src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwCSTab_GeomW3W3_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3W3_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Water3
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW3W3_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  81     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  82     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  83     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  84     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  86     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  87     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  88     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  89     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  90     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  92     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  93     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 100     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 101     __m256i          vfitab;
 102     __m128i          vfitab_lo,vfitab_hi;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 106     __m256i          ewitab;
 107     __m128i          ewitab_lo,ewitab_hi;
 108     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 109     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 110     real             *ewtab;
 111     __m256           dummy_mask,cutoff_mask;
 112     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 113     __m256           one     = _mm256_set1_ps(1.0);
 114     __m256           two     = _mm256_set1_ps(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm256_set1_ps(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 134
 135     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 136     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 137     beta2            = _mm256_mul_ps(beta,beta);
 138     beta3            = _mm256_mul_ps(beta,beta2);
 139
 140     ewtab            = fr->ic->tabq_coul_FDV0;
 141     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 142     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 143
 144     /* Setup water-specific parameters */
 145     inr              = nlist->iinr[0];
 146     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 147     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 148     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 149     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 150
 151     jq0              = _mm256_set1_ps(charge[inr+0]);
 152     jq1              = _mm256_set1_ps(charge[inr+1]);
 153     jq2              = _mm256_set1_ps(charge[inr+2]);
 154     vdwjidx0A        = 2*vdwtype[inr+0];
 155     qq00             = _mm256_mul_ps(iq0,jq0);
 156     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 157     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 158     qq01             = _mm256_mul_ps(iq0,jq1);
 159     qq02             = _mm256_mul_ps(iq0,jq2);
 160     qq10             = _mm256_mul_ps(iq1,jq0);
 161     qq11             = _mm256_mul_ps(iq1,jq1);
 162     qq12             = _mm256_mul_ps(iq1,jq2);
 163     qq20             = _mm256_mul_ps(iq2,jq0);
 164     qq21             = _mm256_mul_ps(iq2,jq1);
 165     qq22             = _mm256_mul_ps(iq2,jq2);
 166
 167     /* Avoid stupid compiler warnings */
 168     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 169     j_coord_offsetA = 0;
 170     j_coord_offsetB = 0;
 171     j_coord_offsetC = 0;
 172     j_coord_offsetD = 0;
 173     j_coord_offsetE = 0;
 174     j_coord_offsetF = 0;
 175     j_coord_offsetG = 0;
 176     j_coord_offsetH = 0;
 177
 178     outeriter        = 0;
 179     inneriter        = 0;
 180
 181     for(iidx=0;iidx<4*DIM;iidx++)
 182     {
 183         scratch[iidx] = 0.0;
 184     }
 185
 186     /* Start outer loop over neighborlists */
 187     for(iidx=0; iidx<nri; iidx++)
 188     {
 189         /* Load shift vector for this list */
 190         i_shift_offset   = DIM*shiftidx[iidx];
 191
 192         /* Load limits for loop over neighbors */
 193         j_index_start    = jindex[iidx];
 194         j_index_end      = jindex[iidx+1];
 195
 196         /* Get outer coordinate index */
 197         inr              = iinr[iidx];
 198         i_coord_offset   = DIM*inr;
 199
 200         /* Load i particle coords and add shift vector */
 201         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 202                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 203
 204         fix0             = _mm256_setzero_ps();
 205         fiy0             = _mm256_setzero_ps();
 206         fiz0             = _mm256_setzero_ps();
 207         fix1             = _mm256_setzero_ps();
 208         fiy1             = _mm256_setzero_ps();
 209         fiz1             = _mm256_setzero_ps();
 210         fix2             = _mm256_setzero_ps();
 211         fiy2             = _mm256_setzero_ps();
 212         fiz2             = _mm256_setzero_ps();
 213
 214         /* Reset potential sums */
 215         velecsum         = _mm256_setzero_ps();
 216         vvdwsum          = _mm256_setzero_ps();
 217
 218         /* Start inner kernel loop */
 219         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 220         {
 221
 222             /* Get j neighbor index, and coordinate index */
 223             jnrA             = jjnr[jidx];
 224             jnrB             = jjnr[jidx+1];
 225             jnrC             = jjnr[jidx+2];
 226             jnrD             = jjnr[jidx+3];
 227             jnrE             = jjnr[jidx+4];
 228             jnrF             = jjnr[jidx+5];
 229             jnrG             = jjnr[jidx+6];
 230             jnrH             = jjnr[jidx+7];
 231             j_coord_offsetA  = DIM*jnrA;
 232             j_coord_offsetB  = DIM*jnrB;
 233             j_coord_offsetC  = DIM*jnrC;
 234             j_coord_offsetD  = DIM*jnrD;
 235             j_coord_offsetE  = DIM*jnrE;
 236             j_coord_offsetF  = DIM*jnrF;
 237             j_coord_offsetG  = DIM*jnrG;
 238             j_coord_offsetH  = DIM*jnrH;
 239
 240             /* load j atom coordinates */
 241             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 242                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 243                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 244                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 245                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 246
 247             /* Calculate displacement vector */
 248             dx00             = _mm256_sub_ps(ix0,jx0);
 249             dy00             = _mm256_sub_ps(iy0,jy0);
 250             dz00             = _mm256_sub_ps(iz0,jz0);
 251             dx01             = _mm256_sub_ps(ix0,jx1);
 252             dy01             = _mm256_sub_ps(iy0,jy1);
 253             dz01             = _mm256_sub_ps(iz0,jz1);
 254             dx02             = _mm256_sub_ps(ix0,jx2);
 255             dy02             = _mm256_sub_ps(iy0,jy2);
 256             dz02             = _mm256_sub_ps(iz0,jz2);
 257             dx10             = _mm256_sub_ps(ix1,jx0);
 258             dy10             = _mm256_sub_ps(iy1,jy0);
 259             dz10             = _mm256_sub_ps(iz1,jz0);
 260             dx11             = _mm256_sub_ps(ix1,jx1);
 261             dy11             = _mm256_sub_ps(iy1,jy1);
 262             dz11             = _mm256_sub_ps(iz1,jz1);
 263             dx12             = _mm256_sub_ps(ix1,jx2);
 264             dy12             = _mm256_sub_ps(iy1,jy2);
 265             dz12             = _mm256_sub_ps(iz1,jz2);
 266             dx20             = _mm256_sub_ps(ix2,jx0);
 267             dy20             = _mm256_sub_ps(iy2,jy0);
 268             dz20             = _mm256_sub_ps(iz2,jz0);
 269             dx21             = _mm256_sub_ps(ix2,jx1);
 270             dy21             = _mm256_sub_ps(iy2,jy1);
 271             dz21             = _mm256_sub_ps(iz2,jz1);
 272             dx22             = _mm256_sub_ps(ix2,jx2);
 273             dy22             = _mm256_sub_ps(iy2,jy2);
 274             dz22             = _mm256_sub_ps(iz2,jz2);
 275
 276             /* Calculate squared distance and things based on it */
 277             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 278             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 279             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 280             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 281             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 282             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 283             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 284             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 285             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 286
 287             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 288             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 289             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 290             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 291             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 292             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 293             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 294             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 295             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 296
 297             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 298             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 299             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 300             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 301             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 302             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 303             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 304             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 305             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 306
 307             fjx0             = _mm256_setzero_ps();
 308             fjy0             = _mm256_setzero_ps();
 309             fjz0             = _mm256_setzero_ps();
 310             fjx1             = _mm256_setzero_ps();
 311             fjy1             = _mm256_setzero_ps();
 312             fjz1             = _mm256_setzero_ps();
 313             fjx2             = _mm256_setzero_ps();
 314             fjy2             = _mm256_setzero_ps();
 315             fjz2             = _mm256_setzero_ps();
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             r00              = _mm256_mul_ps(rsq00,rinv00);
 322
 323             /* Calculate table index by multiplying r with table scale and truncate to integer */
 324             rt               = _mm256_mul_ps(r00,vftabscale);
 325             vfitab           = _mm256_cvttps_epi32(rt);
 326             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 327             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 328             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 329             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 330             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 331             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 332
 333             /* EWALD ELECTROSTATICS */
 334
 335             /* Analytical PME correction */
 336             zeta2            = _mm256_mul_ps(beta2,rsq00);
 337             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 338             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 339             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 340             felec            = _mm256_mul_ps(qq00,felec);
 341             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 342             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 343             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 344             velec            = _mm256_mul_ps(qq00,velec);
 345
 346             /* CUBIC SPLINE TABLE DISPERSION */
 347             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 348                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 349             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 350                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 351             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 352                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 353             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 354                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 355             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 356             Heps             = _mm256_mul_ps(vfeps,H);
 357             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 358             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 359             vvdw6            = _mm256_mul_ps(c6_00,VV);
 360             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 361             fvdw6            = _mm256_mul_ps(c6_00,FF);
 362
 363             /* CUBIC SPLINE TABLE REPULSION */
 364             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 365             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 366             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 367                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 368             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 369                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 370             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 371                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 372             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 373                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 374             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 375             Heps             = _mm256_mul_ps(vfeps,H);
 376             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 377             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 378             vvdw12           = _mm256_mul_ps(c12_00,VV);
 379             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 380             fvdw12           = _mm256_mul_ps(c12_00,FF);
 381             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 382             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 383
 384             /* Update potential sum for this i atom from the interaction with this j atom. */
 385             velecsum         = _mm256_add_ps(velecsum,velec);
 386             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 387
 388             fscal            = _mm256_add_ps(felec,fvdw);
 389
 390             /* Calculate temporary vectorial force */
 391             tx               = _mm256_mul_ps(fscal,dx00);
 392             ty               = _mm256_mul_ps(fscal,dy00);
 393             tz               = _mm256_mul_ps(fscal,dz00);
 394
 395             /* Update vectorial force */
 396             fix0             = _mm256_add_ps(fix0,tx);
 397             fiy0             = _mm256_add_ps(fiy0,ty);
 398             fiz0             = _mm256_add_ps(fiz0,tz);
 399
 400             fjx0             = _mm256_add_ps(fjx0,tx);
 401             fjy0             = _mm256_add_ps(fjy0,ty);
 402             fjz0             = _mm256_add_ps(fjz0,tz);
 403
 404             /**************************
 405              * CALCULATE INTERACTIONS *
 406              **************************/
 407
 408             r01              = _mm256_mul_ps(rsq01,rinv01);
 409
 410             /* EWALD ELECTROSTATICS */
 411
 412             /* Analytical PME correction */
 413             zeta2            = _mm256_mul_ps(beta2,rsq01);
 414             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 415             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 416             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 417             felec            = _mm256_mul_ps(qq01,felec);
 418             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 419             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 420             velec            = _mm256_sub_ps(rinv01,pmecorrV);
 421             velec            = _mm256_mul_ps(qq01,velec);
 422
 423             /* Update potential sum for this i atom from the interaction with this j atom. */
 424             velecsum         = _mm256_add_ps(velecsum,velec);
 425
 426             fscal            = felec;
 427
 428             /* Calculate temporary vectorial force */
 429             tx               = _mm256_mul_ps(fscal,dx01);
 430             ty               = _mm256_mul_ps(fscal,dy01);
 431             tz               = _mm256_mul_ps(fscal,dz01);
 432
 433             /* Update vectorial force */
 434             fix0             = _mm256_add_ps(fix0,tx);
 435             fiy0             = _mm256_add_ps(fiy0,ty);
 436             fiz0             = _mm256_add_ps(fiz0,tz);
 437
 438             fjx1             = _mm256_add_ps(fjx1,tx);
 439             fjy1             = _mm256_add_ps(fjy1,ty);
 440             fjz1             = _mm256_add_ps(fjz1,tz);
 441
 442             /**************************
 443              * CALCULATE INTERACTIONS *
 444              **************************/
 445
 446             r02              = _mm256_mul_ps(rsq02,rinv02);
 447
 448             /* EWALD ELECTROSTATICS */
 449
 450             /* Analytical PME correction */
 451             zeta2            = _mm256_mul_ps(beta2,rsq02);
 452             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 453             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 454             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 455             felec            = _mm256_mul_ps(qq02,felec);
 456             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 457             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 458             velec            = _mm256_sub_ps(rinv02,pmecorrV);
 459             velec            = _mm256_mul_ps(qq02,velec);
 460
 461             /* Update potential sum for this i atom from the interaction with this j atom. */
 462             velecsum         = _mm256_add_ps(velecsum,velec);
 463
 464             fscal            = felec;
 465
 466             /* Calculate temporary vectorial force */
 467             tx               = _mm256_mul_ps(fscal,dx02);
 468             ty               = _mm256_mul_ps(fscal,dy02);
 469             tz               = _mm256_mul_ps(fscal,dz02);
 470
 471             /* Update vectorial force */
 472             fix0             = _mm256_add_ps(fix0,tx);
 473             fiy0             = _mm256_add_ps(fiy0,ty);
 474             fiz0             = _mm256_add_ps(fiz0,tz);
 475
 476             fjx2             = _mm256_add_ps(fjx2,tx);
 477             fjy2             = _mm256_add_ps(fjy2,ty);
 478             fjz2             = _mm256_add_ps(fjz2,tz);
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             r10              = _mm256_mul_ps(rsq10,rinv10);
 485
 486             /* EWALD ELECTROSTATICS */
 487
 488             /* Analytical PME correction */
 489             zeta2            = _mm256_mul_ps(beta2,rsq10);
 490             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 491             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 492             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 493             felec            = _mm256_mul_ps(qq10,felec);
 494             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 495             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 496             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 497             velec            = _mm256_mul_ps(qq10,velec);
 498
 499             /* Update potential sum for this i atom from the interaction with this j atom. */
 500             velecsum         = _mm256_add_ps(velecsum,velec);
 501
 502             fscal            = felec;
 503
 504             /* Calculate temporary vectorial force */
 505             tx               = _mm256_mul_ps(fscal,dx10);
 506             ty               = _mm256_mul_ps(fscal,dy10);
 507             tz               = _mm256_mul_ps(fscal,dz10);
 508
 509             /* Update vectorial force */
 510             fix1             = _mm256_add_ps(fix1,tx);
 511             fiy1             = _mm256_add_ps(fiy1,ty);
 512             fiz1             = _mm256_add_ps(fiz1,tz);
 513
 514             fjx0             = _mm256_add_ps(fjx0,tx);
 515             fjy0             = _mm256_add_ps(fjy0,ty);
 516             fjz0             = _mm256_add_ps(fjz0,tz);
 517
 518             /**************************
 519              * CALCULATE INTERACTIONS *
 520              **************************/
 521
 522             r11              = _mm256_mul_ps(rsq11,rinv11);
 523
 524             /* EWALD ELECTROSTATICS */
 525
 526             /* Analytical PME correction */
 527             zeta2            = _mm256_mul_ps(beta2,rsq11);
 528             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 529             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 530             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 531             felec            = _mm256_mul_ps(qq11,felec);
 532             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 533             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 534             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 535             velec            = _mm256_mul_ps(qq11,velec);
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velecsum         = _mm256_add_ps(velecsum,velec);
 539
 540             fscal            = felec;
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm256_mul_ps(fscal,dx11);
 544             ty               = _mm256_mul_ps(fscal,dy11);
 545             tz               = _mm256_mul_ps(fscal,dz11);
 546
 547             /* Update vectorial force */
 548             fix1             = _mm256_add_ps(fix1,tx);
 549             fiy1             = _mm256_add_ps(fiy1,ty);
 550             fiz1             = _mm256_add_ps(fiz1,tz);
 551
 552             fjx1             = _mm256_add_ps(fjx1,tx);
 553             fjy1             = _mm256_add_ps(fjy1,ty);
 554             fjz1             = _mm256_add_ps(fjz1,tz);
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             r12              = _mm256_mul_ps(rsq12,rinv12);
 561
 562             /* EWALD ELECTROSTATICS */
 563
 564             /* Analytical PME correction */
 565             zeta2            = _mm256_mul_ps(beta2,rsq12);
 566             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 567             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 568             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 569             felec            = _mm256_mul_ps(qq12,felec);
 570             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 571             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 572             velec            = _mm256_sub_ps(rinv12,pmecorrV);
 573             velec            = _mm256_mul_ps(qq12,velec);
 574
 575             /* Update potential sum for this i atom from the interaction with this j atom. */
 576             velecsum         = _mm256_add_ps(velecsum,velec);
 577
 578             fscal            = felec;
 579
 580             /* Calculate temporary vectorial force */
 581             tx               = _mm256_mul_ps(fscal,dx12);
 582             ty               = _mm256_mul_ps(fscal,dy12);
 583             tz               = _mm256_mul_ps(fscal,dz12);
 584
 585             /* Update vectorial force */
 586             fix1             = _mm256_add_ps(fix1,tx);
 587             fiy1             = _mm256_add_ps(fiy1,ty);
 588             fiz1             = _mm256_add_ps(fiz1,tz);
 589
 590             fjx2             = _mm256_add_ps(fjx2,tx);
 591             fjy2             = _mm256_add_ps(fjy2,ty);
 592             fjz2             = _mm256_add_ps(fjz2,tz);
 593
 594             /**************************
 595              * CALCULATE INTERACTIONS *
 596              **************************/
 597
 598             r20              = _mm256_mul_ps(rsq20,rinv20);
 599
 600             /* EWALD ELECTROSTATICS */
 601
 602             /* Analytical PME correction */
 603             zeta2            = _mm256_mul_ps(beta2,rsq20);
 604             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 605             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 606             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 607             felec            = _mm256_mul_ps(qq20,felec);
 608             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 609             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 610             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 611             velec            = _mm256_mul_ps(qq20,velec);
 612
 613             /* Update potential sum for this i atom from the interaction with this j atom. */
 614             velecsum         = _mm256_add_ps(velecsum,velec);
 615
 616             fscal            = felec;
 617
 618             /* Calculate temporary vectorial force */
 619             tx               = _mm256_mul_ps(fscal,dx20);
 620             ty               = _mm256_mul_ps(fscal,dy20);
 621             tz               = _mm256_mul_ps(fscal,dz20);
 622
 623             /* Update vectorial force */
 624             fix2             = _mm256_add_ps(fix2,tx);
 625             fiy2             = _mm256_add_ps(fiy2,ty);
 626             fiz2             = _mm256_add_ps(fiz2,tz);
 627
 628             fjx0             = _mm256_add_ps(fjx0,tx);
 629             fjy0             = _mm256_add_ps(fjy0,ty);
 630             fjz0             = _mm256_add_ps(fjz0,tz);
 631
 632             /**************************
 633              * CALCULATE INTERACTIONS *
 634              **************************/
 635
 636             r21              = _mm256_mul_ps(rsq21,rinv21);
 637
 638             /* EWALD ELECTROSTATICS */
 639
 640             /* Analytical PME correction */
 641             zeta2            = _mm256_mul_ps(beta2,rsq21);
 642             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 643             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 644             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 645             felec            = _mm256_mul_ps(qq21,felec);
 646             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 647             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 648             velec            = _mm256_sub_ps(rinv21,pmecorrV);
 649             velec            = _mm256_mul_ps(qq21,velec);
 650
 651             /* Update potential sum for this i atom from the interaction with this j atom. */
 652             velecsum         = _mm256_add_ps(velecsum,velec);
 653
 654             fscal            = felec;
 655
 656             /* Calculate temporary vectorial force */
 657             tx               = _mm256_mul_ps(fscal,dx21);
 658             ty               = _mm256_mul_ps(fscal,dy21);
 659             tz               = _mm256_mul_ps(fscal,dz21);
 660
 661             /* Update vectorial force */
 662             fix2             = _mm256_add_ps(fix2,tx);
 663             fiy2             = _mm256_add_ps(fiy2,ty);
 664             fiz2             = _mm256_add_ps(fiz2,tz);
 665
 666             fjx1             = _mm256_add_ps(fjx1,tx);
 667             fjy1             = _mm256_add_ps(fjy1,ty);
 668             fjz1             = _mm256_add_ps(fjz1,tz);
 669
 670             /**************************
 671              * CALCULATE INTERACTIONS *
 672              **************************/
 673
 674             r22              = _mm256_mul_ps(rsq22,rinv22);
 675
 676             /* EWALD ELECTROSTATICS */
 677
 678             /* Analytical PME correction */
 679             zeta2            = _mm256_mul_ps(beta2,rsq22);
 680             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 681             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 682             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 683             felec            = _mm256_mul_ps(qq22,felec);
 684             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 685             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 686             velec            = _mm256_sub_ps(rinv22,pmecorrV);
 687             velec            = _mm256_mul_ps(qq22,velec);
 688
 689             /* Update potential sum for this i atom from the interaction with this j atom. */
 690             velecsum         = _mm256_add_ps(velecsum,velec);
 691
 692             fscal            = felec;
 693
 694             /* Calculate temporary vectorial force */
 695             tx               = _mm256_mul_ps(fscal,dx22);
 696             ty               = _mm256_mul_ps(fscal,dy22);
 697             tz               = _mm256_mul_ps(fscal,dz22);
 698
 699             /* Update vectorial force */
 700             fix2             = _mm256_add_ps(fix2,tx);
 701             fiy2             = _mm256_add_ps(fiy2,ty);
 702             fiz2             = _mm256_add_ps(fiz2,tz);
 703
 704             fjx2             = _mm256_add_ps(fjx2,tx);
 705             fjy2             = _mm256_add_ps(fjy2,ty);
 706             fjz2             = _mm256_add_ps(fjz2,tz);
 707
 708             fjptrA             = f+j_coord_offsetA;
 709             fjptrB             = f+j_coord_offsetB;
 710             fjptrC             = f+j_coord_offsetC;
 711             fjptrD             = f+j_coord_offsetD;
 712             fjptrE             = f+j_coord_offsetE;
 713             fjptrF             = f+j_coord_offsetF;
 714             fjptrG             = f+j_coord_offsetG;
 715             fjptrH             = f+j_coord_offsetH;
 716
 717             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 718                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 719
 720             /* Inner loop uses 790 flops */
 721         }
 722
 723         if(jidx<j_index_end)
 724         {
 725
 726             /* Get j neighbor index, and coordinate index */
 727             jnrlistA         = jjnr[jidx];
 728             jnrlistB         = jjnr[jidx+1];
 729             jnrlistC         = jjnr[jidx+2];
 730             jnrlistD         = jjnr[jidx+3];
 731             jnrlistE         = jjnr[jidx+4];
 732             jnrlistF         = jjnr[jidx+5];
 733             jnrlistG         = jjnr[jidx+6];
 734             jnrlistH         = jjnr[jidx+7];
 735             /* Sign of each element will be negative for non-real atoms.
 736              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 737              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 738              */
 739             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 740                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 741
 742             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 743             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 744             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 745             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 746             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 747             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 748             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 749             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 750             j_coord_offsetA  = DIM*jnrA;
 751             j_coord_offsetB  = DIM*jnrB;
 752             j_coord_offsetC  = DIM*jnrC;
 753             j_coord_offsetD  = DIM*jnrD;
 754             j_coord_offsetE  = DIM*jnrE;
 755             j_coord_offsetF  = DIM*jnrF;
 756             j_coord_offsetG  = DIM*jnrG;
 757             j_coord_offsetH  = DIM*jnrH;
 758
 759             /* load j atom coordinates */
 760             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 761                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 762                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 763                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 764                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 765
 766             /* Calculate displacement vector */
 767             dx00             = _mm256_sub_ps(ix0,jx0);
 768             dy00             = _mm256_sub_ps(iy0,jy0);
 769             dz00             = _mm256_sub_ps(iz0,jz0);
 770             dx01             = _mm256_sub_ps(ix0,jx1);
 771             dy01             = _mm256_sub_ps(iy0,jy1);
 772             dz01             = _mm256_sub_ps(iz0,jz1);
 773             dx02             = _mm256_sub_ps(ix0,jx2);
 774             dy02             = _mm256_sub_ps(iy0,jy2);
 775             dz02             = _mm256_sub_ps(iz0,jz2);
 776             dx10             = _mm256_sub_ps(ix1,jx0);
 777             dy10             = _mm256_sub_ps(iy1,jy0);
 778             dz10             = _mm256_sub_ps(iz1,jz0);
 779             dx11             = _mm256_sub_ps(ix1,jx1);
 780             dy11             = _mm256_sub_ps(iy1,jy1);
 781             dz11             = _mm256_sub_ps(iz1,jz1);
 782             dx12             = _mm256_sub_ps(ix1,jx2);
 783             dy12             = _mm256_sub_ps(iy1,jy2);
 784             dz12             = _mm256_sub_ps(iz1,jz2);
 785             dx20             = _mm256_sub_ps(ix2,jx0);
 786             dy20             = _mm256_sub_ps(iy2,jy0);
 787             dz20             = _mm256_sub_ps(iz2,jz0);
 788             dx21             = _mm256_sub_ps(ix2,jx1);
 789             dy21             = _mm256_sub_ps(iy2,jy1);
 790             dz21             = _mm256_sub_ps(iz2,jz1);
 791             dx22             = _mm256_sub_ps(ix2,jx2);
 792             dy22             = _mm256_sub_ps(iy2,jy2);
 793             dz22             = _mm256_sub_ps(iz2,jz2);
 794
 795             /* Calculate squared distance and things based on it */
 796             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 797             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 798             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 799             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 800             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 801             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 802             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 803             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 804             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 805
 806             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 807             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 808             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 809             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 810             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 811             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 812             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 813             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 814             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 815
 816             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 817             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 818             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 819             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 820             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 821             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 822             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 823             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 824             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 825
 826             fjx0             = _mm256_setzero_ps();
 827             fjy0             = _mm256_setzero_ps();
 828             fjz0             = _mm256_setzero_ps();
 829             fjx1             = _mm256_setzero_ps();
 830             fjy1             = _mm256_setzero_ps();
 831             fjz1             = _mm256_setzero_ps();
 832             fjx2             = _mm256_setzero_ps();
 833             fjy2             = _mm256_setzero_ps();
 834             fjz2             = _mm256_setzero_ps();
 835
 836             /**************************
 837              * CALCULATE INTERACTIONS *
 838              **************************/
 839
 840             r00              = _mm256_mul_ps(rsq00,rinv00);
 841             r00              = _mm256_andnot_ps(dummy_mask,r00);
 842
 843             /* Calculate table index by multiplying r with table scale and truncate to integer */
 844             rt               = _mm256_mul_ps(r00,vftabscale);
 845             vfitab           = _mm256_cvttps_epi32(rt);
 846             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 847             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 848             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 849             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 850             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 851             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 852
 853             /* EWALD ELECTROSTATICS */
 854
 855             /* Analytical PME correction */
 856             zeta2            = _mm256_mul_ps(beta2,rsq00);
 857             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 858             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 859             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 860             felec            = _mm256_mul_ps(qq00,felec);
 861             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 862             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 863             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 864             velec            = _mm256_mul_ps(qq00,velec);
 865
 866             /* CUBIC SPLINE TABLE DISPERSION */
 867             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 868                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 869             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 870                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 871             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 872                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 873             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 874                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 875             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 876             Heps             = _mm256_mul_ps(vfeps,H);
 877             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 878             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 879             vvdw6            = _mm256_mul_ps(c6_00,VV);
 880             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 881             fvdw6            = _mm256_mul_ps(c6_00,FF);
 882
 883             /* CUBIC SPLINE TABLE REPULSION */
 884             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 885             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 886             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 887                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 888             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 889                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 890             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 891                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 892             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 893                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 894             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 895             Heps             = _mm256_mul_ps(vfeps,H);
 896             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 897             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 898             vvdw12           = _mm256_mul_ps(c12_00,VV);
 899             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 900             fvdw12           = _mm256_mul_ps(c12_00,FF);
 901             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 902             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 903
 904             /* Update potential sum for this i atom from the interaction with this j atom. */
 905             velec            = _mm256_andnot_ps(dummy_mask,velec);
 906             velecsum         = _mm256_add_ps(velecsum,velec);
 907             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 908             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 909
 910             fscal            = _mm256_add_ps(felec,fvdw);
 911
 912             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 913
 914             /* Calculate temporary vectorial force */
 915             tx               = _mm256_mul_ps(fscal,dx00);
 916             ty               = _mm256_mul_ps(fscal,dy00);
 917             tz               = _mm256_mul_ps(fscal,dz00);
 918
 919             /* Update vectorial force */
 920             fix0             = _mm256_add_ps(fix0,tx);
 921             fiy0             = _mm256_add_ps(fiy0,ty);
 922             fiz0             = _mm256_add_ps(fiz0,tz);
 923
 924             fjx0             = _mm256_add_ps(fjx0,tx);
 925             fjy0             = _mm256_add_ps(fjy0,ty);
 926             fjz0             = _mm256_add_ps(fjz0,tz);
 927
 928             /**************************
 929              * CALCULATE INTERACTIONS *
 930              **************************/
 931
 932             r01              = _mm256_mul_ps(rsq01,rinv01);
 933             r01              = _mm256_andnot_ps(dummy_mask,r01);
 934
 935             /* EWALD ELECTROSTATICS */
 936
 937             /* Analytical PME correction */
 938             zeta2            = _mm256_mul_ps(beta2,rsq01);
 939             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 940             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 941             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 942             felec            = _mm256_mul_ps(qq01,felec);
 943             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 944             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 945             velec            = _mm256_sub_ps(rinv01,pmecorrV);
 946             velec            = _mm256_mul_ps(qq01,velec);
 947
 948             /* Update potential sum for this i atom from the interaction with this j atom. */
 949             velec            = _mm256_andnot_ps(dummy_mask,velec);
 950             velecsum         = _mm256_add_ps(velecsum,velec);
 951
 952             fscal            = felec;
 953
 954             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 955
 956             /* Calculate temporary vectorial force */
 957             tx               = _mm256_mul_ps(fscal,dx01);
 958             ty               = _mm256_mul_ps(fscal,dy01);
 959             tz               = _mm256_mul_ps(fscal,dz01);
 960
 961             /* Update vectorial force */
 962             fix0             = _mm256_add_ps(fix0,tx);
 963             fiy0             = _mm256_add_ps(fiy0,ty);
 964             fiz0             = _mm256_add_ps(fiz0,tz);
 965
 966             fjx1             = _mm256_add_ps(fjx1,tx);
 967             fjy1             = _mm256_add_ps(fjy1,ty);
 968             fjz1             = _mm256_add_ps(fjz1,tz);
 969
 970             /**************************
 971              * CALCULATE INTERACTIONS *
 972              **************************/
 973
 974             r02              = _mm256_mul_ps(rsq02,rinv02);
 975             r02              = _mm256_andnot_ps(dummy_mask,r02);
 976
 977             /* EWALD ELECTROSTATICS */
 978
 979             /* Analytical PME correction */
 980             zeta2            = _mm256_mul_ps(beta2,rsq02);
 981             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 982             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 983             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 984             felec            = _mm256_mul_ps(qq02,felec);
 985             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 986             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 987             velec            = _mm256_sub_ps(rinv02,pmecorrV);
 988             velec            = _mm256_mul_ps(qq02,velec);
 989
 990             /* Update potential sum for this i atom from the interaction with this j atom. */
 991             velec            = _mm256_andnot_ps(dummy_mask,velec);
 992             velecsum         = _mm256_add_ps(velecsum,velec);
 993
 994             fscal            = felec;
 995
 996             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 997
 998             /* Calculate temporary vectorial force */
 999             tx               = _mm256_mul_ps(fscal,dx02);
1000             ty               = _mm256_mul_ps(fscal,dy02);
1001             tz               = _mm256_mul_ps(fscal,dz02);
1002
1003             /* Update vectorial force */
1004             fix0             = _mm256_add_ps(fix0,tx);
1005             fiy0             = _mm256_add_ps(fiy0,ty);
1006             fiz0             = _mm256_add_ps(fiz0,tz);
1007
1008             fjx2             = _mm256_add_ps(fjx2,tx);
1009             fjy2             = _mm256_add_ps(fjy2,ty);
1010             fjz2             = _mm256_add_ps(fjz2,tz);
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             r10              = _mm256_mul_ps(rsq10,rinv10);
1017             r10              = _mm256_andnot_ps(dummy_mask,r10);
1018
1019             /* EWALD ELECTROSTATICS */
1020
1021             /* Analytical PME correction */
1022             zeta2            = _mm256_mul_ps(beta2,rsq10);
1023             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1024             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1025             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1026             felec            = _mm256_mul_ps(qq10,felec);
1027             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1028             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1029             velec            = _mm256_sub_ps(rinv10,pmecorrV);
1030             velec            = _mm256_mul_ps(qq10,velec);
1031
1032             /* Update potential sum for this i atom from the interaction with this j atom. */
1033             velec            = _mm256_andnot_ps(dummy_mask,velec);
1034             velecsum         = _mm256_add_ps(velecsum,velec);
1035
1036             fscal            = felec;
1037
1038             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1039
1040             /* Calculate temporary vectorial force */
1041             tx               = _mm256_mul_ps(fscal,dx10);
1042             ty               = _mm256_mul_ps(fscal,dy10);
1043             tz               = _mm256_mul_ps(fscal,dz10);
1044
1045             /* Update vectorial force */
1046             fix1             = _mm256_add_ps(fix1,tx);
1047             fiy1             = _mm256_add_ps(fiy1,ty);
1048             fiz1             = _mm256_add_ps(fiz1,tz);
1049
1050             fjx0             = _mm256_add_ps(fjx0,tx);
1051             fjy0             = _mm256_add_ps(fjy0,ty);
1052             fjz0             = _mm256_add_ps(fjz0,tz);
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             r11              = _mm256_mul_ps(rsq11,rinv11);
1059             r11              = _mm256_andnot_ps(dummy_mask,r11);
1060
1061             /* EWALD ELECTROSTATICS */
1062
1063             /* Analytical PME correction */
1064             zeta2            = _mm256_mul_ps(beta2,rsq11);
1065             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1066             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1067             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1068             felec            = _mm256_mul_ps(qq11,felec);
1069             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1070             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1071             velec            = _mm256_sub_ps(rinv11,pmecorrV);
1072             velec            = _mm256_mul_ps(qq11,velec);
1073
1074             /* Update potential sum for this i atom from the interaction with this j atom. */
1075             velec            = _mm256_andnot_ps(dummy_mask,velec);
1076             velecsum         = _mm256_add_ps(velecsum,velec);
1077
1078             fscal            = felec;
1079
1080             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1081
1082             /* Calculate temporary vectorial force */
1083             tx               = _mm256_mul_ps(fscal,dx11);
1084             ty               = _mm256_mul_ps(fscal,dy11);
1085             tz               = _mm256_mul_ps(fscal,dz11);
1086
1087             /* Update vectorial force */
1088             fix1             = _mm256_add_ps(fix1,tx);
1089             fiy1             = _mm256_add_ps(fiy1,ty);
1090             fiz1             = _mm256_add_ps(fiz1,tz);
1091
1092             fjx1             = _mm256_add_ps(fjx1,tx);
1093             fjy1             = _mm256_add_ps(fjy1,ty);
1094             fjz1             = _mm256_add_ps(fjz1,tz);
1095
1096             /**************************
1097              * CALCULATE INTERACTIONS *
1098              **************************/
1099
1100             r12              = _mm256_mul_ps(rsq12,rinv12);
1101             r12              = _mm256_andnot_ps(dummy_mask,r12);
1102
1103             /* EWALD ELECTROSTATICS */
1104
1105             /* Analytical PME correction */
1106             zeta2            = _mm256_mul_ps(beta2,rsq12);
1107             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1108             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1109             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1110             felec            = _mm256_mul_ps(qq12,felec);
1111             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1112             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1113             velec            = _mm256_sub_ps(rinv12,pmecorrV);
1114             velec            = _mm256_mul_ps(qq12,velec);
1115
1116             /* Update potential sum for this i atom from the interaction with this j atom. */
1117             velec            = _mm256_andnot_ps(dummy_mask,velec);
1118             velecsum         = _mm256_add_ps(velecsum,velec);
1119
1120             fscal            = felec;
1121
1122             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1123
1124             /* Calculate temporary vectorial force */
1125             tx               = _mm256_mul_ps(fscal,dx12);
1126             ty               = _mm256_mul_ps(fscal,dy12);
1127             tz               = _mm256_mul_ps(fscal,dz12);
1128
1129             /* Update vectorial force */
1130             fix1             = _mm256_add_ps(fix1,tx);
1131             fiy1             = _mm256_add_ps(fiy1,ty);
1132             fiz1             = _mm256_add_ps(fiz1,tz);
1133
1134             fjx2             = _mm256_add_ps(fjx2,tx);
1135             fjy2             = _mm256_add_ps(fjy2,ty);
1136             fjz2             = _mm256_add_ps(fjz2,tz);
1137
1138             /**************************
1139              * CALCULATE INTERACTIONS *
1140              **************************/
1141
1142             r20              = _mm256_mul_ps(rsq20,rinv20);
1143             r20              = _mm256_andnot_ps(dummy_mask,r20);
1144
1145             /* EWALD ELECTROSTATICS */
1146
1147             /* Analytical PME correction */
1148             zeta2            = _mm256_mul_ps(beta2,rsq20);
1149             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1150             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1151             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1152             felec            = _mm256_mul_ps(qq20,felec);
1153             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1154             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1155             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1156             velec            = _mm256_mul_ps(qq20,velec);
1157
1158             /* Update potential sum for this i atom from the interaction with this j atom. */
1159             velec            = _mm256_andnot_ps(dummy_mask,velec);
1160             velecsum         = _mm256_add_ps(velecsum,velec);
1161
1162             fscal            = felec;
1163
1164             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1165
1166             /* Calculate temporary vectorial force */
1167             tx               = _mm256_mul_ps(fscal,dx20);
1168             ty               = _mm256_mul_ps(fscal,dy20);
1169             tz               = _mm256_mul_ps(fscal,dz20);
1170
1171             /* Update vectorial force */
1172             fix2             = _mm256_add_ps(fix2,tx);
1173             fiy2             = _mm256_add_ps(fiy2,ty);
1174             fiz2             = _mm256_add_ps(fiz2,tz);
1175
1176             fjx0             = _mm256_add_ps(fjx0,tx);
1177             fjy0             = _mm256_add_ps(fjy0,ty);
1178             fjz0             = _mm256_add_ps(fjz0,tz);
1179
1180             /**************************
1181              * CALCULATE INTERACTIONS *
1182              **************************/
1183
1184             r21              = _mm256_mul_ps(rsq21,rinv21);
1185             r21              = _mm256_andnot_ps(dummy_mask,r21);
1186
1187             /* EWALD ELECTROSTATICS */
1188
1189             /* Analytical PME correction */
1190             zeta2            = _mm256_mul_ps(beta2,rsq21);
1191             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1192             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1193             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1194             felec            = _mm256_mul_ps(qq21,felec);
1195             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1196             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1197             velec            = _mm256_sub_ps(rinv21,pmecorrV);
1198             velec            = _mm256_mul_ps(qq21,velec);
1199
1200             /* Update potential sum for this i atom from the interaction with this j atom. */
1201             velec            = _mm256_andnot_ps(dummy_mask,velec);
1202             velecsum         = _mm256_add_ps(velecsum,velec);
1203
1204             fscal            = felec;
1205
1206             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1207
1208             /* Calculate temporary vectorial force */
1209             tx               = _mm256_mul_ps(fscal,dx21);
1210             ty               = _mm256_mul_ps(fscal,dy21);
1211             tz               = _mm256_mul_ps(fscal,dz21);
1212
1213             /* Update vectorial force */
1214             fix2             = _mm256_add_ps(fix2,tx);
1215             fiy2             = _mm256_add_ps(fiy2,ty);
1216             fiz2             = _mm256_add_ps(fiz2,tz);
1217
1218             fjx1             = _mm256_add_ps(fjx1,tx);
1219             fjy1             = _mm256_add_ps(fjy1,ty);
1220             fjz1             = _mm256_add_ps(fjz1,tz);
1221
1222             /**************************
1223              * CALCULATE INTERACTIONS *
1224              **************************/
1225
1226             r22              = _mm256_mul_ps(rsq22,rinv22);
1227             r22              = _mm256_andnot_ps(dummy_mask,r22);
1228
1229             /* EWALD ELECTROSTATICS */
1230
1231             /* Analytical PME correction */
1232             zeta2            = _mm256_mul_ps(beta2,rsq22);
1233             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1234             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1235             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1236             felec            = _mm256_mul_ps(qq22,felec);
1237             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1238             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1239             velec            = _mm256_sub_ps(rinv22,pmecorrV);
1240             velec            = _mm256_mul_ps(qq22,velec);
1241
1242             /* Update potential sum for this i atom from the interaction with this j atom. */
1243             velec            = _mm256_andnot_ps(dummy_mask,velec);
1244             velecsum         = _mm256_add_ps(velecsum,velec);
1245
1246             fscal            = felec;
1247
1248             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1249
1250             /* Calculate temporary vectorial force */
1251             tx               = _mm256_mul_ps(fscal,dx22);
1252             ty               = _mm256_mul_ps(fscal,dy22);
1253             tz               = _mm256_mul_ps(fscal,dz22);
1254
1255             /* Update vectorial force */
1256             fix2             = _mm256_add_ps(fix2,tx);
1257             fiy2             = _mm256_add_ps(fiy2,ty);
1258             fiz2             = _mm256_add_ps(fiz2,tz);
1259
1260             fjx2             = _mm256_add_ps(fjx2,tx);
1261             fjy2             = _mm256_add_ps(fjy2,ty);
1262             fjz2             = _mm256_add_ps(fjz2,tz);
1263
1264             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1265             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1266             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1267             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1268             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1269             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1270             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1271             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1272
1273             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1274                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1275
1276             /* Inner loop uses 799 flops */
1277         }
1278
1279         /* End of innermost loop */
1280
1281         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1282                                                  f+i_coord_offset,fshift+i_shift_offset);
1283
1284         ggid                        = gid[iidx];
1285         /* Update potential energies */
1286         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1287         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1288
1289         /* Increment number of inner iterations */
1290         inneriter                  += j_index_end - j_index_start;
1291
1292         /* Outer loop uses 20 flops */
1293     }
1294
1295     /* Increment number of outer iterations */
1296     outeriter        += nri;
1297
1298     /* Update outer/inner flops */
1299
1300     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*799);
1301 }
1302 /*
1303  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3W3_F_avx_256_single
1304  * Electrostatics interaction: Ewald
1305  * VdW interaction:            CubicSplineTable
1306  * Geometry:                   Water3-Water3
1307  * Calculate force/pot:        Force
1308  */
1309 void
1310 nb_kernel_ElecEw_VdwCSTab_GeomW3W3_F_avx_256_single
1311                     (t_nblist * gmx_restrict                nlist,
1312                      rvec * gmx_restrict                    xx,
1313                      rvec * gmx_restrict                    ff,
1314                      t_forcerec * gmx_restrict              fr,
1315                      t_mdatoms * gmx_restrict               mdatoms,
1316                      nb_kernel_data_t * gmx_restrict        kernel_data,
1317                      t_nrnb * gmx_restrict                  nrnb)
1318 {
1319     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1320      * just 0 for non-waters.
1321      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1322      * jnr indices corresponding to data put in the four positions in the SIMD register.
1323      */
1324     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1325     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1326     int              jnrA,jnrB,jnrC,jnrD;
1327     int              jnrE,jnrF,jnrG,jnrH;
1328     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1329     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1330     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1331     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1332     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1333     real             rcutoff_scalar;
1334     real             *shiftvec,*fshift,*x,*f;
1335     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1336     real             scratch[4*DIM];
1337     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1338     real *           vdwioffsetptr0;
1339     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1340     real *           vdwioffsetptr1;
1341     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1342     real *           vdwioffsetptr2;
1343     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1344     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1345     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1346     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1347     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1348     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1349     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1350     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1351     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1352     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1353     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1354     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1355     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1356     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1357     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1358     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1359     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1360     real             *charge;
1361     int              nvdwtype;
1362     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1363     int              *vdwtype;
1364     real             *vdwparam;
1365     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1366     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1367     __m256i          vfitab;
1368     __m128i          vfitab_lo,vfitab_hi;
1369     __m128i          ifour       = _mm_set1_epi32(4);
1370     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1371     real             *vftab;
1372     __m256i          ewitab;
1373     __m128i          ewitab_lo,ewitab_hi;
1374     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1375     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1376     real             *ewtab;
1377     __m256           dummy_mask,cutoff_mask;
1378     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1379     __m256           one     = _mm256_set1_ps(1.0);
1380     __m256           two     = _mm256_set1_ps(2.0);
1381     x                = xx[0];
1382     f                = ff[0];
1383
1384     nri              = nlist->nri;
1385     iinr             = nlist->iinr;
1386     jindex           = nlist->jindex;
1387     jjnr             = nlist->jjnr;
1388     shiftidx         = nlist->shift;
1389     gid              = nlist->gid;
1390     shiftvec         = fr->shift_vec[0];
1391     fshift           = fr->fshift[0];
1392     facel            = _mm256_set1_ps(fr->epsfac);
1393     charge           = mdatoms->chargeA;
1394     nvdwtype         = fr->ntype;
1395     vdwparam         = fr->nbfp;
1396     vdwtype          = mdatoms->typeA;
1397
1398     vftab            = kernel_data->table_vdw->data;
1399     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
1400
1401     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1402     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
1403     beta2            = _mm256_mul_ps(beta,beta);
1404     beta3            = _mm256_mul_ps(beta,beta2);
1405
1406     ewtab            = fr->ic->tabq_coul_F;
1407     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1408     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1409
1410     /* Setup water-specific parameters */
1411     inr              = nlist->iinr[0];
1412     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1413     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1414     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1415     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1416
1417     jq0              = _mm256_set1_ps(charge[inr+0]);
1418     jq1              = _mm256_set1_ps(charge[inr+1]);
1419     jq2              = _mm256_set1_ps(charge[inr+2]);
1420     vdwjidx0A        = 2*vdwtype[inr+0];
1421     qq00             = _mm256_mul_ps(iq0,jq0);
1422     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1423     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1424     qq01             = _mm256_mul_ps(iq0,jq1);
1425     qq02             = _mm256_mul_ps(iq0,jq2);
1426     qq10             = _mm256_mul_ps(iq1,jq0);
1427     qq11             = _mm256_mul_ps(iq1,jq1);
1428     qq12             = _mm256_mul_ps(iq1,jq2);
1429     qq20             = _mm256_mul_ps(iq2,jq0);
1430     qq21             = _mm256_mul_ps(iq2,jq1);
1431     qq22             = _mm256_mul_ps(iq2,jq2);
1432
1433     /* Avoid stupid compiler warnings */
1434     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1435     j_coord_offsetA = 0;
1436     j_coord_offsetB = 0;
1437     j_coord_offsetC = 0;
1438     j_coord_offsetD = 0;
1439     j_coord_offsetE = 0;
1440     j_coord_offsetF = 0;
1441     j_coord_offsetG = 0;
1442     j_coord_offsetH = 0;
1443
1444     outeriter        = 0;
1445     inneriter        = 0;
1446
1447     for(iidx=0;iidx<4*DIM;iidx++)
1448     {
1449         scratch[iidx] = 0.0;
1450     }
1451
1452     /* Start outer loop over neighborlists */
1453     for(iidx=0; iidx<nri; iidx++)
1454     {
1455         /* Load shift vector for this list */
1456         i_shift_offset   = DIM*shiftidx[iidx];
1457
1458         /* Load limits for loop over neighbors */
1459         j_index_start    = jindex[iidx];
1460         j_index_end      = jindex[iidx+1];
1461
1462         /* Get outer coordinate index */
1463         inr              = iinr[iidx];
1464         i_coord_offset   = DIM*inr;
1465
1466         /* Load i particle coords and add shift vector */
1467         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1468                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1469
1470         fix0             = _mm256_setzero_ps();
1471         fiy0             = _mm256_setzero_ps();
1472         fiz0             = _mm256_setzero_ps();
1473         fix1             = _mm256_setzero_ps();
1474         fiy1             = _mm256_setzero_ps();
1475         fiz1             = _mm256_setzero_ps();
1476         fix2             = _mm256_setzero_ps();
1477         fiy2             = _mm256_setzero_ps();
1478         fiz2             = _mm256_setzero_ps();
1479
1480         /* Start inner kernel loop */
1481         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1482         {
1483
1484             /* Get j neighbor index, and coordinate index */
1485             jnrA             = jjnr[jidx];
1486             jnrB             = jjnr[jidx+1];
1487             jnrC             = jjnr[jidx+2];
1488             jnrD             = jjnr[jidx+3];
1489             jnrE             = jjnr[jidx+4];
1490             jnrF             = jjnr[jidx+5];
1491             jnrG             = jjnr[jidx+6];
1492             jnrH             = jjnr[jidx+7];
1493             j_coord_offsetA  = DIM*jnrA;
1494             j_coord_offsetB  = DIM*jnrB;
1495             j_coord_offsetC  = DIM*jnrC;
1496             j_coord_offsetD  = DIM*jnrD;
1497             j_coord_offsetE  = DIM*jnrE;
1498             j_coord_offsetF  = DIM*jnrF;
1499             j_coord_offsetG  = DIM*jnrG;
1500             j_coord_offsetH  = DIM*jnrH;
1501
1502             /* load j atom coordinates */
1503             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1504                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1505                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1506                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1507                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1508
1509             /* Calculate displacement vector */
1510             dx00             = _mm256_sub_ps(ix0,jx0);
1511             dy00             = _mm256_sub_ps(iy0,jy0);
1512             dz00             = _mm256_sub_ps(iz0,jz0);
1513             dx01             = _mm256_sub_ps(ix0,jx1);
1514             dy01             = _mm256_sub_ps(iy0,jy1);
1515             dz01             = _mm256_sub_ps(iz0,jz1);
1516             dx02             = _mm256_sub_ps(ix0,jx2);
1517             dy02             = _mm256_sub_ps(iy0,jy2);
1518             dz02             = _mm256_sub_ps(iz0,jz2);
1519             dx10             = _mm256_sub_ps(ix1,jx0);
1520             dy10             = _mm256_sub_ps(iy1,jy0);
1521             dz10             = _mm256_sub_ps(iz1,jz0);
1522             dx11             = _mm256_sub_ps(ix1,jx1);
1523             dy11             = _mm256_sub_ps(iy1,jy1);
1524             dz11             = _mm256_sub_ps(iz1,jz1);
1525             dx12             = _mm256_sub_ps(ix1,jx2);
1526             dy12             = _mm256_sub_ps(iy1,jy2);
1527             dz12             = _mm256_sub_ps(iz1,jz2);
1528             dx20             = _mm256_sub_ps(ix2,jx0);
1529             dy20             = _mm256_sub_ps(iy2,jy0);
1530             dz20             = _mm256_sub_ps(iz2,jz0);
1531             dx21             = _mm256_sub_ps(ix2,jx1);
1532             dy21             = _mm256_sub_ps(iy2,jy1);
1533             dz21             = _mm256_sub_ps(iz2,jz1);
1534             dx22             = _mm256_sub_ps(ix2,jx2);
1535             dy22             = _mm256_sub_ps(iy2,jy2);
1536             dz22             = _mm256_sub_ps(iz2,jz2);
1537
1538             /* Calculate squared distance and things based on it */
1539             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1540             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1541             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1542             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1543             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1544             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1545             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1546             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1547             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1548
1549             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1550             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1551             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1552             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1553             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1554             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1555             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1556             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1557             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1558
1559             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1560             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1561             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1562             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1563             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1564             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1565             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1566             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1567             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1568
1569             fjx0             = _mm256_setzero_ps();
1570             fjy0             = _mm256_setzero_ps();
1571             fjz0             = _mm256_setzero_ps();
1572             fjx1             = _mm256_setzero_ps();
1573             fjy1             = _mm256_setzero_ps();
1574             fjz1             = _mm256_setzero_ps();
1575             fjx2             = _mm256_setzero_ps();
1576             fjy2             = _mm256_setzero_ps();
1577             fjz2             = _mm256_setzero_ps();
1578
1579             /**************************
1580              * CALCULATE INTERACTIONS *
1581              **************************/
1582
1583             r00              = _mm256_mul_ps(rsq00,rinv00);
1584
1585             /* Calculate table index by multiplying r with table scale and truncate to integer */
1586             rt               = _mm256_mul_ps(r00,vftabscale);
1587             vfitab           = _mm256_cvttps_epi32(rt);
1588             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1589             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1590             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1591             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1592             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1593             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1594
1595             /* EWALD ELECTROSTATICS */
1596
1597             /* Analytical PME correction */
1598             zeta2            = _mm256_mul_ps(beta2,rsq00);
1599             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1600             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1601             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1602             felec            = _mm256_mul_ps(qq00,felec);
1603
1604             /* CUBIC SPLINE TABLE DISPERSION */
1605             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1606                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1607             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1608                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1609             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1610                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1611             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1612                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1613             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1614             Heps             = _mm256_mul_ps(vfeps,H);
1615             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1616             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1617             fvdw6            = _mm256_mul_ps(c6_00,FF);
1618
1619             /* CUBIC SPLINE TABLE REPULSION */
1620             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1621             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1622             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1623                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1624             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1625                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1626             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1627                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1628             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1629                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1630             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1631             Heps             = _mm256_mul_ps(vfeps,H);
1632             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1633             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1634             fvdw12           = _mm256_mul_ps(c12_00,FF);
1635             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1636
1637             fscal            = _mm256_add_ps(felec,fvdw);
1638
1639             /* Calculate temporary vectorial force */
1640             tx               = _mm256_mul_ps(fscal,dx00);
1641             ty               = _mm256_mul_ps(fscal,dy00);
1642             tz               = _mm256_mul_ps(fscal,dz00);
1643
1644             /* Update vectorial force */
1645             fix0             = _mm256_add_ps(fix0,tx);
1646             fiy0             = _mm256_add_ps(fiy0,ty);
1647             fiz0             = _mm256_add_ps(fiz0,tz);
1648
1649             fjx0             = _mm256_add_ps(fjx0,tx);
1650             fjy0             = _mm256_add_ps(fjy0,ty);
1651             fjz0             = _mm256_add_ps(fjz0,tz);
1652
1653             /**************************
1654              * CALCULATE INTERACTIONS *
1655              **************************/
1656
1657             r01              = _mm256_mul_ps(rsq01,rinv01);
1658
1659             /* EWALD ELECTROSTATICS */
1660
1661             /* Analytical PME correction */
1662             zeta2            = _mm256_mul_ps(beta2,rsq01);
1663             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
1664             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1665             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1666             felec            = _mm256_mul_ps(qq01,felec);
1667
1668             fscal            = felec;
1669
1670             /* Calculate temporary vectorial force */
1671             tx               = _mm256_mul_ps(fscal,dx01);
1672             ty               = _mm256_mul_ps(fscal,dy01);
1673             tz               = _mm256_mul_ps(fscal,dz01);
1674
1675             /* Update vectorial force */
1676             fix0             = _mm256_add_ps(fix0,tx);
1677             fiy0             = _mm256_add_ps(fiy0,ty);
1678             fiz0             = _mm256_add_ps(fiz0,tz);
1679
1680             fjx1             = _mm256_add_ps(fjx1,tx);
1681             fjy1             = _mm256_add_ps(fjy1,ty);
1682             fjz1             = _mm256_add_ps(fjz1,tz);
1683
1684             /**************************
1685              * CALCULATE INTERACTIONS *
1686              **************************/
1687
1688             r02              = _mm256_mul_ps(rsq02,rinv02);
1689
1690             /* EWALD ELECTROSTATICS */
1691
1692             /* Analytical PME correction */
1693             zeta2            = _mm256_mul_ps(beta2,rsq02);
1694             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
1695             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1696             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1697             felec            = _mm256_mul_ps(qq02,felec);
1698
1699             fscal            = felec;
1700
1701             /* Calculate temporary vectorial force */
1702             tx               = _mm256_mul_ps(fscal,dx02);
1703             ty               = _mm256_mul_ps(fscal,dy02);
1704             tz               = _mm256_mul_ps(fscal,dz02);
1705
1706             /* Update vectorial force */
1707             fix0             = _mm256_add_ps(fix0,tx);
1708             fiy0             = _mm256_add_ps(fiy0,ty);
1709             fiz0             = _mm256_add_ps(fiz0,tz);
1710
1711             fjx2             = _mm256_add_ps(fjx2,tx);
1712             fjy2             = _mm256_add_ps(fjy2,ty);
1713             fjz2             = _mm256_add_ps(fjz2,tz);
1714
1715             /**************************
1716              * CALCULATE INTERACTIONS *
1717              **************************/
1718
1719             r10              = _mm256_mul_ps(rsq10,rinv10);
1720
1721             /* EWALD ELECTROSTATICS */
1722
1723             /* Analytical PME correction */
1724             zeta2            = _mm256_mul_ps(beta2,rsq10);
1725             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1726             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1727             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1728             felec            = _mm256_mul_ps(qq10,felec);
1729
1730             fscal            = felec;
1731
1732             /* Calculate temporary vectorial force */
1733             tx               = _mm256_mul_ps(fscal,dx10);
1734             ty               = _mm256_mul_ps(fscal,dy10);
1735             tz               = _mm256_mul_ps(fscal,dz10);
1736
1737             /* Update vectorial force */
1738             fix1             = _mm256_add_ps(fix1,tx);
1739             fiy1             = _mm256_add_ps(fiy1,ty);
1740             fiz1             = _mm256_add_ps(fiz1,tz);
1741
1742             fjx0             = _mm256_add_ps(fjx0,tx);
1743             fjy0             = _mm256_add_ps(fjy0,ty);
1744             fjz0             = _mm256_add_ps(fjz0,tz);
1745
1746             /**************************
1747              * CALCULATE INTERACTIONS *
1748              **************************/
1749
1750             r11              = _mm256_mul_ps(rsq11,rinv11);
1751
1752             /* EWALD ELECTROSTATICS */
1753
1754             /* Analytical PME correction */
1755             zeta2            = _mm256_mul_ps(beta2,rsq11);
1756             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1757             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1758             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1759             felec            = _mm256_mul_ps(qq11,felec);
1760
1761             fscal            = felec;
1762
1763             /* Calculate temporary vectorial force */
1764             tx               = _mm256_mul_ps(fscal,dx11);
1765             ty               = _mm256_mul_ps(fscal,dy11);
1766             tz               = _mm256_mul_ps(fscal,dz11);
1767
1768             /* Update vectorial force */
1769             fix1             = _mm256_add_ps(fix1,tx);
1770             fiy1             = _mm256_add_ps(fiy1,ty);
1771             fiz1             = _mm256_add_ps(fiz1,tz);
1772
1773             fjx1             = _mm256_add_ps(fjx1,tx);
1774             fjy1             = _mm256_add_ps(fjy1,ty);
1775             fjz1             = _mm256_add_ps(fjz1,tz);
1776
1777             /**************************
1778              * CALCULATE INTERACTIONS *
1779              **************************/
1780
1781             r12              = _mm256_mul_ps(rsq12,rinv12);
1782
1783             /* EWALD ELECTROSTATICS */
1784
1785             /* Analytical PME correction */
1786             zeta2            = _mm256_mul_ps(beta2,rsq12);
1787             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1788             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1789             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1790             felec            = _mm256_mul_ps(qq12,felec);
1791
1792             fscal            = felec;
1793
1794             /* Calculate temporary vectorial force */
1795             tx               = _mm256_mul_ps(fscal,dx12);
1796             ty               = _mm256_mul_ps(fscal,dy12);
1797             tz               = _mm256_mul_ps(fscal,dz12);
1798
1799             /* Update vectorial force */
1800             fix1             = _mm256_add_ps(fix1,tx);
1801             fiy1             = _mm256_add_ps(fiy1,ty);
1802             fiz1             = _mm256_add_ps(fiz1,tz);
1803
1804             fjx2             = _mm256_add_ps(fjx2,tx);
1805             fjy2             = _mm256_add_ps(fjy2,ty);
1806             fjz2             = _mm256_add_ps(fjz2,tz);
1807
1808             /**************************
1809              * CALCULATE INTERACTIONS *
1810              **************************/
1811
1812             r20              = _mm256_mul_ps(rsq20,rinv20);
1813
1814             /* EWALD ELECTROSTATICS */
1815
1816             /* Analytical PME correction */
1817             zeta2            = _mm256_mul_ps(beta2,rsq20);
1818             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1819             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1820             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1821             felec            = _mm256_mul_ps(qq20,felec);
1822
1823             fscal            = felec;
1824
1825             /* Calculate temporary vectorial force */
1826             tx               = _mm256_mul_ps(fscal,dx20);
1827             ty               = _mm256_mul_ps(fscal,dy20);
1828             tz               = _mm256_mul_ps(fscal,dz20);
1829
1830             /* Update vectorial force */
1831             fix2             = _mm256_add_ps(fix2,tx);
1832             fiy2             = _mm256_add_ps(fiy2,ty);
1833             fiz2             = _mm256_add_ps(fiz2,tz);
1834
1835             fjx0             = _mm256_add_ps(fjx0,tx);
1836             fjy0             = _mm256_add_ps(fjy0,ty);
1837             fjz0             = _mm256_add_ps(fjz0,tz);
1838
1839             /**************************
1840              * CALCULATE INTERACTIONS *
1841              **************************/
1842
1843             r21              = _mm256_mul_ps(rsq21,rinv21);
1844
1845             /* EWALD ELECTROSTATICS */
1846
1847             /* Analytical PME correction */
1848             zeta2            = _mm256_mul_ps(beta2,rsq21);
1849             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1850             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1851             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1852             felec            = _mm256_mul_ps(qq21,felec);
1853
1854             fscal            = felec;
1855
1856             /* Calculate temporary vectorial force */
1857             tx               = _mm256_mul_ps(fscal,dx21);
1858             ty               = _mm256_mul_ps(fscal,dy21);
1859             tz               = _mm256_mul_ps(fscal,dz21);
1860
1861             /* Update vectorial force */
1862             fix2             = _mm256_add_ps(fix2,tx);
1863             fiy2             = _mm256_add_ps(fiy2,ty);
1864             fiz2             = _mm256_add_ps(fiz2,tz);
1865
1866             fjx1             = _mm256_add_ps(fjx1,tx);
1867             fjy1             = _mm256_add_ps(fjy1,ty);
1868             fjz1             = _mm256_add_ps(fjz1,tz);
1869
1870             /**************************
1871              * CALCULATE INTERACTIONS *
1872              **************************/
1873
1874             r22              = _mm256_mul_ps(rsq22,rinv22);
1875
1876             /* EWALD ELECTROSTATICS */
1877
1878             /* Analytical PME correction */
1879             zeta2            = _mm256_mul_ps(beta2,rsq22);
1880             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1881             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1882             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1883             felec            = _mm256_mul_ps(qq22,felec);
1884
1885             fscal            = felec;
1886
1887             /* Calculate temporary vectorial force */
1888             tx               = _mm256_mul_ps(fscal,dx22);
1889             ty               = _mm256_mul_ps(fscal,dy22);
1890             tz               = _mm256_mul_ps(fscal,dz22);
1891
1892             /* Update vectorial force */
1893             fix2             = _mm256_add_ps(fix2,tx);
1894             fiy2             = _mm256_add_ps(fiy2,ty);
1895             fiz2             = _mm256_add_ps(fiz2,tz);
1896
1897             fjx2             = _mm256_add_ps(fjx2,tx);
1898             fjy2             = _mm256_add_ps(fjy2,ty);
1899             fjz2             = _mm256_add_ps(fjz2,tz);
1900
1901             fjptrA             = f+j_coord_offsetA;
1902             fjptrB             = f+j_coord_offsetB;
1903             fjptrC             = f+j_coord_offsetC;
1904             fjptrD             = f+j_coord_offsetD;
1905             fjptrE             = f+j_coord_offsetE;
1906             fjptrF             = f+j_coord_offsetF;
1907             fjptrG             = f+j_coord_offsetG;
1908             fjptrH             = f+j_coord_offsetH;
1909
1910             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1911                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1912
1913             /* Inner loop uses 530 flops */
1914         }
1915
1916         if(jidx<j_index_end)
1917         {
1918
1919             /* Get j neighbor index, and coordinate index */
1920             jnrlistA         = jjnr[jidx];
1921             jnrlistB         = jjnr[jidx+1];
1922             jnrlistC         = jjnr[jidx+2];
1923             jnrlistD         = jjnr[jidx+3];
1924             jnrlistE         = jjnr[jidx+4];
1925             jnrlistF         = jjnr[jidx+5];
1926             jnrlistG         = jjnr[jidx+6];
1927             jnrlistH         = jjnr[jidx+7];
1928             /* Sign of each element will be negative for non-real atoms.
1929              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1930              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1931              */
1932             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1933                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1934
1935             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1936             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1937             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1938             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1939             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1940             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1941             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1942             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1943             j_coord_offsetA  = DIM*jnrA;
1944             j_coord_offsetB  = DIM*jnrB;
1945             j_coord_offsetC  = DIM*jnrC;
1946             j_coord_offsetD  = DIM*jnrD;
1947             j_coord_offsetE  = DIM*jnrE;
1948             j_coord_offsetF  = DIM*jnrF;
1949             j_coord_offsetG  = DIM*jnrG;
1950             j_coord_offsetH  = DIM*jnrH;
1951
1952             /* load j atom coordinates */
1953             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1954                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1955                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1956                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1957                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1958
1959             /* Calculate displacement vector */
1960             dx00             = _mm256_sub_ps(ix0,jx0);
1961             dy00             = _mm256_sub_ps(iy0,jy0);
1962             dz00             = _mm256_sub_ps(iz0,jz0);
1963             dx01             = _mm256_sub_ps(ix0,jx1);
1964             dy01             = _mm256_sub_ps(iy0,jy1);
1965             dz01             = _mm256_sub_ps(iz0,jz1);
1966             dx02             = _mm256_sub_ps(ix0,jx2);
1967             dy02             = _mm256_sub_ps(iy0,jy2);
1968             dz02             = _mm256_sub_ps(iz0,jz2);
1969             dx10             = _mm256_sub_ps(ix1,jx0);
1970             dy10             = _mm256_sub_ps(iy1,jy0);
1971             dz10             = _mm256_sub_ps(iz1,jz0);
1972             dx11             = _mm256_sub_ps(ix1,jx1);
1973             dy11             = _mm256_sub_ps(iy1,jy1);
1974             dz11             = _mm256_sub_ps(iz1,jz1);
1975             dx12             = _mm256_sub_ps(ix1,jx2);
1976             dy12             = _mm256_sub_ps(iy1,jy2);
1977             dz12             = _mm256_sub_ps(iz1,jz2);
1978             dx20             = _mm256_sub_ps(ix2,jx0);
1979             dy20             = _mm256_sub_ps(iy2,jy0);
1980             dz20             = _mm256_sub_ps(iz2,jz0);
1981             dx21             = _mm256_sub_ps(ix2,jx1);
1982             dy21             = _mm256_sub_ps(iy2,jy1);
1983             dz21             = _mm256_sub_ps(iz2,jz1);
1984             dx22             = _mm256_sub_ps(ix2,jx2);
1985             dy22             = _mm256_sub_ps(iy2,jy2);
1986             dz22             = _mm256_sub_ps(iz2,jz2);
1987
1988             /* Calculate squared distance and things based on it */
1989             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1990             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1991             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1992             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1993             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1994             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1995             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1996             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1997             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1998
1999             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
2000             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
2001             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
2002             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
2003             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2004             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2005             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
2006             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2007             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2008
2009             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
2010             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
2011             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
2012             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
2013             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2014             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2015             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
2016             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2017             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2018
2019             fjx0             = _mm256_setzero_ps();
2020             fjy0             = _mm256_setzero_ps();
2021             fjz0             = _mm256_setzero_ps();
2022             fjx1             = _mm256_setzero_ps();
2023             fjy1             = _mm256_setzero_ps();
2024             fjz1             = _mm256_setzero_ps();
2025             fjx2             = _mm256_setzero_ps();
2026             fjy2             = _mm256_setzero_ps();
2027             fjz2             = _mm256_setzero_ps();
2028
2029             /**************************
2030              * CALCULATE INTERACTIONS *
2031              **************************/
2032
2033             r00              = _mm256_mul_ps(rsq00,rinv00);
2034             r00              = _mm256_andnot_ps(dummy_mask,r00);
2035
2036             /* Calculate table index by multiplying r with table scale and truncate to integer */
2037             rt               = _mm256_mul_ps(r00,vftabscale);
2038             vfitab           = _mm256_cvttps_epi32(rt);
2039             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2040             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2041             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2042             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2043             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
2044             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
2045
2046             /* EWALD ELECTROSTATICS */
2047
2048             /* Analytical PME correction */
2049             zeta2            = _mm256_mul_ps(beta2,rsq00);
2050             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
2051             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2052             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2053             felec            = _mm256_mul_ps(qq00,felec);
2054
2055             /* CUBIC SPLINE TABLE DISPERSION */
2056             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2057                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2058             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2059                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2060             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2061                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2062             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2063                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2064             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2065             Heps             = _mm256_mul_ps(vfeps,H);
2066             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2067             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2068             fvdw6            = _mm256_mul_ps(c6_00,FF);
2069
2070             /* CUBIC SPLINE TABLE REPULSION */
2071             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
2072             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
2073             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2074                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2075             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2076                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2077             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2078                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2079             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2080                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2081             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2082             Heps             = _mm256_mul_ps(vfeps,H);
2083             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2084             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2085             fvdw12           = _mm256_mul_ps(c12_00,FF);
2086             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
2087
2088             fscal            = _mm256_add_ps(felec,fvdw);
2089
2090             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2091
2092             /* Calculate temporary vectorial force */
2093             tx               = _mm256_mul_ps(fscal,dx00);
2094             ty               = _mm256_mul_ps(fscal,dy00);
2095             tz               = _mm256_mul_ps(fscal,dz00);
2096
2097             /* Update vectorial force */
2098             fix0             = _mm256_add_ps(fix0,tx);
2099             fiy0             = _mm256_add_ps(fiy0,ty);
2100             fiz0             = _mm256_add_ps(fiz0,tz);
2101
2102             fjx0             = _mm256_add_ps(fjx0,tx);
2103             fjy0             = _mm256_add_ps(fjy0,ty);
2104             fjz0             = _mm256_add_ps(fjz0,tz);
2105
2106             /**************************
2107              * CALCULATE INTERACTIONS *
2108              **************************/
2109
2110             r01              = _mm256_mul_ps(rsq01,rinv01);
2111             r01              = _mm256_andnot_ps(dummy_mask,r01);
2112
2113             /* EWALD ELECTROSTATICS */
2114
2115             /* Analytical PME correction */
2116             zeta2            = _mm256_mul_ps(beta2,rsq01);
2117             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
2118             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2119             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2120             felec            = _mm256_mul_ps(qq01,felec);
2121
2122             fscal            = felec;
2123
2124             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2125
2126             /* Calculate temporary vectorial force */
2127             tx               = _mm256_mul_ps(fscal,dx01);
2128             ty               = _mm256_mul_ps(fscal,dy01);
2129             tz               = _mm256_mul_ps(fscal,dz01);
2130
2131             /* Update vectorial force */
2132             fix0             = _mm256_add_ps(fix0,tx);
2133             fiy0             = _mm256_add_ps(fiy0,ty);
2134             fiz0             = _mm256_add_ps(fiz0,tz);
2135
2136             fjx1             = _mm256_add_ps(fjx1,tx);
2137             fjy1             = _mm256_add_ps(fjy1,ty);
2138             fjz1             = _mm256_add_ps(fjz1,tz);
2139
2140             /**************************
2141              * CALCULATE INTERACTIONS *
2142              **************************/
2143
2144             r02              = _mm256_mul_ps(rsq02,rinv02);
2145             r02              = _mm256_andnot_ps(dummy_mask,r02);
2146
2147             /* EWALD ELECTROSTATICS */
2148
2149             /* Analytical PME correction */
2150             zeta2            = _mm256_mul_ps(beta2,rsq02);
2151             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
2152             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2153             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2154             felec            = _mm256_mul_ps(qq02,felec);
2155
2156             fscal            = felec;
2157
2158             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2159
2160             /* Calculate temporary vectorial force */
2161             tx               = _mm256_mul_ps(fscal,dx02);
2162             ty               = _mm256_mul_ps(fscal,dy02);
2163             tz               = _mm256_mul_ps(fscal,dz02);
2164
2165             /* Update vectorial force */
2166             fix0             = _mm256_add_ps(fix0,tx);
2167             fiy0             = _mm256_add_ps(fiy0,ty);
2168             fiz0             = _mm256_add_ps(fiz0,tz);
2169
2170             fjx2             = _mm256_add_ps(fjx2,tx);
2171             fjy2             = _mm256_add_ps(fjy2,ty);
2172             fjz2             = _mm256_add_ps(fjz2,tz);
2173
2174             /**************************
2175              * CALCULATE INTERACTIONS *
2176              **************************/
2177
2178             r10              = _mm256_mul_ps(rsq10,rinv10);
2179             r10              = _mm256_andnot_ps(dummy_mask,r10);
2180
2181             /* EWALD ELECTROSTATICS */
2182
2183             /* Analytical PME correction */
2184             zeta2            = _mm256_mul_ps(beta2,rsq10);
2185             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
2186             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2187             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2188             felec            = _mm256_mul_ps(qq10,felec);
2189
2190             fscal            = felec;
2191
2192             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2193
2194             /* Calculate temporary vectorial force */
2195             tx               = _mm256_mul_ps(fscal,dx10);
2196             ty               = _mm256_mul_ps(fscal,dy10);
2197             tz               = _mm256_mul_ps(fscal,dz10);
2198
2199             /* Update vectorial force */
2200             fix1             = _mm256_add_ps(fix1,tx);
2201             fiy1             = _mm256_add_ps(fiy1,ty);
2202             fiz1             = _mm256_add_ps(fiz1,tz);
2203
2204             fjx0             = _mm256_add_ps(fjx0,tx);
2205             fjy0             = _mm256_add_ps(fjy0,ty);
2206             fjz0             = _mm256_add_ps(fjz0,tz);
2207
2208             /**************************
2209              * CALCULATE INTERACTIONS *
2210              **************************/
2211
2212             r11              = _mm256_mul_ps(rsq11,rinv11);
2213             r11              = _mm256_andnot_ps(dummy_mask,r11);
2214
2215             /* EWALD ELECTROSTATICS */
2216
2217             /* Analytical PME correction */
2218             zeta2            = _mm256_mul_ps(beta2,rsq11);
2219             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
2220             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2221             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2222             felec            = _mm256_mul_ps(qq11,felec);
2223
2224             fscal            = felec;
2225
2226             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2227
2228             /* Calculate temporary vectorial force */
2229             tx               = _mm256_mul_ps(fscal,dx11);
2230             ty               = _mm256_mul_ps(fscal,dy11);
2231             tz               = _mm256_mul_ps(fscal,dz11);
2232
2233             /* Update vectorial force */
2234             fix1             = _mm256_add_ps(fix1,tx);
2235             fiy1             = _mm256_add_ps(fiy1,ty);
2236             fiz1             = _mm256_add_ps(fiz1,tz);
2237
2238             fjx1             = _mm256_add_ps(fjx1,tx);
2239             fjy1             = _mm256_add_ps(fjy1,ty);
2240             fjz1             = _mm256_add_ps(fjz1,tz);
2241
2242             /**************************
2243              * CALCULATE INTERACTIONS *
2244              **************************/
2245
2246             r12              = _mm256_mul_ps(rsq12,rinv12);
2247             r12              = _mm256_andnot_ps(dummy_mask,r12);
2248
2249             /* EWALD ELECTROSTATICS */
2250
2251             /* Analytical PME correction */
2252             zeta2            = _mm256_mul_ps(beta2,rsq12);
2253             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2254             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2255             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2256             felec            = _mm256_mul_ps(qq12,felec);
2257
2258             fscal            = felec;
2259
2260             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2261
2262             /* Calculate temporary vectorial force */
2263             tx               = _mm256_mul_ps(fscal,dx12);
2264             ty               = _mm256_mul_ps(fscal,dy12);
2265             tz               = _mm256_mul_ps(fscal,dz12);
2266
2267             /* Update vectorial force */
2268             fix1             = _mm256_add_ps(fix1,tx);
2269             fiy1             = _mm256_add_ps(fiy1,ty);
2270             fiz1             = _mm256_add_ps(fiz1,tz);
2271
2272             fjx2             = _mm256_add_ps(fjx2,tx);
2273             fjy2             = _mm256_add_ps(fjy2,ty);
2274             fjz2             = _mm256_add_ps(fjz2,tz);
2275
2276             /**************************
2277              * CALCULATE INTERACTIONS *
2278              **************************/
2279
2280             r20              = _mm256_mul_ps(rsq20,rinv20);
2281             r20              = _mm256_andnot_ps(dummy_mask,r20);
2282
2283             /* EWALD ELECTROSTATICS */
2284
2285             /* Analytical PME correction */
2286             zeta2            = _mm256_mul_ps(beta2,rsq20);
2287             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
2288             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2289             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2290             felec            = _mm256_mul_ps(qq20,felec);
2291
2292             fscal            = felec;
2293
2294             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2295
2296             /* Calculate temporary vectorial force */
2297             tx               = _mm256_mul_ps(fscal,dx20);
2298             ty               = _mm256_mul_ps(fscal,dy20);
2299             tz               = _mm256_mul_ps(fscal,dz20);
2300
2301             /* Update vectorial force */
2302             fix2             = _mm256_add_ps(fix2,tx);
2303             fiy2             = _mm256_add_ps(fiy2,ty);
2304             fiz2             = _mm256_add_ps(fiz2,tz);
2305
2306             fjx0             = _mm256_add_ps(fjx0,tx);
2307             fjy0             = _mm256_add_ps(fjy0,ty);
2308             fjz0             = _mm256_add_ps(fjz0,tz);
2309
2310             /**************************
2311              * CALCULATE INTERACTIONS *
2312              **************************/
2313
2314             r21              = _mm256_mul_ps(rsq21,rinv21);
2315             r21              = _mm256_andnot_ps(dummy_mask,r21);
2316
2317             /* EWALD ELECTROSTATICS */
2318
2319             /* Analytical PME correction */
2320             zeta2            = _mm256_mul_ps(beta2,rsq21);
2321             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2322             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2323             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2324             felec            = _mm256_mul_ps(qq21,felec);
2325
2326             fscal            = felec;
2327
2328             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2329
2330             /* Calculate temporary vectorial force */
2331             tx               = _mm256_mul_ps(fscal,dx21);
2332             ty               = _mm256_mul_ps(fscal,dy21);
2333             tz               = _mm256_mul_ps(fscal,dz21);
2334
2335             /* Update vectorial force */
2336             fix2             = _mm256_add_ps(fix2,tx);
2337             fiy2             = _mm256_add_ps(fiy2,ty);
2338             fiz2             = _mm256_add_ps(fiz2,tz);
2339
2340             fjx1             = _mm256_add_ps(fjx1,tx);
2341             fjy1             = _mm256_add_ps(fjy1,ty);
2342             fjz1             = _mm256_add_ps(fjz1,tz);
2343
2344             /**************************
2345              * CALCULATE INTERACTIONS *
2346              **************************/
2347
2348             r22              = _mm256_mul_ps(rsq22,rinv22);
2349             r22              = _mm256_andnot_ps(dummy_mask,r22);
2350
2351             /* EWALD ELECTROSTATICS */
2352
2353             /* Analytical PME correction */
2354             zeta2            = _mm256_mul_ps(beta2,rsq22);
2355             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2356             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2357             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2358             felec            = _mm256_mul_ps(qq22,felec);
2359
2360             fscal            = felec;
2361
2362             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2363
2364             /* Calculate temporary vectorial force */
2365             tx               = _mm256_mul_ps(fscal,dx22);
2366             ty               = _mm256_mul_ps(fscal,dy22);
2367             tz               = _mm256_mul_ps(fscal,dz22);
2368
2369             /* Update vectorial force */
2370             fix2             = _mm256_add_ps(fix2,tx);
2371             fiy2             = _mm256_add_ps(fiy2,ty);
2372             fiz2             = _mm256_add_ps(fiz2,tz);
2373
2374             fjx2             = _mm256_add_ps(fjx2,tx);
2375             fjy2             = _mm256_add_ps(fjy2,ty);
2376             fjz2             = _mm256_add_ps(fjz2,tz);
2377
2378             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2379             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2380             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2381             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2382             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2383             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2384             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2385             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2386
2387             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2388                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2389
2390             /* Inner loop uses 539 flops */
2391         }
2392
2393         /* End of innermost loop */
2394
2395         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2396                                                  f+i_coord_offset,fshift+i_shift_offset);
2397
2398         /* Increment number of inner iterations */
2399         inneriter                  += j_index_end - j_index_start;
2400
2401         /* Outer loop uses 18 flops */
2402     }
2403
2404     /* Increment number of outer iterations */
2405     outeriter        += nri;
2406
2407     /* Update outer/inner flops */
2408
2409     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*539);
2410 }