src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwCSTab_GeomW4P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     real *           vdwioffsetptr3;
  79     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  80     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  81     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  82     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  83     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  84     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  85     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  86     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  87     real             *charge;
  88     int              nvdwtype;
  89     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  93     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  94     __m256i          vfitab;
  95     __m128i          vfitab_lo,vfitab_hi;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  98     real             *vftab;
  99     __m256i          ewitab;
 100     __m128i          ewitab_lo,ewitab_hi;
 101     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 102     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 103     real             *ewtab;
 104     __m256           dummy_mask,cutoff_mask;
 105     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 106     __m256           one     = _mm256_set1_ps(1.0);
 107     __m256           two     = _mm256_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     vftab            = kernel_data->table_vdw->data;
 126     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 127
 128     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 129     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 130     beta2            = _mm256_mul_ps(beta,beta);
 131     beta3            = _mm256_mul_ps(beta,beta2);
 132
 133     ewtab            = fr->ic->tabq_coul_FDV0;
 134     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 135     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 140     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 141     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 142     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150     j_coord_offsetE = 0;
 151     j_coord_offsetF = 0;
 152     j_coord_offsetG = 0;
 153     j_coord_offsetH = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 180
 181         fix0             = _mm256_setzero_ps();
 182         fiy0             = _mm256_setzero_ps();
 183         fiz0             = _mm256_setzero_ps();
 184         fix1             = _mm256_setzero_ps();
 185         fiy1             = _mm256_setzero_ps();
 186         fiz1             = _mm256_setzero_ps();
 187         fix2             = _mm256_setzero_ps();
 188         fiy2             = _mm256_setzero_ps();
 189         fiz2             = _mm256_setzero_ps();
 190         fix3             = _mm256_setzero_ps();
 191         fiy3             = _mm256_setzero_ps();
 192         fiz3             = _mm256_setzero_ps();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm256_setzero_ps();
 196         vvdwsum          = _mm256_setzero_ps();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             jnrE             = jjnr[jidx+4];
 208             jnrF             = jjnr[jidx+5];
 209             jnrG             = jjnr[jidx+6];
 210             jnrH             = jjnr[jidx+7];
 211             j_coord_offsetA  = DIM*jnrA;
 212             j_coord_offsetB  = DIM*jnrB;
 213             j_coord_offsetC  = DIM*jnrC;
 214             j_coord_offsetD  = DIM*jnrD;
 215             j_coord_offsetE  = DIM*jnrE;
 216             j_coord_offsetF  = DIM*jnrF;
 217             j_coord_offsetG  = DIM*jnrG;
 218             j_coord_offsetH  = DIM*jnrH;
 219
 220             /* load j atom coordinates */
 221             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 222                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 223                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 224                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 225                                                  &jx0,&jy0,&jz0);
 226
 227             /* Calculate displacement vector */
 228             dx00             = _mm256_sub_ps(ix0,jx0);
 229             dy00             = _mm256_sub_ps(iy0,jy0);
 230             dz00             = _mm256_sub_ps(iz0,jz0);
 231             dx10             = _mm256_sub_ps(ix1,jx0);
 232             dy10             = _mm256_sub_ps(iy1,jy0);
 233             dz10             = _mm256_sub_ps(iz1,jz0);
 234             dx20             = _mm256_sub_ps(ix2,jx0);
 235             dy20             = _mm256_sub_ps(iy2,jy0);
 236             dz20             = _mm256_sub_ps(iz2,jz0);
 237             dx30             = _mm256_sub_ps(ix3,jx0);
 238             dy30             = _mm256_sub_ps(iy3,jy0);
 239             dz30             = _mm256_sub_ps(iz3,jz0);
 240
 241             /* Calculate squared distance and things based on it */
 242             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 243             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 244             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 245             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 246
 247             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 248             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 249             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 250             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 251
 252             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 253             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 254             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 255
 256             /* Load parameters for j particles */
 257             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 258                                                                  charge+jnrC+0,charge+jnrD+0,
 259                                                                  charge+jnrE+0,charge+jnrF+0,
 260                                                                  charge+jnrG+0,charge+jnrH+0);
 261             vdwjidx0A        = 2*vdwtype[jnrA+0];
 262             vdwjidx0B        = 2*vdwtype[jnrB+0];
 263             vdwjidx0C        = 2*vdwtype[jnrC+0];
 264             vdwjidx0D        = 2*vdwtype[jnrD+0];
 265             vdwjidx0E        = 2*vdwtype[jnrE+0];
 266             vdwjidx0F        = 2*vdwtype[jnrF+0];
 267             vdwjidx0G        = 2*vdwtype[jnrG+0];
 268             vdwjidx0H        = 2*vdwtype[jnrH+0];
 269
 270             fjx0             = _mm256_setzero_ps();
 271             fjy0             = _mm256_setzero_ps();
 272             fjz0             = _mm256_setzero_ps();
 273
 274             /**************************
 275              * CALCULATE INTERACTIONS *
 276              **************************/
 277
 278             r00              = _mm256_mul_ps(rsq00,rinv00);
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 282                                             vdwioffsetptr0+vdwjidx0B,
 283                                             vdwioffsetptr0+vdwjidx0C,
 284                                             vdwioffsetptr0+vdwjidx0D,
 285                                             vdwioffsetptr0+vdwjidx0E,
 286                                             vdwioffsetptr0+vdwjidx0F,
 287                                             vdwioffsetptr0+vdwjidx0G,
 288                                             vdwioffsetptr0+vdwjidx0H,
 289                                             &c6_00,&c12_00);
 290
 291             /* Calculate table index by multiplying r with table scale and truncate to integer */
 292             rt               = _mm256_mul_ps(r00,vftabscale);
 293             vfitab           = _mm256_cvttps_epi32(rt);
 294             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 295             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 296             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 297             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 298             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 299             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 300
 301             /* CUBIC SPLINE TABLE DISPERSION */
 302             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 303                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 304             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 305                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 306             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 307                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 308             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 309                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 310             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 311             Heps             = _mm256_mul_ps(vfeps,H);
 312             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 313             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 314             vvdw6            = _mm256_mul_ps(c6_00,VV);
 315             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 316             fvdw6            = _mm256_mul_ps(c6_00,FF);
 317
 318             /* CUBIC SPLINE TABLE REPULSION */
 319             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 320             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 321             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 322                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 323             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 324                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 325             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 326                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 327             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 328                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 329             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 330             Heps             = _mm256_mul_ps(vfeps,H);
 331             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 332             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 333             vvdw12           = _mm256_mul_ps(c12_00,VV);
 334             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 335             fvdw12           = _mm256_mul_ps(c12_00,FF);
 336             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 337             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 338
 339             /* Update potential sum for this i atom from the interaction with this j atom. */
 340             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 341
 342             fscal            = fvdw;
 343
 344             /* Calculate temporary vectorial force */
 345             tx               = _mm256_mul_ps(fscal,dx00);
 346             ty               = _mm256_mul_ps(fscal,dy00);
 347             tz               = _mm256_mul_ps(fscal,dz00);
 348
 349             /* Update vectorial force */
 350             fix0             = _mm256_add_ps(fix0,tx);
 351             fiy0             = _mm256_add_ps(fiy0,ty);
 352             fiz0             = _mm256_add_ps(fiz0,tz);
 353
 354             fjx0             = _mm256_add_ps(fjx0,tx);
 355             fjy0             = _mm256_add_ps(fjy0,ty);
 356             fjz0             = _mm256_add_ps(fjz0,tz);
 357
 358             /**************************
 359              * CALCULATE INTERACTIONS *
 360              **************************/
 361
 362             r10              = _mm256_mul_ps(rsq10,rinv10);
 363
 364             /* Compute parameters for interactions between i and j atoms */
 365             qq10             = _mm256_mul_ps(iq1,jq0);
 366
 367             /* EWALD ELECTROSTATICS */
 368
 369             /* Analytical PME correction */
 370             zeta2            = _mm256_mul_ps(beta2,rsq10);
 371             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 372             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 373             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 374             felec            = _mm256_mul_ps(qq10,felec);
 375             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 376             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 377             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 378             velec            = _mm256_mul_ps(qq10,velec);
 379
 380             /* Update potential sum for this i atom from the interaction with this j atom. */
 381             velecsum         = _mm256_add_ps(velecsum,velec);
 382
 383             fscal            = felec;
 384
 385             /* Calculate temporary vectorial force */
 386             tx               = _mm256_mul_ps(fscal,dx10);
 387             ty               = _mm256_mul_ps(fscal,dy10);
 388             tz               = _mm256_mul_ps(fscal,dz10);
 389
 390             /* Update vectorial force */
 391             fix1             = _mm256_add_ps(fix1,tx);
 392             fiy1             = _mm256_add_ps(fiy1,ty);
 393             fiz1             = _mm256_add_ps(fiz1,tz);
 394
 395             fjx0             = _mm256_add_ps(fjx0,tx);
 396             fjy0             = _mm256_add_ps(fjy0,ty);
 397             fjz0             = _mm256_add_ps(fjz0,tz);
 398
 399             /**************************
 400              * CALCULATE INTERACTIONS *
 401              **************************/
 402
 403             r20              = _mm256_mul_ps(rsq20,rinv20);
 404
 405             /* Compute parameters for interactions between i and j atoms */
 406             qq20             = _mm256_mul_ps(iq2,jq0);
 407
 408             /* EWALD ELECTROSTATICS */
 409
 410             /* Analytical PME correction */
 411             zeta2            = _mm256_mul_ps(beta2,rsq20);
 412             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 413             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 414             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 415             felec            = _mm256_mul_ps(qq20,felec);
 416             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 417             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 418             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 419             velec            = _mm256_mul_ps(qq20,velec);
 420
 421             /* Update potential sum for this i atom from the interaction with this j atom. */
 422             velecsum         = _mm256_add_ps(velecsum,velec);
 423
 424             fscal            = felec;
 425
 426             /* Calculate temporary vectorial force */
 427             tx               = _mm256_mul_ps(fscal,dx20);
 428             ty               = _mm256_mul_ps(fscal,dy20);
 429             tz               = _mm256_mul_ps(fscal,dz20);
 430
 431             /* Update vectorial force */
 432             fix2             = _mm256_add_ps(fix2,tx);
 433             fiy2             = _mm256_add_ps(fiy2,ty);
 434             fiz2             = _mm256_add_ps(fiz2,tz);
 435
 436             fjx0             = _mm256_add_ps(fjx0,tx);
 437             fjy0             = _mm256_add_ps(fjy0,ty);
 438             fjz0             = _mm256_add_ps(fjz0,tz);
 439
 440             /**************************
 441              * CALCULATE INTERACTIONS *
 442              **************************/
 443
 444             r30              = _mm256_mul_ps(rsq30,rinv30);
 445
 446             /* Compute parameters for interactions between i and j atoms */
 447             qq30             = _mm256_mul_ps(iq3,jq0);
 448
 449             /* EWALD ELECTROSTATICS */
 450
 451             /* Analytical PME correction */
 452             zeta2            = _mm256_mul_ps(beta2,rsq30);
 453             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 454             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 455             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 456             felec            = _mm256_mul_ps(qq30,felec);
 457             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 458             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 459             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 460             velec            = _mm256_mul_ps(qq30,velec);
 461
 462             /* Update potential sum for this i atom from the interaction with this j atom. */
 463             velecsum         = _mm256_add_ps(velecsum,velec);
 464
 465             fscal            = felec;
 466
 467             /* Calculate temporary vectorial force */
 468             tx               = _mm256_mul_ps(fscal,dx30);
 469             ty               = _mm256_mul_ps(fscal,dy30);
 470             tz               = _mm256_mul_ps(fscal,dz30);
 471
 472             /* Update vectorial force */
 473             fix3             = _mm256_add_ps(fix3,tx);
 474             fiy3             = _mm256_add_ps(fiy3,ty);
 475             fiz3             = _mm256_add_ps(fiz3,tz);
 476
 477             fjx0             = _mm256_add_ps(fjx0,tx);
 478             fjy0             = _mm256_add_ps(fjy0,ty);
 479             fjz0             = _mm256_add_ps(fjz0,tz);
 480
 481             fjptrA             = f+j_coord_offsetA;
 482             fjptrB             = f+j_coord_offsetB;
 483             fjptrC             = f+j_coord_offsetC;
 484             fjptrD             = f+j_coord_offsetD;
 485             fjptrE             = f+j_coord_offsetE;
 486             fjptrF             = f+j_coord_offsetF;
 487             fjptrG             = f+j_coord_offsetG;
 488             fjptrH             = f+j_coord_offsetH;
 489
 490             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 491
 492             /* Inner loop uses 311 flops */
 493         }
 494
 495         if(jidx<j_index_end)
 496         {
 497
 498             /* Get j neighbor index, and coordinate index */
 499             jnrlistA         = jjnr[jidx];
 500             jnrlistB         = jjnr[jidx+1];
 501             jnrlistC         = jjnr[jidx+2];
 502             jnrlistD         = jjnr[jidx+3];
 503             jnrlistE         = jjnr[jidx+4];
 504             jnrlistF         = jjnr[jidx+5];
 505             jnrlistG         = jjnr[jidx+6];
 506             jnrlistH         = jjnr[jidx+7];
 507             /* Sign of each element will be negative for non-real atoms.
 508              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 509              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 510              */
 511             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 512                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 513
 514             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 515             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 516             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 517             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 518             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 519             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 520             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 521             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 522             j_coord_offsetA  = DIM*jnrA;
 523             j_coord_offsetB  = DIM*jnrB;
 524             j_coord_offsetC  = DIM*jnrC;
 525             j_coord_offsetD  = DIM*jnrD;
 526             j_coord_offsetE  = DIM*jnrE;
 527             j_coord_offsetF  = DIM*jnrF;
 528             j_coord_offsetG  = DIM*jnrG;
 529             j_coord_offsetH  = DIM*jnrH;
 530
 531             /* load j atom coordinates */
 532             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 533                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 534                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 535                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 536                                                  &jx0,&jy0,&jz0);
 537
 538             /* Calculate displacement vector */
 539             dx00             = _mm256_sub_ps(ix0,jx0);
 540             dy00             = _mm256_sub_ps(iy0,jy0);
 541             dz00             = _mm256_sub_ps(iz0,jz0);
 542             dx10             = _mm256_sub_ps(ix1,jx0);
 543             dy10             = _mm256_sub_ps(iy1,jy0);
 544             dz10             = _mm256_sub_ps(iz1,jz0);
 545             dx20             = _mm256_sub_ps(ix2,jx0);
 546             dy20             = _mm256_sub_ps(iy2,jy0);
 547             dz20             = _mm256_sub_ps(iz2,jz0);
 548             dx30             = _mm256_sub_ps(ix3,jx0);
 549             dy30             = _mm256_sub_ps(iy3,jy0);
 550             dz30             = _mm256_sub_ps(iz3,jz0);
 551
 552             /* Calculate squared distance and things based on it */
 553             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 554             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 555             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 556             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 557
 558             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 559             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 560             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 561             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 562
 563             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 564             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 565             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 566
 567             /* Load parameters for j particles */
 568             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 569                                                                  charge+jnrC+0,charge+jnrD+0,
 570                                                                  charge+jnrE+0,charge+jnrF+0,
 571                                                                  charge+jnrG+0,charge+jnrH+0);
 572             vdwjidx0A        = 2*vdwtype[jnrA+0];
 573             vdwjidx0B        = 2*vdwtype[jnrB+0];
 574             vdwjidx0C        = 2*vdwtype[jnrC+0];
 575             vdwjidx0D        = 2*vdwtype[jnrD+0];
 576             vdwjidx0E        = 2*vdwtype[jnrE+0];
 577             vdwjidx0F        = 2*vdwtype[jnrF+0];
 578             vdwjidx0G        = 2*vdwtype[jnrG+0];
 579             vdwjidx0H        = 2*vdwtype[jnrH+0];
 580
 581             fjx0             = _mm256_setzero_ps();
 582             fjy0             = _mm256_setzero_ps();
 583             fjz0             = _mm256_setzero_ps();
 584
 585             /**************************
 586              * CALCULATE INTERACTIONS *
 587              **************************/
 588
 589             r00              = _mm256_mul_ps(rsq00,rinv00);
 590             r00              = _mm256_andnot_ps(dummy_mask,r00);
 591
 592             /* Compute parameters for interactions between i and j atoms */
 593             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 594                                             vdwioffsetptr0+vdwjidx0B,
 595                                             vdwioffsetptr0+vdwjidx0C,
 596                                             vdwioffsetptr0+vdwjidx0D,
 597                                             vdwioffsetptr0+vdwjidx0E,
 598                                             vdwioffsetptr0+vdwjidx0F,
 599                                             vdwioffsetptr0+vdwjidx0G,
 600                                             vdwioffsetptr0+vdwjidx0H,
 601                                             &c6_00,&c12_00);
 602
 603             /* Calculate table index by multiplying r with table scale and truncate to integer */
 604             rt               = _mm256_mul_ps(r00,vftabscale);
 605             vfitab           = _mm256_cvttps_epi32(rt);
 606             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 607             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 608             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 609             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 610             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 611             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 612
 613             /* CUBIC SPLINE TABLE DISPERSION */
 614             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 615                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 616             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 617                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 618             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 619                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 620             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 621                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 622             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 623             Heps             = _mm256_mul_ps(vfeps,H);
 624             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 625             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 626             vvdw6            = _mm256_mul_ps(c6_00,VV);
 627             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 628             fvdw6            = _mm256_mul_ps(c6_00,FF);
 629
 630             /* CUBIC SPLINE TABLE REPULSION */
 631             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 632             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 633             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 634                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 635             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 636                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 637             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 638                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 639             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 640                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 641             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 642             Heps             = _mm256_mul_ps(vfeps,H);
 643             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 644             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 645             vvdw12           = _mm256_mul_ps(c12_00,VV);
 646             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 647             fvdw12           = _mm256_mul_ps(c12_00,FF);
 648             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 649             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 650
 651             /* Update potential sum for this i atom from the interaction with this j atom. */
 652             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 653             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 654
 655             fscal            = fvdw;
 656
 657             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 658
 659             /* Calculate temporary vectorial force */
 660             tx               = _mm256_mul_ps(fscal,dx00);
 661             ty               = _mm256_mul_ps(fscal,dy00);
 662             tz               = _mm256_mul_ps(fscal,dz00);
 663
 664             /* Update vectorial force */
 665             fix0             = _mm256_add_ps(fix0,tx);
 666             fiy0             = _mm256_add_ps(fiy0,ty);
 667             fiz0             = _mm256_add_ps(fiz0,tz);
 668
 669             fjx0             = _mm256_add_ps(fjx0,tx);
 670             fjy0             = _mm256_add_ps(fjy0,ty);
 671             fjz0             = _mm256_add_ps(fjz0,tz);
 672
 673             /**************************
 674              * CALCULATE INTERACTIONS *
 675              **************************/
 676
 677             r10              = _mm256_mul_ps(rsq10,rinv10);
 678             r10              = _mm256_andnot_ps(dummy_mask,r10);
 679
 680             /* Compute parameters for interactions between i and j atoms */
 681             qq10             = _mm256_mul_ps(iq1,jq0);
 682
 683             /* EWALD ELECTROSTATICS */
 684
 685             /* Analytical PME correction */
 686             zeta2            = _mm256_mul_ps(beta2,rsq10);
 687             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 688             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 689             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 690             felec            = _mm256_mul_ps(qq10,felec);
 691             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 692             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 693             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 694             velec            = _mm256_mul_ps(qq10,velec);
 695
 696             /* Update potential sum for this i atom from the interaction with this j atom. */
 697             velec            = _mm256_andnot_ps(dummy_mask,velec);
 698             velecsum         = _mm256_add_ps(velecsum,velec);
 699
 700             fscal            = felec;
 701
 702             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 703
 704             /* Calculate temporary vectorial force */
 705             tx               = _mm256_mul_ps(fscal,dx10);
 706             ty               = _mm256_mul_ps(fscal,dy10);
 707             tz               = _mm256_mul_ps(fscal,dz10);
 708
 709             /* Update vectorial force */
 710             fix1             = _mm256_add_ps(fix1,tx);
 711             fiy1             = _mm256_add_ps(fiy1,ty);
 712             fiz1             = _mm256_add_ps(fiz1,tz);
 713
 714             fjx0             = _mm256_add_ps(fjx0,tx);
 715             fjy0             = _mm256_add_ps(fjy0,ty);
 716             fjz0             = _mm256_add_ps(fjz0,tz);
 717
 718             /**************************
 719              * CALCULATE INTERACTIONS *
 720              **************************/
 721
 722             r20              = _mm256_mul_ps(rsq20,rinv20);
 723             r20              = _mm256_andnot_ps(dummy_mask,r20);
 724
 725             /* Compute parameters for interactions between i and j atoms */
 726             qq20             = _mm256_mul_ps(iq2,jq0);
 727
 728             /* EWALD ELECTROSTATICS */
 729
 730             /* Analytical PME correction */
 731             zeta2            = _mm256_mul_ps(beta2,rsq20);
 732             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 733             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 734             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 735             felec            = _mm256_mul_ps(qq20,felec);
 736             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 737             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 738             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 739             velec            = _mm256_mul_ps(qq20,velec);
 740
 741             /* Update potential sum for this i atom from the interaction with this j atom. */
 742             velec            = _mm256_andnot_ps(dummy_mask,velec);
 743             velecsum         = _mm256_add_ps(velecsum,velec);
 744
 745             fscal            = felec;
 746
 747             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 748
 749             /* Calculate temporary vectorial force */
 750             tx               = _mm256_mul_ps(fscal,dx20);
 751             ty               = _mm256_mul_ps(fscal,dy20);
 752             tz               = _mm256_mul_ps(fscal,dz20);
 753
 754             /* Update vectorial force */
 755             fix2             = _mm256_add_ps(fix2,tx);
 756             fiy2             = _mm256_add_ps(fiy2,ty);
 757             fiz2             = _mm256_add_ps(fiz2,tz);
 758
 759             fjx0             = _mm256_add_ps(fjx0,tx);
 760             fjy0             = _mm256_add_ps(fjy0,ty);
 761             fjz0             = _mm256_add_ps(fjz0,tz);
 762
 763             /**************************
 764              * CALCULATE INTERACTIONS *
 765              **************************/
 766
 767             r30              = _mm256_mul_ps(rsq30,rinv30);
 768             r30              = _mm256_andnot_ps(dummy_mask,r30);
 769
 770             /* Compute parameters for interactions between i and j atoms */
 771             qq30             = _mm256_mul_ps(iq3,jq0);
 772
 773             /* EWALD ELECTROSTATICS */
 774
 775             /* Analytical PME correction */
 776             zeta2            = _mm256_mul_ps(beta2,rsq30);
 777             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 778             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 779             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 780             felec            = _mm256_mul_ps(qq30,felec);
 781             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 782             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 783             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 784             velec            = _mm256_mul_ps(qq30,velec);
 785
 786             /* Update potential sum for this i atom from the interaction with this j atom. */
 787             velec            = _mm256_andnot_ps(dummy_mask,velec);
 788             velecsum         = _mm256_add_ps(velecsum,velec);
 789
 790             fscal            = felec;
 791
 792             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 793
 794             /* Calculate temporary vectorial force */
 795             tx               = _mm256_mul_ps(fscal,dx30);
 796             ty               = _mm256_mul_ps(fscal,dy30);
 797             tz               = _mm256_mul_ps(fscal,dz30);
 798
 799             /* Update vectorial force */
 800             fix3             = _mm256_add_ps(fix3,tx);
 801             fiy3             = _mm256_add_ps(fiy3,ty);
 802             fiz3             = _mm256_add_ps(fiz3,tz);
 803
 804             fjx0             = _mm256_add_ps(fjx0,tx);
 805             fjy0             = _mm256_add_ps(fjy0,ty);
 806             fjz0             = _mm256_add_ps(fjz0,tz);
 807
 808             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 809             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 810             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 811             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 812             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 813             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 814             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 815             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 816
 817             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 818
 819             /* Inner loop uses 315 flops */
 820         }
 821
 822         /* End of innermost loop */
 823
 824         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 825                                                  f+i_coord_offset,fshift+i_shift_offset);
 826
 827         ggid                        = gid[iidx];
 828         /* Update potential energies */
 829         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 830         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 831
 832         /* Increment number of inner iterations */
 833         inneriter                  += j_index_end - j_index_start;
 834
 835         /* Outer loop uses 26 flops */
 836     }
 837
 838     /* Increment number of outer iterations */
 839     outeriter        += nri;
 840
 841     /* Update outer/inner flops */
 842
 843     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*315);
 844 }
 845 /*
 846  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_256_single
 847  * Electrostatics interaction: Ewald
 848  * VdW interaction:            CubicSplineTable
 849  * Geometry:                   Water4-Particle
 850  * Calculate force/pot:        Force
 851  */
 852 void
 853 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_256_single
 854                     (t_nblist * gmx_restrict                nlist,
 855                      rvec * gmx_restrict                    xx,
 856                      rvec * gmx_restrict                    ff,
 857                      t_forcerec * gmx_restrict              fr,
 858                      t_mdatoms * gmx_restrict               mdatoms,
 859                      nb_kernel_data_t * gmx_restrict        kernel_data,
 860                      t_nrnb * gmx_restrict                  nrnb)
 861 {
 862     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 863      * just 0 for non-waters.
 864      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 865      * jnr indices corresponding to data put in the four positions in the SIMD register.
 866      */
 867     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 868     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 869     int              jnrA,jnrB,jnrC,jnrD;
 870     int              jnrE,jnrF,jnrG,jnrH;
 871     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 872     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 873     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 874     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 875     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 876     real             rcutoff_scalar;
 877     real             *shiftvec,*fshift,*x,*f;
 878     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 879     real             scratch[4*DIM];
 880     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 881     real *           vdwioffsetptr0;
 882     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 883     real *           vdwioffsetptr1;
 884     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 885     real *           vdwioffsetptr2;
 886     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 887     real *           vdwioffsetptr3;
 888     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 889     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 890     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 891     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 892     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 893     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 894     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 895     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 896     real             *charge;
 897     int              nvdwtype;
 898     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 899     int              *vdwtype;
 900     real             *vdwparam;
 901     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 902     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 903     __m256i          vfitab;
 904     __m128i          vfitab_lo,vfitab_hi;
 905     __m128i          ifour       = _mm_set1_epi32(4);
 906     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 907     real             *vftab;
 908     __m256i          ewitab;
 909     __m128i          ewitab_lo,ewitab_hi;
 910     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 911     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 912     real             *ewtab;
 913     __m256           dummy_mask,cutoff_mask;
 914     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 915     __m256           one     = _mm256_set1_ps(1.0);
 916     __m256           two     = _mm256_set1_ps(2.0);
 917     x                = xx[0];
 918     f                = ff[0];
 919
 920     nri              = nlist->nri;
 921     iinr             = nlist->iinr;
 922     jindex           = nlist->jindex;
 923     jjnr             = nlist->jjnr;
 924     shiftidx         = nlist->shift;
 925     gid              = nlist->gid;
 926     shiftvec         = fr->shift_vec[0];
 927     fshift           = fr->fshift[0];
 928     facel            = _mm256_set1_ps(fr->epsfac);
 929     charge           = mdatoms->chargeA;
 930     nvdwtype         = fr->ntype;
 931     vdwparam         = fr->nbfp;
 932     vdwtype          = mdatoms->typeA;
 933
 934     vftab            = kernel_data->table_vdw->data;
 935     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 936
 937     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 938     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 939     beta2            = _mm256_mul_ps(beta,beta);
 940     beta3            = _mm256_mul_ps(beta,beta2);
 941
 942     ewtab            = fr->ic->tabq_coul_F;
 943     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 944     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 945
 946     /* Setup water-specific parameters */
 947     inr              = nlist->iinr[0];
 948     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 949     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 950     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 951     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 952
 953     /* Avoid stupid compiler warnings */
 954     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 955     j_coord_offsetA = 0;
 956     j_coord_offsetB = 0;
 957     j_coord_offsetC = 0;
 958     j_coord_offsetD = 0;
 959     j_coord_offsetE = 0;
 960     j_coord_offsetF = 0;
 961     j_coord_offsetG = 0;
 962     j_coord_offsetH = 0;
 963
 964     outeriter        = 0;
 965     inneriter        = 0;
 966
 967     for(iidx=0;iidx<4*DIM;iidx++)
 968     {
 969         scratch[iidx] = 0.0;
 970     }
 971
 972     /* Start outer loop over neighborlists */
 973     for(iidx=0; iidx<nri; iidx++)
 974     {
 975         /* Load shift vector for this list */
 976         i_shift_offset   = DIM*shiftidx[iidx];
 977
 978         /* Load limits for loop over neighbors */
 979         j_index_start    = jindex[iidx];
 980         j_index_end      = jindex[iidx+1];
 981
 982         /* Get outer coordinate index */
 983         inr              = iinr[iidx];
 984         i_coord_offset   = DIM*inr;
 985
 986         /* Load i particle coords and add shift vector */
 987         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 988                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 989
 990         fix0             = _mm256_setzero_ps();
 991         fiy0             = _mm256_setzero_ps();
 992         fiz0             = _mm256_setzero_ps();
 993         fix1             = _mm256_setzero_ps();
 994         fiy1             = _mm256_setzero_ps();
 995         fiz1             = _mm256_setzero_ps();
 996         fix2             = _mm256_setzero_ps();
 997         fiy2             = _mm256_setzero_ps();
 998         fiz2             = _mm256_setzero_ps();
 999         fix3             = _mm256_setzero_ps();
1000         fiy3             = _mm256_setzero_ps();
1001         fiz3             = _mm256_setzero_ps();
1002
1003         /* Start inner kernel loop */
1004         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1005         {
1006
1007             /* Get j neighbor index, and coordinate index */
1008             jnrA             = jjnr[jidx];
1009             jnrB             = jjnr[jidx+1];
1010             jnrC             = jjnr[jidx+2];
1011             jnrD             = jjnr[jidx+3];
1012             jnrE             = jjnr[jidx+4];
1013             jnrF             = jjnr[jidx+5];
1014             jnrG             = jjnr[jidx+6];
1015             jnrH             = jjnr[jidx+7];
1016             j_coord_offsetA  = DIM*jnrA;
1017             j_coord_offsetB  = DIM*jnrB;
1018             j_coord_offsetC  = DIM*jnrC;
1019             j_coord_offsetD  = DIM*jnrD;
1020             j_coord_offsetE  = DIM*jnrE;
1021             j_coord_offsetF  = DIM*jnrF;
1022             j_coord_offsetG  = DIM*jnrG;
1023             j_coord_offsetH  = DIM*jnrH;
1024
1025             /* load j atom coordinates */
1026             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1027                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1028                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1029                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1030                                                  &jx0,&jy0,&jz0);
1031
1032             /* Calculate displacement vector */
1033             dx00             = _mm256_sub_ps(ix0,jx0);
1034             dy00             = _mm256_sub_ps(iy0,jy0);
1035             dz00             = _mm256_sub_ps(iz0,jz0);
1036             dx10             = _mm256_sub_ps(ix1,jx0);
1037             dy10             = _mm256_sub_ps(iy1,jy0);
1038             dz10             = _mm256_sub_ps(iz1,jz0);
1039             dx20             = _mm256_sub_ps(ix2,jx0);
1040             dy20             = _mm256_sub_ps(iy2,jy0);
1041             dz20             = _mm256_sub_ps(iz2,jz0);
1042             dx30             = _mm256_sub_ps(ix3,jx0);
1043             dy30             = _mm256_sub_ps(iy3,jy0);
1044             dz30             = _mm256_sub_ps(iz3,jz0);
1045
1046             /* Calculate squared distance and things based on it */
1047             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1048             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1049             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1050             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1051
1052             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1053             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1054             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1055             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1056
1057             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1058             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1059             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1060
1061             /* Load parameters for j particles */
1062             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1063                                                                  charge+jnrC+0,charge+jnrD+0,
1064                                                                  charge+jnrE+0,charge+jnrF+0,
1065                                                                  charge+jnrG+0,charge+jnrH+0);
1066             vdwjidx0A        = 2*vdwtype[jnrA+0];
1067             vdwjidx0B        = 2*vdwtype[jnrB+0];
1068             vdwjidx0C        = 2*vdwtype[jnrC+0];
1069             vdwjidx0D        = 2*vdwtype[jnrD+0];
1070             vdwjidx0E        = 2*vdwtype[jnrE+0];
1071             vdwjidx0F        = 2*vdwtype[jnrF+0];
1072             vdwjidx0G        = 2*vdwtype[jnrG+0];
1073             vdwjidx0H        = 2*vdwtype[jnrH+0];
1074
1075             fjx0             = _mm256_setzero_ps();
1076             fjy0             = _mm256_setzero_ps();
1077             fjz0             = _mm256_setzero_ps();
1078
1079             /**************************
1080              * CALCULATE INTERACTIONS *
1081              **************************/
1082
1083             r00              = _mm256_mul_ps(rsq00,rinv00);
1084
1085             /* Compute parameters for interactions between i and j atoms */
1086             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1087                                             vdwioffsetptr0+vdwjidx0B,
1088                                             vdwioffsetptr0+vdwjidx0C,
1089                                             vdwioffsetptr0+vdwjidx0D,
1090                                             vdwioffsetptr0+vdwjidx0E,
1091                                             vdwioffsetptr0+vdwjidx0F,
1092                                             vdwioffsetptr0+vdwjidx0G,
1093                                             vdwioffsetptr0+vdwjidx0H,
1094                                             &c6_00,&c12_00);
1095
1096             /* Calculate table index by multiplying r with table scale and truncate to integer */
1097             rt               = _mm256_mul_ps(r00,vftabscale);
1098             vfitab           = _mm256_cvttps_epi32(rt);
1099             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1100             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1101             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1102             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1103             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1104             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1105
1106             /* CUBIC SPLINE TABLE DISPERSION */
1107             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1108                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1109             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1110                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1111             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1112                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1113             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1114                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1115             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1116             Heps             = _mm256_mul_ps(vfeps,H);
1117             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1118             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1119             fvdw6            = _mm256_mul_ps(c6_00,FF);
1120
1121             /* CUBIC SPLINE TABLE REPULSION */
1122             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1123             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1124             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1125                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1126             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1127                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1128             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1129                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1130             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1131                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1132             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1133             Heps             = _mm256_mul_ps(vfeps,H);
1134             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1135             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1136             fvdw12           = _mm256_mul_ps(c12_00,FF);
1137             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1138
1139             fscal            = fvdw;
1140
1141             /* Calculate temporary vectorial force */
1142             tx               = _mm256_mul_ps(fscal,dx00);
1143             ty               = _mm256_mul_ps(fscal,dy00);
1144             tz               = _mm256_mul_ps(fscal,dz00);
1145
1146             /* Update vectorial force */
1147             fix0             = _mm256_add_ps(fix0,tx);
1148             fiy0             = _mm256_add_ps(fiy0,ty);
1149             fiz0             = _mm256_add_ps(fiz0,tz);
1150
1151             fjx0             = _mm256_add_ps(fjx0,tx);
1152             fjy0             = _mm256_add_ps(fjy0,ty);
1153             fjz0             = _mm256_add_ps(fjz0,tz);
1154
1155             /**************************
1156              * CALCULATE INTERACTIONS *
1157              **************************/
1158
1159             r10              = _mm256_mul_ps(rsq10,rinv10);
1160
1161             /* Compute parameters for interactions between i and j atoms */
1162             qq10             = _mm256_mul_ps(iq1,jq0);
1163
1164             /* EWALD ELECTROSTATICS */
1165
1166             /* Analytical PME correction */
1167             zeta2            = _mm256_mul_ps(beta2,rsq10);
1168             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1169             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1170             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1171             felec            = _mm256_mul_ps(qq10,felec);
1172
1173             fscal            = felec;
1174
1175             /* Calculate temporary vectorial force */
1176             tx               = _mm256_mul_ps(fscal,dx10);
1177             ty               = _mm256_mul_ps(fscal,dy10);
1178             tz               = _mm256_mul_ps(fscal,dz10);
1179
1180             /* Update vectorial force */
1181             fix1             = _mm256_add_ps(fix1,tx);
1182             fiy1             = _mm256_add_ps(fiy1,ty);
1183             fiz1             = _mm256_add_ps(fiz1,tz);
1184
1185             fjx0             = _mm256_add_ps(fjx0,tx);
1186             fjy0             = _mm256_add_ps(fjy0,ty);
1187             fjz0             = _mm256_add_ps(fjz0,tz);
1188
1189             /**************************
1190              * CALCULATE INTERACTIONS *
1191              **************************/
1192
1193             r20              = _mm256_mul_ps(rsq20,rinv20);
1194
1195             /* Compute parameters for interactions between i and j atoms */
1196             qq20             = _mm256_mul_ps(iq2,jq0);
1197
1198             /* EWALD ELECTROSTATICS */
1199
1200             /* Analytical PME correction */
1201             zeta2            = _mm256_mul_ps(beta2,rsq20);
1202             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1203             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1204             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1205             felec            = _mm256_mul_ps(qq20,felec);
1206
1207             fscal            = felec;
1208
1209             /* Calculate temporary vectorial force */
1210             tx               = _mm256_mul_ps(fscal,dx20);
1211             ty               = _mm256_mul_ps(fscal,dy20);
1212             tz               = _mm256_mul_ps(fscal,dz20);
1213
1214             /* Update vectorial force */
1215             fix2             = _mm256_add_ps(fix2,tx);
1216             fiy2             = _mm256_add_ps(fiy2,ty);
1217             fiz2             = _mm256_add_ps(fiz2,tz);
1218
1219             fjx0             = _mm256_add_ps(fjx0,tx);
1220             fjy0             = _mm256_add_ps(fjy0,ty);
1221             fjz0             = _mm256_add_ps(fjz0,tz);
1222
1223             /**************************
1224              * CALCULATE INTERACTIONS *
1225              **************************/
1226
1227             r30              = _mm256_mul_ps(rsq30,rinv30);
1228
1229             /* Compute parameters for interactions between i and j atoms */
1230             qq30             = _mm256_mul_ps(iq3,jq0);
1231
1232             /* EWALD ELECTROSTATICS */
1233
1234             /* Analytical PME correction */
1235             zeta2            = _mm256_mul_ps(beta2,rsq30);
1236             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1237             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1238             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1239             felec            = _mm256_mul_ps(qq30,felec);
1240
1241             fscal            = felec;
1242
1243             /* Calculate temporary vectorial force */
1244             tx               = _mm256_mul_ps(fscal,dx30);
1245             ty               = _mm256_mul_ps(fscal,dy30);
1246             tz               = _mm256_mul_ps(fscal,dz30);
1247
1248             /* Update vectorial force */
1249             fix3             = _mm256_add_ps(fix3,tx);
1250             fiy3             = _mm256_add_ps(fiy3,ty);
1251             fiz3             = _mm256_add_ps(fiz3,tz);
1252
1253             fjx0             = _mm256_add_ps(fjx0,tx);
1254             fjy0             = _mm256_add_ps(fjy0,ty);
1255             fjz0             = _mm256_add_ps(fjz0,tz);
1256
1257             fjptrA             = f+j_coord_offsetA;
1258             fjptrB             = f+j_coord_offsetB;
1259             fjptrC             = f+j_coord_offsetC;
1260             fjptrD             = f+j_coord_offsetD;
1261             fjptrE             = f+j_coord_offsetE;
1262             fjptrF             = f+j_coord_offsetF;
1263             fjptrG             = f+j_coord_offsetG;
1264             fjptrH             = f+j_coord_offsetH;
1265
1266             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1267
1268             /* Inner loop uses 219 flops */
1269         }
1270
1271         if(jidx<j_index_end)
1272         {
1273
1274             /* Get j neighbor index, and coordinate index */
1275             jnrlistA         = jjnr[jidx];
1276             jnrlistB         = jjnr[jidx+1];
1277             jnrlistC         = jjnr[jidx+2];
1278             jnrlistD         = jjnr[jidx+3];
1279             jnrlistE         = jjnr[jidx+4];
1280             jnrlistF         = jjnr[jidx+5];
1281             jnrlistG         = jjnr[jidx+6];
1282             jnrlistH         = jjnr[jidx+7];
1283             /* Sign of each element will be negative for non-real atoms.
1284              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1285              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1286              */
1287             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1288                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1289
1290             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1291             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1292             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1293             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1294             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1295             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1296             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1297             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1298             j_coord_offsetA  = DIM*jnrA;
1299             j_coord_offsetB  = DIM*jnrB;
1300             j_coord_offsetC  = DIM*jnrC;
1301             j_coord_offsetD  = DIM*jnrD;
1302             j_coord_offsetE  = DIM*jnrE;
1303             j_coord_offsetF  = DIM*jnrF;
1304             j_coord_offsetG  = DIM*jnrG;
1305             j_coord_offsetH  = DIM*jnrH;
1306
1307             /* load j atom coordinates */
1308             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1309                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1310                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1311                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1312                                                  &jx0,&jy0,&jz0);
1313
1314             /* Calculate displacement vector */
1315             dx00             = _mm256_sub_ps(ix0,jx0);
1316             dy00             = _mm256_sub_ps(iy0,jy0);
1317             dz00             = _mm256_sub_ps(iz0,jz0);
1318             dx10             = _mm256_sub_ps(ix1,jx0);
1319             dy10             = _mm256_sub_ps(iy1,jy0);
1320             dz10             = _mm256_sub_ps(iz1,jz0);
1321             dx20             = _mm256_sub_ps(ix2,jx0);
1322             dy20             = _mm256_sub_ps(iy2,jy0);
1323             dz20             = _mm256_sub_ps(iz2,jz0);
1324             dx30             = _mm256_sub_ps(ix3,jx0);
1325             dy30             = _mm256_sub_ps(iy3,jy0);
1326             dz30             = _mm256_sub_ps(iz3,jz0);
1327
1328             /* Calculate squared distance and things based on it */
1329             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1330             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1331             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1332             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1333
1334             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1335             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1336             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1337             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1338
1339             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1340             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1341             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1342
1343             /* Load parameters for j particles */
1344             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1345                                                                  charge+jnrC+0,charge+jnrD+0,
1346                                                                  charge+jnrE+0,charge+jnrF+0,
1347                                                                  charge+jnrG+0,charge+jnrH+0);
1348             vdwjidx0A        = 2*vdwtype[jnrA+0];
1349             vdwjidx0B        = 2*vdwtype[jnrB+0];
1350             vdwjidx0C        = 2*vdwtype[jnrC+0];
1351             vdwjidx0D        = 2*vdwtype[jnrD+0];
1352             vdwjidx0E        = 2*vdwtype[jnrE+0];
1353             vdwjidx0F        = 2*vdwtype[jnrF+0];
1354             vdwjidx0G        = 2*vdwtype[jnrG+0];
1355             vdwjidx0H        = 2*vdwtype[jnrH+0];
1356
1357             fjx0             = _mm256_setzero_ps();
1358             fjy0             = _mm256_setzero_ps();
1359             fjz0             = _mm256_setzero_ps();
1360
1361             /**************************
1362              * CALCULATE INTERACTIONS *
1363              **************************/
1364
1365             r00              = _mm256_mul_ps(rsq00,rinv00);
1366             r00              = _mm256_andnot_ps(dummy_mask,r00);
1367
1368             /* Compute parameters for interactions between i and j atoms */
1369             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1370                                             vdwioffsetptr0+vdwjidx0B,
1371                                             vdwioffsetptr0+vdwjidx0C,
1372                                             vdwioffsetptr0+vdwjidx0D,
1373                                             vdwioffsetptr0+vdwjidx0E,
1374                                             vdwioffsetptr0+vdwjidx0F,
1375                                             vdwioffsetptr0+vdwjidx0G,
1376                                             vdwioffsetptr0+vdwjidx0H,
1377                                             &c6_00,&c12_00);
1378
1379             /* Calculate table index by multiplying r with table scale and truncate to integer */
1380             rt               = _mm256_mul_ps(r00,vftabscale);
1381             vfitab           = _mm256_cvttps_epi32(rt);
1382             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1383             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1384             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1385             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1386             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1387             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1388
1389             /* CUBIC SPLINE TABLE DISPERSION */
1390             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1391                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1392             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1393                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1394             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1395                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1396             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1397                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1398             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1399             Heps             = _mm256_mul_ps(vfeps,H);
1400             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1401             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1402             fvdw6            = _mm256_mul_ps(c6_00,FF);
1403
1404             /* CUBIC SPLINE TABLE REPULSION */
1405             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1406             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1407             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1408                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1409             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1410                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1411             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1412                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1413             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1414                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1415             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1416             Heps             = _mm256_mul_ps(vfeps,H);
1417             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1418             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1419             fvdw12           = _mm256_mul_ps(c12_00,FF);
1420             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1421
1422             fscal            = fvdw;
1423
1424             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1425
1426             /* Calculate temporary vectorial force */
1427             tx               = _mm256_mul_ps(fscal,dx00);
1428             ty               = _mm256_mul_ps(fscal,dy00);
1429             tz               = _mm256_mul_ps(fscal,dz00);
1430
1431             /* Update vectorial force */
1432             fix0             = _mm256_add_ps(fix0,tx);
1433             fiy0             = _mm256_add_ps(fiy0,ty);
1434             fiz0             = _mm256_add_ps(fiz0,tz);
1435
1436             fjx0             = _mm256_add_ps(fjx0,tx);
1437             fjy0             = _mm256_add_ps(fjy0,ty);
1438             fjz0             = _mm256_add_ps(fjz0,tz);
1439
1440             /**************************
1441              * CALCULATE INTERACTIONS *
1442              **************************/
1443
1444             r10              = _mm256_mul_ps(rsq10,rinv10);
1445             r10              = _mm256_andnot_ps(dummy_mask,r10);
1446
1447             /* Compute parameters for interactions between i and j atoms */
1448             qq10             = _mm256_mul_ps(iq1,jq0);
1449
1450             /* EWALD ELECTROSTATICS */
1451
1452             /* Analytical PME correction */
1453             zeta2            = _mm256_mul_ps(beta2,rsq10);
1454             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1455             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1456             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1457             felec            = _mm256_mul_ps(qq10,felec);
1458
1459             fscal            = felec;
1460
1461             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1462
1463             /* Calculate temporary vectorial force */
1464             tx               = _mm256_mul_ps(fscal,dx10);
1465             ty               = _mm256_mul_ps(fscal,dy10);
1466             tz               = _mm256_mul_ps(fscal,dz10);
1467
1468             /* Update vectorial force */
1469             fix1             = _mm256_add_ps(fix1,tx);
1470             fiy1             = _mm256_add_ps(fiy1,ty);
1471             fiz1             = _mm256_add_ps(fiz1,tz);
1472
1473             fjx0             = _mm256_add_ps(fjx0,tx);
1474             fjy0             = _mm256_add_ps(fjy0,ty);
1475             fjz0             = _mm256_add_ps(fjz0,tz);
1476
1477             /**************************
1478              * CALCULATE INTERACTIONS *
1479              **************************/
1480
1481             r20              = _mm256_mul_ps(rsq20,rinv20);
1482             r20              = _mm256_andnot_ps(dummy_mask,r20);
1483
1484             /* Compute parameters for interactions between i and j atoms */
1485             qq20             = _mm256_mul_ps(iq2,jq0);
1486
1487             /* EWALD ELECTROSTATICS */
1488
1489             /* Analytical PME correction */
1490             zeta2            = _mm256_mul_ps(beta2,rsq20);
1491             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1492             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1493             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1494             felec            = _mm256_mul_ps(qq20,felec);
1495
1496             fscal            = felec;
1497
1498             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1499
1500             /* Calculate temporary vectorial force */
1501             tx               = _mm256_mul_ps(fscal,dx20);
1502             ty               = _mm256_mul_ps(fscal,dy20);
1503             tz               = _mm256_mul_ps(fscal,dz20);
1504
1505             /* Update vectorial force */
1506             fix2             = _mm256_add_ps(fix2,tx);
1507             fiy2             = _mm256_add_ps(fiy2,ty);
1508             fiz2             = _mm256_add_ps(fiz2,tz);
1509
1510             fjx0             = _mm256_add_ps(fjx0,tx);
1511             fjy0             = _mm256_add_ps(fjy0,ty);
1512             fjz0             = _mm256_add_ps(fjz0,tz);
1513
1514             /**************************
1515              * CALCULATE INTERACTIONS *
1516              **************************/
1517
1518             r30              = _mm256_mul_ps(rsq30,rinv30);
1519             r30              = _mm256_andnot_ps(dummy_mask,r30);
1520
1521             /* Compute parameters for interactions between i and j atoms */
1522             qq30             = _mm256_mul_ps(iq3,jq0);
1523
1524             /* EWALD ELECTROSTATICS */
1525
1526             /* Analytical PME correction */
1527             zeta2            = _mm256_mul_ps(beta2,rsq30);
1528             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1529             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1530             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1531             felec            = _mm256_mul_ps(qq30,felec);
1532
1533             fscal            = felec;
1534
1535             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1536
1537             /* Calculate temporary vectorial force */
1538             tx               = _mm256_mul_ps(fscal,dx30);
1539             ty               = _mm256_mul_ps(fscal,dy30);
1540             tz               = _mm256_mul_ps(fscal,dz30);
1541
1542             /* Update vectorial force */
1543             fix3             = _mm256_add_ps(fix3,tx);
1544             fiy3             = _mm256_add_ps(fiy3,ty);
1545             fiz3             = _mm256_add_ps(fiz3,tz);
1546
1547             fjx0             = _mm256_add_ps(fjx0,tx);
1548             fjy0             = _mm256_add_ps(fjy0,ty);
1549             fjz0             = _mm256_add_ps(fjz0,tz);
1550
1551             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1552             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1553             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1554             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1555             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1556             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1557             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1558             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1559
1560             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1561
1562             /* Inner loop uses 223 flops */
1563         }
1564
1565         /* End of innermost loop */
1566
1567         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1568                                                  f+i_coord_offset,fshift+i_shift_offset);
1569
1570         /* Increment number of inner iterations */
1571         inneriter                  += j_index_end - j_index_start;
1572
1573         /* Outer loop uses 24 flops */
1574     }
1575
1576     /* Increment number of outer iterations */
1577     outeriter        += nri;
1578
1579     /* Update outer/inner flops */
1580
1581     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*223);
1582 }