src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_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_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_256_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_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     __m256           dummy_mask,cutoff_mask;
 100     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 101     __m256           one     = _mm256_set1_ps(1.0);
 102     __m256           two     = _mm256_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm256_set1_ps(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     vftab            = kernel_data->table_elec_vdw->data;
 121     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 122
 123     /* Setup water-specific parameters */
 124     inr              = nlist->iinr[0];
 125     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 126     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 127     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 128     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136     j_coord_offsetE = 0;
 137     j_coord_offsetF = 0;
 138     j_coord_offsetG = 0;
 139     j_coord_offsetH = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 165                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 166
 167         fix0             = _mm256_setzero_ps();
 168         fiy0             = _mm256_setzero_ps();
 169         fiz0             = _mm256_setzero_ps();
 170         fix1             = _mm256_setzero_ps();
 171         fiy1             = _mm256_setzero_ps();
 172         fiz1             = _mm256_setzero_ps();
 173         fix2             = _mm256_setzero_ps();
 174         fiy2             = _mm256_setzero_ps();
 175         fiz2             = _mm256_setzero_ps();
 176         fix3             = _mm256_setzero_ps();
 177         fiy3             = _mm256_setzero_ps();
 178         fiz3             = _mm256_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm256_setzero_ps();
 182         vvdwsum          = _mm256_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             jnrE             = jjnr[jidx+4];
 194             jnrF             = jjnr[jidx+5];
 195             jnrG             = jjnr[jidx+6];
 196             jnrH             = jjnr[jidx+7];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201             j_coord_offsetE  = DIM*jnrE;
 202             j_coord_offsetF  = DIM*jnrF;
 203             j_coord_offsetG  = DIM*jnrG;
 204             j_coord_offsetH  = DIM*jnrH;
 205
 206             /* load j atom coordinates */
 207             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 209                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 210                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 211                                                  &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm256_sub_ps(ix0,jx0);
 215             dy00             = _mm256_sub_ps(iy0,jy0);
 216             dz00             = _mm256_sub_ps(iz0,jz0);
 217             dx10             = _mm256_sub_ps(ix1,jx0);
 218             dy10             = _mm256_sub_ps(iy1,jy0);
 219             dz10             = _mm256_sub_ps(iz1,jz0);
 220             dx20             = _mm256_sub_ps(ix2,jx0);
 221             dy20             = _mm256_sub_ps(iy2,jy0);
 222             dz20             = _mm256_sub_ps(iz2,jz0);
 223             dx30             = _mm256_sub_ps(ix3,jx0);
 224             dy30             = _mm256_sub_ps(iy3,jy0);
 225             dz30             = _mm256_sub_ps(iz3,jz0);
 226
 227             /* Calculate squared distance and things based on it */
 228             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 229             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 230             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 231             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 232
 233             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 234             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 235             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 236             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 237
 238             /* Load parameters for j particles */
 239             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 240                                                                  charge+jnrC+0,charge+jnrD+0,
 241                                                                  charge+jnrE+0,charge+jnrF+0,
 242                                                                  charge+jnrG+0,charge+jnrH+0);
 243             vdwjidx0A        = 2*vdwtype[jnrA+0];
 244             vdwjidx0B        = 2*vdwtype[jnrB+0];
 245             vdwjidx0C        = 2*vdwtype[jnrC+0];
 246             vdwjidx0D        = 2*vdwtype[jnrD+0];
 247             vdwjidx0E        = 2*vdwtype[jnrE+0];
 248             vdwjidx0F        = 2*vdwtype[jnrF+0];
 249             vdwjidx0G        = 2*vdwtype[jnrG+0];
 250             vdwjidx0H        = 2*vdwtype[jnrH+0];
 251
 252             fjx0             = _mm256_setzero_ps();
 253             fjy0             = _mm256_setzero_ps();
 254             fjz0             = _mm256_setzero_ps();
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             r00              = _mm256_mul_ps(rsq00,rinv00);
 261
 262             /* Compute parameters for interactions between i and j atoms */
 263             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 264                                             vdwioffsetptr0+vdwjidx0B,
 265                                             vdwioffsetptr0+vdwjidx0C,
 266                                             vdwioffsetptr0+vdwjidx0D,
 267                                             vdwioffsetptr0+vdwjidx0E,
 268                                             vdwioffsetptr0+vdwjidx0F,
 269                                             vdwioffsetptr0+vdwjidx0G,
 270                                             vdwioffsetptr0+vdwjidx0H,
 271                                             &c6_00,&c12_00);
 272
 273             /* Calculate table index by multiplying r with table scale and truncate to integer */
 274             rt               = _mm256_mul_ps(r00,vftabscale);
 275             vfitab           = _mm256_cvttps_epi32(rt);
 276             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 277             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 278             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 279             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 280             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 281             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 282
 283             /* CUBIC SPLINE TABLE DISPERSION */
 284             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 285             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 286             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 287                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 288             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 289                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 290             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 291                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 292             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 293                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 294             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 295             Heps             = _mm256_mul_ps(vfeps,H);
 296             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 297             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 298             vvdw6            = _mm256_mul_ps(c6_00,VV);
 299             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 300             fvdw6            = _mm256_mul_ps(c6_00,FF);
 301
 302             /* CUBIC SPLINE TABLE REPULSION */
 303             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 304             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 305             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 306                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 307             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 309             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 310                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 311             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 312                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 313             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 314             Heps             = _mm256_mul_ps(vfeps,H);
 315             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 316             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 317             vvdw12           = _mm256_mul_ps(c12_00,VV);
 318             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 319             fvdw12           = _mm256_mul_ps(c12_00,FF);
 320             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 321             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 322
 323             /* Update potential sum for this i atom from the interaction with this j atom. */
 324             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 325
 326             fscal            = fvdw;
 327
 328             /* Calculate temporary vectorial force */
 329             tx               = _mm256_mul_ps(fscal,dx00);
 330             ty               = _mm256_mul_ps(fscal,dy00);
 331             tz               = _mm256_mul_ps(fscal,dz00);
 332
 333             /* Update vectorial force */
 334             fix0             = _mm256_add_ps(fix0,tx);
 335             fiy0             = _mm256_add_ps(fiy0,ty);
 336             fiz0             = _mm256_add_ps(fiz0,tz);
 337
 338             fjx0             = _mm256_add_ps(fjx0,tx);
 339             fjy0             = _mm256_add_ps(fjy0,ty);
 340             fjz0             = _mm256_add_ps(fjz0,tz);
 341
 342             /**************************
 343              * CALCULATE INTERACTIONS *
 344              **************************/
 345
 346             r10              = _mm256_mul_ps(rsq10,rinv10);
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq10             = _mm256_mul_ps(iq1,jq0);
 350
 351             /* Calculate table index by multiplying r with table scale and truncate to integer */
 352             rt               = _mm256_mul_ps(r10,vftabscale);
 353             vfitab           = _mm256_cvttps_epi32(rt);
 354             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 355             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 356             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 357             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 358             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 359             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 360
 361             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 362             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 363                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 364             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 365                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 366             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 367                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 368             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 369                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 370             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 371             Heps             = _mm256_mul_ps(vfeps,H);
 372             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 373             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 374             velec            = _mm256_mul_ps(qq10,VV);
 375             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 376             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 377
 378             /* Update potential sum for this i atom from the interaction with this j atom. */
 379             velecsum         = _mm256_add_ps(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             /* Calculate temporary vectorial force */
 384             tx               = _mm256_mul_ps(fscal,dx10);
 385             ty               = _mm256_mul_ps(fscal,dy10);
 386             tz               = _mm256_mul_ps(fscal,dz10);
 387
 388             /* Update vectorial force */
 389             fix1             = _mm256_add_ps(fix1,tx);
 390             fiy1             = _mm256_add_ps(fiy1,ty);
 391             fiz1             = _mm256_add_ps(fiz1,tz);
 392
 393             fjx0             = _mm256_add_ps(fjx0,tx);
 394             fjy0             = _mm256_add_ps(fjy0,ty);
 395             fjz0             = _mm256_add_ps(fjz0,tz);
 396
 397             /**************************
 398              * CALCULATE INTERACTIONS *
 399              **************************/
 400
 401             r20              = _mm256_mul_ps(rsq20,rinv20);
 402
 403             /* Compute parameters for interactions between i and j atoms */
 404             qq20             = _mm256_mul_ps(iq2,jq0);
 405
 406             /* Calculate table index by multiplying r with table scale and truncate to integer */
 407             rt               = _mm256_mul_ps(r20,vftabscale);
 408             vfitab           = _mm256_cvttps_epi32(rt);
 409             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 410             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 411             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 412             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 413             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 414             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 415
 416             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 417             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 418                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 419             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 420                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 421             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 422                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 423             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 424                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 425             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 426             Heps             = _mm256_mul_ps(vfeps,H);
 427             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 428             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 429             velec            = _mm256_mul_ps(qq20,VV);
 430             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 431             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 432
 433             /* Update potential sum for this i atom from the interaction with this j atom. */
 434             velecsum         = _mm256_add_ps(velecsum,velec);
 435
 436             fscal            = felec;
 437
 438             /* Calculate temporary vectorial force */
 439             tx               = _mm256_mul_ps(fscal,dx20);
 440             ty               = _mm256_mul_ps(fscal,dy20);
 441             tz               = _mm256_mul_ps(fscal,dz20);
 442
 443             /* Update vectorial force */
 444             fix2             = _mm256_add_ps(fix2,tx);
 445             fiy2             = _mm256_add_ps(fiy2,ty);
 446             fiz2             = _mm256_add_ps(fiz2,tz);
 447
 448             fjx0             = _mm256_add_ps(fjx0,tx);
 449             fjy0             = _mm256_add_ps(fjy0,ty);
 450             fjz0             = _mm256_add_ps(fjz0,tz);
 451
 452             /**************************
 453              * CALCULATE INTERACTIONS *
 454              **************************/
 455
 456             r30              = _mm256_mul_ps(rsq30,rinv30);
 457
 458             /* Compute parameters for interactions between i and j atoms */
 459             qq30             = _mm256_mul_ps(iq3,jq0);
 460
 461             /* Calculate table index by multiplying r with table scale and truncate to integer */
 462             rt               = _mm256_mul_ps(r30,vftabscale);
 463             vfitab           = _mm256_cvttps_epi32(rt);
 464             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 465             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 466             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 467             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 468             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 469             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 470
 471             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 472             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 473                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 474             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 475                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 476             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 477                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 478             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 479                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 480             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 481             Heps             = _mm256_mul_ps(vfeps,H);
 482             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 483             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 484             velec            = _mm256_mul_ps(qq30,VV);
 485             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 486             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 487
 488             /* Update potential sum for this i atom from the interaction with this j atom. */
 489             velecsum         = _mm256_add_ps(velecsum,velec);
 490
 491             fscal            = felec;
 492
 493             /* Calculate temporary vectorial force */
 494             tx               = _mm256_mul_ps(fscal,dx30);
 495             ty               = _mm256_mul_ps(fscal,dy30);
 496             tz               = _mm256_mul_ps(fscal,dz30);
 497
 498             /* Update vectorial force */
 499             fix3             = _mm256_add_ps(fix3,tx);
 500             fiy3             = _mm256_add_ps(fiy3,ty);
 501             fiz3             = _mm256_add_ps(fiz3,tz);
 502
 503             fjx0             = _mm256_add_ps(fjx0,tx);
 504             fjy0             = _mm256_add_ps(fjy0,ty);
 505             fjz0             = _mm256_add_ps(fjz0,tz);
 506
 507             fjptrA             = f+j_coord_offsetA;
 508             fjptrB             = f+j_coord_offsetB;
 509             fjptrC             = f+j_coord_offsetC;
 510             fjptrD             = f+j_coord_offsetD;
 511             fjptrE             = f+j_coord_offsetE;
 512             fjptrF             = f+j_coord_offsetF;
 513             fjptrG             = f+j_coord_offsetG;
 514             fjptrH             = f+j_coord_offsetH;
 515
 516             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 517
 518             /* Inner loop uses 188 flops */
 519         }
 520
 521         if(jidx<j_index_end)
 522         {
 523
 524             /* Get j neighbor index, and coordinate index */
 525             jnrlistA         = jjnr[jidx];
 526             jnrlistB         = jjnr[jidx+1];
 527             jnrlistC         = jjnr[jidx+2];
 528             jnrlistD         = jjnr[jidx+3];
 529             jnrlistE         = jjnr[jidx+4];
 530             jnrlistF         = jjnr[jidx+5];
 531             jnrlistG         = jjnr[jidx+6];
 532             jnrlistH         = jjnr[jidx+7];
 533             /* Sign of each element will be negative for non-real atoms.
 534              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 535              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 536              */
 537             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 538                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 539
 540             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 541             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 542             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 543             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 544             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 545             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 546             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 547             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 548             j_coord_offsetA  = DIM*jnrA;
 549             j_coord_offsetB  = DIM*jnrB;
 550             j_coord_offsetC  = DIM*jnrC;
 551             j_coord_offsetD  = DIM*jnrD;
 552             j_coord_offsetE  = DIM*jnrE;
 553             j_coord_offsetF  = DIM*jnrF;
 554             j_coord_offsetG  = DIM*jnrG;
 555             j_coord_offsetH  = DIM*jnrH;
 556
 557             /* load j atom coordinates */
 558             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 559                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 560                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 561                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 562                                                  &jx0,&jy0,&jz0);
 563
 564             /* Calculate displacement vector */
 565             dx00             = _mm256_sub_ps(ix0,jx0);
 566             dy00             = _mm256_sub_ps(iy0,jy0);
 567             dz00             = _mm256_sub_ps(iz0,jz0);
 568             dx10             = _mm256_sub_ps(ix1,jx0);
 569             dy10             = _mm256_sub_ps(iy1,jy0);
 570             dz10             = _mm256_sub_ps(iz1,jz0);
 571             dx20             = _mm256_sub_ps(ix2,jx0);
 572             dy20             = _mm256_sub_ps(iy2,jy0);
 573             dz20             = _mm256_sub_ps(iz2,jz0);
 574             dx30             = _mm256_sub_ps(ix3,jx0);
 575             dy30             = _mm256_sub_ps(iy3,jy0);
 576             dz30             = _mm256_sub_ps(iz3,jz0);
 577
 578             /* Calculate squared distance and things based on it */
 579             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 580             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 581             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 582             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 583
 584             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 585             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 586             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 587             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 588
 589             /* Load parameters for j particles */
 590             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 591                                                                  charge+jnrC+0,charge+jnrD+0,
 592                                                                  charge+jnrE+0,charge+jnrF+0,
 593                                                                  charge+jnrG+0,charge+jnrH+0);
 594             vdwjidx0A        = 2*vdwtype[jnrA+0];
 595             vdwjidx0B        = 2*vdwtype[jnrB+0];
 596             vdwjidx0C        = 2*vdwtype[jnrC+0];
 597             vdwjidx0D        = 2*vdwtype[jnrD+0];
 598             vdwjidx0E        = 2*vdwtype[jnrE+0];
 599             vdwjidx0F        = 2*vdwtype[jnrF+0];
 600             vdwjidx0G        = 2*vdwtype[jnrG+0];
 601             vdwjidx0H        = 2*vdwtype[jnrH+0];
 602
 603             fjx0             = _mm256_setzero_ps();
 604             fjy0             = _mm256_setzero_ps();
 605             fjz0             = _mm256_setzero_ps();
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             r00              = _mm256_mul_ps(rsq00,rinv00);
 612             r00              = _mm256_andnot_ps(dummy_mask,r00);
 613
 614             /* Compute parameters for interactions between i and j atoms */
 615             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 616                                             vdwioffsetptr0+vdwjidx0B,
 617                                             vdwioffsetptr0+vdwjidx0C,
 618                                             vdwioffsetptr0+vdwjidx0D,
 619                                             vdwioffsetptr0+vdwjidx0E,
 620                                             vdwioffsetptr0+vdwjidx0F,
 621                                             vdwioffsetptr0+vdwjidx0G,
 622                                             vdwioffsetptr0+vdwjidx0H,
 623                                             &c6_00,&c12_00);
 624
 625             /* Calculate table index by multiplying r with table scale and truncate to integer */
 626             rt               = _mm256_mul_ps(r00,vftabscale);
 627             vfitab           = _mm256_cvttps_epi32(rt);
 628             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 629             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 630             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 631             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 632             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 633             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 634
 635             /* CUBIC SPLINE TABLE DISPERSION */
 636             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 637             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 638             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 639                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 640             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 641                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 642             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 643                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 644             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 645                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 646             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 647             Heps             = _mm256_mul_ps(vfeps,H);
 648             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 649             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 650             vvdw6            = _mm256_mul_ps(c6_00,VV);
 651             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 652             fvdw6            = _mm256_mul_ps(c6_00,FF);
 653
 654             /* CUBIC SPLINE TABLE REPULSION */
 655             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 656             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 657             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 658                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 659             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 660                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 661             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 662                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 663             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 664                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 665             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 666             Heps             = _mm256_mul_ps(vfeps,H);
 667             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 668             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 669             vvdw12           = _mm256_mul_ps(c12_00,VV);
 670             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 671             fvdw12           = _mm256_mul_ps(c12_00,FF);
 672             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 673             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 674
 675             /* Update potential sum for this i atom from the interaction with this j atom. */
 676             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 677             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 678
 679             fscal            = fvdw;
 680
 681             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 682
 683             /* Calculate temporary vectorial force */
 684             tx               = _mm256_mul_ps(fscal,dx00);
 685             ty               = _mm256_mul_ps(fscal,dy00);
 686             tz               = _mm256_mul_ps(fscal,dz00);
 687
 688             /* Update vectorial force */
 689             fix0             = _mm256_add_ps(fix0,tx);
 690             fiy0             = _mm256_add_ps(fiy0,ty);
 691             fiz0             = _mm256_add_ps(fiz0,tz);
 692
 693             fjx0             = _mm256_add_ps(fjx0,tx);
 694             fjy0             = _mm256_add_ps(fjy0,ty);
 695             fjz0             = _mm256_add_ps(fjz0,tz);
 696
 697             /**************************
 698              * CALCULATE INTERACTIONS *
 699              **************************/
 700
 701             r10              = _mm256_mul_ps(rsq10,rinv10);
 702             r10              = _mm256_andnot_ps(dummy_mask,r10);
 703
 704             /* Compute parameters for interactions between i and j atoms */
 705             qq10             = _mm256_mul_ps(iq1,jq0);
 706
 707             /* Calculate table index by multiplying r with table scale and truncate to integer */
 708             rt               = _mm256_mul_ps(r10,vftabscale);
 709             vfitab           = _mm256_cvttps_epi32(rt);
 710             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 711             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 712             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 713             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 714             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 715             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 716
 717             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 718             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 719                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 720             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 721                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 722             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 723                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 724             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 725                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 726             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 727             Heps             = _mm256_mul_ps(vfeps,H);
 728             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 729             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 730             velec            = _mm256_mul_ps(qq10,VV);
 731             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 732             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 733
 734             /* Update potential sum for this i atom from the interaction with this j atom. */
 735             velec            = _mm256_andnot_ps(dummy_mask,velec);
 736             velecsum         = _mm256_add_ps(velecsum,velec);
 737
 738             fscal            = felec;
 739
 740             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 741
 742             /* Calculate temporary vectorial force */
 743             tx               = _mm256_mul_ps(fscal,dx10);
 744             ty               = _mm256_mul_ps(fscal,dy10);
 745             tz               = _mm256_mul_ps(fscal,dz10);
 746
 747             /* Update vectorial force */
 748             fix1             = _mm256_add_ps(fix1,tx);
 749             fiy1             = _mm256_add_ps(fiy1,ty);
 750             fiz1             = _mm256_add_ps(fiz1,tz);
 751
 752             fjx0             = _mm256_add_ps(fjx0,tx);
 753             fjy0             = _mm256_add_ps(fjy0,ty);
 754             fjz0             = _mm256_add_ps(fjz0,tz);
 755
 756             /**************************
 757              * CALCULATE INTERACTIONS *
 758              **************************/
 759
 760             r20              = _mm256_mul_ps(rsq20,rinv20);
 761             r20              = _mm256_andnot_ps(dummy_mask,r20);
 762
 763             /* Compute parameters for interactions between i and j atoms */
 764             qq20             = _mm256_mul_ps(iq2,jq0);
 765
 766             /* Calculate table index by multiplying r with table scale and truncate to integer */
 767             rt               = _mm256_mul_ps(r20,vftabscale);
 768             vfitab           = _mm256_cvttps_epi32(rt);
 769             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 770             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 771             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 772             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 773             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 774             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 775
 776             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 777             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 778                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 779             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 780                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 781             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 782                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 783             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 784                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 785             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 786             Heps             = _mm256_mul_ps(vfeps,H);
 787             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 788             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 789             velec            = _mm256_mul_ps(qq20,VV);
 790             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 791             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 792
 793             /* Update potential sum for this i atom from the interaction with this j atom. */
 794             velec            = _mm256_andnot_ps(dummy_mask,velec);
 795             velecsum         = _mm256_add_ps(velecsum,velec);
 796
 797             fscal            = felec;
 798
 799             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 800
 801             /* Calculate temporary vectorial force */
 802             tx               = _mm256_mul_ps(fscal,dx20);
 803             ty               = _mm256_mul_ps(fscal,dy20);
 804             tz               = _mm256_mul_ps(fscal,dz20);
 805
 806             /* Update vectorial force */
 807             fix2             = _mm256_add_ps(fix2,tx);
 808             fiy2             = _mm256_add_ps(fiy2,ty);
 809             fiz2             = _mm256_add_ps(fiz2,tz);
 810
 811             fjx0             = _mm256_add_ps(fjx0,tx);
 812             fjy0             = _mm256_add_ps(fjy0,ty);
 813             fjz0             = _mm256_add_ps(fjz0,tz);
 814
 815             /**************************
 816              * CALCULATE INTERACTIONS *
 817              **************************/
 818
 819             r30              = _mm256_mul_ps(rsq30,rinv30);
 820             r30              = _mm256_andnot_ps(dummy_mask,r30);
 821
 822             /* Compute parameters for interactions between i and j atoms */
 823             qq30             = _mm256_mul_ps(iq3,jq0);
 824
 825             /* Calculate table index by multiplying r with table scale and truncate to integer */
 826             rt               = _mm256_mul_ps(r30,vftabscale);
 827             vfitab           = _mm256_cvttps_epi32(rt);
 828             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 829             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 830             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 831             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 832             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 833             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 834
 835             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 836             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 837                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 838             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 839                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 840             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 841                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 842             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 843                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 844             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 845             Heps             = _mm256_mul_ps(vfeps,H);
 846             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 847             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 848             velec            = _mm256_mul_ps(qq30,VV);
 849             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 850             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 851
 852             /* Update potential sum for this i atom from the interaction with this j atom. */
 853             velec            = _mm256_andnot_ps(dummy_mask,velec);
 854             velecsum         = _mm256_add_ps(velecsum,velec);
 855
 856             fscal            = felec;
 857
 858             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 859
 860             /* Calculate temporary vectorial force */
 861             tx               = _mm256_mul_ps(fscal,dx30);
 862             ty               = _mm256_mul_ps(fscal,dy30);
 863             tz               = _mm256_mul_ps(fscal,dz30);
 864
 865             /* Update vectorial force */
 866             fix3             = _mm256_add_ps(fix3,tx);
 867             fiy3             = _mm256_add_ps(fiy3,ty);
 868             fiz3             = _mm256_add_ps(fiz3,tz);
 869
 870             fjx0             = _mm256_add_ps(fjx0,tx);
 871             fjy0             = _mm256_add_ps(fjy0,ty);
 872             fjz0             = _mm256_add_ps(fjz0,tz);
 873
 874             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 875             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 876             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 877             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 878             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 879             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 880             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 881             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 882
 883             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 884
 885             /* Inner loop uses 192 flops */
 886         }
 887
 888         /* End of innermost loop */
 889
 890         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 891                                                  f+i_coord_offset,fshift+i_shift_offset);
 892
 893         ggid                        = gid[iidx];
 894         /* Update potential energies */
 895         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 896         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 897
 898         /* Increment number of inner iterations */
 899         inneriter                  += j_index_end - j_index_start;
 900
 901         /* Outer loop uses 26 flops */
 902     }
 903
 904     /* Increment number of outer iterations */
 905     outeriter        += nri;
 906
 907     /* Update outer/inner flops */
 908
 909     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*192);
 910 }
 911 /*
 912  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_256_single
 913  * Electrostatics interaction: CubicSplineTable
 914  * VdW interaction:            CubicSplineTable
 915  * Geometry:                   Water4-Particle
 916  * Calculate force/pot:        Force
 917  */
 918 void
 919 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_256_single
 920                     (t_nblist * gmx_restrict                nlist,
 921                      rvec * gmx_restrict                    xx,
 922                      rvec * gmx_restrict                    ff,
 923                      t_forcerec * gmx_restrict              fr,
 924                      t_mdatoms * gmx_restrict               mdatoms,
 925                      nb_kernel_data_t * gmx_restrict        kernel_data,
 926                      t_nrnb * gmx_restrict                  nrnb)
 927 {
 928     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 929      * just 0 for non-waters.
 930      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 931      * jnr indices corresponding to data put in the four positions in the SIMD register.
 932      */
 933     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 934     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 935     int              jnrA,jnrB,jnrC,jnrD;
 936     int              jnrE,jnrF,jnrG,jnrH;
 937     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 938     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 939     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 940     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 941     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 942     real             rcutoff_scalar;
 943     real             *shiftvec,*fshift,*x,*f;
 944     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 945     real             scratch[4*DIM];
 946     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 947     real *           vdwioffsetptr0;
 948     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 949     real *           vdwioffsetptr1;
 950     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 951     real *           vdwioffsetptr2;
 952     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 953     real *           vdwioffsetptr3;
 954     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 955     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 956     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 957     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 958     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 959     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 960     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 961     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 962     real             *charge;
 963     int              nvdwtype;
 964     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 965     int              *vdwtype;
 966     real             *vdwparam;
 967     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 968     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 969     __m256i          vfitab;
 970     __m128i          vfitab_lo,vfitab_hi;
 971     __m128i          ifour       = _mm_set1_epi32(4);
 972     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 973     real             *vftab;
 974     __m256           dummy_mask,cutoff_mask;
 975     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 976     __m256           one     = _mm256_set1_ps(1.0);
 977     __m256           two     = _mm256_set1_ps(2.0);
 978     x                = xx[0];
 979     f                = ff[0];
 980
 981     nri              = nlist->nri;
 982     iinr             = nlist->iinr;
 983     jindex           = nlist->jindex;
 984     jjnr             = nlist->jjnr;
 985     shiftidx         = nlist->shift;
 986     gid              = nlist->gid;
 987     shiftvec         = fr->shift_vec[0];
 988     fshift           = fr->fshift[0];
 989     facel            = _mm256_set1_ps(fr->epsfac);
 990     charge           = mdatoms->chargeA;
 991     nvdwtype         = fr->ntype;
 992     vdwparam         = fr->nbfp;
 993     vdwtype          = mdatoms->typeA;
 994
 995     vftab            = kernel_data->table_elec_vdw->data;
 996     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 997
 998     /* Setup water-specific parameters */
 999     inr              = nlist->iinr[0];
1000     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1001     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1002     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1003     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1004
1005     /* Avoid stupid compiler warnings */
1006     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1007     j_coord_offsetA = 0;
1008     j_coord_offsetB = 0;
1009     j_coord_offsetC = 0;
1010     j_coord_offsetD = 0;
1011     j_coord_offsetE = 0;
1012     j_coord_offsetF = 0;
1013     j_coord_offsetG = 0;
1014     j_coord_offsetH = 0;
1015
1016     outeriter        = 0;
1017     inneriter        = 0;
1018
1019     for(iidx=0;iidx<4*DIM;iidx++)
1020     {
1021         scratch[iidx] = 0.0;
1022     }
1023
1024     /* Start outer loop over neighborlists */
1025     for(iidx=0; iidx<nri; iidx++)
1026     {
1027         /* Load shift vector for this list */
1028         i_shift_offset   = DIM*shiftidx[iidx];
1029
1030         /* Load limits for loop over neighbors */
1031         j_index_start    = jindex[iidx];
1032         j_index_end      = jindex[iidx+1];
1033
1034         /* Get outer coordinate index */
1035         inr              = iinr[iidx];
1036         i_coord_offset   = DIM*inr;
1037
1038         /* Load i particle coords and add shift vector */
1039         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1040                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1041
1042         fix0             = _mm256_setzero_ps();
1043         fiy0             = _mm256_setzero_ps();
1044         fiz0             = _mm256_setzero_ps();
1045         fix1             = _mm256_setzero_ps();
1046         fiy1             = _mm256_setzero_ps();
1047         fiz1             = _mm256_setzero_ps();
1048         fix2             = _mm256_setzero_ps();
1049         fiy2             = _mm256_setzero_ps();
1050         fiz2             = _mm256_setzero_ps();
1051         fix3             = _mm256_setzero_ps();
1052         fiy3             = _mm256_setzero_ps();
1053         fiz3             = _mm256_setzero_ps();
1054
1055         /* Start inner kernel loop */
1056         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1057         {
1058
1059             /* Get j neighbor index, and coordinate index */
1060             jnrA             = jjnr[jidx];
1061             jnrB             = jjnr[jidx+1];
1062             jnrC             = jjnr[jidx+2];
1063             jnrD             = jjnr[jidx+3];
1064             jnrE             = jjnr[jidx+4];
1065             jnrF             = jjnr[jidx+5];
1066             jnrG             = jjnr[jidx+6];
1067             jnrH             = jjnr[jidx+7];
1068             j_coord_offsetA  = DIM*jnrA;
1069             j_coord_offsetB  = DIM*jnrB;
1070             j_coord_offsetC  = DIM*jnrC;
1071             j_coord_offsetD  = DIM*jnrD;
1072             j_coord_offsetE  = DIM*jnrE;
1073             j_coord_offsetF  = DIM*jnrF;
1074             j_coord_offsetG  = DIM*jnrG;
1075             j_coord_offsetH  = DIM*jnrH;
1076
1077             /* load j atom coordinates */
1078             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1079                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1080                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1081                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1082                                                  &jx0,&jy0,&jz0);
1083
1084             /* Calculate displacement vector */
1085             dx00             = _mm256_sub_ps(ix0,jx0);
1086             dy00             = _mm256_sub_ps(iy0,jy0);
1087             dz00             = _mm256_sub_ps(iz0,jz0);
1088             dx10             = _mm256_sub_ps(ix1,jx0);
1089             dy10             = _mm256_sub_ps(iy1,jy0);
1090             dz10             = _mm256_sub_ps(iz1,jz0);
1091             dx20             = _mm256_sub_ps(ix2,jx0);
1092             dy20             = _mm256_sub_ps(iy2,jy0);
1093             dz20             = _mm256_sub_ps(iz2,jz0);
1094             dx30             = _mm256_sub_ps(ix3,jx0);
1095             dy30             = _mm256_sub_ps(iy3,jy0);
1096             dz30             = _mm256_sub_ps(iz3,jz0);
1097
1098             /* Calculate squared distance and things based on it */
1099             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1100             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1101             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1102             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1103
1104             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1105             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1106             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1107             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1108
1109             /* Load parameters for j particles */
1110             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1111                                                                  charge+jnrC+0,charge+jnrD+0,
1112                                                                  charge+jnrE+0,charge+jnrF+0,
1113                                                                  charge+jnrG+0,charge+jnrH+0);
1114             vdwjidx0A        = 2*vdwtype[jnrA+0];
1115             vdwjidx0B        = 2*vdwtype[jnrB+0];
1116             vdwjidx0C        = 2*vdwtype[jnrC+0];
1117             vdwjidx0D        = 2*vdwtype[jnrD+0];
1118             vdwjidx0E        = 2*vdwtype[jnrE+0];
1119             vdwjidx0F        = 2*vdwtype[jnrF+0];
1120             vdwjidx0G        = 2*vdwtype[jnrG+0];
1121             vdwjidx0H        = 2*vdwtype[jnrH+0];
1122
1123             fjx0             = _mm256_setzero_ps();
1124             fjy0             = _mm256_setzero_ps();
1125             fjz0             = _mm256_setzero_ps();
1126
1127             /**************************
1128              * CALCULATE INTERACTIONS *
1129              **************************/
1130
1131             r00              = _mm256_mul_ps(rsq00,rinv00);
1132
1133             /* Compute parameters for interactions between i and j atoms */
1134             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1135                                             vdwioffsetptr0+vdwjidx0B,
1136                                             vdwioffsetptr0+vdwjidx0C,
1137                                             vdwioffsetptr0+vdwjidx0D,
1138                                             vdwioffsetptr0+vdwjidx0E,
1139                                             vdwioffsetptr0+vdwjidx0F,
1140                                             vdwioffsetptr0+vdwjidx0G,
1141                                             vdwioffsetptr0+vdwjidx0H,
1142                                             &c6_00,&c12_00);
1143
1144             /* Calculate table index by multiplying r with table scale and truncate to integer */
1145             rt               = _mm256_mul_ps(r00,vftabscale);
1146             vfitab           = _mm256_cvttps_epi32(rt);
1147             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1148             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1149             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1150             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1151             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1152             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1153
1154             /* CUBIC SPLINE TABLE DISPERSION */
1155             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1156             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1157             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1158                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1159             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1160                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1161             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1162                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1163             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1164                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1165             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1166             Heps             = _mm256_mul_ps(vfeps,H);
1167             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1168             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1169             fvdw6            = _mm256_mul_ps(c6_00,FF);
1170
1171             /* CUBIC SPLINE TABLE REPULSION */
1172             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1173             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1174             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1175                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1176             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1177                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1178             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1179                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1180             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1181                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1182             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1183             Heps             = _mm256_mul_ps(vfeps,H);
1184             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1185             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1186             fvdw12           = _mm256_mul_ps(c12_00,FF);
1187             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1188
1189             fscal            = fvdw;
1190
1191             /* Calculate temporary vectorial force */
1192             tx               = _mm256_mul_ps(fscal,dx00);
1193             ty               = _mm256_mul_ps(fscal,dy00);
1194             tz               = _mm256_mul_ps(fscal,dz00);
1195
1196             /* Update vectorial force */
1197             fix0             = _mm256_add_ps(fix0,tx);
1198             fiy0             = _mm256_add_ps(fiy0,ty);
1199             fiz0             = _mm256_add_ps(fiz0,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             r10              = _mm256_mul_ps(rsq10,rinv10);
1210
1211             /* Compute parameters for interactions between i and j atoms */
1212             qq10             = _mm256_mul_ps(iq1,jq0);
1213
1214             /* Calculate table index by multiplying r with table scale and truncate to integer */
1215             rt               = _mm256_mul_ps(r10,vftabscale);
1216             vfitab           = _mm256_cvttps_epi32(rt);
1217             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1218             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1219             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1220             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1221             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1222             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1223
1224             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1225             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1226                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1227             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1228                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1229             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1230                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1231             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1232                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1233             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1234             Heps             = _mm256_mul_ps(vfeps,H);
1235             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1236             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1237             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1238
1239             fscal            = felec;
1240
1241             /* Calculate temporary vectorial force */
1242             tx               = _mm256_mul_ps(fscal,dx10);
1243             ty               = _mm256_mul_ps(fscal,dy10);
1244             tz               = _mm256_mul_ps(fscal,dz10);
1245
1246             /* Update vectorial force */
1247             fix1             = _mm256_add_ps(fix1,tx);
1248             fiy1             = _mm256_add_ps(fiy1,ty);
1249             fiz1             = _mm256_add_ps(fiz1,tz);
1250
1251             fjx0             = _mm256_add_ps(fjx0,tx);
1252             fjy0             = _mm256_add_ps(fjy0,ty);
1253             fjz0             = _mm256_add_ps(fjz0,tz);
1254
1255             /**************************
1256              * CALCULATE INTERACTIONS *
1257              **************************/
1258
1259             r20              = _mm256_mul_ps(rsq20,rinv20);
1260
1261             /* Compute parameters for interactions between i and j atoms */
1262             qq20             = _mm256_mul_ps(iq2,jq0);
1263
1264             /* Calculate table index by multiplying r with table scale and truncate to integer */
1265             rt               = _mm256_mul_ps(r20,vftabscale);
1266             vfitab           = _mm256_cvttps_epi32(rt);
1267             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1268             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1269             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1270             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1271             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1272             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1273
1274             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1275             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1276                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1277             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1278                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1279             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1280                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1281             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1282                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1283             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1284             Heps             = _mm256_mul_ps(vfeps,H);
1285             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1286             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1287             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1288
1289             fscal            = felec;
1290
1291             /* Calculate temporary vectorial force */
1292             tx               = _mm256_mul_ps(fscal,dx20);
1293             ty               = _mm256_mul_ps(fscal,dy20);
1294             tz               = _mm256_mul_ps(fscal,dz20);
1295
1296             /* Update vectorial force */
1297             fix2             = _mm256_add_ps(fix2,tx);
1298             fiy2             = _mm256_add_ps(fiy2,ty);
1299             fiz2             = _mm256_add_ps(fiz2,tz);
1300
1301             fjx0             = _mm256_add_ps(fjx0,tx);
1302             fjy0             = _mm256_add_ps(fjy0,ty);
1303             fjz0             = _mm256_add_ps(fjz0,tz);
1304
1305             /**************************
1306              * CALCULATE INTERACTIONS *
1307              **************************/
1308
1309             r30              = _mm256_mul_ps(rsq30,rinv30);
1310
1311             /* Compute parameters for interactions between i and j atoms */
1312             qq30             = _mm256_mul_ps(iq3,jq0);
1313
1314             /* Calculate table index by multiplying r with table scale and truncate to integer */
1315             rt               = _mm256_mul_ps(r30,vftabscale);
1316             vfitab           = _mm256_cvttps_epi32(rt);
1317             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1318             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1319             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1320             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1321             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1322             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1323
1324             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1325             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1326                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1327             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1328                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1329             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1330                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1331             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1332                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1333             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1334             Heps             = _mm256_mul_ps(vfeps,H);
1335             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1336             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1337             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1338
1339             fscal            = felec;
1340
1341             /* Calculate temporary vectorial force */
1342             tx               = _mm256_mul_ps(fscal,dx30);
1343             ty               = _mm256_mul_ps(fscal,dy30);
1344             tz               = _mm256_mul_ps(fscal,dz30);
1345
1346             /* Update vectorial force */
1347             fix3             = _mm256_add_ps(fix3,tx);
1348             fiy3             = _mm256_add_ps(fiy3,ty);
1349             fiz3             = _mm256_add_ps(fiz3,tz);
1350
1351             fjx0             = _mm256_add_ps(fjx0,tx);
1352             fjy0             = _mm256_add_ps(fjy0,ty);
1353             fjz0             = _mm256_add_ps(fjz0,tz);
1354
1355             fjptrA             = f+j_coord_offsetA;
1356             fjptrB             = f+j_coord_offsetB;
1357             fjptrC             = f+j_coord_offsetC;
1358             fjptrD             = f+j_coord_offsetD;
1359             fjptrE             = f+j_coord_offsetE;
1360             fjptrF             = f+j_coord_offsetF;
1361             fjptrG             = f+j_coord_offsetG;
1362             fjptrH             = f+j_coord_offsetH;
1363
1364             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1365
1366             /* Inner loop uses 168 flops */
1367         }
1368
1369         if(jidx<j_index_end)
1370         {
1371
1372             /* Get j neighbor index, and coordinate index */
1373             jnrlistA         = jjnr[jidx];
1374             jnrlistB         = jjnr[jidx+1];
1375             jnrlistC         = jjnr[jidx+2];
1376             jnrlistD         = jjnr[jidx+3];
1377             jnrlistE         = jjnr[jidx+4];
1378             jnrlistF         = jjnr[jidx+5];
1379             jnrlistG         = jjnr[jidx+6];
1380             jnrlistH         = jjnr[jidx+7];
1381             /* Sign of each element will be negative for non-real atoms.
1382              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1383              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1384              */
1385             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1386                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1387
1388             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1389             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1390             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1391             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1392             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1393             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1394             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1395             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1396             j_coord_offsetA  = DIM*jnrA;
1397             j_coord_offsetB  = DIM*jnrB;
1398             j_coord_offsetC  = DIM*jnrC;
1399             j_coord_offsetD  = DIM*jnrD;
1400             j_coord_offsetE  = DIM*jnrE;
1401             j_coord_offsetF  = DIM*jnrF;
1402             j_coord_offsetG  = DIM*jnrG;
1403             j_coord_offsetH  = DIM*jnrH;
1404
1405             /* load j atom coordinates */
1406             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1407                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1408                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1409                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1410                                                  &jx0,&jy0,&jz0);
1411
1412             /* Calculate displacement vector */
1413             dx00             = _mm256_sub_ps(ix0,jx0);
1414             dy00             = _mm256_sub_ps(iy0,jy0);
1415             dz00             = _mm256_sub_ps(iz0,jz0);
1416             dx10             = _mm256_sub_ps(ix1,jx0);
1417             dy10             = _mm256_sub_ps(iy1,jy0);
1418             dz10             = _mm256_sub_ps(iz1,jz0);
1419             dx20             = _mm256_sub_ps(ix2,jx0);
1420             dy20             = _mm256_sub_ps(iy2,jy0);
1421             dz20             = _mm256_sub_ps(iz2,jz0);
1422             dx30             = _mm256_sub_ps(ix3,jx0);
1423             dy30             = _mm256_sub_ps(iy3,jy0);
1424             dz30             = _mm256_sub_ps(iz3,jz0);
1425
1426             /* Calculate squared distance and things based on it */
1427             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1428             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1429             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1430             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1431
1432             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1433             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1434             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1435             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1436
1437             /* Load parameters for j particles */
1438             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1439                                                                  charge+jnrC+0,charge+jnrD+0,
1440                                                                  charge+jnrE+0,charge+jnrF+0,
1441                                                                  charge+jnrG+0,charge+jnrH+0);
1442             vdwjidx0A        = 2*vdwtype[jnrA+0];
1443             vdwjidx0B        = 2*vdwtype[jnrB+0];
1444             vdwjidx0C        = 2*vdwtype[jnrC+0];
1445             vdwjidx0D        = 2*vdwtype[jnrD+0];
1446             vdwjidx0E        = 2*vdwtype[jnrE+0];
1447             vdwjidx0F        = 2*vdwtype[jnrF+0];
1448             vdwjidx0G        = 2*vdwtype[jnrG+0];
1449             vdwjidx0H        = 2*vdwtype[jnrH+0];
1450
1451             fjx0             = _mm256_setzero_ps();
1452             fjy0             = _mm256_setzero_ps();
1453             fjz0             = _mm256_setzero_ps();
1454
1455             /**************************
1456              * CALCULATE INTERACTIONS *
1457              **************************/
1458
1459             r00              = _mm256_mul_ps(rsq00,rinv00);
1460             r00              = _mm256_andnot_ps(dummy_mask,r00);
1461
1462             /* Compute parameters for interactions between i and j atoms */
1463             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1464                                             vdwioffsetptr0+vdwjidx0B,
1465                                             vdwioffsetptr0+vdwjidx0C,
1466                                             vdwioffsetptr0+vdwjidx0D,
1467                                             vdwioffsetptr0+vdwjidx0E,
1468                                             vdwioffsetptr0+vdwjidx0F,
1469                                             vdwioffsetptr0+vdwjidx0G,
1470                                             vdwioffsetptr0+vdwjidx0H,
1471                                             &c6_00,&c12_00);
1472
1473             /* Calculate table index by multiplying r with table scale and truncate to integer */
1474             rt               = _mm256_mul_ps(r00,vftabscale);
1475             vfitab           = _mm256_cvttps_epi32(rt);
1476             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1477             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1478             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1479             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1480             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1481             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1482
1483             /* CUBIC SPLINE TABLE DISPERSION */
1484             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1485             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1486             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1487                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1488             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1489                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1490             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1491                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1492             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1493                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1494             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1495             Heps             = _mm256_mul_ps(vfeps,H);
1496             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1497             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1498             fvdw6            = _mm256_mul_ps(c6_00,FF);
1499
1500             /* CUBIC SPLINE TABLE REPULSION */
1501             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1502             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1503             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1504                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1505             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1506                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1507             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1508                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1509             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1510                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1511             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1512             Heps             = _mm256_mul_ps(vfeps,H);
1513             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1514             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1515             fvdw12           = _mm256_mul_ps(c12_00,FF);
1516             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1517
1518             fscal            = fvdw;
1519
1520             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1521
1522             /* Calculate temporary vectorial force */
1523             tx               = _mm256_mul_ps(fscal,dx00);
1524             ty               = _mm256_mul_ps(fscal,dy00);
1525             tz               = _mm256_mul_ps(fscal,dz00);
1526
1527             /* Update vectorial force */
1528             fix0             = _mm256_add_ps(fix0,tx);
1529             fiy0             = _mm256_add_ps(fiy0,ty);
1530             fiz0             = _mm256_add_ps(fiz0,tz);
1531
1532             fjx0             = _mm256_add_ps(fjx0,tx);
1533             fjy0             = _mm256_add_ps(fjy0,ty);
1534             fjz0             = _mm256_add_ps(fjz0,tz);
1535
1536             /**************************
1537              * CALCULATE INTERACTIONS *
1538              **************************/
1539
1540             r10              = _mm256_mul_ps(rsq10,rinv10);
1541             r10              = _mm256_andnot_ps(dummy_mask,r10);
1542
1543             /* Compute parameters for interactions between i and j atoms */
1544             qq10             = _mm256_mul_ps(iq1,jq0);
1545
1546             /* Calculate table index by multiplying r with table scale and truncate to integer */
1547             rt               = _mm256_mul_ps(r10,vftabscale);
1548             vfitab           = _mm256_cvttps_epi32(rt);
1549             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1550             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1551             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1552             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1553             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1554             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1555
1556             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1557             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1558                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1559             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1560                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1561             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1562                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1563             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1564                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1565             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1566             Heps             = _mm256_mul_ps(vfeps,H);
1567             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1568             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1569             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1570
1571             fscal            = felec;
1572
1573             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1574
1575             /* Calculate temporary vectorial force */
1576             tx               = _mm256_mul_ps(fscal,dx10);
1577             ty               = _mm256_mul_ps(fscal,dy10);
1578             tz               = _mm256_mul_ps(fscal,dz10);
1579
1580             /* Update vectorial force */
1581             fix1             = _mm256_add_ps(fix1,tx);
1582             fiy1             = _mm256_add_ps(fiy1,ty);
1583             fiz1             = _mm256_add_ps(fiz1,tz);
1584
1585             fjx0             = _mm256_add_ps(fjx0,tx);
1586             fjy0             = _mm256_add_ps(fjy0,ty);
1587             fjz0             = _mm256_add_ps(fjz0,tz);
1588
1589             /**************************
1590              * CALCULATE INTERACTIONS *
1591              **************************/
1592
1593             r20              = _mm256_mul_ps(rsq20,rinv20);
1594             r20              = _mm256_andnot_ps(dummy_mask,r20);
1595
1596             /* Compute parameters for interactions between i and j atoms */
1597             qq20             = _mm256_mul_ps(iq2,jq0);
1598
1599             /* Calculate table index by multiplying r with table scale and truncate to integer */
1600             rt               = _mm256_mul_ps(r20,vftabscale);
1601             vfitab           = _mm256_cvttps_epi32(rt);
1602             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1603             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1604             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1605             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1606             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1607             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1608
1609             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1610             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1611                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1612             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1613                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1614             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1615                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1616             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1617                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1618             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1619             Heps             = _mm256_mul_ps(vfeps,H);
1620             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1621             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1622             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1623
1624             fscal            = felec;
1625
1626             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1627
1628             /* Calculate temporary vectorial force */
1629             tx               = _mm256_mul_ps(fscal,dx20);
1630             ty               = _mm256_mul_ps(fscal,dy20);
1631             tz               = _mm256_mul_ps(fscal,dz20);
1632
1633             /* Update vectorial force */
1634             fix2             = _mm256_add_ps(fix2,tx);
1635             fiy2             = _mm256_add_ps(fiy2,ty);
1636             fiz2             = _mm256_add_ps(fiz2,tz);
1637
1638             fjx0             = _mm256_add_ps(fjx0,tx);
1639             fjy0             = _mm256_add_ps(fjy0,ty);
1640             fjz0             = _mm256_add_ps(fjz0,tz);
1641
1642             /**************************
1643              * CALCULATE INTERACTIONS *
1644              **************************/
1645
1646             r30              = _mm256_mul_ps(rsq30,rinv30);
1647             r30              = _mm256_andnot_ps(dummy_mask,r30);
1648
1649             /* Compute parameters for interactions between i and j atoms */
1650             qq30             = _mm256_mul_ps(iq3,jq0);
1651
1652             /* Calculate table index by multiplying r with table scale and truncate to integer */
1653             rt               = _mm256_mul_ps(r30,vftabscale);
1654             vfitab           = _mm256_cvttps_epi32(rt);
1655             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1656             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1657             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1658             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1659             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1660             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1661
1662             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1663             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1664                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1665             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1666                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1667             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1668                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1669             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1670                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1671             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1672             Heps             = _mm256_mul_ps(vfeps,H);
1673             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1674             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1675             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1676
1677             fscal            = felec;
1678
1679             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1680
1681             /* Calculate temporary vectorial force */
1682             tx               = _mm256_mul_ps(fscal,dx30);
1683             ty               = _mm256_mul_ps(fscal,dy30);
1684             tz               = _mm256_mul_ps(fscal,dz30);
1685
1686             /* Update vectorial force */
1687             fix3             = _mm256_add_ps(fix3,tx);
1688             fiy3             = _mm256_add_ps(fiy3,ty);
1689             fiz3             = _mm256_add_ps(fiz3,tz);
1690
1691             fjx0             = _mm256_add_ps(fjx0,tx);
1692             fjy0             = _mm256_add_ps(fjy0,ty);
1693             fjz0             = _mm256_add_ps(fjz0,tz);
1694
1695             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1696             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1697             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1698             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1699             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1700             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1701             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1702             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1703
1704             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1705
1706             /* Inner loop uses 172 flops */
1707         }
1708
1709         /* End of innermost loop */
1710
1711         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1712                                                  f+i_coord_offset,fshift+i_shift_offset);
1713
1714         /* Increment number of inner iterations */
1715         inneriter                  += j_index_end - j_index_start;
1716
1717         /* Outer loop uses 24 flops */
1718     }
1719
1720     /* Increment number of outer iterations */
1721     outeriter        += nri;
1722
1723     /* Update outer/inner flops */
1724
1725     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*172);
1726 }