src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_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_ElecRFCut_VdwCSTab_GeomW4W4_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Water4
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_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     real *           vdwioffsetptr3;
  79     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  80     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  81     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  82     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  83     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  84     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  85     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  86     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
  87     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  88     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  90     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  91     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  92     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  93     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  94     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  95     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  96     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
  97     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
  98     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  99     real             *charge;
 100     int              nvdwtype;
 101     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 102     int              *vdwtype;
 103     real             *vdwparam;
 104     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 105     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 106     __m256i          vfitab;
 107     __m128i          vfitab_lo,vfitab_hi;
 108     __m128i          ifour       = _mm_set1_epi32(4);
 109     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 110     real             *vftab;
 111     __m256           dummy_mask,cutoff_mask;
 112     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 113     __m256           one     = _mm256_set1_ps(1.0);
 114     __m256           two     = _mm256_set1_ps(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm256_set1_ps(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     krf              = _mm256_set1_ps(fr->ic->k_rf);
 129     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 130     crf              = _mm256_set1_ps(fr->ic->c_rf);
 131     nvdwtype         = fr->ntype;
 132     vdwparam         = fr->nbfp;
 133     vdwtype          = mdatoms->typeA;
 134
 135     vftab            = kernel_data->table_vdw->data;
 136     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 137
 138     /* Setup water-specific parameters */
 139     inr              = nlist->iinr[0];
 140     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 141     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 142     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 143     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 144
 145     jq1              = _mm256_set1_ps(charge[inr+1]);
 146     jq2              = _mm256_set1_ps(charge[inr+2]);
 147     jq3              = _mm256_set1_ps(charge[inr+3]);
 148     vdwjidx0A        = 2*vdwtype[inr+0];
 149     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 150     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 151     qq11             = _mm256_mul_ps(iq1,jq1);
 152     qq12             = _mm256_mul_ps(iq1,jq2);
 153     qq13             = _mm256_mul_ps(iq1,jq3);
 154     qq21             = _mm256_mul_ps(iq2,jq1);
 155     qq22             = _mm256_mul_ps(iq2,jq2);
 156     qq23             = _mm256_mul_ps(iq2,jq3);
 157     qq31             = _mm256_mul_ps(iq3,jq1);
 158     qq32             = _mm256_mul_ps(iq3,jq2);
 159     qq33             = _mm256_mul_ps(iq3,jq3);
 160
 161     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 162     rcutoff_scalar   = fr->rcoulomb;
 163     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 164     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 165
 166     /* Avoid stupid compiler warnings */
 167     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 168     j_coord_offsetA = 0;
 169     j_coord_offsetB = 0;
 170     j_coord_offsetC = 0;
 171     j_coord_offsetD = 0;
 172     j_coord_offsetE = 0;
 173     j_coord_offsetF = 0;
 174     j_coord_offsetG = 0;
 175     j_coord_offsetH = 0;
 176
 177     outeriter        = 0;
 178     inneriter        = 0;
 179
 180     for(iidx=0;iidx<4*DIM;iidx++)
 181     {
 182         scratch[iidx] = 0.0;
 183     }
 184
 185     /* Start outer loop over neighborlists */
 186     for(iidx=0; iidx<nri; iidx++)
 187     {
 188         /* Load shift vector for this list */
 189         i_shift_offset   = DIM*shiftidx[iidx];
 190
 191         /* Load limits for loop over neighbors */
 192         j_index_start    = jindex[iidx];
 193         j_index_end      = jindex[iidx+1];
 194
 195         /* Get outer coordinate index */
 196         inr              = iinr[iidx];
 197         i_coord_offset   = DIM*inr;
 198
 199         /* Load i particle coords and add shift vector */
 200         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 201                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 202
 203         fix0             = _mm256_setzero_ps();
 204         fiy0             = _mm256_setzero_ps();
 205         fiz0             = _mm256_setzero_ps();
 206         fix1             = _mm256_setzero_ps();
 207         fiy1             = _mm256_setzero_ps();
 208         fiz1             = _mm256_setzero_ps();
 209         fix2             = _mm256_setzero_ps();
 210         fiy2             = _mm256_setzero_ps();
 211         fiz2             = _mm256_setzero_ps();
 212         fix3             = _mm256_setzero_ps();
 213         fiy3             = _mm256_setzero_ps();
 214         fiz3             = _mm256_setzero_ps();
 215
 216         /* Reset potential sums */
 217         velecsum         = _mm256_setzero_ps();
 218         vvdwsum          = _mm256_setzero_ps();
 219
 220         /* Start inner kernel loop */
 221         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 222         {
 223
 224             /* Get j neighbor index, and coordinate index */
 225             jnrA             = jjnr[jidx];
 226             jnrB             = jjnr[jidx+1];
 227             jnrC             = jjnr[jidx+2];
 228             jnrD             = jjnr[jidx+3];
 229             jnrE             = jjnr[jidx+4];
 230             jnrF             = jjnr[jidx+5];
 231             jnrG             = jjnr[jidx+6];
 232             jnrH             = jjnr[jidx+7];
 233             j_coord_offsetA  = DIM*jnrA;
 234             j_coord_offsetB  = DIM*jnrB;
 235             j_coord_offsetC  = DIM*jnrC;
 236             j_coord_offsetD  = DIM*jnrD;
 237             j_coord_offsetE  = DIM*jnrE;
 238             j_coord_offsetF  = DIM*jnrF;
 239             j_coord_offsetG  = DIM*jnrG;
 240             j_coord_offsetH  = DIM*jnrH;
 241
 242             /* load j atom coordinates */
 243             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 244                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 245                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 246                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 247                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 248                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 249
 250             /* Calculate displacement vector */
 251             dx00             = _mm256_sub_ps(ix0,jx0);
 252             dy00             = _mm256_sub_ps(iy0,jy0);
 253             dz00             = _mm256_sub_ps(iz0,jz0);
 254             dx11             = _mm256_sub_ps(ix1,jx1);
 255             dy11             = _mm256_sub_ps(iy1,jy1);
 256             dz11             = _mm256_sub_ps(iz1,jz1);
 257             dx12             = _mm256_sub_ps(ix1,jx2);
 258             dy12             = _mm256_sub_ps(iy1,jy2);
 259             dz12             = _mm256_sub_ps(iz1,jz2);
 260             dx13             = _mm256_sub_ps(ix1,jx3);
 261             dy13             = _mm256_sub_ps(iy1,jy3);
 262             dz13             = _mm256_sub_ps(iz1,jz3);
 263             dx21             = _mm256_sub_ps(ix2,jx1);
 264             dy21             = _mm256_sub_ps(iy2,jy1);
 265             dz21             = _mm256_sub_ps(iz2,jz1);
 266             dx22             = _mm256_sub_ps(ix2,jx2);
 267             dy22             = _mm256_sub_ps(iy2,jy2);
 268             dz22             = _mm256_sub_ps(iz2,jz2);
 269             dx23             = _mm256_sub_ps(ix2,jx3);
 270             dy23             = _mm256_sub_ps(iy2,jy3);
 271             dz23             = _mm256_sub_ps(iz2,jz3);
 272             dx31             = _mm256_sub_ps(ix3,jx1);
 273             dy31             = _mm256_sub_ps(iy3,jy1);
 274             dz31             = _mm256_sub_ps(iz3,jz1);
 275             dx32             = _mm256_sub_ps(ix3,jx2);
 276             dy32             = _mm256_sub_ps(iy3,jy2);
 277             dz32             = _mm256_sub_ps(iz3,jz2);
 278             dx33             = _mm256_sub_ps(ix3,jx3);
 279             dy33             = _mm256_sub_ps(iy3,jy3);
 280             dz33             = _mm256_sub_ps(iz3,jz3);
 281
 282             /* Calculate squared distance and things based on it */
 283             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 284             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 285             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 286             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 287             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 288             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 289             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 290             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 291             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 292             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 293
 294             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 295             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 296             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 297             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 298             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 299             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 300             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 301             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 302             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 303             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 304
 305             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 306             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 307             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 308             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 309             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 310             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 311             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 312             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 313             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 314
 315             fjx0             = _mm256_setzero_ps();
 316             fjy0             = _mm256_setzero_ps();
 317             fjz0             = _mm256_setzero_ps();
 318             fjx1             = _mm256_setzero_ps();
 319             fjy1             = _mm256_setzero_ps();
 320             fjz1             = _mm256_setzero_ps();
 321             fjx2             = _mm256_setzero_ps();
 322             fjy2             = _mm256_setzero_ps();
 323             fjz2             = _mm256_setzero_ps();
 324             fjx3             = _mm256_setzero_ps();
 325             fjy3             = _mm256_setzero_ps();
 326             fjz3             = _mm256_setzero_ps();
 327
 328             /**************************
 329              * CALCULATE INTERACTIONS *
 330              **************************/
 331
 332             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 333             {
 334
 335             r00              = _mm256_mul_ps(rsq00,rinv00);
 336
 337             /* Calculate table index by multiplying r with table scale and truncate to integer */
 338             rt               = _mm256_mul_ps(r00,vftabscale);
 339             vfitab           = _mm256_cvttps_epi32(rt);
 340             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 341             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 342             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 343             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 344             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 345             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 346
 347             /* CUBIC SPLINE TABLE DISPERSION */
 348             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 349                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 350             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 351                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 352             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 353                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 354             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 355                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 356             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 357             Heps             = _mm256_mul_ps(vfeps,H);
 358             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 359             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 360             vvdw6            = _mm256_mul_ps(c6_00,VV);
 361             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 362             fvdw6            = _mm256_mul_ps(c6_00,FF);
 363
 364             /* CUBIC SPLINE TABLE REPULSION */
 365             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 366             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 367             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 368                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 369             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 370                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 371             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 372                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 373             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 374                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 375             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 376             Heps             = _mm256_mul_ps(vfeps,H);
 377             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 378             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 379             vvdw12           = _mm256_mul_ps(c12_00,VV);
 380             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 381             fvdw12           = _mm256_mul_ps(c12_00,FF);
 382             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 383             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 384
 385             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 389             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 390
 391             fscal            = fvdw;
 392
 393             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm256_mul_ps(fscal,dx00);
 397             ty               = _mm256_mul_ps(fscal,dy00);
 398             tz               = _mm256_mul_ps(fscal,dz00);
 399
 400             /* Update vectorial force */
 401             fix0             = _mm256_add_ps(fix0,tx);
 402             fiy0             = _mm256_add_ps(fiy0,ty);
 403             fiz0             = _mm256_add_ps(fiz0,tz);
 404
 405             fjx0             = _mm256_add_ps(fjx0,tx);
 406             fjy0             = _mm256_add_ps(fjy0,ty);
 407             fjz0             = _mm256_add_ps(fjz0,tz);
 408
 409             }
 410
 411             /**************************
 412              * CALCULATE INTERACTIONS *
 413              **************************/
 414
 415             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 416             {
 417
 418             /* REACTION-FIELD ELECTROSTATICS */
 419             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 420             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 421
 422             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 423
 424             /* Update potential sum for this i atom from the interaction with this j atom. */
 425             velec            = _mm256_and_ps(velec,cutoff_mask);
 426             velecsum         = _mm256_add_ps(velecsum,velec);
 427
 428             fscal            = felec;
 429
 430             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 431
 432             /* Calculate temporary vectorial force */
 433             tx               = _mm256_mul_ps(fscal,dx11);
 434             ty               = _mm256_mul_ps(fscal,dy11);
 435             tz               = _mm256_mul_ps(fscal,dz11);
 436
 437             /* Update vectorial force */
 438             fix1             = _mm256_add_ps(fix1,tx);
 439             fiy1             = _mm256_add_ps(fiy1,ty);
 440             fiz1             = _mm256_add_ps(fiz1,tz);
 441
 442             fjx1             = _mm256_add_ps(fjx1,tx);
 443             fjy1             = _mm256_add_ps(fjy1,ty);
 444             fjz1             = _mm256_add_ps(fjz1,tz);
 445
 446             }
 447
 448             /**************************
 449              * CALCULATE INTERACTIONS *
 450              **************************/
 451
 452             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 453             {
 454
 455             /* REACTION-FIELD ELECTROSTATICS */
 456             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 457             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 458
 459             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 460
 461             /* Update potential sum for this i atom from the interaction with this j atom. */
 462             velec            = _mm256_and_ps(velec,cutoff_mask);
 463             velecsum         = _mm256_add_ps(velecsum,velec);
 464
 465             fscal            = felec;
 466
 467             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 468
 469             /* Calculate temporary vectorial force */
 470             tx               = _mm256_mul_ps(fscal,dx12);
 471             ty               = _mm256_mul_ps(fscal,dy12);
 472             tz               = _mm256_mul_ps(fscal,dz12);
 473
 474             /* Update vectorial force */
 475             fix1             = _mm256_add_ps(fix1,tx);
 476             fiy1             = _mm256_add_ps(fiy1,ty);
 477             fiz1             = _mm256_add_ps(fiz1,tz);
 478
 479             fjx2             = _mm256_add_ps(fjx2,tx);
 480             fjy2             = _mm256_add_ps(fjy2,ty);
 481             fjz2             = _mm256_add_ps(fjz2,tz);
 482
 483             }
 484
 485             /**************************
 486              * CALCULATE INTERACTIONS *
 487              **************************/
 488
 489             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 490             {
 491
 492             /* REACTION-FIELD ELECTROSTATICS */
 493             velec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_add_ps(rinv13,_mm256_mul_ps(krf,rsq13)),crf));
 494             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
 495
 496             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
 497
 498             /* Update potential sum for this i atom from the interaction with this j atom. */
 499             velec            = _mm256_and_ps(velec,cutoff_mask);
 500             velecsum         = _mm256_add_ps(velecsum,velec);
 501
 502             fscal            = felec;
 503
 504             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 505
 506             /* Calculate temporary vectorial force */
 507             tx               = _mm256_mul_ps(fscal,dx13);
 508             ty               = _mm256_mul_ps(fscal,dy13);
 509             tz               = _mm256_mul_ps(fscal,dz13);
 510
 511             /* Update vectorial force */
 512             fix1             = _mm256_add_ps(fix1,tx);
 513             fiy1             = _mm256_add_ps(fiy1,ty);
 514             fiz1             = _mm256_add_ps(fiz1,tz);
 515
 516             fjx3             = _mm256_add_ps(fjx3,tx);
 517             fjy3             = _mm256_add_ps(fjy3,ty);
 518             fjz3             = _mm256_add_ps(fjz3,tz);
 519
 520             }
 521
 522             /**************************
 523              * CALCULATE INTERACTIONS *
 524              **************************/
 525
 526             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 527             {
 528
 529             /* REACTION-FIELD ELECTROSTATICS */
 530             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
 531             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
 532
 533             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
 534
 535             /* Update potential sum for this i atom from the interaction with this j atom. */
 536             velec            = _mm256_and_ps(velec,cutoff_mask);
 537             velecsum         = _mm256_add_ps(velecsum,velec);
 538
 539             fscal            = felec;
 540
 541             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 542
 543             /* Calculate temporary vectorial force */
 544             tx               = _mm256_mul_ps(fscal,dx21);
 545             ty               = _mm256_mul_ps(fscal,dy21);
 546             tz               = _mm256_mul_ps(fscal,dz21);
 547
 548             /* Update vectorial force */
 549             fix2             = _mm256_add_ps(fix2,tx);
 550             fiy2             = _mm256_add_ps(fiy2,ty);
 551             fiz2             = _mm256_add_ps(fiz2,tz);
 552
 553             fjx1             = _mm256_add_ps(fjx1,tx);
 554             fjy1             = _mm256_add_ps(fjy1,ty);
 555             fjz1             = _mm256_add_ps(fjz1,tz);
 556
 557             }
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 564             {
 565
 566             /* REACTION-FIELD ELECTROSTATICS */
 567             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
 568             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
 569
 570             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
 571
 572             /* Update potential sum for this i atom from the interaction with this j atom. */
 573             velec            = _mm256_and_ps(velec,cutoff_mask);
 574             velecsum         = _mm256_add_ps(velecsum,velec);
 575
 576             fscal            = felec;
 577
 578             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 579
 580             /* Calculate temporary vectorial force */
 581             tx               = _mm256_mul_ps(fscal,dx22);
 582             ty               = _mm256_mul_ps(fscal,dy22);
 583             tz               = _mm256_mul_ps(fscal,dz22);
 584
 585             /* Update vectorial force */
 586             fix2             = _mm256_add_ps(fix2,tx);
 587             fiy2             = _mm256_add_ps(fiy2,ty);
 588             fiz2             = _mm256_add_ps(fiz2,tz);
 589
 590             fjx2             = _mm256_add_ps(fjx2,tx);
 591             fjy2             = _mm256_add_ps(fjy2,ty);
 592             fjz2             = _mm256_add_ps(fjz2,tz);
 593
 594             }
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 601             {
 602
 603             /* REACTION-FIELD ELECTROSTATICS */
 604             velec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_add_ps(rinv23,_mm256_mul_ps(krf,rsq23)),crf));
 605             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
 606
 607             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
 608
 609             /* Update potential sum for this i atom from the interaction with this j atom. */
 610             velec            = _mm256_and_ps(velec,cutoff_mask);
 611             velecsum         = _mm256_add_ps(velecsum,velec);
 612
 613             fscal            = felec;
 614
 615             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 616
 617             /* Calculate temporary vectorial force */
 618             tx               = _mm256_mul_ps(fscal,dx23);
 619             ty               = _mm256_mul_ps(fscal,dy23);
 620             tz               = _mm256_mul_ps(fscal,dz23);
 621
 622             /* Update vectorial force */
 623             fix2             = _mm256_add_ps(fix2,tx);
 624             fiy2             = _mm256_add_ps(fiy2,ty);
 625             fiz2             = _mm256_add_ps(fiz2,tz);
 626
 627             fjx3             = _mm256_add_ps(fjx3,tx);
 628             fjy3             = _mm256_add_ps(fjy3,ty);
 629             fjz3             = _mm256_add_ps(fjz3,tz);
 630
 631             }
 632
 633             /**************************
 634              * CALCULATE INTERACTIONS *
 635              **************************/
 636
 637             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 638             {
 639
 640             /* REACTION-FIELD ELECTROSTATICS */
 641             velec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_add_ps(rinv31,_mm256_mul_ps(krf,rsq31)),crf));
 642             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
 643
 644             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
 645
 646             /* Update potential sum for this i atom from the interaction with this j atom. */
 647             velec            = _mm256_and_ps(velec,cutoff_mask);
 648             velecsum         = _mm256_add_ps(velecsum,velec);
 649
 650             fscal            = felec;
 651
 652             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 653
 654             /* Calculate temporary vectorial force */
 655             tx               = _mm256_mul_ps(fscal,dx31);
 656             ty               = _mm256_mul_ps(fscal,dy31);
 657             tz               = _mm256_mul_ps(fscal,dz31);
 658
 659             /* Update vectorial force */
 660             fix3             = _mm256_add_ps(fix3,tx);
 661             fiy3             = _mm256_add_ps(fiy3,ty);
 662             fiz3             = _mm256_add_ps(fiz3,tz);
 663
 664             fjx1             = _mm256_add_ps(fjx1,tx);
 665             fjy1             = _mm256_add_ps(fjy1,ty);
 666             fjz1             = _mm256_add_ps(fjz1,tz);
 667
 668             }
 669
 670             /**************************
 671              * CALCULATE INTERACTIONS *
 672              **************************/
 673
 674             if (gmx_mm256_any_lt(rsq32,rcutoff2))
 675             {
 676
 677             /* REACTION-FIELD ELECTROSTATICS */
 678             velec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_add_ps(rinv32,_mm256_mul_ps(krf,rsq32)),crf));
 679             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
 680
 681             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
 682
 683             /* Update potential sum for this i atom from the interaction with this j atom. */
 684             velec            = _mm256_and_ps(velec,cutoff_mask);
 685             velecsum         = _mm256_add_ps(velecsum,velec);
 686
 687             fscal            = felec;
 688
 689             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 690
 691             /* Calculate temporary vectorial force */
 692             tx               = _mm256_mul_ps(fscal,dx32);
 693             ty               = _mm256_mul_ps(fscal,dy32);
 694             tz               = _mm256_mul_ps(fscal,dz32);
 695
 696             /* Update vectorial force */
 697             fix3             = _mm256_add_ps(fix3,tx);
 698             fiy3             = _mm256_add_ps(fiy3,ty);
 699             fiz3             = _mm256_add_ps(fiz3,tz);
 700
 701             fjx2             = _mm256_add_ps(fjx2,tx);
 702             fjy2             = _mm256_add_ps(fjy2,ty);
 703             fjz2             = _mm256_add_ps(fjz2,tz);
 704
 705             }
 706
 707             /**************************
 708              * CALCULATE INTERACTIONS *
 709              **************************/
 710
 711             if (gmx_mm256_any_lt(rsq33,rcutoff2))
 712             {
 713
 714             /* REACTION-FIELD ELECTROSTATICS */
 715             velec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_add_ps(rinv33,_mm256_mul_ps(krf,rsq33)),crf));
 716             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
 717
 718             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
 719
 720             /* Update potential sum for this i atom from the interaction with this j atom. */
 721             velec            = _mm256_and_ps(velec,cutoff_mask);
 722             velecsum         = _mm256_add_ps(velecsum,velec);
 723
 724             fscal            = felec;
 725
 726             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 727
 728             /* Calculate temporary vectorial force */
 729             tx               = _mm256_mul_ps(fscal,dx33);
 730             ty               = _mm256_mul_ps(fscal,dy33);
 731             tz               = _mm256_mul_ps(fscal,dz33);
 732
 733             /* Update vectorial force */
 734             fix3             = _mm256_add_ps(fix3,tx);
 735             fiy3             = _mm256_add_ps(fiy3,ty);
 736             fiz3             = _mm256_add_ps(fiz3,tz);
 737
 738             fjx3             = _mm256_add_ps(fjx3,tx);
 739             fjy3             = _mm256_add_ps(fjy3,ty);
 740             fjz3             = _mm256_add_ps(fjz3,tz);
 741
 742             }
 743
 744             fjptrA             = f+j_coord_offsetA;
 745             fjptrB             = f+j_coord_offsetB;
 746             fjptrC             = f+j_coord_offsetC;
 747             fjptrD             = f+j_coord_offsetD;
 748             fjptrE             = f+j_coord_offsetE;
 749             fjptrF             = f+j_coord_offsetF;
 750             fjptrG             = f+j_coord_offsetG;
 751             fjptrH             = f+j_coord_offsetH;
 752
 753             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 754                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 755                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 756
 757             /* Inner loop uses 387 flops */
 758         }
 759
 760         if(jidx<j_index_end)
 761         {
 762
 763             /* Get j neighbor index, and coordinate index */
 764             jnrlistA         = jjnr[jidx];
 765             jnrlistB         = jjnr[jidx+1];
 766             jnrlistC         = jjnr[jidx+2];
 767             jnrlistD         = jjnr[jidx+3];
 768             jnrlistE         = jjnr[jidx+4];
 769             jnrlistF         = jjnr[jidx+5];
 770             jnrlistG         = jjnr[jidx+6];
 771             jnrlistH         = jjnr[jidx+7];
 772             /* Sign of each element will be negative for non-real atoms.
 773              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 774              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 775              */
 776             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 777                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 778
 779             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 780             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 781             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 782             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 783             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 784             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 785             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 786             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 787             j_coord_offsetA  = DIM*jnrA;
 788             j_coord_offsetB  = DIM*jnrB;
 789             j_coord_offsetC  = DIM*jnrC;
 790             j_coord_offsetD  = DIM*jnrD;
 791             j_coord_offsetE  = DIM*jnrE;
 792             j_coord_offsetF  = DIM*jnrF;
 793             j_coord_offsetG  = DIM*jnrG;
 794             j_coord_offsetH  = DIM*jnrH;
 795
 796             /* load j atom coordinates */
 797             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 798                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 799                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 800                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 801                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 802                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 803
 804             /* Calculate displacement vector */
 805             dx00             = _mm256_sub_ps(ix0,jx0);
 806             dy00             = _mm256_sub_ps(iy0,jy0);
 807             dz00             = _mm256_sub_ps(iz0,jz0);
 808             dx11             = _mm256_sub_ps(ix1,jx1);
 809             dy11             = _mm256_sub_ps(iy1,jy1);
 810             dz11             = _mm256_sub_ps(iz1,jz1);
 811             dx12             = _mm256_sub_ps(ix1,jx2);
 812             dy12             = _mm256_sub_ps(iy1,jy2);
 813             dz12             = _mm256_sub_ps(iz1,jz2);
 814             dx13             = _mm256_sub_ps(ix1,jx3);
 815             dy13             = _mm256_sub_ps(iy1,jy3);
 816             dz13             = _mm256_sub_ps(iz1,jz3);
 817             dx21             = _mm256_sub_ps(ix2,jx1);
 818             dy21             = _mm256_sub_ps(iy2,jy1);
 819             dz21             = _mm256_sub_ps(iz2,jz1);
 820             dx22             = _mm256_sub_ps(ix2,jx2);
 821             dy22             = _mm256_sub_ps(iy2,jy2);
 822             dz22             = _mm256_sub_ps(iz2,jz2);
 823             dx23             = _mm256_sub_ps(ix2,jx3);
 824             dy23             = _mm256_sub_ps(iy2,jy3);
 825             dz23             = _mm256_sub_ps(iz2,jz3);
 826             dx31             = _mm256_sub_ps(ix3,jx1);
 827             dy31             = _mm256_sub_ps(iy3,jy1);
 828             dz31             = _mm256_sub_ps(iz3,jz1);
 829             dx32             = _mm256_sub_ps(ix3,jx2);
 830             dy32             = _mm256_sub_ps(iy3,jy2);
 831             dz32             = _mm256_sub_ps(iz3,jz2);
 832             dx33             = _mm256_sub_ps(ix3,jx3);
 833             dy33             = _mm256_sub_ps(iy3,jy3);
 834             dz33             = _mm256_sub_ps(iz3,jz3);
 835
 836             /* Calculate squared distance and things based on it */
 837             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 838             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 839             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 840             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 841             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 842             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 843             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 844             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 845             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 846             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 847
 848             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 849             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 850             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 851             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 852             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 853             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 854             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 855             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 856             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 857             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 858
 859             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 860             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 861             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
 862             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 863             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 864             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
 865             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
 866             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
 867             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
 868
 869             fjx0             = _mm256_setzero_ps();
 870             fjy0             = _mm256_setzero_ps();
 871             fjz0             = _mm256_setzero_ps();
 872             fjx1             = _mm256_setzero_ps();
 873             fjy1             = _mm256_setzero_ps();
 874             fjz1             = _mm256_setzero_ps();
 875             fjx2             = _mm256_setzero_ps();
 876             fjy2             = _mm256_setzero_ps();
 877             fjz2             = _mm256_setzero_ps();
 878             fjx3             = _mm256_setzero_ps();
 879             fjy3             = _mm256_setzero_ps();
 880             fjz3             = _mm256_setzero_ps();
 881
 882             /**************************
 883              * CALCULATE INTERACTIONS *
 884              **************************/
 885
 886             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 887             {
 888
 889             r00              = _mm256_mul_ps(rsq00,rinv00);
 890             r00              = _mm256_andnot_ps(dummy_mask,r00);
 891
 892             /* Calculate table index by multiplying r with table scale and truncate to integer */
 893             rt               = _mm256_mul_ps(r00,vftabscale);
 894             vfitab           = _mm256_cvttps_epi32(rt);
 895             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 896             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 897             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 898             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 899             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 900             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 901
 902             /* CUBIC SPLINE TABLE DISPERSION */
 903             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 904                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 905             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 906                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 907             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 908                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 909             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 910                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 911             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 912             Heps             = _mm256_mul_ps(vfeps,H);
 913             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 914             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 915             vvdw6            = _mm256_mul_ps(c6_00,VV);
 916             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 917             fvdw6            = _mm256_mul_ps(c6_00,FF);
 918
 919             /* CUBIC SPLINE TABLE REPULSION */
 920             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 921             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 922             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 923                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 924             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 925                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 926             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 927                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 928             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 929                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 930             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 931             Heps             = _mm256_mul_ps(vfeps,H);
 932             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 933             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 934             vvdw12           = _mm256_mul_ps(c12_00,VV);
 935             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 936             fvdw12           = _mm256_mul_ps(c12_00,FF);
 937             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 938             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 939
 940             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 941
 942             /* Update potential sum for this i atom from the interaction with this j atom. */
 943             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 944             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 945             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 946
 947             fscal            = fvdw;
 948
 949             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 950
 951             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 952
 953             /* Calculate temporary vectorial force */
 954             tx               = _mm256_mul_ps(fscal,dx00);
 955             ty               = _mm256_mul_ps(fscal,dy00);
 956             tz               = _mm256_mul_ps(fscal,dz00);
 957
 958             /* Update vectorial force */
 959             fix0             = _mm256_add_ps(fix0,tx);
 960             fiy0             = _mm256_add_ps(fiy0,ty);
 961             fiz0             = _mm256_add_ps(fiz0,tz);
 962
 963             fjx0             = _mm256_add_ps(fjx0,tx);
 964             fjy0             = _mm256_add_ps(fjy0,ty);
 965             fjz0             = _mm256_add_ps(fjz0,tz);
 966
 967             }
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 974             {
 975
 976             /* REACTION-FIELD ELECTROSTATICS */
 977             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 978             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 979
 980             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 981
 982             /* Update potential sum for this i atom from the interaction with this j atom. */
 983             velec            = _mm256_and_ps(velec,cutoff_mask);
 984             velec            = _mm256_andnot_ps(dummy_mask,velec);
 985             velecsum         = _mm256_add_ps(velecsum,velec);
 986
 987             fscal            = felec;
 988
 989             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 990
 991             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 992
 993             /* Calculate temporary vectorial force */
 994             tx               = _mm256_mul_ps(fscal,dx11);
 995             ty               = _mm256_mul_ps(fscal,dy11);
 996             tz               = _mm256_mul_ps(fscal,dz11);
 997
 998             /* Update vectorial force */
 999             fix1             = _mm256_add_ps(fix1,tx);
1000             fiy1             = _mm256_add_ps(fiy1,ty);
1001             fiz1             = _mm256_add_ps(fiz1,tz);
1002
1003             fjx1             = _mm256_add_ps(fjx1,tx);
1004             fjy1             = _mm256_add_ps(fjy1,ty);
1005             fjz1             = _mm256_add_ps(fjz1,tz);
1006
1007             }
1008
1009             /**************************
1010              * CALCULATE INTERACTIONS *
1011              **************************/
1012
1013             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1014             {
1015
1016             /* REACTION-FIELD ELECTROSTATICS */
1017             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
1018             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1019
1020             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1021
1022             /* Update potential sum for this i atom from the interaction with this j atom. */
1023             velec            = _mm256_and_ps(velec,cutoff_mask);
1024             velec            = _mm256_andnot_ps(dummy_mask,velec);
1025             velecsum         = _mm256_add_ps(velecsum,velec);
1026
1027             fscal            = felec;
1028
1029             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1030
1031             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1032
1033             /* Calculate temporary vectorial force */
1034             tx               = _mm256_mul_ps(fscal,dx12);
1035             ty               = _mm256_mul_ps(fscal,dy12);
1036             tz               = _mm256_mul_ps(fscal,dz12);
1037
1038             /* Update vectorial force */
1039             fix1             = _mm256_add_ps(fix1,tx);
1040             fiy1             = _mm256_add_ps(fiy1,ty);
1041             fiz1             = _mm256_add_ps(fiz1,tz);
1042
1043             fjx2             = _mm256_add_ps(fjx2,tx);
1044             fjy2             = _mm256_add_ps(fjy2,ty);
1045             fjz2             = _mm256_add_ps(fjz2,tz);
1046
1047             }
1048
1049             /**************************
1050              * CALCULATE INTERACTIONS *
1051              **************************/
1052
1053             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1054             {
1055
1056             /* REACTION-FIELD ELECTROSTATICS */
1057             velec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_add_ps(rinv13,_mm256_mul_ps(krf,rsq13)),crf));
1058             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
1059
1060             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
1061
1062             /* Update potential sum for this i atom from the interaction with this j atom. */
1063             velec            = _mm256_and_ps(velec,cutoff_mask);
1064             velec            = _mm256_andnot_ps(dummy_mask,velec);
1065             velecsum         = _mm256_add_ps(velecsum,velec);
1066
1067             fscal            = felec;
1068
1069             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1070
1071             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1072
1073             /* Calculate temporary vectorial force */
1074             tx               = _mm256_mul_ps(fscal,dx13);
1075             ty               = _mm256_mul_ps(fscal,dy13);
1076             tz               = _mm256_mul_ps(fscal,dz13);
1077
1078             /* Update vectorial force */
1079             fix1             = _mm256_add_ps(fix1,tx);
1080             fiy1             = _mm256_add_ps(fiy1,ty);
1081             fiz1             = _mm256_add_ps(fiz1,tz);
1082
1083             fjx3             = _mm256_add_ps(fjx3,tx);
1084             fjy3             = _mm256_add_ps(fjy3,ty);
1085             fjz3             = _mm256_add_ps(fjz3,tz);
1086
1087             }
1088
1089             /**************************
1090              * CALCULATE INTERACTIONS *
1091              **************************/
1092
1093             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1094             {
1095
1096             /* REACTION-FIELD ELECTROSTATICS */
1097             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
1098             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1099
1100             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1101
1102             /* Update potential sum for this i atom from the interaction with this j atom. */
1103             velec            = _mm256_and_ps(velec,cutoff_mask);
1104             velec            = _mm256_andnot_ps(dummy_mask,velec);
1105             velecsum         = _mm256_add_ps(velecsum,velec);
1106
1107             fscal            = felec;
1108
1109             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1110
1111             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1112
1113             /* Calculate temporary vectorial force */
1114             tx               = _mm256_mul_ps(fscal,dx21);
1115             ty               = _mm256_mul_ps(fscal,dy21);
1116             tz               = _mm256_mul_ps(fscal,dz21);
1117
1118             /* Update vectorial force */
1119             fix2             = _mm256_add_ps(fix2,tx);
1120             fiy2             = _mm256_add_ps(fiy2,ty);
1121             fiz2             = _mm256_add_ps(fiz2,tz);
1122
1123             fjx1             = _mm256_add_ps(fjx1,tx);
1124             fjy1             = _mm256_add_ps(fjy1,ty);
1125             fjz1             = _mm256_add_ps(fjz1,tz);
1126
1127             }
1128
1129             /**************************
1130              * CALCULATE INTERACTIONS *
1131              **************************/
1132
1133             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1134             {
1135
1136             /* REACTION-FIELD ELECTROSTATICS */
1137             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
1138             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1139
1140             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1141
1142             /* Update potential sum for this i atom from the interaction with this j atom. */
1143             velec            = _mm256_and_ps(velec,cutoff_mask);
1144             velec            = _mm256_andnot_ps(dummy_mask,velec);
1145             velecsum         = _mm256_add_ps(velecsum,velec);
1146
1147             fscal            = felec;
1148
1149             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1150
1151             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1152
1153             /* Calculate temporary vectorial force */
1154             tx               = _mm256_mul_ps(fscal,dx22);
1155             ty               = _mm256_mul_ps(fscal,dy22);
1156             tz               = _mm256_mul_ps(fscal,dz22);
1157
1158             /* Update vectorial force */
1159             fix2             = _mm256_add_ps(fix2,tx);
1160             fiy2             = _mm256_add_ps(fiy2,ty);
1161             fiz2             = _mm256_add_ps(fiz2,tz);
1162
1163             fjx2             = _mm256_add_ps(fjx2,tx);
1164             fjy2             = _mm256_add_ps(fjy2,ty);
1165             fjz2             = _mm256_add_ps(fjz2,tz);
1166
1167             }
1168
1169             /**************************
1170              * CALCULATE INTERACTIONS *
1171              **************************/
1172
1173             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1174             {
1175
1176             /* REACTION-FIELD ELECTROSTATICS */
1177             velec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_add_ps(rinv23,_mm256_mul_ps(krf,rsq23)),crf));
1178             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
1179
1180             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
1181
1182             /* Update potential sum for this i atom from the interaction with this j atom. */
1183             velec            = _mm256_and_ps(velec,cutoff_mask);
1184             velec            = _mm256_andnot_ps(dummy_mask,velec);
1185             velecsum         = _mm256_add_ps(velecsum,velec);
1186
1187             fscal            = felec;
1188
1189             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1190
1191             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1192
1193             /* Calculate temporary vectorial force */
1194             tx               = _mm256_mul_ps(fscal,dx23);
1195             ty               = _mm256_mul_ps(fscal,dy23);
1196             tz               = _mm256_mul_ps(fscal,dz23);
1197
1198             /* Update vectorial force */
1199             fix2             = _mm256_add_ps(fix2,tx);
1200             fiy2             = _mm256_add_ps(fiy2,ty);
1201             fiz2             = _mm256_add_ps(fiz2,tz);
1202
1203             fjx3             = _mm256_add_ps(fjx3,tx);
1204             fjy3             = _mm256_add_ps(fjy3,ty);
1205             fjz3             = _mm256_add_ps(fjz3,tz);
1206
1207             }
1208
1209             /**************************
1210              * CALCULATE INTERACTIONS *
1211              **************************/
1212
1213             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1214             {
1215
1216             /* REACTION-FIELD ELECTROSTATICS */
1217             velec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_add_ps(rinv31,_mm256_mul_ps(krf,rsq31)),crf));
1218             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
1219
1220             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
1221
1222             /* Update potential sum for this i atom from the interaction with this j atom. */
1223             velec            = _mm256_and_ps(velec,cutoff_mask);
1224             velec            = _mm256_andnot_ps(dummy_mask,velec);
1225             velecsum         = _mm256_add_ps(velecsum,velec);
1226
1227             fscal            = felec;
1228
1229             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1230
1231             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1232
1233             /* Calculate temporary vectorial force */
1234             tx               = _mm256_mul_ps(fscal,dx31);
1235             ty               = _mm256_mul_ps(fscal,dy31);
1236             tz               = _mm256_mul_ps(fscal,dz31);
1237
1238             /* Update vectorial force */
1239             fix3             = _mm256_add_ps(fix3,tx);
1240             fiy3             = _mm256_add_ps(fiy3,ty);
1241             fiz3             = _mm256_add_ps(fiz3,tz);
1242
1243             fjx1             = _mm256_add_ps(fjx1,tx);
1244             fjy1             = _mm256_add_ps(fjy1,ty);
1245             fjz1             = _mm256_add_ps(fjz1,tz);
1246
1247             }
1248
1249             /**************************
1250              * CALCULATE INTERACTIONS *
1251              **************************/
1252
1253             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1254             {
1255
1256             /* REACTION-FIELD ELECTROSTATICS */
1257             velec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_add_ps(rinv32,_mm256_mul_ps(krf,rsq32)),crf));
1258             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
1259
1260             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
1261
1262             /* Update potential sum for this i atom from the interaction with this j atom. */
1263             velec            = _mm256_and_ps(velec,cutoff_mask);
1264             velec            = _mm256_andnot_ps(dummy_mask,velec);
1265             velecsum         = _mm256_add_ps(velecsum,velec);
1266
1267             fscal            = felec;
1268
1269             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1270
1271             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1272
1273             /* Calculate temporary vectorial force */
1274             tx               = _mm256_mul_ps(fscal,dx32);
1275             ty               = _mm256_mul_ps(fscal,dy32);
1276             tz               = _mm256_mul_ps(fscal,dz32);
1277
1278             /* Update vectorial force */
1279             fix3             = _mm256_add_ps(fix3,tx);
1280             fiy3             = _mm256_add_ps(fiy3,ty);
1281             fiz3             = _mm256_add_ps(fiz3,tz);
1282
1283             fjx2             = _mm256_add_ps(fjx2,tx);
1284             fjy2             = _mm256_add_ps(fjy2,ty);
1285             fjz2             = _mm256_add_ps(fjz2,tz);
1286
1287             }
1288
1289             /**************************
1290              * CALCULATE INTERACTIONS *
1291              **************************/
1292
1293             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1294             {
1295
1296             /* REACTION-FIELD ELECTROSTATICS */
1297             velec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_add_ps(rinv33,_mm256_mul_ps(krf,rsq33)),crf));
1298             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
1299
1300             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
1301
1302             /* Update potential sum for this i atom from the interaction with this j atom. */
1303             velec            = _mm256_and_ps(velec,cutoff_mask);
1304             velec            = _mm256_andnot_ps(dummy_mask,velec);
1305             velecsum         = _mm256_add_ps(velecsum,velec);
1306
1307             fscal            = felec;
1308
1309             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1310
1311             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1312
1313             /* Calculate temporary vectorial force */
1314             tx               = _mm256_mul_ps(fscal,dx33);
1315             ty               = _mm256_mul_ps(fscal,dy33);
1316             tz               = _mm256_mul_ps(fscal,dz33);
1317
1318             /* Update vectorial force */
1319             fix3             = _mm256_add_ps(fix3,tx);
1320             fiy3             = _mm256_add_ps(fiy3,ty);
1321             fiz3             = _mm256_add_ps(fiz3,tz);
1322
1323             fjx3             = _mm256_add_ps(fjx3,tx);
1324             fjy3             = _mm256_add_ps(fjy3,ty);
1325             fjz3             = _mm256_add_ps(fjz3,tz);
1326
1327             }
1328
1329             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1330             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1331             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1332             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1333             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1334             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1335             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1336             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1337
1338             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1339                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1340                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1341
1342             /* Inner loop uses 388 flops */
1343         }
1344
1345         /* End of innermost loop */
1346
1347         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1348                                                  f+i_coord_offset,fshift+i_shift_offset);
1349
1350         ggid                        = gid[iidx];
1351         /* Update potential energies */
1352         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1353         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1354
1355         /* Increment number of inner iterations */
1356         inneriter                  += j_index_end - j_index_start;
1357
1358         /* Outer loop uses 26 flops */
1359     }
1360
1361     /* Increment number of outer iterations */
1362     outeriter        += nri;
1363
1364     /* Update outer/inner flops */
1365
1366     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*388);
1367 }
1368 /*
1369  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_F_avx_256_single
1370  * Electrostatics interaction: ReactionField
1371  * VdW interaction:            CubicSplineTable
1372  * Geometry:                   Water4-Water4
1373  * Calculate force/pot:        Force
1374  */
1375 void
1376 nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_F_avx_256_single
1377                     (t_nblist * gmx_restrict                nlist,
1378                      rvec * gmx_restrict                    xx,
1379                      rvec * gmx_restrict                    ff,
1380                      t_forcerec * gmx_restrict              fr,
1381                      t_mdatoms * gmx_restrict               mdatoms,
1382                      nb_kernel_data_t * gmx_restrict        kernel_data,
1383                      t_nrnb * gmx_restrict                  nrnb)
1384 {
1385     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1386      * just 0 for non-waters.
1387      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1388      * jnr indices corresponding to data put in the four positions in the SIMD register.
1389      */
1390     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1391     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1392     int              jnrA,jnrB,jnrC,jnrD;
1393     int              jnrE,jnrF,jnrG,jnrH;
1394     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1395     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1396     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1397     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1398     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1399     real             rcutoff_scalar;
1400     real             *shiftvec,*fshift,*x,*f;
1401     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1402     real             scratch[4*DIM];
1403     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1404     real *           vdwioffsetptr0;
1405     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1406     real *           vdwioffsetptr1;
1407     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1408     real *           vdwioffsetptr2;
1409     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1410     real *           vdwioffsetptr3;
1411     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1412     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1413     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1414     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1415     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1416     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1417     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1418     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1419     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1420     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1421     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1422     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1423     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1424     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1425     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1426     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1427     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1428     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1429     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1430     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1431     real             *charge;
1432     int              nvdwtype;
1433     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1434     int              *vdwtype;
1435     real             *vdwparam;
1436     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1437     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1438     __m256i          vfitab;
1439     __m128i          vfitab_lo,vfitab_hi;
1440     __m128i          ifour       = _mm_set1_epi32(4);
1441     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1442     real             *vftab;
1443     __m256           dummy_mask,cutoff_mask;
1444     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1445     __m256           one     = _mm256_set1_ps(1.0);
1446     __m256           two     = _mm256_set1_ps(2.0);
1447     x                = xx[0];
1448     f                = ff[0];
1449
1450     nri              = nlist->nri;
1451     iinr             = nlist->iinr;
1452     jindex           = nlist->jindex;
1453     jjnr             = nlist->jjnr;
1454     shiftidx         = nlist->shift;
1455     gid              = nlist->gid;
1456     shiftvec         = fr->shift_vec[0];
1457     fshift           = fr->fshift[0];
1458     facel            = _mm256_set1_ps(fr->epsfac);
1459     charge           = mdatoms->chargeA;
1460     krf              = _mm256_set1_ps(fr->ic->k_rf);
1461     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
1462     crf              = _mm256_set1_ps(fr->ic->c_rf);
1463     nvdwtype         = fr->ntype;
1464     vdwparam         = fr->nbfp;
1465     vdwtype          = mdatoms->typeA;
1466
1467     vftab            = kernel_data->table_vdw->data;
1468     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
1469
1470     /* Setup water-specific parameters */
1471     inr              = nlist->iinr[0];
1472     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1473     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1474     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1475     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1476
1477     jq1              = _mm256_set1_ps(charge[inr+1]);
1478     jq2              = _mm256_set1_ps(charge[inr+2]);
1479     jq3              = _mm256_set1_ps(charge[inr+3]);
1480     vdwjidx0A        = 2*vdwtype[inr+0];
1481     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1482     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1483     qq11             = _mm256_mul_ps(iq1,jq1);
1484     qq12             = _mm256_mul_ps(iq1,jq2);
1485     qq13             = _mm256_mul_ps(iq1,jq3);
1486     qq21             = _mm256_mul_ps(iq2,jq1);
1487     qq22             = _mm256_mul_ps(iq2,jq2);
1488     qq23             = _mm256_mul_ps(iq2,jq3);
1489     qq31             = _mm256_mul_ps(iq3,jq1);
1490     qq32             = _mm256_mul_ps(iq3,jq2);
1491     qq33             = _mm256_mul_ps(iq3,jq3);
1492
1493     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1494     rcutoff_scalar   = fr->rcoulomb;
1495     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1496     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1497
1498     /* Avoid stupid compiler warnings */
1499     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1500     j_coord_offsetA = 0;
1501     j_coord_offsetB = 0;
1502     j_coord_offsetC = 0;
1503     j_coord_offsetD = 0;
1504     j_coord_offsetE = 0;
1505     j_coord_offsetF = 0;
1506     j_coord_offsetG = 0;
1507     j_coord_offsetH = 0;
1508
1509     outeriter        = 0;
1510     inneriter        = 0;
1511
1512     for(iidx=0;iidx<4*DIM;iidx++)
1513     {
1514         scratch[iidx] = 0.0;
1515     }
1516
1517     /* Start outer loop over neighborlists */
1518     for(iidx=0; iidx<nri; iidx++)
1519     {
1520         /* Load shift vector for this list */
1521         i_shift_offset   = DIM*shiftidx[iidx];
1522
1523         /* Load limits for loop over neighbors */
1524         j_index_start    = jindex[iidx];
1525         j_index_end      = jindex[iidx+1];
1526
1527         /* Get outer coordinate index */
1528         inr              = iinr[iidx];
1529         i_coord_offset   = DIM*inr;
1530
1531         /* Load i particle coords and add shift vector */
1532         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1533                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1534
1535         fix0             = _mm256_setzero_ps();
1536         fiy0             = _mm256_setzero_ps();
1537         fiz0             = _mm256_setzero_ps();
1538         fix1             = _mm256_setzero_ps();
1539         fiy1             = _mm256_setzero_ps();
1540         fiz1             = _mm256_setzero_ps();
1541         fix2             = _mm256_setzero_ps();
1542         fiy2             = _mm256_setzero_ps();
1543         fiz2             = _mm256_setzero_ps();
1544         fix3             = _mm256_setzero_ps();
1545         fiy3             = _mm256_setzero_ps();
1546         fiz3             = _mm256_setzero_ps();
1547
1548         /* Start inner kernel loop */
1549         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1550         {
1551
1552             /* Get j neighbor index, and coordinate index */
1553             jnrA             = jjnr[jidx];
1554             jnrB             = jjnr[jidx+1];
1555             jnrC             = jjnr[jidx+2];
1556             jnrD             = jjnr[jidx+3];
1557             jnrE             = jjnr[jidx+4];
1558             jnrF             = jjnr[jidx+5];
1559             jnrG             = jjnr[jidx+6];
1560             jnrH             = jjnr[jidx+7];
1561             j_coord_offsetA  = DIM*jnrA;
1562             j_coord_offsetB  = DIM*jnrB;
1563             j_coord_offsetC  = DIM*jnrC;
1564             j_coord_offsetD  = DIM*jnrD;
1565             j_coord_offsetE  = DIM*jnrE;
1566             j_coord_offsetF  = DIM*jnrF;
1567             j_coord_offsetG  = DIM*jnrG;
1568             j_coord_offsetH  = DIM*jnrH;
1569
1570             /* load j atom coordinates */
1571             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1572                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1573                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1574                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1575                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1576                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1577
1578             /* Calculate displacement vector */
1579             dx00             = _mm256_sub_ps(ix0,jx0);
1580             dy00             = _mm256_sub_ps(iy0,jy0);
1581             dz00             = _mm256_sub_ps(iz0,jz0);
1582             dx11             = _mm256_sub_ps(ix1,jx1);
1583             dy11             = _mm256_sub_ps(iy1,jy1);
1584             dz11             = _mm256_sub_ps(iz1,jz1);
1585             dx12             = _mm256_sub_ps(ix1,jx2);
1586             dy12             = _mm256_sub_ps(iy1,jy2);
1587             dz12             = _mm256_sub_ps(iz1,jz2);
1588             dx13             = _mm256_sub_ps(ix1,jx3);
1589             dy13             = _mm256_sub_ps(iy1,jy3);
1590             dz13             = _mm256_sub_ps(iz1,jz3);
1591             dx21             = _mm256_sub_ps(ix2,jx1);
1592             dy21             = _mm256_sub_ps(iy2,jy1);
1593             dz21             = _mm256_sub_ps(iz2,jz1);
1594             dx22             = _mm256_sub_ps(ix2,jx2);
1595             dy22             = _mm256_sub_ps(iy2,jy2);
1596             dz22             = _mm256_sub_ps(iz2,jz2);
1597             dx23             = _mm256_sub_ps(ix2,jx3);
1598             dy23             = _mm256_sub_ps(iy2,jy3);
1599             dz23             = _mm256_sub_ps(iz2,jz3);
1600             dx31             = _mm256_sub_ps(ix3,jx1);
1601             dy31             = _mm256_sub_ps(iy3,jy1);
1602             dz31             = _mm256_sub_ps(iz3,jz1);
1603             dx32             = _mm256_sub_ps(ix3,jx2);
1604             dy32             = _mm256_sub_ps(iy3,jy2);
1605             dz32             = _mm256_sub_ps(iz3,jz2);
1606             dx33             = _mm256_sub_ps(ix3,jx3);
1607             dy33             = _mm256_sub_ps(iy3,jy3);
1608             dz33             = _mm256_sub_ps(iz3,jz3);
1609
1610             /* Calculate squared distance and things based on it */
1611             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1612             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1613             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1614             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1615             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1616             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1617             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1618             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1619             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1620             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1621
1622             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1623             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1624             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1625             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
1626             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1627             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1628             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
1629             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
1630             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
1631             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
1632
1633             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1634             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1635             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
1636             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1637             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1638             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
1639             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
1640             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
1641             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
1642
1643             fjx0             = _mm256_setzero_ps();
1644             fjy0             = _mm256_setzero_ps();
1645             fjz0             = _mm256_setzero_ps();
1646             fjx1             = _mm256_setzero_ps();
1647             fjy1             = _mm256_setzero_ps();
1648             fjz1             = _mm256_setzero_ps();
1649             fjx2             = _mm256_setzero_ps();
1650             fjy2             = _mm256_setzero_ps();
1651             fjz2             = _mm256_setzero_ps();
1652             fjx3             = _mm256_setzero_ps();
1653             fjy3             = _mm256_setzero_ps();
1654             fjz3             = _mm256_setzero_ps();
1655
1656             /**************************
1657              * CALCULATE INTERACTIONS *
1658              **************************/
1659
1660             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1661             {
1662
1663             r00              = _mm256_mul_ps(rsq00,rinv00);
1664
1665             /* Calculate table index by multiplying r with table scale and truncate to integer */
1666             rt               = _mm256_mul_ps(r00,vftabscale);
1667             vfitab           = _mm256_cvttps_epi32(rt);
1668             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1669             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1670             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1671             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1672             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1673             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1674
1675             /* CUBIC SPLINE TABLE DISPERSION */
1676             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1677                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1678             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1679                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1680             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1681                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1682             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1683                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1684             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1685             Heps             = _mm256_mul_ps(vfeps,H);
1686             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1687             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1688             fvdw6            = _mm256_mul_ps(c6_00,FF);
1689
1690             /* CUBIC SPLINE TABLE REPULSION */
1691             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1692             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1693             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1694                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1695             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1696                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1697             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1698                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1699             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1700                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1701             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1702             Heps             = _mm256_mul_ps(vfeps,H);
1703             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1704             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1705             fvdw12           = _mm256_mul_ps(c12_00,FF);
1706             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1707
1708             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1709
1710             fscal            = fvdw;
1711
1712             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1713
1714             /* Calculate temporary vectorial force */
1715             tx               = _mm256_mul_ps(fscal,dx00);
1716             ty               = _mm256_mul_ps(fscal,dy00);
1717             tz               = _mm256_mul_ps(fscal,dz00);
1718
1719             /* Update vectorial force */
1720             fix0             = _mm256_add_ps(fix0,tx);
1721             fiy0             = _mm256_add_ps(fiy0,ty);
1722             fiz0             = _mm256_add_ps(fiz0,tz);
1723
1724             fjx0             = _mm256_add_ps(fjx0,tx);
1725             fjy0             = _mm256_add_ps(fjy0,ty);
1726             fjz0             = _mm256_add_ps(fjz0,tz);
1727
1728             }
1729
1730             /**************************
1731              * CALCULATE INTERACTIONS *
1732              **************************/
1733
1734             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1735             {
1736
1737             /* REACTION-FIELD ELECTROSTATICS */
1738             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1739
1740             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1741
1742             fscal            = felec;
1743
1744             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1745
1746             /* Calculate temporary vectorial force */
1747             tx               = _mm256_mul_ps(fscal,dx11);
1748             ty               = _mm256_mul_ps(fscal,dy11);
1749             tz               = _mm256_mul_ps(fscal,dz11);
1750
1751             /* Update vectorial force */
1752             fix1             = _mm256_add_ps(fix1,tx);
1753             fiy1             = _mm256_add_ps(fiy1,ty);
1754             fiz1             = _mm256_add_ps(fiz1,tz);
1755
1756             fjx1             = _mm256_add_ps(fjx1,tx);
1757             fjy1             = _mm256_add_ps(fjy1,ty);
1758             fjz1             = _mm256_add_ps(fjz1,tz);
1759
1760             }
1761
1762             /**************************
1763              * CALCULATE INTERACTIONS *
1764              **************************/
1765
1766             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1767             {
1768
1769             /* REACTION-FIELD ELECTROSTATICS */
1770             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1771
1772             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1773
1774             fscal            = felec;
1775
1776             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1777
1778             /* Calculate temporary vectorial force */
1779             tx               = _mm256_mul_ps(fscal,dx12);
1780             ty               = _mm256_mul_ps(fscal,dy12);
1781             tz               = _mm256_mul_ps(fscal,dz12);
1782
1783             /* Update vectorial force */
1784             fix1             = _mm256_add_ps(fix1,tx);
1785             fiy1             = _mm256_add_ps(fiy1,ty);
1786             fiz1             = _mm256_add_ps(fiz1,tz);
1787
1788             fjx2             = _mm256_add_ps(fjx2,tx);
1789             fjy2             = _mm256_add_ps(fjy2,ty);
1790             fjz2             = _mm256_add_ps(fjz2,tz);
1791
1792             }
1793
1794             /**************************
1795              * CALCULATE INTERACTIONS *
1796              **************************/
1797
1798             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1799             {
1800
1801             /* REACTION-FIELD ELECTROSTATICS */
1802             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
1803
1804             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
1805
1806             fscal            = felec;
1807
1808             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1809
1810             /* Calculate temporary vectorial force */
1811             tx               = _mm256_mul_ps(fscal,dx13);
1812             ty               = _mm256_mul_ps(fscal,dy13);
1813             tz               = _mm256_mul_ps(fscal,dz13);
1814
1815             /* Update vectorial force */
1816             fix1             = _mm256_add_ps(fix1,tx);
1817             fiy1             = _mm256_add_ps(fiy1,ty);
1818             fiz1             = _mm256_add_ps(fiz1,tz);
1819
1820             fjx3             = _mm256_add_ps(fjx3,tx);
1821             fjy3             = _mm256_add_ps(fjy3,ty);
1822             fjz3             = _mm256_add_ps(fjz3,tz);
1823
1824             }
1825
1826             /**************************
1827              * CALCULATE INTERACTIONS *
1828              **************************/
1829
1830             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1831             {
1832
1833             /* REACTION-FIELD ELECTROSTATICS */
1834             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1835
1836             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1837
1838             fscal            = felec;
1839
1840             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1841
1842             /* Calculate temporary vectorial force */
1843             tx               = _mm256_mul_ps(fscal,dx21);
1844             ty               = _mm256_mul_ps(fscal,dy21);
1845             tz               = _mm256_mul_ps(fscal,dz21);
1846
1847             /* Update vectorial force */
1848             fix2             = _mm256_add_ps(fix2,tx);
1849             fiy2             = _mm256_add_ps(fiy2,ty);
1850             fiz2             = _mm256_add_ps(fiz2,tz);
1851
1852             fjx1             = _mm256_add_ps(fjx1,tx);
1853             fjy1             = _mm256_add_ps(fjy1,ty);
1854             fjz1             = _mm256_add_ps(fjz1,tz);
1855
1856             }
1857
1858             /**************************
1859              * CALCULATE INTERACTIONS *
1860              **************************/
1861
1862             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1863             {
1864
1865             /* REACTION-FIELD ELECTROSTATICS */
1866             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1867
1868             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1869
1870             fscal            = felec;
1871
1872             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1873
1874             /* Calculate temporary vectorial force */
1875             tx               = _mm256_mul_ps(fscal,dx22);
1876             ty               = _mm256_mul_ps(fscal,dy22);
1877             tz               = _mm256_mul_ps(fscal,dz22);
1878
1879             /* Update vectorial force */
1880             fix2             = _mm256_add_ps(fix2,tx);
1881             fiy2             = _mm256_add_ps(fiy2,ty);
1882             fiz2             = _mm256_add_ps(fiz2,tz);
1883
1884             fjx2             = _mm256_add_ps(fjx2,tx);
1885             fjy2             = _mm256_add_ps(fjy2,ty);
1886             fjz2             = _mm256_add_ps(fjz2,tz);
1887
1888             }
1889
1890             /**************************
1891              * CALCULATE INTERACTIONS *
1892              **************************/
1893
1894             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1895             {
1896
1897             /* REACTION-FIELD ELECTROSTATICS */
1898             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
1899
1900             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
1901
1902             fscal            = felec;
1903
1904             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1905
1906             /* Calculate temporary vectorial force */
1907             tx               = _mm256_mul_ps(fscal,dx23);
1908             ty               = _mm256_mul_ps(fscal,dy23);
1909             tz               = _mm256_mul_ps(fscal,dz23);
1910
1911             /* Update vectorial force */
1912             fix2             = _mm256_add_ps(fix2,tx);
1913             fiy2             = _mm256_add_ps(fiy2,ty);
1914             fiz2             = _mm256_add_ps(fiz2,tz);
1915
1916             fjx3             = _mm256_add_ps(fjx3,tx);
1917             fjy3             = _mm256_add_ps(fjy3,ty);
1918             fjz3             = _mm256_add_ps(fjz3,tz);
1919
1920             }
1921
1922             /**************************
1923              * CALCULATE INTERACTIONS *
1924              **************************/
1925
1926             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1927             {
1928
1929             /* REACTION-FIELD ELECTROSTATICS */
1930             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
1931
1932             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
1933
1934             fscal            = felec;
1935
1936             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1937
1938             /* Calculate temporary vectorial force */
1939             tx               = _mm256_mul_ps(fscal,dx31);
1940             ty               = _mm256_mul_ps(fscal,dy31);
1941             tz               = _mm256_mul_ps(fscal,dz31);
1942
1943             /* Update vectorial force */
1944             fix3             = _mm256_add_ps(fix3,tx);
1945             fiy3             = _mm256_add_ps(fiy3,ty);
1946             fiz3             = _mm256_add_ps(fiz3,tz);
1947
1948             fjx1             = _mm256_add_ps(fjx1,tx);
1949             fjy1             = _mm256_add_ps(fjy1,ty);
1950             fjz1             = _mm256_add_ps(fjz1,tz);
1951
1952             }
1953
1954             /**************************
1955              * CALCULATE INTERACTIONS *
1956              **************************/
1957
1958             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1959             {
1960
1961             /* REACTION-FIELD ELECTROSTATICS */
1962             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
1963
1964             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
1965
1966             fscal            = felec;
1967
1968             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1969
1970             /* Calculate temporary vectorial force */
1971             tx               = _mm256_mul_ps(fscal,dx32);
1972             ty               = _mm256_mul_ps(fscal,dy32);
1973             tz               = _mm256_mul_ps(fscal,dz32);
1974
1975             /* Update vectorial force */
1976             fix3             = _mm256_add_ps(fix3,tx);
1977             fiy3             = _mm256_add_ps(fiy3,ty);
1978             fiz3             = _mm256_add_ps(fiz3,tz);
1979
1980             fjx2             = _mm256_add_ps(fjx2,tx);
1981             fjy2             = _mm256_add_ps(fjy2,ty);
1982             fjz2             = _mm256_add_ps(fjz2,tz);
1983
1984             }
1985
1986             /**************************
1987              * CALCULATE INTERACTIONS *
1988              **************************/
1989
1990             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1991             {
1992
1993             /* REACTION-FIELD ELECTROSTATICS */
1994             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
1995
1996             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
1997
1998             fscal            = felec;
1999
2000             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2001
2002             /* Calculate temporary vectorial force */
2003             tx               = _mm256_mul_ps(fscal,dx33);
2004             ty               = _mm256_mul_ps(fscal,dy33);
2005             tz               = _mm256_mul_ps(fscal,dz33);
2006
2007             /* Update vectorial force */
2008             fix3             = _mm256_add_ps(fix3,tx);
2009             fiy3             = _mm256_add_ps(fiy3,ty);
2010             fiz3             = _mm256_add_ps(fiz3,tz);
2011
2012             fjx3             = _mm256_add_ps(fjx3,tx);
2013             fjy3             = _mm256_add_ps(fjy3,ty);
2014             fjz3             = _mm256_add_ps(fjz3,tz);
2015
2016             }
2017
2018             fjptrA             = f+j_coord_offsetA;
2019             fjptrB             = f+j_coord_offsetB;
2020             fjptrC             = f+j_coord_offsetC;
2021             fjptrD             = f+j_coord_offsetD;
2022             fjptrE             = f+j_coord_offsetE;
2023             fjptrF             = f+j_coord_offsetF;
2024             fjptrG             = f+j_coord_offsetG;
2025             fjptrH             = f+j_coord_offsetH;
2026
2027             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2028                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2029                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2030
2031             /* Inner loop uses 324 flops */
2032         }
2033
2034         if(jidx<j_index_end)
2035         {
2036
2037             /* Get j neighbor index, and coordinate index */
2038             jnrlistA         = jjnr[jidx];
2039             jnrlistB         = jjnr[jidx+1];
2040             jnrlistC         = jjnr[jidx+2];
2041             jnrlistD         = jjnr[jidx+3];
2042             jnrlistE         = jjnr[jidx+4];
2043             jnrlistF         = jjnr[jidx+5];
2044             jnrlistG         = jjnr[jidx+6];
2045             jnrlistH         = jjnr[jidx+7];
2046             /* Sign of each element will be negative for non-real atoms.
2047              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2048              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2049              */
2050             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2051                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2052
2053             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2054             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2055             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2056             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2057             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2058             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2059             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2060             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2061             j_coord_offsetA  = DIM*jnrA;
2062             j_coord_offsetB  = DIM*jnrB;
2063             j_coord_offsetC  = DIM*jnrC;
2064             j_coord_offsetD  = DIM*jnrD;
2065             j_coord_offsetE  = DIM*jnrE;
2066             j_coord_offsetF  = DIM*jnrF;
2067             j_coord_offsetG  = DIM*jnrG;
2068             j_coord_offsetH  = DIM*jnrH;
2069
2070             /* load j atom coordinates */
2071             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2072                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2073                                                  x+j_coord_offsetE,x+j_coord_offsetF,
2074                                                  x+j_coord_offsetG,x+j_coord_offsetH,
2075                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2076                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
2077
2078             /* Calculate displacement vector */
2079             dx00             = _mm256_sub_ps(ix0,jx0);
2080             dy00             = _mm256_sub_ps(iy0,jy0);
2081             dz00             = _mm256_sub_ps(iz0,jz0);
2082             dx11             = _mm256_sub_ps(ix1,jx1);
2083             dy11             = _mm256_sub_ps(iy1,jy1);
2084             dz11             = _mm256_sub_ps(iz1,jz1);
2085             dx12             = _mm256_sub_ps(ix1,jx2);
2086             dy12             = _mm256_sub_ps(iy1,jy2);
2087             dz12             = _mm256_sub_ps(iz1,jz2);
2088             dx13             = _mm256_sub_ps(ix1,jx3);
2089             dy13             = _mm256_sub_ps(iy1,jy3);
2090             dz13             = _mm256_sub_ps(iz1,jz3);
2091             dx21             = _mm256_sub_ps(ix2,jx1);
2092             dy21             = _mm256_sub_ps(iy2,jy1);
2093             dz21             = _mm256_sub_ps(iz2,jz1);
2094             dx22             = _mm256_sub_ps(ix2,jx2);
2095             dy22             = _mm256_sub_ps(iy2,jy2);
2096             dz22             = _mm256_sub_ps(iz2,jz2);
2097             dx23             = _mm256_sub_ps(ix2,jx3);
2098             dy23             = _mm256_sub_ps(iy2,jy3);
2099             dz23             = _mm256_sub_ps(iz2,jz3);
2100             dx31             = _mm256_sub_ps(ix3,jx1);
2101             dy31             = _mm256_sub_ps(iy3,jy1);
2102             dz31             = _mm256_sub_ps(iz3,jz1);
2103             dx32             = _mm256_sub_ps(ix3,jx2);
2104             dy32             = _mm256_sub_ps(iy3,jy2);
2105             dz32             = _mm256_sub_ps(iz3,jz2);
2106             dx33             = _mm256_sub_ps(ix3,jx3);
2107             dy33             = _mm256_sub_ps(iy3,jy3);
2108             dz33             = _mm256_sub_ps(iz3,jz3);
2109
2110             /* Calculate squared distance and things based on it */
2111             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2112             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2113             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2114             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
2115             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2116             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2117             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
2118             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
2119             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
2120             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
2121
2122             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
2123             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2124             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2125             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
2126             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2127             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2128             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
2129             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
2130             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
2131             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
2132
2133             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2134             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2135             rinvsq13         = _mm256_mul_ps(rinv13,rinv13);
2136             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2137             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2138             rinvsq23         = _mm256_mul_ps(rinv23,rinv23);
2139             rinvsq31         = _mm256_mul_ps(rinv31,rinv31);
2140             rinvsq32         = _mm256_mul_ps(rinv32,rinv32);
2141             rinvsq33         = _mm256_mul_ps(rinv33,rinv33);
2142
2143             fjx0             = _mm256_setzero_ps();
2144             fjy0             = _mm256_setzero_ps();
2145             fjz0             = _mm256_setzero_ps();
2146             fjx1             = _mm256_setzero_ps();
2147             fjy1             = _mm256_setzero_ps();
2148             fjz1             = _mm256_setzero_ps();
2149             fjx2             = _mm256_setzero_ps();
2150             fjy2             = _mm256_setzero_ps();
2151             fjz2             = _mm256_setzero_ps();
2152             fjx3             = _mm256_setzero_ps();
2153             fjy3             = _mm256_setzero_ps();
2154             fjz3             = _mm256_setzero_ps();
2155
2156             /**************************
2157              * CALCULATE INTERACTIONS *
2158              **************************/
2159
2160             if (gmx_mm256_any_lt(rsq00,rcutoff2))
2161             {
2162
2163             r00              = _mm256_mul_ps(rsq00,rinv00);
2164             r00              = _mm256_andnot_ps(dummy_mask,r00);
2165
2166             /* Calculate table index by multiplying r with table scale and truncate to integer */
2167             rt               = _mm256_mul_ps(r00,vftabscale);
2168             vfitab           = _mm256_cvttps_epi32(rt);
2169             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2170             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2171             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2172             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2173             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
2174             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
2175
2176             /* CUBIC SPLINE TABLE DISPERSION */
2177             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2178                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2179             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2180                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2181             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2182                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2183             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2184                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2185             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2186             Heps             = _mm256_mul_ps(vfeps,H);
2187             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2188             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2189             fvdw6            = _mm256_mul_ps(c6_00,FF);
2190
2191             /* CUBIC SPLINE TABLE REPULSION */
2192             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
2193             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
2194             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2195                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2196             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2197                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2198             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2199                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2200             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2201                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2202             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2203             Heps             = _mm256_mul_ps(vfeps,H);
2204             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2205             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2206             fvdw12           = _mm256_mul_ps(c12_00,FF);
2207             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
2208
2209             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
2210
2211             fscal            = fvdw;
2212
2213             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2214
2215             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2216
2217             /* Calculate temporary vectorial force */
2218             tx               = _mm256_mul_ps(fscal,dx00);
2219             ty               = _mm256_mul_ps(fscal,dy00);
2220             tz               = _mm256_mul_ps(fscal,dz00);
2221
2222             /* Update vectorial force */
2223             fix0             = _mm256_add_ps(fix0,tx);
2224             fiy0             = _mm256_add_ps(fiy0,ty);
2225             fiz0             = _mm256_add_ps(fiz0,tz);
2226
2227             fjx0             = _mm256_add_ps(fjx0,tx);
2228             fjy0             = _mm256_add_ps(fjy0,ty);
2229             fjz0             = _mm256_add_ps(fjz0,tz);
2230
2231             }
2232
2233             /**************************
2234              * CALCULATE INTERACTIONS *
2235              **************************/
2236
2237             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2238             {
2239
2240             /* REACTION-FIELD ELECTROSTATICS */
2241             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
2242
2243             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2244
2245             fscal            = felec;
2246
2247             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2248
2249             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2250
2251             /* Calculate temporary vectorial force */
2252             tx               = _mm256_mul_ps(fscal,dx11);
2253             ty               = _mm256_mul_ps(fscal,dy11);
2254             tz               = _mm256_mul_ps(fscal,dz11);
2255
2256             /* Update vectorial force */
2257             fix1             = _mm256_add_ps(fix1,tx);
2258             fiy1             = _mm256_add_ps(fiy1,ty);
2259             fiz1             = _mm256_add_ps(fiz1,tz);
2260
2261             fjx1             = _mm256_add_ps(fjx1,tx);
2262             fjy1             = _mm256_add_ps(fjy1,ty);
2263             fjz1             = _mm256_add_ps(fjz1,tz);
2264
2265             }
2266
2267             /**************************
2268              * CALCULATE INTERACTIONS *
2269              **************************/
2270
2271             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2272             {
2273
2274             /* REACTION-FIELD ELECTROSTATICS */
2275             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
2276
2277             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2278
2279             fscal            = felec;
2280
2281             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2282
2283             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2284
2285             /* Calculate temporary vectorial force */
2286             tx               = _mm256_mul_ps(fscal,dx12);
2287             ty               = _mm256_mul_ps(fscal,dy12);
2288             tz               = _mm256_mul_ps(fscal,dz12);
2289
2290             /* Update vectorial force */
2291             fix1             = _mm256_add_ps(fix1,tx);
2292             fiy1             = _mm256_add_ps(fiy1,ty);
2293             fiz1             = _mm256_add_ps(fiz1,tz);
2294
2295             fjx2             = _mm256_add_ps(fjx2,tx);
2296             fjy2             = _mm256_add_ps(fjy2,ty);
2297             fjz2             = _mm256_add_ps(fjz2,tz);
2298
2299             }
2300
2301             /**************************
2302              * CALCULATE INTERACTIONS *
2303              **************************/
2304
2305             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2306             {
2307
2308             /* REACTION-FIELD ELECTROSTATICS */
2309             felec            = _mm256_mul_ps(qq13,_mm256_sub_ps(_mm256_mul_ps(rinv13,rinvsq13),krf2));
2310
2311             cutoff_mask      = _mm256_cmp_ps(rsq13,rcutoff2,_CMP_LT_OQ);
2312
2313             fscal            = felec;
2314
2315             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2316
2317             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2318
2319             /* Calculate temporary vectorial force */
2320             tx               = _mm256_mul_ps(fscal,dx13);
2321             ty               = _mm256_mul_ps(fscal,dy13);
2322             tz               = _mm256_mul_ps(fscal,dz13);
2323
2324             /* Update vectorial force */
2325             fix1             = _mm256_add_ps(fix1,tx);
2326             fiy1             = _mm256_add_ps(fiy1,ty);
2327             fiz1             = _mm256_add_ps(fiz1,tz);
2328
2329             fjx3             = _mm256_add_ps(fjx3,tx);
2330             fjy3             = _mm256_add_ps(fjy3,ty);
2331             fjz3             = _mm256_add_ps(fjz3,tz);
2332
2333             }
2334
2335             /**************************
2336              * CALCULATE INTERACTIONS *
2337              **************************/
2338
2339             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2340             {
2341
2342             /* REACTION-FIELD ELECTROSTATICS */
2343             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
2344
2345             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2346
2347             fscal            = felec;
2348
2349             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2350
2351             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2352
2353             /* Calculate temporary vectorial force */
2354             tx               = _mm256_mul_ps(fscal,dx21);
2355             ty               = _mm256_mul_ps(fscal,dy21);
2356             tz               = _mm256_mul_ps(fscal,dz21);
2357
2358             /* Update vectorial force */
2359             fix2             = _mm256_add_ps(fix2,tx);
2360             fiy2             = _mm256_add_ps(fiy2,ty);
2361             fiz2             = _mm256_add_ps(fiz2,tz);
2362
2363             fjx1             = _mm256_add_ps(fjx1,tx);
2364             fjy1             = _mm256_add_ps(fjy1,ty);
2365             fjz1             = _mm256_add_ps(fjz1,tz);
2366
2367             }
2368
2369             /**************************
2370              * CALCULATE INTERACTIONS *
2371              **************************/
2372
2373             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2374             {
2375
2376             /* REACTION-FIELD ELECTROSTATICS */
2377             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
2378
2379             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2380
2381             fscal            = felec;
2382
2383             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2384
2385             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2386
2387             /* Calculate temporary vectorial force */
2388             tx               = _mm256_mul_ps(fscal,dx22);
2389             ty               = _mm256_mul_ps(fscal,dy22);
2390             tz               = _mm256_mul_ps(fscal,dz22);
2391
2392             /* Update vectorial force */
2393             fix2             = _mm256_add_ps(fix2,tx);
2394             fiy2             = _mm256_add_ps(fiy2,ty);
2395             fiz2             = _mm256_add_ps(fiz2,tz);
2396
2397             fjx2             = _mm256_add_ps(fjx2,tx);
2398             fjy2             = _mm256_add_ps(fjy2,ty);
2399             fjz2             = _mm256_add_ps(fjz2,tz);
2400
2401             }
2402
2403             /**************************
2404              * CALCULATE INTERACTIONS *
2405              **************************/
2406
2407             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2408             {
2409
2410             /* REACTION-FIELD ELECTROSTATICS */
2411             felec            = _mm256_mul_ps(qq23,_mm256_sub_ps(_mm256_mul_ps(rinv23,rinvsq23),krf2));
2412
2413             cutoff_mask      = _mm256_cmp_ps(rsq23,rcutoff2,_CMP_LT_OQ);
2414
2415             fscal            = felec;
2416
2417             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2418
2419             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2420
2421             /* Calculate temporary vectorial force */
2422             tx               = _mm256_mul_ps(fscal,dx23);
2423             ty               = _mm256_mul_ps(fscal,dy23);
2424             tz               = _mm256_mul_ps(fscal,dz23);
2425
2426             /* Update vectorial force */
2427             fix2             = _mm256_add_ps(fix2,tx);
2428             fiy2             = _mm256_add_ps(fiy2,ty);
2429             fiz2             = _mm256_add_ps(fiz2,tz);
2430
2431             fjx3             = _mm256_add_ps(fjx3,tx);
2432             fjy3             = _mm256_add_ps(fjy3,ty);
2433             fjz3             = _mm256_add_ps(fjz3,tz);
2434
2435             }
2436
2437             /**************************
2438              * CALCULATE INTERACTIONS *
2439              **************************/
2440
2441             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2442             {
2443
2444             /* REACTION-FIELD ELECTROSTATICS */
2445             felec            = _mm256_mul_ps(qq31,_mm256_sub_ps(_mm256_mul_ps(rinv31,rinvsq31),krf2));
2446
2447             cutoff_mask      = _mm256_cmp_ps(rsq31,rcutoff2,_CMP_LT_OQ);
2448
2449             fscal            = felec;
2450
2451             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2452
2453             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2454
2455             /* Calculate temporary vectorial force */
2456             tx               = _mm256_mul_ps(fscal,dx31);
2457             ty               = _mm256_mul_ps(fscal,dy31);
2458             tz               = _mm256_mul_ps(fscal,dz31);
2459
2460             /* Update vectorial force */
2461             fix3             = _mm256_add_ps(fix3,tx);
2462             fiy3             = _mm256_add_ps(fiy3,ty);
2463             fiz3             = _mm256_add_ps(fiz3,tz);
2464
2465             fjx1             = _mm256_add_ps(fjx1,tx);
2466             fjy1             = _mm256_add_ps(fjy1,ty);
2467             fjz1             = _mm256_add_ps(fjz1,tz);
2468
2469             }
2470
2471             /**************************
2472              * CALCULATE INTERACTIONS *
2473              **************************/
2474
2475             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2476             {
2477
2478             /* REACTION-FIELD ELECTROSTATICS */
2479             felec            = _mm256_mul_ps(qq32,_mm256_sub_ps(_mm256_mul_ps(rinv32,rinvsq32),krf2));
2480
2481             cutoff_mask      = _mm256_cmp_ps(rsq32,rcutoff2,_CMP_LT_OQ);
2482
2483             fscal            = felec;
2484
2485             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2486
2487             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2488
2489             /* Calculate temporary vectorial force */
2490             tx               = _mm256_mul_ps(fscal,dx32);
2491             ty               = _mm256_mul_ps(fscal,dy32);
2492             tz               = _mm256_mul_ps(fscal,dz32);
2493
2494             /* Update vectorial force */
2495             fix3             = _mm256_add_ps(fix3,tx);
2496             fiy3             = _mm256_add_ps(fiy3,ty);
2497             fiz3             = _mm256_add_ps(fiz3,tz);
2498
2499             fjx2             = _mm256_add_ps(fjx2,tx);
2500             fjy2             = _mm256_add_ps(fjy2,ty);
2501             fjz2             = _mm256_add_ps(fjz2,tz);
2502
2503             }
2504
2505             /**************************
2506              * CALCULATE INTERACTIONS *
2507              **************************/
2508
2509             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2510             {
2511
2512             /* REACTION-FIELD ELECTROSTATICS */
2513             felec            = _mm256_mul_ps(qq33,_mm256_sub_ps(_mm256_mul_ps(rinv33,rinvsq33),krf2));
2514
2515             cutoff_mask      = _mm256_cmp_ps(rsq33,rcutoff2,_CMP_LT_OQ);
2516
2517             fscal            = felec;
2518
2519             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2520
2521             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2522
2523             /* Calculate temporary vectorial force */
2524             tx               = _mm256_mul_ps(fscal,dx33);
2525             ty               = _mm256_mul_ps(fscal,dy33);
2526             tz               = _mm256_mul_ps(fscal,dz33);
2527
2528             /* Update vectorial force */
2529             fix3             = _mm256_add_ps(fix3,tx);
2530             fiy3             = _mm256_add_ps(fiy3,ty);
2531             fiz3             = _mm256_add_ps(fiz3,tz);
2532
2533             fjx3             = _mm256_add_ps(fjx3,tx);
2534             fjy3             = _mm256_add_ps(fjy3,ty);
2535             fjz3             = _mm256_add_ps(fjz3,tz);
2536
2537             }
2538
2539             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2540             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2541             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2542             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2543             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2544             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2545             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2546             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2547
2548             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2549                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2550                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2551
2552             /* Inner loop uses 325 flops */
2553         }
2554
2555         /* End of innermost loop */
2556
2557         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2558                                                  f+i_coord_offset,fshift+i_shift_offset);
2559
2560         /* Increment number of inner iterations */
2561         inneriter                  += j_index_end - j_index_start;
2562
2563         /* Outer loop uses 24 flops */
2564     }
2565
2566     /* Increment number of outer iterations */
2567     outeriter        += nri;
2568
2569     /* Update outer/inner flops */
2570
2571     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*325);
2572 }