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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_avx_256_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     real *           vdwioffsetptr3;
  95     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  96     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  97     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  98     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 100     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 101     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 102     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 103     real             *charge;
 104     int              nvdwtype;
 105     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 106     int              *vdwtype;
 107     real             *vdwparam;
 108     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 109     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 110     __m256i          vfitab;
 111     __m128i          vfitab_lo,vfitab_hi;
 112     __m128i          ifour       = _mm_set1_epi32(4);
 113     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 114     real             *vftab;
 115     __m256           dummy_mask,cutoff_mask;
 116     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 117     __m256           one     = _mm256_set1_ps(1.0);
 118     __m256           two     = _mm256_set1_ps(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm256_set1_ps(fr->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135
 136     vftab            = kernel_data->table_elec->data;
 137     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 138
 139     /* Setup water-specific parameters */
 140     inr              = nlist->iinr[0];
 141     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 142     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 143     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 144     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150     j_coord_offsetC = 0;
 151     j_coord_offsetD = 0;
 152     j_coord_offsetE = 0;
 153     j_coord_offsetF = 0;
 154     j_coord_offsetG = 0;
 155     j_coord_offsetH = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 181                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 182
 183         fix0             = _mm256_setzero_ps();
 184         fiy0             = _mm256_setzero_ps();
 185         fiz0             = _mm256_setzero_ps();
 186         fix1             = _mm256_setzero_ps();
 187         fiy1             = _mm256_setzero_ps();
 188         fiz1             = _mm256_setzero_ps();
 189         fix2             = _mm256_setzero_ps();
 190         fiy2             = _mm256_setzero_ps();
 191         fiz2             = _mm256_setzero_ps();
 192         fix3             = _mm256_setzero_ps();
 193         fiy3             = _mm256_setzero_ps();
 194         fiz3             = _mm256_setzero_ps();
 195
 196         /* Reset potential sums */
 197         velecsum         = _mm256_setzero_ps();
 198         vvdwsum          = _mm256_setzero_ps();
 199
 200         /* Start inner kernel loop */
 201         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 202         {
 203
 204             /* Get j neighbor index, and coordinate index */
 205             jnrA             = jjnr[jidx];
 206             jnrB             = jjnr[jidx+1];
 207             jnrC             = jjnr[jidx+2];
 208             jnrD             = jjnr[jidx+3];
 209             jnrE             = jjnr[jidx+4];
 210             jnrF             = jjnr[jidx+5];
 211             jnrG             = jjnr[jidx+6];
 212             jnrH             = jjnr[jidx+7];
 213             j_coord_offsetA  = DIM*jnrA;
 214             j_coord_offsetB  = DIM*jnrB;
 215             j_coord_offsetC  = DIM*jnrC;
 216             j_coord_offsetD  = DIM*jnrD;
 217             j_coord_offsetE  = DIM*jnrE;
 218             j_coord_offsetF  = DIM*jnrF;
 219             j_coord_offsetG  = DIM*jnrG;
 220             j_coord_offsetH  = DIM*jnrH;
 221
 222             /* load j atom coordinates */
 223             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 224                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 225                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 226                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 227                                                  &jx0,&jy0,&jz0);
 228
 229             /* Calculate displacement vector */
 230             dx00             = _mm256_sub_ps(ix0,jx0);
 231             dy00             = _mm256_sub_ps(iy0,jy0);
 232             dz00             = _mm256_sub_ps(iz0,jz0);
 233             dx10             = _mm256_sub_ps(ix1,jx0);
 234             dy10             = _mm256_sub_ps(iy1,jy0);
 235             dz10             = _mm256_sub_ps(iz1,jz0);
 236             dx20             = _mm256_sub_ps(ix2,jx0);
 237             dy20             = _mm256_sub_ps(iy2,jy0);
 238             dz20             = _mm256_sub_ps(iz2,jz0);
 239             dx30             = _mm256_sub_ps(ix3,jx0);
 240             dy30             = _mm256_sub_ps(iy3,jy0);
 241             dz30             = _mm256_sub_ps(iz3,jz0);
 242
 243             /* Calculate squared distance and things based on it */
 244             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 245             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 246             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 247             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 248
 249             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 250             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 251             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 252
 253             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 254
 255             /* Load parameters for j particles */
 256             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 257                                                                  charge+jnrC+0,charge+jnrD+0,
 258                                                                  charge+jnrE+0,charge+jnrF+0,
 259                                                                  charge+jnrG+0,charge+jnrH+0);
 260             vdwjidx0A        = 2*vdwtype[jnrA+0];
 261             vdwjidx0B        = 2*vdwtype[jnrB+0];
 262             vdwjidx0C        = 2*vdwtype[jnrC+0];
 263             vdwjidx0D        = 2*vdwtype[jnrD+0];
 264             vdwjidx0E        = 2*vdwtype[jnrE+0];
 265             vdwjidx0F        = 2*vdwtype[jnrF+0];
 266             vdwjidx0G        = 2*vdwtype[jnrG+0];
 267             vdwjidx0H        = 2*vdwtype[jnrH+0];
 268
 269             fjx0             = _mm256_setzero_ps();
 270             fjy0             = _mm256_setzero_ps();
 271             fjz0             = _mm256_setzero_ps();
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             /* Compute parameters for interactions between i and j atoms */
 278             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 279                                             vdwioffsetptr0+vdwjidx0B,
 280                                             vdwioffsetptr0+vdwjidx0C,
 281                                             vdwioffsetptr0+vdwjidx0D,
 282                                             vdwioffsetptr0+vdwjidx0E,
 283                                             vdwioffsetptr0+vdwjidx0F,
 284                                             vdwioffsetptr0+vdwjidx0G,
 285                                             vdwioffsetptr0+vdwjidx0H,
 286                                             &c6_00,&c12_00);
 287
 288             /* LENNARD-JONES DISPERSION/REPULSION */
 289
 290             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 291             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 292             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 293             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 294             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 298
 299             fscal            = fvdw;
 300
 301             /* Calculate temporary vectorial force */
 302             tx               = _mm256_mul_ps(fscal,dx00);
 303             ty               = _mm256_mul_ps(fscal,dy00);
 304             tz               = _mm256_mul_ps(fscal,dz00);
 305
 306             /* Update vectorial force */
 307             fix0             = _mm256_add_ps(fix0,tx);
 308             fiy0             = _mm256_add_ps(fiy0,ty);
 309             fiz0             = _mm256_add_ps(fiz0,tz);
 310
 311             fjx0             = _mm256_add_ps(fjx0,tx);
 312             fjy0             = _mm256_add_ps(fjy0,ty);
 313             fjz0             = _mm256_add_ps(fjz0,tz);
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             r10              = _mm256_mul_ps(rsq10,rinv10);
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq10             = _mm256_mul_ps(iq1,jq0);
 323
 324             /* Calculate table index by multiplying r with table scale and truncate to integer */
 325             rt               = _mm256_mul_ps(r10,vftabscale);
 326             vfitab           = _mm256_cvttps_epi32(rt);
 327             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 328             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 329             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 330             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 331             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 332             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 333
 334             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 335             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 336                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 337             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 338                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 339             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 340                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 341             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 342                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 343             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 344             Heps             = _mm256_mul_ps(vfeps,H);
 345             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 346             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 347             velec            = _mm256_mul_ps(qq10,VV);
 348             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 349             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velecsum         = _mm256_add_ps(velecsum,velec);
 353
 354             fscal            = felec;
 355
 356             /* Calculate temporary vectorial force */
 357             tx               = _mm256_mul_ps(fscal,dx10);
 358             ty               = _mm256_mul_ps(fscal,dy10);
 359             tz               = _mm256_mul_ps(fscal,dz10);
 360
 361             /* Update vectorial force */
 362             fix1             = _mm256_add_ps(fix1,tx);
 363             fiy1             = _mm256_add_ps(fiy1,ty);
 364             fiz1             = _mm256_add_ps(fiz1,tz);
 365
 366             fjx0             = _mm256_add_ps(fjx0,tx);
 367             fjy0             = _mm256_add_ps(fjy0,ty);
 368             fjz0             = _mm256_add_ps(fjz0,tz);
 369
 370             /**************************
 371              * CALCULATE INTERACTIONS *
 372              **************************/
 373
 374             r20              = _mm256_mul_ps(rsq20,rinv20);
 375
 376             /* Compute parameters for interactions between i and j atoms */
 377             qq20             = _mm256_mul_ps(iq2,jq0);
 378
 379             /* Calculate table index by multiplying r with table scale and truncate to integer */
 380             rt               = _mm256_mul_ps(r20,vftabscale);
 381             vfitab           = _mm256_cvttps_epi32(rt);
 382             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 383             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 384             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 385             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 386             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 387             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 388
 389             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 390             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 391                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 392             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 393                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 394             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 395                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 396             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 397                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 398             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 399             Heps             = _mm256_mul_ps(vfeps,H);
 400             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 401             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 402             velec            = _mm256_mul_ps(qq20,VV);
 403             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 404             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 405
 406             /* Update potential sum for this i atom from the interaction with this j atom. */
 407             velecsum         = _mm256_add_ps(velecsum,velec);
 408
 409             fscal            = felec;
 410
 411             /* Calculate temporary vectorial force */
 412             tx               = _mm256_mul_ps(fscal,dx20);
 413             ty               = _mm256_mul_ps(fscal,dy20);
 414             tz               = _mm256_mul_ps(fscal,dz20);
 415
 416             /* Update vectorial force */
 417             fix2             = _mm256_add_ps(fix2,tx);
 418             fiy2             = _mm256_add_ps(fiy2,ty);
 419             fiz2             = _mm256_add_ps(fiz2,tz);
 420
 421             fjx0             = _mm256_add_ps(fjx0,tx);
 422             fjy0             = _mm256_add_ps(fjy0,ty);
 423             fjz0             = _mm256_add_ps(fjz0,tz);
 424
 425             /**************************
 426              * CALCULATE INTERACTIONS *
 427              **************************/
 428
 429             r30              = _mm256_mul_ps(rsq30,rinv30);
 430
 431             /* Compute parameters for interactions between i and j atoms */
 432             qq30             = _mm256_mul_ps(iq3,jq0);
 433
 434             /* Calculate table index by multiplying r with table scale and truncate to integer */
 435             rt               = _mm256_mul_ps(r30,vftabscale);
 436             vfitab           = _mm256_cvttps_epi32(rt);
 437             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 438             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 439             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 440             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 441             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 442             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 443
 444             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 445             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 446                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 447             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 448                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 449             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 450                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 451             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 452                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 453             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 454             Heps             = _mm256_mul_ps(vfeps,H);
 455             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 456             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 457             velec            = _mm256_mul_ps(qq30,VV);
 458             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 459             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 460
 461             /* Update potential sum for this i atom from the interaction with this j atom. */
 462             velecsum         = _mm256_add_ps(velecsum,velec);
 463
 464             fscal            = felec;
 465
 466             /* Calculate temporary vectorial force */
 467             tx               = _mm256_mul_ps(fscal,dx30);
 468             ty               = _mm256_mul_ps(fscal,dy30);
 469             tz               = _mm256_mul_ps(fscal,dz30);
 470
 471             /* Update vectorial force */
 472             fix3             = _mm256_add_ps(fix3,tx);
 473             fiy3             = _mm256_add_ps(fiy3,ty);
 474             fiz3             = _mm256_add_ps(fiz3,tz);
 475
 476             fjx0             = _mm256_add_ps(fjx0,tx);
 477             fjy0             = _mm256_add_ps(fjy0,ty);
 478             fjz0             = _mm256_add_ps(fjz0,tz);
 479
 480             fjptrA             = f+j_coord_offsetA;
 481             fjptrB             = f+j_coord_offsetB;
 482             fjptrC             = f+j_coord_offsetC;
 483             fjptrD             = f+j_coord_offsetD;
 484             fjptrE             = f+j_coord_offsetE;
 485             fjptrF             = f+j_coord_offsetF;
 486             fjptrG             = f+j_coord_offsetG;
 487             fjptrH             = f+j_coord_offsetH;
 488
 489             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 490
 491             /* Inner loop uses 164 flops */
 492         }
 493
 494         if(jidx<j_index_end)
 495         {
 496
 497             /* Get j neighbor index, and coordinate index */
 498             jnrlistA         = jjnr[jidx];
 499             jnrlistB         = jjnr[jidx+1];
 500             jnrlistC         = jjnr[jidx+2];
 501             jnrlistD         = jjnr[jidx+3];
 502             jnrlistE         = jjnr[jidx+4];
 503             jnrlistF         = jjnr[jidx+5];
 504             jnrlistG         = jjnr[jidx+6];
 505             jnrlistH         = jjnr[jidx+7];
 506             /* Sign of each element will be negative for non-real atoms.
 507              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 508              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 509              */
 510             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 511                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 512
 513             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 514             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 515             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 516             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 517             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 518             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 519             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 520             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 521             j_coord_offsetA  = DIM*jnrA;
 522             j_coord_offsetB  = DIM*jnrB;
 523             j_coord_offsetC  = DIM*jnrC;
 524             j_coord_offsetD  = DIM*jnrD;
 525             j_coord_offsetE  = DIM*jnrE;
 526             j_coord_offsetF  = DIM*jnrF;
 527             j_coord_offsetG  = DIM*jnrG;
 528             j_coord_offsetH  = DIM*jnrH;
 529
 530             /* load j atom coordinates */
 531             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 532                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 533                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 534                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 535                                                  &jx0,&jy0,&jz0);
 536
 537             /* Calculate displacement vector */
 538             dx00             = _mm256_sub_ps(ix0,jx0);
 539             dy00             = _mm256_sub_ps(iy0,jy0);
 540             dz00             = _mm256_sub_ps(iz0,jz0);
 541             dx10             = _mm256_sub_ps(ix1,jx0);
 542             dy10             = _mm256_sub_ps(iy1,jy0);
 543             dz10             = _mm256_sub_ps(iz1,jz0);
 544             dx20             = _mm256_sub_ps(ix2,jx0);
 545             dy20             = _mm256_sub_ps(iy2,jy0);
 546             dz20             = _mm256_sub_ps(iz2,jz0);
 547             dx30             = _mm256_sub_ps(ix3,jx0);
 548             dy30             = _mm256_sub_ps(iy3,jy0);
 549             dz30             = _mm256_sub_ps(iz3,jz0);
 550
 551             /* Calculate squared distance and things based on it */
 552             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 553             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 554             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 555             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 556
 557             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 558             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 559             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 560
 561             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 562
 563             /* Load parameters for j particles */
 564             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 565                                                                  charge+jnrC+0,charge+jnrD+0,
 566                                                                  charge+jnrE+0,charge+jnrF+0,
 567                                                                  charge+jnrG+0,charge+jnrH+0);
 568             vdwjidx0A        = 2*vdwtype[jnrA+0];
 569             vdwjidx0B        = 2*vdwtype[jnrB+0];
 570             vdwjidx0C        = 2*vdwtype[jnrC+0];
 571             vdwjidx0D        = 2*vdwtype[jnrD+0];
 572             vdwjidx0E        = 2*vdwtype[jnrE+0];
 573             vdwjidx0F        = 2*vdwtype[jnrF+0];
 574             vdwjidx0G        = 2*vdwtype[jnrG+0];
 575             vdwjidx0H        = 2*vdwtype[jnrH+0];
 576
 577             fjx0             = _mm256_setzero_ps();
 578             fjy0             = _mm256_setzero_ps();
 579             fjz0             = _mm256_setzero_ps();
 580
 581             /**************************
 582              * CALCULATE INTERACTIONS *
 583              **************************/
 584
 585             /* Compute parameters for interactions between i and j atoms */
 586             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 587                                             vdwioffsetptr0+vdwjidx0B,
 588                                             vdwioffsetptr0+vdwjidx0C,
 589                                             vdwioffsetptr0+vdwjidx0D,
 590                                             vdwioffsetptr0+vdwjidx0E,
 591                                             vdwioffsetptr0+vdwjidx0F,
 592                                             vdwioffsetptr0+vdwjidx0G,
 593                                             vdwioffsetptr0+vdwjidx0H,
 594                                             &c6_00,&c12_00);
 595
 596             /* LENNARD-JONES DISPERSION/REPULSION */
 597
 598             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 599             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 600             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 601             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 602             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 603
 604             /* Update potential sum for this i atom from the interaction with this j atom. */
 605             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 606             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 607
 608             fscal            = fvdw;
 609
 610             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 611
 612             /* Calculate temporary vectorial force */
 613             tx               = _mm256_mul_ps(fscal,dx00);
 614             ty               = _mm256_mul_ps(fscal,dy00);
 615             tz               = _mm256_mul_ps(fscal,dz00);
 616
 617             /* Update vectorial force */
 618             fix0             = _mm256_add_ps(fix0,tx);
 619             fiy0             = _mm256_add_ps(fiy0,ty);
 620             fiz0             = _mm256_add_ps(fiz0,tz);
 621
 622             fjx0             = _mm256_add_ps(fjx0,tx);
 623             fjy0             = _mm256_add_ps(fjy0,ty);
 624             fjz0             = _mm256_add_ps(fjz0,tz);
 625
 626             /**************************
 627              * CALCULATE INTERACTIONS *
 628              **************************/
 629
 630             r10              = _mm256_mul_ps(rsq10,rinv10);
 631             r10              = _mm256_andnot_ps(dummy_mask,r10);
 632
 633             /* Compute parameters for interactions between i and j atoms */
 634             qq10             = _mm256_mul_ps(iq1,jq0);
 635
 636             /* Calculate table index by multiplying r with table scale and truncate to integer */
 637             rt               = _mm256_mul_ps(r10,vftabscale);
 638             vfitab           = _mm256_cvttps_epi32(rt);
 639             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 640             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 641             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 642             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 643             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 644             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 645
 646             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 647             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 648                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 649             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 650                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 651             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 652                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 653             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 654                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 655             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 656             Heps             = _mm256_mul_ps(vfeps,H);
 657             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 658             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 659             velec            = _mm256_mul_ps(qq10,VV);
 660             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 661             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 662
 663             /* Update potential sum for this i atom from the interaction with this j atom. */
 664             velec            = _mm256_andnot_ps(dummy_mask,velec);
 665             velecsum         = _mm256_add_ps(velecsum,velec);
 666
 667             fscal            = felec;
 668
 669             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 670
 671             /* Calculate temporary vectorial force */
 672             tx               = _mm256_mul_ps(fscal,dx10);
 673             ty               = _mm256_mul_ps(fscal,dy10);
 674             tz               = _mm256_mul_ps(fscal,dz10);
 675
 676             /* Update vectorial force */
 677             fix1             = _mm256_add_ps(fix1,tx);
 678             fiy1             = _mm256_add_ps(fiy1,ty);
 679             fiz1             = _mm256_add_ps(fiz1,tz);
 680
 681             fjx0             = _mm256_add_ps(fjx0,tx);
 682             fjy0             = _mm256_add_ps(fjy0,ty);
 683             fjz0             = _mm256_add_ps(fjz0,tz);
 684
 685             /**************************
 686              * CALCULATE INTERACTIONS *
 687              **************************/
 688
 689             r20              = _mm256_mul_ps(rsq20,rinv20);
 690             r20              = _mm256_andnot_ps(dummy_mask,r20);
 691
 692             /* Compute parameters for interactions between i and j atoms */
 693             qq20             = _mm256_mul_ps(iq2,jq0);
 694
 695             /* Calculate table index by multiplying r with table scale and truncate to integer */
 696             rt               = _mm256_mul_ps(r20,vftabscale);
 697             vfitab           = _mm256_cvttps_epi32(rt);
 698             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 699             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 700             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 701             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 702             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 703             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 704
 705             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 706             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 707                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 708             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 709                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 710             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 711                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 712             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 713                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 714             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 715             Heps             = _mm256_mul_ps(vfeps,H);
 716             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 717             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 718             velec            = _mm256_mul_ps(qq20,VV);
 719             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 720             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 721
 722             /* Update potential sum for this i atom from the interaction with this j atom. */
 723             velec            = _mm256_andnot_ps(dummy_mask,velec);
 724             velecsum         = _mm256_add_ps(velecsum,velec);
 725
 726             fscal            = felec;
 727
 728             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 729
 730             /* Calculate temporary vectorial force */
 731             tx               = _mm256_mul_ps(fscal,dx20);
 732             ty               = _mm256_mul_ps(fscal,dy20);
 733             tz               = _mm256_mul_ps(fscal,dz20);
 734
 735             /* Update vectorial force */
 736             fix2             = _mm256_add_ps(fix2,tx);
 737             fiy2             = _mm256_add_ps(fiy2,ty);
 738             fiz2             = _mm256_add_ps(fiz2,tz);
 739
 740             fjx0             = _mm256_add_ps(fjx0,tx);
 741             fjy0             = _mm256_add_ps(fjy0,ty);
 742             fjz0             = _mm256_add_ps(fjz0,tz);
 743
 744             /**************************
 745              * CALCULATE INTERACTIONS *
 746              **************************/
 747
 748             r30              = _mm256_mul_ps(rsq30,rinv30);
 749             r30              = _mm256_andnot_ps(dummy_mask,r30);
 750
 751             /* Compute parameters for interactions between i and j atoms */
 752             qq30             = _mm256_mul_ps(iq3,jq0);
 753
 754             /* Calculate table index by multiplying r with table scale and truncate to integer */
 755             rt               = _mm256_mul_ps(r30,vftabscale);
 756             vfitab           = _mm256_cvttps_epi32(rt);
 757             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 758             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 759             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 760             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 761             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 762             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 763
 764             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 765             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 766                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 767             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 768                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 769             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 770                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 771             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 772                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 773             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 774             Heps             = _mm256_mul_ps(vfeps,H);
 775             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 776             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 777             velec            = _mm256_mul_ps(qq30,VV);
 778             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 779             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 780
 781             /* Update potential sum for this i atom from the interaction with this j atom. */
 782             velec            = _mm256_andnot_ps(dummy_mask,velec);
 783             velecsum         = _mm256_add_ps(velecsum,velec);
 784
 785             fscal            = felec;
 786
 787             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 788
 789             /* Calculate temporary vectorial force */
 790             tx               = _mm256_mul_ps(fscal,dx30);
 791             ty               = _mm256_mul_ps(fscal,dy30);
 792             tz               = _mm256_mul_ps(fscal,dz30);
 793
 794             /* Update vectorial force */
 795             fix3             = _mm256_add_ps(fix3,tx);
 796             fiy3             = _mm256_add_ps(fiy3,ty);
 797             fiz3             = _mm256_add_ps(fiz3,tz);
 798
 799             fjx0             = _mm256_add_ps(fjx0,tx);
 800             fjy0             = _mm256_add_ps(fjy0,ty);
 801             fjz0             = _mm256_add_ps(fjz0,tz);
 802
 803             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 804             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 805             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 806             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 807             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 808             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 809             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 810             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 811
 812             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 813
 814             /* Inner loop uses 167 flops */
 815         }
 816
 817         /* End of innermost loop */
 818
 819         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 820                                                  f+i_coord_offset,fshift+i_shift_offset);
 821
 822         ggid                        = gid[iidx];
 823         /* Update potential energies */
 824         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 825         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 826
 827         /* Increment number of inner iterations */
 828         inneriter                  += j_index_end - j_index_start;
 829
 830         /* Outer loop uses 26 flops */
 831     }
 832
 833     /* Increment number of outer iterations */
 834     outeriter        += nri;
 835
 836     /* Update outer/inner flops */
 837
 838     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*167);
 839 }
 840 /*
 841  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_256_single
 842  * Electrostatics interaction: CubicSplineTable
 843  * VdW interaction:            LennardJones
 844  * Geometry:                   Water4-Particle
 845  * Calculate force/pot:        Force
 846  */
 847 void
 848 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_256_single
 849                     (t_nblist                    * gmx_restrict       nlist,
 850                      rvec                        * gmx_restrict          xx,
 851                      rvec                        * gmx_restrict          ff,
 852                      t_forcerec                  * gmx_restrict          fr,
 853                      t_mdatoms                   * gmx_restrict     mdatoms,
 854                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 855                      t_nrnb                      * gmx_restrict        nrnb)
 856 {
 857     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 858      * just 0 for non-waters.
 859      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 860      * jnr indices corresponding to data put in the four positions in the SIMD register.
 861      */
 862     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 863     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 864     int              jnrA,jnrB,jnrC,jnrD;
 865     int              jnrE,jnrF,jnrG,jnrH;
 866     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 867     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 868     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 869     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 870     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 871     real             rcutoff_scalar;
 872     real             *shiftvec,*fshift,*x,*f;
 873     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 874     real             scratch[4*DIM];
 875     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 876     real *           vdwioffsetptr0;
 877     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 878     real *           vdwioffsetptr1;
 879     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 880     real *           vdwioffsetptr2;
 881     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 882     real *           vdwioffsetptr3;
 883     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 884     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 885     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 886     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 887     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 888     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 889     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 890     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 891     real             *charge;
 892     int              nvdwtype;
 893     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 894     int              *vdwtype;
 895     real             *vdwparam;
 896     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 897     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 898     __m256i          vfitab;
 899     __m128i          vfitab_lo,vfitab_hi;
 900     __m128i          ifour       = _mm_set1_epi32(4);
 901     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 902     real             *vftab;
 903     __m256           dummy_mask,cutoff_mask;
 904     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 905     __m256           one     = _mm256_set1_ps(1.0);
 906     __m256           two     = _mm256_set1_ps(2.0);
 907     x                = xx[0];
 908     f                = ff[0];
 909
 910     nri              = nlist->nri;
 911     iinr             = nlist->iinr;
 912     jindex           = nlist->jindex;
 913     jjnr             = nlist->jjnr;
 914     shiftidx         = nlist->shift;
 915     gid              = nlist->gid;
 916     shiftvec         = fr->shift_vec[0];
 917     fshift           = fr->fshift[0];
 918     facel            = _mm256_set1_ps(fr->epsfac);
 919     charge           = mdatoms->chargeA;
 920     nvdwtype         = fr->ntype;
 921     vdwparam         = fr->nbfp;
 922     vdwtype          = mdatoms->typeA;
 923
 924     vftab            = kernel_data->table_elec->data;
 925     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 926
 927     /* Setup water-specific parameters */
 928     inr              = nlist->iinr[0];
 929     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 930     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 931     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 932     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 933
 934     /* Avoid stupid compiler warnings */
 935     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 936     j_coord_offsetA = 0;
 937     j_coord_offsetB = 0;
 938     j_coord_offsetC = 0;
 939     j_coord_offsetD = 0;
 940     j_coord_offsetE = 0;
 941     j_coord_offsetF = 0;
 942     j_coord_offsetG = 0;
 943     j_coord_offsetH = 0;
 944
 945     outeriter        = 0;
 946     inneriter        = 0;
 947
 948     for(iidx=0;iidx<4*DIM;iidx++)
 949     {
 950         scratch[iidx] = 0.0;
 951     }
 952
 953     /* Start outer loop over neighborlists */
 954     for(iidx=0; iidx<nri; iidx++)
 955     {
 956         /* Load shift vector for this list */
 957         i_shift_offset   = DIM*shiftidx[iidx];
 958
 959         /* Load limits for loop over neighbors */
 960         j_index_start    = jindex[iidx];
 961         j_index_end      = jindex[iidx+1];
 962
 963         /* Get outer coordinate index */
 964         inr              = iinr[iidx];
 965         i_coord_offset   = DIM*inr;
 966
 967         /* Load i particle coords and add shift vector */
 968         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 969                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 970
 971         fix0             = _mm256_setzero_ps();
 972         fiy0             = _mm256_setzero_ps();
 973         fiz0             = _mm256_setzero_ps();
 974         fix1             = _mm256_setzero_ps();
 975         fiy1             = _mm256_setzero_ps();
 976         fiz1             = _mm256_setzero_ps();
 977         fix2             = _mm256_setzero_ps();
 978         fiy2             = _mm256_setzero_ps();
 979         fiz2             = _mm256_setzero_ps();
 980         fix3             = _mm256_setzero_ps();
 981         fiy3             = _mm256_setzero_ps();
 982         fiz3             = _mm256_setzero_ps();
 983
 984         /* Start inner kernel loop */
 985         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 986         {
 987
 988             /* Get j neighbor index, and coordinate index */
 989             jnrA             = jjnr[jidx];
 990             jnrB             = jjnr[jidx+1];
 991             jnrC             = jjnr[jidx+2];
 992             jnrD             = jjnr[jidx+3];
 993             jnrE             = jjnr[jidx+4];
 994             jnrF             = jjnr[jidx+5];
 995             jnrG             = jjnr[jidx+6];
 996             jnrH             = jjnr[jidx+7];
 997             j_coord_offsetA  = DIM*jnrA;
 998             j_coord_offsetB  = DIM*jnrB;
 999             j_coord_offsetC  = DIM*jnrC;
1000             j_coord_offsetD  = DIM*jnrD;
1001             j_coord_offsetE  = DIM*jnrE;
1002             j_coord_offsetF  = DIM*jnrF;
1003             j_coord_offsetG  = DIM*jnrG;
1004             j_coord_offsetH  = DIM*jnrH;
1005
1006             /* load j atom coordinates */
1007             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1008                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1009                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1010                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1011                                                  &jx0,&jy0,&jz0);
1012
1013             /* Calculate displacement vector */
1014             dx00             = _mm256_sub_ps(ix0,jx0);
1015             dy00             = _mm256_sub_ps(iy0,jy0);
1016             dz00             = _mm256_sub_ps(iz0,jz0);
1017             dx10             = _mm256_sub_ps(ix1,jx0);
1018             dy10             = _mm256_sub_ps(iy1,jy0);
1019             dz10             = _mm256_sub_ps(iz1,jz0);
1020             dx20             = _mm256_sub_ps(ix2,jx0);
1021             dy20             = _mm256_sub_ps(iy2,jy0);
1022             dz20             = _mm256_sub_ps(iz2,jz0);
1023             dx30             = _mm256_sub_ps(ix3,jx0);
1024             dy30             = _mm256_sub_ps(iy3,jy0);
1025             dz30             = _mm256_sub_ps(iz3,jz0);
1026
1027             /* Calculate squared distance and things based on it */
1028             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1029             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1030             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1031             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1032
1033             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1034             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1035             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1036
1037             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1038
1039             /* Load parameters for j particles */
1040             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1041                                                                  charge+jnrC+0,charge+jnrD+0,
1042                                                                  charge+jnrE+0,charge+jnrF+0,
1043                                                                  charge+jnrG+0,charge+jnrH+0);
1044             vdwjidx0A        = 2*vdwtype[jnrA+0];
1045             vdwjidx0B        = 2*vdwtype[jnrB+0];
1046             vdwjidx0C        = 2*vdwtype[jnrC+0];
1047             vdwjidx0D        = 2*vdwtype[jnrD+0];
1048             vdwjidx0E        = 2*vdwtype[jnrE+0];
1049             vdwjidx0F        = 2*vdwtype[jnrF+0];
1050             vdwjidx0G        = 2*vdwtype[jnrG+0];
1051             vdwjidx0H        = 2*vdwtype[jnrH+0];
1052
1053             fjx0             = _mm256_setzero_ps();
1054             fjy0             = _mm256_setzero_ps();
1055             fjz0             = _mm256_setzero_ps();
1056
1057             /**************************
1058              * CALCULATE INTERACTIONS *
1059              **************************/
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1063                                             vdwioffsetptr0+vdwjidx0B,
1064                                             vdwioffsetptr0+vdwjidx0C,
1065                                             vdwioffsetptr0+vdwjidx0D,
1066                                             vdwioffsetptr0+vdwjidx0E,
1067                                             vdwioffsetptr0+vdwjidx0F,
1068                                             vdwioffsetptr0+vdwjidx0G,
1069                                             vdwioffsetptr0+vdwjidx0H,
1070                                             &c6_00,&c12_00);
1071
1072             /* LENNARD-JONES DISPERSION/REPULSION */
1073
1074             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1075             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1076
1077             fscal            = fvdw;
1078
1079             /* Calculate temporary vectorial force */
1080             tx               = _mm256_mul_ps(fscal,dx00);
1081             ty               = _mm256_mul_ps(fscal,dy00);
1082             tz               = _mm256_mul_ps(fscal,dz00);
1083
1084             /* Update vectorial force */
1085             fix0             = _mm256_add_ps(fix0,tx);
1086             fiy0             = _mm256_add_ps(fiy0,ty);
1087             fiz0             = _mm256_add_ps(fiz0,tz);
1088
1089             fjx0             = _mm256_add_ps(fjx0,tx);
1090             fjy0             = _mm256_add_ps(fjy0,ty);
1091             fjz0             = _mm256_add_ps(fjz0,tz);
1092
1093             /**************************
1094              * CALCULATE INTERACTIONS *
1095              **************************/
1096
1097             r10              = _mm256_mul_ps(rsq10,rinv10);
1098
1099             /* Compute parameters for interactions between i and j atoms */
1100             qq10             = _mm256_mul_ps(iq1,jq0);
1101
1102             /* Calculate table index by multiplying r with table scale and truncate to integer */
1103             rt               = _mm256_mul_ps(r10,vftabscale);
1104             vfitab           = _mm256_cvttps_epi32(rt);
1105             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1106             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1107             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1108             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1109             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1110             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1111
1112             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1113             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1114                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1115             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1116                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1117             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1118                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1119             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1120                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1121             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1122             Heps             = _mm256_mul_ps(vfeps,H);
1123             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1124             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1125             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1126
1127             fscal            = felec;
1128
1129             /* Calculate temporary vectorial force */
1130             tx               = _mm256_mul_ps(fscal,dx10);
1131             ty               = _mm256_mul_ps(fscal,dy10);
1132             tz               = _mm256_mul_ps(fscal,dz10);
1133
1134             /* Update vectorial force */
1135             fix1             = _mm256_add_ps(fix1,tx);
1136             fiy1             = _mm256_add_ps(fiy1,ty);
1137             fiz1             = _mm256_add_ps(fiz1,tz);
1138
1139             fjx0             = _mm256_add_ps(fjx0,tx);
1140             fjy0             = _mm256_add_ps(fjy0,ty);
1141             fjz0             = _mm256_add_ps(fjz0,tz);
1142
1143             /**************************
1144              * CALCULATE INTERACTIONS *
1145              **************************/
1146
1147             r20              = _mm256_mul_ps(rsq20,rinv20);
1148
1149             /* Compute parameters for interactions between i and j atoms */
1150             qq20             = _mm256_mul_ps(iq2,jq0);
1151
1152             /* Calculate table index by multiplying r with table scale and truncate to integer */
1153             rt               = _mm256_mul_ps(r20,vftabscale);
1154             vfitab           = _mm256_cvttps_epi32(rt);
1155             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1156             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1157             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1158             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1159             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1160             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1161
1162             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1163             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1164                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1165             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1166                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1167             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1168                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1169             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1170                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1171             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1172             Heps             = _mm256_mul_ps(vfeps,H);
1173             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1174             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1175             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1176
1177             fscal            = felec;
1178
1179             /* Calculate temporary vectorial force */
1180             tx               = _mm256_mul_ps(fscal,dx20);
1181             ty               = _mm256_mul_ps(fscal,dy20);
1182             tz               = _mm256_mul_ps(fscal,dz20);
1183
1184             /* Update vectorial force */
1185             fix2             = _mm256_add_ps(fix2,tx);
1186             fiy2             = _mm256_add_ps(fiy2,ty);
1187             fiz2             = _mm256_add_ps(fiz2,tz);
1188
1189             fjx0             = _mm256_add_ps(fjx0,tx);
1190             fjy0             = _mm256_add_ps(fjy0,ty);
1191             fjz0             = _mm256_add_ps(fjz0,tz);
1192
1193             /**************************
1194              * CALCULATE INTERACTIONS *
1195              **************************/
1196
1197             r30              = _mm256_mul_ps(rsq30,rinv30);
1198
1199             /* Compute parameters for interactions between i and j atoms */
1200             qq30             = _mm256_mul_ps(iq3,jq0);
1201
1202             /* Calculate table index by multiplying r with table scale and truncate to integer */
1203             rt               = _mm256_mul_ps(r30,vftabscale);
1204             vfitab           = _mm256_cvttps_epi32(rt);
1205             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1206             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1207             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1208             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1209             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1210             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1211
1212             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1213             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1214                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1215             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1216                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1217             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1218                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1219             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1220                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1221             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1222             Heps             = _mm256_mul_ps(vfeps,H);
1223             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1224             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1225             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1226
1227             fscal            = felec;
1228
1229             /* Calculate temporary vectorial force */
1230             tx               = _mm256_mul_ps(fscal,dx30);
1231             ty               = _mm256_mul_ps(fscal,dy30);
1232             tz               = _mm256_mul_ps(fscal,dz30);
1233
1234             /* Update vectorial force */
1235             fix3             = _mm256_add_ps(fix3,tx);
1236             fiy3             = _mm256_add_ps(fiy3,ty);
1237             fiz3             = _mm256_add_ps(fiz3,tz);
1238
1239             fjx0             = _mm256_add_ps(fjx0,tx);
1240             fjy0             = _mm256_add_ps(fjy0,ty);
1241             fjz0             = _mm256_add_ps(fjz0,tz);
1242
1243             fjptrA             = f+j_coord_offsetA;
1244             fjptrB             = f+j_coord_offsetB;
1245             fjptrC             = f+j_coord_offsetC;
1246             fjptrD             = f+j_coord_offsetD;
1247             fjptrE             = f+j_coord_offsetE;
1248             fjptrF             = f+j_coord_offsetF;
1249             fjptrG             = f+j_coord_offsetG;
1250             fjptrH             = f+j_coord_offsetH;
1251
1252             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1253
1254             /* Inner loop uses 147 flops */
1255         }
1256
1257         if(jidx<j_index_end)
1258         {
1259
1260             /* Get j neighbor index, and coordinate index */
1261             jnrlistA         = jjnr[jidx];
1262             jnrlistB         = jjnr[jidx+1];
1263             jnrlistC         = jjnr[jidx+2];
1264             jnrlistD         = jjnr[jidx+3];
1265             jnrlistE         = jjnr[jidx+4];
1266             jnrlistF         = jjnr[jidx+5];
1267             jnrlistG         = jjnr[jidx+6];
1268             jnrlistH         = jjnr[jidx+7];
1269             /* Sign of each element will be negative for non-real atoms.
1270              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1271              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1272              */
1273             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1274                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1275
1276             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1277             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1278             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1279             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1280             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1281             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1282             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1283             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1284             j_coord_offsetA  = DIM*jnrA;
1285             j_coord_offsetB  = DIM*jnrB;
1286             j_coord_offsetC  = DIM*jnrC;
1287             j_coord_offsetD  = DIM*jnrD;
1288             j_coord_offsetE  = DIM*jnrE;
1289             j_coord_offsetF  = DIM*jnrF;
1290             j_coord_offsetG  = DIM*jnrG;
1291             j_coord_offsetH  = DIM*jnrH;
1292
1293             /* load j atom coordinates */
1294             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1295                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1296                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1297                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1298                                                  &jx0,&jy0,&jz0);
1299
1300             /* Calculate displacement vector */
1301             dx00             = _mm256_sub_ps(ix0,jx0);
1302             dy00             = _mm256_sub_ps(iy0,jy0);
1303             dz00             = _mm256_sub_ps(iz0,jz0);
1304             dx10             = _mm256_sub_ps(ix1,jx0);
1305             dy10             = _mm256_sub_ps(iy1,jy0);
1306             dz10             = _mm256_sub_ps(iz1,jz0);
1307             dx20             = _mm256_sub_ps(ix2,jx0);
1308             dy20             = _mm256_sub_ps(iy2,jy0);
1309             dz20             = _mm256_sub_ps(iz2,jz0);
1310             dx30             = _mm256_sub_ps(ix3,jx0);
1311             dy30             = _mm256_sub_ps(iy3,jy0);
1312             dz30             = _mm256_sub_ps(iz3,jz0);
1313
1314             /* Calculate squared distance and things based on it */
1315             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1316             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1317             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1318             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1319
1320             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1321             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1322             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1323
1324             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1325
1326             /* Load parameters for j particles */
1327             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1328                                                                  charge+jnrC+0,charge+jnrD+0,
1329                                                                  charge+jnrE+0,charge+jnrF+0,
1330                                                                  charge+jnrG+0,charge+jnrH+0);
1331             vdwjidx0A        = 2*vdwtype[jnrA+0];
1332             vdwjidx0B        = 2*vdwtype[jnrB+0];
1333             vdwjidx0C        = 2*vdwtype[jnrC+0];
1334             vdwjidx0D        = 2*vdwtype[jnrD+0];
1335             vdwjidx0E        = 2*vdwtype[jnrE+0];
1336             vdwjidx0F        = 2*vdwtype[jnrF+0];
1337             vdwjidx0G        = 2*vdwtype[jnrG+0];
1338             vdwjidx0H        = 2*vdwtype[jnrH+0];
1339
1340             fjx0             = _mm256_setzero_ps();
1341             fjy0             = _mm256_setzero_ps();
1342             fjz0             = _mm256_setzero_ps();
1343
1344             /**************************
1345              * CALCULATE INTERACTIONS *
1346              **************************/
1347
1348             /* Compute parameters for interactions between i and j atoms */
1349             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1350                                             vdwioffsetptr0+vdwjidx0B,
1351                                             vdwioffsetptr0+vdwjidx0C,
1352                                             vdwioffsetptr0+vdwjidx0D,
1353                                             vdwioffsetptr0+vdwjidx0E,
1354                                             vdwioffsetptr0+vdwjidx0F,
1355                                             vdwioffsetptr0+vdwjidx0G,
1356                                             vdwioffsetptr0+vdwjidx0H,
1357                                             &c6_00,&c12_00);
1358
1359             /* LENNARD-JONES DISPERSION/REPULSION */
1360
1361             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1362             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1363
1364             fscal            = fvdw;
1365
1366             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1367
1368             /* Calculate temporary vectorial force */
1369             tx               = _mm256_mul_ps(fscal,dx00);
1370             ty               = _mm256_mul_ps(fscal,dy00);
1371             tz               = _mm256_mul_ps(fscal,dz00);
1372
1373             /* Update vectorial force */
1374             fix0             = _mm256_add_ps(fix0,tx);
1375             fiy0             = _mm256_add_ps(fiy0,ty);
1376             fiz0             = _mm256_add_ps(fiz0,tz);
1377
1378             fjx0             = _mm256_add_ps(fjx0,tx);
1379             fjy0             = _mm256_add_ps(fjy0,ty);
1380             fjz0             = _mm256_add_ps(fjz0,tz);
1381
1382             /**************************
1383              * CALCULATE INTERACTIONS *
1384              **************************/
1385
1386             r10              = _mm256_mul_ps(rsq10,rinv10);
1387             r10              = _mm256_andnot_ps(dummy_mask,r10);
1388
1389             /* Compute parameters for interactions between i and j atoms */
1390             qq10             = _mm256_mul_ps(iq1,jq0);
1391
1392             /* Calculate table index by multiplying r with table scale and truncate to integer */
1393             rt               = _mm256_mul_ps(r10,vftabscale);
1394             vfitab           = _mm256_cvttps_epi32(rt);
1395             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1396             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1397             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1398             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1399             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1400             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1401
1402             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1403             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1404                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1405             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1406                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1407             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1408                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1409             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1410                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1411             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1412             Heps             = _mm256_mul_ps(vfeps,H);
1413             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1414             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1415             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1416
1417             fscal            = felec;
1418
1419             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1420
1421             /* Calculate temporary vectorial force */
1422             tx               = _mm256_mul_ps(fscal,dx10);
1423             ty               = _mm256_mul_ps(fscal,dy10);
1424             tz               = _mm256_mul_ps(fscal,dz10);
1425
1426             /* Update vectorial force */
1427             fix1             = _mm256_add_ps(fix1,tx);
1428             fiy1             = _mm256_add_ps(fiy1,ty);
1429             fiz1             = _mm256_add_ps(fiz1,tz);
1430
1431             fjx0             = _mm256_add_ps(fjx0,tx);
1432             fjy0             = _mm256_add_ps(fjy0,ty);
1433             fjz0             = _mm256_add_ps(fjz0,tz);
1434
1435             /**************************
1436              * CALCULATE INTERACTIONS *
1437              **************************/
1438
1439             r20              = _mm256_mul_ps(rsq20,rinv20);
1440             r20              = _mm256_andnot_ps(dummy_mask,r20);
1441
1442             /* Compute parameters for interactions between i and j atoms */
1443             qq20             = _mm256_mul_ps(iq2,jq0);
1444
1445             /* Calculate table index by multiplying r with table scale and truncate to integer */
1446             rt               = _mm256_mul_ps(r20,vftabscale);
1447             vfitab           = _mm256_cvttps_epi32(rt);
1448             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1449             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1450             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1451             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1452             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1453             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1454
1455             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1456             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1457                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1458             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1459                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1460             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1461                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1462             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1463                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1464             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1465             Heps             = _mm256_mul_ps(vfeps,H);
1466             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1467             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1468             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1469
1470             fscal            = felec;
1471
1472             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1473
1474             /* Calculate temporary vectorial force */
1475             tx               = _mm256_mul_ps(fscal,dx20);
1476             ty               = _mm256_mul_ps(fscal,dy20);
1477             tz               = _mm256_mul_ps(fscal,dz20);
1478
1479             /* Update vectorial force */
1480             fix2             = _mm256_add_ps(fix2,tx);
1481             fiy2             = _mm256_add_ps(fiy2,ty);
1482             fiz2             = _mm256_add_ps(fiz2,tz);
1483
1484             fjx0             = _mm256_add_ps(fjx0,tx);
1485             fjy0             = _mm256_add_ps(fjy0,ty);
1486             fjz0             = _mm256_add_ps(fjz0,tz);
1487
1488             /**************************
1489              * CALCULATE INTERACTIONS *
1490              **************************/
1491
1492             r30              = _mm256_mul_ps(rsq30,rinv30);
1493             r30              = _mm256_andnot_ps(dummy_mask,r30);
1494
1495             /* Compute parameters for interactions between i and j atoms */
1496             qq30             = _mm256_mul_ps(iq3,jq0);
1497
1498             /* Calculate table index by multiplying r with table scale and truncate to integer */
1499             rt               = _mm256_mul_ps(r30,vftabscale);
1500             vfitab           = _mm256_cvttps_epi32(rt);
1501             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1502             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1503             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1504             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1505             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1506             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1507
1508             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1509             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1510                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1511             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1512                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1513             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1514                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1515             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1516                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1517             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1518             Heps             = _mm256_mul_ps(vfeps,H);
1519             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1520             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1521             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1522
1523             fscal            = felec;
1524
1525             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1526
1527             /* Calculate temporary vectorial force */
1528             tx               = _mm256_mul_ps(fscal,dx30);
1529             ty               = _mm256_mul_ps(fscal,dy30);
1530             tz               = _mm256_mul_ps(fscal,dz30);
1531
1532             /* Update vectorial force */
1533             fix3             = _mm256_add_ps(fix3,tx);
1534             fiy3             = _mm256_add_ps(fiy3,ty);
1535             fiz3             = _mm256_add_ps(fiz3,tz);
1536
1537             fjx0             = _mm256_add_ps(fjx0,tx);
1538             fjy0             = _mm256_add_ps(fjy0,ty);
1539             fjz0             = _mm256_add_ps(fjz0,tz);
1540
1541             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1542             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1543             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1544             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1545             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1546             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1547             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1548             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1549
1550             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1551
1552             /* Inner loop uses 150 flops */
1553         }
1554
1555         /* End of innermost loop */
1556
1557         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1558                                                  f+i_coord_offset,fshift+i_shift_offset);
1559
1560         /* Increment number of inner iterations */
1561         inneriter                  += j_index_end - j_index_start;
1562
1563         /* Outer loop uses 24 flops */
1564     }
1565
1566     /* Increment number of outer iterations */
1567     outeriter        += nri;
1568
1569     /* Update outer/inner flops */
1570
1571     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1572 }