_coord_offsetG = 0;
 190     j_coord_offsetH = 0;
 191
 192     outeriter        = 0;
 193     inneriter        = 0;
 194
 195     for(iidx=0;iidx<4*DIM;iidx++)
 196     {
 197         scratch[iidx] = 0.0;
 198     }
 199
 200     /* Start outer loop over neighborlists */
 201     for(iidx=0; iidx<nri; iidx++)
 202     {
 203         /* Load shift vector for this list */
 204         i_shift_offset   = DIM*shiftidx[iidx];
 205
 206         /* Load limits for loop over neighbors */
 207         j_index_start    = jindex[iidx];
 208         j_index_end      = jindex[iidx+1];
 209
 210         /* Get outer coordinate index */
 211         inr              = iinr[iidx];
 212         i_coord_offset   = DIM*inr;
 213
 214         /* Load i particle coords and add shift vector */
 215         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 216                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 217
 218         fix0             = _mm256_setzero_ps();
 219         fiy0             = _mm256_setzero_ps();
 220         fiz0             = _mm256_setzero_ps();
 221         fix1             = _mm256_setzero_ps();
 222         fiy1             = _mm256_setzero_ps();
 223         fiz1             = _mm256_setzero_ps();
 224         fix2             = _mm256_setzero_ps();
 225         fiy2             = _mm256_setzero_ps();
 226         fiz2             = _mm256_setzero_ps();
 227
 228         /* Reset potential sums */
 229         velecsum         = _mm256_setzero_ps();
 230         vvdwsum          = _mm256_setzero_ps();
 231
 232         /* Start inner kernel loop */
 233         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 234         {
 235
 236             /* Get j neighbor index, and coordinate index */
 237             jnrA             = jjnr[jidx];
 238             jnrB             = jjnr[jidx+1];
 239             jnrC             = jjnr[jidx+2];
 240             jnrD             = jjnr[jidx+3];
 241             jnrE             = jjnr[jidx+4];
 242             jnrF             = jjnr[jidx+5];
 243             jnrG             = jjnr[jidx+6];
 244             jnrH             = jjnr[jidx+7];
 245             j_coord_offsetA  = DIM*jnrA;
 246             j_coord_offsetB  = DIM*jnrB;
 247             j_coord_offsetC  = DIM*jnrC;
 248             j_coord_offsetD  = DIM*jnrD;
 249             j_coord_offsetE  = DIM*jnrE;
 250             j_coord_offsetF  = DIM*jnrF;
 251             j_coord_offsetG  = DIM*jnrG;
 252             j_coord_offsetH  = DIM*jnrH;
 253
 254             /* load j atom coordinates */
 255             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 256                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 257                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 258                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 259                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 260
 261             /* Calculate displacement vector */
 262             dx00             = _mm256_sub_ps(ix0,jx0);
 263             dy00             = _mm256_sub_ps(iy0,jy0);
 264             dz00             = _mm256_sub_ps(iz0,jz0);
 265             dx01             = _mm256_sub_ps(ix0,jx1);
 266             dy01             = _mm256_sub_ps(iy0,jy1);
 267             dz01             = _mm256_sub_ps(iz0,jz1);
 268             dx02             = _mm256_sub_ps(ix0,jx2);
 269             dy02             = _mm256_sub_ps(iy0,jy2);
 270             dz02             = _mm256_sub_ps(iz0,jz2);
 271             dx10             = _mm256_sub_ps(ix1,jx0);
 272             dy10             = _mm256_sub_ps(iy1,jy0);
 273             dz10             = _mm256_sub_ps(iz1,jz0);
 274             dx11             = _mm256_sub_ps(ix1,jx1);
 275             dy11             = _mm256_sub_ps(iy1,jy1);
 276             dz11             = _mm256_sub_ps(iz1,jz1);
 277             dx12             = _mm256_sub_ps(ix1,jx2);
 278             dy12             = _mm256_sub_ps(iy1,jy2);
 279             dz12             = _mm256_sub_ps(iz1,jz2);
 280             dx20             = _mm256_sub_ps(ix2,jx0);
 281             dy20             = _mm256_sub_ps(iy2,jy0);
 282             dz20             = _mm256_sub_ps(iz2,jz0);
 283             dx21             = _mm256_sub_ps(ix2,jx1);
 284             dy21             = _mm256_sub_ps(iy2,jy1);
 285             dz21             = _mm256_sub_ps(iz2,jz1);
 286             dx22             = _mm256_sub_ps(ix2,jx2);
 287             dy22             = _mm256_sub_ps(iy2,jy2);
 288             dz22             = _mm256_sub_ps(iz2,jz2);
 289
 290             /* Calculate squared distance and things based on it */
 291             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 292             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 293             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 294             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 295             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 296             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 297             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 298             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 299             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 300
 301             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 302             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 303             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 304             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 305             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 306             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 307             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 308             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 309             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 310
 311             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 312             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 313             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 314             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 315             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 316             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 317             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 318             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 319             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 320
 321             fjx0             = _mm256_setzero_ps();
 322             fjy0             = _mm256_setzero_ps();
 323             fjz0             = _mm256_setzero_ps();
 324             fjx1             = _mm256_setzero_ps();
 325             fjy1             = _mm256_setzero_ps();
 326             fjz1             = _mm256_setzero_ps();
 327             fjx2             = _mm256_setzero_ps();
 328             fjy2             = _mm256_setzero_ps();
 329             fjz2             = _mm256_setzero_ps();
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             r00              = _mm256_mul_ps(rsq00,rinv00);
 336
 337             /* Calculate table index by multiplying r with table scale and truncate to integer */
 338             rt               = _mm256_mul_ps(r00,vftabscale);
 339             vfitab           = _mm256_cvttps_epi32(rt);
 340             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 341             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 342             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 343             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 344             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 345             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 346
 347             /* EWALD ELECTROSTATICS */
 348
 349             /* Analytical PME correction */
 350             zeta2            = _mm256_mul_ps(beta2,rsq00);
 351             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 352             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 353             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 354             felec            = _mm256_mul_ps(qq00,felec);
 355             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 356             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 357             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 358             velec            = _mm256_mul_ps(qq00,velec);
 359
 360             /* CUBIC SPLINE TABLE DISPERSION */
 361             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 362                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 363             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 364                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 365             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 366                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 367             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 368                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 369             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 370             Heps             = _mm256_mul_ps(vfeps,H);
 371             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 372             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 373             vvdw6            = _mm256_mul_ps(c6_00,VV);
 374             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 375             fvdw6            = _mm256_mul_ps(c6_00,FF);
 376
 377             /* CUBIC SPLINE TABLE REPULSION */
 378             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 379             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 380             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 381                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 382             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 383                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 384             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 385                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 386             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 387                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 388             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 389             Heps             = _mm256_mul_ps(vfeps,H);
 390             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 391             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 392             vvdw12           = _mm256_mul_ps(c12_00,VV);
 393             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 394             fvdw12           = _mm256_mul_ps(c12_00,FF);
 395             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 396             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 397
 398             /* Update potential sum for this i atom from the interaction with this j atom. */
 399             velecsum         = _mm256_add_ps(velecsum,velec);
 400             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 401
 402             fscal            = _mm256_add_ps(felec,fvdw);
 403
 404             /* Calculate temporary vectorial force */
 405             tx               = _mm256_mul_ps(fscal,dx00);
 406             ty               = _mm256_mul_ps(fscal,dy00);
 407             tz               = _mm256_mul_ps(fscal,dz00);
 408
 409             /* Update vectorial force */
 410             fix0             = _mm256_add_ps(fix0,tx);
 411             fiy0             = _mm256_add_ps(fiy0,ty);
 412             fiz0             = _mm256_add_ps(fiz0,tz);
 413
 414             fjx0             = _mm256_add_ps(fjx0,tx);
 415             fjy0             = _mm256_add_ps(fjy0,ty);
 416             fjz0             = _mm256_add_ps(fjz0,tz);
 417
 418             /**************************
 419              * CALCULATE INTERACTIONS *
 420              **************************/
 421
 422             r01              = _mm256_mul_ps(rsq01,rinv01);
 423
 424             /* EWALD ELECTROSTATICS */
 425
 426             /* Analytical PME correction */
 427             zeta2            = _mm256_mul_ps(beta2,rsq01);
 428             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 429             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 430             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 431             felec            = _mm256_mul_ps(qq01,felec);
 432             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 433             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 434             velec            = _mm256_sub_ps(rinv01,pmecorrV);
 435             velec            = _mm256_mul_ps(qq01,velec);
 436
 437             /* Update potential sum for this i atom from the interaction with this j atom. */
 438             velecsum         = _mm256_add_ps(velecsum,velec);
 439
 440             fscal            = felec;
 441
 442             /* Calculate temporary vectorial force */
 443             tx               = _mm256_mul_ps(fscal,dx01);
 444             ty               = _mm256_mul_ps(fscal,dy01);
 445             tz               = _mm256_mul_ps(fscal,dz01);
 446
 447             /* Update vectorial force */
 448             fix0             = _mm256_add_ps(fix0,tx);
 449             fiy0             = _mm256_add_ps(fiy0,ty);
 450             fiz0             = _mm256_add_ps(fiz0,tz);
 451
 452             fjx1             = _mm256_add_ps(fjx1,tx);
 453             fjy1             = _mm256_add_ps(fjy1,ty);
 454             fjz1             = _mm256_add_ps(fjz1,tz);
 455
 456             /**************************
 457              * CALCULATE INTERACTIONS *
 458              **************************/
 459
 460             r02              = _mm256_mul_ps(rsq02,rinv02);
 461
 462             /* EWALD ELECTROSTATICS */
 463
 464             /* Analytical PME correction */
 465             zeta2            = _mm256_mul_ps(beta2,rsq02);
 466             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 467             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 468             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 469             felec            = _mm256_mul_ps(qq02,felec);
 470             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 471             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 472             velec            = _mm256_sub_ps(rinv02,pmecorrV);
 473             velec            = _mm256_mul_ps(qq02,velec);
 474
 475             /* Update potential sum for this i atom from the interaction with this j atom. */
 476             velecsum         = _mm256_add_ps(velecsum,velec);
 477
 478             fscal            = felec;
 479
 480             /* Calculate temporary vectorial force */
 481             tx               = _mm256_mul_ps(fscal,dx02);
 482             ty               = _mm256_mul_ps(fscal,dy02);
 483             tz               = _mm256_mul_ps(fscal,dz02);
 484
 485             /* Update vectorial force */
 486             fix0             = _mm256_add_ps(fix0,tx);
 487             fiy0             = _mm256_add_ps(fiy0,ty);
 488             fiz0             = _mm256_add_ps(fiz0,tz);
 489
 490             fjx2             = _mm256_add_ps(fjx2,tx);
 491             fjy2             = _mm256_add_ps(fjy2,ty);
 492             fjz2             = _mm256_add_ps(fjz2,tz);
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             r10              = _mm256_mul_ps(rsq10,rinv10);
 499
 500             /* EWALD ELECTROSTATICS */
 501
 502             /* Analytical PME correction */
 503             zeta2            = _mm256_mul_ps(beta2,rsq10);
 504             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 505             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 506             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 507             felec            = _mm256_mul_ps(qq10,felec);
 508             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 509             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 510             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 511             velec            = _mm256_mul_ps(qq10,velec);
 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,dx10);
 520             ty               = _mm256_mul_ps(fscal,dy10);
 521             tz               = _mm256_mul_ps(fscal,dz10);
 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             fjx0             = _mm256_add_ps(fjx0,tx);
 529             fjy0             = _mm256_add_ps(fjy0,ty);
 530             fjz0             = _mm256_add_ps(fjz0,tz);
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             r11              = _mm256_mul_ps(rsq11,rinv11);
 537
 538             /* EWALD ELECTROSTATICS */
 539
 540             /* Analytical PME correction */
 541             zeta2            = _mm256_mul_ps(beta2,rsq11);
 542             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 543             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 544             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 545             felec            = _mm256_mul_ps(qq11,felec);
 546             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 547             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 548             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 549             velec            = _mm256_mul_ps(qq11,velec);
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             velecsum         = _mm256_add_ps(velecsum,velec);
 553
 554             fscal            = felec;
 555
 556             /* Calculate temporary vectorial force */
 557             tx               = _mm256_mul_ps(fscal,dx11);
 558             ty               = _mm256_mul_ps(fscal,dy11);
 559             tz               = _mm256_mul_ps(fscal,dz11);
 560
 561             /* Update vectorial force */
 562             fix1             = _mm256_add_ps(fix1,tx);
 563             fiy1             = _mm256_add_ps(fiy1,ty);
 564             fiz1             = _mm256_add_ps(fiz1,tz);
 565
 566             fjx1             = _mm256_add_ps(fjx1,tx);
 567             fjy1             = _mm256_add_ps(fjy1,ty);
 568             fjz1             = _mm256_add_ps(fjz1,tz);
 569
 570             /**************************
 571              * CALCULATE INTERACTIONS *
 572              **************************/
 573
 574             r12              = _mm256_mul_ps(rsq12,rinv12);
 575
 576             /* EWALD ELECTROSTATICS */
 577
 578             /* Analytical PME correction */
 579             zeta2            = _mm256_mul_ps(beta2,rsq12);
 580             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 581             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 582             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 583             felec            = _mm256_mul_ps(qq12,felec);
 584             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 585             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 586             velec            = _mm256_sub_ps(rinv12,pmecorrV);
 587             velec            = _mm256_mul_ps(qq12,velec);
 588
 589             /* Update potential sum for this i atom from the interaction with this j atom. */
 590             velecsum         = _mm256_add_ps(velecsum,velec);
 591
 592             fscal            = felec;
 593
 594             /* Calculate temporary vectorial force */
 595             tx               = _mm256_mul_ps(fscal,dx12);
 596             ty               = _mm256_mul_ps(fscal,dy12);
 597             tz               = _mm256_mul_ps(fscal,dz12);
 598
 599             /* Update vectorial force */
 600             fix1             = _mm256_add_ps(fix1,tx);
 601             fiy1             = _mm256_add_ps(fiy1,ty);
 602             fiz1             = _mm256_add_ps(fiz1,tz);
 603
 604             fjx2             = _mm256_add_ps(fjx2,tx);
 605             fjy2             = _mm256_add_ps(fjy2,ty);
 606             fjz2             = _mm256_add_ps(fjz2,tz);
 607
 608             /**************************
 609              * CALCULATE INTERACTIONS *
 610              **************************/
 611
 612             r20              = _mm256_mul_ps(rsq20,rinv20);
 613
 614             /* EWALD ELECTROSTATICS */
 615
 616             /* Analytical PME correction */
 617             zeta2            = _mm256_mul_ps(beta2,rsq20);
 618             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 619             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 620             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 621             felec            = _mm256_mul_ps(qq20,felec);
 622             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 623             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 624             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 625             velec            = _mm256_mul_ps(qq20,velec);
 626
 627             /* Update potential sum for this i atom from the interaction with this j atom. */
 628             velecsum         = _mm256_add_ps(velecsum,velec);
 629
 630             fscal            = felec;
 631
 632             /* Calculate temporary vectorial force */
 633             tx               = _mm256_mul_ps(fscal,dx20);
 634             ty               = _mm256_mul_ps(fscal,dy20);
 635             tz               = _mm256_mul_ps(fscal,dz20);
 636
 637             /* Update vectorial force */
 638             fix2             = _mm256_add_ps(fix2,tx);
 639             fiy2             = _mm256_add_ps(fiy2,ty);
 640             fiz2             = _mm256_add_ps(fiz2,tz);
 641
 642             fjx0             = _mm256_add_ps(fjx0,tx);
 643             fjy0             = _mm256_add_ps(fjy0,ty);
 644             fjz0             = _mm256_add_ps(fjz0,tz);
 645
 646             /**************************
 647              * CALCULATE INTERACTIONS *
 648              **************************/
 649
 650             r21              = _mm256_mul_ps(rsq21,rinv21);
 651
 652             /* EWALD ELECTROSTATICS */
 653
 654             /* Analytical PME correction */
 655             zeta2            = _mm256_mul_ps(beta2,rsq21);
 656             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 657             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 658             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 659             felec            = _mm256_mul_ps(qq21,felec);
 660             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 661             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 662             velec            = _mm256_sub_ps(rinv21,pmecorrV);
 663             velec            = _mm256_mul_ps(qq21,velec);
 664
 665             /* Update potential sum for this i atom from the interaction with this j atom. */
 666             velecsum         = _mm256_add_ps(velecsum,velec);
 667
 668             fscal            = felec;
 669
 670             /* Calculate temporary vectorial force */
 671             tx               = _mm256_mul_ps(fscal,dx21);
 672             ty               = _mm256_mul_ps(fscal,dy21);
 673             tz               = _mm256_mul_ps(fscal,dz21);
 674
 675             /* Update vectorial force */
 676             fix2             = _mm256_add_ps(fix2,tx);
 677             fiy2             = _mm256_add_ps(fiy2,ty);
 678             fiz2             = _mm256_add_ps(fiz2,tz);
 679
 680             fjx1             = _mm256_add_ps(fjx1,tx);
 681             fjy1             = _mm256_add_ps(fjy1,ty);
 682             fjz1             = _mm256_add_ps(fjz1,tz);
 683
 684             /**************************
 685              * CALCULATE INTERACTIONS *
 686              **************************/
 687
 688             r22              = _mm256_mul_ps(rsq22,rinv22);
 689
 690             /* EWALD ELECTROSTATICS */
 691
 692             /* Analytical PME correction */
 693             zeta2            = _mm256_mul_ps(beta2,rsq22);
 694             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 695             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 696             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 697             felec            = _mm256_mul_ps(qq22,felec);
 698             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 699             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 700             velec            = _mm256_sub_ps(rinv22,pmecorrV);
 701             velec            = _mm256_mul_ps(qq22,velec);
 702
 703             /* Update potential sum for this i atom from the interaction with this j atom. */
 704             velecsum         = _mm256_add_ps(velecsum,velec);
 705
 706             fscal            = felec;
 707
 708             /* Calculate temporary vectorial force */
 709             tx               = _mm256_mul_ps(fscal,dx22);
 710             ty               = _mm256_mul_ps(fscal,dy22);
 711             tz               = _mm256_mul_ps(fscal,dz22);
 712
 713             /* Update vectorial force */
 714             fix2             = _mm256_add_ps(fix2,tx);
 715             fiy2             = _mm256_add_ps(fiy2,ty);
 716             fiz2             = _mm256_add_ps(fiz2,tz);
 717
 718             fjx2             = _mm256_add_ps(fjx2,tx);
 719             fjy2             = _mm256_add_ps(fjy2,ty);
 720             fjz2             = _mm256_add_ps(fjz2,tz);
 721
 722             fjptrA             = f+j_coord_offsetA;
 723             fjptrB             = f+j_coord_offsetB;
 724             fjptrC             = f+j_coord_offsetC;
 725             fjptrD             = f+j_coord_offsetD;
 726             fjptrE             = f+j_coord_offsetE;
 727             fjptrF             = f+j_coord_offsetF;
 728             fjptrG             = f+j_coord_offsetG;
 729             fjptrH             = f+j_coord_offsetH;
 730
 731             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 732                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 733
 734             /* Inner loop uses 790 flops */
 735         }
 736
 737         if(jidx<j_index_end)
 738         {
 739
 740             /* Get j neighbor index, and coordinate index */
 741             jnrlistA         = jjnr[jidx];
 742             jnrlistB         = jjnr[jidx+1];
 743             jnrlistC         = jjnr[jidx+2];
 744             jnrlistD         = jjnr[jidx+3];
 745             jnrlistE         = jjnr[jidx+4];
 746             jnrlistF         = jjnr[jidx+5];
 747             jnrlistG         = jjnr[jidx+6];
 748             jnrlistH         = jjnr[jidx+7];
 749             /* Sign of each element will be negative for non-real atoms.
 750              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 751              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 752              */
 753             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 754                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 755
 756             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 757             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 758             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 759             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 760             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 761             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 762             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 763             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 764             j_coord_offsetA  = DIM*jnrA;
 765             j_coord_offsetB  = DIM*jnrB;
 766             j_coord_offsetC  = DIM*jnrC;
 767             j_coord_offsetD  = DIM*jnrD;
 768             j_coord_offsetE  = DIM*jnrE;
 769             j_coord_offsetF  = DIM*jnrF;
 770             j_coord_offsetG  = DIM*jnrG;
 771             j_coord_offsetH  = DIM*jnrH;
 772
 773             /* load j atom coordinates */
 774             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 775                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 776                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 777                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 778                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 779
 780             /* Calculate displacement vector */
 781             dx00             = _mm256_sub_ps(ix0,jx0);
 782             dy00             = _mm256_sub_ps(iy0,jy0);
 783             dz00             = _mm256_sub_ps(iz0,jz0);
 784             dx01             = _mm256_sub_ps(ix0,jx1);
 785             dy01             = _mm256_sub_ps(iy0,jy1);
 786             dz01             = _mm256_sub_ps(iz0,jz1);
 787             dx02             = _mm256_sub_ps(ix0,jx2);
 788             dy02             = _mm256_sub_ps(iy0,jy2);
 789             dz02             = _mm256_sub_ps(iz0,jz2);
 790             dx10             = _mm256_sub_ps(ix1,jx0);
 791             dy10             = _mm256_sub_ps(iy1,jy0);
 792             dz10             = _mm256_sub_ps(iz1,jz0);
 793             dx11             = _mm256_sub_ps(ix1,jx1);
 794             dy11             = _mm256_sub_ps(iy1,jy1);
 795             dz11             = _mm256_sub_ps(iz1,jz1);
 796             dx12             = _mm256_sub_ps(ix1,jx2);
 797             dy12             = _mm256_sub_ps(iy1,jy2);
 798             dz12             = _mm256_sub_ps(iz1,jz2);
 799             dx20             = _mm256_sub_ps(ix2,jx0);
 800             dy20             = _mm256_sub_ps(iy2,jy0);
 801             dz20             = _mm256_sub_ps(iz2,jz0);
 802             dx21             = _mm256_sub_ps(ix2,jx1);
 803             dy21             = _mm256_sub_ps(iy2,jy1);
 804             dz21             = _mm256_sub_ps(iz2,jz1);
 805             dx22             = _mm256_sub_ps(ix2,jx2);
 806             dy22             = _mm256_sub_ps(iy2,jy2);
 807             dz22             = _mm256_sub_ps(iz2,jz2);
 808
 809             /* Calculate squared distance and things based on it */
 810             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 811             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 812             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 813             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 814             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 815             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 816             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 817             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 818             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 819
 820             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 821             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 822             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 823             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 824             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 825             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 826             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 827             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 828             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 829
 830             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 831             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 832             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 833             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 834             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 835             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 836             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 837             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 838             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 839
 840             fjx0             = _mm256_setzero_ps();
 841             fjy0             = _mm256_setzero_ps();
 842             fjz0             = _mm256_setzero_ps();
 843             fjx1             = _mm256_setzero_ps();
 844             fjy1             = _mm256_setzero_ps();
 845             fjz1             = _mm256_setzero_ps();
 846             fjx2             = _mm256_setzero_ps();
 847             fjy2             = _mm256_setzero_ps();
 848             fjz2             = _mm256_setzero_ps();
 849
 850             /**************************
 851              * CALCULATE INTERACTIONS *
 852              **************************/
 853
 854             r00              = _mm256_mul_ps(rsq00,rinv00);
 855             r00              = _mm256_andnot_ps(dummy_mask,r00);
 856
 857             /* Calculate table index by multiplying r with table scale and truncate to integer */
 858             rt               = _mm256_mul_ps(r00,vftabscale);
 859             vfitab           = _mm256_cvttps_epi32(rt);
 860             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 861             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 862             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 863             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 864             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 865             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 866
 867             /* EWALD ELECTROSTATICS */
 868
 869             /* Analytical PME correction */
 870             zeta2            = _mm256_mul_ps(beta2,rsq00);
 871             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 872             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 873             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 874             felec            = _mm256_mul_ps(qq00,felec);
 875             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 876             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 877             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 878             velec            = _mm256_mul_ps(qq00,velec);
 879
 880             /* CUBIC SPLINE TABLE DISPERSION */
 881             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 882                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 883             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 884                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 885             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 886                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 887             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 888                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 889             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 890             Heps             = _mm256_mul_ps(vfeps,H);
 891             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 892             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 893             vvdw6            = _mm256_mul_ps(c6_00,VV);
 894             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 895             fvdw6            = _mm256_mul_ps(c6_00,FF);
 896
 897             /* CUBIC SPLINE TABLE REPULSION */
 898             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 899             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 900             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 901                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 902             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 903                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 904             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 905                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 906             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 907                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 908             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 909             Heps             = _mm256_mul_ps(vfeps,H);
 910             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 911             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 912             vvdw12           = _mm256_mul_ps(c12_00,VV);
 913             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 914             fvdw12           = _mm256_mul_ps(c12_00,FF);
 915             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 916             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 917
 918             /* Update potential sum for this i atom from the interaction with this j atom. */
 919             velec            = _mm256_andnot_ps(dummy_mask,velec);
 920             velecsum         = _mm256_add_ps(velecsum,velec);
 921             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 922             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 923
 924             fscal            = _mm256_add_ps(felec,fvdw);
 925
 926             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 927
 928             /* Calculate temporary vectorial force */
 929             tx               = _mm256_mul_ps(fscal,dx00);
 930             ty               = _mm256_mul_ps(fscal,dy00);
 931             tz               = _mm256_mul_ps(fscal,dz00);
 932
 933             /* Update vectorial force */
 934             fix0             = _mm256_add_ps(fix0,tx);
 935             fiy0             = _mm256_add_ps(fiy0,ty);
 936             fiz0             = _mm256_add_ps(fiz0,tz);
 937
 938             fjx0             = _mm256_add_ps(fjx0,tx);
 939             fjy0             = _mm256_add_ps(fjy0,ty);
 940             fjz0             = _mm256_add_ps(fjz0,tz);
 941
 942             /**************************
 943              * CALCULATE INTERACTIONS *
 944              **************************/
 945
 946             r01              = _mm256_mul_ps(rsq01,rinv01);
 947             r01              = _mm256_andnot_ps(dummy_mask,r01);
 948
 949             /* EWALD ELECTROSTATICS */
 950
 951             /* Analytical PME correction */
 952             zeta2            = _mm256_mul_ps(beta2,rsq01);
 953             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 954             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 955             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 956             felec            = _mm256_mul_ps(qq01,felec);
 957             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 958             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 959             velec            = _mm256_sub_ps(rinv01,pmecorrV);
 960             velec            = _mm256_mul_ps(qq01,velec);
 961
 962             /* Update potential sum for this i atom from the interaction with this j atom. */
 963             velec            = _mm256_andnot_ps(dummy_mask,velec);
 964             velecsum         = _mm256_add_ps(velecsum,velec);
 965
 966             fscal            = felec;
 967
 968             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 969
 970             /* Calculate temporary vectorial force */
 971             tx               = _mm256_mul_ps(fscal,dx01);
 972             ty               = _mm256_mul_ps(fscal,dy01);
 973             tz               = _mm256_mul_ps(fscal,dz01);
 974
 975             /* Update vectorial force */
 976             fix0             = _mm256_add_ps(fix0,tx);
 977             fiy0             = _mm256_add_ps(fiy0,ty);
 978             fiz0             = _mm256_add_ps(fiz0,tz);
 979
 980             fjx1             = _mm256_add_ps(fjx1,tx);
 981             fjy1             = _mm256_add_ps(fjy1,ty);
 982             fjz1             = _mm256_add_ps(fjz1,tz);
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             r02              = _mm256_mul_ps(rsq02,rinv02);
 989             r02              = _mm256_andnot_ps(dummy_mask,r02);
 990
 991             /* EWALD ELECTROSTATICS */
 992
 993             /* Analytical PME correction */
 994             zeta2            = _mm256_mul_ps(beta2,rsq02);
 995             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 996             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 997             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 998             felec            = _mm256_mul_ps(qq02,felec);
 999             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1000             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1001             velec            = _mm256_sub_ps(rinv02,pmecorrV);
1002             velec            = _mm256_mul_ps(qq02,velec);
1003
1004             /* Update potential sum for this i atom from the interaction with this j atom. */
1005             velec            = _mm256_andnot_ps(dummy_mask,velec);
1006             velecsum         = _mm256_add_ps(velecsum,velec);
1007
1008             fscal            = felec;
1009
1010             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1011
1012             /* Calculate temporary vectorial force */
1013             tx               = _mm256_mul_ps(fscal,dx02);
1014             ty               = _mm256_mul_ps(fscal,dy02);
1015             tz               = _mm256_mul_ps(fscal,dz02);
1016
1017             /* Update vectorial force */
1018             fix0             = _mm256_add_ps(fix0,tx);
1019             fiy0             = _mm256_add_ps(fiy0,ty);
1020             fiz0             = _mm256_add_ps(fiz0,tz);
1021
1022             fjx2             = _mm256_add_ps(fjx2,tx);
1023             fjy2             = _mm256_add_ps(fjy2,ty);
1024             fjz2             = _mm256_add_ps(fjz2,tz);
1025
1026             /**************************
1027              * CALCULATE INTERACTIONS *
1028              **************************/
1029
1030             r10              = _mm256_mul_ps(rsq10,rinv10);
1031             r10              = _mm256_andnot_ps(dummy_mask,r10);
1032
1033             /* EWALD ELECTROSTATICS */
1034
1035             /* Analytical PME correction */
1036             zeta2            = _mm256_mul_ps(beta2,rsq10);
1037             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1038             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1039             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1040             felec            = _mm256_mul_ps(qq10,felec);
1041             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1042             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1043             velec            = _mm256_sub_ps(rinv10,pmecorrV);
1044             velec            = _mm256_mul_ps(qq10,velec);
1045
1046             /* Update potential sum for this i atom from the interaction with this j atom. */
1047             velec            = _mm256_andnot_ps(dummy_mask,velec);
1048             velecsum         = _mm256_add_ps(velecsum,velec);
1049
1050             fscal            = felec;
1051
1052             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1053
1054             /* Calculate temporary vectorial force */
1055             tx               = _mm256_mul_ps(fscal,dx10);
1056             ty               = _mm256_mul_ps(fscal,dy10);
1057             tz               = _mm256_mul_ps(fscal,dz10);
1058
1059             /* Update vectorial force */
1060             fix1             = _mm256_add_ps(fix1,tx);
1061             fiy1             = _mm256_add_ps(fiy1,ty);
1062             fiz1             = _mm256_add_ps(fiz1,tz);
1063
1064             fjx0             = _mm256_add_ps(fjx0,tx);
1065             fjy0             = _mm256_add_ps(fjy0,ty);
1066             fjz0             = _mm256_add_ps(fjz0,tz);
1067
1068             /**************************
1069              * CALCULATE INTERACTIONS *
1070              **************************/
1071
1072             r11              = _mm256_mul_ps(rsq11,rinv11);
1073             r11              = _mm256_andnot_ps(dummy_mask,r11);
1074
1075             /* EWALD ELECTROSTATICS */
1076
1077             /* Analytical PME correction */
1078             zeta2            = _mm256_mul_ps(beta2,rsq11);
1079             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1080             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1081             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1082             felec            = _mm256_mul_ps(qq11,felec);
1083             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1084             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1085             velec            = _mm256_sub_ps(rinv11,pmecorrV);
1086             velec            = _mm256_mul_ps(qq11,velec);
1087
1088             /* Update potential sum for this i atom from the interaction with this j atom. */
1089             velec            = _mm256_andnot_ps(dummy_mask,velec);
1090             velecsum         = _mm256_add_ps(velecsum,velec);
1091
1092             fscal            = felec;
1093
1094             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1095
1096             /* Calculate temporary vectorial force */
1097             tx               = _mm256_mul_ps(fscal,dx11);
1098             ty               = _mm256_mul_ps(fscal,dy11);
1099             tz               = _mm256_mul_ps(fscal,dz11);
1100
1101             /* Update vectorial force */
1102             fix1             = _mm256_add_ps(fix1,tx);
1103             fiy1             = _mm256_add_ps(fiy1,ty);
1104             fiz1             = _mm256_add_ps(fiz1,tz);
1105
1106             fjx1             = _mm256_add_ps(fjx1,tx);
1107             fjy1             = _mm256_add_ps(fjy1,ty);
1108             fjz1             = _mm256_add_ps(fjz1,tz);
1109
1110             /**************************
1111              * CALCULATE INTERACTIONS *
1112              **************************/
1113
1114             r12              = _mm256_mul_ps(rsq12,rinv12);
1115             r12              = _mm256_andnot_ps(dummy_mask,r12);
1116
1117             /* EWALD ELECTROSTATICS */
1118
1119             /* Analytical PME correction */
1120             zeta2            = _mm256_mul_ps(beta2,rsq12);
1121             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1122             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1123             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1124             felec            = _mm256_mul_ps(qq12,felec);
1125             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1126             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1127             velec            = _mm256_sub_ps(rinv12,pmecorrV);
1128             velec            = _mm256_mul_ps(qq12,velec);
1129
1130             /* Update potential sum for this i atom from the interaction with this j atom. */
1131             velec            = _mm256_andnot_ps(dummy_mask,velec);
1132             velecsum         = _mm256_add_ps(velecsum,velec);
1133
1134             fscal            = felec;
1135
1136             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1137
1138             /* Calculate temporary vectorial force */
1139             tx               = _mm256_mul_ps(fscal,dx12);
1140             ty               = _mm256_mul_ps(fscal,dy12);
1141             tz               = _mm256_mul_ps(fscal,dz12);
1142
1143             /* Update vectorial force */
1144             fix1             = _mm256_add_ps(fix1,tx);
1145             fiy1             = _mm256_add_ps(fiy1,ty);
1146             fiz1             = _mm256_add_ps(fiz1,tz);
1147
1148             fjx2             = _mm256_add_ps(fjx2,tx);
1149             fjy2             = _mm256_add_ps(fjy2,ty);
1150             fjz2             = _mm256_add_ps(fjz2,tz);
1151
1152             /**************************
1153              * CALCULATE INTERACTIONS *
1154              **************************/
1155
1156             r20              = _mm256_mul_ps(rsq20,rinv20);
1157             r20              = _mm256_andnot_ps(dummy_mask,r20);
1158
1159             /* EWALD ELECTROSTATICS */
1160
1161             /* Analytical PME correction */
1162             zeta2            = _mm256_mul_ps(beta2,rsq20);
1163             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1164             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1165             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1166             felec            = _mm256_mul_ps(qq20,felec);
1167             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1168             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1169             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1170             velec            = _mm256_mul_ps(qq20,velec);
1171
1172             /* Update potential sum for this i atom from the interaction with this j atom. */
1173             velec            = _mm256_andnot_ps(dummy_mask,velec);
1174             velecsum         = _mm256_add_ps(velecsum,velec);
1175
1176             fscal            = felec;
1177
1178             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1179
1180             /* Calculate temporary vectorial force */
1181             tx               = _mm256_mul_ps(fscal,dx20);
1182             ty               = _mm256_mul_ps(fscal,dy20);
1183             tz               = _mm256_mul_ps(fscal,dz20);
1184
1185             /* Update vectorial force */
1186             fix2             = _mm256_add_ps(fix2,tx);
1187             fiy2             = _mm256_add_ps(fiy2,ty);
1188             fiz2             = _mm256_add_ps(fiz2,tz);
1189
1190             fjx0             = _mm256_add_ps(fjx0,tx);
1191             fjy0             = _mm256_add_ps(fjy0,ty);
1192             fjz0             = _mm256_add_ps(fjz0,tz);
1193
1194             /**************************
1195              * CALCULATE INTERACTIONS *
1196              **************************/
1197
1198             r21              = _mm256_mul_ps(rsq21,rinv21);
1199             r21              = _mm256_andnot_ps(dummy_mask,r21);
1200
1201             /* EWALD ELECTROSTATICS */
1202
1203             /* Analytical PME correction */
1204             zeta2            = _mm256_mul_ps(beta2,rsq21);
1205             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1206             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1207             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1208             felec            = _mm256_mul_ps(qq21,felec);
1209             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1210             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1211             velec            = _mm256_sub_ps(rinv21,pmecorrV);
1212             velec            = _mm256_mul_ps(qq21,velec);
1213
1214             /* Update potential sum for this i atom from the interaction with this j atom. */
1215             velec            = _mm256_andnot_ps(dummy_mask,velec);
1216             velecsum         = _mm256_add_ps(velecsum,velec);
1217
1218             fscal            = felec;
1219
1220             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1221
1222             /* Calculate temporary vectorial force */
1223             tx               = _mm256_mul_ps(fscal,dx21);
1224             ty               = _mm256_mul_ps(fscal,dy21);
1225             tz               = _mm256_mul_ps(fscal,dz21);
1226
1227             /* Update vectorial force */
1228             fix2             = _mm256_add_ps(fix2,tx);
1229             fiy2             = _mm256_add_ps(fiy2,ty);
1230             fiz2             = _mm256_add_ps(fiz2,tz);
1231
1232             fjx1             = _mm256_add_ps(fjx1,tx);
1233             fjy1             = _mm256_add_ps(fjy1,ty);
1234             fjz1             = _mm256_add_ps(fjz1,tz);
1235
1236             /**************************
1237              * CALCULATE INTERACTIONS *
1238              **************************/
1239
1240             r22              = _mm256_mul_ps(rsq22,rinv22);
1241             r22              = _mm256_andnot_ps(dummy_mask,r22);
1242
1243             /* EWALD ELECTROSTATICS */
1244
1245             /* Analytical PME correction */
1246             zeta2            = _mm256_mul_ps(beta2,rsq22);
1247             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1248             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1249             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1250             felec            = _mm256_mul_ps(qq22,felec);
1251             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1252             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1253             velec            = _mm256_sub_ps(rinv22,pmecorrV);
1254             velec            = _mm256_mul_ps(qq22,velec);
1255
1256             /* Update potential sum for this i atom from the interaction with this j atom. */
1257             velec            = _mm256_andnot_ps(dummy_mask,velec);
1258             velecsum         = _mm256_add_ps(velecsum,velec);
1259
1260             fscal            = felec;
1261
1262             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1263
1264             /* Calculate temporary vectorial force */
1265             tx               = _mm256_mul_ps(fscal,dx22);
1266             ty               = _mm256_mul_ps(fscal,dy22);
1267             tz               = _mm256_mul_ps(fscal,dz22);
1268
1269             /* Update vectorial force */
1270             fix2             = _mm256_add_ps(fix2,tx);
1271             fiy2             = _mm256_add_ps(fiy2,ty);
1272             fiz2             = _mm256_add_ps(fiz2,tz);
1273
1274             fjx2             = _mm256_add_ps(fjx2,tx);
1275             fjy2             = _mm256_add_ps(fjy2,ty);
1276             fjz2             = _mm256_add_ps(fjz2,tz);
1277
1278             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1279             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1280             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1281             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1282             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1283             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1284             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1285             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1286
1287             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1288                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1289
1290             /* Inner loop uses 799 flops */
1291         }
1292
1293         /* End of innermost loop */
1294
1295         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1296                                                  f+i_coord_offset,fshift+i_shift_offset);
1297
1298         ggid                        = gid[iidx];
1299         /* Update potential energies */
1300         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1301         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1302
1303         /* Increment number of inner iterations */
1304         inneriter                  += j_index_end - j_index_start;
1305
1306         /* Outer loop uses 20 flops */
1307     }
1308
1309     /* Increment number of outer iterations */
1310     outeriter        += nri;
1311
1312     /* Update outer/inner flops */
1313
1314     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*799);
1315 }
1316 /*
1317  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3W3_F_avx_256_single
1318  * Electrostatics interaction: Ewald
1319  * VdW interaction:            CubicSplineTable
1320  * Geometry:                   Water3-Water3
1321  * Calculate force/pot:        Force
1322  */
1323 void
1324 nb_kernel_ElecEw_VdwCSTab_GeomW3W3_F_avx_256_single
1325                     (t_nblist                    * gmx_restrict       nlist,
1326                      rvec                        * gmx_restrict          xx,
1327                      rvec                        * gmx_restrict          ff,
1328                      t_forcerec                  * gmx_restrict          fr,
1329                      t_mdatoms                   * gmx_restrict     mdatoms,
1330                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1331                      t_nrnb                      * gmx_restrict        nrnb)
1332 {
1333     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1334      * just 0 for non-waters.
1335      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1336      * jnr indices corresponding to data put in the four positions in the SIMD register.
1337      */
1338     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1339     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1340     int              jnrA,jnrB,jnrC,jnrD;
1341     int              jnrE,jnrF,jnrG,jnrH;
1342     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1343     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1344     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1345     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1346     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1347     real             rcutoff_scalar;
1348     real             *shiftvec,*fshift,*x,*f;
1349     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1350     real             scratch[4*DIM];
1351     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1352     real *           vdwioffsetptr0;
1353     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1354     real *           vdwioffsetptr1;
1355     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1356     real *           vdwioffsetptr2;
1357     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1358     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1359     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1360     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1361     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1362     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1363     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1364     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1365     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1366     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1367     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1368     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1369     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1370     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1371     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1372     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1373     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1374     real             *charge;
1375     int              nvdwtype;
1376     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1377     int              *vdwtype;
1378     real             *vdwparam;
1379     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1380     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1381     __m256i          vfitab;
1382     __m128i          vfitab_lo,vfitab_hi;
1383     __m128i          ifour       = _mm_set1_epi32(4);
1384     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1385     real             *vftab;
1386     __m256i          ewitab;
1387     __m128i          ewitab_lo,ewitab_hi;
1388     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1389     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1390     real             *ewtab;
1391     __m256           dummy_mask,cutoff_mask;
1392     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1393     __m256           one     = _mm256_set1_ps(1.0);
1394     __m256           two     = _mm256_set1_ps(2.0);
1395     x                = xx[0];
1396     f                = ff[0];
1397
1398     nri              = nlist->nri;
1399     iinr             = nlist->iinr;
1400     jindex           = nlist->jindex;
1401     jjnr             = nlist->jjnr;
1402     shiftidx         = nlist->shift;
1403     gid              = nlist->gid;
1404     shiftvec         = fr->shift_vec[0];
1405     fshift           = fr->fshift[0];
1406     facel            = _mm256_set1_ps(fr->epsfac);
1407     charge           = mdatoms->chargeA;
1408     nvdwtype         = fr->ntype;
1409     vdwparam         = fr->nbfp;
1410     vdwtype          = mdatoms->typeA;
1411
1412     vftab            = kernel_data->table_vdw->data;
1413     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
1414
1415     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1416     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1417     beta2            = _mm256_mul_ps(beta,beta);
1418     beta3            = _mm256_mul_ps(beta,beta2);
1419
1420     ewtab            = fr->ic->tabq_coul_F;
1421     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1422     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1423
1424     /* Setup water-specific parameters */
1425     inr              = nlist->iinr[0];
1426     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1427     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1428     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1429     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1430
1431     jq0              = _mm256_set1_ps(charge[inr+0]);
1432     jq1              = _mm256_set1_ps(charge[inr+1]);
1433     jq2              = _mm256_set1_ps(charge[inr+2]);
1434     vdwjidx0A        = 2*vdwtype[inr+0];
1435     qq00             = _mm256_mul_ps(iq0,jq0);
1436     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1437     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1438     qq01             = _mm256_mul_ps(iq0,jq1);
1439     qq02             = _mm256_mul_ps(iq0,jq2);
1440     qq10             = _mm256_mul_ps(iq1,jq0);
1441     qq11             = _mm256_mul_ps(iq1,jq1);
1442     qq12             = _mm256_mul_ps(iq1,jq2);
1443     qq20             = _mm256_mul_ps(iq2,jq0);
1444     qq21             = _mm256_mul_ps(iq2,jq1);
1445     qq22             = _mm256_mul_ps(iq2,jq2);
1446
1447     /* Avoid stupid compiler warnings */
1448     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1449     j_coord_offsetA = 0;
1450     j_coord_offsetB = 0;
1451     j_coord_offsetC = 0;
1452     j_coord_offsetD = 0;
1453     j_coord_offsetE = 0;
1454     j_coord_offsetF = 0;
1455     j_coord_offsetG = 0;
1456     j_coord_offsetH = 0;
1457
1458     outeriter        = 0;
1459     inneriter        = 0;
1460
1461     for(iidx=0;iidx<4*DIM;iidx++)
1462     {
1463         scratch[iidx] = 0.0;
1464     }
1465
1466     /* Start outer loop over neighborlists */
1467     for(iidx=0; iidx<nri; iidx++)
1468     {
1469         /* Load shift vector for this list */
1470         i_shift_offset   = DIM*shiftidx[iidx];
1471
1472         /* Load limits for loop over neighbors */
1473         j_index_start    = jindex[iidx];
1474         j_index_end      = jindex[iidx+1];
1475
1476         /* Get outer coordinate index */
1477         inr              = iinr[iidx];
1478         i_coord_offset   = DIM*inr;
1479
1480         /* Load i particle coords and add shift vector */
1481         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1482                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1483
1484         fix0             = _mm256_setzero_ps();
1485         fiy0             = _mm256_setzero_ps();
1486         fiz0             = _mm256_setzero_ps();
1487         fix1             = _mm256_setzero_ps();
1488         fiy1             = _mm256_setzero_ps();
1489         fiz1             = _mm256_setzero_ps();
1490         fix2             = _mm256_setzero_ps();
1491         fiy2             = _mm256_setzero_ps();
1492         fiz2             = _mm256_setzero_ps();
1493
1494         /* Start inner kernel loop */
1495         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1496         {
1497
1498             /* Get j neighbor index, and coordinate index */
1499             jnrA             = jjnr[jidx];
1500             jnrB             = jjnr[jidx+1];
1501             jnrC             = jjnr[jidx+2];
1502             jnrD             = jjnr[jidx+3];
1503             jnrE             = jjnr[jidx+4];
1504             jnrF             = jjnr[jidx+5];
1505             jnrG             = jjnr[jidx+6];
1506             jnrH             = jjnr[jidx+7];
1507             j_coord_offsetA  = DIM*jnrA;
1508             j_coord_offsetB  = DIM*jnrB;
1509             j_coord_offsetC  = DIM*jnrC;
1510             j_coord_offsetD  = DIM*jnrD;
1511             j_coord_offsetE  = DIM*jnrE;
1512             j_coord_offsetF  = DIM*jnrF;
1513             j_coord_offsetG  = DIM*jnrG;
1514             j_coord_offsetH  = DIM*jnrH;
1515
1516             /* load j atom coordinates */
1517             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1518                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1519                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1520                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1521                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1522
1523             /* Calculate displacement vector */
1524             dx00             = _mm256_sub_ps(ix0,jx0);
1525             dy00             = _mm256_sub_ps(iy0,jy0);
1526             dz00             = _mm256_sub_ps(iz0,jz0);
1527             dx01             = _mm256_sub_ps(ix0,jx1);
1528             dy01             = _mm256_sub_ps(iy0,jy1);
1529             dz01             = _mm256_sub_ps(iz0,jz1);
1530             dx02             = _mm256_sub_ps(ix0,jx2);
1531             dy02             = _mm256_sub_ps(iy0,jy2);
1532             dz02             = _mm256_sub_ps(iz0,jz2);
1533             dx10             = _mm256_sub_ps(ix1,jx0);
1534             dy10             = _mm256_sub_ps(iy1,jy0);
1535             dz10             = _mm256_sub_ps(iz1,jz0);
1536             dx11             = _mm256_sub_ps(ix1,jx1);
1537             dy11             = _mm256_sub_ps(iy1,jy1);
1538             dz11             = _mm256_sub_ps(iz1,jz1);
1539             dx12             = _mm256_sub_ps(ix1,jx2);
1540             dy12             = _mm256_sub_ps(iy1,jy2);
1541             dz12             = _mm256_sub_ps(iz1,jz2);
1542             dx20             = _mm256_sub_ps(ix2,jx0);
1543             dy20             = _mm256_sub_ps(iy2,jy0);
1544             dz20             = _mm256_sub_ps(iz2,jz0);
1545             dx21             = _mm256_sub_ps(ix2,jx1);
1546             dy21             = _mm256_sub_ps(iy2,jy1);
1547             dz21             = _mm256_sub_ps(iz2,jz1);
1548             dx22             = _mm256_sub_ps(ix2,jx2);
1549             dy22             = _mm256_sub_ps(iy2,jy2);
1550             dz22             = _mm256_sub_ps(iz2,jz2);
1551
1552             /* Calculate squared distance and things based on it */
1553             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1554             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1555             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1556             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1557             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1558             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1559             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1560             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1561             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1562
1563             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1564             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1565             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1566             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1567             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1568             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1569             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1570             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1571             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1572
1573             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1574             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1575             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1576             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1577             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1578             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1579             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1580             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1581             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1582
1583             fjx0             = _mm256_setzero_ps();
1584             fjy0             = _mm256_setzero_ps();
1585             fjz0             = _mm256_setzero_ps();
1586             fjx1             = _mm256_setzero_ps();
1587             fjy1             = _mm256_setzero_ps();
1588             fjz1             = _mm256_setzero_ps();
1589             fjx2             = _mm256_setzero_ps();
1590             fjy2             = _mm256_setzero_ps();
1591             fjz2             = _mm256_setzero_ps();
1592
1593             /**************************
1594              * CALCULATE INTERACTIONS *
1595              **************************/
1596
1597             r00              = _mm256_mul_ps(rsq00,rinv00);
1598
1599             /* Calculate table index by multiplying r with table scale and truncate to integer */
1600             rt               = _mm256_mul_ps(r00,vftabscale);
1601             vfitab           = _mm256_cvttps_epi32(rt);
1602             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1603             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1604             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1605             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1606             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1607             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1608
1609             /* EWALD ELECTROSTATICS */
1610
1611             /* Analytical PME correction */
1612             zeta2            = _mm256_mul_ps(beta2,rsq00);
1613             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1614             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1615             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1616             felec            = _mm256_mul_ps(qq00,felec);
1617
1618             /* CUBIC SPLINE TABLE DISPERSION */
1619             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1620                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1621             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1622                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1623             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1624                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1625             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1626                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1627             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1628             Heps             = _mm256_mul_ps(vfeps,H);
1629             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1630             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1631             fvdw6            = _mm256_mul_ps(c6_00,FF);
1632
1633             /* CUBIC SPLINE TABLE REPULSION */
1634             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1635             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1636             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1637                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1638             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1639                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1640             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1641                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1642             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1643                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1644             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1645             Heps             = _mm256_mul_ps(vfeps,H);
1646             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1647             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1648             fvdw12           = _mm256_mul_ps(c12_00,FF);
1649             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1650
1651             fscal            = _mm256_add_ps(felec,fvdw);
1652
1653             /* Calculate temporary vectorial force */
1654             tx               = _mm256_mul_ps(fscal,dx00);
1655             ty               = _mm256_mul_ps(fscal,dy00);
1656             tz               = _mm256_mul_ps(fscal,dz00);
1657
1658             /* Update vectorial force */
1659             fix0             = _mm256_add_ps(fix0,tx);
1660             fiy0             = _mm256_add_ps(fiy0,ty);
1661             fiz0             = _mm256_add_ps(fiz0,tz);
1662
1663             fjx0             = _mm256_add_ps(fjx0,tx);
1664             fjy0             = _mm256_add_ps(fjy0,ty);
1665             fjz0             = _mm256_add_ps(fjz0,tz);
1666
1667             /**************************
1668              * CALCULATE INTERACTIONS *
1669              **************************/
1670
1671             r01              = _mm256_mul_ps(rsq01,rinv01);
1672
1673             /* EWALD ELECTROSTATICS */
1674
1675             /* Analytical PME correction */
1676             zeta2            = _mm256_mul_ps(beta2,rsq01);
1677             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
1678             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1679             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1680             felec            = _mm256_mul_ps(qq01,felec);
1681
1682             fscal            = felec;
1683
1684             /* Calculate temporary vectorial force */
1685             tx               = _mm256_mul_ps(fscal,dx01);
1686             ty               = _mm256_mul_ps(fscal,dy01);
1687             tz               = _mm256_mul_ps(fscal,dz01);
1688
1689             /* Update vectorial force */
1690             fix0             = _mm256_add_ps(fix0,tx);
1691             fiy0             = _mm256_add_ps(fiy0,ty);
1692             fiz0             = _mm256_add_ps(fiz0,tz);
1693
1694             fjx1             = _mm256_add_ps(fjx1,tx);
1695             fjy1             = _mm256_add_ps(fjy1,ty);
1696             fjz1             = _mm256_add_ps(fjz1,tz);
1697
1698             /**************************
1699              * CALCULATE INTERACTIONS *
1700              **************************/
1701
1702             r02              = _mm256_mul_ps(rsq02,rinv02);
1703
1704             /* EWALD ELECTROSTATICS */
1705
1706             /* Analytical PME correction */
1707             zeta2            = _mm256_mul_ps(beta2,rsq02);
1708             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
1709             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1710             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1711             felec            = _mm256_mul_ps(qq02,felec);
1712
1713             fscal            = felec;
1714
1715             /* Calculate temporary vectorial force */
1716             tx               = _mm256_mul_ps(fscal,dx02);
1717             ty               = _mm256_mul_ps(fscal,dy02);
1718             tz               = _mm256_mul_ps(fscal,dz02);
1719
1720             /* Update vectorial force */
1721             fix0             = _mm256_add_ps(fix0,tx);
1722             fiy0             = _mm256_add_ps(fiy0,ty);
1723             fiz0             = _mm256_add_ps(fiz0,tz);
1724
1725             fjx2             = _mm256_add_ps(fjx2,tx);
1726             fjy2             = _mm256_add_ps(fjy2,ty);
1727             fjz2             = _mm256_add_ps(fjz2,tz);
1728
1729             /**************************
1730              * CALCULATE INTERACTIONS *
1731              **************************/
1732
1733             r10              = _mm256_mul_ps(rsq10,rinv10);
1734
1735             /* EWALD ELECTROSTATICS */
1736
1737             /* Analytical PME correction */
1738             zeta2            = _mm256_mul_ps(beta2,rsq10);
1739             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1740             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1741             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1742             felec            = _mm256_mul_ps(qq10,felec);
1743
1744             fscal            = felec;
1745
1746             /* Calculate temporary vectorial force */
1747             tx               = _mm256_mul_ps(fscal,dx10);
1748             ty               = _mm256_mul_ps(fscal,dy10);
1749             tz               = _mm256_mul_ps(fscal,dz10);
1750
1751             /* Update vectorial force */
1752             fix1             = _mm256_add_ps(fix1,tx);
1753             fiy1             = _mm256_add_ps(fiy1,ty);
1754             fiz1             = _mm256_add_ps(fiz1,tz);
1755
1756             fjx0             = _mm256_add_ps(fjx0,tx);
1757             fjy0             = _mm256_add_ps(fjy0,ty);
1758             fjz0             = _mm256_add_ps(fjz0,tz);
1759
1760             /**************************
1761              * CALCULATE INTERACTIONS *
1762              **************************/
1763
1764             r11              = _mm256_mul_ps(rsq11,rinv11);
1765
1766             /* EWALD ELECTROSTATICS */
1767
1768             /* Analytical PME correction */
1769             zeta2            = _mm256_mul_ps(beta2,rsq11);
1770             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1771             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1772             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1773             felec            = _mm256_mul_ps(qq11,felec);
1774
1775             fscal            = felec;
1776
1777             /* Calculate temporary vectorial force */
1778             tx               = _mm256_mul_ps(fscal,dx11);
1779             ty               = _mm256_mul_ps(fscal,dy11);
1780             tz               = _mm256_mul_ps(fscal,dz11);
1781
1782             /* Update vectorial force */
1783             fix1             = _mm256_add_ps(fix1,tx);
1784             fiy1             = _mm256_add_ps(fiy1,ty);
1785             fiz1             = _mm256_add_ps(fiz1,tz);
1786
1787             fjx1             = _mm256_add_ps(fjx1,tx);
1788             fjy1             = _mm256_add_ps(fjy1,ty);
1789             fjz1             = _mm256_add_ps(fjz1,tz);
1790
1791             /**************************
1792              * CALCULATE INTERACTIONS *
1793              **************************/
1794
1795             r12              = _mm256_mul_ps(rsq12,rinv12);
1796
1797             /* EWALD ELECTROSTATICS */
1798
1799             /* Analytical PME correction */
1800             zeta2            = _mm256_mul_ps(beta2,rsq12);
1801             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1802             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1803             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1804             felec            = _mm256_mul_ps(qq12,felec);
1805
1806             fscal            = felec;
1807
1808             /* Calculate temporary vectorial force */
1809             tx               = _mm256_mul_ps(fscal,dx12);
1810             ty               = _mm256_mul_ps(fscal,dy12);
1811             tz               = _mm256_mul_ps(fscal,dz12);
1812
1813             /* Update vectorial force */
1814             fix1             = _mm256_add_ps(fix1,tx);
1815             fiy1             = _mm256_add_ps(fiy1,ty);
1816             fiz1             = _mm256_add_ps(fiz1,tz);
1817
1818             fjx2             = _mm256_add_ps(fjx2,tx);
1819             fjy2             = _mm256_add_ps(fjy2,ty);
1820             fjz2             = _mm256_add_ps(fjz2,tz);
1821
1822             /**************************
1823              * CALCULATE INTERACTIONS *
1824              **************************/
1825
1826             r20              = _mm256_mul_ps(rsq20,rinv20);
1827
1828             /* EWALD ELECTROSTATICS */
1829
1830             /* Analytical PME correction */
1831             zeta2            = _mm256_mul_ps(beta2,rsq20);
1832             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1833             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1834             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1835             felec            = _mm256_mul_ps(qq20,felec);
1836
1837             fscal            = felec;
1838
1839             /* Calculate temporary vectorial force */
1840             tx               = _mm256_mul_ps(fscal,dx20);
1841             ty               = _mm256_mul_ps(fscal,dy20);
1842             tz               = _mm256_mul_ps(fscal,dz20);
1843
1844             /* Update vectorial force */
1845             fix2             = _mm256_add_ps(fix2,tx);
1846             fiy2             = _mm256_add_ps(fiy2,ty);
1847             fiz2             = _mm256_add_ps(fiz2,tz);
1848
1849             fjx0             = _mm256_add_ps(fjx0,tx);
1850             fjy0             = _mm256_add_ps(fjy0,ty);
1851             fjz0             = _mm256_add_ps(fjz0,tz);
1852
1853             /**************************
1854              * CALCULATE INTERACTIONS *
1855              **************************/
1856
1857             r21              = _mm256_mul_ps(rsq21,rinv21);
1858
1859             /* EWALD ELECTROSTATICS */
1860
1861             /* Analytical PME correction */
1862             zeta2            = _mm256_mul_ps(beta2,rsq21);
1863             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1864             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1865             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1866             felec            = _mm256_mul_ps(qq21,felec);
1867
1868             fscal            = felec;
1869
1870             /* Calculate temporary vectorial force */
1871             tx               = _mm256_mul_ps(fscal,dx21);
1872             ty               = _mm256_mul_ps(fscal,dy21);
1873             tz               = _mm256_mul_ps(fscal,dz21);
1874
1875             /* Update vectorial force */
1876             fix2             = _mm256_add_ps(fix2,tx);
1877             fiy2             = _mm256_add_ps(fiy2,ty);
1878             fiz2             = _mm256_add_ps(fiz2,tz);
1879
1880             fjx1             = _mm256_add_ps(fjx1,tx);
1881             fjy1             = _mm256_add_ps(fjy1,ty);
1882             fjz1             = _mm256_add_ps(fjz1,tz);
1883
1884             /**************************
1885              * CALCULATE INTERACTIONS *
1886              **************************/
1887
1888             r22              = _mm256_mul_ps(rsq22,rinv22);
1889
1890             /* EWALD ELECTROSTATICS */
1891
1892             /* Analytical PME correction */
1893             zeta2            = _mm256_mul_ps(beta2,rsq22);
1894             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1895             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1896             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1897             felec            = _mm256_mul_ps(qq22,felec);
1898
1899             fscal            = felec;
1900
1901             /* Calculate temporary vectorial force */
1902             tx               = _mm256_mul_ps(fscal,dx22);
1903             ty               = _mm256_mul_ps(fscal,dy22);
1904             tz               = _mm256_mul_ps(fscal,dz22);
1905
1906             /* Update vectorial force */
1907             fix2             = _mm256_add_ps(fix2,tx);
1908             fiy2             = _mm256_add_ps(fiy2,ty);
1909             fiz2             = _mm256_add_ps(fiz2,tz);
1910
1911             fjx2             = _mm256_add_ps(fjx2,tx);
1912             fjy2             = _mm256_add_ps(fjy2,ty);
1913             fjz2             = _mm256_add_ps(fjz2,tz);
1914
1915             fjptrA             = f+j_coord_offsetA;
1916             fjptrB             = f+j_coord_offsetB;
1917             fjptrC             = f+j_coord_offsetC;
1918             fjptrD             = f+j_coord_offsetD;
1919             fjptrE             = f+j_coord_offsetE;
1920             fjptrF             = f+j_coord_offsetF;
1921             fjptrG             = f+j_coord_offsetG;
1922             fjptrH             = f+j_coord_offsetH;
1923
1924             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1925                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1926
1927             /* Inner loop uses 530 flops */
1928         }
1929
1930         if(jidx<j_index_end)
1931         {
1932
1933             /* Get j neighbor index, and coordinate index */
1934             jnrlistA         = jjnr[jidx];
1935             jnrlistB         = jjnr[jidx+1];
1936             jnrlistC         = jjnr[jidx+2];
1937             jnrlistD         = jjnr[jidx+3];
1938             jnrlistE         = jjnr[jidx+4];
1939             jnrlistF         = jjnr[jidx+5];
1940             jnrlistG         = jjnr[jidx+6];
1941             jnrlistH         = jjnr[jidx+7];
1942             /* Sign of each element will be negative for non-real atoms.
1943              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1944              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1945              */
1946             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1947                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1948
1949             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1950             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1951             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1952             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1953             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1954             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1955             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1956             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1957             j_coord_offsetA  = DIM*jnrA;
1958             j_coord_offsetB  = DIM*jnrB;
1959             j_coord_offsetC  = DIM*jnrC;
1960             j_coord_offsetD  = DIM*jnrD;
1961             j_coord_offsetE  = DIM*jnrE;
1962             j_coord_offsetF  = DIM*jnrF;
1963             j_coord_offsetG  = DIM*jnrG;
1964             j_coord_offsetH  = DIM*jnrH;
1965
1966             /* load j atom coordinates */
1967             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1968                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1969                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1970                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1971                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1972
1973             /* Calculate displacement vector */
1974             dx00             = _mm256_sub_ps(ix0,jx0);
1975             dy00             = _mm256_sub_ps(iy0,jy0);
1976             dz00             = _mm256_sub_ps(iz0,jz0);
1977             dx01             = _mm256_sub_ps(ix0,jx1);
1978             dy01             = _mm256_sub_ps(iy0,jy1);
1979             dz01             = _mm256_sub_ps(iz0,jz1);
1980             dx02             = _mm256_sub_ps(ix0,jx2);
1981             dy02             = _mm256_sub_ps(iy0,jy2);
1982             dz02             = _mm256_sub_ps(iz0,jz2);
1983             dx10             = _mm256_sub_ps(ix1,jx0);
1984             dy10             = _mm256_sub_ps(iy1,jy0);
1985             dz10             = _mm256_sub_ps(iz1,jz0);
1986             dx11             = _mm256_sub_ps(ix1,jx1);
1987             dy11             = _mm256_sub_ps(iy1,jy1);
1988             dz11             = _mm256_sub_ps(iz1,jz1);
1989             dx12             = _mm256_sub_ps(ix1,jx2);
1990             dy12             = _mm256_sub_ps(iy1,jy2);
1991             dz12             = _mm256_sub_ps(iz1,jz2);
1992             dx20             = _mm256_sub_ps(ix2,jx0);
1993             dy20             = _mm256_sub_ps(iy2,jy0);
1994             dz20             = _mm256_sub_ps(iz2,jz0);
1995             dx21             = _mm256_sub_ps(ix2,jx1);
1996             dy21             = _mm256_sub_ps(iy2,jy1);
1997             dz21             = _mm256_sub_ps(iz2,jz1);
1998             dx22             = _mm256_sub_ps(ix2,jx2);
1999             dy22             = _mm256_sub_ps(iy2,jy2);
2000             dz22             = _mm256_sub_ps(iz2,jz2);
2001
2002             /* Calculate squared distance and things based on it */
2003             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2004             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
2005             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
2006             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
2007             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2008             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2009             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
2010             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2011             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2012
2013             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
2014             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
2015             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
2016             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
2017             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2018             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2019             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
2020             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2021             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2022
2023             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
2024             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
2025             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
2026             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
2027             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2028             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2029             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
2030             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2031             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2032
2033             fjx0             = _mm256_setzero_ps();
2034             fjy0             = _mm256_setzero_ps();
2035             fjz0             = _mm256_setzero_ps();
2036             fjx1             = _mm256_setzero_ps();
2037             fjy1             = _mm256_setzero_ps();
2038             fjz1             = _mm256_setzero_ps();
2039             fjx2             = _mm256_setzero_ps();
2040             fjy2             = _mm256_setzero_ps();
2041             fjz2             = _mm256_setzero_ps();
2042
2043             /**************************
2044              * CALCULATE INTERACTIONS *
2045              **************************/
2046
2047             r00              = _mm256_mul_ps(rsq00,rinv00);
2048             r00              = _mm256_andnot_ps(dummy_mask,r00);
2049
2050             /* Calculate table index by multiplying r with table scale and truncate to integer */
2051             rt               = _mm256_mul_ps(r00,vftabscale);
2052             vfitab           = _mm256_cvttps_epi32(rt);
2053             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2054             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2055             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2056             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2057             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
2058             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
2059
2060             /* EWALD ELECTROSTATICS */
2061
2062             /* Analytical PME correction */
2063             zeta2            = _mm256_mul_ps(beta2,rsq00);
2064             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
2065             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2066             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2067             felec            = _mm256_mul_ps(qq00,felec);
2068
2069             /* CUBIC SPLINE TABLE DISPERSION */
2070             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2071                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2072             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2073                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2074             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2075                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2076             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2077                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2078             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2079             Heps             = _mm256_mul_ps(vfeps,H);
2080             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2081             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2082             fvdw6            = _mm256_mul_ps(c6_00,FF);
2083
2084             /* CUBIC SPLINE TABLE REPULSION */
2085             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
2086             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
2087             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2088                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2089             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2090                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2091             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2092                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2093             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2094                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2095             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2096             Heps             = _mm256_mul_ps(vfeps,H);
2097             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2098             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2099             fvdw12           = _mm256_mul_ps(c12_00,FF);
2100             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
2101
2102             fscal            = _mm256_add_ps(felec,fvdw);
2103
2104             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2105
2106             /* Calculate temporary vectorial force */
2107             tx               = _mm256_mul_ps(fscal,dx00);
2108             ty               = _mm256_mul_ps(fscal,dy00);
2109             tz               = _mm256_mul_ps(fscal,dz00);
2110
2111             /* Update vectorial force */
2112             fix0             = _mm256_add_ps(fix0,tx);
2113             fiy0             = _mm256_add_ps(fiy0,ty);
2114             fiz0             = _mm256_add_ps(fiz0,tz);
2115
2116             fjx0             = _mm256_add_ps(fjx0,tx);
2117             fjy0             = _mm256_add_ps(fjy0,ty);
2118             fjz0             = _mm256_add_ps(fjz0,tz);
2119
2120             /**************************
2121              * CALCULATE INTERACTIONS *
2122              **************************/
2123
2124             r01              = _mm256_mul_ps(rsq01,rinv01);
2125             r01              = _mm256_andnot_ps(dummy_mask,r01);
2126
2127             /* EWALD ELECTROSTATICS */
2128
2129             /* Analytical PME correction */
2130             zeta2            = _mm256_mul_ps(beta2,rsq01);
2131             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
2132             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2133             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2134             felec            = _mm256_mul_ps(qq01,felec);
2135
2136             fscal            = felec;
2137
2138             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2139
2140             /* Calculate temporary vectorial force */
2141             tx               = _mm256_mul_ps(fscal,dx01);
2142             ty               = _mm256_mul_ps(fscal,dy01);
2143             tz               = _mm256_mul_ps(fscal,dz01);
2144
2145             /* Update vectorial force */
2146             fix0             = _mm256_add_ps(fix0,tx);
2147             fiy0             = _mm256_add_ps(fiy0,ty);
2148             fiz0             = _mm256_add_ps(fiz0,tz);
2149
2150             fjx1             = _mm256_add_ps(fjx1,tx);
2151             fjy1             = _mm256_add_ps(fjy1,ty);
2152             fjz1             = _mm256_add_ps(fjz1,tz);
2153
2154             /**************************
2155              * CALCULATE INTERACTIONS *
2156              **************************/
2157
2158             r02              = _mm256_mul_ps(rsq02,rinv02);
2159             r02              = _mm256_andnot_ps(dummy_mask,r02);
2160
2161             /* EWALD ELECTROSTATICS */
2162
2163             /* Analytical PME correction */
2164             zeta2            = _mm256_mul_ps(beta2,rsq02);
2165             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
2166             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2167             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2168             felec            = _mm256_mul_ps(qq02,felec);
2169
2170             fscal            = felec;
2171
2172             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2173
2174             /* Calculate temporary vectorial force */
2175             tx               = _mm256_mul_ps(fscal,dx02);
2176             ty               = _mm256_mul_ps(fscal,dy02);
2177             tz               = _mm256_mul_ps(fscal,dz02);
2178
2179             /* Update vectorial force */
2180             fix0             = _mm256_add_ps(fix0,tx);
2181             fiy0             = _mm256_add_ps(fiy0,ty);
2182             fiz0             = _mm256_add_ps(fiz0,tz);
2183
2184             fjx2             = _mm256_add_ps(fjx2,tx);
2185             fjy2             = _mm256_add_ps(fjy2,ty);
2186             fjz2             = _mm256_add_ps(fjz2,tz);
2187
2188             /**************************
2189              * CALCULATE INTERACTIONS *
2190              **************************/
2191
2192             r10              = _mm256_mul_ps(rsq10,rinv10);
2193             r10              = _mm256_andnot_ps(dummy_mask,r10);
2194
2195             /* EWALD ELECTROSTATICS */
2196
2197             /* Analytical PME correction */
2198             zeta2            = _mm256_mul_ps(beta2,rsq10);
2199             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
2200             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2201             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2202             felec            = _mm256_mul_ps(qq10,felec);
2203
2204             fscal            = felec;
2205
2206             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2207
2208             /* Calculate temporary vectorial force */
2209             tx               = _mm256_mul_ps(fscal,dx10);
2210             ty               = _mm256_mul_ps(fscal,dy10);
2211             tz               = _mm256_mul_ps(fscal,dz10);
2212
2213             /* Update vectorial force */
2214             fix1             = _mm256_add_ps(fix1,tx);
2215             fiy1             = _mm256_add_ps(fiy1,ty);
2216             fiz1             = _mm256_add_ps(fiz1,tz);
2217
2218             fjx0             = _mm256_add_ps(fjx0,tx);
2219             fjy0             = _mm256_add_ps(fjy0,ty);
2220             fjz0             = _mm256_add_ps(fjz0,tz);
2221
2222             /**************************
2223              * CALCULATE INTERACTIONS *
2224              **************************/
2225
2226             r11              = _mm256_mul_ps(rsq11,rinv11);
2227             r11              = _mm256_andnot_ps(dummy_mask,r11);
2228
2229             /* EWALD ELECTROSTATICS */
2230
2231             /* Analytical PME correction */
2232             zeta2            = _mm256_mul_ps(beta2,rsq11);
2233             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
2234             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2235             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2236             felec            = _mm256_mul_ps(qq11,felec);
2237
2238             fscal            = felec;
2239
2240             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2241
2242             /* Calculate temporary vectorial force */
2243             tx               = _mm256_mul_ps(fscal,dx11);
2244             ty               = _mm256_mul_ps(fscal,dy11);
2245             tz               = _mm256_mul_ps(fscal,dz11);
2246
2247             /* Update vectorial force */
2248             fix1             = _mm256_add_ps(fix1,tx);
2249             fiy1             = _mm256_add_ps(fiy1,ty);
2250             fiz1             = _mm256_add_ps(fiz1,tz);
2251
2252             fjx1             = _mm256_add_ps(fjx1,tx);
2253             fjy1             = _mm256_add_ps(fjy1,ty);
2254             fjz1             = _mm256_add_ps(fjz1,tz);
2255
2256             /**************************
2257              * CALCULATE INTERACTIONS *
2258              **************************/
2259
2260             r12              = _mm256_mul_ps(rsq12,rinv12);
2261             r12              = _mm256_andnot_ps(dummy_mask,r12);
2262
2263             /* EWALD ELECTROSTATICS */
2264
2265             /* Analytical PME correction */
2266             zeta2            = _mm256_mul_ps(beta2,rsq12);
2267             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2268             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2269             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2270             felec            = _mm256_mul_ps(qq12,felec);
2271
2272             fscal            = felec;
2273
2274             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2275
2276             /* Calculate temporary vectorial force */
2277             tx               = _mm256_mul_ps(fscal,dx12);
2278             ty               = _mm256_mul_ps(fscal,dy12);
2279             tz               = _mm256_mul_ps(fscal,dz12);
2280
2281             /* Update vectorial force */
2282             fix1             = _mm256_add_ps(fix1,tx);
2283             fiy1             = _mm256_add_ps(fiy1,ty);
2284             fiz1             = _mm256_add_ps(fiz1,tz);
2285
2286             fjx2             = _mm256_add_ps(fjx2,tx);
2287             fjy2             = _mm256_add_ps(fjy2,ty);
2288             fjz2             = _mm256_add_ps(fjz2,tz);
2289
2290             /**************************
2291              * CALCULATE INTERACTIONS *
2292              **************************/
2293
2294             r20              = _mm256_mul_ps(rsq20,rinv20);
2295             r20              = _mm256_andnot_ps(dummy_mask,r20);
2296
2297             /* EWALD ELECTROSTATICS */
2298
2299             /* Analytical PME correction */
2300             zeta2            = _mm256_mul_ps(beta2,rsq20);
2301             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
2302             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2303             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2304             felec            = _mm256_mul_ps(qq20,felec);
2305
2306             fscal            = felec;
2307
2308             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2309
2310             /* Calculate temporary vectorial force */
2311             tx               = _mm256_mul_ps(fscal,dx20);
2312             ty               = _mm256_mul_ps(fscal,dy20);
2313             tz               = _mm256_mul_ps(fscal,dz20);
2314
2315             /* Update vectorial force */
2316             fix2             = _mm256_add_ps(fix2,tx);
2317             fiy2             = _mm256_add_ps(fiy2,ty);
2318             fiz2             = _mm256_add_ps(fiz2,tz);
2319
2320             fjx0             = _mm256_add_ps(fjx0,tx);
2321             fjy0             = _mm256_add_ps(fjy0,ty);
2322             fjz0             = _mm256_add_ps(fjz0,tz);
2323
2324             /**************************
2325              * CALCULATE INTERACTIONS *
2326              **************************/
2327
2328             r21              = _mm256_mul_ps(rsq21,rinv21);
2329             r21              = _mm256_andnot_ps(dummy_mask,r21);
2330
2331             /* EWALD ELECTROSTATICS */
2332
2333             /* Analytical PME correction */
2334             zeta2            = _mm256_mul_ps(beta2,rsq21);
2335             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2336             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2337             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2338             felec            = _mm256_mul_ps(qq21,felec);
2339
2340             fscal            = felec;
2341
2342             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2343
2344             /* Calculate temporary vectorial force */
2345             tx               = _mm256_mul_ps(fscal,dx21);
2346             ty               = _mm256_mul_ps(fscal,dy21);
2347             tz               = _mm256_mul_ps(fscal,dz21);
2348
2349             /* Update vectorial force */
2350             fix2             = _mm256_add_ps(fix2,tx);
2351             fiy2             = _mm256_add_ps(fiy2,ty);
2352             fiz2             = _mm256_add_ps(fiz2,tz);
2353
2354             fjx1             = _mm256_add_ps(fjx1,tx);
2355             fjy1             = _mm256_add_ps(fjy1,ty);
2356             fjz1             = _mm256_add_ps(fjz1,tz);
2357
2358             /**************************
2359              * CALCULATE INTERACTIONS *
2360              **************************/
2361
2362             r22              = _mm256_mul_ps(rsq22,rinv22);
2363             r22              = _mm256_andnot_ps(dummy_mask,r22);
2364
2365             /* EWALD ELECTROSTATICS */
2366
2367             /* Analytical PME correction */
2368             zeta2            = _mm256_mul_ps(beta2,rsq22);
2369             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2370             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2371             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2372             felec            = _mm256_mul_ps(qq22,felec);
2373
2374             fscal            = felec;
2375
2376             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2377
2378             /* Calculate temporary vectorial force */
2379             tx               = _mm256_mul_ps(fscal,dx22);
2380             ty               = _mm256_mul_ps(fscal,dy22);
2381             tz               = _mm256_mul_ps(fscal,dz22);
2382
2383             /* Update vectorial force */
2384             fix2             = _mm256_add_ps(fix2,tx);
2385             fiy2             = _mm256_add_ps(fiy2,ty);
2386             fiz2             = _mm256_add_ps(fiz2,tz);
2387
2388             fjx2             = _mm256_add_ps(fjx2,tx);
2389             fjy2             = _mm256_add_ps(fjy2,ty);
2390             fjz2             = _mm256_add_ps(fjz2,tz);
2391
2392             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2393             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2394             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2395             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2396             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2397             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2398             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2399             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2400
2401             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2402                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2403
2404             /* Inner loop uses 539 flops */
2405         }
2406
2407         /* End of innermost loop */
2408
2409         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2410                                                  f+i_coord_offset,fshift+i_shift_offset);
2411
2412         /* Increment number of inner iterations */
2413         inneriter                  += j_index_end - j_index_start;
2414
2415         /* Outer loop uses 18 flops */
2416     }
2417
2418     /* Increment number of outer iterations */
2419     outeriter        += nri;
2420
2421     /* Update outer/inner flops */
2422
2423     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*539);
2424 }