src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwNone_GeomW4P1_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/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_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_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 *           vdwioffsetptr1;
  87     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     __m256i          vfitab;
 100     __m128i          vfitab_lo,vfitab_hi;
 101     __m128i          ifour       = _mm_set1_epi32(4);
 102     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 103     real             *vftab;
 104     __m256           dummy_mask,cutoff_mask;
 105     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 106     __m256           one     = _mm256_set1_ps(1.0);
 107     __m256           two     = _mm256_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121
 122     vftab            = kernel_data->table_elec->data;
 123     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 124
 125     /* Setup water-specific parameters */
 126     inr              = nlist->iinr[0];
 127     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 128     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 129     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137     j_coord_offsetE = 0;
 138     j_coord_offsetF = 0;
 139     j_coord_offsetG = 0;
 140     j_coord_offsetH = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     for(iidx=0;iidx<4*DIM;iidx++)
 146     {
 147         scratch[iidx] = 0.0;
 148     }
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 166                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 167
 168         fix1             = _mm256_setzero_ps();
 169         fiy1             = _mm256_setzero_ps();
 170         fiz1             = _mm256_setzero_ps();
 171         fix2             = _mm256_setzero_ps();
 172         fiy2             = _mm256_setzero_ps();
 173         fiz2             = _mm256_setzero_ps();
 174         fix3             = _mm256_setzero_ps();
 175         fiy3             = _mm256_setzero_ps();
 176         fiz3             = _mm256_setzero_ps();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm256_setzero_ps();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             jnrC             = jjnr[jidx+2];
 189             jnrD             = jjnr[jidx+3];
 190             jnrE             = jjnr[jidx+4];
 191             jnrF             = jjnr[jidx+5];
 192             jnrG             = jjnr[jidx+6];
 193             jnrH             = jjnr[jidx+7];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198             j_coord_offsetE  = DIM*jnrE;
 199             j_coord_offsetF  = DIM*jnrF;
 200             j_coord_offsetG  = DIM*jnrG;
 201             j_coord_offsetH  = DIM*jnrH;
 202
 203             /* load j atom coordinates */
 204             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 206                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 207                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 208                                                  &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx10             = _mm256_sub_ps(ix1,jx0);
 212             dy10             = _mm256_sub_ps(iy1,jy0);
 213             dz10             = _mm256_sub_ps(iz1,jz0);
 214             dx20             = _mm256_sub_ps(ix2,jx0);
 215             dy20             = _mm256_sub_ps(iy2,jy0);
 216             dz20             = _mm256_sub_ps(iz2,jz0);
 217             dx30             = _mm256_sub_ps(ix3,jx0);
 218             dy30             = _mm256_sub_ps(iy3,jy0);
 219             dz30             = _mm256_sub_ps(iz3,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 223             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 224             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 225
 226             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 227             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 228             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 229
 230             /* Load parameters for j particles */
 231             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 232                                                                  charge+jnrC+0,charge+jnrD+0,
 233                                                                  charge+jnrE+0,charge+jnrF+0,
 234                                                                  charge+jnrG+0,charge+jnrH+0);
 235
 236             fjx0             = _mm256_setzero_ps();
 237             fjy0             = _mm256_setzero_ps();
 238             fjz0             = _mm256_setzero_ps();
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             r10              = _mm256_mul_ps(rsq10,rinv10);
 245
 246             /* Compute parameters for interactions between i and j atoms */
 247             qq10             = _mm256_mul_ps(iq1,jq0);
 248
 249             /* Calculate table index by multiplying r with table scale and truncate to integer */
 250             rt               = _mm256_mul_ps(r10,vftabscale);
 251             vfitab           = _mm256_cvttps_epi32(rt);
 252             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 253             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 254             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 255             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 256             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 257             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 258
 259             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 260             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 261                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 262             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 263                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 264             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 265                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 266             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 267                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 268             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 269             Heps             = _mm256_mul_ps(vfeps,H);
 270             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 271             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 272             velec            = _mm256_mul_ps(qq10,VV);
 273             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 274             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             velecsum         = _mm256_add_ps(velecsum,velec);
 278
 279             fscal            = felec;
 280
 281             /* Calculate temporary vectorial force */
 282             tx               = _mm256_mul_ps(fscal,dx10);
 283             ty               = _mm256_mul_ps(fscal,dy10);
 284             tz               = _mm256_mul_ps(fscal,dz10);
 285
 286             /* Update vectorial force */
 287             fix1             = _mm256_add_ps(fix1,tx);
 288             fiy1             = _mm256_add_ps(fiy1,ty);
 289             fiz1             = _mm256_add_ps(fiz1,tz);
 290
 291             fjx0             = _mm256_add_ps(fjx0,tx);
 292             fjy0             = _mm256_add_ps(fjy0,ty);
 293             fjz0             = _mm256_add_ps(fjz0,tz);
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             r20              = _mm256_mul_ps(rsq20,rinv20);
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq20             = _mm256_mul_ps(iq2,jq0);
 303
 304             /* Calculate table index by multiplying r with table scale and truncate to integer */
 305             rt               = _mm256_mul_ps(r20,vftabscale);
 306             vfitab           = _mm256_cvttps_epi32(rt);
 307             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 308             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 309             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 310             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 311             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 312             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 313
 314             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 315             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 316                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 317             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 318                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 319             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 320                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 321             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 322                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 323             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 324             Heps             = _mm256_mul_ps(vfeps,H);
 325             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 326             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 327             velec            = _mm256_mul_ps(qq20,VV);
 328             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 329             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 330
 331             /* Update potential sum for this i atom from the interaction with this j atom. */
 332             velecsum         = _mm256_add_ps(velecsum,velec);
 333
 334             fscal            = felec;
 335
 336             /* Calculate temporary vectorial force */
 337             tx               = _mm256_mul_ps(fscal,dx20);
 338             ty               = _mm256_mul_ps(fscal,dy20);
 339             tz               = _mm256_mul_ps(fscal,dz20);
 340
 341             /* Update vectorial force */
 342             fix2             = _mm256_add_ps(fix2,tx);
 343             fiy2             = _mm256_add_ps(fiy2,ty);
 344             fiz2             = _mm256_add_ps(fiz2,tz);
 345
 346             fjx0             = _mm256_add_ps(fjx0,tx);
 347             fjy0             = _mm256_add_ps(fjy0,ty);
 348             fjz0             = _mm256_add_ps(fjz0,tz);
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             r30              = _mm256_mul_ps(rsq30,rinv30);
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq30             = _mm256_mul_ps(iq3,jq0);
 358
 359             /* Calculate table index by multiplying r with table scale and truncate to integer */
 360             rt               = _mm256_mul_ps(r30,vftabscale);
 361             vfitab           = _mm256_cvttps_epi32(rt);
 362             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 363             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 364             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 365             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 366             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 367             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 368
 369             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 370             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 371                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 372             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 373                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 374             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 375                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 376             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 377                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 378             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 379             Heps             = _mm256_mul_ps(vfeps,H);
 380             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 381             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 382             velec            = _mm256_mul_ps(qq30,VV);
 383             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 384             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velecsum         = _mm256_add_ps(velecsum,velec);
 388
 389             fscal            = felec;
 390
 391             /* Calculate temporary vectorial force */
 392             tx               = _mm256_mul_ps(fscal,dx30);
 393             ty               = _mm256_mul_ps(fscal,dy30);
 394             tz               = _mm256_mul_ps(fscal,dz30);
 395
 396             /* Update vectorial force */
 397             fix3             = _mm256_add_ps(fix3,tx);
 398             fiy3             = _mm256_add_ps(fiy3,ty);
 399             fiz3             = _mm256_add_ps(fiz3,tz);
 400
 401             fjx0             = _mm256_add_ps(fjx0,tx);
 402             fjy0             = _mm256_add_ps(fjy0,ty);
 403             fjz0             = _mm256_add_ps(fjz0,tz);
 404
 405             fjptrA             = f+j_coord_offsetA;
 406             fjptrB             = f+j_coord_offsetB;
 407             fjptrC             = f+j_coord_offsetC;
 408             fjptrD             = f+j_coord_offsetD;
 409             fjptrE             = f+j_coord_offsetE;
 410             fjptrF             = f+j_coord_offsetF;
 411             fjptrG             = f+j_coord_offsetG;
 412             fjptrH             = f+j_coord_offsetH;
 413
 414             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 415
 416             /* Inner loop uses 132 flops */
 417         }
 418
 419         if(jidx<j_index_end)
 420         {
 421
 422             /* Get j neighbor index, and coordinate index */
 423             jnrlistA         = jjnr[jidx];
 424             jnrlistB         = jjnr[jidx+1];
 425             jnrlistC         = jjnr[jidx+2];
 426             jnrlistD         = jjnr[jidx+3];
 427             jnrlistE         = jjnr[jidx+4];
 428             jnrlistF         = jjnr[jidx+5];
 429             jnrlistG         = jjnr[jidx+6];
 430             jnrlistH         = jjnr[jidx+7];
 431             /* Sign of each element will be negative for non-real atoms.
 432              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 433              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 434              */
 435             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 436                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 437
 438             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 439             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 440             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 441             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 442             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 443             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 444             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 445             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 446             j_coord_offsetA  = DIM*jnrA;
 447             j_coord_offsetB  = DIM*jnrB;
 448             j_coord_offsetC  = DIM*jnrC;
 449             j_coord_offsetD  = DIM*jnrD;
 450             j_coord_offsetE  = DIM*jnrE;
 451             j_coord_offsetF  = DIM*jnrF;
 452             j_coord_offsetG  = DIM*jnrG;
 453             j_coord_offsetH  = DIM*jnrH;
 454
 455             /* load j atom coordinates */
 456             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 457                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 458                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 459                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 460                                                  &jx0,&jy0,&jz0);
 461
 462             /* Calculate displacement vector */
 463             dx10             = _mm256_sub_ps(ix1,jx0);
 464             dy10             = _mm256_sub_ps(iy1,jy0);
 465             dz10             = _mm256_sub_ps(iz1,jz0);
 466             dx20             = _mm256_sub_ps(ix2,jx0);
 467             dy20             = _mm256_sub_ps(iy2,jy0);
 468             dz20             = _mm256_sub_ps(iz2,jz0);
 469             dx30             = _mm256_sub_ps(ix3,jx0);
 470             dy30             = _mm256_sub_ps(iy3,jy0);
 471             dz30             = _mm256_sub_ps(iz3,jz0);
 472
 473             /* Calculate squared distance and things based on it */
 474             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 475             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 476             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 477
 478             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 479             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 480             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 481
 482             /* Load parameters for j particles */
 483             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 484                                                                  charge+jnrC+0,charge+jnrD+0,
 485                                                                  charge+jnrE+0,charge+jnrF+0,
 486                                                                  charge+jnrG+0,charge+jnrH+0);
 487
 488             fjx0             = _mm256_setzero_ps();
 489             fjy0             = _mm256_setzero_ps();
 490             fjz0             = _mm256_setzero_ps();
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             r10              = _mm256_mul_ps(rsq10,rinv10);
 497             r10              = _mm256_andnot_ps(dummy_mask,r10);
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq10             = _mm256_mul_ps(iq1,jq0);
 501
 502             /* Calculate table index by multiplying r with table scale and truncate to integer */
 503             rt               = _mm256_mul_ps(r10,vftabscale);
 504             vfitab           = _mm256_cvttps_epi32(rt);
 505             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 506             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 507             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 508             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 509             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 510             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 511
 512             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 513             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 514                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 515             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 516                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 517             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 518                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 519             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 520                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 521             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 522             Heps             = _mm256_mul_ps(vfeps,H);
 523             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 524             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 525             velec            = _mm256_mul_ps(qq10,VV);
 526             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 527             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             velec            = _mm256_andnot_ps(dummy_mask,velec);
 531             velecsum         = _mm256_add_ps(velecsum,velec);
 532
 533             fscal            = felec;
 534
 535             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 536
 537             /* Calculate temporary vectorial force */
 538             tx               = _mm256_mul_ps(fscal,dx10);
 539             ty               = _mm256_mul_ps(fscal,dy10);
 540             tz               = _mm256_mul_ps(fscal,dz10);
 541
 542             /* Update vectorial force */
 543             fix1             = _mm256_add_ps(fix1,tx);
 544             fiy1             = _mm256_add_ps(fiy1,ty);
 545             fiz1             = _mm256_add_ps(fiz1,tz);
 546
 547             fjx0             = _mm256_add_ps(fjx0,tx);
 548             fjy0             = _mm256_add_ps(fjy0,ty);
 549             fjz0             = _mm256_add_ps(fjz0,tz);
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             r20              = _mm256_mul_ps(rsq20,rinv20);
 556             r20              = _mm256_andnot_ps(dummy_mask,r20);
 557
 558             /* Compute parameters for interactions between i and j atoms */
 559             qq20             = _mm256_mul_ps(iq2,jq0);
 560
 561             /* Calculate table index by multiplying r with table scale and truncate to integer */
 562             rt               = _mm256_mul_ps(r20,vftabscale);
 563             vfitab           = _mm256_cvttps_epi32(rt);
 564             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 565             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 566             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 567             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 568             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 569             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 570
 571             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 572             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 573                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 574             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 575                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 576             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 577                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 578             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 579                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 580             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 581             Heps             = _mm256_mul_ps(vfeps,H);
 582             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 583             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 584             velec            = _mm256_mul_ps(qq20,VV);
 585             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 586             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 587
 588             /* Update potential sum for this i atom from the interaction with this j atom. */
 589             velec            = _mm256_andnot_ps(dummy_mask,velec);
 590             velecsum         = _mm256_add_ps(velecsum,velec);
 591
 592             fscal            = felec;
 593
 594             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 595
 596             /* Calculate temporary vectorial force */
 597             tx               = _mm256_mul_ps(fscal,dx20);
 598             ty               = _mm256_mul_ps(fscal,dy20);
 599             tz               = _mm256_mul_ps(fscal,dz20);
 600
 601             /* Update vectorial force */
 602             fix2             = _mm256_add_ps(fix2,tx);
 603             fiy2             = _mm256_add_ps(fiy2,ty);
 604             fiz2             = _mm256_add_ps(fiz2,tz);
 605
 606             fjx0             = _mm256_add_ps(fjx0,tx);
 607             fjy0             = _mm256_add_ps(fjy0,ty);
 608             fjz0             = _mm256_add_ps(fjz0,tz);
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             r30              = _mm256_mul_ps(rsq30,rinv30);
 615             r30              = _mm256_andnot_ps(dummy_mask,r30);
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             qq30             = _mm256_mul_ps(iq3,jq0);
 619
 620             /* Calculate table index by multiplying r with table scale and truncate to integer */
 621             rt               = _mm256_mul_ps(r30,vftabscale);
 622             vfitab           = _mm256_cvttps_epi32(rt);
 623             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 624             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 625             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 626             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 627             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 628             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 629
 630             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 631             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 632                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 633             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 634                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 635             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 636                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 637             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 638                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 639             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 640             Heps             = _mm256_mul_ps(vfeps,H);
 641             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 642             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 643             velec            = _mm256_mul_ps(qq30,VV);
 644             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 645             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 646
 647             /* Update potential sum for this i atom from the interaction with this j atom. */
 648             velec            = _mm256_andnot_ps(dummy_mask,velec);
 649             velecsum         = _mm256_add_ps(velecsum,velec);
 650
 651             fscal            = felec;
 652
 653             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 654
 655             /* Calculate temporary vectorial force */
 656             tx               = _mm256_mul_ps(fscal,dx30);
 657             ty               = _mm256_mul_ps(fscal,dy30);
 658             tz               = _mm256_mul_ps(fscal,dz30);
 659
 660             /* Update vectorial force */
 661             fix3             = _mm256_add_ps(fix3,tx);
 662             fiy3             = _mm256_add_ps(fiy3,ty);
 663             fiz3             = _mm256_add_ps(fiz3,tz);
 664
 665             fjx0             = _mm256_add_ps(fjx0,tx);
 666             fjy0             = _mm256_add_ps(fjy0,ty);
 667             fjz0             = _mm256_add_ps(fjz0,tz);
 668
 669             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 670             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 671             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 672             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 673             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 674             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 675             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 676             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 677
 678             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 679
 680             /* Inner loop uses 135 flops */
 681         }
 682
 683         /* End of innermost loop */
 684
 685         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 686                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 687
 688         ggid                        = gid[iidx];
 689         /* Update potential energies */
 690         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 691
 692         /* Increment number of inner iterations */
 693         inneriter                  += j_index_end - j_index_start;
 694
 695         /* Outer loop uses 19 flops */
 696     }
 697
 698     /* Increment number of outer iterations */
 699     outeriter        += nri;
 700
 701     /* Update outer/inner flops */
 702
 703     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*135);
 704 }
 705 /*
 706  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_256_single
 707  * Electrostatics interaction: CubicSplineTable
 708  * VdW interaction:            None
 709  * Geometry:                   Water4-Particle
 710  * Calculate force/pot:        Force
 711  */
 712 void
 713 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_256_single
 714                     (t_nblist                    * gmx_restrict       nlist,
 715                      rvec                        * gmx_restrict          xx,
 716                      rvec                        * gmx_restrict          ff,
 717                      t_forcerec                  * gmx_restrict          fr,
 718                      t_mdatoms                   * gmx_restrict     mdatoms,
 719                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 720                      t_nrnb                      * gmx_restrict        nrnb)
 721 {
 722     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 723      * just 0 for non-waters.
 724      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 725      * jnr indices corresponding to data put in the four positions in the SIMD register.
 726      */
 727     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 728     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 729     int              jnrA,jnrB,jnrC,jnrD;
 730     int              jnrE,jnrF,jnrG,jnrH;
 731     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 732     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 733     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 734     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 735     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 736     real             rcutoff_scalar;
 737     real             *shiftvec,*fshift,*x,*f;
 738     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 739     real             scratch[4*DIM];
 740     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 741     real *           vdwioffsetptr1;
 742     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 743     real *           vdwioffsetptr2;
 744     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 745     real *           vdwioffsetptr3;
 746     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 747     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 748     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 749     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 750     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 751     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 752     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 753     real             *charge;
 754     __m256i          vfitab;
 755     __m128i          vfitab_lo,vfitab_hi;
 756     __m128i          ifour       = _mm_set1_epi32(4);
 757     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 758     real             *vftab;
 759     __m256           dummy_mask,cutoff_mask;
 760     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 761     __m256           one     = _mm256_set1_ps(1.0);
 762     __m256           two     = _mm256_set1_ps(2.0);
 763     x                = xx[0];
 764     f                = ff[0];
 765
 766     nri              = nlist->nri;
 767     iinr             = nlist->iinr;
 768     jindex           = nlist->jindex;
 769     jjnr             = nlist->jjnr;
 770     shiftidx         = nlist->shift;
 771     gid              = nlist->gid;
 772     shiftvec         = fr->shift_vec[0];
 773     fshift           = fr->fshift[0];
 774     facel            = _mm256_set1_ps(fr->epsfac);
 775     charge           = mdatoms->chargeA;
 776
 777     vftab            = kernel_data->table_elec->data;
 778     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 779
 780     /* Setup water-specific parameters */
 781     inr              = nlist->iinr[0];
 782     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 783     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 784     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 785
 786     /* Avoid stupid compiler warnings */
 787     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 788     j_coord_offsetA = 0;
 789     j_coord_offsetB = 0;
 790     j_coord_offsetC = 0;
 791     j_coord_offsetD = 0;
 792     j_coord_offsetE = 0;
 793     j_coord_offsetF = 0;
 794     j_coord_offsetG = 0;
 795     j_coord_offsetH = 0;
 796
 797     outeriter        = 0;
 798     inneriter        = 0;
 799
 800     for(iidx=0;iidx<4*DIM;iidx++)
 801     {
 802         scratch[iidx] = 0.0;
 803     }
 804
 805     /* Start outer loop over neighborlists */
 806     for(iidx=0; iidx<nri; iidx++)
 807     {
 808         /* Load shift vector for this list */
 809         i_shift_offset   = DIM*shiftidx[iidx];
 810
 811         /* Load limits for loop over neighbors */
 812         j_index_start    = jindex[iidx];
 813         j_index_end      = jindex[iidx+1];
 814
 815         /* Get outer coordinate index */
 816         inr              = iinr[iidx];
 817         i_coord_offset   = DIM*inr;
 818
 819         /* Load i particle coords and add shift vector */
 820         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 821                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 822
 823         fix1             = _mm256_setzero_ps();
 824         fiy1             = _mm256_setzero_ps();
 825         fiz1             = _mm256_setzero_ps();
 826         fix2             = _mm256_setzero_ps();
 827         fiy2             = _mm256_setzero_ps();
 828         fiz2             = _mm256_setzero_ps();
 829         fix3             = _mm256_setzero_ps();
 830         fiy3             = _mm256_setzero_ps();
 831         fiz3             = _mm256_setzero_ps();
 832
 833         /* Start inner kernel loop */
 834         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 835         {
 836
 837             /* Get j neighbor index, and coordinate index */
 838             jnrA             = jjnr[jidx];
 839             jnrB             = jjnr[jidx+1];
 840             jnrC             = jjnr[jidx+2];
 841             jnrD             = jjnr[jidx+3];
 842             jnrE             = jjnr[jidx+4];
 843             jnrF             = jjnr[jidx+5];
 844             jnrG             = jjnr[jidx+6];
 845             jnrH             = jjnr[jidx+7];
 846             j_coord_offsetA  = DIM*jnrA;
 847             j_coord_offsetB  = DIM*jnrB;
 848             j_coord_offsetC  = DIM*jnrC;
 849             j_coord_offsetD  = DIM*jnrD;
 850             j_coord_offsetE  = DIM*jnrE;
 851             j_coord_offsetF  = DIM*jnrF;
 852             j_coord_offsetG  = DIM*jnrG;
 853             j_coord_offsetH  = DIM*jnrH;
 854
 855             /* load j atom coordinates */
 856             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 857                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 858                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 859                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 860                                                  &jx0,&jy0,&jz0);
 861
 862             /* Calculate displacement vector */
 863             dx10             = _mm256_sub_ps(ix1,jx0);
 864             dy10             = _mm256_sub_ps(iy1,jy0);
 865             dz10             = _mm256_sub_ps(iz1,jz0);
 866             dx20             = _mm256_sub_ps(ix2,jx0);
 867             dy20             = _mm256_sub_ps(iy2,jy0);
 868             dz20             = _mm256_sub_ps(iz2,jz0);
 869             dx30             = _mm256_sub_ps(ix3,jx0);
 870             dy30             = _mm256_sub_ps(iy3,jy0);
 871             dz30             = _mm256_sub_ps(iz3,jz0);
 872
 873             /* Calculate squared distance and things based on it */
 874             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 875             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 876             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 877
 878             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 879             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 880             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 881
 882             /* Load parameters for j particles */
 883             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 884                                                                  charge+jnrC+0,charge+jnrD+0,
 885                                                                  charge+jnrE+0,charge+jnrF+0,
 886                                                                  charge+jnrG+0,charge+jnrH+0);
 887
 888             fjx0             = _mm256_setzero_ps();
 889             fjy0             = _mm256_setzero_ps();
 890             fjz0             = _mm256_setzero_ps();
 891
 892             /**************************
 893              * CALCULATE INTERACTIONS *
 894              **************************/
 895
 896             r10              = _mm256_mul_ps(rsq10,rinv10);
 897
 898             /* Compute parameters for interactions between i and j atoms */
 899             qq10             = _mm256_mul_ps(iq1,jq0);
 900
 901             /* Calculate table index by multiplying r with table scale and truncate to integer */
 902             rt               = _mm256_mul_ps(r10,vftabscale);
 903             vfitab           = _mm256_cvttps_epi32(rt);
 904             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 905             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 906             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 907             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 908             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 909             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 910
 911             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 912             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 913                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 914             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 915                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 916             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 917                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 918             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 919                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 920             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 921             Heps             = _mm256_mul_ps(vfeps,H);
 922             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 923             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 924             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 925
 926             fscal            = felec;
 927
 928             /* Calculate temporary vectorial force */
 929             tx               = _mm256_mul_ps(fscal,dx10);
 930             ty               = _mm256_mul_ps(fscal,dy10);
 931             tz               = _mm256_mul_ps(fscal,dz10);
 932
 933             /* Update vectorial force */
 934             fix1             = _mm256_add_ps(fix1,tx);
 935             fiy1             = _mm256_add_ps(fiy1,ty);
 936             fiz1             = _mm256_add_ps(fiz1,tz);
 937
 938             fjx0             = _mm256_add_ps(fjx0,tx);
 939             fjy0             = _mm256_add_ps(fjy0,ty);
 940             fjz0             = _mm256_add_ps(fjz0,tz);
 941
 942             /**************************
 943              * CALCULATE INTERACTIONS *
 944              **************************/
 945
 946             r20              = _mm256_mul_ps(rsq20,rinv20);
 947
 948             /* Compute parameters for interactions between i and j atoms */
 949             qq20             = _mm256_mul_ps(iq2,jq0);
 950
 951             /* Calculate table index by multiplying r with table scale and truncate to integer */
 952             rt               = _mm256_mul_ps(r20,vftabscale);
 953             vfitab           = _mm256_cvttps_epi32(rt);
 954             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 955             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 956             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 957             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 958             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 959             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 960
 961             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 962             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 963                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 964             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 965                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 966             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 967                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 968             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 969                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 970             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 971             Heps             = _mm256_mul_ps(vfeps,H);
 972             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 973             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 974             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 975
 976             fscal            = felec;
 977
 978             /* Calculate temporary vectorial force */
 979             tx               = _mm256_mul_ps(fscal,dx20);
 980             ty               = _mm256_mul_ps(fscal,dy20);
 981             tz               = _mm256_mul_ps(fscal,dz20);
 982
 983             /* Update vectorial force */
 984             fix2             = _mm256_add_ps(fix2,tx);
 985             fiy2             = _mm256_add_ps(fiy2,ty);
 986             fiz2             = _mm256_add_ps(fiz2,tz);
 987
 988             fjx0             = _mm256_add_ps(fjx0,tx);
 989             fjy0             = _mm256_add_ps(fjy0,ty);
 990             fjz0             = _mm256_add_ps(fjz0,tz);
 991
 992             /**************************
 993              * CALCULATE INTERACTIONS *
 994              **************************/
 995
 996             r30              = _mm256_mul_ps(rsq30,rinv30);
 997
 998             /* Compute parameters for interactions between i and j atoms */
 999             qq30             = _mm256_mul_ps(iq3,jq0);
1000
1001             /* Calculate table index by multiplying r with table scale and truncate to integer */
1002             rt               = _mm256_mul_ps(r30,vftabscale);
1003             vfitab           = _mm256_cvttps_epi32(rt);
1004             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1005             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1006             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1007             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1008             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1009             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1010
1011             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1012             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1013                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1014             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1015                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1016             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1017                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1018             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1019                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1020             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1021             Heps             = _mm256_mul_ps(vfeps,H);
1022             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1023             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1024             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1025
1026             fscal            = felec;
1027
1028             /* Calculate temporary vectorial force */
1029             tx               = _mm256_mul_ps(fscal,dx30);
1030             ty               = _mm256_mul_ps(fscal,dy30);
1031             tz               = _mm256_mul_ps(fscal,dz30);
1032
1033             /* Update vectorial force */
1034             fix3             = _mm256_add_ps(fix3,tx);
1035             fiy3             = _mm256_add_ps(fiy3,ty);
1036             fiz3             = _mm256_add_ps(fiz3,tz);
1037
1038             fjx0             = _mm256_add_ps(fjx0,tx);
1039             fjy0             = _mm256_add_ps(fjy0,ty);
1040             fjz0             = _mm256_add_ps(fjz0,tz);
1041
1042             fjptrA             = f+j_coord_offsetA;
1043             fjptrB             = f+j_coord_offsetB;
1044             fjptrC             = f+j_coord_offsetC;
1045             fjptrD             = f+j_coord_offsetD;
1046             fjptrE             = f+j_coord_offsetE;
1047             fjptrF             = f+j_coord_offsetF;
1048             fjptrG             = f+j_coord_offsetG;
1049             fjptrH             = f+j_coord_offsetH;
1050
1051             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1052
1053             /* Inner loop uses 120 flops */
1054         }
1055
1056         if(jidx<j_index_end)
1057         {
1058
1059             /* Get j neighbor index, and coordinate index */
1060             jnrlistA         = jjnr[jidx];
1061             jnrlistB         = jjnr[jidx+1];
1062             jnrlistC         = jjnr[jidx+2];
1063             jnrlistD         = jjnr[jidx+3];
1064             jnrlistE         = jjnr[jidx+4];
1065             jnrlistF         = jjnr[jidx+5];
1066             jnrlistG         = jjnr[jidx+6];
1067             jnrlistH         = jjnr[jidx+7];
1068             /* Sign of each element will be negative for non-real atoms.
1069              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1070              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1071              */
1072             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1073                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1074
1075             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1076             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1077             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1078             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1079             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1080             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1081             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1082             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1083             j_coord_offsetA  = DIM*jnrA;
1084             j_coord_offsetB  = DIM*jnrB;
1085             j_coord_offsetC  = DIM*jnrC;
1086             j_coord_offsetD  = DIM*jnrD;
1087             j_coord_offsetE  = DIM*jnrE;
1088             j_coord_offsetF  = DIM*jnrF;
1089             j_coord_offsetG  = DIM*jnrG;
1090             j_coord_offsetH  = DIM*jnrH;
1091
1092             /* load j atom coordinates */
1093             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1094                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1095                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1096                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1097                                                  &jx0,&jy0,&jz0);
1098
1099             /* Calculate displacement vector */
1100             dx10             = _mm256_sub_ps(ix1,jx0);
1101             dy10             = _mm256_sub_ps(iy1,jy0);
1102             dz10             = _mm256_sub_ps(iz1,jz0);
1103             dx20             = _mm256_sub_ps(ix2,jx0);
1104             dy20             = _mm256_sub_ps(iy2,jy0);
1105             dz20             = _mm256_sub_ps(iz2,jz0);
1106             dx30             = _mm256_sub_ps(ix3,jx0);
1107             dy30             = _mm256_sub_ps(iy3,jy0);
1108             dz30             = _mm256_sub_ps(iz3,jz0);
1109
1110             /* Calculate squared distance and things based on it */
1111             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1112             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1113             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1114
1115             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1116             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1117             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1118
1119             /* Load parameters for j particles */
1120             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1121                                                                  charge+jnrC+0,charge+jnrD+0,
1122                                                                  charge+jnrE+0,charge+jnrF+0,
1123                                                                  charge+jnrG+0,charge+jnrH+0);
1124
1125             fjx0             = _mm256_setzero_ps();
1126             fjy0             = _mm256_setzero_ps();
1127             fjz0             = _mm256_setzero_ps();
1128
1129             /**************************
1130              * CALCULATE INTERACTIONS *
1131              **************************/
1132
1133             r10              = _mm256_mul_ps(rsq10,rinv10);
1134             r10              = _mm256_andnot_ps(dummy_mask,r10);
1135
1136             /* Compute parameters for interactions between i and j atoms */
1137             qq10             = _mm256_mul_ps(iq1,jq0);
1138
1139             /* Calculate table index by multiplying r with table scale and truncate to integer */
1140             rt               = _mm256_mul_ps(r10,vftabscale);
1141             vfitab           = _mm256_cvttps_epi32(rt);
1142             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1143             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1144             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1145             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1146             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1147             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1148
1149             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1150             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1151                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1152             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1153                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1154             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1155                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1156             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1157                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1158             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1159             Heps             = _mm256_mul_ps(vfeps,H);
1160             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1161             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1162             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1163
1164             fscal            = felec;
1165
1166             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1167
1168             /* Calculate temporary vectorial force */
1169             tx               = _mm256_mul_ps(fscal,dx10);
1170             ty               = _mm256_mul_ps(fscal,dy10);
1171             tz               = _mm256_mul_ps(fscal,dz10);
1172
1173             /* Update vectorial force */
1174             fix1             = _mm256_add_ps(fix1,tx);
1175             fiy1             = _mm256_add_ps(fiy1,ty);
1176             fiz1             = _mm256_add_ps(fiz1,tz);
1177
1178             fjx0             = _mm256_add_ps(fjx0,tx);
1179             fjy0             = _mm256_add_ps(fjy0,ty);
1180             fjz0             = _mm256_add_ps(fjz0,tz);
1181
1182             /**************************
1183              * CALCULATE INTERACTIONS *
1184              **************************/
1185
1186             r20              = _mm256_mul_ps(rsq20,rinv20);
1187             r20              = _mm256_andnot_ps(dummy_mask,r20);
1188
1189             /* Compute parameters for interactions between i and j atoms */
1190             qq20             = _mm256_mul_ps(iq2,jq0);
1191
1192             /* Calculate table index by multiplying r with table scale and truncate to integer */
1193             rt               = _mm256_mul_ps(r20,vftabscale);
1194             vfitab           = _mm256_cvttps_epi32(rt);
1195             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1196             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1197             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1198             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1199             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1200             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1201
1202             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1203             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1204                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1205             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1206                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1207             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1208                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1209             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1210                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1211             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1212             Heps             = _mm256_mul_ps(vfeps,H);
1213             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1214             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1215             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1216
1217             fscal            = felec;
1218
1219             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1220
1221             /* Calculate temporary vectorial force */
1222             tx               = _mm256_mul_ps(fscal,dx20);
1223             ty               = _mm256_mul_ps(fscal,dy20);
1224             tz               = _mm256_mul_ps(fscal,dz20);
1225
1226             /* Update vectorial force */
1227             fix2             = _mm256_add_ps(fix2,tx);
1228             fiy2             = _mm256_add_ps(fiy2,ty);
1229             fiz2             = _mm256_add_ps(fiz2,tz);
1230
1231             fjx0             = _mm256_add_ps(fjx0,tx);
1232             fjy0             = _mm256_add_ps(fjy0,ty);
1233             fjz0             = _mm256_add_ps(fjz0,tz);
1234
1235             /**************************
1236              * CALCULATE INTERACTIONS *
1237              **************************/
1238
1239             r30              = _mm256_mul_ps(rsq30,rinv30);
1240             r30              = _mm256_andnot_ps(dummy_mask,r30);
1241
1242             /* Compute parameters for interactions between i and j atoms */
1243             qq30             = _mm256_mul_ps(iq3,jq0);
1244
1245             /* Calculate table index by multiplying r with table scale and truncate to integer */
1246             rt               = _mm256_mul_ps(r30,vftabscale);
1247             vfitab           = _mm256_cvttps_epi32(rt);
1248             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1249             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1250             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1251             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1252             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1253             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1254
1255             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1256             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1257                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1258             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1259                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1260             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1261                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1262             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1263                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1264             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1265             Heps             = _mm256_mul_ps(vfeps,H);
1266             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1267             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1268             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1269
1270             fscal            = felec;
1271
1272             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1273
1274             /* Calculate temporary vectorial force */
1275             tx               = _mm256_mul_ps(fscal,dx30);
1276             ty               = _mm256_mul_ps(fscal,dy30);
1277             tz               = _mm256_mul_ps(fscal,dz30);
1278
1279             /* Update vectorial force */
1280             fix3             = _mm256_add_ps(fix3,tx);
1281             fiy3             = _mm256_add_ps(fiy3,ty);
1282             fiz3             = _mm256_add_ps(fiz3,tz);
1283
1284             fjx0             = _mm256_add_ps(fjx0,tx);
1285             fjy0             = _mm256_add_ps(fjy0,ty);
1286             fjz0             = _mm256_add_ps(fjz0,tz);
1287
1288             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1289             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1290             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1291             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1292             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1293             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1294             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1295             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1296
1297             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1298
1299             /* Inner loop uses 123 flops */
1300         }
1301
1302         /* End of innermost loop */
1303
1304         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1305                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1306
1307         /* Increment number of inner iterations */
1308         inneriter                  += j_index_end - j_index_start;
1309
1310         /* Outer loop uses 18 flops */
1311     }
1312
1313     /* Increment number of outer iterations */
1314     outeriter        += nri;
1315
1316     /* Update outer/inner flops */
1317
1318     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*123);
1319 }