src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_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_ElecRF_VdwCSTab_GeomW3W3_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Water3
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_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     __m256           dummy_mask,cutoff_mask;
 107     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 108     __m256           one     = _mm256_set1_ps(1.0);
 109     __m256           two     = _mm256_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm256_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm256_set1_ps(fr->ic->k_rf);
 124     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 125     crf              = _mm256_set1_ps(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 136     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 137     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 138     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 139
 140     jq0              = _mm256_set1_ps(charge[inr+0]);
 141     jq1              = _mm256_set1_ps(charge[inr+1]);
 142     jq2              = _mm256_set1_ps(charge[inr+2]);
 143     vdwjidx0A        = 2*vdwtype[inr+0];
 144     qq00             = _mm256_mul_ps(iq0,jq0);
 145     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 146     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 147     qq01             = _mm256_mul_ps(iq0,jq1);
 148     qq02             = _mm256_mul_ps(iq0,jq2);
 149     qq10             = _mm256_mul_ps(iq1,jq0);
 150     qq11             = _mm256_mul_ps(iq1,jq1);
 151     qq12             = _mm256_mul_ps(iq1,jq2);
 152     qq20             = _mm256_mul_ps(iq2,jq0);
 153     qq21             = _mm256_mul_ps(iq2,jq1);
 154     qq22             = _mm256_mul_ps(iq2,jq2);
 155
 156     /* Avoid stupid compiler warnings */
 157     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 158     j_coord_offsetA = 0;
 159     j_coord_offsetB = 0;
 160     j_coord_offsetC = 0;
 161     j_coord_offsetD = 0;
 162     j_coord_offsetE = 0;
 163     j_coord_offsetF = 0;
 164     j_coord_offsetG = 0;
 165     j_coord_offsetH = 0;
 166
 167     outeriter        = 0;
 168     inneriter        = 0;
 169
 170     for(iidx=0;iidx<4*DIM;iidx++)
 171     {
 172         scratch[iidx] = 0.0;
 173     }
 174
 175     /* Start outer loop over neighborlists */
 176     for(iidx=0; iidx<nri; iidx++)
 177     {
 178         /* Load shift vector for this list */
 179         i_shift_offset   = DIM*shiftidx[iidx];
 180
 181         /* Load limits for loop over neighbors */
 182         j_index_start    = jindex[iidx];
 183         j_index_end      = jindex[iidx+1];
 184
 185         /* Get outer coordinate index */
 186         inr              = iinr[iidx];
 187         i_coord_offset   = DIM*inr;
 188
 189         /* Load i particle coords and add shift vector */
 190         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 191                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 192
 193         fix0             = _mm256_setzero_ps();
 194         fiy0             = _mm256_setzero_ps();
 195         fiz0             = _mm256_setzero_ps();
 196         fix1             = _mm256_setzero_ps();
 197         fiy1             = _mm256_setzero_ps();
 198         fiz1             = _mm256_setzero_ps();
 199         fix2             = _mm256_setzero_ps();
 200         fiy2             = _mm256_setzero_ps();
 201         fiz2             = _mm256_setzero_ps();
 202
 203         /* Reset potential sums */
 204         velecsum         = _mm256_setzero_ps();
 205         vvdwsum          = _mm256_setzero_ps();
 206
 207         /* Start inner kernel loop */
 208         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 209         {
 210
 211             /* Get j neighbor index, and coordinate index */
 212             jnrA             = jjnr[jidx];
 213             jnrB             = jjnr[jidx+1];
 214             jnrC             = jjnr[jidx+2];
 215             jnrD             = jjnr[jidx+3];
 216             jnrE             = jjnr[jidx+4];
 217             jnrF             = jjnr[jidx+5];
 218             jnrG             = jjnr[jidx+6];
 219             jnrH             = jjnr[jidx+7];
 220             j_coord_offsetA  = DIM*jnrA;
 221             j_coord_offsetB  = DIM*jnrB;
 222             j_coord_offsetC  = DIM*jnrC;
 223             j_coord_offsetD  = DIM*jnrD;
 224             j_coord_offsetE  = DIM*jnrE;
 225             j_coord_offsetF  = DIM*jnrF;
 226             j_coord_offsetG  = DIM*jnrG;
 227             j_coord_offsetH  = DIM*jnrH;
 228
 229             /* load j atom coordinates */
 230             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 231                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 232                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 233                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 234                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 235
 236             /* Calculate displacement vector */
 237             dx00             = _mm256_sub_ps(ix0,jx0);
 238             dy00             = _mm256_sub_ps(iy0,jy0);
 239             dz00             = _mm256_sub_ps(iz0,jz0);
 240             dx01             = _mm256_sub_ps(ix0,jx1);
 241             dy01             = _mm256_sub_ps(iy0,jy1);
 242             dz01             = _mm256_sub_ps(iz0,jz1);
 243             dx02             = _mm256_sub_ps(ix0,jx2);
 244             dy02             = _mm256_sub_ps(iy0,jy2);
 245             dz02             = _mm256_sub_ps(iz0,jz2);
 246             dx10             = _mm256_sub_ps(ix1,jx0);
 247             dy10             = _mm256_sub_ps(iy1,jy0);
 248             dz10             = _mm256_sub_ps(iz1,jz0);
 249             dx11             = _mm256_sub_ps(ix1,jx1);
 250             dy11             = _mm256_sub_ps(iy1,jy1);
 251             dz11             = _mm256_sub_ps(iz1,jz1);
 252             dx12             = _mm256_sub_ps(ix1,jx2);
 253             dy12             = _mm256_sub_ps(iy1,jy2);
 254             dz12             = _mm256_sub_ps(iz1,jz2);
 255             dx20             = _mm256_sub_ps(ix2,jx0);
 256             dy20             = _mm256_sub_ps(iy2,jy0);
 257             dz20             = _mm256_sub_ps(iz2,jz0);
 258             dx21             = _mm256_sub_ps(ix2,jx1);
 259             dy21             = _mm256_sub_ps(iy2,jy1);
 260             dz21             = _mm256_sub_ps(iz2,jz1);
 261             dx22             = _mm256_sub_ps(ix2,jx2);
 262             dy22             = _mm256_sub_ps(iy2,jy2);
 263             dz22             = _mm256_sub_ps(iz2,jz2);
 264
 265             /* Calculate squared distance and things based on it */
 266             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 267             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 268             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 269             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 270             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 271             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 272             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 273             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 274             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 275
 276             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 277             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 278             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 279             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 280             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 281             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 282             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 283             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 284             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 285
 286             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 287             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 288             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 289             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 290             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 291             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 292             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 293             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 294             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 295
 296             fjx0             = _mm256_setzero_ps();
 297             fjy0             = _mm256_setzero_ps();
 298             fjz0             = _mm256_setzero_ps();
 299             fjx1             = _mm256_setzero_ps();
 300             fjy1             = _mm256_setzero_ps();
 301             fjz1             = _mm256_setzero_ps();
 302             fjx2             = _mm256_setzero_ps();
 303             fjy2             = _mm256_setzero_ps();
 304             fjz2             = _mm256_setzero_ps();
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             r00              = _mm256_mul_ps(rsq00,rinv00);
 311
 312             /* Calculate table index by multiplying r with table scale and truncate to integer */
 313             rt               = _mm256_mul_ps(r00,vftabscale);
 314             vfitab           = _mm256_cvttps_epi32(rt);
 315             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 316             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 317             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 318             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 319             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 320             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 321
 322             /* REACTION-FIELD ELECTROSTATICS */
 323             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 324             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 325
 326             /* CUBIC SPLINE TABLE DISPERSION */
 327             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 328                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 329             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 330                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 331             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 332                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 333             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 334                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 335             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 336             Heps             = _mm256_mul_ps(vfeps,H);
 337             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 338             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 339             vvdw6            = _mm256_mul_ps(c6_00,VV);
 340             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 341             fvdw6            = _mm256_mul_ps(c6_00,FF);
 342
 343             /* CUBIC SPLINE TABLE REPULSION */
 344             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 345             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 346             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 347                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 348             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 349                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 350             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 351                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 352             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 353                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 354             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 355             Heps             = _mm256_mul_ps(vfeps,H);
 356             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 357             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 358             vvdw12           = _mm256_mul_ps(c12_00,VV);
 359             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 360             fvdw12           = _mm256_mul_ps(c12_00,FF);
 361             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 362             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velecsum         = _mm256_add_ps(velecsum,velec);
 366             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 367
 368             fscal            = _mm256_add_ps(felec,fvdw);
 369
 370             /* Calculate temporary vectorial force */
 371             tx               = _mm256_mul_ps(fscal,dx00);
 372             ty               = _mm256_mul_ps(fscal,dy00);
 373             tz               = _mm256_mul_ps(fscal,dz00);
 374
 375             /* Update vectorial force */
 376             fix0             = _mm256_add_ps(fix0,tx);
 377             fiy0             = _mm256_add_ps(fiy0,ty);
 378             fiz0             = _mm256_add_ps(fiz0,tz);
 379
 380             fjx0             = _mm256_add_ps(fjx0,tx);
 381             fjy0             = _mm256_add_ps(fjy0,ty);
 382             fjz0             = _mm256_add_ps(fjz0,tz);
 383
 384             /**************************
 385              * CALCULATE INTERACTIONS *
 386              **************************/
 387
 388             /* REACTION-FIELD ELECTROSTATICS */
 389             velec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
 390             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
 391
 392             /* Update potential sum for this i atom from the interaction with this j atom. */
 393             velecsum         = _mm256_add_ps(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397             /* Calculate temporary vectorial force */
 398             tx               = _mm256_mul_ps(fscal,dx01);
 399             ty               = _mm256_mul_ps(fscal,dy01);
 400             tz               = _mm256_mul_ps(fscal,dz01);
 401
 402             /* Update vectorial force */
 403             fix0             = _mm256_add_ps(fix0,tx);
 404             fiy0             = _mm256_add_ps(fiy0,ty);
 405             fiz0             = _mm256_add_ps(fiz0,tz);
 406
 407             fjx1             = _mm256_add_ps(fjx1,tx);
 408             fjy1             = _mm256_add_ps(fjy1,ty);
 409             fjz1             = _mm256_add_ps(fjz1,tz);
 410
 411             /**************************
 412              * CALCULATE INTERACTIONS *
 413              **************************/
 414
 415             /* REACTION-FIELD ELECTROSTATICS */
 416             velec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
 417             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
 418
 419             /* Update potential sum for this i atom from the interaction with this j atom. */
 420             velecsum         = _mm256_add_ps(velecsum,velec);
 421
 422             fscal            = felec;
 423
 424             /* Calculate temporary vectorial force */
 425             tx               = _mm256_mul_ps(fscal,dx02);
 426             ty               = _mm256_mul_ps(fscal,dy02);
 427             tz               = _mm256_mul_ps(fscal,dz02);
 428
 429             /* Update vectorial force */
 430             fix0             = _mm256_add_ps(fix0,tx);
 431             fiy0             = _mm256_add_ps(fiy0,ty);
 432             fiz0             = _mm256_add_ps(fiz0,tz);
 433
 434             fjx2             = _mm256_add_ps(fjx2,tx);
 435             fjy2             = _mm256_add_ps(fjy2,ty);
 436             fjz2             = _mm256_add_ps(fjz2,tz);
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             /* REACTION-FIELD ELECTROSTATICS */
 443             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 444             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 445
 446             /* Update potential sum for this i atom from the interaction with this j atom. */
 447             velecsum         = _mm256_add_ps(velecsum,velec);
 448
 449             fscal            = felec;
 450
 451             /* Calculate temporary vectorial force */
 452             tx               = _mm256_mul_ps(fscal,dx10);
 453             ty               = _mm256_mul_ps(fscal,dy10);
 454             tz               = _mm256_mul_ps(fscal,dz10);
 455
 456             /* Update vectorial force */
 457             fix1             = _mm256_add_ps(fix1,tx);
 458             fiy1             = _mm256_add_ps(fiy1,ty);
 459             fiz1             = _mm256_add_ps(fiz1,tz);
 460
 461             fjx0             = _mm256_add_ps(fjx0,tx);
 462             fjy0             = _mm256_add_ps(fjy0,ty);
 463             fjz0             = _mm256_add_ps(fjz0,tz);
 464
 465             /**************************
 466              * CALCULATE INTERACTIONS *
 467              **************************/
 468
 469             /* REACTION-FIELD ELECTROSTATICS */
 470             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 471             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 472
 473             /* Update potential sum for this i atom from the interaction with this j atom. */
 474             velecsum         = _mm256_add_ps(velecsum,velec);
 475
 476             fscal            = felec;
 477
 478             /* Calculate temporary vectorial force */
 479             tx               = _mm256_mul_ps(fscal,dx11);
 480             ty               = _mm256_mul_ps(fscal,dy11);
 481             tz               = _mm256_mul_ps(fscal,dz11);
 482
 483             /* Update vectorial force */
 484             fix1             = _mm256_add_ps(fix1,tx);
 485             fiy1             = _mm256_add_ps(fiy1,ty);
 486             fiz1             = _mm256_add_ps(fiz1,tz);
 487
 488             fjx1             = _mm256_add_ps(fjx1,tx);
 489             fjy1             = _mm256_add_ps(fjy1,ty);
 490             fjz1             = _mm256_add_ps(fjz1,tz);
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             /* REACTION-FIELD ELECTROSTATICS */
 497             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 498             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 499
 500             /* Update potential sum for this i atom from the interaction with this j atom. */
 501             velecsum         = _mm256_add_ps(velecsum,velec);
 502
 503             fscal            = felec;
 504
 505             /* Calculate temporary vectorial force */
 506             tx               = _mm256_mul_ps(fscal,dx12);
 507             ty               = _mm256_mul_ps(fscal,dy12);
 508             tz               = _mm256_mul_ps(fscal,dz12);
 509
 510             /* Update vectorial force */
 511             fix1             = _mm256_add_ps(fix1,tx);
 512             fiy1             = _mm256_add_ps(fiy1,ty);
 513             fiz1             = _mm256_add_ps(fiz1,tz);
 514
 515             fjx2             = _mm256_add_ps(fjx2,tx);
 516             fjy2             = _mm256_add_ps(fjy2,ty);
 517             fjz2             = _mm256_add_ps(fjz2,tz);
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             /* REACTION-FIELD ELECTROSTATICS */
 524             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 525             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 526
 527             /* Update potential sum for this i atom from the interaction with this j atom. */
 528             velecsum         = _mm256_add_ps(velecsum,velec);
 529
 530             fscal            = felec;
 531
 532             /* Calculate temporary vectorial force */
 533             tx               = _mm256_mul_ps(fscal,dx20);
 534             ty               = _mm256_mul_ps(fscal,dy20);
 535             tz               = _mm256_mul_ps(fscal,dz20);
 536
 537             /* Update vectorial force */
 538             fix2             = _mm256_add_ps(fix2,tx);
 539             fiy2             = _mm256_add_ps(fiy2,ty);
 540             fiz2             = _mm256_add_ps(fiz2,tz);
 541
 542             fjx0             = _mm256_add_ps(fjx0,tx);
 543             fjy0             = _mm256_add_ps(fjy0,ty);
 544             fjz0             = _mm256_add_ps(fjz0,tz);
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             /* REACTION-FIELD ELECTROSTATICS */
 551             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
 552             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velecsum         = _mm256_add_ps(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm256_mul_ps(fscal,dx21);
 561             ty               = _mm256_mul_ps(fscal,dy21);
 562             tz               = _mm256_mul_ps(fscal,dz21);
 563
 564             /* Update vectorial force */
 565             fix2             = _mm256_add_ps(fix2,tx);
 566             fiy2             = _mm256_add_ps(fiy2,ty);
 567             fiz2             = _mm256_add_ps(fiz2,tz);
 568
 569             fjx1             = _mm256_add_ps(fjx1,tx);
 570             fjy1             = _mm256_add_ps(fjy1,ty);
 571             fjz1             = _mm256_add_ps(fjz1,tz);
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             /* REACTION-FIELD ELECTROSTATICS */
 578             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
 579             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velecsum         = _mm256_add_ps(velecsum,velec);
 583
 584             fscal            = felec;
 585
 586             /* Calculate temporary vectorial force */
 587             tx               = _mm256_mul_ps(fscal,dx22);
 588             ty               = _mm256_mul_ps(fscal,dy22);
 589             tz               = _mm256_mul_ps(fscal,dz22);
 590
 591             /* Update vectorial force */
 592             fix2             = _mm256_add_ps(fix2,tx);
 593             fiy2             = _mm256_add_ps(fiy2,ty);
 594             fiz2             = _mm256_add_ps(fiz2,tz);
 595
 596             fjx2             = _mm256_add_ps(fjx2,tx);
 597             fjy2             = _mm256_add_ps(fjy2,ty);
 598             fjz2             = _mm256_add_ps(fjz2,tz);
 599
 600             fjptrA             = f+j_coord_offsetA;
 601             fjptrB             = f+j_coord_offsetB;
 602             fjptrC             = f+j_coord_offsetC;
 603             fjptrD             = f+j_coord_offsetD;
 604             fjptrE             = f+j_coord_offsetE;
 605             fjptrF             = f+j_coord_offsetF;
 606             fjptrG             = f+j_coord_offsetG;
 607             fjptrH             = f+j_coord_offsetH;
 608
 609             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 610                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 611
 612             /* Inner loop uses 323 flops */
 613         }
 614
 615         if(jidx<j_index_end)
 616         {
 617
 618             /* Get j neighbor index, and coordinate index */
 619             jnrlistA         = jjnr[jidx];
 620             jnrlistB         = jjnr[jidx+1];
 621             jnrlistC         = jjnr[jidx+2];
 622             jnrlistD         = jjnr[jidx+3];
 623             jnrlistE         = jjnr[jidx+4];
 624             jnrlistF         = jjnr[jidx+5];
 625             jnrlistG         = jjnr[jidx+6];
 626             jnrlistH         = jjnr[jidx+7];
 627             /* Sign of each element will be negative for non-real atoms.
 628              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 629              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 630              */
 631             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 632                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 633
 634             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 635             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 636             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 637             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 638             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 639             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 640             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 641             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 642             j_coord_offsetA  = DIM*jnrA;
 643             j_coord_offsetB  = DIM*jnrB;
 644             j_coord_offsetC  = DIM*jnrC;
 645             j_coord_offsetD  = DIM*jnrD;
 646             j_coord_offsetE  = DIM*jnrE;
 647             j_coord_offsetF  = DIM*jnrF;
 648             j_coord_offsetG  = DIM*jnrG;
 649             j_coord_offsetH  = DIM*jnrH;
 650
 651             /* load j atom coordinates */
 652             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 653                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 654                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 655                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 656                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 657
 658             /* Calculate displacement vector */
 659             dx00             = _mm256_sub_ps(ix0,jx0);
 660             dy00             = _mm256_sub_ps(iy0,jy0);
 661             dz00             = _mm256_sub_ps(iz0,jz0);
 662             dx01             = _mm256_sub_ps(ix0,jx1);
 663             dy01             = _mm256_sub_ps(iy0,jy1);
 664             dz01             = _mm256_sub_ps(iz0,jz1);
 665             dx02             = _mm256_sub_ps(ix0,jx2);
 666             dy02             = _mm256_sub_ps(iy0,jy2);
 667             dz02             = _mm256_sub_ps(iz0,jz2);
 668             dx10             = _mm256_sub_ps(ix1,jx0);
 669             dy10             = _mm256_sub_ps(iy1,jy0);
 670             dz10             = _mm256_sub_ps(iz1,jz0);
 671             dx11             = _mm256_sub_ps(ix1,jx1);
 672             dy11             = _mm256_sub_ps(iy1,jy1);
 673             dz11             = _mm256_sub_ps(iz1,jz1);
 674             dx12             = _mm256_sub_ps(ix1,jx2);
 675             dy12             = _mm256_sub_ps(iy1,jy2);
 676             dz12             = _mm256_sub_ps(iz1,jz2);
 677             dx20             = _mm256_sub_ps(ix2,jx0);
 678             dy20             = _mm256_sub_ps(iy2,jy0);
 679             dz20             = _mm256_sub_ps(iz2,jz0);
 680             dx21             = _mm256_sub_ps(ix2,jx1);
 681             dy21             = _mm256_sub_ps(iy2,jy1);
 682             dz21             = _mm256_sub_ps(iz2,jz1);
 683             dx22             = _mm256_sub_ps(ix2,jx2);
 684             dy22             = _mm256_sub_ps(iy2,jy2);
 685             dz22             = _mm256_sub_ps(iz2,jz2);
 686
 687             /* Calculate squared distance and things based on it */
 688             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 689             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 690             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 691             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 692             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 693             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 694             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 695             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 696             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 697
 698             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 699             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 700             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 701             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 702             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 703             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 704             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 705             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 706             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 707
 708             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 709             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 710             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 711             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 712             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 713             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 714             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 715             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 716             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 717
 718             fjx0             = _mm256_setzero_ps();
 719             fjy0             = _mm256_setzero_ps();
 720             fjz0             = _mm256_setzero_ps();
 721             fjx1             = _mm256_setzero_ps();
 722             fjy1             = _mm256_setzero_ps();
 723             fjz1             = _mm256_setzero_ps();
 724             fjx2             = _mm256_setzero_ps();
 725             fjy2             = _mm256_setzero_ps();
 726             fjz2             = _mm256_setzero_ps();
 727
 728             /**************************
 729              * CALCULATE INTERACTIONS *
 730              **************************/
 731
 732             r00              = _mm256_mul_ps(rsq00,rinv00);
 733             r00              = _mm256_andnot_ps(dummy_mask,r00);
 734
 735             /* Calculate table index by multiplying r with table scale and truncate to integer */
 736             rt               = _mm256_mul_ps(r00,vftabscale);
 737             vfitab           = _mm256_cvttps_epi32(rt);
 738             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 739             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 740             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 741             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 742             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 743             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 744
 745             /* REACTION-FIELD ELECTROSTATICS */
 746             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 747             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 748
 749             /* CUBIC SPLINE TABLE DISPERSION */
 750             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 751                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 752             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 753                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 754             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 755                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 756             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 757                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 758             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 759             Heps             = _mm256_mul_ps(vfeps,H);
 760             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 761             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 762             vvdw6            = _mm256_mul_ps(c6_00,VV);
 763             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 764             fvdw6            = _mm256_mul_ps(c6_00,FF);
 765
 766             /* CUBIC SPLINE TABLE REPULSION */
 767             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 768             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 769             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 770                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 771             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 772                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 773             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 774                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 775             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 776                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 777             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 778             Heps             = _mm256_mul_ps(vfeps,H);
 779             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 780             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 781             vvdw12           = _mm256_mul_ps(c12_00,VV);
 782             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 783             fvdw12           = _mm256_mul_ps(c12_00,FF);
 784             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 785             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 786
 787             /* Update potential sum for this i atom from the interaction with this j atom. */
 788             velec            = _mm256_andnot_ps(dummy_mask,velec);
 789             velecsum         = _mm256_add_ps(velecsum,velec);
 790             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 791             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 792
 793             fscal            = _mm256_add_ps(felec,fvdw);
 794
 795             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 796
 797             /* Calculate temporary vectorial force */
 798             tx               = _mm256_mul_ps(fscal,dx00);
 799             ty               = _mm256_mul_ps(fscal,dy00);
 800             tz               = _mm256_mul_ps(fscal,dz00);
 801
 802             /* Update vectorial force */
 803             fix0             = _mm256_add_ps(fix0,tx);
 804             fiy0             = _mm256_add_ps(fiy0,ty);
 805             fiz0             = _mm256_add_ps(fiz0,tz);
 806
 807             fjx0             = _mm256_add_ps(fjx0,tx);
 808             fjy0             = _mm256_add_ps(fjy0,ty);
 809             fjz0             = _mm256_add_ps(fjz0,tz);
 810
 811             /**************************
 812              * CALCULATE INTERACTIONS *
 813              **************************/
 814
 815             /* REACTION-FIELD ELECTROSTATICS */
 816             velec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
 817             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
 818
 819             /* Update potential sum for this i atom from the interaction with this j atom. */
 820             velec            = _mm256_andnot_ps(dummy_mask,velec);
 821             velecsum         = _mm256_add_ps(velecsum,velec);
 822
 823             fscal            = felec;
 824
 825             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 826
 827             /* Calculate temporary vectorial force */
 828             tx               = _mm256_mul_ps(fscal,dx01);
 829             ty               = _mm256_mul_ps(fscal,dy01);
 830             tz               = _mm256_mul_ps(fscal,dz01);
 831
 832             /* Update vectorial force */
 833             fix0             = _mm256_add_ps(fix0,tx);
 834             fiy0             = _mm256_add_ps(fiy0,ty);
 835             fiz0             = _mm256_add_ps(fiz0,tz);
 836
 837             fjx1             = _mm256_add_ps(fjx1,tx);
 838             fjy1             = _mm256_add_ps(fjy1,ty);
 839             fjz1             = _mm256_add_ps(fjz1,tz);
 840
 841             /**************************
 842              * CALCULATE INTERACTIONS *
 843              **************************/
 844
 845             /* REACTION-FIELD ELECTROSTATICS */
 846             velec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
 847             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
 848
 849             /* Update potential sum for this i atom from the interaction with this j atom. */
 850             velec            = _mm256_andnot_ps(dummy_mask,velec);
 851             velecsum         = _mm256_add_ps(velecsum,velec);
 852
 853             fscal            = felec;
 854
 855             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 856
 857             /* Calculate temporary vectorial force */
 858             tx               = _mm256_mul_ps(fscal,dx02);
 859             ty               = _mm256_mul_ps(fscal,dy02);
 860             tz               = _mm256_mul_ps(fscal,dz02);
 861
 862             /* Update vectorial force */
 863             fix0             = _mm256_add_ps(fix0,tx);
 864             fiy0             = _mm256_add_ps(fiy0,ty);
 865             fiz0             = _mm256_add_ps(fiz0,tz);
 866
 867             fjx2             = _mm256_add_ps(fjx2,tx);
 868             fjy2             = _mm256_add_ps(fjy2,ty);
 869             fjz2             = _mm256_add_ps(fjz2,tz);
 870
 871             /**************************
 872              * CALCULATE INTERACTIONS *
 873              **************************/
 874
 875             /* REACTION-FIELD ELECTROSTATICS */
 876             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 877             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 878
 879             /* Update potential sum for this i atom from the interaction with this j atom. */
 880             velec            = _mm256_andnot_ps(dummy_mask,velec);
 881             velecsum         = _mm256_add_ps(velecsum,velec);
 882
 883             fscal            = felec;
 884
 885             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 886
 887             /* Calculate temporary vectorial force */
 888             tx               = _mm256_mul_ps(fscal,dx10);
 889             ty               = _mm256_mul_ps(fscal,dy10);
 890             tz               = _mm256_mul_ps(fscal,dz10);
 891
 892             /* Update vectorial force */
 893             fix1             = _mm256_add_ps(fix1,tx);
 894             fiy1             = _mm256_add_ps(fiy1,ty);
 895             fiz1             = _mm256_add_ps(fiz1,tz);
 896
 897             fjx0             = _mm256_add_ps(fjx0,tx);
 898             fjy0             = _mm256_add_ps(fjy0,ty);
 899             fjz0             = _mm256_add_ps(fjz0,tz);
 900
 901             /**************************
 902              * CALCULATE INTERACTIONS *
 903              **************************/
 904
 905             /* REACTION-FIELD ELECTROSTATICS */
 906             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 907             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 908
 909             /* Update potential sum for this i atom from the interaction with this j atom. */
 910             velec            = _mm256_andnot_ps(dummy_mask,velec);
 911             velecsum         = _mm256_add_ps(velecsum,velec);
 912
 913             fscal            = felec;
 914
 915             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 916
 917             /* Calculate temporary vectorial force */
 918             tx               = _mm256_mul_ps(fscal,dx11);
 919             ty               = _mm256_mul_ps(fscal,dy11);
 920             tz               = _mm256_mul_ps(fscal,dz11);
 921
 922             /* Update vectorial force */
 923             fix1             = _mm256_add_ps(fix1,tx);
 924             fiy1             = _mm256_add_ps(fiy1,ty);
 925             fiz1             = _mm256_add_ps(fiz1,tz);
 926
 927             fjx1             = _mm256_add_ps(fjx1,tx);
 928             fjy1             = _mm256_add_ps(fjy1,ty);
 929             fjz1             = _mm256_add_ps(fjz1,tz);
 930
 931             /**************************
 932              * CALCULATE INTERACTIONS *
 933              **************************/
 934
 935             /* REACTION-FIELD ELECTROSTATICS */
 936             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 937             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 938
 939             /* Update potential sum for this i atom from the interaction with this j atom. */
 940             velec            = _mm256_andnot_ps(dummy_mask,velec);
 941             velecsum         = _mm256_add_ps(velecsum,velec);
 942
 943             fscal            = felec;
 944
 945             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 946
 947             /* Calculate temporary vectorial force */
 948             tx               = _mm256_mul_ps(fscal,dx12);
 949             ty               = _mm256_mul_ps(fscal,dy12);
 950             tz               = _mm256_mul_ps(fscal,dz12);
 951
 952             /* Update vectorial force */
 953             fix1             = _mm256_add_ps(fix1,tx);
 954             fiy1             = _mm256_add_ps(fiy1,ty);
 955             fiz1             = _mm256_add_ps(fiz1,tz);
 956
 957             fjx2             = _mm256_add_ps(fjx2,tx);
 958             fjy2             = _mm256_add_ps(fjy2,ty);
 959             fjz2             = _mm256_add_ps(fjz2,tz);
 960
 961             /**************************
 962              * CALCULATE INTERACTIONS *
 963              **************************/
 964
 965             /* REACTION-FIELD ELECTROSTATICS */
 966             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 967             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 968
 969             /* Update potential sum for this i atom from the interaction with this j atom. */
 970             velec            = _mm256_andnot_ps(dummy_mask,velec);
 971             velecsum         = _mm256_add_ps(velecsum,velec);
 972
 973             fscal            = felec;
 974
 975             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 976
 977             /* Calculate temporary vectorial force */
 978             tx               = _mm256_mul_ps(fscal,dx20);
 979             ty               = _mm256_mul_ps(fscal,dy20);
 980             tz               = _mm256_mul_ps(fscal,dz20);
 981
 982             /* Update vectorial force */
 983             fix2             = _mm256_add_ps(fix2,tx);
 984             fiy2             = _mm256_add_ps(fiy2,ty);
 985             fiz2             = _mm256_add_ps(fiz2,tz);
 986
 987             fjx0             = _mm256_add_ps(fjx0,tx);
 988             fjy0             = _mm256_add_ps(fjy0,ty);
 989             fjz0             = _mm256_add_ps(fjz0,tz);
 990
 991             /**************************
 992              * CALCULATE INTERACTIONS *
 993              **************************/
 994
 995             /* REACTION-FIELD ELECTROSTATICS */
 996             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
 997             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
 998
 999             /* Update potential sum for this i atom from the interaction with this j atom. */
1000             velec            = _mm256_andnot_ps(dummy_mask,velec);
1001             velecsum         = _mm256_add_ps(velecsum,velec);
1002
1003             fscal            = felec;
1004
1005             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1006
1007             /* Calculate temporary vectorial force */
1008             tx               = _mm256_mul_ps(fscal,dx21);
1009             ty               = _mm256_mul_ps(fscal,dy21);
1010             tz               = _mm256_mul_ps(fscal,dz21);
1011
1012             /* Update vectorial force */
1013             fix2             = _mm256_add_ps(fix2,tx);
1014             fiy2             = _mm256_add_ps(fiy2,ty);
1015             fiz2             = _mm256_add_ps(fiz2,tz);
1016
1017             fjx1             = _mm256_add_ps(fjx1,tx);
1018             fjy1             = _mm256_add_ps(fjy1,ty);
1019             fjz1             = _mm256_add_ps(fjz1,tz);
1020
1021             /**************************
1022              * CALCULATE INTERACTIONS *
1023              **************************/
1024
1025             /* REACTION-FIELD ELECTROSTATICS */
1026             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
1027             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1028
1029             /* Update potential sum for this i atom from the interaction with this j atom. */
1030             velec            = _mm256_andnot_ps(dummy_mask,velec);
1031             velecsum         = _mm256_add_ps(velecsum,velec);
1032
1033             fscal            = felec;
1034
1035             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1036
1037             /* Calculate temporary vectorial force */
1038             tx               = _mm256_mul_ps(fscal,dx22);
1039             ty               = _mm256_mul_ps(fscal,dy22);
1040             tz               = _mm256_mul_ps(fscal,dz22);
1041
1042             /* Update vectorial force */
1043             fix2             = _mm256_add_ps(fix2,tx);
1044             fiy2             = _mm256_add_ps(fiy2,ty);
1045             fiz2             = _mm256_add_ps(fiz2,tz);
1046
1047             fjx2             = _mm256_add_ps(fjx2,tx);
1048             fjy2             = _mm256_add_ps(fjy2,ty);
1049             fjz2             = _mm256_add_ps(fjz2,tz);
1050
1051             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1052             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1053             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1054             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1055             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1056             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1057             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1058             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1059
1060             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1061                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1062
1063             /* Inner loop uses 324 flops */
1064         }
1065
1066         /* End of innermost loop */
1067
1068         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1069                                                  f+i_coord_offset,fshift+i_shift_offset);
1070
1071         ggid                        = gid[iidx];
1072         /* Update potential energies */
1073         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1074         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1075
1076         /* Increment number of inner iterations */
1077         inneriter                  += j_index_end - j_index_start;
1078
1079         /* Outer loop uses 20 flops */
1080     }
1081
1082     /* Increment number of outer iterations */
1083     outeriter        += nri;
1084
1085     /* Update outer/inner flops */
1086
1087     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*324);
1088 }
1089 /*
1090  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3W3_F_avx_256_single
1091  * Electrostatics interaction: ReactionField
1092  * VdW interaction:            CubicSplineTable
1093  * Geometry:                   Water3-Water3
1094  * Calculate force/pot:        Force
1095  */
1096 void
1097 nb_kernel_ElecRF_VdwCSTab_GeomW3W3_F_avx_256_single
1098                     (t_nblist * gmx_restrict                nlist,
1099                      rvec * gmx_restrict                    xx,
1100                      rvec * gmx_restrict                    ff,
1101                      t_forcerec * gmx_restrict              fr,
1102                      t_mdatoms * gmx_restrict               mdatoms,
1103                      nb_kernel_data_t * gmx_restrict        kernel_data,
1104                      t_nrnb * gmx_restrict                  nrnb)
1105 {
1106     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1107      * just 0 for non-waters.
1108      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1109      * jnr indices corresponding to data put in the four positions in the SIMD register.
1110      */
1111     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1112     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1113     int              jnrA,jnrB,jnrC,jnrD;
1114     int              jnrE,jnrF,jnrG,jnrH;
1115     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1116     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1117     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1118     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1119     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1120     real             rcutoff_scalar;
1121     real             *shiftvec,*fshift,*x,*f;
1122     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1123     real             scratch[4*DIM];
1124     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1125     real *           vdwioffsetptr0;
1126     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1127     real *           vdwioffsetptr1;
1128     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1129     real *           vdwioffsetptr2;
1130     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1131     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1132     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1133     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1134     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1135     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1136     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1137     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1138     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1139     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1140     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1141     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1142     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1143     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1144     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1145     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1146     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1147     real             *charge;
1148     int              nvdwtype;
1149     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1150     int              *vdwtype;
1151     real             *vdwparam;
1152     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1153     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1154     __m256i          vfitab;
1155     __m128i          vfitab_lo,vfitab_hi;
1156     __m128i          ifour       = _mm_set1_epi32(4);
1157     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1158     real             *vftab;
1159     __m256           dummy_mask,cutoff_mask;
1160     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1161     __m256           one     = _mm256_set1_ps(1.0);
1162     __m256           two     = _mm256_set1_ps(2.0);
1163     x                = xx[0];
1164     f                = ff[0];
1165
1166     nri              = nlist->nri;
1167     iinr             = nlist->iinr;
1168     jindex           = nlist->jindex;
1169     jjnr             = nlist->jjnr;
1170     shiftidx         = nlist->shift;
1171     gid              = nlist->gid;
1172     shiftvec         = fr->shift_vec[0];
1173     fshift           = fr->fshift[0];
1174     facel            = _mm256_set1_ps(fr->epsfac);
1175     charge           = mdatoms->chargeA;
1176     krf              = _mm256_set1_ps(fr->ic->k_rf);
1177     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
1178     crf              = _mm256_set1_ps(fr->ic->c_rf);
1179     nvdwtype         = fr->ntype;
1180     vdwparam         = fr->nbfp;
1181     vdwtype          = mdatoms->typeA;
1182
1183     vftab            = kernel_data->table_vdw->data;
1184     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
1185
1186     /* Setup water-specific parameters */
1187     inr              = nlist->iinr[0];
1188     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1189     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1190     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1191     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1192
1193     jq0              = _mm256_set1_ps(charge[inr+0]);
1194     jq1              = _mm256_set1_ps(charge[inr+1]);
1195     jq2              = _mm256_set1_ps(charge[inr+2]);
1196     vdwjidx0A        = 2*vdwtype[inr+0];
1197     qq00             = _mm256_mul_ps(iq0,jq0);
1198     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1199     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1200     qq01             = _mm256_mul_ps(iq0,jq1);
1201     qq02             = _mm256_mul_ps(iq0,jq2);
1202     qq10             = _mm256_mul_ps(iq1,jq0);
1203     qq11             = _mm256_mul_ps(iq1,jq1);
1204     qq12             = _mm256_mul_ps(iq1,jq2);
1205     qq20             = _mm256_mul_ps(iq2,jq0);
1206     qq21             = _mm256_mul_ps(iq2,jq1);
1207     qq22             = _mm256_mul_ps(iq2,jq2);
1208
1209     /* Avoid stupid compiler warnings */
1210     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1211     j_coord_offsetA = 0;
1212     j_coord_offsetB = 0;
1213     j_coord_offsetC = 0;
1214     j_coord_offsetD = 0;
1215     j_coord_offsetE = 0;
1216     j_coord_offsetF = 0;
1217     j_coord_offsetG = 0;
1218     j_coord_offsetH = 0;
1219
1220     outeriter        = 0;
1221     inneriter        = 0;
1222
1223     for(iidx=0;iidx<4*DIM;iidx++)
1224     {
1225         scratch[iidx] = 0.0;
1226     }
1227
1228     /* Start outer loop over neighborlists */
1229     for(iidx=0; iidx<nri; iidx++)
1230     {
1231         /* Load shift vector for this list */
1232         i_shift_offset   = DIM*shiftidx[iidx];
1233
1234         /* Load limits for loop over neighbors */
1235         j_index_start    = jindex[iidx];
1236         j_index_end      = jindex[iidx+1];
1237
1238         /* Get outer coordinate index */
1239         inr              = iinr[iidx];
1240         i_coord_offset   = DIM*inr;
1241
1242         /* Load i particle coords and add shift vector */
1243         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1244                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1245
1246         fix0             = _mm256_setzero_ps();
1247         fiy0             = _mm256_setzero_ps();
1248         fiz0             = _mm256_setzero_ps();
1249         fix1             = _mm256_setzero_ps();
1250         fiy1             = _mm256_setzero_ps();
1251         fiz1             = _mm256_setzero_ps();
1252         fix2             = _mm256_setzero_ps();
1253         fiy2             = _mm256_setzero_ps();
1254         fiz2             = _mm256_setzero_ps();
1255
1256         /* Start inner kernel loop */
1257         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1258         {
1259
1260             /* Get j neighbor index, and coordinate index */
1261             jnrA             = jjnr[jidx];
1262             jnrB             = jjnr[jidx+1];
1263             jnrC             = jjnr[jidx+2];
1264             jnrD             = jjnr[jidx+3];
1265             jnrE             = jjnr[jidx+4];
1266             jnrF             = jjnr[jidx+5];
1267             jnrG             = jjnr[jidx+6];
1268             jnrH             = jjnr[jidx+7];
1269             j_coord_offsetA  = DIM*jnrA;
1270             j_coord_offsetB  = DIM*jnrB;
1271             j_coord_offsetC  = DIM*jnrC;
1272             j_coord_offsetD  = DIM*jnrD;
1273             j_coord_offsetE  = DIM*jnrE;
1274             j_coord_offsetF  = DIM*jnrF;
1275             j_coord_offsetG  = DIM*jnrG;
1276             j_coord_offsetH  = DIM*jnrH;
1277
1278             /* load j atom coordinates */
1279             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1280                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1281                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1282                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1283                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1284
1285             /* Calculate displacement vector */
1286             dx00             = _mm256_sub_ps(ix0,jx0);
1287             dy00             = _mm256_sub_ps(iy0,jy0);
1288             dz00             = _mm256_sub_ps(iz0,jz0);
1289             dx01             = _mm256_sub_ps(ix0,jx1);
1290             dy01             = _mm256_sub_ps(iy0,jy1);
1291             dz01             = _mm256_sub_ps(iz0,jz1);
1292             dx02             = _mm256_sub_ps(ix0,jx2);
1293             dy02             = _mm256_sub_ps(iy0,jy2);
1294             dz02             = _mm256_sub_ps(iz0,jz2);
1295             dx10             = _mm256_sub_ps(ix1,jx0);
1296             dy10             = _mm256_sub_ps(iy1,jy0);
1297             dz10             = _mm256_sub_ps(iz1,jz0);
1298             dx11             = _mm256_sub_ps(ix1,jx1);
1299             dy11             = _mm256_sub_ps(iy1,jy1);
1300             dz11             = _mm256_sub_ps(iz1,jz1);
1301             dx12             = _mm256_sub_ps(ix1,jx2);
1302             dy12             = _mm256_sub_ps(iy1,jy2);
1303             dz12             = _mm256_sub_ps(iz1,jz2);
1304             dx20             = _mm256_sub_ps(ix2,jx0);
1305             dy20             = _mm256_sub_ps(iy2,jy0);
1306             dz20             = _mm256_sub_ps(iz2,jz0);
1307             dx21             = _mm256_sub_ps(ix2,jx1);
1308             dy21             = _mm256_sub_ps(iy2,jy1);
1309             dz21             = _mm256_sub_ps(iz2,jz1);
1310             dx22             = _mm256_sub_ps(ix2,jx2);
1311             dy22             = _mm256_sub_ps(iy2,jy2);
1312             dz22             = _mm256_sub_ps(iz2,jz2);
1313
1314             /* Calculate squared distance and things based on it */
1315             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1316             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1317             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1318             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1319             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1320             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1321             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1322             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1323             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1324
1325             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1326             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1327             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1328             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1329             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1330             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1331             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1332             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1333             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1334
1335             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1336             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1337             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1338             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1339             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1340             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1341             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1342             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1343             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1344
1345             fjx0             = _mm256_setzero_ps();
1346             fjy0             = _mm256_setzero_ps();
1347             fjz0             = _mm256_setzero_ps();
1348             fjx1             = _mm256_setzero_ps();
1349             fjy1             = _mm256_setzero_ps();
1350             fjz1             = _mm256_setzero_ps();
1351             fjx2             = _mm256_setzero_ps();
1352             fjy2             = _mm256_setzero_ps();
1353             fjz2             = _mm256_setzero_ps();
1354
1355             /**************************
1356              * CALCULATE INTERACTIONS *
1357              **************************/
1358
1359             r00              = _mm256_mul_ps(rsq00,rinv00);
1360
1361             /* Calculate table index by multiplying r with table scale and truncate to integer */
1362             rt               = _mm256_mul_ps(r00,vftabscale);
1363             vfitab           = _mm256_cvttps_epi32(rt);
1364             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1365             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1366             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1367             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1368             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1369             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1370
1371             /* REACTION-FIELD ELECTROSTATICS */
1372             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1373
1374             /* CUBIC SPLINE TABLE DISPERSION */
1375             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1376                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1377             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1378                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1379             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1380                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1381             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1382                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1383             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1384             Heps             = _mm256_mul_ps(vfeps,H);
1385             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1386             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1387             fvdw6            = _mm256_mul_ps(c6_00,FF);
1388
1389             /* CUBIC SPLINE TABLE REPULSION */
1390             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1391             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1392             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1393                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1394             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1395                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1396             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1397                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1398             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1399                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1400             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1401             Heps             = _mm256_mul_ps(vfeps,H);
1402             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1403             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1404             fvdw12           = _mm256_mul_ps(c12_00,FF);
1405             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1406
1407             fscal            = _mm256_add_ps(felec,fvdw);
1408
1409             /* Calculate temporary vectorial force */
1410             tx               = _mm256_mul_ps(fscal,dx00);
1411             ty               = _mm256_mul_ps(fscal,dy00);
1412             tz               = _mm256_mul_ps(fscal,dz00);
1413
1414             /* Update vectorial force */
1415             fix0             = _mm256_add_ps(fix0,tx);
1416             fiy0             = _mm256_add_ps(fiy0,ty);
1417             fiz0             = _mm256_add_ps(fiz0,tz);
1418
1419             fjx0             = _mm256_add_ps(fjx0,tx);
1420             fjy0             = _mm256_add_ps(fjy0,ty);
1421             fjz0             = _mm256_add_ps(fjz0,tz);
1422
1423             /**************************
1424              * CALCULATE INTERACTIONS *
1425              **************************/
1426
1427             /* REACTION-FIELD ELECTROSTATICS */
1428             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1429
1430             fscal            = felec;
1431
1432             /* Calculate temporary vectorial force */
1433             tx               = _mm256_mul_ps(fscal,dx01);
1434             ty               = _mm256_mul_ps(fscal,dy01);
1435             tz               = _mm256_mul_ps(fscal,dz01);
1436
1437             /* Update vectorial force */
1438             fix0             = _mm256_add_ps(fix0,tx);
1439             fiy0             = _mm256_add_ps(fiy0,ty);
1440             fiz0             = _mm256_add_ps(fiz0,tz);
1441
1442             fjx1             = _mm256_add_ps(fjx1,tx);
1443             fjy1             = _mm256_add_ps(fjy1,ty);
1444             fjz1             = _mm256_add_ps(fjz1,tz);
1445
1446             /**************************
1447              * CALCULATE INTERACTIONS *
1448              **************************/
1449
1450             /* REACTION-FIELD ELECTROSTATICS */
1451             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1452
1453             fscal            = felec;
1454
1455             /* Calculate temporary vectorial force */
1456             tx               = _mm256_mul_ps(fscal,dx02);
1457             ty               = _mm256_mul_ps(fscal,dy02);
1458             tz               = _mm256_mul_ps(fscal,dz02);
1459
1460             /* Update vectorial force */
1461             fix0             = _mm256_add_ps(fix0,tx);
1462             fiy0             = _mm256_add_ps(fiy0,ty);
1463             fiz0             = _mm256_add_ps(fiz0,tz);
1464
1465             fjx2             = _mm256_add_ps(fjx2,tx);
1466             fjy2             = _mm256_add_ps(fjy2,ty);
1467             fjz2             = _mm256_add_ps(fjz2,tz);
1468
1469             /**************************
1470              * CALCULATE INTERACTIONS *
1471              **************************/
1472
1473             /* REACTION-FIELD ELECTROSTATICS */
1474             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1475
1476             fscal            = felec;
1477
1478             /* Calculate temporary vectorial force */
1479             tx               = _mm256_mul_ps(fscal,dx10);
1480             ty               = _mm256_mul_ps(fscal,dy10);
1481             tz               = _mm256_mul_ps(fscal,dz10);
1482
1483             /* Update vectorial force */
1484             fix1             = _mm256_add_ps(fix1,tx);
1485             fiy1             = _mm256_add_ps(fiy1,ty);
1486             fiz1             = _mm256_add_ps(fiz1,tz);
1487
1488             fjx0             = _mm256_add_ps(fjx0,tx);
1489             fjy0             = _mm256_add_ps(fjy0,ty);
1490             fjz0             = _mm256_add_ps(fjz0,tz);
1491
1492             /**************************
1493              * CALCULATE INTERACTIONS *
1494              **************************/
1495
1496             /* REACTION-FIELD ELECTROSTATICS */
1497             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1498
1499             fscal            = felec;
1500
1501             /* Calculate temporary vectorial force */
1502             tx               = _mm256_mul_ps(fscal,dx11);
1503             ty               = _mm256_mul_ps(fscal,dy11);
1504             tz               = _mm256_mul_ps(fscal,dz11);
1505
1506             /* Update vectorial force */
1507             fix1             = _mm256_add_ps(fix1,tx);
1508             fiy1             = _mm256_add_ps(fiy1,ty);
1509             fiz1             = _mm256_add_ps(fiz1,tz);
1510
1511             fjx1             = _mm256_add_ps(fjx1,tx);
1512             fjy1             = _mm256_add_ps(fjy1,ty);
1513             fjz1             = _mm256_add_ps(fjz1,tz);
1514
1515             /**************************
1516              * CALCULATE INTERACTIONS *
1517              **************************/
1518
1519             /* REACTION-FIELD ELECTROSTATICS */
1520             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1521
1522             fscal            = felec;
1523
1524             /* Calculate temporary vectorial force */
1525             tx               = _mm256_mul_ps(fscal,dx12);
1526             ty               = _mm256_mul_ps(fscal,dy12);
1527             tz               = _mm256_mul_ps(fscal,dz12);
1528
1529             /* Update vectorial force */
1530             fix1             = _mm256_add_ps(fix1,tx);
1531             fiy1             = _mm256_add_ps(fiy1,ty);
1532             fiz1             = _mm256_add_ps(fiz1,tz);
1533
1534             fjx2             = _mm256_add_ps(fjx2,tx);
1535             fjy2             = _mm256_add_ps(fjy2,ty);
1536             fjz2             = _mm256_add_ps(fjz2,tz);
1537
1538             /**************************
1539              * CALCULATE INTERACTIONS *
1540              **************************/
1541
1542             /* REACTION-FIELD ELECTROSTATICS */
1543             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1544
1545             fscal            = felec;
1546
1547             /* Calculate temporary vectorial force */
1548             tx               = _mm256_mul_ps(fscal,dx20);
1549             ty               = _mm256_mul_ps(fscal,dy20);
1550             tz               = _mm256_mul_ps(fscal,dz20);
1551
1552             /* Update vectorial force */
1553             fix2             = _mm256_add_ps(fix2,tx);
1554             fiy2             = _mm256_add_ps(fiy2,ty);
1555             fiz2             = _mm256_add_ps(fiz2,tz);
1556
1557             fjx0             = _mm256_add_ps(fjx0,tx);
1558             fjy0             = _mm256_add_ps(fjy0,ty);
1559             fjz0             = _mm256_add_ps(fjz0,tz);
1560
1561             /**************************
1562              * CALCULATE INTERACTIONS *
1563              **************************/
1564
1565             /* REACTION-FIELD ELECTROSTATICS */
1566             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1567
1568             fscal            = felec;
1569
1570             /* Calculate temporary vectorial force */
1571             tx               = _mm256_mul_ps(fscal,dx21);
1572             ty               = _mm256_mul_ps(fscal,dy21);
1573             tz               = _mm256_mul_ps(fscal,dz21);
1574
1575             /* Update vectorial force */
1576             fix2             = _mm256_add_ps(fix2,tx);
1577             fiy2             = _mm256_add_ps(fiy2,ty);
1578             fiz2             = _mm256_add_ps(fiz2,tz);
1579
1580             fjx1             = _mm256_add_ps(fjx1,tx);
1581             fjy1             = _mm256_add_ps(fjy1,ty);
1582             fjz1             = _mm256_add_ps(fjz1,tz);
1583
1584             /**************************
1585              * CALCULATE INTERACTIONS *
1586              **************************/
1587
1588             /* REACTION-FIELD ELECTROSTATICS */
1589             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1590
1591             fscal            = felec;
1592
1593             /* Calculate temporary vectorial force */
1594             tx               = _mm256_mul_ps(fscal,dx22);
1595             ty               = _mm256_mul_ps(fscal,dy22);
1596             tz               = _mm256_mul_ps(fscal,dz22);
1597
1598             /* Update vectorial force */
1599             fix2             = _mm256_add_ps(fix2,tx);
1600             fiy2             = _mm256_add_ps(fiy2,ty);
1601             fiz2             = _mm256_add_ps(fiz2,tz);
1602
1603             fjx2             = _mm256_add_ps(fjx2,tx);
1604             fjy2             = _mm256_add_ps(fjy2,ty);
1605             fjz2             = _mm256_add_ps(fjz2,tz);
1606
1607             fjptrA             = f+j_coord_offsetA;
1608             fjptrB             = f+j_coord_offsetB;
1609             fjptrC             = f+j_coord_offsetC;
1610             fjptrD             = f+j_coord_offsetD;
1611             fjptrE             = f+j_coord_offsetE;
1612             fjptrF             = f+j_coord_offsetF;
1613             fjptrG             = f+j_coord_offsetG;
1614             fjptrH             = f+j_coord_offsetH;
1615
1616             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1617                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1618
1619             /* Inner loop uses 270 flops */
1620         }
1621
1622         if(jidx<j_index_end)
1623         {
1624
1625             /* Get j neighbor index, and coordinate index */
1626             jnrlistA         = jjnr[jidx];
1627             jnrlistB         = jjnr[jidx+1];
1628             jnrlistC         = jjnr[jidx+2];
1629             jnrlistD         = jjnr[jidx+3];
1630             jnrlistE         = jjnr[jidx+4];
1631             jnrlistF         = jjnr[jidx+5];
1632             jnrlistG         = jjnr[jidx+6];
1633             jnrlistH         = jjnr[jidx+7];
1634             /* Sign of each element will be negative for non-real atoms.
1635              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1636              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1637              */
1638             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1639                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1640
1641             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1642             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1643             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1644             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1645             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1646             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1647             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1648             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1649             j_coord_offsetA  = DIM*jnrA;
1650             j_coord_offsetB  = DIM*jnrB;
1651             j_coord_offsetC  = DIM*jnrC;
1652             j_coord_offsetD  = DIM*jnrD;
1653             j_coord_offsetE  = DIM*jnrE;
1654             j_coord_offsetF  = DIM*jnrF;
1655             j_coord_offsetG  = DIM*jnrG;
1656             j_coord_offsetH  = DIM*jnrH;
1657
1658             /* load j atom coordinates */
1659             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1660                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1661                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1662                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1663                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1664
1665             /* Calculate displacement vector */
1666             dx00             = _mm256_sub_ps(ix0,jx0);
1667             dy00             = _mm256_sub_ps(iy0,jy0);
1668             dz00             = _mm256_sub_ps(iz0,jz0);
1669             dx01             = _mm256_sub_ps(ix0,jx1);
1670             dy01             = _mm256_sub_ps(iy0,jy1);
1671             dz01             = _mm256_sub_ps(iz0,jz1);
1672             dx02             = _mm256_sub_ps(ix0,jx2);
1673             dy02             = _mm256_sub_ps(iy0,jy2);
1674             dz02             = _mm256_sub_ps(iz0,jz2);
1675             dx10             = _mm256_sub_ps(ix1,jx0);
1676             dy10             = _mm256_sub_ps(iy1,jy0);
1677             dz10             = _mm256_sub_ps(iz1,jz0);
1678             dx11             = _mm256_sub_ps(ix1,jx1);
1679             dy11             = _mm256_sub_ps(iy1,jy1);
1680             dz11             = _mm256_sub_ps(iz1,jz1);
1681             dx12             = _mm256_sub_ps(ix1,jx2);
1682             dy12             = _mm256_sub_ps(iy1,jy2);
1683             dz12             = _mm256_sub_ps(iz1,jz2);
1684             dx20             = _mm256_sub_ps(ix2,jx0);
1685             dy20             = _mm256_sub_ps(iy2,jy0);
1686             dz20             = _mm256_sub_ps(iz2,jz0);
1687             dx21             = _mm256_sub_ps(ix2,jx1);
1688             dy21             = _mm256_sub_ps(iy2,jy1);
1689             dz21             = _mm256_sub_ps(iz2,jz1);
1690             dx22             = _mm256_sub_ps(ix2,jx2);
1691             dy22             = _mm256_sub_ps(iy2,jy2);
1692             dz22             = _mm256_sub_ps(iz2,jz2);
1693
1694             /* Calculate squared distance and things based on it */
1695             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1696             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1697             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1698             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1699             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1700             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1701             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1702             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1703             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1704
1705             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1706             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1707             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1708             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1709             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1710             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1711             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1712             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1713             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1714
1715             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1716             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1717             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1718             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1719             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1720             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1721             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1722             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1723             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1724
1725             fjx0             = _mm256_setzero_ps();
1726             fjy0             = _mm256_setzero_ps();
1727             fjz0             = _mm256_setzero_ps();
1728             fjx1             = _mm256_setzero_ps();
1729             fjy1             = _mm256_setzero_ps();
1730             fjz1             = _mm256_setzero_ps();
1731             fjx2             = _mm256_setzero_ps();
1732             fjy2             = _mm256_setzero_ps();
1733             fjz2             = _mm256_setzero_ps();
1734
1735             /**************************
1736              * CALCULATE INTERACTIONS *
1737              **************************/
1738
1739             r00              = _mm256_mul_ps(rsq00,rinv00);
1740             r00              = _mm256_andnot_ps(dummy_mask,r00);
1741
1742             /* Calculate table index by multiplying r with table scale and truncate to integer */
1743             rt               = _mm256_mul_ps(r00,vftabscale);
1744             vfitab           = _mm256_cvttps_epi32(rt);
1745             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1746             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1747             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1748             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1749             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1750             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1751
1752             /* REACTION-FIELD ELECTROSTATICS */
1753             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1754
1755             /* CUBIC SPLINE TABLE DISPERSION */
1756             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1757                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1758             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1759                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1760             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1761                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1762             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1763                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1764             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1765             Heps             = _mm256_mul_ps(vfeps,H);
1766             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1767             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1768             fvdw6            = _mm256_mul_ps(c6_00,FF);
1769
1770             /* CUBIC SPLINE TABLE REPULSION */
1771             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1772             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1773             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1774                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1775             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1776                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1777             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1778                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1779             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1780                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1781             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1782             Heps             = _mm256_mul_ps(vfeps,H);
1783             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1784             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1785             fvdw12           = _mm256_mul_ps(c12_00,FF);
1786             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1787
1788             fscal            = _mm256_add_ps(felec,fvdw);
1789
1790             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1791
1792             /* Calculate temporary vectorial force */
1793             tx               = _mm256_mul_ps(fscal,dx00);
1794             ty               = _mm256_mul_ps(fscal,dy00);
1795             tz               = _mm256_mul_ps(fscal,dz00);
1796
1797             /* Update vectorial force */
1798             fix0             = _mm256_add_ps(fix0,tx);
1799             fiy0             = _mm256_add_ps(fiy0,ty);
1800             fiz0             = _mm256_add_ps(fiz0,tz);
1801
1802             fjx0             = _mm256_add_ps(fjx0,tx);
1803             fjy0             = _mm256_add_ps(fjy0,ty);
1804             fjz0             = _mm256_add_ps(fjz0,tz);
1805
1806             /**************************
1807              * CALCULATE INTERACTIONS *
1808              **************************/
1809
1810             /* REACTION-FIELD ELECTROSTATICS */
1811             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1812
1813             fscal            = felec;
1814
1815             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1816
1817             /* Calculate temporary vectorial force */
1818             tx               = _mm256_mul_ps(fscal,dx01);
1819             ty               = _mm256_mul_ps(fscal,dy01);
1820             tz               = _mm256_mul_ps(fscal,dz01);
1821
1822             /* Update vectorial force */
1823             fix0             = _mm256_add_ps(fix0,tx);
1824             fiy0             = _mm256_add_ps(fiy0,ty);
1825             fiz0             = _mm256_add_ps(fiz0,tz);
1826
1827             fjx1             = _mm256_add_ps(fjx1,tx);
1828             fjy1             = _mm256_add_ps(fjy1,ty);
1829             fjz1             = _mm256_add_ps(fjz1,tz);
1830
1831             /**************************
1832              * CALCULATE INTERACTIONS *
1833              **************************/
1834
1835             /* REACTION-FIELD ELECTROSTATICS */
1836             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1837
1838             fscal            = felec;
1839
1840             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1841
1842             /* Calculate temporary vectorial force */
1843             tx               = _mm256_mul_ps(fscal,dx02);
1844             ty               = _mm256_mul_ps(fscal,dy02);
1845             tz               = _mm256_mul_ps(fscal,dz02);
1846
1847             /* Update vectorial force */
1848             fix0             = _mm256_add_ps(fix0,tx);
1849             fiy0             = _mm256_add_ps(fiy0,ty);
1850             fiz0             = _mm256_add_ps(fiz0,tz);
1851
1852             fjx2             = _mm256_add_ps(fjx2,tx);
1853             fjy2             = _mm256_add_ps(fjy2,ty);
1854             fjz2             = _mm256_add_ps(fjz2,tz);
1855
1856             /**************************
1857              * CALCULATE INTERACTIONS *
1858              **************************/
1859
1860             /* REACTION-FIELD ELECTROSTATICS */
1861             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1862
1863             fscal            = felec;
1864
1865             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1866
1867             /* Calculate temporary vectorial force */
1868             tx               = _mm256_mul_ps(fscal,dx10);
1869             ty               = _mm256_mul_ps(fscal,dy10);
1870             tz               = _mm256_mul_ps(fscal,dz10);
1871
1872             /* Update vectorial force */
1873             fix1             = _mm256_add_ps(fix1,tx);
1874             fiy1             = _mm256_add_ps(fiy1,ty);
1875             fiz1             = _mm256_add_ps(fiz1,tz);
1876
1877             fjx0             = _mm256_add_ps(fjx0,tx);
1878             fjy0             = _mm256_add_ps(fjy0,ty);
1879             fjz0             = _mm256_add_ps(fjz0,tz);
1880
1881             /**************************
1882              * CALCULATE INTERACTIONS *
1883              **************************/
1884
1885             /* REACTION-FIELD ELECTROSTATICS */
1886             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1887
1888             fscal            = felec;
1889
1890             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1891
1892             /* Calculate temporary vectorial force */
1893             tx               = _mm256_mul_ps(fscal,dx11);
1894             ty               = _mm256_mul_ps(fscal,dy11);
1895             tz               = _mm256_mul_ps(fscal,dz11);
1896
1897             /* Update vectorial force */
1898             fix1             = _mm256_add_ps(fix1,tx);
1899             fiy1             = _mm256_add_ps(fiy1,ty);
1900             fiz1             = _mm256_add_ps(fiz1,tz);
1901
1902             fjx1             = _mm256_add_ps(fjx1,tx);
1903             fjy1             = _mm256_add_ps(fjy1,ty);
1904             fjz1             = _mm256_add_ps(fjz1,tz);
1905
1906             /**************************
1907              * CALCULATE INTERACTIONS *
1908              **************************/
1909
1910             /* REACTION-FIELD ELECTROSTATICS */
1911             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1912
1913             fscal            = felec;
1914
1915             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1916
1917             /* Calculate temporary vectorial force */
1918             tx               = _mm256_mul_ps(fscal,dx12);
1919             ty               = _mm256_mul_ps(fscal,dy12);
1920             tz               = _mm256_mul_ps(fscal,dz12);
1921
1922             /* Update vectorial force */
1923             fix1             = _mm256_add_ps(fix1,tx);
1924             fiy1             = _mm256_add_ps(fiy1,ty);
1925             fiz1             = _mm256_add_ps(fiz1,tz);
1926
1927             fjx2             = _mm256_add_ps(fjx2,tx);
1928             fjy2             = _mm256_add_ps(fjy2,ty);
1929             fjz2             = _mm256_add_ps(fjz2,tz);
1930
1931             /**************************
1932              * CALCULATE INTERACTIONS *
1933              **************************/
1934
1935             /* REACTION-FIELD ELECTROSTATICS */
1936             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1937
1938             fscal            = felec;
1939
1940             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1941
1942             /* Calculate temporary vectorial force */
1943             tx               = _mm256_mul_ps(fscal,dx20);
1944             ty               = _mm256_mul_ps(fscal,dy20);
1945             tz               = _mm256_mul_ps(fscal,dz20);
1946
1947             /* Update vectorial force */
1948             fix2             = _mm256_add_ps(fix2,tx);
1949             fiy2             = _mm256_add_ps(fiy2,ty);
1950             fiz2             = _mm256_add_ps(fiz2,tz);
1951
1952             fjx0             = _mm256_add_ps(fjx0,tx);
1953             fjy0             = _mm256_add_ps(fjy0,ty);
1954             fjz0             = _mm256_add_ps(fjz0,tz);
1955
1956             /**************************
1957              * CALCULATE INTERACTIONS *
1958              **************************/
1959
1960             /* REACTION-FIELD ELECTROSTATICS */
1961             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1962
1963             fscal            = felec;
1964
1965             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1966
1967             /* Calculate temporary vectorial force */
1968             tx               = _mm256_mul_ps(fscal,dx21);
1969             ty               = _mm256_mul_ps(fscal,dy21);
1970             tz               = _mm256_mul_ps(fscal,dz21);
1971
1972             /* Update vectorial force */
1973             fix2             = _mm256_add_ps(fix2,tx);
1974             fiy2             = _mm256_add_ps(fiy2,ty);
1975             fiz2             = _mm256_add_ps(fiz2,tz);
1976
1977             fjx1             = _mm256_add_ps(fjx1,tx);
1978             fjy1             = _mm256_add_ps(fjy1,ty);
1979             fjz1             = _mm256_add_ps(fjz1,tz);
1980
1981             /**************************
1982              * CALCULATE INTERACTIONS *
1983              **************************/
1984
1985             /* REACTION-FIELD ELECTROSTATICS */
1986             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1987
1988             fscal            = felec;
1989
1990             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1991
1992             /* Calculate temporary vectorial force */
1993             tx               = _mm256_mul_ps(fscal,dx22);
1994             ty               = _mm256_mul_ps(fscal,dy22);
1995             tz               = _mm256_mul_ps(fscal,dz22);
1996
1997             /* Update vectorial force */
1998             fix2             = _mm256_add_ps(fix2,tx);
1999             fiy2             = _mm256_add_ps(fiy2,ty);
2000             fiz2             = _mm256_add_ps(fiz2,tz);
2001
2002             fjx2             = _mm256_add_ps(fjx2,tx);
2003             fjy2             = _mm256_add_ps(fjy2,ty);
2004             fjz2             = _mm256_add_ps(fjz2,tz);
2005
2006             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2007             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2008             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2009             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2010             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2011             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2012             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2013             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2014
2015             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2016                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2017
2018             /* Inner loop uses 271 flops */
2019         }
2020
2021         /* End of innermost loop */
2022
2023         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2024                                                  f+i_coord_offset,fshift+i_shift_offset);
2025
2026         /* Increment number of inner iterations */
2027         inneriter                  += j_index_end - j_index_start;
2028
2029         /* Outer loop uses 18 flops */
2030     }
2031
2032     /* Increment number of outer iterations */
2033     outeriter        += nri;
2034
2035     /* Update outer/inner flops */
2036
2037     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*271);
2038 }