src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwNone_GeomW4P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_256_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr1;
  73     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     real *           vdwioffsetptr3;
  77     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     __m256i          vfitab;
  86     __m128i          vfitab_lo,vfitab_hi;
  87     __m128i          ifour       = _mm_set1_epi32(4);
  88     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  89     real             *vftab;
  90     __m256           dummy_mask,cutoff_mask;
  91     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  92     __m256           one     = _mm256_set1_ps(1.0);
  93     __m256           two     = _mm256_set1_ps(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm256_set1_ps(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107
 108     vftab            = kernel_data->table_elec->data;
 109     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 110
 111     /* Setup water-specific parameters */
 112     inr              = nlist->iinr[0];
 113     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 114     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 115     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121     j_coord_offsetC = 0;
 122     j_coord_offsetD = 0;
 123     j_coord_offsetE = 0;
 124     j_coord_offsetF = 0;
 125     j_coord_offsetG = 0;
 126     j_coord_offsetH = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     for(iidx=0;iidx<4*DIM;iidx++)
 132     {
 133         scratch[iidx] = 0.0;
 134     }
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 152                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 153
 154         fix1             = _mm256_setzero_ps();
 155         fiy1             = _mm256_setzero_ps();
 156         fiz1             = _mm256_setzero_ps();
 157         fix2             = _mm256_setzero_ps();
 158         fiy2             = _mm256_setzero_ps();
 159         fiz2             = _mm256_setzero_ps();
 160         fix3             = _mm256_setzero_ps();
 161         fiy3             = _mm256_setzero_ps();
 162         fiz3             = _mm256_setzero_ps();
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm256_setzero_ps();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             jnrC             = jjnr[jidx+2];
 175             jnrD             = jjnr[jidx+3];
 176             jnrE             = jjnr[jidx+4];
 177             jnrF             = jjnr[jidx+5];
 178             jnrG             = jjnr[jidx+6];
 179             jnrH             = jjnr[jidx+7];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182             j_coord_offsetC  = DIM*jnrC;
 183             j_coord_offsetD  = DIM*jnrD;
 184             j_coord_offsetE  = DIM*jnrE;
 185             j_coord_offsetF  = DIM*jnrF;
 186             j_coord_offsetG  = DIM*jnrG;
 187             j_coord_offsetH  = DIM*jnrH;
 188
 189             /* load j atom coordinates */
 190             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 192                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 193                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 194                                                  &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx10             = _mm256_sub_ps(ix1,jx0);
 198             dy10             = _mm256_sub_ps(iy1,jy0);
 199             dz10             = _mm256_sub_ps(iz1,jz0);
 200             dx20             = _mm256_sub_ps(ix2,jx0);
 201             dy20             = _mm256_sub_ps(iy2,jy0);
 202             dz20             = _mm256_sub_ps(iz2,jz0);
 203             dx30             = _mm256_sub_ps(ix3,jx0);
 204             dy30             = _mm256_sub_ps(iy3,jy0);
 205             dz30             = _mm256_sub_ps(iz3,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 209             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 210             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 211
 212             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 213             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 214             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 215
 216             /* Load parameters for j particles */
 217             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 218                                                                  charge+jnrC+0,charge+jnrD+0,
 219                                                                  charge+jnrE+0,charge+jnrF+0,
 220                                                                  charge+jnrG+0,charge+jnrH+0);
 221
 222             fjx0             = _mm256_setzero_ps();
 223             fjy0             = _mm256_setzero_ps();
 224             fjz0             = _mm256_setzero_ps();
 225
 226             /**************************
 227              * CALCULATE INTERACTIONS *
 228              **************************/
 229
 230             r10              = _mm256_mul_ps(rsq10,rinv10);
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             qq10             = _mm256_mul_ps(iq1,jq0);
 234
 235             /* Calculate table index by multiplying r with table scale and truncate to integer */
 236             rt               = _mm256_mul_ps(r10,vftabscale);
 237             vfitab           = _mm256_cvttps_epi32(rt);
 238             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 239             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 240             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 241             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 242             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 243             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 244
 245             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 246             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 247                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 248             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 249                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 250             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 251                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 252             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 253                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 254             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 255             Heps             = _mm256_mul_ps(vfeps,H);
 256             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 257             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 258             velec            = _mm256_mul_ps(qq10,VV);
 259             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 260             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 261
 262             /* Update potential sum for this i atom from the interaction with this j atom. */
 263             velecsum         = _mm256_add_ps(velecsum,velec);
 264
 265             fscal            = felec;
 266
 267             /* Calculate temporary vectorial force */
 268             tx               = _mm256_mul_ps(fscal,dx10);
 269             ty               = _mm256_mul_ps(fscal,dy10);
 270             tz               = _mm256_mul_ps(fscal,dz10);
 271
 272             /* Update vectorial force */
 273             fix1             = _mm256_add_ps(fix1,tx);
 274             fiy1             = _mm256_add_ps(fiy1,ty);
 275             fiz1             = _mm256_add_ps(fiz1,tz);
 276
 277             fjx0             = _mm256_add_ps(fjx0,tx);
 278             fjy0             = _mm256_add_ps(fjy0,ty);
 279             fjz0             = _mm256_add_ps(fjz0,tz);
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             r20              = _mm256_mul_ps(rsq20,rinv20);
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             qq20             = _mm256_mul_ps(iq2,jq0);
 289
 290             /* Calculate table index by multiplying r with table scale and truncate to integer */
 291             rt               = _mm256_mul_ps(r20,vftabscale);
 292             vfitab           = _mm256_cvttps_epi32(rt);
 293             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 294             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 295             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 296             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 297             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 298             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 299
 300             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 301             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 302                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 303             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 304                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 305             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 306                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 307             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 309             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 310             Heps             = _mm256_mul_ps(vfeps,H);
 311             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 312             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 313             velec            = _mm256_mul_ps(qq20,VV);
 314             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 315             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velecsum         = _mm256_add_ps(velecsum,velec);
 319
 320             fscal            = felec;
 321
 322             /* Calculate temporary vectorial force */
 323             tx               = _mm256_mul_ps(fscal,dx20);
 324             ty               = _mm256_mul_ps(fscal,dy20);
 325             tz               = _mm256_mul_ps(fscal,dz20);
 326
 327             /* Update vectorial force */
 328             fix2             = _mm256_add_ps(fix2,tx);
 329             fiy2             = _mm256_add_ps(fiy2,ty);
 330             fiz2             = _mm256_add_ps(fiz2,tz);
 331
 332             fjx0             = _mm256_add_ps(fjx0,tx);
 333             fjy0             = _mm256_add_ps(fjy0,ty);
 334             fjz0             = _mm256_add_ps(fjz0,tz);
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             r30              = _mm256_mul_ps(rsq30,rinv30);
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq30             = _mm256_mul_ps(iq3,jq0);
 344
 345             /* Calculate table index by multiplying r with table scale and truncate to integer */
 346             rt               = _mm256_mul_ps(r30,vftabscale);
 347             vfitab           = _mm256_cvttps_epi32(rt);
 348             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 349             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 350             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 351             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 352             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 353             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 354
 355             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 356             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 357                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 358             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 359                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 360             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 361                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 362             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 363                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 364             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 365             Heps             = _mm256_mul_ps(vfeps,H);
 366             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 367             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 368             velec            = _mm256_mul_ps(qq30,VV);
 369             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 370             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velecsum         = _mm256_add_ps(velecsum,velec);
 374
 375             fscal            = felec;
 376
 377             /* Calculate temporary vectorial force */
 378             tx               = _mm256_mul_ps(fscal,dx30);
 379             ty               = _mm256_mul_ps(fscal,dy30);
 380             tz               = _mm256_mul_ps(fscal,dz30);
 381
 382             /* Update vectorial force */
 383             fix3             = _mm256_add_ps(fix3,tx);
 384             fiy3             = _mm256_add_ps(fiy3,ty);
 385             fiz3             = _mm256_add_ps(fiz3,tz);
 386
 387             fjx0             = _mm256_add_ps(fjx0,tx);
 388             fjy0             = _mm256_add_ps(fjy0,ty);
 389             fjz0             = _mm256_add_ps(fjz0,tz);
 390
 391             fjptrA             = f+j_coord_offsetA;
 392             fjptrB             = f+j_coord_offsetB;
 393             fjptrC             = f+j_coord_offsetC;
 394             fjptrD             = f+j_coord_offsetD;
 395             fjptrE             = f+j_coord_offsetE;
 396             fjptrF             = f+j_coord_offsetF;
 397             fjptrG             = f+j_coord_offsetG;
 398             fjptrH             = f+j_coord_offsetH;
 399
 400             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 401
 402             /* Inner loop uses 132 flops */
 403         }
 404
 405         if(jidx<j_index_end)
 406         {
 407
 408             /* Get j neighbor index, and coordinate index */
 409             jnrlistA         = jjnr[jidx];
 410             jnrlistB         = jjnr[jidx+1];
 411             jnrlistC         = jjnr[jidx+2];
 412             jnrlistD         = jjnr[jidx+3];
 413             jnrlistE         = jjnr[jidx+4];
 414             jnrlistF         = jjnr[jidx+5];
 415             jnrlistG         = jjnr[jidx+6];
 416             jnrlistH         = jjnr[jidx+7];
 417             /* Sign of each element will be negative for non-real atoms.
 418              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 419              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 420              */
 421             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 422                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 423
 424             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 425             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 426             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 427             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 428             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 429             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 430             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 431             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 432             j_coord_offsetA  = DIM*jnrA;
 433             j_coord_offsetB  = DIM*jnrB;
 434             j_coord_offsetC  = DIM*jnrC;
 435             j_coord_offsetD  = DIM*jnrD;
 436             j_coord_offsetE  = DIM*jnrE;
 437             j_coord_offsetF  = DIM*jnrF;
 438             j_coord_offsetG  = DIM*jnrG;
 439             j_coord_offsetH  = DIM*jnrH;
 440
 441             /* load j atom coordinates */
 442             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 443                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 444                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 445                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 446                                                  &jx0,&jy0,&jz0);
 447
 448             /* Calculate displacement vector */
 449             dx10             = _mm256_sub_ps(ix1,jx0);
 450             dy10             = _mm256_sub_ps(iy1,jy0);
 451             dz10             = _mm256_sub_ps(iz1,jz0);
 452             dx20             = _mm256_sub_ps(ix2,jx0);
 453             dy20             = _mm256_sub_ps(iy2,jy0);
 454             dz20             = _mm256_sub_ps(iz2,jz0);
 455             dx30             = _mm256_sub_ps(ix3,jx0);
 456             dy30             = _mm256_sub_ps(iy3,jy0);
 457             dz30             = _mm256_sub_ps(iz3,jz0);
 458
 459             /* Calculate squared distance and things based on it */
 460             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 461             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 462             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 463
 464             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 465             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 466             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 467
 468             /* Load parameters for j particles */
 469             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 470                                                                  charge+jnrC+0,charge+jnrD+0,
 471                                                                  charge+jnrE+0,charge+jnrF+0,
 472                                                                  charge+jnrG+0,charge+jnrH+0);
 473
 474             fjx0             = _mm256_setzero_ps();
 475             fjy0             = _mm256_setzero_ps();
 476             fjz0             = _mm256_setzero_ps();
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             r10              = _mm256_mul_ps(rsq10,rinv10);
 483             r10              = _mm256_andnot_ps(dummy_mask,r10);
 484
 485             /* Compute parameters for interactions between i and j atoms */
 486             qq10             = _mm256_mul_ps(iq1,jq0);
 487
 488             /* Calculate table index by multiplying r with table scale and truncate to integer */
 489             rt               = _mm256_mul_ps(r10,vftabscale);
 490             vfitab           = _mm256_cvttps_epi32(rt);
 491             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 492             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 493             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 494             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 495             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 496             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 497
 498             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 499             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 500                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 501             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 502                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 503             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 504                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 505             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 506                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 507             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 508             Heps             = _mm256_mul_ps(vfeps,H);
 509             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 510             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 511             velec            = _mm256_mul_ps(qq10,VV);
 512             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 513             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 514
 515             /* Update potential sum for this i atom from the interaction with this j atom. */
 516             velec            = _mm256_andnot_ps(dummy_mask,velec);
 517             velecsum         = _mm256_add_ps(velecsum,velec);
 518
 519             fscal            = felec;
 520
 521             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 522
 523             /* Calculate temporary vectorial force */
 524             tx               = _mm256_mul_ps(fscal,dx10);
 525             ty               = _mm256_mul_ps(fscal,dy10);
 526             tz               = _mm256_mul_ps(fscal,dz10);
 527
 528             /* Update vectorial force */
 529             fix1             = _mm256_add_ps(fix1,tx);
 530             fiy1             = _mm256_add_ps(fiy1,ty);
 531             fiz1             = _mm256_add_ps(fiz1,tz);
 532
 533             fjx0             = _mm256_add_ps(fjx0,tx);
 534             fjy0             = _mm256_add_ps(fjy0,ty);
 535             fjz0             = _mm256_add_ps(fjz0,tz);
 536
 537             /**************************
 538              * CALCULATE INTERACTIONS *
 539              **************************/
 540
 541             r20              = _mm256_mul_ps(rsq20,rinv20);
 542             r20              = _mm256_andnot_ps(dummy_mask,r20);
 543
 544             /* Compute parameters for interactions between i and j atoms */
 545             qq20             = _mm256_mul_ps(iq2,jq0);
 546
 547             /* Calculate table index by multiplying r with table scale and truncate to integer */
 548             rt               = _mm256_mul_ps(r20,vftabscale);
 549             vfitab           = _mm256_cvttps_epi32(rt);
 550             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 551             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 552             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 553             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 554             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 555             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 556
 557             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 558             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 559                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 560             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 561                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 562             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 563                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 564             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 565                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 566             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 567             Heps             = _mm256_mul_ps(vfeps,H);
 568             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 569             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 570             velec            = _mm256_mul_ps(qq20,VV);
 571             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 572             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 573
 574             /* Update potential sum for this i atom from the interaction with this j atom. */
 575             velec            = _mm256_andnot_ps(dummy_mask,velec);
 576             velecsum         = _mm256_add_ps(velecsum,velec);
 577
 578             fscal            = felec;
 579
 580             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm256_mul_ps(fscal,dx20);
 584             ty               = _mm256_mul_ps(fscal,dy20);
 585             tz               = _mm256_mul_ps(fscal,dz20);
 586
 587             /* Update vectorial force */
 588             fix2             = _mm256_add_ps(fix2,tx);
 589             fiy2             = _mm256_add_ps(fiy2,ty);
 590             fiz2             = _mm256_add_ps(fiz2,tz);
 591
 592             fjx0             = _mm256_add_ps(fjx0,tx);
 593             fjy0             = _mm256_add_ps(fjy0,ty);
 594             fjz0             = _mm256_add_ps(fjz0,tz);
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             r30              = _mm256_mul_ps(rsq30,rinv30);
 601             r30              = _mm256_andnot_ps(dummy_mask,r30);
 602
 603             /* Compute parameters for interactions between i and j atoms */
 604             qq30             = _mm256_mul_ps(iq3,jq0);
 605
 606             /* Calculate table index by multiplying r with table scale and truncate to integer */
 607             rt               = _mm256_mul_ps(r30,vftabscale);
 608             vfitab           = _mm256_cvttps_epi32(rt);
 609             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 610             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 611             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 612             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 613             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 614             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 615
 616             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 617             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 618                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 619             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 620                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 621             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 622                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 623             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 624                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 625             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 626             Heps             = _mm256_mul_ps(vfeps,H);
 627             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 628             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 629             velec            = _mm256_mul_ps(qq30,VV);
 630             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 631             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 632
 633             /* Update potential sum for this i atom from the interaction with this j atom. */
 634             velec            = _mm256_andnot_ps(dummy_mask,velec);
 635             velecsum         = _mm256_add_ps(velecsum,velec);
 636
 637             fscal            = felec;
 638
 639             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 640
 641             /* Calculate temporary vectorial force */
 642             tx               = _mm256_mul_ps(fscal,dx30);
 643             ty               = _mm256_mul_ps(fscal,dy30);
 644             tz               = _mm256_mul_ps(fscal,dz30);
 645
 646             /* Update vectorial force */
 647             fix3             = _mm256_add_ps(fix3,tx);
 648             fiy3             = _mm256_add_ps(fiy3,ty);
 649             fiz3             = _mm256_add_ps(fiz3,tz);
 650
 651             fjx0             = _mm256_add_ps(fjx0,tx);
 652             fjy0             = _mm256_add_ps(fjy0,ty);
 653             fjz0             = _mm256_add_ps(fjz0,tz);
 654
 655             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 656             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 657             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 658             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 659             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 660             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 661             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 662             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 663
 664             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 665
 666             /* Inner loop uses 135 flops */
 667         }
 668
 669         /* End of innermost loop */
 670
 671         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 672                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 673
 674         ggid                        = gid[iidx];
 675         /* Update potential energies */
 676         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 677
 678         /* Increment number of inner iterations */
 679         inneriter                  += j_index_end - j_index_start;
 680
 681         /* Outer loop uses 19 flops */
 682     }
 683
 684     /* Increment number of outer iterations */
 685     outeriter        += nri;
 686
 687     /* Update outer/inner flops */
 688
 689     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*135);
 690 }
 691 /*
 692  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_256_single
 693  * Electrostatics interaction: CubicSplineTable
 694  * VdW interaction:            None
 695  * Geometry:                   Water4-Particle
 696  * Calculate force/pot:        Force
 697  */
 698 void
 699 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_256_single
 700                     (t_nblist * gmx_restrict                nlist,
 701                      rvec * gmx_restrict                    xx,
 702                      rvec * gmx_restrict                    ff,
 703                      t_forcerec * gmx_restrict              fr,
 704                      t_mdatoms * gmx_restrict               mdatoms,
 705                      nb_kernel_data_t * gmx_restrict        kernel_data,
 706                      t_nrnb * gmx_restrict                  nrnb)
 707 {
 708     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 709      * just 0 for non-waters.
 710      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 711      * jnr indices corresponding to data put in the four positions in the SIMD register.
 712      */
 713     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 714     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 715     int              jnrA,jnrB,jnrC,jnrD;
 716     int              jnrE,jnrF,jnrG,jnrH;
 717     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 718     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 719     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 720     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 721     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 722     real             rcutoff_scalar;
 723     real             *shiftvec,*fshift,*x,*f;
 724     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 725     real             scratch[4*DIM];
 726     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 727     real *           vdwioffsetptr1;
 728     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 729     real *           vdwioffsetptr2;
 730     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 731     real *           vdwioffsetptr3;
 732     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 733     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 734     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 735     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 736     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 737     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 738     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 739     real             *charge;
 740     __m256i          vfitab;
 741     __m128i          vfitab_lo,vfitab_hi;
 742     __m128i          ifour       = _mm_set1_epi32(4);
 743     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 744     real             *vftab;
 745     __m256           dummy_mask,cutoff_mask;
 746     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 747     __m256           one     = _mm256_set1_ps(1.0);
 748     __m256           two     = _mm256_set1_ps(2.0);
 749     x                = xx[0];
 750     f                = ff[0];
 751
 752     nri              = nlist->nri;
 753     iinr             = nlist->iinr;
 754     jindex           = nlist->jindex;
 755     jjnr             = nlist->jjnr;
 756     shiftidx         = nlist->shift;
 757     gid              = nlist->gid;
 758     shiftvec         = fr->shift_vec[0];
 759     fshift           = fr->fshift[0];
 760     facel            = _mm256_set1_ps(fr->epsfac);
 761     charge           = mdatoms->chargeA;
 762
 763     vftab            = kernel_data->table_elec->data;
 764     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 765
 766     /* Setup water-specific parameters */
 767     inr              = nlist->iinr[0];
 768     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 769     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 770     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 771
 772     /* Avoid stupid compiler warnings */
 773     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 774     j_coord_offsetA = 0;
 775     j_coord_offsetB = 0;
 776     j_coord_offsetC = 0;
 777     j_coord_offsetD = 0;
 778     j_coord_offsetE = 0;
 779     j_coord_offsetF = 0;
 780     j_coord_offsetG = 0;
 781     j_coord_offsetH = 0;
 782
 783     outeriter        = 0;
 784     inneriter        = 0;
 785
 786     for(iidx=0;iidx<4*DIM;iidx++)
 787     {
 788         scratch[iidx] = 0.0;
 789     }
 790
 791     /* Start outer loop over neighborlists */
 792     for(iidx=0; iidx<nri; iidx++)
 793     {
 794         /* Load shift vector for this list */
 795         i_shift_offset   = DIM*shiftidx[iidx];
 796
 797         /* Load limits for loop over neighbors */
 798         j_index_start    = jindex[iidx];
 799         j_index_end      = jindex[iidx+1];
 800
 801         /* Get outer coordinate index */
 802         inr              = iinr[iidx];
 803         i_coord_offset   = DIM*inr;
 804
 805         /* Load i particle coords and add shift vector */
 806         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 807                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 808
 809         fix1             = _mm256_setzero_ps();
 810         fiy1             = _mm256_setzero_ps();
 811         fiz1             = _mm256_setzero_ps();
 812         fix2             = _mm256_setzero_ps();
 813         fiy2             = _mm256_setzero_ps();
 814         fiz2             = _mm256_setzero_ps();
 815         fix3             = _mm256_setzero_ps();
 816         fiy3             = _mm256_setzero_ps();
 817         fiz3             = _mm256_setzero_ps();
 818
 819         /* Start inner kernel loop */
 820         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 821         {
 822
 823             /* Get j neighbor index, and coordinate index */
 824             jnrA             = jjnr[jidx];
 825             jnrB             = jjnr[jidx+1];
 826             jnrC             = jjnr[jidx+2];
 827             jnrD             = jjnr[jidx+3];
 828             jnrE             = jjnr[jidx+4];
 829             jnrF             = jjnr[jidx+5];
 830             jnrG             = jjnr[jidx+6];
 831             jnrH             = jjnr[jidx+7];
 832             j_coord_offsetA  = DIM*jnrA;
 833             j_coord_offsetB  = DIM*jnrB;
 834             j_coord_offsetC  = DIM*jnrC;
 835             j_coord_offsetD  = DIM*jnrD;
 836             j_coord_offsetE  = DIM*jnrE;
 837             j_coord_offsetF  = DIM*jnrF;
 838             j_coord_offsetG  = DIM*jnrG;
 839             j_coord_offsetH  = DIM*jnrH;
 840
 841             /* load j atom coordinates */
 842             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 843                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 844                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 845                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 846                                                  &jx0,&jy0,&jz0);
 847
 848             /* Calculate displacement vector */
 849             dx10             = _mm256_sub_ps(ix1,jx0);
 850             dy10             = _mm256_sub_ps(iy1,jy0);
 851             dz10             = _mm256_sub_ps(iz1,jz0);
 852             dx20             = _mm256_sub_ps(ix2,jx0);
 853             dy20             = _mm256_sub_ps(iy2,jy0);
 854             dz20             = _mm256_sub_ps(iz2,jz0);
 855             dx30             = _mm256_sub_ps(ix3,jx0);
 856             dy30             = _mm256_sub_ps(iy3,jy0);
 857             dz30             = _mm256_sub_ps(iz3,jz0);
 858
 859             /* Calculate squared distance and things based on it */
 860             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 861             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 862             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 863
 864             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 865             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 866             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 867
 868             /* Load parameters for j particles */
 869             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 870                                                                  charge+jnrC+0,charge+jnrD+0,
 871                                                                  charge+jnrE+0,charge+jnrF+0,
 872                                                                  charge+jnrG+0,charge+jnrH+0);
 873
 874             fjx0             = _mm256_setzero_ps();
 875             fjy0             = _mm256_setzero_ps();
 876             fjz0             = _mm256_setzero_ps();
 877
 878             /**************************
 879              * CALCULATE INTERACTIONS *
 880              **************************/
 881
 882             r10              = _mm256_mul_ps(rsq10,rinv10);
 883
 884             /* Compute parameters for interactions between i and j atoms */
 885             qq10             = _mm256_mul_ps(iq1,jq0);
 886
 887             /* Calculate table index by multiplying r with table scale and truncate to integer */
 888             rt               = _mm256_mul_ps(r10,vftabscale);
 889             vfitab           = _mm256_cvttps_epi32(rt);
 890             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 891             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 892             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 893             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 894             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 895             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 896
 897             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 898             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 899                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 900             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 901                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 902             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 903                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 904             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 905                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 906             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 907             Heps             = _mm256_mul_ps(vfeps,H);
 908             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 909             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 910             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 911
 912             fscal            = felec;
 913
 914             /* Calculate temporary vectorial force */
 915             tx               = _mm256_mul_ps(fscal,dx10);
 916             ty               = _mm256_mul_ps(fscal,dy10);
 917             tz               = _mm256_mul_ps(fscal,dz10);
 918
 919             /* Update vectorial force */
 920             fix1             = _mm256_add_ps(fix1,tx);
 921             fiy1             = _mm256_add_ps(fiy1,ty);
 922             fiz1             = _mm256_add_ps(fiz1,tz);
 923
 924             fjx0             = _mm256_add_ps(fjx0,tx);
 925             fjy0             = _mm256_add_ps(fjy0,ty);
 926             fjz0             = _mm256_add_ps(fjz0,tz);
 927
 928             /**************************
 929              * CALCULATE INTERACTIONS *
 930              **************************/
 931
 932             r20              = _mm256_mul_ps(rsq20,rinv20);
 933
 934             /* Compute parameters for interactions between i and j atoms */
 935             qq20             = _mm256_mul_ps(iq2,jq0);
 936
 937             /* Calculate table index by multiplying r with table scale and truncate to integer */
 938             rt               = _mm256_mul_ps(r20,vftabscale);
 939             vfitab           = _mm256_cvttps_epi32(rt);
 940             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 941             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 942             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 943             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 944             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 945             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 946
 947             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 948             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 949                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 950             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 951                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 952             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 953                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 954             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 955                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 956             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 957             Heps             = _mm256_mul_ps(vfeps,H);
 958             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 959             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 960             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 961
 962             fscal            = felec;
 963
 964             /* Calculate temporary vectorial force */
 965             tx               = _mm256_mul_ps(fscal,dx20);
 966             ty               = _mm256_mul_ps(fscal,dy20);
 967             tz               = _mm256_mul_ps(fscal,dz20);
 968
 969             /* Update vectorial force */
 970             fix2             = _mm256_add_ps(fix2,tx);
 971             fiy2             = _mm256_add_ps(fiy2,ty);
 972             fiz2             = _mm256_add_ps(fiz2,tz);
 973
 974             fjx0             = _mm256_add_ps(fjx0,tx);
 975             fjy0             = _mm256_add_ps(fjy0,ty);
 976             fjz0             = _mm256_add_ps(fjz0,tz);
 977
 978             /**************************
 979              * CALCULATE INTERACTIONS *
 980              **************************/
 981
 982             r30              = _mm256_mul_ps(rsq30,rinv30);
 983
 984             /* Compute parameters for interactions between i and j atoms */
 985             qq30             = _mm256_mul_ps(iq3,jq0);
 986
 987             /* Calculate table index by multiplying r with table scale and truncate to integer */
 988             rt               = _mm256_mul_ps(r30,vftabscale);
 989             vfitab           = _mm256_cvttps_epi32(rt);
 990             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 991             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 992             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 993             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 994             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 995             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 996
 997             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 998             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 999                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1000             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1001                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1002             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1003                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1004             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1005                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1006             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1007             Heps             = _mm256_mul_ps(vfeps,H);
1008             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1009             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1010             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1011
1012             fscal            = felec;
1013
1014             /* Calculate temporary vectorial force */
1015             tx               = _mm256_mul_ps(fscal,dx30);
1016             ty               = _mm256_mul_ps(fscal,dy30);
1017             tz               = _mm256_mul_ps(fscal,dz30);
1018
1019             /* Update vectorial force */
1020             fix3             = _mm256_add_ps(fix3,tx);
1021             fiy3             = _mm256_add_ps(fiy3,ty);
1022             fiz3             = _mm256_add_ps(fiz3,tz);
1023
1024             fjx0             = _mm256_add_ps(fjx0,tx);
1025             fjy0             = _mm256_add_ps(fjy0,ty);
1026             fjz0             = _mm256_add_ps(fjz0,tz);
1027
1028             fjptrA             = f+j_coord_offsetA;
1029             fjptrB             = f+j_coord_offsetB;
1030             fjptrC             = f+j_coord_offsetC;
1031             fjptrD             = f+j_coord_offsetD;
1032             fjptrE             = f+j_coord_offsetE;
1033             fjptrF             = f+j_coord_offsetF;
1034             fjptrG             = f+j_coord_offsetG;
1035             fjptrH             = f+j_coord_offsetH;
1036
1037             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1038
1039             /* Inner loop uses 120 flops */
1040         }
1041
1042         if(jidx<j_index_end)
1043         {
1044
1045             /* Get j neighbor index, and coordinate index */
1046             jnrlistA         = jjnr[jidx];
1047             jnrlistB         = jjnr[jidx+1];
1048             jnrlistC         = jjnr[jidx+2];
1049             jnrlistD         = jjnr[jidx+3];
1050             jnrlistE         = jjnr[jidx+4];
1051             jnrlistF         = jjnr[jidx+5];
1052             jnrlistG         = jjnr[jidx+6];
1053             jnrlistH         = jjnr[jidx+7];
1054             /* Sign of each element will be negative for non-real atoms.
1055              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1056              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1057              */
1058             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1059                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1060
1061             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1062             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1063             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1064             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1065             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1066             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1067             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1068             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1069             j_coord_offsetA  = DIM*jnrA;
1070             j_coord_offsetB  = DIM*jnrB;
1071             j_coord_offsetC  = DIM*jnrC;
1072             j_coord_offsetD  = DIM*jnrD;
1073             j_coord_offsetE  = DIM*jnrE;
1074             j_coord_offsetF  = DIM*jnrF;
1075             j_coord_offsetG  = DIM*jnrG;
1076             j_coord_offsetH  = DIM*jnrH;
1077
1078             /* load j atom coordinates */
1079             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1080                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1081                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1082                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1083                                                  &jx0,&jy0,&jz0);
1084
1085             /* Calculate displacement vector */
1086             dx10             = _mm256_sub_ps(ix1,jx0);
1087             dy10             = _mm256_sub_ps(iy1,jy0);
1088             dz10             = _mm256_sub_ps(iz1,jz0);
1089             dx20             = _mm256_sub_ps(ix2,jx0);
1090             dy20             = _mm256_sub_ps(iy2,jy0);
1091             dz20             = _mm256_sub_ps(iz2,jz0);
1092             dx30             = _mm256_sub_ps(ix3,jx0);
1093             dy30             = _mm256_sub_ps(iy3,jy0);
1094             dz30             = _mm256_sub_ps(iz3,jz0);
1095
1096             /* Calculate squared distance and things based on it */
1097             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1098             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1099             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1100
1101             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1102             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1103             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1104
1105             /* Load parameters for j particles */
1106             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1107                                                                  charge+jnrC+0,charge+jnrD+0,
1108                                                                  charge+jnrE+0,charge+jnrF+0,
1109                                                                  charge+jnrG+0,charge+jnrH+0);
1110
1111             fjx0             = _mm256_setzero_ps();
1112             fjy0             = _mm256_setzero_ps();
1113             fjz0             = _mm256_setzero_ps();
1114
1115             /**************************
1116              * CALCULATE INTERACTIONS *
1117              **************************/
1118
1119             r10              = _mm256_mul_ps(rsq10,rinv10);
1120             r10              = _mm256_andnot_ps(dummy_mask,r10);
1121
1122             /* Compute parameters for interactions between i and j atoms */
1123             qq10             = _mm256_mul_ps(iq1,jq0);
1124
1125             /* Calculate table index by multiplying r with table scale and truncate to integer */
1126             rt               = _mm256_mul_ps(r10,vftabscale);
1127             vfitab           = _mm256_cvttps_epi32(rt);
1128             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1129             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1130             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1131             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1132             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1133             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1134
1135             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1136             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1137                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1138             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1139                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1140             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1141                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1142             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1143                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1144             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1145             Heps             = _mm256_mul_ps(vfeps,H);
1146             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1147             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1148             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1149
1150             fscal            = felec;
1151
1152             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1153
1154             /* Calculate temporary vectorial force */
1155             tx               = _mm256_mul_ps(fscal,dx10);
1156             ty               = _mm256_mul_ps(fscal,dy10);
1157             tz               = _mm256_mul_ps(fscal,dz10);
1158
1159             /* Update vectorial force */
1160             fix1             = _mm256_add_ps(fix1,tx);
1161             fiy1             = _mm256_add_ps(fiy1,ty);
1162             fiz1             = _mm256_add_ps(fiz1,tz);
1163
1164             fjx0             = _mm256_add_ps(fjx0,tx);
1165             fjy0             = _mm256_add_ps(fjy0,ty);
1166             fjz0             = _mm256_add_ps(fjz0,tz);
1167
1168             /**************************
1169              * CALCULATE INTERACTIONS *
1170              **************************/
1171
1172             r20              = _mm256_mul_ps(rsq20,rinv20);
1173             r20              = _mm256_andnot_ps(dummy_mask,r20);
1174
1175             /* Compute parameters for interactions between i and j atoms */
1176             qq20             = _mm256_mul_ps(iq2,jq0);
1177
1178             /* Calculate table index by multiplying r with table scale and truncate to integer */
1179             rt               = _mm256_mul_ps(r20,vftabscale);
1180             vfitab           = _mm256_cvttps_epi32(rt);
1181             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1182             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1183             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1184             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1185             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1186             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1187
1188             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1189             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1190                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1191             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1192                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1193             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1194                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1195             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1196                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1197             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1198             Heps             = _mm256_mul_ps(vfeps,H);
1199             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1200             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1201             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1202
1203             fscal            = felec;
1204
1205             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1206
1207             /* Calculate temporary vectorial force */
1208             tx               = _mm256_mul_ps(fscal,dx20);
1209             ty               = _mm256_mul_ps(fscal,dy20);
1210             tz               = _mm256_mul_ps(fscal,dz20);
1211
1212             /* Update vectorial force */
1213             fix2             = _mm256_add_ps(fix2,tx);
1214             fiy2             = _mm256_add_ps(fiy2,ty);
1215             fiz2             = _mm256_add_ps(fiz2,tz);
1216
1217             fjx0             = _mm256_add_ps(fjx0,tx);
1218             fjy0             = _mm256_add_ps(fjy0,ty);
1219             fjz0             = _mm256_add_ps(fjz0,tz);
1220
1221             /**************************
1222              * CALCULATE INTERACTIONS *
1223              **************************/
1224
1225             r30              = _mm256_mul_ps(rsq30,rinv30);
1226             r30              = _mm256_andnot_ps(dummy_mask,r30);
1227
1228             /* Compute parameters for interactions between i and j atoms */
1229             qq30             = _mm256_mul_ps(iq3,jq0);
1230
1231             /* Calculate table index by multiplying r with table scale and truncate to integer */
1232             rt               = _mm256_mul_ps(r30,vftabscale);
1233             vfitab           = _mm256_cvttps_epi32(rt);
1234             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1235             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1236             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1237             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1238             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1239             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1240
1241             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1242             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1243                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1244             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1245                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1246             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1247                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1248             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1249                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1250             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1251             Heps             = _mm256_mul_ps(vfeps,H);
1252             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1253             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1254             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1255
1256             fscal            = felec;
1257
1258             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1259
1260             /* Calculate temporary vectorial force */
1261             tx               = _mm256_mul_ps(fscal,dx30);
1262             ty               = _mm256_mul_ps(fscal,dy30);
1263             tz               = _mm256_mul_ps(fscal,dz30);
1264
1265             /* Update vectorial force */
1266             fix3             = _mm256_add_ps(fix3,tx);
1267             fiy3             = _mm256_add_ps(fiy3,ty);
1268             fiz3             = _mm256_add_ps(fiz3,tz);
1269
1270             fjx0             = _mm256_add_ps(fjx0,tx);
1271             fjy0             = _mm256_add_ps(fjy0,ty);
1272             fjz0             = _mm256_add_ps(fjz0,tz);
1273
1274             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1275             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1276             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1277             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1278             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1279             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1280             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1281             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1282
1283             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1284
1285             /* Inner loop uses 123 flops */
1286         }
1287
1288         /* End of innermost loop */
1289
1290         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1291                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1292
1293         /* Increment number of inner iterations */
1294         inneriter                  += j_index_end - j_index_start;
1295
1296         /* Outer loop uses 18 flops */
1297     }
1298
1299     /* Increment number of outer iterations */
1300     outeriter        += nri;
1301
1302     /* Update outer/inner flops */
1303
1304     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*123);
1305 }