src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwNone_GeomW3W3_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW3W3_VF_avx_256_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwNone_GeomW3W3_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  95     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  97     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 100     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 101     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 102     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 105     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     __m256i          vfitab;
 110     __m128i          vfitab_lo,vfitab_hi;
 111     __m128i          ifour       = _mm_set1_epi32(4);
 112     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 113     real             *vftab;
 114     __m256           dummy_mask,cutoff_mask;
 115     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 116     __m256           one     = _mm256_set1_ps(1.0);
 117     __m256           two     = _mm256_set1_ps(2.0);
 118     x                = xx[0];
 119     f                = ff[0];
 120
 121     nri              = nlist->nri;
 122     iinr             = nlist->iinr;
 123     jindex           = nlist->jindex;
 124     jjnr             = nlist->jjnr;
 125     shiftidx         = nlist->shift;
 126     gid              = nlist->gid;
 127     shiftvec         = fr->shift_vec[0];
 128     fshift           = fr->fshift[0];
 129     facel            = _mm256_set1_ps(fr->epsfac);
 130     charge           = mdatoms->chargeA;
 131
 132     vftab            = kernel_data->table_elec->data;
 133     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 138     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 139     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 140
 141     jq0              = _mm256_set1_ps(charge[inr+0]);
 142     jq1              = _mm256_set1_ps(charge[inr+1]);
 143     jq2              = _mm256_set1_ps(charge[inr+2]);
 144     qq00             = _mm256_mul_ps(iq0,jq0);
 145     qq01             = _mm256_mul_ps(iq0,jq1);
 146     qq02             = _mm256_mul_ps(iq0,jq2);
 147     qq10             = _mm256_mul_ps(iq1,jq0);
 148     qq11             = _mm256_mul_ps(iq1,jq1);
 149     qq12             = _mm256_mul_ps(iq1,jq2);
 150     qq20             = _mm256_mul_ps(iq2,jq0);
 151     qq21             = _mm256_mul_ps(iq2,jq1);
 152     qq22             = _mm256_mul_ps(iq2,jq2);
 153
 154     /* Avoid stupid compiler warnings */
 155     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 156     j_coord_offsetA = 0;
 157     j_coord_offsetB = 0;
 158     j_coord_offsetC = 0;
 159     j_coord_offsetD = 0;
 160     j_coord_offsetE = 0;
 161     j_coord_offsetF = 0;
 162     j_coord_offsetG = 0;
 163     j_coord_offsetH = 0;
 164
 165     outeriter        = 0;
 166     inneriter        = 0;
 167
 168     for(iidx=0;iidx<4*DIM;iidx++)
 169     {
 170         scratch[iidx] = 0.0;
 171     }
 172
 173     /* Start outer loop over neighborlists */
 174     for(iidx=0; iidx<nri; iidx++)
 175     {
 176         /* Load shift vector for this list */
 177         i_shift_offset   = DIM*shiftidx[iidx];
 178
 179         /* Load limits for loop over neighbors */
 180         j_index_start    = jindex[iidx];
 181         j_index_end      = jindex[iidx+1];
 182
 183         /* Get outer coordinate index */
 184         inr              = iinr[iidx];
 185         i_coord_offset   = DIM*inr;
 186
 187         /* Load i particle coords and add shift vector */
 188         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 189                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 190
 191         fix0             = _mm256_setzero_ps();
 192         fiy0             = _mm256_setzero_ps();
 193         fiz0             = _mm256_setzero_ps();
 194         fix1             = _mm256_setzero_ps();
 195         fiy1             = _mm256_setzero_ps();
 196         fiz1             = _mm256_setzero_ps();
 197         fix2             = _mm256_setzero_ps();
 198         fiy2             = _mm256_setzero_ps();
 199         fiz2             = _mm256_setzero_ps();
 200
 201         /* Reset potential sums */
 202         velecsum         = _mm256_setzero_ps();
 203
 204         /* Start inner kernel loop */
 205         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 206         {
 207
 208             /* Get j neighbor index, and coordinate index */
 209             jnrA             = jjnr[jidx];
 210             jnrB             = jjnr[jidx+1];
 211             jnrC             = jjnr[jidx+2];
 212             jnrD             = jjnr[jidx+3];
 213             jnrE             = jjnr[jidx+4];
 214             jnrF             = jjnr[jidx+5];
 215             jnrG             = jjnr[jidx+6];
 216             jnrH             = jjnr[jidx+7];
 217             j_coord_offsetA  = DIM*jnrA;
 218             j_coord_offsetB  = DIM*jnrB;
 219             j_coord_offsetC  = DIM*jnrC;
 220             j_coord_offsetD  = DIM*jnrD;
 221             j_coord_offsetE  = DIM*jnrE;
 222             j_coord_offsetF  = DIM*jnrF;
 223             j_coord_offsetG  = DIM*jnrG;
 224             j_coord_offsetH  = DIM*jnrH;
 225
 226             /* load j atom coordinates */
 227             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 228                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 229                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 230                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 231                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 232
 233             /* Calculate displacement vector */
 234             dx00             = _mm256_sub_ps(ix0,jx0);
 235             dy00             = _mm256_sub_ps(iy0,jy0);
 236             dz00             = _mm256_sub_ps(iz0,jz0);
 237             dx01             = _mm256_sub_ps(ix0,jx1);
 238             dy01             = _mm256_sub_ps(iy0,jy1);
 239             dz01             = _mm256_sub_ps(iz0,jz1);
 240             dx02             = _mm256_sub_ps(ix0,jx2);
 241             dy02             = _mm256_sub_ps(iy0,jy2);
 242             dz02             = _mm256_sub_ps(iz0,jz2);
 243             dx10             = _mm256_sub_ps(ix1,jx0);
 244             dy10             = _mm256_sub_ps(iy1,jy0);
 245             dz10             = _mm256_sub_ps(iz1,jz0);
 246             dx11             = _mm256_sub_ps(ix1,jx1);
 247             dy11             = _mm256_sub_ps(iy1,jy1);
 248             dz11             = _mm256_sub_ps(iz1,jz1);
 249             dx12             = _mm256_sub_ps(ix1,jx2);
 250             dy12             = _mm256_sub_ps(iy1,jy2);
 251             dz12             = _mm256_sub_ps(iz1,jz2);
 252             dx20             = _mm256_sub_ps(ix2,jx0);
 253             dy20             = _mm256_sub_ps(iy2,jy0);
 254             dz20             = _mm256_sub_ps(iz2,jz0);
 255             dx21             = _mm256_sub_ps(ix2,jx1);
 256             dy21             = _mm256_sub_ps(iy2,jy1);
 257             dz21             = _mm256_sub_ps(iz2,jz1);
 258             dx22             = _mm256_sub_ps(ix2,jx2);
 259             dy22             = _mm256_sub_ps(iy2,jy2);
 260             dz22             = _mm256_sub_ps(iz2,jz2);
 261
 262             /* Calculate squared distance and things based on it */
 263             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 264             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 265             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 266             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 267             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 268             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 269             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 270             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 271             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 272
 273             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 274             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 275             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 276             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 277             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 278             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 279             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 280             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 281             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 282
 283             fjx0             = _mm256_setzero_ps();
 284             fjy0             = _mm256_setzero_ps();
 285             fjz0             = _mm256_setzero_ps();
 286             fjx1             = _mm256_setzero_ps();
 287             fjy1             = _mm256_setzero_ps();
 288             fjz1             = _mm256_setzero_ps();
 289             fjx2             = _mm256_setzero_ps();
 290             fjy2             = _mm256_setzero_ps();
 291             fjz2             = _mm256_setzero_ps();
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             r00              = _mm256_mul_ps(rsq00,rinv00);
 298
 299             /* Calculate table index by multiplying r with table scale and truncate to integer */
 300             rt               = _mm256_mul_ps(r00,vftabscale);
 301             vfitab           = _mm256_cvttps_epi32(rt);
 302             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 303             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 304             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 305             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 306             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 307             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 308
 309             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 310             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 311                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 312             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 313                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 314             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 315                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 316             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 317                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 318             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 319             Heps             = _mm256_mul_ps(vfeps,H);
 320             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 321             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 322             velec            = _mm256_mul_ps(qq00,VV);
 323             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 324             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velecsum         = _mm256_add_ps(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331             /* Calculate temporary vectorial force */
 332             tx               = _mm256_mul_ps(fscal,dx00);
 333             ty               = _mm256_mul_ps(fscal,dy00);
 334             tz               = _mm256_mul_ps(fscal,dz00);
 335
 336             /* Update vectorial force */
 337             fix0             = _mm256_add_ps(fix0,tx);
 338             fiy0             = _mm256_add_ps(fiy0,ty);
 339             fiz0             = _mm256_add_ps(fiz0,tz);
 340
 341             fjx0             = _mm256_add_ps(fjx0,tx);
 342             fjy0             = _mm256_add_ps(fjy0,ty);
 343             fjz0             = _mm256_add_ps(fjz0,tz);
 344
 345             /**************************
 346              * CALCULATE INTERACTIONS *
 347              **************************/
 348
 349             r01              = _mm256_mul_ps(rsq01,rinv01);
 350
 351             /* Calculate table index by multiplying r with table scale and truncate to integer */
 352             rt               = _mm256_mul_ps(r01,vftabscale);
 353             vfitab           = _mm256_cvttps_epi32(rt);
 354             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 355             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 356             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 357             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 358             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 359             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 360
 361             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 362             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 363                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 364             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 365                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 366             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 367                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 368             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 369                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 370             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 371             Heps             = _mm256_mul_ps(vfeps,H);
 372             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 373             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 374             velec            = _mm256_mul_ps(qq01,VV);
 375             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 376             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq01,FF),_mm256_mul_ps(vftabscale,rinv01)));
 377
 378             /* Update potential sum for this i atom from the interaction with this j atom. */
 379             velecsum         = _mm256_add_ps(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             /* Calculate temporary vectorial force */
 384             tx               = _mm256_mul_ps(fscal,dx01);
 385             ty               = _mm256_mul_ps(fscal,dy01);
 386             tz               = _mm256_mul_ps(fscal,dz01);
 387
 388             /* Update vectorial force */
 389             fix0             = _mm256_add_ps(fix0,tx);
 390             fiy0             = _mm256_add_ps(fiy0,ty);
 391             fiz0             = _mm256_add_ps(fiz0,tz);
 392
 393             fjx1             = _mm256_add_ps(fjx1,tx);
 394             fjy1             = _mm256_add_ps(fjy1,ty);
 395             fjz1             = _mm256_add_ps(fjz1,tz);
 396
 397             /**************************
 398              * CALCULATE INTERACTIONS *
 399              **************************/
 400
 401             r02              = _mm256_mul_ps(rsq02,rinv02);
 402
 403             /* Calculate table index by multiplying r with table scale and truncate to integer */
 404             rt               = _mm256_mul_ps(r02,vftabscale);
 405             vfitab           = _mm256_cvttps_epi32(rt);
 406             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 407             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 408             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 409             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 410             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 411             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 412
 413             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 414             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 415                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 416             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 417                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 418             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 419                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 420             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 421                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 422             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 423             Heps             = _mm256_mul_ps(vfeps,H);
 424             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 425             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 426             velec            = _mm256_mul_ps(qq02,VV);
 427             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 428             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq02,FF),_mm256_mul_ps(vftabscale,rinv02)));
 429
 430             /* Update potential sum for this i atom from the interaction with this j atom. */
 431             velecsum         = _mm256_add_ps(velecsum,velec);
 432
 433             fscal            = felec;
 434
 435             /* Calculate temporary vectorial force */
 436             tx               = _mm256_mul_ps(fscal,dx02);
 437             ty               = _mm256_mul_ps(fscal,dy02);
 438             tz               = _mm256_mul_ps(fscal,dz02);
 439
 440             /* Update vectorial force */
 441             fix0             = _mm256_add_ps(fix0,tx);
 442             fiy0             = _mm256_add_ps(fiy0,ty);
 443             fiz0             = _mm256_add_ps(fiz0,tz);
 444
 445             fjx2             = _mm256_add_ps(fjx2,tx);
 446             fjy2             = _mm256_add_ps(fjy2,ty);
 447             fjz2             = _mm256_add_ps(fjz2,tz);
 448
 449             /**************************
 450              * CALCULATE INTERACTIONS *
 451              **************************/
 452
 453             r10              = _mm256_mul_ps(rsq10,rinv10);
 454
 455             /* Calculate table index by multiplying r with table scale and truncate to integer */
 456             rt               = _mm256_mul_ps(r10,vftabscale);
 457             vfitab           = _mm256_cvttps_epi32(rt);
 458             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 459             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 460             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 461             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 462             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 463             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 464
 465             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 466             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 467                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 468             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 469                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 470             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 471                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 472             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 473                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 474             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 475             Heps             = _mm256_mul_ps(vfeps,H);
 476             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 477             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 478             velec            = _mm256_mul_ps(qq10,VV);
 479             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 480             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velecsum         = _mm256_add_ps(velecsum,velec);
 484
 485             fscal            = felec;
 486
 487             /* Calculate temporary vectorial force */
 488             tx               = _mm256_mul_ps(fscal,dx10);
 489             ty               = _mm256_mul_ps(fscal,dy10);
 490             tz               = _mm256_mul_ps(fscal,dz10);
 491
 492             /* Update vectorial force */
 493             fix1             = _mm256_add_ps(fix1,tx);
 494             fiy1             = _mm256_add_ps(fiy1,ty);
 495             fiz1             = _mm256_add_ps(fiz1,tz);
 496
 497             fjx0             = _mm256_add_ps(fjx0,tx);
 498             fjy0             = _mm256_add_ps(fjy0,ty);
 499             fjz0             = _mm256_add_ps(fjz0,tz);
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             r11              = _mm256_mul_ps(rsq11,rinv11);
 506
 507             /* Calculate table index by multiplying r with table scale and truncate to integer */
 508             rt               = _mm256_mul_ps(r11,vftabscale);
 509             vfitab           = _mm256_cvttps_epi32(rt);
 510             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 511             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 512             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 513             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 514             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 515             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 516
 517             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 518             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 519                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 520             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 521                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 522             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 523                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 524             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 525                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 526             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 527             Heps             = _mm256_mul_ps(vfeps,H);
 528             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 529             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 530             velec            = _mm256_mul_ps(qq11,VV);
 531             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 532             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             velecsum         = _mm256_add_ps(velecsum,velec);
 536
 537             fscal            = felec;
 538
 539             /* Calculate temporary vectorial force */
 540             tx               = _mm256_mul_ps(fscal,dx11);
 541             ty               = _mm256_mul_ps(fscal,dy11);
 542             tz               = _mm256_mul_ps(fscal,dz11);
 543
 544             /* Update vectorial force */
 545             fix1             = _mm256_add_ps(fix1,tx);
 546             fiy1             = _mm256_add_ps(fiy1,ty);
 547             fiz1             = _mm256_add_ps(fiz1,tz);
 548
 549             fjx1             = _mm256_add_ps(fjx1,tx);
 550             fjy1             = _mm256_add_ps(fjy1,ty);
 551             fjz1             = _mm256_add_ps(fjz1,tz);
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             r12              = _mm256_mul_ps(rsq12,rinv12);
 558
 559             /* Calculate table index by multiplying r with table scale and truncate to integer */
 560             rt               = _mm256_mul_ps(r12,vftabscale);
 561             vfitab           = _mm256_cvttps_epi32(rt);
 562             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 563             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 564             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 565             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 566             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 567             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 568
 569             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 570             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 571                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 572             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 573                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 574             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 575                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 576             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 577                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 578             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 579             Heps             = _mm256_mul_ps(vfeps,H);
 580             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 581             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 582             velec            = _mm256_mul_ps(qq12,VV);
 583             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 584             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             velecsum         = _mm256_add_ps(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             /* Calculate temporary vectorial force */
 592             tx               = _mm256_mul_ps(fscal,dx12);
 593             ty               = _mm256_mul_ps(fscal,dy12);
 594             tz               = _mm256_mul_ps(fscal,dz12);
 595
 596             /* Update vectorial force */
 597             fix1             = _mm256_add_ps(fix1,tx);
 598             fiy1             = _mm256_add_ps(fiy1,ty);
 599             fiz1             = _mm256_add_ps(fiz1,tz);
 600
 601             fjx2             = _mm256_add_ps(fjx2,tx);
 602             fjy2             = _mm256_add_ps(fjy2,ty);
 603             fjz2             = _mm256_add_ps(fjz2,tz);
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             r20              = _mm256_mul_ps(rsq20,rinv20);
 610
 611             /* Calculate table index by multiplying r with table scale and truncate to integer */
 612             rt               = _mm256_mul_ps(r20,vftabscale);
 613             vfitab           = _mm256_cvttps_epi32(rt);
 614             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 615             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 616             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 617             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 618             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 619             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 620
 621             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 622             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 623                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 624             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 625                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 626             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 627                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 628             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 629                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 630             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 631             Heps             = _mm256_mul_ps(vfeps,H);
 632             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 633             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 634             velec            = _mm256_mul_ps(qq20,VV);
 635             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 636             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 637
 638             /* Update potential sum for this i atom from the interaction with this j atom. */
 639             velecsum         = _mm256_add_ps(velecsum,velec);
 640
 641             fscal            = felec;
 642
 643             /* Calculate temporary vectorial force */
 644             tx               = _mm256_mul_ps(fscal,dx20);
 645             ty               = _mm256_mul_ps(fscal,dy20);
 646             tz               = _mm256_mul_ps(fscal,dz20);
 647
 648             /* Update vectorial force */
 649             fix2             = _mm256_add_ps(fix2,tx);
 650             fiy2             = _mm256_add_ps(fiy2,ty);
 651             fiz2             = _mm256_add_ps(fiz2,tz);
 652
 653             fjx0             = _mm256_add_ps(fjx0,tx);
 654             fjy0             = _mm256_add_ps(fjy0,ty);
 655             fjz0             = _mm256_add_ps(fjz0,tz);
 656
 657             /**************************
 658              * CALCULATE INTERACTIONS *
 659              **************************/
 660
 661             r21              = _mm256_mul_ps(rsq21,rinv21);
 662
 663             /* Calculate table index by multiplying r with table scale and truncate to integer */
 664             rt               = _mm256_mul_ps(r21,vftabscale);
 665             vfitab           = _mm256_cvttps_epi32(rt);
 666             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 667             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 668             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 669             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 670             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 671             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 672
 673             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 674             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 675                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 676             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 677                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 678             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 679                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 680             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 681                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 682             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 683             Heps             = _mm256_mul_ps(vfeps,H);
 684             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 685             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 686             velec            = _mm256_mul_ps(qq21,VV);
 687             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 688             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
 689
 690             /* Update potential sum for this i atom from the interaction with this j atom. */
 691             velecsum         = _mm256_add_ps(velecsum,velec);
 692
 693             fscal            = felec;
 694
 695             /* Calculate temporary vectorial force */
 696             tx               = _mm256_mul_ps(fscal,dx21);
 697             ty               = _mm256_mul_ps(fscal,dy21);
 698             tz               = _mm256_mul_ps(fscal,dz21);
 699
 700             /* Update vectorial force */
 701             fix2             = _mm256_add_ps(fix2,tx);
 702             fiy2             = _mm256_add_ps(fiy2,ty);
 703             fiz2             = _mm256_add_ps(fiz2,tz);
 704
 705             fjx1             = _mm256_add_ps(fjx1,tx);
 706             fjy1             = _mm256_add_ps(fjy1,ty);
 707             fjz1             = _mm256_add_ps(fjz1,tz);
 708
 709             /**************************
 710              * CALCULATE INTERACTIONS *
 711              **************************/
 712
 713             r22              = _mm256_mul_ps(rsq22,rinv22);
 714
 715             /* Calculate table index by multiplying r with table scale and truncate to integer */
 716             rt               = _mm256_mul_ps(r22,vftabscale);
 717             vfitab           = _mm256_cvttps_epi32(rt);
 718             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 719             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 720             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 721             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 722             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 723             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 724
 725             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 726             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 727                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 728             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 729                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 730             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 731                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 732             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 733                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 734             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 735             Heps             = _mm256_mul_ps(vfeps,H);
 736             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 737             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 738             velec            = _mm256_mul_ps(qq22,VV);
 739             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 740             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
 741
 742             /* Update potential sum for this i atom from the interaction with this j atom. */
 743             velecsum         = _mm256_add_ps(velecsum,velec);
 744
 745             fscal            = felec;
 746
 747             /* Calculate temporary vectorial force */
 748             tx               = _mm256_mul_ps(fscal,dx22);
 749             ty               = _mm256_mul_ps(fscal,dy22);
 750             tz               = _mm256_mul_ps(fscal,dz22);
 751
 752             /* Update vectorial force */
 753             fix2             = _mm256_add_ps(fix2,tx);
 754             fiy2             = _mm256_add_ps(fiy2,ty);
 755             fiz2             = _mm256_add_ps(fiz2,tz);
 756
 757             fjx2             = _mm256_add_ps(fjx2,tx);
 758             fjy2             = _mm256_add_ps(fjy2,ty);
 759             fjz2             = _mm256_add_ps(fjz2,tz);
 760
 761             fjptrA             = f+j_coord_offsetA;
 762             fjptrB             = f+j_coord_offsetB;
 763             fjptrC             = f+j_coord_offsetC;
 764             fjptrD             = f+j_coord_offsetD;
 765             fjptrE             = f+j_coord_offsetE;
 766             fjptrF             = f+j_coord_offsetF;
 767             fjptrG             = f+j_coord_offsetG;
 768             fjptrH             = f+j_coord_offsetH;
 769
 770             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 771                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 772
 773             /* Inner loop uses 387 flops */
 774         }
 775
 776         if(jidx<j_index_end)
 777         {
 778
 779             /* Get j neighbor index, and coordinate index */
 780             jnrlistA         = jjnr[jidx];
 781             jnrlistB         = jjnr[jidx+1];
 782             jnrlistC         = jjnr[jidx+2];
 783             jnrlistD         = jjnr[jidx+3];
 784             jnrlistE         = jjnr[jidx+4];
 785             jnrlistF         = jjnr[jidx+5];
 786             jnrlistG         = jjnr[jidx+6];
 787             jnrlistH         = jjnr[jidx+7];
 788             /* Sign of each element will be negative for non-real atoms.
 789              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 790              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 791              */
 792             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 793                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 794
 795             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 796             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 797             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 798             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 799             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 800             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 801             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 802             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 803             j_coord_offsetA  = DIM*jnrA;
 804             j_coord_offsetB  = DIM*jnrB;
 805             j_coord_offsetC  = DIM*jnrC;
 806             j_coord_offsetD  = DIM*jnrD;
 807             j_coord_offsetE  = DIM*jnrE;
 808             j_coord_offsetF  = DIM*jnrF;
 809             j_coord_offsetG  = DIM*jnrG;
 810             j_coord_offsetH  = DIM*jnrH;
 811
 812             /* load j atom coordinates */
 813             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 814                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 815                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 816                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 817                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 818
 819             /* Calculate displacement vector */
 820             dx00             = _mm256_sub_ps(ix0,jx0);
 821             dy00             = _mm256_sub_ps(iy0,jy0);
 822             dz00             = _mm256_sub_ps(iz0,jz0);
 823             dx01             = _mm256_sub_ps(ix0,jx1);
 824             dy01             = _mm256_sub_ps(iy0,jy1);
 825             dz01             = _mm256_sub_ps(iz0,jz1);
 826             dx02             = _mm256_sub_ps(ix0,jx2);
 827             dy02             = _mm256_sub_ps(iy0,jy2);
 828             dz02             = _mm256_sub_ps(iz0,jz2);
 829             dx10             = _mm256_sub_ps(ix1,jx0);
 830             dy10             = _mm256_sub_ps(iy1,jy0);
 831             dz10             = _mm256_sub_ps(iz1,jz0);
 832             dx11             = _mm256_sub_ps(ix1,jx1);
 833             dy11             = _mm256_sub_ps(iy1,jy1);
 834             dz11             = _mm256_sub_ps(iz1,jz1);
 835             dx12             = _mm256_sub_ps(ix1,jx2);
 836             dy12             = _mm256_sub_ps(iy1,jy2);
 837             dz12             = _mm256_sub_ps(iz1,jz2);
 838             dx20             = _mm256_sub_ps(ix2,jx0);
 839             dy20             = _mm256_sub_ps(iy2,jy0);
 840             dz20             = _mm256_sub_ps(iz2,jz0);
 841             dx21             = _mm256_sub_ps(ix2,jx1);
 842             dy21             = _mm256_sub_ps(iy2,jy1);
 843             dz21             = _mm256_sub_ps(iz2,jz1);
 844             dx22             = _mm256_sub_ps(ix2,jx2);
 845             dy22             = _mm256_sub_ps(iy2,jy2);
 846             dz22             = _mm256_sub_ps(iz2,jz2);
 847
 848             /* Calculate squared distance and things based on it */
 849             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 850             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 851             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 852             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 853             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 854             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 855             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 856             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 857             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 858
 859             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 860             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 861             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 862             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 863             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 864             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 865             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 866             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 867             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 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
 879             /**************************
 880              * CALCULATE INTERACTIONS *
 881              **************************/
 882
 883             r00              = _mm256_mul_ps(rsq00,rinv00);
 884             r00              = _mm256_andnot_ps(dummy_mask,r00);
 885
 886             /* Calculate table index by multiplying r with table scale and truncate to integer */
 887             rt               = _mm256_mul_ps(r00,vftabscale);
 888             vfitab           = _mm256_cvttps_epi32(rt);
 889             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 890             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 891             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 892             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 893             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 894             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 895
 896             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 897             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 898                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 899             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 900                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 901             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 902                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 903             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 904                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 905             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 906             Heps             = _mm256_mul_ps(vfeps,H);
 907             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 908             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 909             velec            = _mm256_mul_ps(qq00,VV);
 910             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 911             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 912
 913             /* Update potential sum for this i atom from the interaction with this j atom. */
 914             velec            = _mm256_andnot_ps(dummy_mask,velec);
 915             velecsum         = _mm256_add_ps(velecsum,velec);
 916
 917             fscal            = felec;
 918
 919             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 920
 921             /* Calculate temporary vectorial force */
 922             tx               = _mm256_mul_ps(fscal,dx00);
 923             ty               = _mm256_mul_ps(fscal,dy00);
 924             tz               = _mm256_mul_ps(fscal,dz00);
 925
 926             /* Update vectorial force */
 927             fix0             = _mm256_add_ps(fix0,tx);
 928             fiy0             = _mm256_add_ps(fiy0,ty);
 929             fiz0             = _mm256_add_ps(fiz0,tz);
 930
 931             fjx0             = _mm256_add_ps(fjx0,tx);
 932             fjy0             = _mm256_add_ps(fjy0,ty);
 933             fjz0             = _mm256_add_ps(fjz0,tz);
 934
 935             /**************************
 936              * CALCULATE INTERACTIONS *
 937              **************************/
 938
 939             r01              = _mm256_mul_ps(rsq01,rinv01);
 940             r01              = _mm256_andnot_ps(dummy_mask,r01);
 941
 942             /* Calculate table index by multiplying r with table scale and truncate to integer */
 943             rt               = _mm256_mul_ps(r01,vftabscale);
 944             vfitab           = _mm256_cvttps_epi32(rt);
 945             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 946             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 947             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 948             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 949             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 950             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 951
 952             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 953             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 954                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 955             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 956                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 957             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 958                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 959             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 960                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 961             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 962             Heps             = _mm256_mul_ps(vfeps,H);
 963             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 964             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 965             velec            = _mm256_mul_ps(qq01,VV);
 966             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 967             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq01,FF),_mm256_mul_ps(vftabscale,rinv01)));
 968
 969             /* Update potential sum for this i atom from the interaction with this j atom. */
 970             velec            = _mm256_andnot_ps(dummy_mask,velec);
 971             velecsum         = _mm256_add_ps(velecsum,velec);
 972
 973             fscal            = felec;
 974
 975             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 976
 977             /* Calculate temporary vectorial force */
 978             tx               = _mm256_mul_ps(fscal,dx01);
 979             ty               = _mm256_mul_ps(fscal,dy01);
 980             tz               = _mm256_mul_ps(fscal,dz01);
 981
 982             /* Update vectorial force */
 983             fix0             = _mm256_add_ps(fix0,tx);
 984             fiy0             = _mm256_add_ps(fiy0,ty);
 985             fiz0             = _mm256_add_ps(fiz0,tz);
 986
 987             fjx1             = _mm256_add_ps(fjx1,tx);
 988             fjy1             = _mm256_add_ps(fjy1,ty);
 989             fjz1             = _mm256_add_ps(fjz1,tz);
 990
 991             /**************************
 992              * CALCULATE INTERACTIONS *
 993              **************************/
 994
 995             r02              = _mm256_mul_ps(rsq02,rinv02);
 996             r02              = _mm256_andnot_ps(dummy_mask,r02);
 997
 998             /* Calculate table index by multiplying r with table scale and truncate to integer */
 999             rt               = _mm256_mul_ps(r02,vftabscale);
1000             vfitab           = _mm256_cvttps_epi32(rt);
1001             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1002             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1003             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1004             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1005             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1006             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1007
1008             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1009             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1010                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1011             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1012                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1013             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1014                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1015             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1016                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1017             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1018             Heps             = _mm256_mul_ps(vfeps,H);
1019             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1020             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1021             velec            = _mm256_mul_ps(qq02,VV);
1022             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1023             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq02,FF),_mm256_mul_ps(vftabscale,rinv02)));
1024
1025             /* Update potential sum for this i atom from the interaction with this j atom. */
1026             velec            = _mm256_andnot_ps(dummy_mask,velec);
1027             velecsum         = _mm256_add_ps(velecsum,velec);
1028
1029             fscal            = felec;
1030
1031             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1032
1033             /* Calculate temporary vectorial force */
1034             tx               = _mm256_mul_ps(fscal,dx02);
1035             ty               = _mm256_mul_ps(fscal,dy02);
1036             tz               = _mm256_mul_ps(fscal,dz02);
1037
1038             /* Update vectorial force */
1039             fix0             = _mm256_add_ps(fix0,tx);
1040             fiy0             = _mm256_add_ps(fiy0,ty);
1041             fiz0             = _mm256_add_ps(fiz0,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              * CALCULATE INTERACTIONS *
1049              **************************/
1050
1051             r10              = _mm256_mul_ps(rsq10,rinv10);
1052             r10              = _mm256_andnot_ps(dummy_mask,r10);
1053
1054             /* Calculate table index by multiplying r with table scale and truncate to integer */
1055             rt               = _mm256_mul_ps(r10,vftabscale);
1056             vfitab           = _mm256_cvttps_epi32(rt);
1057             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1058             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1059             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1060             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1061             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1062             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1063
1064             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1065             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1066                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1067             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1068                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1069             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1070                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1071             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1072                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1073             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1074             Heps             = _mm256_mul_ps(vfeps,H);
1075             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1076             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1077             velec            = _mm256_mul_ps(qq10,VV);
1078             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1079             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1080
1081             /* Update potential sum for this i atom from the interaction with this j atom. */
1082             velec            = _mm256_andnot_ps(dummy_mask,velec);
1083             velecsum         = _mm256_add_ps(velecsum,velec);
1084
1085             fscal            = felec;
1086
1087             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1088
1089             /* Calculate temporary vectorial force */
1090             tx               = _mm256_mul_ps(fscal,dx10);
1091             ty               = _mm256_mul_ps(fscal,dy10);
1092             tz               = _mm256_mul_ps(fscal,dz10);
1093
1094             /* Update vectorial force */
1095             fix1             = _mm256_add_ps(fix1,tx);
1096             fiy1             = _mm256_add_ps(fiy1,ty);
1097             fiz1             = _mm256_add_ps(fiz1,tz);
1098
1099             fjx0             = _mm256_add_ps(fjx0,tx);
1100             fjy0             = _mm256_add_ps(fjy0,ty);
1101             fjz0             = _mm256_add_ps(fjz0,tz);
1102
1103             /**************************
1104              * CALCULATE INTERACTIONS *
1105              **************************/
1106
1107             r11              = _mm256_mul_ps(rsq11,rinv11);
1108             r11              = _mm256_andnot_ps(dummy_mask,r11);
1109
1110             /* Calculate table index by multiplying r with table scale and truncate to integer */
1111             rt               = _mm256_mul_ps(r11,vftabscale);
1112             vfitab           = _mm256_cvttps_epi32(rt);
1113             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1114             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1115             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1116             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1117             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1118             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1119
1120             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1121             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1122                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1123             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1124                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1125             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1126                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1127             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1128                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1129             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1130             Heps             = _mm256_mul_ps(vfeps,H);
1131             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1132             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1133             velec            = _mm256_mul_ps(qq11,VV);
1134             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1135             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
1136
1137             /* Update potential sum for this i atom from the interaction with this j atom. */
1138             velec            = _mm256_andnot_ps(dummy_mask,velec);
1139             velecsum         = _mm256_add_ps(velecsum,velec);
1140
1141             fscal            = felec;
1142
1143             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1144
1145             /* Calculate temporary vectorial force */
1146             tx               = _mm256_mul_ps(fscal,dx11);
1147             ty               = _mm256_mul_ps(fscal,dy11);
1148             tz               = _mm256_mul_ps(fscal,dz11);
1149
1150             /* Update vectorial force */
1151             fix1             = _mm256_add_ps(fix1,tx);
1152             fiy1             = _mm256_add_ps(fiy1,ty);
1153             fiz1             = _mm256_add_ps(fiz1,tz);
1154
1155             fjx1             = _mm256_add_ps(fjx1,tx);
1156             fjy1             = _mm256_add_ps(fjy1,ty);
1157             fjz1             = _mm256_add_ps(fjz1,tz);
1158
1159             /**************************
1160              * CALCULATE INTERACTIONS *
1161              **************************/
1162
1163             r12              = _mm256_mul_ps(rsq12,rinv12);
1164             r12              = _mm256_andnot_ps(dummy_mask,r12);
1165
1166             /* Calculate table index by multiplying r with table scale and truncate to integer */
1167             rt               = _mm256_mul_ps(r12,vftabscale);
1168             vfitab           = _mm256_cvttps_epi32(rt);
1169             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1170             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1171             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1172             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1173             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1174             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1175
1176             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1177             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1178                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1179             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1180                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1181             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1182                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1183             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1184                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1185             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1186             Heps             = _mm256_mul_ps(vfeps,H);
1187             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1188             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1189             velec            = _mm256_mul_ps(qq12,VV);
1190             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1191             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
1192
1193             /* Update potential sum for this i atom from the interaction with this j atom. */
1194             velec            = _mm256_andnot_ps(dummy_mask,velec);
1195             velecsum         = _mm256_add_ps(velecsum,velec);
1196
1197             fscal            = felec;
1198
1199             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1200
1201             /* Calculate temporary vectorial force */
1202             tx               = _mm256_mul_ps(fscal,dx12);
1203             ty               = _mm256_mul_ps(fscal,dy12);
1204             tz               = _mm256_mul_ps(fscal,dz12);
1205
1206             /* Update vectorial force */
1207             fix1             = _mm256_add_ps(fix1,tx);
1208             fiy1             = _mm256_add_ps(fiy1,ty);
1209             fiz1             = _mm256_add_ps(fiz1,tz);
1210
1211             fjx2             = _mm256_add_ps(fjx2,tx);
1212             fjy2             = _mm256_add_ps(fjy2,ty);
1213             fjz2             = _mm256_add_ps(fjz2,tz);
1214
1215             /**************************
1216              * CALCULATE INTERACTIONS *
1217              **************************/
1218
1219             r20              = _mm256_mul_ps(rsq20,rinv20);
1220             r20              = _mm256_andnot_ps(dummy_mask,r20);
1221
1222             /* Calculate table index by multiplying r with table scale and truncate to integer */
1223             rt               = _mm256_mul_ps(r20,vftabscale);
1224             vfitab           = _mm256_cvttps_epi32(rt);
1225             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1226             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1227             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1228             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1229             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1230             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1231
1232             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1233             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1234                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1235             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1236                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1237             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1238                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1239             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1240                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1241             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1242             Heps             = _mm256_mul_ps(vfeps,H);
1243             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1244             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1245             velec            = _mm256_mul_ps(qq20,VV);
1246             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1247             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1248
1249             /* Update potential sum for this i atom from the interaction with this j atom. */
1250             velec            = _mm256_andnot_ps(dummy_mask,velec);
1251             velecsum         = _mm256_add_ps(velecsum,velec);
1252
1253             fscal            = felec;
1254
1255             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1256
1257             /* Calculate temporary vectorial force */
1258             tx               = _mm256_mul_ps(fscal,dx20);
1259             ty               = _mm256_mul_ps(fscal,dy20);
1260             tz               = _mm256_mul_ps(fscal,dz20);
1261
1262             /* Update vectorial force */
1263             fix2             = _mm256_add_ps(fix2,tx);
1264             fiy2             = _mm256_add_ps(fiy2,ty);
1265             fiz2             = _mm256_add_ps(fiz2,tz);
1266
1267             fjx0             = _mm256_add_ps(fjx0,tx);
1268             fjy0             = _mm256_add_ps(fjy0,ty);
1269             fjz0             = _mm256_add_ps(fjz0,tz);
1270
1271             /**************************
1272              * CALCULATE INTERACTIONS *
1273              **************************/
1274
1275             r21              = _mm256_mul_ps(rsq21,rinv21);
1276             r21              = _mm256_andnot_ps(dummy_mask,r21);
1277
1278             /* Calculate table index by multiplying r with table scale and truncate to integer */
1279             rt               = _mm256_mul_ps(r21,vftabscale);
1280             vfitab           = _mm256_cvttps_epi32(rt);
1281             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1282             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1283             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1284             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1285             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1286             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1287
1288             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1289             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1290                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1291             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1292                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1293             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1294                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1295             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1296                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1297             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1298             Heps             = _mm256_mul_ps(vfeps,H);
1299             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1300             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1301             velec            = _mm256_mul_ps(qq21,VV);
1302             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1303             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
1304
1305             /* Update potential sum for this i atom from the interaction with this j atom. */
1306             velec            = _mm256_andnot_ps(dummy_mask,velec);
1307             velecsum         = _mm256_add_ps(velecsum,velec);
1308
1309             fscal            = felec;
1310
1311             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1312
1313             /* Calculate temporary vectorial force */
1314             tx               = _mm256_mul_ps(fscal,dx21);
1315             ty               = _mm256_mul_ps(fscal,dy21);
1316             tz               = _mm256_mul_ps(fscal,dz21);
1317
1318             /* Update vectorial force */
1319             fix2             = _mm256_add_ps(fix2,tx);
1320             fiy2             = _mm256_add_ps(fiy2,ty);
1321             fiz2             = _mm256_add_ps(fiz2,tz);
1322
1323             fjx1             = _mm256_add_ps(fjx1,tx);
1324             fjy1             = _mm256_add_ps(fjy1,ty);
1325             fjz1             = _mm256_add_ps(fjz1,tz);
1326
1327             /**************************
1328              * CALCULATE INTERACTIONS *
1329              **************************/
1330
1331             r22              = _mm256_mul_ps(rsq22,rinv22);
1332             r22              = _mm256_andnot_ps(dummy_mask,r22);
1333
1334             /* Calculate table index by multiplying r with table scale and truncate to integer */
1335             rt               = _mm256_mul_ps(r22,vftabscale);
1336             vfitab           = _mm256_cvttps_epi32(rt);
1337             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1338             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1339             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1340             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1341             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1342             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1343
1344             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1345             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1346                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1347             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1348                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1349             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1350                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1351             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1352                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1353             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1354             Heps             = _mm256_mul_ps(vfeps,H);
1355             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1356             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1357             velec            = _mm256_mul_ps(qq22,VV);
1358             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1359             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
1360
1361             /* Update potential sum for this i atom from the interaction with this j atom. */
1362             velec            = _mm256_andnot_ps(dummy_mask,velec);
1363             velecsum         = _mm256_add_ps(velecsum,velec);
1364
1365             fscal            = felec;
1366
1367             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1368
1369             /* Calculate temporary vectorial force */
1370             tx               = _mm256_mul_ps(fscal,dx22);
1371             ty               = _mm256_mul_ps(fscal,dy22);
1372             tz               = _mm256_mul_ps(fscal,dz22);
1373
1374             /* Update vectorial force */
1375             fix2             = _mm256_add_ps(fix2,tx);
1376             fiy2             = _mm256_add_ps(fiy2,ty);
1377             fiz2             = _mm256_add_ps(fiz2,tz);
1378
1379             fjx2             = _mm256_add_ps(fjx2,tx);
1380             fjy2             = _mm256_add_ps(fjy2,ty);
1381             fjz2             = _mm256_add_ps(fjz2,tz);
1382
1383             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1384             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1385             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1386             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1387             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1388             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1389             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1390             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1391
1392             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1393                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1394
1395             /* Inner loop uses 396 flops */
1396         }
1397
1398         /* End of innermost loop */
1399
1400         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1401                                                  f+i_coord_offset,fshift+i_shift_offset);
1402
1403         ggid                        = gid[iidx];
1404         /* Update potential energies */
1405         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1406
1407         /* Increment number of inner iterations */
1408         inneriter                  += j_index_end - j_index_start;
1409
1410         /* Outer loop uses 19 flops */
1411     }
1412
1413     /* Increment number of outer iterations */
1414     outeriter        += nri;
1415
1416     /* Update outer/inner flops */
1417
1418     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*396);
1419 }
1420 /*
1421  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW3W3_F_avx_256_single
1422  * Electrostatics interaction: CubicSplineTable
1423  * VdW interaction:            None
1424  * Geometry:                   Water3-Water3
1425  * Calculate force/pot:        Force
1426  */
1427 void
1428 nb_kernel_ElecCSTab_VdwNone_GeomW3W3_F_avx_256_single
1429                     (t_nblist                    * gmx_restrict       nlist,
1430                      rvec                        * gmx_restrict          xx,
1431                      rvec                        * gmx_restrict          ff,
1432                      t_forcerec                  * gmx_restrict          fr,
1433                      t_mdatoms                   * gmx_restrict     mdatoms,
1434                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1435                      t_nrnb                      * gmx_restrict        nrnb)
1436 {
1437     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1438      * just 0 for non-waters.
1439      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1440      * jnr indices corresponding to data put in the four positions in the SIMD register.
1441      */
1442     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1443     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1444     int              jnrA,jnrB,jnrC,jnrD;
1445     int              jnrE,jnrF,jnrG,jnrH;
1446     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1447     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1448     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1449     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1450     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1451     real             rcutoff_scalar;
1452     real             *shiftvec,*fshift,*x,*f;
1453     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1454     real             scratch[4*DIM];
1455     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1456     real *           vdwioffsetptr0;
1457     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1458     real *           vdwioffsetptr1;
1459     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1460     real *           vdwioffsetptr2;
1461     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1462     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1463     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1464     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1465     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1466     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1467     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1468     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1469     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1470     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1471     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1472     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1473     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1474     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1475     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1476     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1477     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1478     real             *charge;
1479     __m256i          vfitab;
1480     __m128i          vfitab_lo,vfitab_hi;
1481     __m128i          ifour       = _mm_set1_epi32(4);
1482     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1483     real             *vftab;
1484     __m256           dummy_mask,cutoff_mask;
1485     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1486     __m256           one     = _mm256_set1_ps(1.0);
1487     __m256           two     = _mm256_set1_ps(2.0);
1488     x                = xx[0];
1489     f                = ff[0];
1490
1491     nri              = nlist->nri;
1492     iinr             = nlist->iinr;
1493     jindex           = nlist->jindex;
1494     jjnr             = nlist->jjnr;
1495     shiftidx         = nlist->shift;
1496     gid              = nlist->gid;
1497     shiftvec         = fr->shift_vec[0];
1498     fshift           = fr->fshift[0];
1499     facel            = _mm256_set1_ps(fr->epsfac);
1500     charge           = mdatoms->chargeA;
1501
1502     vftab            = kernel_data->table_elec->data;
1503     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
1504
1505     /* Setup water-specific parameters */
1506     inr              = nlist->iinr[0];
1507     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1508     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1509     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1510
1511     jq0              = _mm256_set1_ps(charge[inr+0]);
1512     jq1              = _mm256_set1_ps(charge[inr+1]);
1513     jq2              = _mm256_set1_ps(charge[inr+2]);
1514     qq00             = _mm256_mul_ps(iq0,jq0);
1515     qq01             = _mm256_mul_ps(iq0,jq1);
1516     qq02             = _mm256_mul_ps(iq0,jq2);
1517     qq10             = _mm256_mul_ps(iq1,jq0);
1518     qq11             = _mm256_mul_ps(iq1,jq1);
1519     qq12             = _mm256_mul_ps(iq1,jq2);
1520     qq20             = _mm256_mul_ps(iq2,jq0);
1521     qq21             = _mm256_mul_ps(iq2,jq1);
1522     qq22             = _mm256_mul_ps(iq2,jq2);
1523
1524     /* Avoid stupid compiler warnings */
1525     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1526     j_coord_offsetA = 0;
1527     j_coord_offsetB = 0;
1528     j_coord_offsetC = 0;
1529     j_coord_offsetD = 0;
1530     j_coord_offsetE = 0;
1531     j_coord_offsetF = 0;
1532     j_coord_offsetG = 0;
1533     j_coord_offsetH = 0;
1534
1535     outeriter        = 0;
1536     inneriter        = 0;
1537
1538     for(iidx=0;iidx<4*DIM;iidx++)
1539     {
1540         scratch[iidx] = 0.0;
1541     }
1542
1543     /* Start outer loop over neighborlists */
1544     for(iidx=0; iidx<nri; iidx++)
1545     {
1546         /* Load shift vector for this list */
1547         i_shift_offset   = DIM*shiftidx[iidx];
1548
1549         /* Load limits for loop over neighbors */
1550         j_index_start    = jindex[iidx];
1551         j_index_end      = jindex[iidx+1];
1552
1553         /* Get outer coordinate index */
1554         inr              = iinr[iidx];
1555         i_coord_offset   = DIM*inr;
1556
1557         /* Load i particle coords and add shift vector */
1558         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1559                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1560
1561         fix0             = _mm256_setzero_ps();
1562         fiy0             = _mm256_setzero_ps();
1563         fiz0             = _mm256_setzero_ps();
1564         fix1             = _mm256_setzero_ps();
1565         fiy1             = _mm256_setzero_ps();
1566         fiz1             = _mm256_setzero_ps();
1567         fix2             = _mm256_setzero_ps();
1568         fiy2             = _mm256_setzero_ps();
1569         fiz2             = _mm256_setzero_ps();
1570
1571         /* Start inner kernel loop */
1572         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1573         {
1574
1575             /* Get j neighbor index, and coordinate index */
1576             jnrA             = jjnr[jidx];
1577             jnrB             = jjnr[jidx+1];
1578             jnrC             = jjnr[jidx+2];
1579             jnrD             = jjnr[jidx+3];
1580             jnrE             = jjnr[jidx+4];
1581             jnrF             = jjnr[jidx+5];
1582             jnrG             = jjnr[jidx+6];
1583             jnrH             = jjnr[jidx+7];
1584             j_coord_offsetA  = DIM*jnrA;
1585             j_coord_offsetB  = DIM*jnrB;
1586             j_coord_offsetC  = DIM*jnrC;
1587             j_coord_offsetD  = DIM*jnrD;
1588             j_coord_offsetE  = DIM*jnrE;
1589             j_coord_offsetF  = DIM*jnrF;
1590             j_coord_offsetG  = DIM*jnrG;
1591             j_coord_offsetH  = DIM*jnrH;
1592
1593             /* load j atom coordinates */
1594             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1595                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1596                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1597                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1598                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1599
1600             /* Calculate displacement vector */
1601             dx00             = _mm256_sub_ps(ix0,jx0);
1602             dy00             = _mm256_sub_ps(iy0,jy0);
1603             dz00             = _mm256_sub_ps(iz0,jz0);
1604             dx01             = _mm256_sub_ps(ix0,jx1);
1605             dy01             = _mm256_sub_ps(iy0,jy1);
1606             dz01             = _mm256_sub_ps(iz0,jz1);
1607             dx02             = _mm256_sub_ps(ix0,jx2);
1608             dy02             = _mm256_sub_ps(iy0,jy2);
1609             dz02             = _mm256_sub_ps(iz0,jz2);
1610             dx10             = _mm256_sub_ps(ix1,jx0);
1611             dy10             = _mm256_sub_ps(iy1,jy0);
1612             dz10             = _mm256_sub_ps(iz1,jz0);
1613             dx11             = _mm256_sub_ps(ix1,jx1);
1614             dy11             = _mm256_sub_ps(iy1,jy1);
1615             dz11             = _mm256_sub_ps(iz1,jz1);
1616             dx12             = _mm256_sub_ps(ix1,jx2);
1617             dy12             = _mm256_sub_ps(iy1,jy2);
1618             dz12             = _mm256_sub_ps(iz1,jz2);
1619             dx20             = _mm256_sub_ps(ix2,jx0);
1620             dy20             = _mm256_sub_ps(iy2,jy0);
1621             dz20             = _mm256_sub_ps(iz2,jz0);
1622             dx21             = _mm256_sub_ps(ix2,jx1);
1623             dy21             = _mm256_sub_ps(iy2,jy1);
1624             dz21             = _mm256_sub_ps(iz2,jz1);
1625             dx22             = _mm256_sub_ps(ix2,jx2);
1626             dy22             = _mm256_sub_ps(iy2,jy2);
1627             dz22             = _mm256_sub_ps(iz2,jz2);
1628
1629             /* Calculate squared distance and things based on it */
1630             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1631             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1632             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1633             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1634             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1635             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1636             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1637             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1638             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1639
1640             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1641             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1642             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1643             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1644             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1645             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1646             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1647             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1648             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1649
1650             fjx0             = _mm256_setzero_ps();
1651             fjy0             = _mm256_setzero_ps();
1652             fjz0             = _mm256_setzero_ps();
1653             fjx1             = _mm256_setzero_ps();
1654             fjy1             = _mm256_setzero_ps();
1655             fjz1             = _mm256_setzero_ps();
1656             fjx2             = _mm256_setzero_ps();
1657             fjy2             = _mm256_setzero_ps();
1658             fjz2             = _mm256_setzero_ps();
1659
1660             /**************************
1661              * CALCULATE INTERACTIONS *
1662              **************************/
1663
1664             r00              = _mm256_mul_ps(rsq00,rinv00);
1665
1666             /* Calculate table index by multiplying r with table scale and truncate to integer */
1667             rt               = _mm256_mul_ps(r00,vftabscale);
1668             vfitab           = _mm256_cvttps_epi32(rt);
1669             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1670             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1671             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1672             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1673             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1674             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1675
1676             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1677             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1678                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1679             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1680                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1681             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1682                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1683             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1684                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1685             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1686             Heps             = _mm256_mul_ps(vfeps,H);
1687             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1688             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1689             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1690
1691             fscal            = felec;
1692
1693             /* Calculate temporary vectorial force */
1694             tx               = _mm256_mul_ps(fscal,dx00);
1695             ty               = _mm256_mul_ps(fscal,dy00);
1696             tz               = _mm256_mul_ps(fscal,dz00);
1697
1698             /* Update vectorial force */
1699             fix0             = _mm256_add_ps(fix0,tx);
1700             fiy0             = _mm256_add_ps(fiy0,ty);
1701             fiz0             = _mm256_add_ps(fiz0,tz);
1702
1703             fjx0             = _mm256_add_ps(fjx0,tx);
1704             fjy0             = _mm256_add_ps(fjy0,ty);
1705             fjz0             = _mm256_add_ps(fjz0,tz);
1706
1707             /**************************
1708              * CALCULATE INTERACTIONS *
1709              **************************/
1710
1711             r01              = _mm256_mul_ps(rsq01,rinv01);
1712
1713             /* Calculate table index by multiplying r with table scale and truncate to integer */
1714             rt               = _mm256_mul_ps(r01,vftabscale);
1715             vfitab           = _mm256_cvttps_epi32(rt);
1716             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1717             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1718             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1719             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1720             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1721             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1722
1723             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1724             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1725                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1726             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1727                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1728             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1729                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1730             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1731                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1732             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1733             Heps             = _mm256_mul_ps(vfeps,H);
1734             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1735             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1736             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq01,FF),_mm256_mul_ps(vftabscale,rinv01)));
1737
1738             fscal            = felec;
1739
1740             /* Calculate temporary vectorial force */
1741             tx               = _mm256_mul_ps(fscal,dx01);
1742             ty               = _mm256_mul_ps(fscal,dy01);
1743             tz               = _mm256_mul_ps(fscal,dz01);
1744
1745             /* Update vectorial force */
1746             fix0             = _mm256_add_ps(fix0,tx);
1747             fiy0             = _mm256_add_ps(fiy0,ty);
1748             fiz0             = _mm256_add_ps(fiz0,tz);
1749
1750             fjx1             = _mm256_add_ps(fjx1,tx);
1751             fjy1             = _mm256_add_ps(fjy1,ty);
1752             fjz1             = _mm256_add_ps(fjz1,tz);
1753
1754             /**************************
1755              * CALCULATE INTERACTIONS *
1756              **************************/
1757
1758             r02              = _mm256_mul_ps(rsq02,rinv02);
1759
1760             /* Calculate table index by multiplying r with table scale and truncate to integer */
1761             rt               = _mm256_mul_ps(r02,vftabscale);
1762             vfitab           = _mm256_cvttps_epi32(rt);
1763             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1764             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1765             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1766             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1767             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1768             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1769
1770             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1771             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1772                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1773             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1774                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1775             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1776                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1777             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1778                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1779             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1780             Heps             = _mm256_mul_ps(vfeps,H);
1781             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1782             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1783             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq02,FF),_mm256_mul_ps(vftabscale,rinv02)));
1784
1785             fscal            = felec;
1786
1787             /* Calculate temporary vectorial force */
1788             tx               = _mm256_mul_ps(fscal,dx02);
1789             ty               = _mm256_mul_ps(fscal,dy02);
1790             tz               = _mm256_mul_ps(fscal,dz02);
1791
1792             /* Update vectorial force */
1793             fix0             = _mm256_add_ps(fix0,tx);
1794             fiy0             = _mm256_add_ps(fiy0,ty);
1795             fiz0             = _mm256_add_ps(fiz0,tz);
1796
1797             fjx2             = _mm256_add_ps(fjx2,tx);
1798             fjy2             = _mm256_add_ps(fjy2,ty);
1799             fjz2             = _mm256_add_ps(fjz2,tz);
1800
1801             /**************************
1802              * CALCULATE INTERACTIONS *
1803              **************************/
1804
1805             r10              = _mm256_mul_ps(rsq10,rinv10);
1806
1807             /* Calculate table index by multiplying r with table scale and truncate to integer */
1808             rt               = _mm256_mul_ps(r10,vftabscale);
1809             vfitab           = _mm256_cvttps_epi32(rt);
1810             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1811             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1812             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1813             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1814             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1815             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1816
1817             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1818             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1819                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1820             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1821                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1822             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1823                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1824             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1825                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1826             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1827             Heps             = _mm256_mul_ps(vfeps,H);
1828             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1829             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1830             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1831
1832             fscal            = felec;
1833
1834             /* Calculate temporary vectorial force */
1835             tx               = _mm256_mul_ps(fscal,dx10);
1836             ty               = _mm256_mul_ps(fscal,dy10);
1837             tz               = _mm256_mul_ps(fscal,dz10);
1838
1839             /* Update vectorial force */
1840             fix1             = _mm256_add_ps(fix1,tx);
1841             fiy1             = _mm256_add_ps(fiy1,ty);
1842             fiz1             = _mm256_add_ps(fiz1,tz);
1843
1844             fjx0             = _mm256_add_ps(fjx0,tx);
1845             fjy0             = _mm256_add_ps(fjy0,ty);
1846             fjz0             = _mm256_add_ps(fjz0,tz);
1847
1848             /**************************
1849              * CALCULATE INTERACTIONS *
1850              **************************/
1851
1852             r11              = _mm256_mul_ps(rsq11,rinv11);
1853
1854             /* Calculate table index by multiplying r with table scale and truncate to integer */
1855             rt               = _mm256_mul_ps(r11,vftabscale);
1856             vfitab           = _mm256_cvttps_epi32(rt);
1857             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1858             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1859             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1860             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1861             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1862             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1863
1864             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1865             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1866                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1867             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1868                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1869             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1870                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1871             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1872                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1873             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1874             Heps             = _mm256_mul_ps(vfeps,H);
1875             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1876             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1877             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
1878
1879             fscal            = felec;
1880
1881             /* Calculate temporary vectorial force */
1882             tx               = _mm256_mul_ps(fscal,dx11);
1883             ty               = _mm256_mul_ps(fscal,dy11);
1884             tz               = _mm256_mul_ps(fscal,dz11);
1885
1886             /* Update vectorial force */
1887             fix1             = _mm256_add_ps(fix1,tx);
1888             fiy1             = _mm256_add_ps(fiy1,ty);
1889             fiz1             = _mm256_add_ps(fiz1,tz);
1890
1891             fjx1             = _mm256_add_ps(fjx1,tx);
1892             fjy1             = _mm256_add_ps(fjy1,ty);
1893             fjz1             = _mm256_add_ps(fjz1,tz);
1894
1895             /**************************
1896              * CALCULATE INTERACTIONS *
1897              **************************/
1898
1899             r12              = _mm256_mul_ps(rsq12,rinv12);
1900
1901             /* Calculate table index by multiplying r with table scale and truncate to integer */
1902             rt               = _mm256_mul_ps(r12,vftabscale);
1903             vfitab           = _mm256_cvttps_epi32(rt);
1904             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1905             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1906             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1907             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1908             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1909             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1910
1911             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1912             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1913                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1914             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1915                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1916             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1917                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1918             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1919                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1920             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1921             Heps             = _mm256_mul_ps(vfeps,H);
1922             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1923             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1924             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
1925
1926             fscal            = felec;
1927
1928             /* Calculate temporary vectorial force */
1929             tx               = _mm256_mul_ps(fscal,dx12);
1930             ty               = _mm256_mul_ps(fscal,dy12);
1931             tz               = _mm256_mul_ps(fscal,dz12);
1932
1933             /* Update vectorial force */
1934             fix1             = _mm256_add_ps(fix1,tx);
1935             fiy1             = _mm256_add_ps(fiy1,ty);
1936             fiz1             = _mm256_add_ps(fiz1,tz);
1937
1938             fjx2             = _mm256_add_ps(fjx2,tx);
1939             fjy2             = _mm256_add_ps(fjy2,ty);
1940             fjz2             = _mm256_add_ps(fjz2,tz);
1941
1942             /**************************
1943              * CALCULATE INTERACTIONS *
1944              **************************/
1945
1946             r20              = _mm256_mul_ps(rsq20,rinv20);
1947
1948             /* Calculate table index by multiplying r with table scale and truncate to integer */
1949             rt               = _mm256_mul_ps(r20,vftabscale);
1950             vfitab           = _mm256_cvttps_epi32(rt);
1951             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1952             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1953             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1954             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1955             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1956             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1957
1958             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1959             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1960                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1961             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1962                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1963             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1964                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1965             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1966                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1967             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1968             Heps             = _mm256_mul_ps(vfeps,H);
1969             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1970             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1971             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1972
1973             fscal            = felec;
1974
1975             /* Calculate temporary vectorial force */
1976             tx               = _mm256_mul_ps(fscal,dx20);
1977             ty               = _mm256_mul_ps(fscal,dy20);
1978             tz               = _mm256_mul_ps(fscal,dz20);
1979
1980             /* Update vectorial force */
1981             fix2             = _mm256_add_ps(fix2,tx);
1982             fiy2             = _mm256_add_ps(fiy2,ty);
1983             fiz2             = _mm256_add_ps(fiz2,tz);
1984
1985             fjx0             = _mm256_add_ps(fjx0,tx);
1986             fjy0             = _mm256_add_ps(fjy0,ty);
1987             fjz0             = _mm256_add_ps(fjz0,tz);
1988
1989             /**************************
1990              * CALCULATE INTERACTIONS *
1991              **************************/
1992
1993             r21              = _mm256_mul_ps(rsq21,rinv21);
1994
1995             /* Calculate table index by multiplying r with table scale and truncate to integer */
1996             rt               = _mm256_mul_ps(r21,vftabscale);
1997             vfitab           = _mm256_cvttps_epi32(rt);
1998             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1999             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2000             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2001             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2002             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2003             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2004
2005             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2006             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2007                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2008             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2009                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2010             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2011                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2012             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2013                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2014             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2015             Heps             = _mm256_mul_ps(vfeps,H);
2016             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2017             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2018             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
2019
2020             fscal            = felec;
2021
2022             /* Calculate temporary vectorial force */
2023             tx               = _mm256_mul_ps(fscal,dx21);
2024             ty               = _mm256_mul_ps(fscal,dy21);
2025             tz               = _mm256_mul_ps(fscal,dz21);
2026
2027             /* Update vectorial force */
2028             fix2             = _mm256_add_ps(fix2,tx);
2029             fiy2             = _mm256_add_ps(fiy2,ty);
2030             fiz2             = _mm256_add_ps(fiz2,tz);
2031
2032             fjx1             = _mm256_add_ps(fjx1,tx);
2033             fjy1             = _mm256_add_ps(fjy1,ty);
2034             fjz1             = _mm256_add_ps(fjz1,tz);
2035
2036             /**************************
2037              * CALCULATE INTERACTIONS *
2038              **************************/
2039
2040             r22              = _mm256_mul_ps(rsq22,rinv22);
2041
2042             /* Calculate table index by multiplying r with table scale and truncate to integer */
2043             rt               = _mm256_mul_ps(r22,vftabscale);
2044             vfitab           = _mm256_cvttps_epi32(rt);
2045             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2046             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2047             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2048             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2049             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2050             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2051
2052             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2053             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2054                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2055             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2056                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2057             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2058                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2059             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2060                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2061             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2062             Heps             = _mm256_mul_ps(vfeps,H);
2063             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2064             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2065             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
2066
2067             fscal            = felec;
2068
2069             /* Calculate temporary vectorial force */
2070             tx               = _mm256_mul_ps(fscal,dx22);
2071             ty               = _mm256_mul_ps(fscal,dy22);
2072             tz               = _mm256_mul_ps(fscal,dz22);
2073
2074             /* Update vectorial force */
2075             fix2             = _mm256_add_ps(fix2,tx);
2076             fiy2             = _mm256_add_ps(fiy2,ty);
2077             fiz2             = _mm256_add_ps(fiz2,tz);
2078
2079             fjx2             = _mm256_add_ps(fjx2,tx);
2080             fjy2             = _mm256_add_ps(fjy2,ty);
2081             fjz2             = _mm256_add_ps(fjz2,tz);
2082
2083             fjptrA             = f+j_coord_offsetA;
2084             fjptrB             = f+j_coord_offsetB;
2085             fjptrC             = f+j_coord_offsetC;
2086             fjptrD             = f+j_coord_offsetD;
2087             fjptrE             = f+j_coord_offsetE;
2088             fjptrF             = f+j_coord_offsetF;
2089             fjptrG             = f+j_coord_offsetG;
2090             fjptrH             = f+j_coord_offsetH;
2091
2092             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2093                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2094
2095             /* Inner loop uses 351 flops */
2096         }
2097
2098         if(jidx<j_index_end)
2099         {
2100
2101             /* Get j neighbor index, and coordinate index */
2102             jnrlistA         = jjnr[jidx];
2103             jnrlistB         = jjnr[jidx+1];
2104             jnrlistC         = jjnr[jidx+2];
2105             jnrlistD         = jjnr[jidx+3];
2106             jnrlistE         = jjnr[jidx+4];
2107             jnrlistF         = jjnr[jidx+5];
2108             jnrlistG         = jjnr[jidx+6];
2109             jnrlistH         = jjnr[jidx+7];
2110             /* Sign of each element will be negative for non-real atoms.
2111              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2112              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2113              */
2114             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2115                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2116
2117             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2118             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2119             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2120             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2121             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2122             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2123             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2124             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2125             j_coord_offsetA  = DIM*jnrA;
2126             j_coord_offsetB  = DIM*jnrB;
2127             j_coord_offsetC  = DIM*jnrC;
2128             j_coord_offsetD  = DIM*jnrD;
2129             j_coord_offsetE  = DIM*jnrE;
2130             j_coord_offsetF  = DIM*jnrF;
2131             j_coord_offsetG  = DIM*jnrG;
2132             j_coord_offsetH  = DIM*jnrH;
2133
2134             /* load j atom coordinates */
2135             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2136                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2137                                                  x+j_coord_offsetE,x+j_coord_offsetF,
2138                                                  x+j_coord_offsetG,x+j_coord_offsetH,
2139                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2140
2141             /* Calculate displacement vector */
2142             dx00             = _mm256_sub_ps(ix0,jx0);
2143             dy00             = _mm256_sub_ps(iy0,jy0);
2144             dz00             = _mm256_sub_ps(iz0,jz0);
2145             dx01             = _mm256_sub_ps(ix0,jx1);
2146             dy01             = _mm256_sub_ps(iy0,jy1);
2147             dz01             = _mm256_sub_ps(iz0,jz1);
2148             dx02             = _mm256_sub_ps(ix0,jx2);
2149             dy02             = _mm256_sub_ps(iy0,jy2);
2150             dz02             = _mm256_sub_ps(iz0,jz2);
2151             dx10             = _mm256_sub_ps(ix1,jx0);
2152             dy10             = _mm256_sub_ps(iy1,jy0);
2153             dz10             = _mm256_sub_ps(iz1,jz0);
2154             dx11             = _mm256_sub_ps(ix1,jx1);
2155             dy11             = _mm256_sub_ps(iy1,jy1);
2156             dz11             = _mm256_sub_ps(iz1,jz1);
2157             dx12             = _mm256_sub_ps(ix1,jx2);
2158             dy12             = _mm256_sub_ps(iy1,jy2);
2159             dz12             = _mm256_sub_ps(iz1,jz2);
2160             dx20             = _mm256_sub_ps(ix2,jx0);
2161             dy20             = _mm256_sub_ps(iy2,jy0);
2162             dz20             = _mm256_sub_ps(iz2,jz0);
2163             dx21             = _mm256_sub_ps(ix2,jx1);
2164             dy21             = _mm256_sub_ps(iy2,jy1);
2165             dz21             = _mm256_sub_ps(iz2,jz1);
2166             dx22             = _mm256_sub_ps(ix2,jx2);
2167             dy22             = _mm256_sub_ps(iy2,jy2);
2168             dz22             = _mm256_sub_ps(iz2,jz2);
2169
2170             /* Calculate squared distance and things based on it */
2171             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2172             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
2173             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
2174             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
2175             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2176             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2177             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
2178             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2179             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2180
2181             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
2182             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
2183             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
2184             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
2185             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2186             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2187             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
2188             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2189             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2190
2191             fjx0             = _mm256_setzero_ps();
2192             fjy0             = _mm256_setzero_ps();
2193             fjz0             = _mm256_setzero_ps();
2194             fjx1             = _mm256_setzero_ps();
2195             fjy1             = _mm256_setzero_ps();
2196             fjz1             = _mm256_setzero_ps();
2197             fjx2             = _mm256_setzero_ps();
2198             fjy2             = _mm256_setzero_ps();
2199             fjz2             = _mm256_setzero_ps();
2200
2201             /**************************
2202              * CALCULATE INTERACTIONS *
2203              **************************/
2204
2205             r00              = _mm256_mul_ps(rsq00,rinv00);
2206             r00              = _mm256_andnot_ps(dummy_mask,r00);
2207
2208             /* Calculate table index by multiplying r with table scale and truncate to integer */
2209             rt               = _mm256_mul_ps(r00,vftabscale);
2210             vfitab           = _mm256_cvttps_epi32(rt);
2211             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2212             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2213             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2214             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2215             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2216             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2217
2218             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2219             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2220                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2221             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2222                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2223             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2224                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2225             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2226                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2227             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2228             Heps             = _mm256_mul_ps(vfeps,H);
2229             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2230             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2231             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
2232
2233             fscal            = felec;
2234
2235             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2236
2237             /* Calculate temporary vectorial force */
2238             tx               = _mm256_mul_ps(fscal,dx00);
2239             ty               = _mm256_mul_ps(fscal,dy00);
2240             tz               = _mm256_mul_ps(fscal,dz00);
2241
2242             /* Update vectorial force */
2243             fix0             = _mm256_add_ps(fix0,tx);
2244             fiy0             = _mm256_add_ps(fiy0,ty);
2245             fiz0             = _mm256_add_ps(fiz0,tz);
2246
2247             fjx0             = _mm256_add_ps(fjx0,tx);
2248             fjy0             = _mm256_add_ps(fjy0,ty);
2249             fjz0             = _mm256_add_ps(fjz0,tz);
2250
2251             /**************************
2252              * CALCULATE INTERACTIONS *
2253              **************************/
2254
2255             r01              = _mm256_mul_ps(rsq01,rinv01);
2256             r01              = _mm256_andnot_ps(dummy_mask,r01);
2257
2258             /* Calculate table index by multiplying r with table scale and truncate to integer */
2259             rt               = _mm256_mul_ps(r01,vftabscale);
2260             vfitab           = _mm256_cvttps_epi32(rt);
2261             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2262             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2263             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2264             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2265             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2266             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2267
2268             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2269             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2270                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2271             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2272                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2273             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2274                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2275             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2276                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2277             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2278             Heps             = _mm256_mul_ps(vfeps,H);
2279             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2280             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2281             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq01,FF),_mm256_mul_ps(vftabscale,rinv01)));
2282
2283             fscal            = felec;
2284
2285             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2286
2287             /* Calculate temporary vectorial force */
2288             tx               = _mm256_mul_ps(fscal,dx01);
2289             ty               = _mm256_mul_ps(fscal,dy01);
2290             tz               = _mm256_mul_ps(fscal,dz01);
2291
2292             /* Update vectorial force */
2293             fix0             = _mm256_add_ps(fix0,tx);
2294             fiy0             = _mm256_add_ps(fiy0,ty);
2295             fiz0             = _mm256_add_ps(fiz0,tz);
2296
2297             fjx1             = _mm256_add_ps(fjx1,tx);
2298             fjy1             = _mm256_add_ps(fjy1,ty);
2299             fjz1             = _mm256_add_ps(fjz1,tz);
2300
2301             /**************************
2302              * CALCULATE INTERACTIONS *
2303              **************************/
2304
2305             r02              = _mm256_mul_ps(rsq02,rinv02);
2306             r02              = _mm256_andnot_ps(dummy_mask,r02);
2307
2308             /* Calculate table index by multiplying r with table scale and truncate to integer */
2309             rt               = _mm256_mul_ps(r02,vftabscale);
2310             vfitab           = _mm256_cvttps_epi32(rt);
2311             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2312             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2313             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2314             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2315             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2316             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2317
2318             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2319             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2320                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2321             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2322                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2323             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2324                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2325             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2326                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2327             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2328             Heps             = _mm256_mul_ps(vfeps,H);
2329             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2330             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2331             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq02,FF),_mm256_mul_ps(vftabscale,rinv02)));
2332
2333             fscal            = felec;
2334
2335             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2336
2337             /* Calculate temporary vectorial force */
2338             tx               = _mm256_mul_ps(fscal,dx02);
2339             ty               = _mm256_mul_ps(fscal,dy02);
2340             tz               = _mm256_mul_ps(fscal,dz02);
2341
2342             /* Update vectorial force */
2343             fix0             = _mm256_add_ps(fix0,tx);
2344             fiy0             = _mm256_add_ps(fiy0,ty);
2345             fiz0             = _mm256_add_ps(fiz0,tz);
2346
2347             fjx2             = _mm256_add_ps(fjx2,tx);
2348             fjy2             = _mm256_add_ps(fjy2,ty);
2349             fjz2             = _mm256_add_ps(fjz2,tz);
2350
2351             /**************************
2352              * CALCULATE INTERACTIONS *
2353              **************************/
2354
2355             r10              = _mm256_mul_ps(rsq10,rinv10);
2356             r10              = _mm256_andnot_ps(dummy_mask,r10);
2357
2358             /* Calculate table index by multiplying r with table scale and truncate to integer */
2359             rt               = _mm256_mul_ps(r10,vftabscale);
2360             vfitab           = _mm256_cvttps_epi32(rt);
2361             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2362             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2363             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2364             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2365             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2366             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2367
2368             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2369             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2370                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2371             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2372                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2373             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2374                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2375             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2376                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2377             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2378             Heps             = _mm256_mul_ps(vfeps,H);
2379             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2380             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2381             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
2382
2383             fscal            = felec;
2384
2385             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2386
2387             /* Calculate temporary vectorial force */
2388             tx               = _mm256_mul_ps(fscal,dx10);
2389             ty               = _mm256_mul_ps(fscal,dy10);
2390             tz               = _mm256_mul_ps(fscal,dz10);
2391
2392             /* Update vectorial force */
2393             fix1             = _mm256_add_ps(fix1,tx);
2394             fiy1             = _mm256_add_ps(fiy1,ty);
2395             fiz1             = _mm256_add_ps(fiz1,tz);
2396
2397             fjx0             = _mm256_add_ps(fjx0,tx);
2398             fjy0             = _mm256_add_ps(fjy0,ty);
2399             fjz0             = _mm256_add_ps(fjz0,tz);
2400
2401             /**************************
2402              * CALCULATE INTERACTIONS *
2403              **************************/
2404
2405             r11              = _mm256_mul_ps(rsq11,rinv11);
2406             r11              = _mm256_andnot_ps(dummy_mask,r11);
2407
2408             /* Calculate table index by multiplying r with table scale and truncate to integer */
2409             rt               = _mm256_mul_ps(r11,vftabscale);
2410             vfitab           = _mm256_cvttps_epi32(rt);
2411             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2412             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2413             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2414             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2415             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2416             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2417
2418             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2419             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2420                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2421             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2422                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2423             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2424                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2425             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2426                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2427             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2428             Heps             = _mm256_mul_ps(vfeps,H);
2429             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2430             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2431             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
2432
2433             fscal            = felec;
2434
2435             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2436
2437             /* Calculate temporary vectorial force */
2438             tx               = _mm256_mul_ps(fscal,dx11);
2439             ty               = _mm256_mul_ps(fscal,dy11);
2440             tz               = _mm256_mul_ps(fscal,dz11);
2441
2442             /* Update vectorial force */
2443             fix1             = _mm256_add_ps(fix1,tx);
2444             fiy1             = _mm256_add_ps(fiy1,ty);
2445             fiz1             = _mm256_add_ps(fiz1,tz);
2446
2447             fjx1             = _mm256_add_ps(fjx1,tx);
2448             fjy1             = _mm256_add_ps(fjy1,ty);
2449             fjz1             = _mm256_add_ps(fjz1,tz);
2450
2451             /**************************
2452              * CALCULATE INTERACTIONS *
2453              **************************/
2454
2455             r12              = _mm256_mul_ps(rsq12,rinv12);
2456             r12              = _mm256_andnot_ps(dummy_mask,r12);
2457
2458             /* Calculate table index by multiplying r with table scale and truncate to integer */
2459             rt               = _mm256_mul_ps(r12,vftabscale);
2460             vfitab           = _mm256_cvttps_epi32(rt);
2461             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2462             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2463             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2464             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2465             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2466             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2467
2468             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2469             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2470                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2471             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2472                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2473             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2474                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2475             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2476                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2477             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2478             Heps             = _mm256_mul_ps(vfeps,H);
2479             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2480             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2481             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
2482
2483             fscal            = felec;
2484
2485             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2486
2487             /* Calculate temporary vectorial force */
2488             tx               = _mm256_mul_ps(fscal,dx12);
2489             ty               = _mm256_mul_ps(fscal,dy12);
2490             tz               = _mm256_mul_ps(fscal,dz12);
2491
2492             /* Update vectorial force */
2493             fix1             = _mm256_add_ps(fix1,tx);
2494             fiy1             = _mm256_add_ps(fiy1,ty);
2495             fiz1             = _mm256_add_ps(fiz1,tz);
2496
2497             fjx2             = _mm256_add_ps(fjx2,tx);
2498             fjy2             = _mm256_add_ps(fjy2,ty);
2499             fjz2             = _mm256_add_ps(fjz2,tz);
2500
2501             /**************************
2502              * CALCULATE INTERACTIONS *
2503              **************************/
2504
2505             r20              = _mm256_mul_ps(rsq20,rinv20);
2506             r20              = _mm256_andnot_ps(dummy_mask,r20);
2507
2508             /* Calculate table index by multiplying r with table scale and truncate to integer */
2509             rt               = _mm256_mul_ps(r20,vftabscale);
2510             vfitab           = _mm256_cvttps_epi32(rt);
2511             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2512             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2513             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2514             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2515             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2516             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2517
2518             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2519             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2520                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2521             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2522                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2523             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2524                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2525             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2526                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2527             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2528             Heps             = _mm256_mul_ps(vfeps,H);
2529             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2530             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2531             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
2532
2533             fscal            = felec;
2534
2535             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2536
2537             /* Calculate temporary vectorial force */
2538             tx               = _mm256_mul_ps(fscal,dx20);
2539             ty               = _mm256_mul_ps(fscal,dy20);
2540             tz               = _mm256_mul_ps(fscal,dz20);
2541
2542             /* Update vectorial force */
2543             fix2             = _mm256_add_ps(fix2,tx);
2544             fiy2             = _mm256_add_ps(fiy2,ty);
2545             fiz2             = _mm256_add_ps(fiz2,tz);
2546
2547             fjx0             = _mm256_add_ps(fjx0,tx);
2548             fjy0             = _mm256_add_ps(fjy0,ty);
2549             fjz0             = _mm256_add_ps(fjz0,tz);
2550
2551             /**************************
2552              * CALCULATE INTERACTIONS *
2553              **************************/
2554
2555             r21              = _mm256_mul_ps(rsq21,rinv21);
2556             r21              = _mm256_andnot_ps(dummy_mask,r21);
2557
2558             /* Calculate table index by multiplying r with table scale and truncate to integer */
2559             rt               = _mm256_mul_ps(r21,vftabscale);
2560             vfitab           = _mm256_cvttps_epi32(rt);
2561             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2562             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2563             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2564             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2565             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2566             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2567
2568             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2569             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2570                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2571             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2572                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2573             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2574                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2575             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2576                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2577             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2578             Heps             = _mm256_mul_ps(vfeps,H);
2579             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2580             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2581             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
2582
2583             fscal            = felec;
2584
2585             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2586
2587             /* Calculate temporary vectorial force */
2588             tx               = _mm256_mul_ps(fscal,dx21);
2589             ty               = _mm256_mul_ps(fscal,dy21);
2590             tz               = _mm256_mul_ps(fscal,dz21);
2591
2592             /* Update vectorial force */
2593             fix2             = _mm256_add_ps(fix2,tx);
2594             fiy2             = _mm256_add_ps(fiy2,ty);
2595             fiz2             = _mm256_add_ps(fiz2,tz);
2596
2597             fjx1             = _mm256_add_ps(fjx1,tx);
2598             fjy1             = _mm256_add_ps(fjy1,ty);
2599             fjz1             = _mm256_add_ps(fjz1,tz);
2600
2601             /**************************
2602              * CALCULATE INTERACTIONS *
2603              **************************/
2604
2605             r22              = _mm256_mul_ps(rsq22,rinv22);
2606             r22              = _mm256_andnot_ps(dummy_mask,r22);
2607
2608             /* Calculate table index by multiplying r with table scale and truncate to integer */
2609             rt               = _mm256_mul_ps(r22,vftabscale);
2610             vfitab           = _mm256_cvttps_epi32(rt);
2611             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2612             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2613             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2614             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2615             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2616             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2617
2618             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2619             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2620                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2621             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2622                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2623             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2624                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2625             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2626                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2627             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2628             Heps             = _mm256_mul_ps(vfeps,H);
2629             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2630             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2631             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
2632
2633             fscal            = felec;
2634
2635             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2636
2637             /* Calculate temporary vectorial force */
2638             tx               = _mm256_mul_ps(fscal,dx22);
2639             ty               = _mm256_mul_ps(fscal,dy22);
2640             tz               = _mm256_mul_ps(fscal,dz22);
2641
2642             /* Update vectorial force */
2643             fix2             = _mm256_add_ps(fix2,tx);
2644             fiy2             = _mm256_add_ps(fiy2,ty);
2645             fiz2             = _mm256_add_ps(fiz2,tz);
2646
2647             fjx2             = _mm256_add_ps(fjx2,tx);
2648             fjy2             = _mm256_add_ps(fjy2,ty);
2649             fjz2             = _mm256_add_ps(fjz2,tz);
2650
2651             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2652             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2653             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2654             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2655             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2656             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2657             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2658             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2659
2660             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2661                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2662
2663             /* Inner loop uses 360 flops */
2664         }
2665
2666         /* End of innermost loop */
2667
2668         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2669                                                  f+i_coord_offset,fshift+i_shift_offset);
2670
2671         /* Increment number of inner iterations */
2672         inneriter                  += j_index_end - j_index_start;
2673
2674         /* Outer loop uses 18 flops */
2675     }
2676
2677     /* Increment number of outer iterations */
2678     outeriter        += nri;
2679
2680     /* Update outer/inner flops */
2681
2682     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*360);
2683 }