src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwCSTab_GeomW4P1_avx_256_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_256_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     real *           vdwioffsetptr3;
  92     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  94     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256i          vfitab;
 108     __m128i          vfitab_lo,vfitab_hi;
 109     __m128i          ifour       = _mm_set1_epi32(4);
 110     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 111     real             *vftab;
 112     __m256i          ewitab;
 113     __m128i          ewitab_lo,ewitab_hi;
 114     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 115     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 116     real             *ewtab;
 117     __m256           dummy_mask,cutoff_mask;
 118     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 119     __m256           one     = _mm256_set1_ps(1.0);
 120     __m256           two     = _mm256_set1_ps(2.0);
 121     x                = xx[0];
 122     f                = ff[0];
 123
 124     nri              = nlist->nri;
 125     iinr             = nlist->iinr;
 126     jindex           = nlist->jindex;
 127     jjnr             = nlist->jjnr;
 128     shiftidx         = nlist->shift;
 129     gid              = nlist->gid;
 130     shiftvec         = fr->shift_vec[0];
 131     fshift           = fr->fshift[0];
 132     facel            = _mm256_set1_ps(fr->ic->epsfac);
 133     charge           = mdatoms->chargeA;
 134     nvdwtype         = fr->ntype;
 135     vdwparam         = fr->nbfp;
 136     vdwtype          = mdatoms->typeA;
 137
 138     vftab            = kernel_data->table_vdw->data;
 139     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 140
 141     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 142     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 143     beta2            = _mm256_mul_ps(beta,beta);
 144     beta3            = _mm256_mul_ps(beta,beta2);
 145
 146     ewtab            = fr->ic->tabq_coul_FDV0;
 147     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 148     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 149
 150     /* Setup water-specific parameters */
 151     inr              = nlist->iinr[0];
 152     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 153     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 154     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 155     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 156
 157     /* Avoid stupid compiler warnings */
 158     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 159     j_coord_offsetA = 0;
 160     j_coord_offsetB = 0;
 161     j_coord_offsetC = 0;
 162     j_coord_offsetD = 0;
 163     j_coord_offsetE = 0;
 164     j_coord_offsetF = 0;
 165     j_coord_offsetG = 0;
 166     j_coord_offsetH = 0;
 167
 168     outeriter        = 0;
 169     inneriter        = 0;
 170
 171     for(iidx=0;iidx<4*DIM;iidx++)
 172     {
 173         scratch[iidx] = 0.0;
 174     }
 175
 176     /* Start outer loop over neighborlists */
 177     for(iidx=0; iidx<nri; iidx++)
 178     {
 179         /* Load shift vector for this list */
 180         i_shift_offset   = DIM*shiftidx[iidx];
 181
 182         /* Load limits for loop over neighbors */
 183         j_index_start    = jindex[iidx];
 184         j_index_end      = jindex[iidx+1];
 185
 186         /* Get outer coordinate index */
 187         inr              = iinr[iidx];
 188         i_coord_offset   = DIM*inr;
 189
 190         /* Load i particle coords and add shift vector */
 191         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 192                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 193
 194         fix0             = _mm256_setzero_ps();
 195         fiy0             = _mm256_setzero_ps();
 196         fiz0             = _mm256_setzero_ps();
 197         fix1             = _mm256_setzero_ps();
 198         fiy1             = _mm256_setzero_ps();
 199         fiz1             = _mm256_setzero_ps();
 200         fix2             = _mm256_setzero_ps();
 201         fiy2             = _mm256_setzero_ps();
 202         fiz2             = _mm256_setzero_ps();
 203         fix3             = _mm256_setzero_ps();
 204         fiy3             = _mm256_setzero_ps();
 205         fiz3             = _mm256_setzero_ps();
 206
 207         /* Reset potential sums */
 208         velecsum         = _mm256_setzero_ps();
 209         vvdwsum          = _mm256_setzero_ps();
 210
 211         /* Start inner kernel loop */
 212         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 213         {
 214
 215             /* Get j neighbor index, and coordinate index */
 216             jnrA             = jjnr[jidx];
 217             jnrB             = jjnr[jidx+1];
 218             jnrC             = jjnr[jidx+2];
 219             jnrD             = jjnr[jidx+3];
 220             jnrE             = jjnr[jidx+4];
 221             jnrF             = jjnr[jidx+5];
 222             jnrG             = jjnr[jidx+6];
 223             jnrH             = jjnr[jidx+7];
 224             j_coord_offsetA  = DIM*jnrA;
 225             j_coord_offsetB  = DIM*jnrB;
 226             j_coord_offsetC  = DIM*jnrC;
 227             j_coord_offsetD  = DIM*jnrD;
 228             j_coord_offsetE  = DIM*jnrE;
 229             j_coord_offsetF  = DIM*jnrF;
 230             j_coord_offsetG  = DIM*jnrG;
 231             j_coord_offsetH  = DIM*jnrH;
 232
 233             /* load j atom coordinates */
 234             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 235                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 236                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 237                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 238                                                  &jx0,&jy0,&jz0);
 239
 240             /* Calculate displacement vector */
 241             dx00             = _mm256_sub_ps(ix0,jx0);
 242             dy00             = _mm256_sub_ps(iy0,jy0);
 243             dz00             = _mm256_sub_ps(iz0,jz0);
 244             dx10             = _mm256_sub_ps(ix1,jx0);
 245             dy10             = _mm256_sub_ps(iy1,jy0);
 246             dz10             = _mm256_sub_ps(iz1,jz0);
 247             dx20             = _mm256_sub_ps(ix2,jx0);
 248             dy20             = _mm256_sub_ps(iy2,jy0);
 249             dz20             = _mm256_sub_ps(iz2,jz0);
 250             dx30             = _mm256_sub_ps(ix3,jx0);
 251             dy30             = _mm256_sub_ps(iy3,jy0);
 252             dz30             = _mm256_sub_ps(iz3,jz0);
 253
 254             /* Calculate squared distance and things based on it */
 255             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 256             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 257             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 258             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 259
 260             rinv00           = avx256_invsqrt_f(rsq00);
 261             rinv10           = avx256_invsqrt_f(rsq10);
 262             rinv20           = avx256_invsqrt_f(rsq20);
 263             rinv30           = avx256_invsqrt_f(rsq30);
 264
 265             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 266             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 267             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 268
 269             /* Load parameters for j particles */
 270             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 271                                                                  charge+jnrC+0,charge+jnrD+0,
 272                                                                  charge+jnrE+0,charge+jnrF+0,
 273                                                                  charge+jnrG+0,charge+jnrH+0);
 274             vdwjidx0A        = 2*vdwtype[jnrA+0];
 275             vdwjidx0B        = 2*vdwtype[jnrB+0];
 276             vdwjidx0C        = 2*vdwtype[jnrC+0];
 277             vdwjidx0D        = 2*vdwtype[jnrD+0];
 278             vdwjidx0E        = 2*vdwtype[jnrE+0];
 279             vdwjidx0F        = 2*vdwtype[jnrF+0];
 280             vdwjidx0G        = 2*vdwtype[jnrG+0];
 281             vdwjidx0H        = 2*vdwtype[jnrH+0];
 282
 283             fjx0             = _mm256_setzero_ps();
 284             fjy0             = _mm256_setzero_ps();
 285             fjz0             = _mm256_setzero_ps();
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             r00              = _mm256_mul_ps(rsq00,rinv00);
 292
 293             /* Compute parameters for interactions between i and j atoms */
 294             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 295                                             vdwioffsetptr0+vdwjidx0B,
 296                                             vdwioffsetptr0+vdwjidx0C,
 297                                             vdwioffsetptr0+vdwjidx0D,
 298                                             vdwioffsetptr0+vdwjidx0E,
 299                                             vdwioffsetptr0+vdwjidx0F,
 300                                             vdwioffsetptr0+vdwjidx0G,
 301                                             vdwioffsetptr0+vdwjidx0H,
 302                                             &c6_00,&c12_00);
 303
 304             /* Calculate table index by multiplying r with table scale and truncate to integer */
 305             rt               = _mm256_mul_ps(r00,vftabscale);
 306             vfitab           = _mm256_cvttps_epi32(rt);
 307             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 308             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 309             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 310             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 311             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 312             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 313
 314             /* CUBIC SPLINE TABLE DISPERSION */
 315             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 316                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 317             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 318                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 319             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 320                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 321             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 322                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 323             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 324             Heps             = _mm256_mul_ps(vfeps,H);
 325             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 326             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 327             vvdw6            = _mm256_mul_ps(c6_00,VV);
 328             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 329             fvdw6            = _mm256_mul_ps(c6_00,FF);
 330
 331             /* CUBIC SPLINE TABLE REPULSION */
 332             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 333             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 334             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 335                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 336             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 337                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 338             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 339                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 340             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 341                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 342             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 343             Heps             = _mm256_mul_ps(vfeps,H);
 344             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 345             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 346             vvdw12           = _mm256_mul_ps(c12_00,VV);
 347             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 348             fvdw12           = _mm256_mul_ps(c12_00,FF);
 349             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 350             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 354
 355             fscal            = fvdw;
 356
 357             /* Calculate temporary vectorial force */
 358             tx               = _mm256_mul_ps(fscal,dx00);
 359             ty               = _mm256_mul_ps(fscal,dy00);
 360             tz               = _mm256_mul_ps(fscal,dz00);
 361
 362             /* Update vectorial force */
 363             fix0             = _mm256_add_ps(fix0,tx);
 364             fiy0             = _mm256_add_ps(fiy0,ty);
 365             fiz0             = _mm256_add_ps(fiz0,tz);
 366
 367             fjx0             = _mm256_add_ps(fjx0,tx);
 368             fjy0             = _mm256_add_ps(fjy0,ty);
 369             fjz0             = _mm256_add_ps(fjz0,tz);
 370
 371             /**************************
 372              * CALCULATE INTERACTIONS *
 373              **************************/
 374
 375             r10              = _mm256_mul_ps(rsq10,rinv10);
 376
 377             /* Compute parameters for interactions between i and j atoms */
 378             qq10             = _mm256_mul_ps(iq1,jq0);
 379
 380             /* EWALD ELECTROSTATICS */
 381
 382             /* Analytical PME correction */
 383             zeta2            = _mm256_mul_ps(beta2,rsq10);
 384             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 385             pmecorrF         = avx256_pmecorrF_f(zeta2);
 386             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 387             felec            = _mm256_mul_ps(qq10,felec);
 388             pmecorrV         = avx256_pmecorrV_f(zeta2);
 389             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 390             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 391             velec            = _mm256_mul_ps(qq10,velec);
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velecsum         = _mm256_add_ps(velecsum,velec);
 395
 396             fscal            = felec;
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm256_mul_ps(fscal,dx10);
 400             ty               = _mm256_mul_ps(fscal,dy10);
 401             tz               = _mm256_mul_ps(fscal,dz10);
 402
 403             /* Update vectorial force */
 404             fix1             = _mm256_add_ps(fix1,tx);
 405             fiy1             = _mm256_add_ps(fiy1,ty);
 406             fiz1             = _mm256_add_ps(fiz1,tz);
 407
 408             fjx0             = _mm256_add_ps(fjx0,tx);
 409             fjy0             = _mm256_add_ps(fjy0,ty);
 410             fjz0             = _mm256_add_ps(fjz0,tz);
 411
 412             /**************************
 413              * CALCULATE INTERACTIONS *
 414              **************************/
 415
 416             r20              = _mm256_mul_ps(rsq20,rinv20);
 417
 418             /* Compute parameters for interactions between i and j atoms */
 419             qq20             = _mm256_mul_ps(iq2,jq0);
 420
 421             /* EWALD ELECTROSTATICS */
 422
 423             /* Analytical PME correction */
 424             zeta2            = _mm256_mul_ps(beta2,rsq20);
 425             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 426             pmecorrF         = avx256_pmecorrF_f(zeta2);
 427             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 428             felec            = _mm256_mul_ps(qq20,felec);
 429             pmecorrV         = avx256_pmecorrV_f(zeta2);
 430             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 431             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 432             velec            = _mm256_mul_ps(qq20,velec);
 433
 434             /* Update potential sum for this i atom from the interaction with this j atom. */
 435             velecsum         = _mm256_add_ps(velecsum,velec);
 436
 437             fscal            = felec;
 438
 439             /* Calculate temporary vectorial force */
 440             tx               = _mm256_mul_ps(fscal,dx20);
 441             ty               = _mm256_mul_ps(fscal,dy20);
 442             tz               = _mm256_mul_ps(fscal,dz20);
 443
 444             /* Update vectorial force */
 445             fix2             = _mm256_add_ps(fix2,tx);
 446             fiy2             = _mm256_add_ps(fiy2,ty);
 447             fiz2             = _mm256_add_ps(fiz2,tz);
 448
 449             fjx0             = _mm256_add_ps(fjx0,tx);
 450             fjy0             = _mm256_add_ps(fjy0,ty);
 451             fjz0             = _mm256_add_ps(fjz0,tz);
 452
 453             /**************************
 454              * CALCULATE INTERACTIONS *
 455              **************************/
 456
 457             r30              = _mm256_mul_ps(rsq30,rinv30);
 458
 459             /* Compute parameters for interactions between i and j atoms */
 460             qq30             = _mm256_mul_ps(iq3,jq0);
 461
 462             /* EWALD ELECTROSTATICS */
 463
 464             /* Analytical PME correction */
 465             zeta2            = _mm256_mul_ps(beta2,rsq30);
 466             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 467             pmecorrF         = avx256_pmecorrF_f(zeta2);
 468             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 469             felec            = _mm256_mul_ps(qq30,felec);
 470             pmecorrV         = avx256_pmecorrV_f(zeta2);
 471             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 472             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 473             velec            = _mm256_mul_ps(qq30,velec);
 474
 475             /* Update potential sum for this i atom from the interaction with this j atom. */
 476             velecsum         = _mm256_add_ps(velecsum,velec);
 477
 478             fscal            = felec;
 479
 480             /* Calculate temporary vectorial force */
 481             tx               = _mm256_mul_ps(fscal,dx30);
 482             ty               = _mm256_mul_ps(fscal,dy30);
 483             tz               = _mm256_mul_ps(fscal,dz30);
 484
 485             /* Update vectorial force */
 486             fix3             = _mm256_add_ps(fix3,tx);
 487             fiy3             = _mm256_add_ps(fiy3,ty);
 488             fiz3             = _mm256_add_ps(fiz3,tz);
 489
 490             fjx0             = _mm256_add_ps(fjx0,tx);
 491             fjy0             = _mm256_add_ps(fjy0,ty);
 492             fjz0             = _mm256_add_ps(fjz0,tz);
 493
 494             fjptrA             = f+j_coord_offsetA;
 495             fjptrB             = f+j_coord_offsetB;
 496             fjptrC             = f+j_coord_offsetC;
 497             fjptrD             = f+j_coord_offsetD;
 498             fjptrE             = f+j_coord_offsetE;
 499             fjptrF             = f+j_coord_offsetF;
 500             fjptrG             = f+j_coord_offsetG;
 501             fjptrH             = f+j_coord_offsetH;
 502
 503             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 504
 505             /* Inner loop uses 311 flops */
 506         }
 507
 508         if(jidx<j_index_end)
 509         {
 510
 511             /* Get j neighbor index, and coordinate index */
 512             jnrlistA         = jjnr[jidx];
 513             jnrlistB         = jjnr[jidx+1];
 514             jnrlistC         = jjnr[jidx+2];
 515             jnrlistD         = jjnr[jidx+3];
 516             jnrlistE         = jjnr[jidx+4];
 517             jnrlistF         = jjnr[jidx+5];
 518             jnrlistG         = jjnr[jidx+6];
 519             jnrlistH         = jjnr[jidx+7];
 520             /* Sign of each element will be negative for non-real atoms.
 521              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 522              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 523              */
 524             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 525                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 526
 527             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 528             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 529             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 530             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 531             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 532             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 533             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 534             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 535             j_coord_offsetA  = DIM*jnrA;
 536             j_coord_offsetB  = DIM*jnrB;
 537             j_coord_offsetC  = DIM*jnrC;
 538             j_coord_offsetD  = DIM*jnrD;
 539             j_coord_offsetE  = DIM*jnrE;
 540             j_coord_offsetF  = DIM*jnrF;
 541             j_coord_offsetG  = DIM*jnrG;
 542             j_coord_offsetH  = DIM*jnrH;
 543
 544             /* load j atom coordinates */
 545             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 546                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 547                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 548                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 549                                                  &jx0,&jy0,&jz0);
 550
 551             /* Calculate displacement vector */
 552             dx00             = _mm256_sub_ps(ix0,jx0);
 553             dy00             = _mm256_sub_ps(iy0,jy0);
 554             dz00             = _mm256_sub_ps(iz0,jz0);
 555             dx10             = _mm256_sub_ps(ix1,jx0);
 556             dy10             = _mm256_sub_ps(iy1,jy0);
 557             dz10             = _mm256_sub_ps(iz1,jz0);
 558             dx20             = _mm256_sub_ps(ix2,jx0);
 559             dy20             = _mm256_sub_ps(iy2,jy0);
 560             dz20             = _mm256_sub_ps(iz2,jz0);
 561             dx30             = _mm256_sub_ps(ix3,jx0);
 562             dy30             = _mm256_sub_ps(iy3,jy0);
 563             dz30             = _mm256_sub_ps(iz3,jz0);
 564
 565             /* Calculate squared distance and things based on it */
 566             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 567             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 568             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 569             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 570
 571             rinv00           = avx256_invsqrt_f(rsq00);
 572             rinv10           = avx256_invsqrt_f(rsq10);
 573             rinv20           = avx256_invsqrt_f(rsq20);
 574             rinv30           = avx256_invsqrt_f(rsq30);
 575
 576             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 577             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 578             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 579
 580             /* Load parameters for j particles */
 581             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 582                                                                  charge+jnrC+0,charge+jnrD+0,
 583                                                                  charge+jnrE+0,charge+jnrF+0,
 584                                                                  charge+jnrG+0,charge+jnrH+0);
 585             vdwjidx0A        = 2*vdwtype[jnrA+0];
 586             vdwjidx0B        = 2*vdwtype[jnrB+0];
 587             vdwjidx0C        = 2*vdwtype[jnrC+0];
 588             vdwjidx0D        = 2*vdwtype[jnrD+0];
 589             vdwjidx0E        = 2*vdwtype[jnrE+0];
 590             vdwjidx0F        = 2*vdwtype[jnrF+0];
 591             vdwjidx0G        = 2*vdwtype[jnrG+0];
 592             vdwjidx0H        = 2*vdwtype[jnrH+0];
 593
 594             fjx0             = _mm256_setzero_ps();
 595             fjy0             = _mm256_setzero_ps();
 596             fjz0             = _mm256_setzero_ps();
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             r00              = _mm256_mul_ps(rsq00,rinv00);
 603             r00              = _mm256_andnot_ps(dummy_mask,r00);
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 607                                             vdwioffsetptr0+vdwjidx0B,
 608                                             vdwioffsetptr0+vdwjidx0C,
 609                                             vdwioffsetptr0+vdwjidx0D,
 610                                             vdwioffsetptr0+vdwjidx0E,
 611                                             vdwioffsetptr0+vdwjidx0F,
 612                                             vdwioffsetptr0+vdwjidx0G,
 613                                             vdwioffsetptr0+vdwjidx0H,
 614                                             &c6_00,&c12_00);
 615
 616             /* Calculate table index by multiplying r with table scale and truncate to integer */
 617             rt               = _mm256_mul_ps(r00,vftabscale);
 618             vfitab           = _mm256_cvttps_epi32(rt);
 619             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 620             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 621             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 622             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 623             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 624             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 625
 626             /* CUBIC SPLINE TABLE DISPERSION */
 627             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 628                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 629             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 630                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 631             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 632                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 633             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 634                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 635             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 636             Heps             = _mm256_mul_ps(vfeps,H);
 637             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 638             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 639             vvdw6            = _mm256_mul_ps(c6_00,VV);
 640             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 641             fvdw6            = _mm256_mul_ps(c6_00,FF);
 642
 643             /* CUBIC SPLINE TABLE REPULSION */
 644             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 645             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 646             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 647                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 648             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 649                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 650             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 651                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 652             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 653                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 654             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 655             Heps             = _mm256_mul_ps(vfeps,H);
 656             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 657             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 658             vvdw12           = _mm256_mul_ps(c12_00,VV);
 659             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 660             fvdw12           = _mm256_mul_ps(c12_00,FF);
 661             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 662             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 663
 664             /* Update potential sum for this i atom from the interaction with this j atom. */
 665             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 666             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 667
 668             fscal            = fvdw;
 669
 670             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 671
 672             /* Calculate temporary vectorial force */
 673             tx               = _mm256_mul_ps(fscal,dx00);
 674             ty               = _mm256_mul_ps(fscal,dy00);
 675             tz               = _mm256_mul_ps(fscal,dz00);
 676
 677             /* Update vectorial force */
 678             fix0             = _mm256_add_ps(fix0,tx);
 679             fiy0             = _mm256_add_ps(fiy0,ty);
 680             fiz0             = _mm256_add_ps(fiz0,tz);
 681
 682             fjx0             = _mm256_add_ps(fjx0,tx);
 683             fjy0             = _mm256_add_ps(fjy0,ty);
 684             fjz0             = _mm256_add_ps(fjz0,tz);
 685
 686             /**************************
 687              * CALCULATE INTERACTIONS *
 688              **************************/
 689
 690             r10              = _mm256_mul_ps(rsq10,rinv10);
 691             r10              = _mm256_andnot_ps(dummy_mask,r10);
 692
 693             /* Compute parameters for interactions between i and j atoms */
 694             qq10             = _mm256_mul_ps(iq1,jq0);
 695
 696             /* EWALD ELECTROSTATICS */
 697
 698             /* Analytical PME correction */
 699             zeta2            = _mm256_mul_ps(beta2,rsq10);
 700             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 701             pmecorrF         = avx256_pmecorrF_f(zeta2);
 702             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 703             felec            = _mm256_mul_ps(qq10,felec);
 704             pmecorrV         = avx256_pmecorrV_f(zeta2);
 705             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 706             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 707             velec            = _mm256_mul_ps(qq10,velec);
 708
 709             /* Update potential sum for this i atom from the interaction with this j atom. */
 710             velec            = _mm256_andnot_ps(dummy_mask,velec);
 711             velecsum         = _mm256_add_ps(velecsum,velec);
 712
 713             fscal            = felec;
 714
 715             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 716
 717             /* Calculate temporary vectorial force */
 718             tx               = _mm256_mul_ps(fscal,dx10);
 719             ty               = _mm256_mul_ps(fscal,dy10);
 720             tz               = _mm256_mul_ps(fscal,dz10);
 721
 722             /* Update vectorial force */
 723             fix1             = _mm256_add_ps(fix1,tx);
 724             fiy1             = _mm256_add_ps(fiy1,ty);
 725             fiz1             = _mm256_add_ps(fiz1,tz);
 726
 727             fjx0             = _mm256_add_ps(fjx0,tx);
 728             fjy0             = _mm256_add_ps(fjy0,ty);
 729             fjz0             = _mm256_add_ps(fjz0,tz);
 730
 731             /**************************
 732              * CALCULATE INTERACTIONS *
 733              **************************/
 734
 735             r20              = _mm256_mul_ps(rsq20,rinv20);
 736             r20              = _mm256_andnot_ps(dummy_mask,r20);
 737
 738             /* Compute parameters for interactions between i and j atoms */
 739             qq20             = _mm256_mul_ps(iq2,jq0);
 740
 741             /* EWALD ELECTROSTATICS */
 742
 743             /* Analytical PME correction */
 744             zeta2            = _mm256_mul_ps(beta2,rsq20);
 745             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 746             pmecorrF         = avx256_pmecorrF_f(zeta2);
 747             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 748             felec            = _mm256_mul_ps(qq20,felec);
 749             pmecorrV         = avx256_pmecorrV_f(zeta2);
 750             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 751             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 752             velec            = _mm256_mul_ps(qq20,velec);
 753
 754             /* Update potential sum for this i atom from the interaction with this j atom. */
 755             velec            = _mm256_andnot_ps(dummy_mask,velec);
 756             velecsum         = _mm256_add_ps(velecsum,velec);
 757
 758             fscal            = felec;
 759
 760             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 761
 762             /* Calculate temporary vectorial force */
 763             tx               = _mm256_mul_ps(fscal,dx20);
 764             ty               = _mm256_mul_ps(fscal,dy20);
 765             tz               = _mm256_mul_ps(fscal,dz20);
 766
 767             /* Update vectorial force */
 768             fix2             = _mm256_add_ps(fix2,tx);
 769             fiy2             = _mm256_add_ps(fiy2,ty);
 770             fiz2             = _mm256_add_ps(fiz2,tz);
 771
 772             fjx0             = _mm256_add_ps(fjx0,tx);
 773             fjy0             = _mm256_add_ps(fjy0,ty);
 774             fjz0             = _mm256_add_ps(fjz0,tz);
 775
 776             /**************************
 777              * CALCULATE INTERACTIONS *
 778              **************************/
 779
 780             r30              = _mm256_mul_ps(rsq30,rinv30);
 781             r30              = _mm256_andnot_ps(dummy_mask,r30);
 782
 783             /* Compute parameters for interactions between i and j atoms */
 784             qq30             = _mm256_mul_ps(iq3,jq0);
 785
 786             /* EWALD ELECTROSTATICS */
 787
 788             /* Analytical PME correction */
 789             zeta2            = _mm256_mul_ps(beta2,rsq30);
 790             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 791             pmecorrF         = avx256_pmecorrF_f(zeta2);
 792             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 793             felec            = _mm256_mul_ps(qq30,felec);
 794             pmecorrV         = avx256_pmecorrV_f(zeta2);
 795             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 796             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 797             velec            = _mm256_mul_ps(qq30,velec);
 798
 799             /* Update potential sum for this i atom from the interaction with this j atom. */
 800             velec            = _mm256_andnot_ps(dummy_mask,velec);
 801             velecsum         = _mm256_add_ps(velecsum,velec);
 802
 803             fscal            = felec;
 804
 805             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 806
 807             /* Calculate temporary vectorial force */
 808             tx               = _mm256_mul_ps(fscal,dx30);
 809             ty               = _mm256_mul_ps(fscal,dy30);
 810             tz               = _mm256_mul_ps(fscal,dz30);
 811
 812             /* Update vectorial force */
 813             fix3             = _mm256_add_ps(fix3,tx);
 814             fiy3             = _mm256_add_ps(fiy3,ty);
 815             fiz3             = _mm256_add_ps(fiz3,tz);
 816
 817             fjx0             = _mm256_add_ps(fjx0,tx);
 818             fjy0             = _mm256_add_ps(fjy0,ty);
 819             fjz0             = _mm256_add_ps(fjz0,tz);
 820
 821             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 822             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 823             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 824             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 825             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 826             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 827             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 828             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 829
 830             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 831
 832             /* Inner loop uses 315 flops */
 833         }
 834
 835         /* End of innermost loop */
 836
 837         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 838                                                  f+i_coord_offset,fshift+i_shift_offset);
 839
 840         ggid                        = gid[iidx];
 841         /* Update potential energies */
 842         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 843         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 844
 845         /* Increment number of inner iterations */
 846         inneriter                  += j_index_end - j_index_start;
 847
 848         /* Outer loop uses 26 flops */
 849     }
 850
 851     /* Increment number of outer iterations */
 852     outeriter        += nri;
 853
 854     /* Update outer/inner flops */
 855
 856     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*315);
 857 }
 858 /*
 859  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_256_single
 860  * Electrostatics interaction: Ewald
 861  * VdW interaction:            CubicSplineTable
 862  * Geometry:                   Water4-Particle
 863  * Calculate force/pot:        Force
 864  */
 865 void
 866 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_256_single
 867                     (t_nblist                    * gmx_restrict       nlist,
 868                      rvec                        * gmx_restrict          xx,
 869                      rvec                        * gmx_restrict          ff,
 870                      struct t_forcerec           * gmx_restrict          fr,
 871                      t_mdatoms                   * gmx_restrict     mdatoms,
 872                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 873                      t_nrnb                      * gmx_restrict        nrnb)
 874 {
 875     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 876      * just 0 for non-waters.
 877      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 878      * jnr indices corresponding to data put in the four positions in the SIMD register.
 879      */
 880     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 881     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 882     int              jnrA,jnrB,jnrC,jnrD;
 883     int              jnrE,jnrF,jnrG,jnrH;
 884     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 885     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 886     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 887     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 888     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 889     real             rcutoff_scalar;
 890     real             *shiftvec,*fshift,*x,*f;
 891     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 892     real             scratch[4*DIM];
 893     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 894     real *           vdwioffsetptr0;
 895     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 896     real *           vdwioffsetptr1;
 897     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 898     real *           vdwioffsetptr2;
 899     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 900     real *           vdwioffsetptr3;
 901     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 902     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 903     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 904     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 905     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 906     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 907     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 908     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 909     real             *charge;
 910     int              nvdwtype;
 911     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 912     int              *vdwtype;
 913     real             *vdwparam;
 914     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 915     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 916     __m256i          vfitab;
 917     __m128i          vfitab_lo,vfitab_hi;
 918     __m128i          ifour       = _mm_set1_epi32(4);
 919     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 920     real             *vftab;
 921     __m256i          ewitab;
 922     __m128i          ewitab_lo,ewitab_hi;
 923     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 924     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 925     real             *ewtab;
 926     __m256           dummy_mask,cutoff_mask;
 927     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 928     __m256           one     = _mm256_set1_ps(1.0);
 929     __m256           two     = _mm256_set1_ps(2.0);
 930     x                = xx[0];
 931     f                = ff[0];
 932
 933     nri              = nlist->nri;
 934     iinr             = nlist->iinr;
 935     jindex           = nlist->jindex;
 936     jjnr             = nlist->jjnr;
 937     shiftidx         = nlist->shift;
 938     gid              = nlist->gid;
 939     shiftvec         = fr->shift_vec[0];
 940     fshift           = fr->fshift[0];
 941     facel            = _mm256_set1_ps(fr->ic->epsfac);
 942     charge           = mdatoms->chargeA;
 943     nvdwtype         = fr->ntype;
 944     vdwparam         = fr->nbfp;
 945     vdwtype          = mdatoms->typeA;
 946
 947     vftab            = kernel_data->table_vdw->data;
 948     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 949
 950     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 951     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 952     beta2            = _mm256_mul_ps(beta,beta);
 953     beta3            = _mm256_mul_ps(beta,beta2);
 954
 955     ewtab            = fr->ic->tabq_coul_F;
 956     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 957     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 958
 959     /* Setup water-specific parameters */
 960     inr              = nlist->iinr[0];
 961     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 962     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 963     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 964     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 965
 966     /* Avoid stupid compiler warnings */
 967     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 968     j_coord_offsetA = 0;
 969     j_coord_offsetB = 0;
 970     j_coord_offsetC = 0;
 971     j_coord_offsetD = 0;
 972     j_coord_offsetE = 0;
 973     j_coord_offsetF = 0;
 974     j_coord_offsetG = 0;
 975     j_coord_offsetH = 0;
 976
 977     outeriter        = 0;
 978     inneriter        = 0;
 979
 980     for(iidx=0;iidx<4*DIM;iidx++)
 981     {
 982         scratch[iidx] = 0.0;
 983     }
 984
 985     /* Start outer loop over neighborlists */
 986     for(iidx=0; iidx<nri; iidx++)
 987     {
 988         /* Load shift vector for this list */
 989         i_shift_offset   = DIM*shiftidx[iidx];
 990
 991         /* Load limits for loop over neighbors */
 992         j_index_start    = jindex[iidx];
 993         j_index_end      = jindex[iidx+1];
 994
 995         /* Get outer coordinate index */
 996         inr              = iinr[iidx];
 997         i_coord_offset   = DIM*inr;
 998
 999         /* Load i particle coords and add shift vector */
1000         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1001                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1002
1003         fix0             = _mm256_setzero_ps();
1004         fiy0             = _mm256_setzero_ps();
1005         fiz0             = _mm256_setzero_ps();
1006         fix1             = _mm256_setzero_ps();
1007         fiy1             = _mm256_setzero_ps();
1008         fiz1             = _mm256_setzero_ps();
1009         fix2             = _mm256_setzero_ps();
1010         fiy2             = _mm256_setzero_ps();
1011         fiz2             = _mm256_setzero_ps();
1012         fix3             = _mm256_setzero_ps();
1013         fiy3             = _mm256_setzero_ps();
1014         fiz3             = _mm256_setzero_ps();
1015
1016         /* Start inner kernel loop */
1017         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1018         {
1019
1020             /* Get j neighbor index, and coordinate index */
1021             jnrA             = jjnr[jidx];
1022             jnrB             = jjnr[jidx+1];
1023             jnrC             = jjnr[jidx+2];
1024             jnrD             = jjnr[jidx+3];
1025             jnrE             = jjnr[jidx+4];
1026             jnrF             = jjnr[jidx+5];
1027             jnrG             = jjnr[jidx+6];
1028             jnrH             = jjnr[jidx+7];
1029             j_coord_offsetA  = DIM*jnrA;
1030             j_coord_offsetB  = DIM*jnrB;
1031             j_coord_offsetC  = DIM*jnrC;
1032             j_coord_offsetD  = DIM*jnrD;
1033             j_coord_offsetE  = DIM*jnrE;
1034             j_coord_offsetF  = DIM*jnrF;
1035             j_coord_offsetG  = DIM*jnrG;
1036             j_coord_offsetH  = DIM*jnrH;
1037
1038             /* load j atom coordinates */
1039             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1040                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1041                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1042                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1043                                                  &jx0,&jy0,&jz0);
1044
1045             /* Calculate displacement vector */
1046             dx00             = _mm256_sub_ps(ix0,jx0);
1047             dy00             = _mm256_sub_ps(iy0,jy0);
1048             dz00             = _mm256_sub_ps(iz0,jz0);
1049             dx10             = _mm256_sub_ps(ix1,jx0);
1050             dy10             = _mm256_sub_ps(iy1,jy0);
1051             dz10             = _mm256_sub_ps(iz1,jz0);
1052             dx20             = _mm256_sub_ps(ix2,jx0);
1053             dy20             = _mm256_sub_ps(iy2,jy0);
1054             dz20             = _mm256_sub_ps(iz2,jz0);
1055             dx30             = _mm256_sub_ps(ix3,jx0);
1056             dy30             = _mm256_sub_ps(iy3,jy0);
1057             dz30             = _mm256_sub_ps(iz3,jz0);
1058
1059             /* Calculate squared distance and things based on it */
1060             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1061             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1062             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1063             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1064
1065             rinv00           = avx256_invsqrt_f(rsq00);
1066             rinv10           = avx256_invsqrt_f(rsq10);
1067             rinv20           = avx256_invsqrt_f(rsq20);
1068             rinv30           = avx256_invsqrt_f(rsq30);
1069
1070             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1071             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1072             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1073
1074             /* Load parameters for j particles */
1075             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1076                                                                  charge+jnrC+0,charge+jnrD+0,
1077                                                                  charge+jnrE+0,charge+jnrF+0,
1078                                                                  charge+jnrG+0,charge+jnrH+0);
1079             vdwjidx0A        = 2*vdwtype[jnrA+0];
1080             vdwjidx0B        = 2*vdwtype[jnrB+0];
1081             vdwjidx0C        = 2*vdwtype[jnrC+0];
1082             vdwjidx0D        = 2*vdwtype[jnrD+0];
1083             vdwjidx0E        = 2*vdwtype[jnrE+0];
1084             vdwjidx0F        = 2*vdwtype[jnrF+0];
1085             vdwjidx0G        = 2*vdwtype[jnrG+0];
1086             vdwjidx0H        = 2*vdwtype[jnrH+0];
1087
1088             fjx0             = _mm256_setzero_ps();
1089             fjy0             = _mm256_setzero_ps();
1090             fjz0             = _mm256_setzero_ps();
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             r00              = _mm256_mul_ps(rsq00,rinv00);
1097
1098             /* Compute parameters for interactions between i and j atoms */
1099             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1100                                             vdwioffsetptr0+vdwjidx0B,
1101                                             vdwioffsetptr0+vdwjidx0C,
1102                                             vdwioffsetptr0+vdwjidx0D,
1103                                             vdwioffsetptr0+vdwjidx0E,
1104                                             vdwioffsetptr0+vdwjidx0F,
1105                                             vdwioffsetptr0+vdwjidx0G,
1106                                             vdwioffsetptr0+vdwjidx0H,
1107                                             &c6_00,&c12_00);
1108
1109             /* Calculate table index by multiplying r with table scale and truncate to integer */
1110             rt               = _mm256_mul_ps(r00,vftabscale);
1111             vfitab           = _mm256_cvttps_epi32(rt);
1112             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1113             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1114             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1115             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1116             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1117             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1118
1119             /* CUBIC SPLINE TABLE DISPERSION */
1120             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1121                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1122             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1123                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1124             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1125                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1126             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1127                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1128             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1129             Heps             = _mm256_mul_ps(vfeps,H);
1130             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1131             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1132             fvdw6            = _mm256_mul_ps(c6_00,FF);
1133
1134             /* CUBIC SPLINE TABLE REPULSION */
1135             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1136             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1137             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1138                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1139             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1140                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1141             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1142                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1143             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1144                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1145             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1146             Heps             = _mm256_mul_ps(vfeps,H);
1147             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1148             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1149             fvdw12           = _mm256_mul_ps(c12_00,FF);
1150             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1151
1152             fscal            = fvdw;
1153
1154             /* Calculate temporary vectorial force */
1155             tx               = _mm256_mul_ps(fscal,dx00);
1156             ty               = _mm256_mul_ps(fscal,dy00);
1157             tz               = _mm256_mul_ps(fscal,dz00);
1158
1159             /* Update vectorial force */
1160             fix0             = _mm256_add_ps(fix0,tx);
1161             fiy0             = _mm256_add_ps(fiy0,ty);
1162             fiz0             = _mm256_add_ps(fiz0,tz);
1163
1164             fjx0             = _mm256_add_ps(fjx0,tx);
1165             fjy0             = _mm256_add_ps(fjy0,ty);
1166             fjz0             = _mm256_add_ps(fjz0,tz);
1167
1168             /**************************
1169              * CALCULATE INTERACTIONS *
1170              **************************/
1171
1172             r10              = _mm256_mul_ps(rsq10,rinv10);
1173
1174             /* Compute parameters for interactions between i and j atoms */
1175             qq10             = _mm256_mul_ps(iq1,jq0);
1176
1177             /* EWALD ELECTROSTATICS */
1178
1179             /* Analytical PME correction */
1180             zeta2            = _mm256_mul_ps(beta2,rsq10);
1181             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1182             pmecorrF         = avx256_pmecorrF_f(zeta2);
1183             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1184             felec            = _mm256_mul_ps(qq10,felec);
1185
1186             fscal            = felec;
1187
1188             /* Calculate temporary vectorial force */
1189             tx               = _mm256_mul_ps(fscal,dx10);
1190             ty               = _mm256_mul_ps(fscal,dy10);
1191             tz               = _mm256_mul_ps(fscal,dz10);
1192
1193             /* Update vectorial force */
1194             fix1             = _mm256_add_ps(fix1,tx);
1195             fiy1             = _mm256_add_ps(fiy1,ty);
1196             fiz1             = _mm256_add_ps(fiz1,tz);
1197
1198             fjx0             = _mm256_add_ps(fjx0,tx);
1199             fjy0             = _mm256_add_ps(fjy0,ty);
1200             fjz0             = _mm256_add_ps(fjz0,tz);
1201
1202             /**************************
1203              * CALCULATE INTERACTIONS *
1204              **************************/
1205
1206             r20              = _mm256_mul_ps(rsq20,rinv20);
1207
1208             /* Compute parameters for interactions between i and j atoms */
1209             qq20             = _mm256_mul_ps(iq2,jq0);
1210
1211             /* EWALD ELECTROSTATICS */
1212
1213             /* Analytical PME correction */
1214             zeta2            = _mm256_mul_ps(beta2,rsq20);
1215             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1216             pmecorrF         = avx256_pmecorrF_f(zeta2);
1217             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1218             felec            = _mm256_mul_ps(qq20,felec);
1219
1220             fscal            = felec;
1221
1222             /* Calculate temporary vectorial force */
1223             tx               = _mm256_mul_ps(fscal,dx20);
1224             ty               = _mm256_mul_ps(fscal,dy20);
1225             tz               = _mm256_mul_ps(fscal,dz20);
1226
1227             /* Update vectorial force */
1228             fix2             = _mm256_add_ps(fix2,tx);
1229             fiy2             = _mm256_add_ps(fiy2,ty);
1230             fiz2             = _mm256_add_ps(fiz2,tz);
1231
1232             fjx0             = _mm256_add_ps(fjx0,tx);
1233             fjy0             = _mm256_add_ps(fjy0,ty);
1234             fjz0             = _mm256_add_ps(fjz0,tz);
1235
1236             /**************************
1237              * CALCULATE INTERACTIONS *
1238              **************************/
1239
1240             r30              = _mm256_mul_ps(rsq30,rinv30);
1241
1242             /* Compute parameters for interactions between i and j atoms */
1243             qq30             = _mm256_mul_ps(iq3,jq0);
1244
1245             /* EWALD ELECTROSTATICS */
1246
1247             /* Analytical PME correction */
1248             zeta2            = _mm256_mul_ps(beta2,rsq30);
1249             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1250             pmecorrF         = avx256_pmecorrF_f(zeta2);
1251             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1252             felec            = _mm256_mul_ps(qq30,felec);
1253
1254             fscal            = felec;
1255
1256             /* Calculate temporary vectorial force */
1257             tx               = _mm256_mul_ps(fscal,dx30);
1258             ty               = _mm256_mul_ps(fscal,dy30);
1259             tz               = _mm256_mul_ps(fscal,dz30);
1260
1261             /* Update vectorial force */
1262             fix3             = _mm256_add_ps(fix3,tx);
1263             fiy3             = _mm256_add_ps(fiy3,ty);
1264             fiz3             = _mm256_add_ps(fiz3,tz);
1265
1266             fjx0             = _mm256_add_ps(fjx0,tx);
1267             fjy0             = _mm256_add_ps(fjy0,ty);
1268             fjz0             = _mm256_add_ps(fjz0,tz);
1269
1270             fjptrA             = f+j_coord_offsetA;
1271             fjptrB             = f+j_coord_offsetB;
1272             fjptrC             = f+j_coord_offsetC;
1273             fjptrD             = f+j_coord_offsetD;
1274             fjptrE             = f+j_coord_offsetE;
1275             fjptrF             = f+j_coord_offsetF;
1276             fjptrG             = f+j_coord_offsetG;
1277             fjptrH             = f+j_coord_offsetH;
1278
1279             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1280
1281             /* Inner loop uses 219 flops */
1282         }
1283
1284         if(jidx<j_index_end)
1285         {
1286
1287             /* Get j neighbor index, and coordinate index */
1288             jnrlistA         = jjnr[jidx];
1289             jnrlistB         = jjnr[jidx+1];
1290             jnrlistC         = jjnr[jidx+2];
1291             jnrlistD         = jjnr[jidx+3];
1292             jnrlistE         = jjnr[jidx+4];
1293             jnrlistF         = jjnr[jidx+5];
1294             jnrlistG         = jjnr[jidx+6];
1295             jnrlistH         = jjnr[jidx+7];
1296             /* Sign of each element will be negative for non-real atoms.
1297              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1298              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1299              */
1300             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1301                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1302
1303             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1304             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1305             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1306             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1307             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1308             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1309             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1310             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1311             j_coord_offsetA  = DIM*jnrA;
1312             j_coord_offsetB  = DIM*jnrB;
1313             j_coord_offsetC  = DIM*jnrC;
1314             j_coord_offsetD  = DIM*jnrD;
1315             j_coord_offsetE  = DIM*jnrE;
1316             j_coord_offsetF  = DIM*jnrF;
1317             j_coord_offsetG  = DIM*jnrG;
1318             j_coord_offsetH  = DIM*jnrH;
1319
1320             /* load j atom coordinates */
1321             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1322                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1323                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1324                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1325                                                  &jx0,&jy0,&jz0);
1326
1327             /* Calculate displacement vector */
1328             dx00             = _mm256_sub_ps(ix0,jx0);
1329             dy00             = _mm256_sub_ps(iy0,jy0);
1330             dz00             = _mm256_sub_ps(iz0,jz0);
1331             dx10             = _mm256_sub_ps(ix1,jx0);
1332             dy10             = _mm256_sub_ps(iy1,jy0);
1333             dz10             = _mm256_sub_ps(iz1,jz0);
1334             dx20             = _mm256_sub_ps(ix2,jx0);
1335             dy20             = _mm256_sub_ps(iy2,jy0);
1336             dz20             = _mm256_sub_ps(iz2,jz0);
1337             dx30             = _mm256_sub_ps(ix3,jx0);
1338             dy30             = _mm256_sub_ps(iy3,jy0);
1339             dz30             = _mm256_sub_ps(iz3,jz0);
1340
1341             /* Calculate squared distance and things based on it */
1342             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1343             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1344             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1345             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1346
1347             rinv00           = avx256_invsqrt_f(rsq00);
1348             rinv10           = avx256_invsqrt_f(rsq10);
1349             rinv20           = avx256_invsqrt_f(rsq20);
1350             rinv30           = avx256_invsqrt_f(rsq30);
1351
1352             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1353             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1354             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1355
1356             /* Load parameters for j particles */
1357             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1358                                                                  charge+jnrC+0,charge+jnrD+0,
1359                                                                  charge+jnrE+0,charge+jnrF+0,
1360                                                                  charge+jnrG+0,charge+jnrH+0);
1361             vdwjidx0A        = 2*vdwtype[jnrA+0];
1362             vdwjidx0B        = 2*vdwtype[jnrB+0];
1363             vdwjidx0C        = 2*vdwtype[jnrC+0];
1364             vdwjidx0D        = 2*vdwtype[jnrD+0];
1365             vdwjidx0E        = 2*vdwtype[jnrE+0];
1366             vdwjidx0F        = 2*vdwtype[jnrF+0];
1367             vdwjidx0G        = 2*vdwtype[jnrG+0];
1368             vdwjidx0H        = 2*vdwtype[jnrH+0];
1369
1370             fjx0             = _mm256_setzero_ps();
1371             fjy0             = _mm256_setzero_ps();
1372             fjz0             = _mm256_setzero_ps();
1373
1374             /**************************
1375              * CALCULATE INTERACTIONS *
1376              **************************/
1377
1378             r00              = _mm256_mul_ps(rsq00,rinv00);
1379             r00              = _mm256_andnot_ps(dummy_mask,r00);
1380
1381             /* Compute parameters for interactions between i and j atoms */
1382             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1383                                             vdwioffsetptr0+vdwjidx0B,
1384                                             vdwioffsetptr0+vdwjidx0C,
1385                                             vdwioffsetptr0+vdwjidx0D,
1386                                             vdwioffsetptr0+vdwjidx0E,
1387                                             vdwioffsetptr0+vdwjidx0F,
1388                                             vdwioffsetptr0+vdwjidx0G,
1389                                             vdwioffsetptr0+vdwjidx0H,
1390                                             &c6_00,&c12_00);
1391
1392             /* Calculate table index by multiplying r with table scale and truncate to integer */
1393             rt               = _mm256_mul_ps(r00,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,3);
1400             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1401
1402             /* CUBIC SPLINE TABLE DISPERSION */
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             fvdw6            = _mm256_mul_ps(c6_00,FF);
1416
1417             /* CUBIC SPLINE TABLE REPULSION */
1418             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1419             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1420             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1421                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1422             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1423                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1424             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1425                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1426             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1427                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1428             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1429             Heps             = _mm256_mul_ps(vfeps,H);
1430             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1431             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1432             fvdw12           = _mm256_mul_ps(c12_00,FF);
1433             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1434
1435             fscal            = fvdw;
1436
1437             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1438
1439             /* Calculate temporary vectorial force */
1440             tx               = _mm256_mul_ps(fscal,dx00);
1441             ty               = _mm256_mul_ps(fscal,dy00);
1442             tz               = _mm256_mul_ps(fscal,dz00);
1443
1444             /* Update vectorial force */
1445             fix0             = _mm256_add_ps(fix0,tx);
1446             fiy0             = _mm256_add_ps(fiy0,ty);
1447             fiz0             = _mm256_add_ps(fiz0,tz);
1448
1449             fjx0             = _mm256_add_ps(fjx0,tx);
1450             fjy0             = _mm256_add_ps(fjy0,ty);
1451             fjz0             = _mm256_add_ps(fjz0,tz);
1452
1453             /**************************
1454              * CALCULATE INTERACTIONS *
1455              **************************/
1456
1457             r10              = _mm256_mul_ps(rsq10,rinv10);
1458             r10              = _mm256_andnot_ps(dummy_mask,r10);
1459
1460             /* Compute parameters for interactions between i and j atoms */
1461             qq10             = _mm256_mul_ps(iq1,jq0);
1462
1463             /* EWALD ELECTROSTATICS */
1464
1465             /* Analytical PME correction */
1466             zeta2            = _mm256_mul_ps(beta2,rsq10);
1467             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1468             pmecorrF         = avx256_pmecorrF_f(zeta2);
1469             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1470             felec            = _mm256_mul_ps(qq10,felec);
1471
1472             fscal            = felec;
1473
1474             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1475
1476             /* Calculate temporary vectorial force */
1477             tx               = _mm256_mul_ps(fscal,dx10);
1478             ty               = _mm256_mul_ps(fscal,dy10);
1479             tz               = _mm256_mul_ps(fscal,dz10);
1480
1481             /* Update vectorial force */
1482             fix1             = _mm256_add_ps(fix1,tx);
1483             fiy1             = _mm256_add_ps(fiy1,ty);
1484             fiz1             = _mm256_add_ps(fiz1,tz);
1485
1486             fjx0             = _mm256_add_ps(fjx0,tx);
1487             fjy0             = _mm256_add_ps(fjy0,ty);
1488             fjz0             = _mm256_add_ps(fjz0,tz);
1489
1490             /**************************
1491              * CALCULATE INTERACTIONS *
1492              **************************/
1493
1494             r20              = _mm256_mul_ps(rsq20,rinv20);
1495             r20              = _mm256_andnot_ps(dummy_mask,r20);
1496
1497             /* Compute parameters for interactions between i and j atoms */
1498             qq20             = _mm256_mul_ps(iq2,jq0);
1499
1500             /* EWALD ELECTROSTATICS */
1501
1502             /* Analytical PME correction */
1503             zeta2            = _mm256_mul_ps(beta2,rsq20);
1504             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1505             pmecorrF         = avx256_pmecorrF_f(zeta2);
1506             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1507             felec            = _mm256_mul_ps(qq20,felec);
1508
1509             fscal            = felec;
1510
1511             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1512
1513             /* Calculate temporary vectorial force */
1514             tx               = _mm256_mul_ps(fscal,dx20);
1515             ty               = _mm256_mul_ps(fscal,dy20);
1516             tz               = _mm256_mul_ps(fscal,dz20);
1517
1518             /* Update vectorial force */
1519             fix2             = _mm256_add_ps(fix2,tx);
1520             fiy2             = _mm256_add_ps(fiy2,ty);
1521             fiz2             = _mm256_add_ps(fiz2,tz);
1522
1523             fjx0             = _mm256_add_ps(fjx0,tx);
1524             fjy0             = _mm256_add_ps(fjy0,ty);
1525             fjz0             = _mm256_add_ps(fjz0,tz);
1526
1527             /**************************
1528              * CALCULATE INTERACTIONS *
1529              **************************/
1530
1531             r30              = _mm256_mul_ps(rsq30,rinv30);
1532             r30              = _mm256_andnot_ps(dummy_mask,r30);
1533
1534             /* Compute parameters for interactions between i and j atoms */
1535             qq30             = _mm256_mul_ps(iq3,jq0);
1536
1537             /* EWALD ELECTROSTATICS */
1538
1539             /* Analytical PME correction */
1540             zeta2            = _mm256_mul_ps(beta2,rsq30);
1541             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1542             pmecorrF         = avx256_pmecorrF_f(zeta2);
1543             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1544             felec            = _mm256_mul_ps(qq30,felec);
1545
1546             fscal            = felec;
1547
1548             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1549
1550             /* Calculate temporary vectorial force */
1551             tx               = _mm256_mul_ps(fscal,dx30);
1552             ty               = _mm256_mul_ps(fscal,dy30);
1553             tz               = _mm256_mul_ps(fscal,dz30);
1554
1555             /* Update vectorial force */
1556             fix3             = _mm256_add_ps(fix3,tx);
1557             fiy3             = _mm256_add_ps(fiy3,ty);
1558             fiz3             = _mm256_add_ps(fiz3,tz);
1559
1560             fjx0             = _mm256_add_ps(fjx0,tx);
1561             fjy0             = _mm256_add_ps(fjy0,ty);
1562             fjz0             = _mm256_add_ps(fjz0,tz);
1563
1564             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1565             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1566             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1567             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1568             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1569             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1570             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1571             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1572
1573             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1574
1575             /* Inner loop uses 223 flops */
1576         }
1577
1578         /* End of innermost loop */
1579
1580         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1581                                                  f+i_coord_offset,fshift+i_shift_offset);
1582
1583         /* Increment number of inner iterations */
1584         inneriter                  += j_index_end - j_index_start;
1585
1586         /* Outer loop uses 24 flops */
1587     }
1588
1589     /* Increment number of outer iterations */
1590     outeriter        += nri;
1591
1592     /* Update outer/inner flops */
1593
1594     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*223);
1595 }