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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_ElecEw_VdwCSTab_GeomW4P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwCSTab_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     __m256i          ewitab;
 116     __m128i          ewitab_lo,ewitab_hi;
 117     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 118     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 119     real             *ewtab;
 120     __m256           dummy_mask,cutoff_mask;
 121     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 122     __m256           one     = _mm256_set1_ps(1.0);
 123     __m256           two     = _mm256_set1_ps(2.0);
 124     x                = xx[0];
 125     f                = ff[0];
 126
 127     nri              = nlist->nri;
 128     iinr             = nlist->iinr;
 129     jindex           = nlist->jindex;
 130     jjnr             = nlist->jjnr;
 131     shiftidx         = nlist->shift;
 132     gid              = nlist->gid;
 133     shiftvec         = fr->shift_vec[0];
 134     fshift           = fr->fshift[0];
 135     facel            = _mm256_set1_ps(fr->epsfac);
 136     charge           = mdatoms->chargeA;
 137     nvdwtype         = fr->ntype;
 138     vdwparam         = fr->nbfp;
 139     vdwtype          = mdatoms->typeA;
 140
 141     vftab            = kernel_data->table_vdw->data;
 142     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 143
 144     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 145     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 146     beta2            = _mm256_mul_ps(beta,beta);
 147     beta3            = _mm256_mul_ps(beta,beta2);
 148
 149     ewtab            = fr->ic->tabq_coul_FDV0;
 150     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 151     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 152
 153     /* Setup water-specific parameters */
 154     inr              = nlist->iinr[0];
 155     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 156     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 157     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 158     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 159
 160     /* Avoid stupid compiler warnings */
 161     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 162     j_coord_offsetA = 0;
 163     j_coord_offsetB = 0;
 164     j_coord_offsetC = 0;
 165     j_coord_offsetD = 0;
 166     j_coord_offsetE = 0;
 167     j_coord_offsetF = 0;
 168     j_coord_offsetG = 0;
 169     j_coord_offsetH = 0;
 170
 171     outeriter        = 0;
 172     inneriter        = 0;
 173
 174     for(iidx=0;iidx<4*DIM;iidx++)
 175     {
 176         scratch[iidx] = 0.0;
 177     }
 178
 179     /* Start outer loop over neighborlists */
 180     for(iidx=0; iidx<nri; iidx++)
 181     {
 182         /* Load shift vector for this list */
 183         i_shift_offset   = DIM*shiftidx[iidx];
 184
 185         /* Load limits for loop over neighbors */
 186         j_index_start    = jindex[iidx];
 187         j_index_end      = jindex[iidx+1];
 188
 189         /* Get outer coordinate index */
 190         inr              = iinr[iidx];
 191         i_coord_offset   = DIM*inr;
 192
 193         /* Load i particle coords and add shift vector */
 194         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 195                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 196
 197         fix0             = _mm256_setzero_ps();
 198         fiy0             = _mm256_setzero_ps();
 199         fiz0             = _mm256_setzero_ps();
 200         fix1             = _mm256_setzero_ps();
 201         fiy1             = _mm256_setzero_ps();
 202         fiz1             = _mm256_setzero_ps();
 203         fix2             = _mm256_setzero_ps();
 204         fiy2             = _mm256_setzero_ps();
 205         fiz2             = _mm256_setzero_ps();
 206         fix3             = _mm256_setzero_ps();
 207         fiy3             = _mm256_setzero_ps();
 208         fiz3             = _mm256_setzero_ps();
 209
 210         /* Reset potential sums */
 211         velecsum         = _mm256_setzero_ps();
 212         vvdwsum          = _mm256_setzero_ps();
 213
 214         /* Start inner kernel loop */
 215         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 216         {
 217
 218             /* Get j neighbor index, and coordinate index */
 219             jnrA             = jjnr[jidx];
 220             jnrB             = jjnr[jidx+1];
 221             jnrC             = jjnr[jidx+2];
 222             jnrD             = jjnr[jidx+3];
 223             jnrE             = jjnr[jidx+4];
 224             jnrF             = jjnr[jidx+5];
 225             jnrG             = jjnr[jidx+6];
 226             jnrH             = jjnr[jidx+7];
 227             j_coord_offsetA  = DIM*jnrA;
 228             j_coord_offsetB  = DIM*jnrB;
 229             j_coord_offsetC  = DIM*jnrC;
 230             j_coord_offsetD  = DIM*jnrD;
 231             j_coord_offsetE  = DIM*jnrE;
 232             j_coord_offsetF  = DIM*jnrF;
 233             j_coord_offsetG  = DIM*jnrG;
 234             j_coord_offsetH  = DIM*jnrH;
 235
 236             /* load j atom coordinates */
 237             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 238                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 239                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 240                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 241                                                  &jx0,&jy0,&jz0);
 242
 243             /* Calculate displacement vector */
 244             dx00             = _mm256_sub_ps(ix0,jx0);
 245             dy00             = _mm256_sub_ps(iy0,jy0);
 246             dz00             = _mm256_sub_ps(iz0,jz0);
 247             dx10             = _mm256_sub_ps(ix1,jx0);
 248             dy10             = _mm256_sub_ps(iy1,jy0);
 249             dz10             = _mm256_sub_ps(iz1,jz0);
 250             dx20             = _mm256_sub_ps(ix2,jx0);
 251             dy20             = _mm256_sub_ps(iy2,jy0);
 252             dz20             = _mm256_sub_ps(iz2,jz0);
 253             dx30             = _mm256_sub_ps(ix3,jx0);
 254             dy30             = _mm256_sub_ps(iy3,jy0);
 255             dz30             = _mm256_sub_ps(iz3,jz0);
 256
 257             /* Calculate squared distance and things based on it */
 258             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 259             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 260             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 261             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 262
 263             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 264             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 265             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 266             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 267
 268             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 269             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 270             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 271
 272             /* Load parameters for j particles */
 273             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 274                                                                  charge+jnrC+0,charge+jnrD+0,
 275                                                                  charge+jnrE+0,charge+jnrF+0,
 276                                                                  charge+jnrG+0,charge+jnrH+0);
 277             vdwjidx0A        = 2*vdwtype[jnrA+0];
 278             vdwjidx0B        = 2*vdwtype[jnrB+0];
 279             vdwjidx0C        = 2*vdwtype[jnrC+0];
 280             vdwjidx0D        = 2*vdwtype[jnrD+0];
 281             vdwjidx0E        = 2*vdwtype[jnrE+0];
 282             vdwjidx0F        = 2*vdwtype[jnrF+0];
 283             vdwjidx0G        = 2*vdwtype[jnrG+0];
 284             vdwjidx0H        = 2*vdwtype[jnrH+0];
 285
 286             fjx0             = _mm256_setzero_ps();
 287             fjy0             = _mm256_setzero_ps();
 288             fjz0             = _mm256_setzero_ps();
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             r00              = _mm256_mul_ps(rsq00,rinv00);
 295
 296             /* Compute parameters for interactions between i and j atoms */
 297             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 298                                             vdwioffsetptr0+vdwjidx0B,
 299                                             vdwioffsetptr0+vdwjidx0C,
 300                                             vdwioffsetptr0+vdwjidx0D,
 301                                             vdwioffsetptr0+vdwjidx0E,
 302                                             vdwioffsetptr0+vdwjidx0F,
 303                                             vdwioffsetptr0+vdwjidx0G,
 304                                             vdwioffsetptr0+vdwjidx0H,
 305                                             &c6_00,&c12_00);
 306
 307             /* Calculate table index by multiplying r with table scale and truncate to integer */
 308             rt               = _mm256_mul_ps(r00,vftabscale);
 309             vfitab           = _mm256_cvttps_epi32(rt);
 310             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 311             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 312             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 313             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 314             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 315             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 316
 317             /* CUBIC SPLINE TABLE DISPERSION */
 318             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 319                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 320             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 321                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 322             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 323                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 324             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 325                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 326             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 327             Heps             = _mm256_mul_ps(vfeps,H);
 328             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 329             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 330             vvdw6            = _mm256_mul_ps(c6_00,VV);
 331             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 332             fvdw6            = _mm256_mul_ps(c6_00,FF);
 333
 334             /* CUBIC SPLINE TABLE REPULSION */
 335             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 336             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 337             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 338                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 339             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 340                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 341             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 342                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 343             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 344                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 345             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 346             Heps             = _mm256_mul_ps(vfeps,H);
 347             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 348             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 349             vvdw12           = _mm256_mul_ps(c12_00,VV);
 350             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 351             fvdw12           = _mm256_mul_ps(c12_00,FF);
 352             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 353             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 357
 358             fscal            = fvdw;
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm256_mul_ps(fscal,dx00);
 362             ty               = _mm256_mul_ps(fscal,dy00);
 363             tz               = _mm256_mul_ps(fscal,dz00);
 364
 365             /* Update vectorial force */
 366             fix0             = _mm256_add_ps(fix0,tx);
 367             fiy0             = _mm256_add_ps(fiy0,ty);
 368             fiz0             = _mm256_add_ps(fiz0,tz);
 369
 370             fjx0             = _mm256_add_ps(fjx0,tx);
 371             fjy0             = _mm256_add_ps(fjy0,ty);
 372             fjz0             = _mm256_add_ps(fjz0,tz);
 373
 374             /**************************
 375              * CALCULATE INTERACTIONS *
 376              **************************/
 377
 378             r10              = _mm256_mul_ps(rsq10,rinv10);
 379
 380             /* Compute parameters for interactions between i and j atoms */
 381             qq10             = _mm256_mul_ps(iq1,jq0);
 382
 383             /* EWALD ELECTROSTATICS */
 384
 385             /* Analytical PME correction */
 386             zeta2            = _mm256_mul_ps(beta2,rsq10);
 387             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 388             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 389             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 390             felec            = _mm256_mul_ps(qq10,felec);
 391             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 392             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 393             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 394             velec            = _mm256_mul_ps(qq10,velec);
 395
 396             /* Update potential sum for this i atom from the interaction with this j atom. */
 397             velecsum         = _mm256_add_ps(velecsum,velec);
 398
 399             fscal            = felec;
 400
 401             /* Calculate temporary vectorial force */
 402             tx               = _mm256_mul_ps(fscal,dx10);
 403             ty               = _mm256_mul_ps(fscal,dy10);
 404             tz               = _mm256_mul_ps(fscal,dz10);
 405
 406             /* Update vectorial force */
 407             fix1             = _mm256_add_ps(fix1,tx);
 408             fiy1             = _mm256_add_ps(fiy1,ty);
 409             fiz1             = _mm256_add_ps(fiz1,tz);
 410
 411             fjx0             = _mm256_add_ps(fjx0,tx);
 412             fjy0             = _mm256_add_ps(fjy0,ty);
 413             fjz0             = _mm256_add_ps(fjz0,tz);
 414
 415             /**************************
 416              * CALCULATE INTERACTIONS *
 417              **************************/
 418
 419             r20              = _mm256_mul_ps(rsq20,rinv20);
 420
 421             /* Compute parameters for interactions between i and j atoms */
 422             qq20             = _mm256_mul_ps(iq2,jq0);
 423
 424             /* EWALD ELECTROSTATICS */
 425
 426             /* Analytical PME correction */
 427             zeta2            = _mm256_mul_ps(beta2,rsq20);
 428             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 429             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 430             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 431             felec            = _mm256_mul_ps(qq20,felec);
 432             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 433             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 434             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 435             velec            = _mm256_mul_ps(qq20,velec);
 436
 437             /* Update potential sum for this i atom from the interaction with this j atom. */
 438             velecsum         = _mm256_add_ps(velecsum,velec);
 439
 440             fscal            = felec;
 441
 442             /* Calculate temporary vectorial force */
 443             tx               = _mm256_mul_ps(fscal,dx20);
 444             ty               = _mm256_mul_ps(fscal,dy20);
 445             tz               = _mm256_mul_ps(fscal,dz20);
 446
 447             /* Update vectorial force */
 448             fix2             = _mm256_add_ps(fix2,tx);
 449             fiy2             = _mm256_add_ps(fiy2,ty);
 450             fiz2             = _mm256_add_ps(fiz2,tz);
 451
 452             fjx0             = _mm256_add_ps(fjx0,tx);
 453             fjy0             = _mm256_add_ps(fjy0,ty);
 454             fjz0             = _mm256_add_ps(fjz0,tz);
 455
 456             /**************************
 457              * CALCULATE INTERACTIONS *
 458              **************************/
 459
 460             r30              = _mm256_mul_ps(rsq30,rinv30);
 461
 462             /* Compute parameters for interactions between i and j atoms */
 463             qq30             = _mm256_mul_ps(iq3,jq0);
 464
 465             /* EWALD ELECTROSTATICS */
 466
 467             /* Analytical PME correction */
 468             zeta2            = _mm256_mul_ps(beta2,rsq30);
 469             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 470             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 471             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 472             felec            = _mm256_mul_ps(qq30,felec);
 473             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 474             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 475             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 476             velec            = _mm256_mul_ps(qq30,velec);
 477
 478             /* Update potential sum for this i atom from the interaction with this j atom. */
 479             velecsum         = _mm256_add_ps(velecsum,velec);
 480
 481             fscal            = felec;
 482
 483             /* Calculate temporary vectorial force */
 484             tx               = _mm256_mul_ps(fscal,dx30);
 485             ty               = _mm256_mul_ps(fscal,dy30);
 486             tz               = _mm256_mul_ps(fscal,dz30);
 487
 488             /* Update vectorial force */
 489             fix3             = _mm256_add_ps(fix3,tx);
 490             fiy3             = _mm256_add_ps(fiy3,ty);
 491             fiz3             = _mm256_add_ps(fiz3,tz);
 492
 493             fjx0             = _mm256_add_ps(fjx0,tx);
 494             fjy0             = _mm256_add_ps(fjy0,ty);
 495             fjz0             = _mm256_add_ps(fjz0,tz);
 496
 497             fjptrA             = f+j_coord_offsetA;
 498             fjptrB             = f+j_coord_offsetB;
 499             fjptrC             = f+j_coord_offsetC;
 500             fjptrD             = f+j_coord_offsetD;
 501             fjptrE             = f+j_coord_offsetE;
 502             fjptrF             = f+j_coord_offsetF;
 503             fjptrG             = f+j_coord_offsetG;
 504             fjptrH             = f+j_coord_offsetH;
 505
 506             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 507
 508             /* Inner loop uses 311 flops */
 509         }
 510
 511         if(jidx<j_index_end)
 512         {
 513
 514             /* Get j neighbor index, and coordinate index */
 515             jnrlistA         = jjnr[jidx];
 516             jnrlistB         = jjnr[jidx+1];
 517             jnrlistC         = jjnr[jidx+2];
 518             jnrlistD         = jjnr[jidx+3];
 519             jnrlistE         = jjnr[jidx+4];
 520             jnrlistF         = jjnr[jidx+5];
 521             jnrlistG         = jjnr[jidx+6];
 522             jnrlistH         = jjnr[jidx+7];
 523             /* Sign of each element will be negative for non-real atoms.
 524              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 525              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 526              */
 527             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 528                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 529
 530             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 531             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 532             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 533             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 534             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 535             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 536             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 537             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 538             j_coord_offsetA  = DIM*jnrA;
 539             j_coord_offsetB  = DIM*jnrB;
 540             j_coord_offsetC  = DIM*jnrC;
 541             j_coord_offsetD  = DIM*jnrD;
 542             j_coord_offsetE  = DIM*jnrE;
 543             j_coord_offsetF  = DIM*jnrF;
 544             j_coord_offsetG  = DIM*jnrG;
 545             j_coord_offsetH  = DIM*jnrH;
 546
 547             /* load j atom coordinates */
 548             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 549                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 550                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 551                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 552                                                  &jx0,&jy0,&jz0);
 553
 554             /* Calculate displacement vector */
 555             dx00             = _mm256_sub_ps(ix0,jx0);
 556             dy00             = _mm256_sub_ps(iy0,jy0);
 557             dz00             = _mm256_sub_ps(iz0,jz0);
 558             dx10             = _mm256_sub_ps(ix1,jx0);
 559             dy10             = _mm256_sub_ps(iy1,jy0);
 560             dz10             = _mm256_sub_ps(iz1,jz0);
 561             dx20             = _mm256_sub_ps(ix2,jx0);
 562             dy20             = _mm256_sub_ps(iy2,jy0);
 563             dz20             = _mm256_sub_ps(iz2,jz0);
 564             dx30             = _mm256_sub_ps(ix3,jx0);
 565             dy30             = _mm256_sub_ps(iy3,jy0);
 566             dz30             = _mm256_sub_ps(iz3,jz0);
 567
 568             /* Calculate squared distance and things based on it */
 569             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 570             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 571             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 572             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 573
 574             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 575             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 576             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 577             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 578
 579             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 580             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 581             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 582
 583             /* Load parameters for j particles */
 584             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 585                                                                  charge+jnrC+0,charge+jnrD+0,
 586                                                                  charge+jnrE+0,charge+jnrF+0,
 587                                                                  charge+jnrG+0,charge+jnrH+0);
 588             vdwjidx0A        = 2*vdwtype[jnrA+0];
 589             vdwjidx0B        = 2*vdwtype[jnrB+0];
 590             vdwjidx0C        = 2*vdwtype[jnrC+0];
 591             vdwjidx0D        = 2*vdwtype[jnrD+0];
 592             vdwjidx0E        = 2*vdwtype[jnrE+0];
 593             vdwjidx0F        = 2*vdwtype[jnrF+0];
 594             vdwjidx0G        = 2*vdwtype[jnrG+0];
 595             vdwjidx0H        = 2*vdwtype[jnrH+0];
 596
 597             fjx0             = _mm256_setzero_ps();
 598             fjy0             = _mm256_setzero_ps();
 599             fjz0             = _mm256_setzero_ps();
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             r00              = _mm256_mul_ps(rsq00,rinv00);
 606             r00              = _mm256_andnot_ps(dummy_mask,r00);
 607
 608             /* Compute parameters for interactions between i and j atoms */
 609             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 610                                             vdwioffsetptr0+vdwjidx0B,
 611                                             vdwioffsetptr0+vdwjidx0C,
 612                                             vdwioffsetptr0+vdwjidx0D,
 613                                             vdwioffsetptr0+vdwjidx0E,
 614                                             vdwioffsetptr0+vdwjidx0F,
 615                                             vdwioffsetptr0+vdwjidx0G,
 616                                             vdwioffsetptr0+vdwjidx0H,
 617                                             &c6_00,&c12_00);
 618
 619             /* Calculate table index by multiplying r with table scale and truncate to integer */
 620             rt               = _mm256_mul_ps(r00,vftabscale);
 621             vfitab           = _mm256_cvttps_epi32(rt);
 622             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 623             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 624             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 625             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 626             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 627             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 628
 629             /* CUBIC SPLINE TABLE DISPERSION */
 630             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 631                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 632             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 633                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 634             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 635                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 636             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 637                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 638             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 639             Heps             = _mm256_mul_ps(vfeps,H);
 640             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 641             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 642             vvdw6            = _mm256_mul_ps(c6_00,VV);
 643             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 644             fvdw6            = _mm256_mul_ps(c6_00,FF);
 645
 646             /* CUBIC SPLINE TABLE REPULSION */
 647             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 648             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 649             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 650                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 651             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 652                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 653             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 654                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 655             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 656                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 657             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 658             Heps             = _mm256_mul_ps(vfeps,H);
 659             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 660             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 661             vvdw12           = _mm256_mul_ps(c12_00,VV);
 662             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 663             fvdw12           = _mm256_mul_ps(c12_00,FF);
 664             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 665             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 666
 667             /* Update potential sum for this i atom from the interaction with this j atom. */
 668             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 669             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 670
 671             fscal            = fvdw;
 672
 673             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 674
 675             /* Calculate temporary vectorial force */
 676             tx               = _mm256_mul_ps(fscal,dx00);
 677             ty               = _mm256_mul_ps(fscal,dy00);
 678             tz               = _mm256_mul_ps(fscal,dz00);
 679
 680             /* Update vectorial force */
 681             fix0             = _mm256_add_ps(fix0,tx);
 682             fiy0             = _mm256_add_ps(fiy0,ty);
 683             fiz0             = _mm256_add_ps(fiz0,tz);
 684
 685             fjx0             = _mm256_add_ps(fjx0,tx);
 686             fjy0             = _mm256_add_ps(fjy0,ty);
 687             fjz0             = _mm256_add_ps(fjz0,tz);
 688
 689             /**************************
 690              * CALCULATE INTERACTIONS *
 691              **************************/
 692
 693             r10              = _mm256_mul_ps(rsq10,rinv10);
 694             r10              = _mm256_andnot_ps(dummy_mask,r10);
 695
 696             /* Compute parameters for interactions between i and j atoms */
 697             qq10             = _mm256_mul_ps(iq1,jq0);
 698
 699             /* EWALD ELECTROSTATICS */
 700
 701             /* Analytical PME correction */
 702             zeta2            = _mm256_mul_ps(beta2,rsq10);
 703             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 704             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 705             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 706             felec            = _mm256_mul_ps(qq10,felec);
 707             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 708             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 709             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 710             velec            = _mm256_mul_ps(qq10,velec);
 711
 712             /* Update potential sum for this i atom from the interaction with this j atom. */
 713             velec            = _mm256_andnot_ps(dummy_mask,velec);
 714             velecsum         = _mm256_add_ps(velecsum,velec);
 715
 716             fscal            = felec;
 717
 718             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 719
 720             /* Calculate temporary vectorial force */
 721             tx               = _mm256_mul_ps(fscal,dx10);
 722             ty               = _mm256_mul_ps(fscal,dy10);
 723             tz               = _mm256_mul_ps(fscal,dz10);
 724
 725             /* Update vectorial force */
 726             fix1             = _mm256_add_ps(fix1,tx);
 727             fiy1             = _mm256_add_ps(fiy1,ty);
 728             fiz1             = _mm256_add_ps(fiz1,tz);
 729
 730             fjx0             = _mm256_add_ps(fjx0,tx);
 731             fjy0             = _mm256_add_ps(fjy0,ty);
 732             fjz0             = _mm256_add_ps(fjz0,tz);
 733
 734             /**************************
 735              * CALCULATE INTERACTIONS *
 736              **************************/
 737
 738             r20              = _mm256_mul_ps(rsq20,rinv20);
 739             r20              = _mm256_andnot_ps(dummy_mask,r20);
 740
 741             /* Compute parameters for interactions between i and j atoms */
 742             qq20             = _mm256_mul_ps(iq2,jq0);
 743
 744             /* EWALD ELECTROSTATICS */
 745
 746             /* Analytical PME correction */
 747             zeta2            = _mm256_mul_ps(beta2,rsq20);
 748             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 749             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 750             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 751             felec            = _mm256_mul_ps(qq20,felec);
 752             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 753             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 754             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 755             velec            = _mm256_mul_ps(qq20,velec);
 756
 757             /* Update potential sum for this i atom from the interaction with this j atom. */
 758             velec            = _mm256_andnot_ps(dummy_mask,velec);
 759             velecsum         = _mm256_add_ps(velecsum,velec);
 760
 761             fscal            = felec;
 762
 763             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 764
 765             /* Calculate temporary vectorial force */
 766             tx               = _mm256_mul_ps(fscal,dx20);
 767             ty               = _mm256_mul_ps(fscal,dy20);
 768             tz               = _mm256_mul_ps(fscal,dz20);
 769
 770             /* Update vectorial force */
 771             fix2             = _mm256_add_ps(fix2,tx);
 772             fiy2             = _mm256_add_ps(fiy2,ty);
 773             fiz2             = _mm256_add_ps(fiz2,tz);
 774
 775             fjx0             = _mm256_add_ps(fjx0,tx);
 776             fjy0             = _mm256_add_ps(fjy0,ty);
 777             fjz0             = _mm256_add_ps(fjz0,tz);
 778
 779             /**************************
 780              * CALCULATE INTERACTIONS *
 781              **************************/
 782
 783             r30              = _mm256_mul_ps(rsq30,rinv30);
 784             r30              = _mm256_andnot_ps(dummy_mask,r30);
 785
 786             /* Compute parameters for interactions between i and j atoms */
 787             qq30             = _mm256_mul_ps(iq3,jq0);
 788
 789             /* EWALD ELECTROSTATICS */
 790
 791             /* Analytical PME correction */
 792             zeta2            = _mm256_mul_ps(beta2,rsq30);
 793             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 794             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 795             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 796             felec            = _mm256_mul_ps(qq30,felec);
 797             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 798             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 799             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 800             velec            = _mm256_mul_ps(qq30,velec);
 801
 802             /* Update potential sum for this i atom from the interaction with this j atom. */
 803             velec            = _mm256_andnot_ps(dummy_mask,velec);
 804             velecsum         = _mm256_add_ps(velecsum,velec);
 805
 806             fscal            = felec;
 807
 808             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 809
 810             /* Calculate temporary vectorial force */
 811             tx               = _mm256_mul_ps(fscal,dx30);
 812             ty               = _mm256_mul_ps(fscal,dy30);
 813             tz               = _mm256_mul_ps(fscal,dz30);
 814
 815             /* Update vectorial force */
 816             fix3             = _mm256_add_ps(fix3,tx);
 817             fiy3             = _mm256_add_ps(fiy3,ty);
 818             fiz3             = _mm256_add_ps(fiz3,tz);
 819
 820             fjx0             = _mm256_add_ps(fjx0,tx);
 821             fjy0             = _mm256_add_ps(fjy0,ty);
 822             fjz0             = _mm256_add_ps(fjz0,tz);
 823
 824             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 825             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 826             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 827             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 828             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 829             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 830             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 831             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 832
 833             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 834
 835             /* Inner loop uses 315 flops */
 836         }
 837
 838         /* End of innermost loop */
 839
 840         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 841                                                  f+i_coord_offset,fshift+i_shift_offset);
 842
 843         ggid                        = gid[iidx];
 844         /* Update potential energies */
 845         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 846         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 847
 848         /* Increment number of inner iterations */
 849         inneriter                  += j_index_end - j_index_start;
 850
 851         /* Outer loop uses 26 flops */
 852     }
 853
 854     /* Increment number of outer iterations */
 855     outeriter        += nri;
 856
 857     /* Update outer/inner flops */
 858
 859     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*315);
 860 }
 861 /*
 862  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_256_single
 863  * Electrostatics interaction: Ewald
 864  * VdW interaction:            CubicSplineTable
 865  * Geometry:                   Water4-Particle
 866  * Calculate force/pot:        Force
 867  */
 868 void
 869 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_256_single
 870                     (t_nblist                    * gmx_restrict       nlist,
 871                      rvec                        * gmx_restrict          xx,
 872                      rvec                        * gmx_restrict          ff,
 873                      t_forcerec                  * gmx_restrict          fr,
 874                      t_mdatoms                   * gmx_restrict     mdatoms,
 875                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 876                      t_nrnb                      * gmx_restrict        nrnb)
 877 {
 878     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 879      * just 0 for non-waters.
 880      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 881      * jnr indices corresponding to data put in the four positions in the SIMD register.
 882      */
 883     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 884     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 885     int              jnrA,jnrB,jnrC,jnrD;
 886     int              jnrE,jnrF,jnrG,jnrH;
 887     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 888     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 889     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 890     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 891     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 892     real             rcutoff_scalar;
 893     real             *shiftvec,*fshift,*x,*f;
 894     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 895     real             scratch[4*DIM];
 896     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 897     real *           vdwioffsetptr0;
 898     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 899     real *           vdwioffsetptr1;
 900     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 901     real *           vdwioffsetptr2;
 902     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 903     real *           vdwioffsetptr3;
 904     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 905     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 906     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 907     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 908     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 909     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 910     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 911     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 912     real             *charge;
 913     int              nvdwtype;
 914     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 915     int              *vdwtype;
 916     real             *vdwparam;
 917     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 918     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 919     __m256i          vfitab;
 920     __m128i          vfitab_lo,vfitab_hi;
 921     __m128i          ifour       = _mm_set1_epi32(4);
 922     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 923     real             *vftab;
 924     __m256i          ewitab;
 925     __m128i          ewitab_lo,ewitab_hi;
 926     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 927     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 928     real             *ewtab;
 929     __m256           dummy_mask,cutoff_mask;
 930     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 931     __m256           one     = _mm256_set1_ps(1.0);
 932     __m256           two     = _mm256_set1_ps(2.0);
 933     x                = xx[0];
 934     f                = ff[0];
 935
 936     nri              = nlist->nri;
 937     iinr             = nlist->iinr;
 938     jindex           = nlist->jindex;
 939     jjnr             = nlist->jjnr;
 940     shiftidx         = nlist->shift;
 941     gid              = nlist->gid;
 942     shiftvec         = fr->shift_vec[0];
 943     fshift           = fr->fshift[0];
 944     facel            = _mm256_set1_ps(fr->epsfac);
 945     charge           = mdatoms->chargeA;
 946     nvdwtype         = fr->ntype;
 947     vdwparam         = fr->nbfp;
 948     vdwtype          = mdatoms->typeA;
 949
 950     vftab            = kernel_data->table_vdw->data;
 951     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 952
 953     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 954     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 955     beta2            = _mm256_mul_ps(beta,beta);
 956     beta3            = _mm256_mul_ps(beta,beta2);
 957
 958     ewtab            = fr->ic->tabq_coul_F;
 959     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 960     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 961
 962     /* Setup water-specific parameters */
 963     inr              = nlist->iinr[0];
 964     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 965     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 966     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 967     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 968
 969     /* Avoid stupid compiler warnings */
 970     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 971     j_coord_offsetA = 0;
 972     j_coord_offsetB = 0;
 973     j_coord_offsetC = 0;
 974     j_coord_offsetD = 0;
 975     j_coord_offsetE = 0;
 976     j_coord_offsetF = 0;
 977     j_coord_offsetG = 0;
 978     j_coord_offsetH = 0;
 979
 980     outeriter        = 0;
 981     inneriter        = 0;
 982
 983     for(iidx=0;iidx<4*DIM;iidx++)
 984     {
 985         scratch[iidx] = 0.0;
 986     }
 987
 988     /* Start outer loop over neighborlists */
 989     for(iidx=0; iidx<nri; iidx++)
 990     {
 991         /* Load shift vector for this list */
 992         i_shift_offset   = DIM*shiftidx[iidx];
 993
 994         /* Load limits for loop over neighbors */
 995         j_index_start    = jindex[iidx];
 996         j_index_end      = jindex[iidx+1];
 997
 998         /* Get outer coordinate index */
 999         inr              = iinr[iidx];
1000         i_coord_offset   = DIM*inr;
1001
1002         /* Load i particle coords and add shift vector */
1003         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1004                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1005
1006         fix0             = _mm256_setzero_ps();
1007         fiy0             = _mm256_setzero_ps();
1008         fiz0             = _mm256_setzero_ps();
1009         fix1             = _mm256_setzero_ps();
1010         fiy1             = _mm256_setzero_ps();
1011         fiz1             = _mm256_setzero_ps();
1012         fix2             = _mm256_setzero_ps();
1013         fiy2             = _mm256_setzero_ps();
1014         fiz2             = _mm256_setzero_ps();
1015         fix3             = _mm256_setzero_ps();
1016         fiy3             = _mm256_setzero_ps();
1017         fiz3             = _mm256_setzero_ps();
1018
1019         /* Start inner kernel loop */
1020         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1021         {
1022
1023             /* Get j neighbor index, and coordinate index */
1024             jnrA             = jjnr[jidx];
1025             jnrB             = jjnr[jidx+1];
1026             jnrC             = jjnr[jidx+2];
1027             jnrD             = jjnr[jidx+3];
1028             jnrE             = jjnr[jidx+4];
1029             jnrF             = jjnr[jidx+5];
1030             jnrG             = jjnr[jidx+6];
1031             jnrH             = jjnr[jidx+7];
1032             j_coord_offsetA  = DIM*jnrA;
1033             j_coord_offsetB  = DIM*jnrB;
1034             j_coord_offsetC  = DIM*jnrC;
1035             j_coord_offsetD  = DIM*jnrD;
1036             j_coord_offsetE  = DIM*jnrE;
1037             j_coord_offsetF  = DIM*jnrF;
1038             j_coord_offsetG  = DIM*jnrG;
1039             j_coord_offsetH  = DIM*jnrH;
1040
1041             /* load j atom coordinates */
1042             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1043                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1044                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1045                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1046                                                  &jx0,&jy0,&jz0);
1047
1048             /* Calculate displacement vector */
1049             dx00             = _mm256_sub_ps(ix0,jx0);
1050             dy00             = _mm256_sub_ps(iy0,jy0);
1051             dz00             = _mm256_sub_ps(iz0,jz0);
1052             dx10             = _mm256_sub_ps(ix1,jx0);
1053             dy10             = _mm256_sub_ps(iy1,jy0);
1054             dz10             = _mm256_sub_ps(iz1,jz0);
1055             dx20             = _mm256_sub_ps(ix2,jx0);
1056             dy20             = _mm256_sub_ps(iy2,jy0);
1057             dz20             = _mm256_sub_ps(iz2,jz0);
1058             dx30             = _mm256_sub_ps(ix3,jx0);
1059             dy30             = _mm256_sub_ps(iy3,jy0);
1060             dz30             = _mm256_sub_ps(iz3,jz0);
1061
1062             /* Calculate squared distance and things based on it */
1063             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1064             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1065             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1066             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1067
1068             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1069             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1070             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1071             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1072
1073             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1074             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1075             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1076
1077             /* Load parameters for j particles */
1078             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1079                                                                  charge+jnrC+0,charge+jnrD+0,
1080                                                                  charge+jnrE+0,charge+jnrF+0,
1081                                                                  charge+jnrG+0,charge+jnrH+0);
1082             vdwjidx0A        = 2*vdwtype[jnrA+0];
1083             vdwjidx0B        = 2*vdwtype[jnrB+0];
1084             vdwjidx0C        = 2*vdwtype[jnrC+0];
1085             vdwjidx0D        = 2*vdwtype[jnrD+0];
1086             vdwjidx0E        = 2*vdwtype[jnrE+0];
1087             vdwjidx0F        = 2*vdwtype[jnrF+0];
1088             vdwjidx0G        = 2*vdwtype[jnrG+0];
1089             vdwjidx0H        = 2*vdwtype[jnrH+0];
1090
1091             fjx0             = _mm256_setzero_ps();
1092             fjy0             = _mm256_setzero_ps();
1093             fjz0             = _mm256_setzero_ps();
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             r00              = _mm256_mul_ps(rsq00,rinv00);
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1103                                             vdwioffsetptr0+vdwjidx0B,
1104                                             vdwioffsetptr0+vdwjidx0C,
1105                                             vdwioffsetptr0+vdwjidx0D,
1106                                             vdwioffsetptr0+vdwjidx0E,
1107                                             vdwioffsetptr0+vdwjidx0F,
1108                                             vdwioffsetptr0+vdwjidx0G,
1109                                             vdwioffsetptr0+vdwjidx0H,
1110                                             &c6_00,&c12_00);
1111
1112             /* Calculate table index by multiplying r with table scale and truncate to integer */
1113             rt               = _mm256_mul_ps(r00,vftabscale);
1114             vfitab           = _mm256_cvttps_epi32(rt);
1115             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1116             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1117             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1118             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1119             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1120             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1121
1122             /* CUBIC SPLINE TABLE DISPERSION */
1123             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1124                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1125             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1126                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1127             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1128                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1129             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1130                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1131             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1132             Heps             = _mm256_mul_ps(vfeps,H);
1133             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1134             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1135             fvdw6            = _mm256_mul_ps(c6_00,FF);
1136
1137             /* CUBIC SPLINE TABLE REPULSION */
1138             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1139             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1140             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1141                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1142             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1143                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1144             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1145                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1146             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1147                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1148             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1149             Heps             = _mm256_mul_ps(vfeps,H);
1150             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1151             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1152             fvdw12           = _mm256_mul_ps(c12_00,FF);
1153             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1154
1155             fscal            = fvdw;
1156
1157             /* Calculate temporary vectorial force */
1158             tx               = _mm256_mul_ps(fscal,dx00);
1159             ty               = _mm256_mul_ps(fscal,dy00);
1160             tz               = _mm256_mul_ps(fscal,dz00);
1161
1162             /* Update vectorial force */
1163             fix0             = _mm256_add_ps(fix0,tx);
1164             fiy0             = _mm256_add_ps(fiy0,ty);
1165             fiz0             = _mm256_add_ps(fiz0,tz);
1166
1167             fjx0             = _mm256_add_ps(fjx0,tx);
1168             fjy0             = _mm256_add_ps(fjy0,ty);
1169             fjz0             = _mm256_add_ps(fjz0,tz);
1170
1171             /**************************
1172              * CALCULATE INTERACTIONS *
1173              **************************/
1174
1175             r10              = _mm256_mul_ps(rsq10,rinv10);
1176
1177             /* Compute parameters for interactions between i and j atoms */
1178             qq10             = _mm256_mul_ps(iq1,jq0);
1179
1180             /* EWALD ELECTROSTATICS */
1181
1182             /* Analytical PME correction */
1183             zeta2            = _mm256_mul_ps(beta2,rsq10);
1184             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1185             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1186             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1187             felec            = _mm256_mul_ps(qq10,felec);
1188
1189             fscal            = felec;
1190
1191             /* Calculate temporary vectorial force */
1192             tx               = _mm256_mul_ps(fscal,dx10);
1193             ty               = _mm256_mul_ps(fscal,dy10);
1194             tz               = _mm256_mul_ps(fscal,dz10);
1195
1196             /* Update vectorial force */
1197             fix1             = _mm256_add_ps(fix1,tx);
1198             fiy1             = _mm256_add_ps(fiy1,ty);
1199             fiz1             = _mm256_add_ps(fiz1,tz);
1200
1201             fjx0             = _mm256_add_ps(fjx0,tx);
1202             fjy0             = _mm256_add_ps(fjy0,ty);
1203             fjz0             = _mm256_add_ps(fjz0,tz);
1204
1205             /**************************
1206              * CALCULATE INTERACTIONS *
1207              **************************/
1208
1209             r20              = _mm256_mul_ps(rsq20,rinv20);
1210
1211             /* Compute parameters for interactions between i and j atoms */
1212             qq20             = _mm256_mul_ps(iq2,jq0);
1213
1214             /* EWALD ELECTROSTATICS */
1215
1216             /* Analytical PME correction */
1217             zeta2            = _mm256_mul_ps(beta2,rsq20);
1218             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1219             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1220             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1221             felec            = _mm256_mul_ps(qq20,felec);
1222
1223             fscal            = felec;
1224
1225             /* Calculate temporary vectorial force */
1226             tx               = _mm256_mul_ps(fscal,dx20);
1227             ty               = _mm256_mul_ps(fscal,dy20);
1228             tz               = _mm256_mul_ps(fscal,dz20);
1229
1230             /* Update vectorial force */
1231             fix2             = _mm256_add_ps(fix2,tx);
1232             fiy2             = _mm256_add_ps(fiy2,ty);
1233             fiz2             = _mm256_add_ps(fiz2,tz);
1234
1235             fjx0             = _mm256_add_ps(fjx0,tx);
1236             fjy0             = _mm256_add_ps(fjy0,ty);
1237             fjz0             = _mm256_add_ps(fjz0,tz);
1238
1239             /**************************
1240              * CALCULATE INTERACTIONS *
1241              **************************/
1242
1243             r30              = _mm256_mul_ps(rsq30,rinv30);
1244
1245             /* Compute parameters for interactions between i and j atoms */
1246             qq30             = _mm256_mul_ps(iq3,jq0);
1247
1248             /* EWALD ELECTROSTATICS */
1249
1250             /* Analytical PME correction */
1251             zeta2            = _mm256_mul_ps(beta2,rsq30);
1252             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1253             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1254             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1255             felec            = _mm256_mul_ps(qq30,felec);
1256
1257             fscal            = felec;
1258
1259             /* Calculate temporary vectorial force */
1260             tx               = _mm256_mul_ps(fscal,dx30);
1261             ty               = _mm256_mul_ps(fscal,dy30);
1262             tz               = _mm256_mul_ps(fscal,dz30);
1263
1264             /* Update vectorial force */
1265             fix3             = _mm256_add_ps(fix3,tx);
1266             fiy3             = _mm256_add_ps(fiy3,ty);
1267             fiz3             = _mm256_add_ps(fiz3,tz);
1268
1269             fjx0             = _mm256_add_ps(fjx0,tx);
1270             fjy0             = _mm256_add_ps(fjy0,ty);
1271             fjz0             = _mm256_add_ps(fjz0,tz);
1272
1273             fjptrA             = f+j_coord_offsetA;
1274             fjptrB             = f+j_coord_offsetB;
1275             fjptrC             = f+j_coord_offsetC;
1276             fjptrD             = f+j_coord_offsetD;
1277             fjptrE             = f+j_coord_offsetE;
1278             fjptrF             = f+j_coord_offsetF;
1279             fjptrG             = f+j_coord_offsetG;
1280             fjptrH             = f+j_coord_offsetH;
1281
1282             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1283
1284             /* Inner loop uses 219 flops */
1285         }
1286
1287         if(jidx<j_index_end)
1288         {
1289
1290             /* Get j neighbor index, and coordinate index */
1291             jnrlistA         = jjnr[jidx];
1292             jnrlistB         = jjnr[jidx+1];
1293             jnrlistC         = jjnr[jidx+2];
1294             jnrlistD         = jjnr[jidx+3];
1295             jnrlistE         = jjnr[jidx+4];
1296             jnrlistF         = jjnr[jidx+5];
1297             jnrlistG         = jjnr[jidx+6];
1298             jnrlistH         = jjnr[jidx+7];
1299             /* Sign of each element will be negative for non-real atoms.
1300              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1301              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1302              */
1303             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1304                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1305
1306             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1307             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1308             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1309             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1310             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1311             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1312             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1313             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1314             j_coord_offsetA  = DIM*jnrA;
1315             j_coord_offsetB  = DIM*jnrB;
1316             j_coord_offsetC  = DIM*jnrC;
1317             j_coord_offsetD  = DIM*jnrD;
1318             j_coord_offsetE  = DIM*jnrE;
1319             j_coord_offsetF  = DIM*jnrF;
1320             j_coord_offsetG  = DIM*jnrG;
1321             j_coord_offsetH  = DIM*jnrH;
1322
1323             /* load j atom coordinates */
1324             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1325                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1326                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1327                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1328                                                  &jx0,&jy0,&jz0);
1329
1330             /* Calculate displacement vector */
1331             dx00             = _mm256_sub_ps(ix0,jx0);
1332             dy00             = _mm256_sub_ps(iy0,jy0);
1333             dz00             = _mm256_sub_ps(iz0,jz0);
1334             dx10             = _mm256_sub_ps(ix1,jx0);
1335             dy10             = _mm256_sub_ps(iy1,jy0);
1336             dz10             = _mm256_sub_ps(iz1,jz0);
1337             dx20             = _mm256_sub_ps(ix2,jx0);
1338             dy20             = _mm256_sub_ps(iy2,jy0);
1339             dz20             = _mm256_sub_ps(iz2,jz0);
1340             dx30             = _mm256_sub_ps(ix3,jx0);
1341             dy30             = _mm256_sub_ps(iy3,jy0);
1342             dz30             = _mm256_sub_ps(iz3,jz0);
1343
1344             /* Calculate squared distance and things based on it */
1345             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1346             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1347             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1348             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1349
1350             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1351             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1352             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1353             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1354
1355             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1356             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1357             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1358
1359             /* Load parameters for j particles */
1360             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1361                                                                  charge+jnrC+0,charge+jnrD+0,
1362                                                                  charge+jnrE+0,charge+jnrF+0,
1363                                                                  charge+jnrG+0,charge+jnrH+0);
1364             vdwjidx0A        = 2*vdwtype[jnrA+0];
1365             vdwjidx0B        = 2*vdwtype[jnrB+0];
1366             vdwjidx0C        = 2*vdwtype[jnrC+0];
1367             vdwjidx0D        = 2*vdwtype[jnrD+0];
1368             vdwjidx0E        = 2*vdwtype[jnrE+0];
1369             vdwjidx0F        = 2*vdwtype[jnrF+0];
1370             vdwjidx0G        = 2*vdwtype[jnrG+0];
1371             vdwjidx0H        = 2*vdwtype[jnrH+0];
1372
1373             fjx0             = _mm256_setzero_ps();
1374             fjy0             = _mm256_setzero_ps();
1375             fjz0             = _mm256_setzero_ps();
1376
1377             /**************************
1378              * CALCULATE INTERACTIONS *
1379              **************************/
1380
1381             r00              = _mm256_mul_ps(rsq00,rinv00);
1382             r00              = _mm256_andnot_ps(dummy_mask,r00);
1383
1384             /* Compute parameters for interactions between i and j atoms */
1385             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1386                                             vdwioffsetptr0+vdwjidx0B,
1387                                             vdwioffsetptr0+vdwjidx0C,
1388                                             vdwioffsetptr0+vdwjidx0D,
1389                                             vdwioffsetptr0+vdwjidx0E,
1390                                             vdwioffsetptr0+vdwjidx0F,
1391                                             vdwioffsetptr0+vdwjidx0G,
1392                                             vdwioffsetptr0+vdwjidx0H,
1393                                             &c6_00,&c12_00);
1394
1395             /* Calculate table index by multiplying r with table scale and truncate to integer */
1396             rt               = _mm256_mul_ps(r00,vftabscale);
1397             vfitab           = _mm256_cvttps_epi32(rt);
1398             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1399             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1400             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1401             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1402             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1403             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1404
1405             /* CUBIC SPLINE TABLE DISPERSION */
1406             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1407                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1408             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1409                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1410             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1411                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1412             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1413                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1414             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1415             Heps             = _mm256_mul_ps(vfeps,H);
1416             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1417             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1418             fvdw6            = _mm256_mul_ps(c6_00,FF);
1419
1420             /* CUBIC SPLINE TABLE REPULSION */
1421             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1422             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1423             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1424                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1425             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1426                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1427             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1428                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1429             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1430                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1431             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1432             Heps             = _mm256_mul_ps(vfeps,H);
1433             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1434             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1435             fvdw12           = _mm256_mul_ps(c12_00,FF);
1436             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1437
1438             fscal            = fvdw;
1439
1440             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1441
1442             /* Calculate temporary vectorial force */
1443             tx               = _mm256_mul_ps(fscal,dx00);
1444             ty               = _mm256_mul_ps(fscal,dy00);
1445             tz               = _mm256_mul_ps(fscal,dz00);
1446
1447             /* Update vectorial force */
1448             fix0             = _mm256_add_ps(fix0,tx);
1449             fiy0             = _mm256_add_ps(fiy0,ty);
1450             fiz0             = _mm256_add_ps(fiz0,tz);
1451
1452             fjx0             = _mm256_add_ps(fjx0,tx);
1453             fjy0             = _mm256_add_ps(fjy0,ty);
1454             fjz0             = _mm256_add_ps(fjz0,tz);
1455
1456             /**************************
1457              * CALCULATE INTERACTIONS *
1458              **************************/
1459
1460             r10              = _mm256_mul_ps(rsq10,rinv10);
1461             r10              = _mm256_andnot_ps(dummy_mask,r10);
1462
1463             /* Compute parameters for interactions between i and j atoms */
1464             qq10             = _mm256_mul_ps(iq1,jq0);
1465
1466             /* EWALD ELECTROSTATICS */
1467
1468             /* Analytical PME correction */
1469             zeta2            = _mm256_mul_ps(beta2,rsq10);
1470             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1471             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1472             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1473             felec            = _mm256_mul_ps(qq10,felec);
1474
1475             fscal            = felec;
1476
1477             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1478
1479             /* Calculate temporary vectorial force */
1480             tx               = _mm256_mul_ps(fscal,dx10);
1481             ty               = _mm256_mul_ps(fscal,dy10);
1482             tz               = _mm256_mul_ps(fscal,dz10);
1483
1484             /* Update vectorial force */
1485             fix1             = _mm256_add_ps(fix1,tx);
1486             fiy1             = _mm256_add_ps(fiy1,ty);
1487             fiz1             = _mm256_add_ps(fiz1,tz);
1488
1489             fjx0             = _mm256_add_ps(fjx0,tx);
1490             fjy0             = _mm256_add_ps(fjy0,ty);
1491             fjz0             = _mm256_add_ps(fjz0,tz);
1492
1493             /**************************
1494              * CALCULATE INTERACTIONS *
1495              **************************/
1496
1497             r20              = _mm256_mul_ps(rsq20,rinv20);
1498             r20              = _mm256_andnot_ps(dummy_mask,r20);
1499
1500             /* Compute parameters for interactions between i and j atoms */
1501             qq20             = _mm256_mul_ps(iq2,jq0);
1502
1503             /* EWALD ELECTROSTATICS */
1504
1505             /* Analytical PME correction */
1506             zeta2            = _mm256_mul_ps(beta2,rsq20);
1507             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1508             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1509             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1510             felec            = _mm256_mul_ps(qq20,felec);
1511
1512             fscal            = felec;
1513
1514             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1515
1516             /* Calculate temporary vectorial force */
1517             tx               = _mm256_mul_ps(fscal,dx20);
1518             ty               = _mm256_mul_ps(fscal,dy20);
1519             tz               = _mm256_mul_ps(fscal,dz20);
1520
1521             /* Update vectorial force */
1522             fix2             = _mm256_add_ps(fix2,tx);
1523             fiy2             = _mm256_add_ps(fiy2,ty);
1524             fiz2             = _mm256_add_ps(fiz2,tz);
1525
1526             fjx0             = _mm256_add_ps(fjx0,tx);
1527             fjy0             = _mm256_add_ps(fjy0,ty);
1528             fjz0             = _mm256_add_ps(fjz0,tz);
1529
1530             /**************************
1531              * CALCULATE INTERACTIONS *
1532              **************************/
1533
1534             r30              = _mm256_mul_ps(rsq30,rinv30);
1535             r30              = _mm256_andnot_ps(dummy_mask,r30);
1536
1537             /* Compute parameters for interactions between i and j atoms */
1538             qq30             = _mm256_mul_ps(iq3,jq0);
1539
1540             /* EWALD ELECTROSTATICS */
1541
1542             /* Analytical PME correction */
1543             zeta2            = _mm256_mul_ps(beta2,rsq30);
1544             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1545             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1546             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1547             felec            = _mm256_mul_ps(qq30,felec);
1548
1549             fscal            = felec;
1550
1551             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1552
1553             /* Calculate temporary vectorial force */
1554             tx               = _mm256_mul_ps(fscal,dx30);
1555             ty               = _mm256_mul_ps(fscal,dy30);
1556             tz               = _mm256_mul_ps(fscal,dz30);
1557
1558             /* Update vectorial force */
1559             fix3             = _mm256_add_ps(fix3,tx);
1560             fiy3             = _mm256_add_ps(fiy3,ty);
1561             fiz3             = _mm256_add_ps(fiz3,tz);
1562
1563             fjx0             = _mm256_add_ps(fjx0,tx);
1564             fjy0             = _mm256_add_ps(fjy0,ty);
1565             fjz0             = _mm256_add_ps(fjz0,tz);
1566
1567             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1568             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1569             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1570             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1571             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1572             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1573             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1574             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1575
1576             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1577
1578             /* Inner loop uses 223 flops */
1579         }
1580
1581         /* End of innermost loop */
1582
1583         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1584                                                  f+i_coord_offset,fshift+i_shift_offset);
1585
1586         /* Increment number of inner iterations */
1587         inneriter                  += j_index_end - j_index_start;
1588
1589         /* Outer loop uses 24 flops */
1590     }
1591
1592     /* Increment number of outer iterations */
1593     outeriter        += nri;
1594
1595     /* Update outer/inner flops */
1596
1597     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*223);
1598 }