src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwLJ_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_VdwLJ_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwLJ_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 *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 107     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 108     __m256i          vfitab;
 109     __m128i          vfitab_lo,vfitab_hi;
 110     __m128i          ifour       = _mm_set1_epi32(4);
 111     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 112     real             *vftab;
 113     __m256           dummy_mask,cutoff_mask;
 114     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 115     __m256           one     = _mm256_set1_ps(1.0);
 116     __m256           two     = _mm256_set1_ps(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_ps(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     vftab            = kernel_data->table_elec->data;
 135     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 140     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 141     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 142     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150     j_coord_offsetE = 0;
 151     j_coord_offsetF = 0;
 152     j_coord_offsetG = 0;
 153     j_coord_offsetH = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 180
 181         fix0             = _mm256_setzero_ps();
 182         fiy0             = _mm256_setzero_ps();
 183         fiz0             = _mm256_setzero_ps();
 184         fix1             = _mm256_setzero_ps();
 185         fiy1             = _mm256_setzero_ps();
 186         fiz1             = _mm256_setzero_ps();
 187         fix2             = _mm256_setzero_ps();
 188         fiy2             = _mm256_setzero_ps();
 189         fiz2             = _mm256_setzero_ps();
 190         fix3             = _mm256_setzero_ps();
 191         fiy3             = _mm256_setzero_ps();
 192         fiz3             = _mm256_setzero_ps();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm256_setzero_ps();
 196         vvdwsum          = _mm256_setzero_ps();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             jnrE             = jjnr[jidx+4];
 208             jnrF             = jjnr[jidx+5];
 209             jnrG             = jjnr[jidx+6];
 210             jnrH             = jjnr[jidx+7];
 211             j_coord_offsetA  = DIM*jnrA;
 212             j_coord_offsetB  = DIM*jnrB;
 213             j_coord_offsetC  = DIM*jnrC;
 214             j_coord_offsetD  = DIM*jnrD;
 215             j_coord_offsetE  = DIM*jnrE;
 216             j_coord_offsetF  = DIM*jnrF;
 217             j_coord_offsetG  = DIM*jnrG;
 218             j_coord_offsetH  = DIM*jnrH;
 219
 220             /* load j atom coordinates */
 221             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 222                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 223                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 224                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 225                                                  &jx0,&jy0,&jz0);
 226
 227             /* Calculate displacement vector */
 228             dx00             = _mm256_sub_ps(ix0,jx0);
 229             dy00             = _mm256_sub_ps(iy0,jy0);
 230             dz00             = _mm256_sub_ps(iz0,jz0);
 231             dx10             = _mm256_sub_ps(ix1,jx0);
 232             dy10             = _mm256_sub_ps(iy1,jy0);
 233             dz10             = _mm256_sub_ps(iz1,jz0);
 234             dx20             = _mm256_sub_ps(ix2,jx0);
 235             dy20             = _mm256_sub_ps(iy2,jy0);
 236             dz20             = _mm256_sub_ps(iz2,jz0);
 237             dx30             = _mm256_sub_ps(ix3,jx0);
 238             dy30             = _mm256_sub_ps(iy3,jy0);
 239             dz30             = _mm256_sub_ps(iz3,jz0);
 240
 241             /* Calculate squared distance and things based on it */
 242             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 243             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 244             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 245             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 246
 247             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 248             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 249             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 250
 251             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 252
 253             /* Load parameters for j particles */
 254             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 255                                                                  charge+jnrC+0,charge+jnrD+0,
 256                                                                  charge+jnrE+0,charge+jnrF+0,
 257                                                                  charge+jnrG+0,charge+jnrH+0);
 258             vdwjidx0A        = 2*vdwtype[jnrA+0];
 259             vdwjidx0B        = 2*vdwtype[jnrB+0];
 260             vdwjidx0C        = 2*vdwtype[jnrC+0];
 261             vdwjidx0D        = 2*vdwtype[jnrD+0];
 262             vdwjidx0E        = 2*vdwtype[jnrE+0];
 263             vdwjidx0F        = 2*vdwtype[jnrF+0];
 264             vdwjidx0G        = 2*vdwtype[jnrG+0];
 265             vdwjidx0H        = 2*vdwtype[jnrH+0];
 266
 267             fjx0             = _mm256_setzero_ps();
 268             fjy0             = _mm256_setzero_ps();
 269             fjz0             = _mm256_setzero_ps();
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             /* Compute parameters for interactions between i and j atoms */
 276             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 277                                             vdwioffsetptr0+vdwjidx0B,
 278                                             vdwioffsetptr0+vdwjidx0C,
 279                                             vdwioffsetptr0+vdwjidx0D,
 280                                             vdwioffsetptr0+vdwjidx0E,
 281                                             vdwioffsetptr0+vdwjidx0F,
 282                                             vdwioffsetptr0+vdwjidx0G,
 283                                             vdwioffsetptr0+vdwjidx0H,
 284                                             &c6_00,&c12_00);
 285
 286             /* LENNARD-JONES DISPERSION/REPULSION */
 287
 288             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 289             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 290             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 291             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 292             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 296
 297             fscal            = fvdw;
 298
 299             /* Calculate temporary vectorial force */
 300             tx               = _mm256_mul_ps(fscal,dx00);
 301             ty               = _mm256_mul_ps(fscal,dy00);
 302             tz               = _mm256_mul_ps(fscal,dz00);
 303
 304             /* Update vectorial force */
 305             fix0             = _mm256_add_ps(fix0,tx);
 306             fiy0             = _mm256_add_ps(fiy0,ty);
 307             fiz0             = _mm256_add_ps(fiz0,tz);
 308
 309             fjx0             = _mm256_add_ps(fjx0,tx);
 310             fjy0             = _mm256_add_ps(fjy0,ty);
 311             fjz0             = _mm256_add_ps(fjz0,tz);
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             r10              = _mm256_mul_ps(rsq10,rinv10);
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq10             = _mm256_mul_ps(iq1,jq0);
 321
 322             /* Calculate table index by multiplying r with table scale and truncate to integer */
 323             rt               = _mm256_mul_ps(r10,vftabscale);
 324             vfitab           = _mm256_cvttps_epi32(rt);
 325             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 326             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 327             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 328             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 329             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 330             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 331
 332             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 333             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 334                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 335             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 336                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 337             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 338                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 339             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 340                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 341             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 342             Heps             = _mm256_mul_ps(vfeps,H);
 343             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 344             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 345             velec            = _mm256_mul_ps(qq10,VV);
 346             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 347             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velecsum         = _mm256_add_ps(velecsum,velec);
 351
 352             fscal            = felec;
 353
 354             /* Calculate temporary vectorial force */
 355             tx               = _mm256_mul_ps(fscal,dx10);
 356             ty               = _mm256_mul_ps(fscal,dy10);
 357             tz               = _mm256_mul_ps(fscal,dz10);
 358
 359             /* Update vectorial force */
 360             fix1             = _mm256_add_ps(fix1,tx);
 361             fiy1             = _mm256_add_ps(fiy1,ty);
 362             fiz1             = _mm256_add_ps(fiz1,tz);
 363
 364             fjx0             = _mm256_add_ps(fjx0,tx);
 365             fjy0             = _mm256_add_ps(fjy0,ty);
 366             fjz0             = _mm256_add_ps(fjz0,tz);
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             r20              = _mm256_mul_ps(rsq20,rinv20);
 373
 374             /* Compute parameters for interactions between i and j atoms */
 375             qq20             = _mm256_mul_ps(iq2,jq0);
 376
 377             /* Calculate table index by multiplying r with table scale and truncate to integer */
 378             rt               = _mm256_mul_ps(r20,vftabscale);
 379             vfitab           = _mm256_cvttps_epi32(rt);
 380             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 381             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 382             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 383             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 384             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 385             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 386
 387             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 388             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 389                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 390             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 391                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 392             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 393                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 394             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 395                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 396             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 397             Heps             = _mm256_mul_ps(vfeps,H);
 398             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 399             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 400             velec            = _mm256_mul_ps(qq20,VV);
 401             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 402             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velecsum         = _mm256_add_ps(velecsum,velec);
 406
 407             fscal            = felec;
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm256_mul_ps(fscal,dx20);
 411             ty               = _mm256_mul_ps(fscal,dy20);
 412             tz               = _mm256_mul_ps(fscal,dz20);
 413
 414             /* Update vectorial force */
 415             fix2             = _mm256_add_ps(fix2,tx);
 416             fiy2             = _mm256_add_ps(fiy2,ty);
 417             fiz2             = _mm256_add_ps(fiz2,tz);
 418
 419             fjx0             = _mm256_add_ps(fjx0,tx);
 420             fjy0             = _mm256_add_ps(fjy0,ty);
 421             fjz0             = _mm256_add_ps(fjz0,tz);
 422
 423             /**************************
 424              * CALCULATE INTERACTIONS *
 425              **************************/
 426
 427             r30              = _mm256_mul_ps(rsq30,rinv30);
 428
 429             /* Compute parameters for interactions between i and j atoms */
 430             qq30             = _mm256_mul_ps(iq3,jq0);
 431
 432             /* Calculate table index by multiplying r with table scale and truncate to integer */
 433             rt               = _mm256_mul_ps(r30,vftabscale);
 434             vfitab           = _mm256_cvttps_epi32(rt);
 435             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 436             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 437             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 438             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 439             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 440             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 441
 442             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 443             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 444                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 445             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 446                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 447             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 448                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 449             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 450                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 451             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 452             Heps             = _mm256_mul_ps(vfeps,H);
 453             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 454             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 455             velec            = _mm256_mul_ps(qq30,VV);
 456             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 457             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 458
 459             /* Update potential sum for this i atom from the interaction with this j atom. */
 460             velecsum         = _mm256_add_ps(velecsum,velec);
 461
 462             fscal            = felec;
 463
 464             /* Calculate temporary vectorial force */
 465             tx               = _mm256_mul_ps(fscal,dx30);
 466             ty               = _mm256_mul_ps(fscal,dy30);
 467             tz               = _mm256_mul_ps(fscal,dz30);
 468
 469             /* Update vectorial force */
 470             fix3             = _mm256_add_ps(fix3,tx);
 471             fiy3             = _mm256_add_ps(fiy3,ty);
 472             fiz3             = _mm256_add_ps(fiz3,tz);
 473
 474             fjx0             = _mm256_add_ps(fjx0,tx);
 475             fjy0             = _mm256_add_ps(fjy0,ty);
 476             fjz0             = _mm256_add_ps(fjz0,tz);
 477
 478             fjptrA             = f+j_coord_offsetA;
 479             fjptrB             = f+j_coord_offsetB;
 480             fjptrC             = f+j_coord_offsetC;
 481             fjptrD             = f+j_coord_offsetD;
 482             fjptrE             = f+j_coord_offsetE;
 483             fjptrF             = f+j_coord_offsetF;
 484             fjptrG             = f+j_coord_offsetG;
 485             fjptrH             = f+j_coord_offsetH;
 486
 487             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 488
 489             /* Inner loop uses 164 flops */
 490         }
 491
 492         if(jidx<j_index_end)
 493         {
 494
 495             /* Get j neighbor index, and coordinate index */
 496             jnrlistA         = jjnr[jidx];
 497             jnrlistB         = jjnr[jidx+1];
 498             jnrlistC         = jjnr[jidx+2];
 499             jnrlistD         = jjnr[jidx+3];
 500             jnrlistE         = jjnr[jidx+4];
 501             jnrlistF         = jjnr[jidx+5];
 502             jnrlistG         = jjnr[jidx+6];
 503             jnrlistH         = jjnr[jidx+7];
 504             /* Sign of each element will be negative for non-real atoms.
 505              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 506              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 507              */
 508             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 509                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 510
 511             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 512             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 513             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 514             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 515             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 516             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 517             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 518             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 519             j_coord_offsetA  = DIM*jnrA;
 520             j_coord_offsetB  = DIM*jnrB;
 521             j_coord_offsetC  = DIM*jnrC;
 522             j_coord_offsetD  = DIM*jnrD;
 523             j_coord_offsetE  = DIM*jnrE;
 524             j_coord_offsetF  = DIM*jnrF;
 525             j_coord_offsetG  = DIM*jnrG;
 526             j_coord_offsetH  = DIM*jnrH;
 527
 528             /* load j atom coordinates */
 529             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 530                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 531                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 532                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 533                                                  &jx0,&jy0,&jz0);
 534
 535             /* Calculate displacement vector */
 536             dx00             = _mm256_sub_ps(ix0,jx0);
 537             dy00             = _mm256_sub_ps(iy0,jy0);
 538             dz00             = _mm256_sub_ps(iz0,jz0);
 539             dx10             = _mm256_sub_ps(ix1,jx0);
 540             dy10             = _mm256_sub_ps(iy1,jy0);
 541             dz10             = _mm256_sub_ps(iz1,jz0);
 542             dx20             = _mm256_sub_ps(ix2,jx0);
 543             dy20             = _mm256_sub_ps(iy2,jy0);
 544             dz20             = _mm256_sub_ps(iz2,jz0);
 545             dx30             = _mm256_sub_ps(ix3,jx0);
 546             dy30             = _mm256_sub_ps(iy3,jy0);
 547             dz30             = _mm256_sub_ps(iz3,jz0);
 548
 549             /* Calculate squared distance and things based on it */
 550             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 551             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 552             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 553             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 554
 555             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 556             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 557             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 558
 559             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 560
 561             /* Load parameters for j particles */
 562             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 563                                                                  charge+jnrC+0,charge+jnrD+0,
 564                                                                  charge+jnrE+0,charge+jnrF+0,
 565                                                                  charge+jnrG+0,charge+jnrH+0);
 566             vdwjidx0A        = 2*vdwtype[jnrA+0];
 567             vdwjidx0B        = 2*vdwtype[jnrB+0];
 568             vdwjidx0C        = 2*vdwtype[jnrC+0];
 569             vdwjidx0D        = 2*vdwtype[jnrD+0];
 570             vdwjidx0E        = 2*vdwtype[jnrE+0];
 571             vdwjidx0F        = 2*vdwtype[jnrF+0];
 572             vdwjidx0G        = 2*vdwtype[jnrG+0];
 573             vdwjidx0H        = 2*vdwtype[jnrH+0];
 574
 575             fjx0             = _mm256_setzero_ps();
 576             fjy0             = _mm256_setzero_ps();
 577             fjz0             = _mm256_setzero_ps();
 578
 579             /**************************
 580              * CALCULATE INTERACTIONS *
 581              **************************/
 582
 583             /* Compute parameters for interactions between i and j atoms */
 584             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 585                                             vdwioffsetptr0+vdwjidx0B,
 586                                             vdwioffsetptr0+vdwjidx0C,
 587                                             vdwioffsetptr0+vdwjidx0D,
 588                                             vdwioffsetptr0+vdwjidx0E,
 589                                             vdwioffsetptr0+vdwjidx0F,
 590                                             vdwioffsetptr0+vdwjidx0G,
 591                                             vdwioffsetptr0+vdwjidx0H,
 592                                             &c6_00,&c12_00);
 593
 594             /* LENNARD-JONES DISPERSION/REPULSION */
 595
 596             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 597             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 598             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 599             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 600             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 601
 602             /* Update potential sum for this i atom from the interaction with this j atom. */
 603             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 604             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 605
 606             fscal            = fvdw;
 607
 608             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 609
 610             /* Calculate temporary vectorial force */
 611             tx               = _mm256_mul_ps(fscal,dx00);
 612             ty               = _mm256_mul_ps(fscal,dy00);
 613             tz               = _mm256_mul_ps(fscal,dz00);
 614
 615             /* Update vectorial force */
 616             fix0             = _mm256_add_ps(fix0,tx);
 617             fiy0             = _mm256_add_ps(fiy0,ty);
 618             fiz0             = _mm256_add_ps(fiz0,tz);
 619
 620             fjx0             = _mm256_add_ps(fjx0,tx);
 621             fjy0             = _mm256_add_ps(fjy0,ty);
 622             fjz0             = _mm256_add_ps(fjz0,tz);
 623
 624             /**************************
 625              * CALCULATE INTERACTIONS *
 626              **************************/
 627
 628             r10              = _mm256_mul_ps(rsq10,rinv10);
 629             r10              = _mm256_andnot_ps(dummy_mask,r10);
 630
 631             /* Compute parameters for interactions between i and j atoms */
 632             qq10             = _mm256_mul_ps(iq1,jq0);
 633
 634             /* Calculate table index by multiplying r with table scale and truncate to integer */
 635             rt               = _mm256_mul_ps(r10,vftabscale);
 636             vfitab           = _mm256_cvttps_epi32(rt);
 637             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 638             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 639             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 640             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 641             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 642             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 643
 644             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 645             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 646                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 647             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 648                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 649             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 650                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 651             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 652                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 653             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 654             Heps             = _mm256_mul_ps(vfeps,H);
 655             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 656             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 657             velec            = _mm256_mul_ps(qq10,VV);
 658             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 659             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 660
 661             /* Update potential sum for this i atom from the interaction with this j atom. */
 662             velec            = _mm256_andnot_ps(dummy_mask,velec);
 663             velecsum         = _mm256_add_ps(velecsum,velec);
 664
 665             fscal            = felec;
 666
 667             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 668
 669             /* Calculate temporary vectorial force */
 670             tx               = _mm256_mul_ps(fscal,dx10);
 671             ty               = _mm256_mul_ps(fscal,dy10);
 672             tz               = _mm256_mul_ps(fscal,dz10);
 673
 674             /* Update vectorial force */
 675             fix1             = _mm256_add_ps(fix1,tx);
 676             fiy1             = _mm256_add_ps(fiy1,ty);
 677             fiz1             = _mm256_add_ps(fiz1,tz);
 678
 679             fjx0             = _mm256_add_ps(fjx0,tx);
 680             fjy0             = _mm256_add_ps(fjy0,ty);
 681             fjz0             = _mm256_add_ps(fjz0,tz);
 682
 683             /**************************
 684              * CALCULATE INTERACTIONS *
 685              **************************/
 686
 687             r20              = _mm256_mul_ps(rsq20,rinv20);
 688             r20              = _mm256_andnot_ps(dummy_mask,r20);
 689
 690             /* Compute parameters for interactions between i and j atoms */
 691             qq20             = _mm256_mul_ps(iq2,jq0);
 692
 693             /* Calculate table index by multiplying r with table scale and truncate to integer */
 694             rt               = _mm256_mul_ps(r20,vftabscale);
 695             vfitab           = _mm256_cvttps_epi32(rt);
 696             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 697             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 698             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 699             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 700             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 701             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 702
 703             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 704             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 705                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 706             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 707                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 708             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 709                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 710             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 711                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 712             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 713             Heps             = _mm256_mul_ps(vfeps,H);
 714             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 715             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 716             velec            = _mm256_mul_ps(qq20,VV);
 717             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 718             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 719
 720             /* Update potential sum for this i atom from the interaction with this j atom. */
 721             velec            = _mm256_andnot_ps(dummy_mask,velec);
 722             velecsum         = _mm256_add_ps(velecsum,velec);
 723
 724             fscal            = felec;
 725
 726             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 727
 728             /* Calculate temporary vectorial force */
 729             tx               = _mm256_mul_ps(fscal,dx20);
 730             ty               = _mm256_mul_ps(fscal,dy20);
 731             tz               = _mm256_mul_ps(fscal,dz20);
 732
 733             /* Update vectorial force */
 734             fix2             = _mm256_add_ps(fix2,tx);
 735             fiy2             = _mm256_add_ps(fiy2,ty);
 736             fiz2             = _mm256_add_ps(fiz2,tz);
 737
 738             fjx0             = _mm256_add_ps(fjx0,tx);
 739             fjy0             = _mm256_add_ps(fjy0,ty);
 740             fjz0             = _mm256_add_ps(fjz0,tz);
 741
 742             /**************************
 743              * CALCULATE INTERACTIONS *
 744              **************************/
 745
 746             r30              = _mm256_mul_ps(rsq30,rinv30);
 747             r30              = _mm256_andnot_ps(dummy_mask,r30);
 748
 749             /* Compute parameters for interactions between i and j atoms */
 750             qq30             = _mm256_mul_ps(iq3,jq0);
 751
 752             /* Calculate table index by multiplying r with table scale and truncate to integer */
 753             rt               = _mm256_mul_ps(r30,vftabscale);
 754             vfitab           = _mm256_cvttps_epi32(rt);
 755             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 756             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 757             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 758             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 759             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 760             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 761
 762             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 763             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 764                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 765             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 766                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 767             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 768                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 769             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 770                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 771             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 772             Heps             = _mm256_mul_ps(vfeps,H);
 773             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 774             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 775             velec            = _mm256_mul_ps(qq30,VV);
 776             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 777             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 778
 779             /* Update potential sum for this i atom from the interaction with this j atom. */
 780             velec            = _mm256_andnot_ps(dummy_mask,velec);
 781             velecsum         = _mm256_add_ps(velecsum,velec);
 782
 783             fscal            = felec;
 784
 785             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 786
 787             /* Calculate temporary vectorial force */
 788             tx               = _mm256_mul_ps(fscal,dx30);
 789             ty               = _mm256_mul_ps(fscal,dy30);
 790             tz               = _mm256_mul_ps(fscal,dz30);
 791
 792             /* Update vectorial force */
 793             fix3             = _mm256_add_ps(fix3,tx);
 794             fiy3             = _mm256_add_ps(fiy3,ty);
 795             fiz3             = _mm256_add_ps(fiz3,tz);
 796
 797             fjx0             = _mm256_add_ps(fjx0,tx);
 798             fjy0             = _mm256_add_ps(fjy0,ty);
 799             fjz0             = _mm256_add_ps(fjz0,tz);
 800
 801             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 802             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 803             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 804             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 805             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 806             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 807             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 808             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 809
 810             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 811
 812             /* Inner loop uses 167 flops */
 813         }
 814
 815         /* End of innermost loop */
 816
 817         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 818                                                  f+i_coord_offset,fshift+i_shift_offset);
 819
 820         ggid                        = gid[iidx];
 821         /* Update potential energies */
 822         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 823         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 824
 825         /* Increment number of inner iterations */
 826         inneriter                  += j_index_end - j_index_start;
 827
 828         /* Outer loop uses 26 flops */
 829     }
 830
 831     /* Increment number of outer iterations */
 832     outeriter        += nri;
 833
 834     /* Update outer/inner flops */
 835
 836     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*167);
 837 }
 838 /*
 839  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_256_single
 840  * Electrostatics interaction: CubicSplineTable
 841  * VdW interaction:            LennardJones
 842  * Geometry:                   Water4-Particle
 843  * Calculate force/pot:        Force
 844  */
 845 void
 846 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_256_single
 847                     (t_nblist                    * gmx_restrict       nlist,
 848                      rvec                        * gmx_restrict          xx,
 849                      rvec                        * gmx_restrict          ff,
 850                      t_forcerec                  * gmx_restrict          fr,
 851                      t_mdatoms                   * gmx_restrict     mdatoms,
 852                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 853                      t_nrnb                      * gmx_restrict        nrnb)
 854 {
 855     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 856      * just 0 for non-waters.
 857      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 858      * jnr indices corresponding to data put in the four positions in the SIMD register.
 859      */
 860     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 861     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 862     int              jnrA,jnrB,jnrC,jnrD;
 863     int              jnrE,jnrF,jnrG,jnrH;
 864     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 865     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 866     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 867     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 868     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 869     real             rcutoff_scalar;
 870     real             *shiftvec,*fshift,*x,*f;
 871     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 872     real             scratch[4*DIM];
 873     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 874     real *           vdwioffsetptr0;
 875     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 876     real *           vdwioffsetptr1;
 877     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 878     real *           vdwioffsetptr2;
 879     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 880     real *           vdwioffsetptr3;
 881     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 882     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 883     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 884     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 885     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 886     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 887     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 888     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 889     real             *charge;
 890     int              nvdwtype;
 891     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 892     int              *vdwtype;
 893     real             *vdwparam;
 894     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 895     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 896     __m256i          vfitab;
 897     __m128i          vfitab_lo,vfitab_hi;
 898     __m128i          ifour       = _mm_set1_epi32(4);
 899     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 900     real             *vftab;
 901     __m256           dummy_mask,cutoff_mask;
 902     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 903     __m256           one     = _mm256_set1_ps(1.0);
 904     __m256           two     = _mm256_set1_ps(2.0);
 905     x                = xx[0];
 906     f                = ff[0];
 907
 908     nri              = nlist->nri;
 909     iinr             = nlist->iinr;
 910     jindex           = nlist->jindex;
 911     jjnr             = nlist->jjnr;
 912     shiftidx         = nlist->shift;
 913     gid              = nlist->gid;
 914     shiftvec         = fr->shift_vec[0];
 915     fshift           = fr->fshift[0];
 916     facel            = _mm256_set1_ps(fr->epsfac);
 917     charge           = mdatoms->chargeA;
 918     nvdwtype         = fr->ntype;
 919     vdwparam         = fr->nbfp;
 920     vdwtype          = mdatoms->typeA;
 921
 922     vftab            = kernel_data->table_elec->data;
 923     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 924
 925     /* Setup water-specific parameters */
 926     inr              = nlist->iinr[0];
 927     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 928     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 929     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 930     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 931
 932     /* Avoid stupid compiler warnings */
 933     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 934     j_coord_offsetA = 0;
 935     j_coord_offsetB = 0;
 936     j_coord_offsetC = 0;
 937     j_coord_offsetD = 0;
 938     j_coord_offsetE = 0;
 939     j_coord_offsetF = 0;
 940     j_coord_offsetG = 0;
 941     j_coord_offsetH = 0;
 942
 943     outeriter        = 0;
 944     inneriter        = 0;
 945
 946     for(iidx=0;iidx<4*DIM;iidx++)
 947     {
 948         scratch[iidx] = 0.0;
 949     }
 950
 951     /* Start outer loop over neighborlists */
 952     for(iidx=0; iidx<nri; iidx++)
 953     {
 954         /* Load shift vector for this list */
 955         i_shift_offset   = DIM*shiftidx[iidx];
 956
 957         /* Load limits for loop over neighbors */
 958         j_index_start    = jindex[iidx];
 959         j_index_end      = jindex[iidx+1];
 960
 961         /* Get outer coordinate index */
 962         inr              = iinr[iidx];
 963         i_coord_offset   = DIM*inr;
 964
 965         /* Load i particle coords and add shift vector */
 966         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 967                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 968
 969         fix0             = _mm256_setzero_ps();
 970         fiy0             = _mm256_setzero_ps();
 971         fiz0             = _mm256_setzero_ps();
 972         fix1             = _mm256_setzero_ps();
 973         fiy1             = _mm256_setzero_ps();
 974         fiz1             = _mm256_setzero_ps();
 975         fix2             = _mm256_setzero_ps();
 976         fiy2             = _mm256_setzero_ps();
 977         fiz2             = _mm256_setzero_ps();
 978         fix3             = _mm256_setzero_ps();
 979         fiy3             = _mm256_setzero_ps();
 980         fiz3             = _mm256_setzero_ps();
 981
 982         /* Start inner kernel loop */
 983         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 984         {
 985
 986             /* Get j neighbor index, and coordinate index */
 987             jnrA             = jjnr[jidx];
 988             jnrB             = jjnr[jidx+1];
 989             jnrC             = jjnr[jidx+2];
 990             jnrD             = jjnr[jidx+3];
 991             jnrE             = jjnr[jidx+4];
 992             jnrF             = jjnr[jidx+5];
 993             jnrG             = jjnr[jidx+6];
 994             jnrH             = jjnr[jidx+7];
 995             j_coord_offsetA  = DIM*jnrA;
 996             j_coord_offsetB  = DIM*jnrB;
 997             j_coord_offsetC  = DIM*jnrC;
 998             j_coord_offsetD  = DIM*jnrD;
 999             j_coord_offsetE  = DIM*jnrE;
1000             j_coord_offsetF  = DIM*jnrF;
1001             j_coord_offsetG  = DIM*jnrG;
1002             j_coord_offsetH  = DIM*jnrH;
1003
1004             /* load j atom coordinates */
1005             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1006                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1007                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1008                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1009                                                  &jx0,&jy0,&jz0);
1010
1011             /* Calculate displacement vector */
1012             dx00             = _mm256_sub_ps(ix0,jx0);
1013             dy00             = _mm256_sub_ps(iy0,jy0);
1014             dz00             = _mm256_sub_ps(iz0,jz0);
1015             dx10             = _mm256_sub_ps(ix1,jx0);
1016             dy10             = _mm256_sub_ps(iy1,jy0);
1017             dz10             = _mm256_sub_ps(iz1,jz0);
1018             dx20             = _mm256_sub_ps(ix2,jx0);
1019             dy20             = _mm256_sub_ps(iy2,jy0);
1020             dz20             = _mm256_sub_ps(iz2,jz0);
1021             dx30             = _mm256_sub_ps(ix3,jx0);
1022             dy30             = _mm256_sub_ps(iy3,jy0);
1023             dz30             = _mm256_sub_ps(iz3,jz0);
1024
1025             /* Calculate squared distance and things based on it */
1026             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1027             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1028             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1029             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1030
1031             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1032             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1033             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1034
1035             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1036
1037             /* Load parameters for j particles */
1038             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1039                                                                  charge+jnrC+0,charge+jnrD+0,
1040                                                                  charge+jnrE+0,charge+jnrF+0,
1041                                                                  charge+jnrG+0,charge+jnrH+0);
1042             vdwjidx0A        = 2*vdwtype[jnrA+0];
1043             vdwjidx0B        = 2*vdwtype[jnrB+0];
1044             vdwjidx0C        = 2*vdwtype[jnrC+0];
1045             vdwjidx0D        = 2*vdwtype[jnrD+0];
1046             vdwjidx0E        = 2*vdwtype[jnrE+0];
1047             vdwjidx0F        = 2*vdwtype[jnrF+0];
1048             vdwjidx0G        = 2*vdwtype[jnrG+0];
1049             vdwjidx0H        = 2*vdwtype[jnrH+0];
1050
1051             fjx0             = _mm256_setzero_ps();
1052             fjy0             = _mm256_setzero_ps();
1053             fjz0             = _mm256_setzero_ps();
1054
1055             /**************************
1056              * CALCULATE INTERACTIONS *
1057              **************************/
1058
1059             /* Compute parameters for interactions between i and j atoms */
1060             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1061                                             vdwioffsetptr0+vdwjidx0B,
1062                                             vdwioffsetptr0+vdwjidx0C,
1063                                             vdwioffsetptr0+vdwjidx0D,
1064                                             vdwioffsetptr0+vdwjidx0E,
1065                                             vdwioffsetptr0+vdwjidx0F,
1066                                             vdwioffsetptr0+vdwjidx0G,
1067                                             vdwioffsetptr0+vdwjidx0H,
1068                                             &c6_00,&c12_00);
1069
1070             /* LENNARD-JONES DISPERSION/REPULSION */
1071
1072             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1073             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1074
1075             fscal            = fvdw;
1076
1077             /* Calculate temporary vectorial force */
1078             tx               = _mm256_mul_ps(fscal,dx00);
1079             ty               = _mm256_mul_ps(fscal,dy00);
1080             tz               = _mm256_mul_ps(fscal,dz00);
1081
1082             /* Update vectorial force */
1083             fix0             = _mm256_add_ps(fix0,tx);
1084             fiy0             = _mm256_add_ps(fiy0,ty);
1085             fiz0             = _mm256_add_ps(fiz0,tz);
1086
1087             fjx0             = _mm256_add_ps(fjx0,tx);
1088             fjy0             = _mm256_add_ps(fjy0,ty);
1089             fjz0             = _mm256_add_ps(fjz0,tz);
1090
1091             /**************************
1092              * CALCULATE INTERACTIONS *
1093              **************************/
1094
1095             r10              = _mm256_mul_ps(rsq10,rinv10);
1096
1097             /* Compute parameters for interactions between i and j atoms */
1098             qq10             = _mm256_mul_ps(iq1,jq0);
1099
1100             /* Calculate table index by multiplying r with table scale and truncate to integer */
1101             rt               = _mm256_mul_ps(r10,vftabscale);
1102             vfitab           = _mm256_cvttps_epi32(rt);
1103             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1104             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1105             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1106             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1107             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1108             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1109
1110             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1111             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1112                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1113             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1114                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1115             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1116                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1117             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1118                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1119             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1120             Heps             = _mm256_mul_ps(vfeps,H);
1121             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1122             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1123             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1124
1125             fscal            = felec;
1126
1127             /* Calculate temporary vectorial force */
1128             tx               = _mm256_mul_ps(fscal,dx10);
1129             ty               = _mm256_mul_ps(fscal,dy10);
1130             tz               = _mm256_mul_ps(fscal,dz10);
1131
1132             /* Update vectorial force */
1133             fix1             = _mm256_add_ps(fix1,tx);
1134             fiy1             = _mm256_add_ps(fiy1,ty);
1135             fiz1             = _mm256_add_ps(fiz1,tz);
1136
1137             fjx0             = _mm256_add_ps(fjx0,tx);
1138             fjy0             = _mm256_add_ps(fjy0,ty);
1139             fjz0             = _mm256_add_ps(fjz0,tz);
1140
1141             /**************************
1142              * CALCULATE INTERACTIONS *
1143              **************************/
1144
1145             r20              = _mm256_mul_ps(rsq20,rinv20);
1146
1147             /* Compute parameters for interactions between i and j atoms */
1148             qq20             = _mm256_mul_ps(iq2,jq0);
1149
1150             /* Calculate table index by multiplying r with table scale and truncate to integer */
1151             rt               = _mm256_mul_ps(r20,vftabscale);
1152             vfitab           = _mm256_cvttps_epi32(rt);
1153             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1154             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1155             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1156             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1157             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1158             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1159
1160             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1161             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1162                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1163             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1164                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1165             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1166                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1167             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1168                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1169             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1170             Heps             = _mm256_mul_ps(vfeps,H);
1171             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1172             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1173             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1174
1175             fscal            = felec;
1176
1177             /* Calculate temporary vectorial force */
1178             tx               = _mm256_mul_ps(fscal,dx20);
1179             ty               = _mm256_mul_ps(fscal,dy20);
1180             tz               = _mm256_mul_ps(fscal,dz20);
1181
1182             /* Update vectorial force */
1183             fix2             = _mm256_add_ps(fix2,tx);
1184             fiy2             = _mm256_add_ps(fiy2,ty);
1185             fiz2             = _mm256_add_ps(fiz2,tz);
1186
1187             fjx0             = _mm256_add_ps(fjx0,tx);
1188             fjy0             = _mm256_add_ps(fjy0,ty);
1189             fjz0             = _mm256_add_ps(fjz0,tz);
1190
1191             /**************************
1192              * CALCULATE INTERACTIONS *
1193              **************************/
1194
1195             r30              = _mm256_mul_ps(rsq30,rinv30);
1196
1197             /* Compute parameters for interactions between i and j atoms */
1198             qq30             = _mm256_mul_ps(iq3,jq0);
1199
1200             /* Calculate table index by multiplying r with table scale and truncate to integer */
1201             rt               = _mm256_mul_ps(r30,vftabscale);
1202             vfitab           = _mm256_cvttps_epi32(rt);
1203             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1204             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1205             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1206             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1207             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1208             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1209
1210             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1211             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1212                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1213             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1214                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1215             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1216                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1217             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1218                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1219             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1220             Heps             = _mm256_mul_ps(vfeps,H);
1221             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1222             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1223             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1224
1225             fscal            = felec;
1226
1227             /* Calculate temporary vectorial force */
1228             tx               = _mm256_mul_ps(fscal,dx30);
1229             ty               = _mm256_mul_ps(fscal,dy30);
1230             tz               = _mm256_mul_ps(fscal,dz30);
1231
1232             /* Update vectorial force */
1233             fix3             = _mm256_add_ps(fix3,tx);
1234             fiy3             = _mm256_add_ps(fiy3,ty);
1235             fiz3             = _mm256_add_ps(fiz3,tz);
1236
1237             fjx0             = _mm256_add_ps(fjx0,tx);
1238             fjy0             = _mm256_add_ps(fjy0,ty);
1239             fjz0             = _mm256_add_ps(fjz0,tz);
1240
1241             fjptrA             = f+j_coord_offsetA;
1242             fjptrB             = f+j_coord_offsetB;
1243             fjptrC             = f+j_coord_offsetC;
1244             fjptrD             = f+j_coord_offsetD;
1245             fjptrE             = f+j_coord_offsetE;
1246             fjptrF             = f+j_coord_offsetF;
1247             fjptrG             = f+j_coord_offsetG;
1248             fjptrH             = f+j_coord_offsetH;
1249
1250             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1251
1252             /* Inner loop uses 147 flops */
1253         }
1254
1255         if(jidx<j_index_end)
1256         {
1257
1258             /* Get j neighbor index, and coordinate index */
1259             jnrlistA         = jjnr[jidx];
1260             jnrlistB         = jjnr[jidx+1];
1261             jnrlistC         = jjnr[jidx+2];
1262             jnrlistD         = jjnr[jidx+3];
1263             jnrlistE         = jjnr[jidx+4];
1264             jnrlistF         = jjnr[jidx+5];
1265             jnrlistG         = jjnr[jidx+6];
1266             jnrlistH         = jjnr[jidx+7];
1267             /* Sign of each element will be negative for non-real atoms.
1268              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1269              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1270              */
1271             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1272                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1273
1274             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1275             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1276             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1277             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1278             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1279             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1280             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1281             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1282             j_coord_offsetA  = DIM*jnrA;
1283             j_coord_offsetB  = DIM*jnrB;
1284             j_coord_offsetC  = DIM*jnrC;
1285             j_coord_offsetD  = DIM*jnrD;
1286             j_coord_offsetE  = DIM*jnrE;
1287             j_coord_offsetF  = DIM*jnrF;
1288             j_coord_offsetG  = DIM*jnrG;
1289             j_coord_offsetH  = DIM*jnrH;
1290
1291             /* load j atom coordinates */
1292             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1293                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1294                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1295                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1296                                                  &jx0,&jy0,&jz0);
1297
1298             /* Calculate displacement vector */
1299             dx00             = _mm256_sub_ps(ix0,jx0);
1300             dy00             = _mm256_sub_ps(iy0,jy0);
1301             dz00             = _mm256_sub_ps(iz0,jz0);
1302             dx10             = _mm256_sub_ps(ix1,jx0);
1303             dy10             = _mm256_sub_ps(iy1,jy0);
1304             dz10             = _mm256_sub_ps(iz1,jz0);
1305             dx20             = _mm256_sub_ps(ix2,jx0);
1306             dy20             = _mm256_sub_ps(iy2,jy0);
1307             dz20             = _mm256_sub_ps(iz2,jz0);
1308             dx30             = _mm256_sub_ps(ix3,jx0);
1309             dy30             = _mm256_sub_ps(iy3,jy0);
1310             dz30             = _mm256_sub_ps(iz3,jz0);
1311
1312             /* Calculate squared distance and things based on it */
1313             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1314             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1315             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1316             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1317
1318             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1319             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1320             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1321
1322             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1323
1324             /* Load parameters for j particles */
1325             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1326                                                                  charge+jnrC+0,charge+jnrD+0,
1327                                                                  charge+jnrE+0,charge+jnrF+0,
1328                                                                  charge+jnrG+0,charge+jnrH+0);
1329             vdwjidx0A        = 2*vdwtype[jnrA+0];
1330             vdwjidx0B        = 2*vdwtype[jnrB+0];
1331             vdwjidx0C        = 2*vdwtype[jnrC+0];
1332             vdwjidx0D        = 2*vdwtype[jnrD+0];
1333             vdwjidx0E        = 2*vdwtype[jnrE+0];
1334             vdwjidx0F        = 2*vdwtype[jnrF+0];
1335             vdwjidx0G        = 2*vdwtype[jnrG+0];
1336             vdwjidx0H        = 2*vdwtype[jnrH+0];
1337
1338             fjx0             = _mm256_setzero_ps();
1339             fjy0             = _mm256_setzero_ps();
1340             fjz0             = _mm256_setzero_ps();
1341
1342             /**************************
1343              * CALCULATE INTERACTIONS *
1344              **************************/
1345
1346             /* Compute parameters for interactions between i and j atoms */
1347             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1348                                             vdwioffsetptr0+vdwjidx0B,
1349                                             vdwioffsetptr0+vdwjidx0C,
1350                                             vdwioffsetptr0+vdwjidx0D,
1351                                             vdwioffsetptr0+vdwjidx0E,
1352                                             vdwioffsetptr0+vdwjidx0F,
1353                                             vdwioffsetptr0+vdwjidx0G,
1354                                             vdwioffsetptr0+vdwjidx0H,
1355                                             &c6_00,&c12_00);
1356
1357             /* LENNARD-JONES DISPERSION/REPULSION */
1358
1359             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1360             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1361
1362             fscal            = fvdw;
1363
1364             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1365
1366             /* Calculate temporary vectorial force */
1367             tx               = _mm256_mul_ps(fscal,dx00);
1368             ty               = _mm256_mul_ps(fscal,dy00);
1369             tz               = _mm256_mul_ps(fscal,dz00);
1370
1371             /* Update vectorial force */
1372             fix0             = _mm256_add_ps(fix0,tx);
1373             fiy0             = _mm256_add_ps(fiy0,ty);
1374             fiz0             = _mm256_add_ps(fiz0,tz);
1375
1376             fjx0             = _mm256_add_ps(fjx0,tx);
1377             fjy0             = _mm256_add_ps(fjy0,ty);
1378             fjz0             = _mm256_add_ps(fjz0,tz);
1379
1380             /**************************
1381              * CALCULATE INTERACTIONS *
1382              **************************/
1383
1384             r10              = _mm256_mul_ps(rsq10,rinv10);
1385             r10              = _mm256_andnot_ps(dummy_mask,r10);
1386
1387             /* Compute parameters for interactions between i and j atoms */
1388             qq10             = _mm256_mul_ps(iq1,jq0);
1389
1390             /* Calculate table index by multiplying r with table scale and truncate to integer */
1391             rt               = _mm256_mul_ps(r10,vftabscale);
1392             vfitab           = _mm256_cvttps_epi32(rt);
1393             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1394             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1395             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1396             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1397             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1398             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1399
1400             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1401             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1402                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1403             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1404                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1405             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1406                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1407             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1408                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1409             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1410             Heps             = _mm256_mul_ps(vfeps,H);
1411             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1412             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1413             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1414
1415             fscal            = felec;
1416
1417             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1418
1419             /* Calculate temporary vectorial force */
1420             tx               = _mm256_mul_ps(fscal,dx10);
1421             ty               = _mm256_mul_ps(fscal,dy10);
1422             tz               = _mm256_mul_ps(fscal,dz10);
1423
1424             /* Update vectorial force */
1425             fix1             = _mm256_add_ps(fix1,tx);
1426             fiy1             = _mm256_add_ps(fiy1,ty);
1427             fiz1             = _mm256_add_ps(fiz1,tz);
1428
1429             fjx0             = _mm256_add_ps(fjx0,tx);
1430             fjy0             = _mm256_add_ps(fjy0,ty);
1431             fjz0             = _mm256_add_ps(fjz0,tz);
1432
1433             /**************************
1434              * CALCULATE INTERACTIONS *
1435              **************************/
1436
1437             r20              = _mm256_mul_ps(rsq20,rinv20);
1438             r20              = _mm256_andnot_ps(dummy_mask,r20);
1439
1440             /* Compute parameters for interactions between i and j atoms */
1441             qq20             = _mm256_mul_ps(iq2,jq0);
1442
1443             /* Calculate table index by multiplying r with table scale and truncate to integer */
1444             rt               = _mm256_mul_ps(r20,vftabscale);
1445             vfitab           = _mm256_cvttps_epi32(rt);
1446             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1447             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1448             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1449             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1450             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1451             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1452
1453             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1454             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1455                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1456             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1457                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1458             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1459                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1460             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1461                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1462             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1463             Heps             = _mm256_mul_ps(vfeps,H);
1464             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1465             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1466             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1467
1468             fscal            = felec;
1469
1470             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1471
1472             /* Calculate temporary vectorial force */
1473             tx               = _mm256_mul_ps(fscal,dx20);
1474             ty               = _mm256_mul_ps(fscal,dy20);
1475             tz               = _mm256_mul_ps(fscal,dz20);
1476
1477             /* Update vectorial force */
1478             fix2             = _mm256_add_ps(fix2,tx);
1479             fiy2             = _mm256_add_ps(fiy2,ty);
1480             fiz2             = _mm256_add_ps(fiz2,tz);
1481
1482             fjx0             = _mm256_add_ps(fjx0,tx);
1483             fjy0             = _mm256_add_ps(fjy0,ty);
1484             fjz0             = _mm256_add_ps(fjz0,tz);
1485
1486             /**************************
1487              * CALCULATE INTERACTIONS *
1488              **************************/
1489
1490             r30              = _mm256_mul_ps(rsq30,rinv30);
1491             r30              = _mm256_andnot_ps(dummy_mask,r30);
1492
1493             /* Compute parameters for interactions between i and j atoms */
1494             qq30             = _mm256_mul_ps(iq3,jq0);
1495
1496             /* Calculate table index by multiplying r with table scale and truncate to integer */
1497             rt               = _mm256_mul_ps(r30,vftabscale);
1498             vfitab           = _mm256_cvttps_epi32(rt);
1499             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1500             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1501             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1502             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1503             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1504             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1505
1506             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1507             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1508                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1509             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1510                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1511             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1512                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1513             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1514                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1515             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1516             Heps             = _mm256_mul_ps(vfeps,H);
1517             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1518             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1519             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1520
1521             fscal            = felec;
1522
1523             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1524
1525             /* Calculate temporary vectorial force */
1526             tx               = _mm256_mul_ps(fscal,dx30);
1527             ty               = _mm256_mul_ps(fscal,dy30);
1528             tz               = _mm256_mul_ps(fscal,dz30);
1529
1530             /* Update vectorial force */
1531             fix3             = _mm256_add_ps(fix3,tx);
1532             fiy3             = _mm256_add_ps(fiy3,ty);
1533             fiz3             = _mm256_add_ps(fiz3,tz);
1534
1535             fjx0             = _mm256_add_ps(fjx0,tx);
1536             fjy0             = _mm256_add_ps(fjy0,ty);
1537             fjz0             = _mm256_add_ps(fjz0,tz);
1538
1539             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1540             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1541             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1542             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1543             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1544             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1545             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1546             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1547
1548             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1549
1550             /* Inner loop uses 150 flops */
1551         }
1552
1553         /* End of innermost loop */
1554
1555         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1556                                                  f+i_coord_offset,fshift+i_shift_offset);
1557
1558         /* Increment number of inner iterations */
1559         inneriter                  += j_index_end - j_index_start;
1560
1561         /* Outer loop uses 24 flops */
1562     }
1563
1564     /* Increment number of outer iterations */
1565     outeriter        += nri;
1566
1567     /* Update outer/inner flops */
1568
1569     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1570 }