src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_VdwCSTab_GeomW3P1_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double 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_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_256_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_256_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  77     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  78     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  79     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  80     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  81     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  82     real             *charge;
  83     int              nvdwtype;
  84     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  85     int              *vdwtype;
  86     real             *vdwparam;
  87     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  88     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  89     __m128i          vfitab;
  90     __m128i          ifour       = _mm_set1_epi32(4);
  91     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  92     real             *vftab;
  93     __m256d          dummy_mask,cutoff_mask;
  94     __m128           tmpmask0,tmpmask1;
  95     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  96     __m256d          one     = _mm256_set1_pd(1.0);
  97     __m256d          two     = _mm256_set1_pd(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = _mm256_set1_pd(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111     krf              = _mm256_set1_pd(fr->ic->k_rf);
 112     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 113     crf              = _mm256_set1_pd(fr->ic->c_rf);
 114     nvdwtype         = fr->ntype;
 115     vdwparam         = fr->nbfp;
 116     vdwtype          = mdatoms->typeA;
 117
 118     vftab            = kernel_data->table_vdw->data;
 119     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 120
 121     /* Setup water-specific parameters */
 122     inr              = nlist->iinr[0];
 123     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 124     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 125     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 126     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = jnrC = jnrD = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132     j_coord_offsetC = 0;
 133     j_coord_offsetD = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     for(iidx=0;iidx<4*DIM;iidx++)
 139     {
 140         scratch[iidx] = 0.0;
 141     }
 142
 143     /* Start outer loop over neighborlists */
 144     for(iidx=0; iidx<nri; iidx++)
 145     {
 146         /* Load shift vector for this list */
 147         i_shift_offset   = DIM*shiftidx[iidx];
 148
 149         /* Load limits for loop over neighbors */
 150         j_index_start    = jindex[iidx];
 151         j_index_end      = jindex[iidx+1];
 152
 153         /* Get outer coordinate index */
 154         inr              = iinr[iidx];
 155         i_coord_offset   = DIM*inr;
 156
 157         /* Load i particle coords and add shift vector */
 158         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 159                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 160
 161         fix0             = _mm256_setzero_pd();
 162         fiy0             = _mm256_setzero_pd();
 163         fiz0             = _mm256_setzero_pd();
 164         fix1             = _mm256_setzero_pd();
 165         fiy1             = _mm256_setzero_pd();
 166         fiz1             = _mm256_setzero_pd();
 167         fix2             = _mm256_setzero_pd();
 168         fiy2             = _mm256_setzero_pd();
 169         fiz2             = _mm256_setzero_pd();
 170
 171         /* Reset potential sums */
 172         velecsum         = _mm256_setzero_pd();
 173         vvdwsum          = _mm256_setzero_pd();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             jnrC             = jjnr[jidx+2];
 183             jnrD             = jjnr[jidx+3];
 184             j_coord_offsetA  = DIM*jnrA;
 185             j_coord_offsetB  = DIM*jnrB;
 186             j_coord_offsetC  = DIM*jnrC;
 187             j_coord_offsetD  = DIM*jnrD;
 188
 189             /* load j atom coordinates */
 190             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 192                                                  &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm256_sub_pd(ix0,jx0);
 196             dy00             = _mm256_sub_pd(iy0,jy0);
 197             dz00             = _mm256_sub_pd(iz0,jz0);
 198             dx10             = _mm256_sub_pd(ix1,jx0);
 199             dy10             = _mm256_sub_pd(iy1,jy0);
 200             dz10             = _mm256_sub_pd(iz1,jz0);
 201             dx20             = _mm256_sub_pd(ix2,jx0);
 202             dy20             = _mm256_sub_pd(iy2,jy0);
 203             dz20             = _mm256_sub_pd(iz2,jz0);
 204
 205             /* Calculate squared distance and things based on it */
 206             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 207             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 208             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 209
 210             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 211             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 212             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 213
 214             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 215             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 216             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 217
 218             /* Load parameters for j particles */
 219             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 220                                                                  charge+jnrC+0,charge+jnrD+0);
 221             vdwjidx0A        = 2*vdwtype[jnrA+0];
 222             vdwjidx0B        = 2*vdwtype[jnrB+0];
 223             vdwjidx0C        = 2*vdwtype[jnrC+0];
 224             vdwjidx0D        = 2*vdwtype[jnrD+0];
 225
 226             fjx0             = _mm256_setzero_pd();
 227             fjy0             = _mm256_setzero_pd();
 228             fjz0             = _mm256_setzero_pd();
 229
 230             /**************************
 231              * CALCULATE INTERACTIONS *
 232              **************************/
 233
 234             r00              = _mm256_mul_pd(rsq00,rinv00);
 235
 236             /* Compute parameters for interactions between i and j atoms */
 237             qq00             = _mm256_mul_pd(iq0,jq0);
 238             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 239                                             vdwioffsetptr0+vdwjidx0B,
 240                                             vdwioffsetptr0+vdwjidx0C,
 241                                             vdwioffsetptr0+vdwjidx0D,
 242                                             &c6_00,&c12_00);
 243
 244             /* Calculate table index by multiplying r with table scale and truncate to integer */
 245             rt               = _mm256_mul_pd(r00,vftabscale);
 246             vfitab           = _mm256_cvttpd_epi32(rt);
 247             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 248             vfitab           = _mm_slli_epi32(vfitab,3);
 249
 250             /* REACTION-FIELD ELECTROSTATICS */
 251             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 252             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 253
 254             /* CUBIC SPLINE TABLE DISPERSION */
 255             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 256             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 257             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 258             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 259             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 260             Heps             = _mm256_mul_pd(vfeps,H);
 261             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 262             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 263             vvdw6            = _mm256_mul_pd(c6_00,VV);
 264             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 265             fvdw6            = _mm256_mul_pd(c6_00,FF);
 266
 267             /* CUBIC SPLINE TABLE REPULSION */
 268             vfitab           = _mm_add_epi32(vfitab,ifour);
 269             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 270             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 271             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 272             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 273             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 274             Heps             = _mm256_mul_pd(vfeps,H);
 275             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 276             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 277             vvdw12           = _mm256_mul_pd(c12_00,VV);
 278             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 279             fvdw12           = _mm256_mul_pd(c12_00,FF);
 280             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 281             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velecsum         = _mm256_add_pd(velecsum,velec);
 285             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 286
 287             fscal            = _mm256_add_pd(felec,fvdw);
 288
 289             /* Calculate temporary vectorial force */
 290             tx               = _mm256_mul_pd(fscal,dx00);
 291             ty               = _mm256_mul_pd(fscal,dy00);
 292             tz               = _mm256_mul_pd(fscal,dz00);
 293
 294             /* Update vectorial force */
 295             fix0             = _mm256_add_pd(fix0,tx);
 296             fiy0             = _mm256_add_pd(fiy0,ty);
 297             fiz0             = _mm256_add_pd(fiz0,tz);
 298
 299             fjx0             = _mm256_add_pd(fjx0,tx);
 300             fjy0             = _mm256_add_pd(fjy0,ty);
 301             fjz0             = _mm256_add_pd(fjz0,tz);
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq10             = _mm256_mul_pd(iq1,jq0);
 309
 310             /* REACTION-FIELD ELECTROSTATICS */
 311             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 312             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velecsum         = _mm256_add_pd(velecsum,velec);
 316
 317             fscal            = felec;
 318
 319             /* Calculate temporary vectorial force */
 320             tx               = _mm256_mul_pd(fscal,dx10);
 321             ty               = _mm256_mul_pd(fscal,dy10);
 322             tz               = _mm256_mul_pd(fscal,dz10);
 323
 324             /* Update vectorial force */
 325             fix1             = _mm256_add_pd(fix1,tx);
 326             fiy1             = _mm256_add_pd(fiy1,ty);
 327             fiz1             = _mm256_add_pd(fiz1,tz);
 328
 329             fjx0             = _mm256_add_pd(fjx0,tx);
 330             fjy0             = _mm256_add_pd(fjy0,ty);
 331             fjz0             = _mm256_add_pd(fjz0,tz);
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq20             = _mm256_mul_pd(iq2,jq0);
 339
 340             /* REACTION-FIELD ELECTROSTATICS */
 341             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 342             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm256_add_pd(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm256_mul_pd(fscal,dx20);
 351             ty               = _mm256_mul_pd(fscal,dy20);
 352             tz               = _mm256_mul_pd(fscal,dz20);
 353
 354             /* Update vectorial force */
 355             fix2             = _mm256_add_pd(fix2,tx);
 356             fiy2             = _mm256_add_pd(fiy2,ty);
 357             fiz2             = _mm256_add_pd(fiz2,tz);
 358
 359             fjx0             = _mm256_add_pd(fjx0,tx);
 360             fjy0             = _mm256_add_pd(fjy0,ty);
 361             fjz0             = _mm256_add_pd(fjz0,tz);
 362
 363             fjptrA             = f+j_coord_offsetA;
 364             fjptrB             = f+j_coord_offsetB;
 365             fjptrC             = f+j_coord_offsetC;
 366             fjptrD             = f+j_coord_offsetD;
 367
 368             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 369
 370             /* Inner loop uses 134 flops */
 371         }
 372
 373         if(jidx<j_index_end)
 374         {
 375
 376             /* Get j neighbor index, and coordinate index */
 377             jnrlistA         = jjnr[jidx];
 378             jnrlistB         = jjnr[jidx+1];
 379             jnrlistC         = jjnr[jidx+2];
 380             jnrlistD         = jjnr[jidx+3];
 381             /* Sign of each element will be negative for non-real atoms.
 382              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 383              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 384              */
 385             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 386
 387             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 388             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 389             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 390
 391             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 392             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 393             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 394             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 395             j_coord_offsetA  = DIM*jnrA;
 396             j_coord_offsetB  = DIM*jnrB;
 397             j_coord_offsetC  = DIM*jnrC;
 398             j_coord_offsetD  = DIM*jnrD;
 399
 400             /* load j atom coordinates */
 401             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 402                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 403                                                  &jx0,&jy0,&jz0);
 404
 405             /* Calculate displacement vector */
 406             dx00             = _mm256_sub_pd(ix0,jx0);
 407             dy00             = _mm256_sub_pd(iy0,jy0);
 408             dz00             = _mm256_sub_pd(iz0,jz0);
 409             dx10             = _mm256_sub_pd(ix1,jx0);
 410             dy10             = _mm256_sub_pd(iy1,jy0);
 411             dz10             = _mm256_sub_pd(iz1,jz0);
 412             dx20             = _mm256_sub_pd(ix2,jx0);
 413             dy20             = _mm256_sub_pd(iy2,jy0);
 414             dz20             = _mm256_sub_pd(iz2,jz0);
 415
 416             /* Calculate squared distance and things based on it */
 417             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 418             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 419             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 420
 421             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 422             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 423             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 424
 425             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 426             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 427             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 428
 429             /* Load parameters for j particles */
 430             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 431                                                                  charge+jnrC+0,charge+jnrD+0);
 432             vdwjidx0A        = 2*vdwtype[jnrA+0];
 433             vdwjidx0B        = 2*vdwtype[jnrB+0];
 434             vdwjidx0C        = 2*vdwtype[jnrC+0];
 435             vdwjidx0D        = 2*vdwtype[jnrD+0];
 436
 437             fjx0             = _mm256_setzero_pd();
 438             fjy0             = _mm256_setzero_pd();
 439             fjz0             = _mm256_setzero_pd();
 440
 441             /**************************
 442              * CALCULATE INTERACTIONS *
 443              **************************/
 444
 445             r00              = _mm256_mul_pd(rsq00,rinv00);
 446             r00              = _mm256_andnot_pd(dummy_mask,r00);
 447
 448             /* Compute parameters for interactions between i and j atoms */
 449             qq00             = _mm256_mul_pd(iq0,jq0);
 450             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 451                                             vdwioffsetptr0+vdwjidx0B,
 452                                             vdwioffsetptr0+vdwjidx0C,
 453                                             vdwioffsetptr0+vdwjidx0D,
 454                                             &c6_00,&c12_00);
 455
 456             /* Calculate table index by multiplying r with table scale and truncate to integer */
 457             rt               = _mm256_mul_pd(r00,vftabscale);
 458             vfitab           = _mm256_cvttpd_epi32(rt);
 459             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 460             vfitab           = _mm_slli_epi32(vfitab,3);
 461
 462             /* REACTION-FIELD ELECTROSTATICS */
 463             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 464             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 465
 466             /* CUBIC SPLINE TABLE DISPERSION */
 467             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 468             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 469             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 470             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 471             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 472             Heps             = _mm256_mul_pd(vfeps,H);
 473             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 474             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 475             vvdw6            = _mm256_mul_pd(c6_00,VV);
 476             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 477             fvdw6            = _mm256_mul_pd(c6_00,FF);
 478
 479             /* CUBIC SPLINE TABLE REPULSION */
 480             vfitab           = _mm_add_epi32(vfitab,ifour);
 481             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 482             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 483             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 484             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 485             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 486             Heps             = _mm256_mul_pd(vfeps,H);
 487             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 488             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 489             vvdw12           = _mm256_mul_pd(c12_00,VV);
 490             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 491             fvdw12           = _mm256_mul_pd(c12_00,FF);
 492             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 493             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 494
 495             /* Update potential sum for this i atom from the interaction with this j atom. */
 496             velec            = _mm256_andnot_pd(dummy_mask,velec);
 497             velecsum         = _mm256_add_pd(velecsum,velec);
 498             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 499             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 500
 501             fscal            = _mm256_add_pd(felec,fvdw);
 502
 503             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 504
 505             /* Calculate temporary vectorial force */
 506             tx               = _mm256_mul_pd(fscal,dx00);
 507             ty               = _mm256_mul_pd(fscal,dy00);
 508             tz               = _mm256_mul_pd(fscal,dz00);
 509
 510             /* Update vectorial force */
 511             fix0             = _mm256_add_pd(fix0,tx);
 512             fiy0             = _mm256_add_pd(fiy0,ty);
 513             fiz0             = _mm256_add_pd(fiz0,tz);
 514
 515             fjx0             = _mm256_add_pd(fjx0,tx);
 516             fjy0             = _mm256_add_pd(fjy0,ty);
 517             fjz0             = _mm256_add_pd(fjz0,tz);
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             /* Compute parameters for interactions between i and j atoms */
 524             qq10             = _mm256_mul_pd(iq1,jq0);
 525
 526             /* REACTION-FIELD ELECTROSTATICS */
 527             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 528             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             velec            = _mm256_andnot_pd(dummy_mask,velec);
 532             velecsum         = _mm256_add_pd(velecsum,velec);
 533
 534             fscal            = felec;
 535
 536             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 537
 538             /* Calculate temporary vectorial force */
 539             tx               = _mm256_mul_pd(fscal,dx10);
 540             ty               = _mm256_mul_pd(fscal,dy10);
 541             tz               = _mm256_mul_pd(fscal,dz10);
 542
 543             /* Update vectorial force */
 544             fix1             = _mm256_add_pd(fix1,tx);
 545             fiy1             = _mm256_add_pd(fiy1,ty);
 546             fiz1             = _mm256_add_pd(fiz1,tz);
 547
 548             fjx0             = _mm256_add_pd(fjx0,tx);
 549             fjy0             = _mm256_add_pd(fjy0,ty);
 550             fjz0             = _mm256_add_pd(fjz0,tz);
 551
 552             /**************************
 553              * CALCULATE INTERACTIONS *
 554              **************************/
 555
 556             /* Compute parameters for interactions between i and j atoms */
 557             qq20             = _mm256_mul_pd(iq2,jq0);
 558
 559             /* REACTION-FIELD ELECTROSTATICS */
 560             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 561             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 562
 563             /* Update potential sum for this i atom from the interaction with this j atom. */
 564             velec            = _mm256_andnot_pd(dummy_mask,velec);
 565             velecsum         = _mm256_add_pd(velecsum,velec);
 566
 567             fscal            = felec;
 568
 569             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 570
 571             /* Calculate temporary vectorial force */
 572             tx               = _mm256_mul_pd(fscal,dx20);
 573             ty               = _mm256_mul_pd(fscal,dy20);
 574             tz               = _mm256_mul_pd(fscal,dz20);
 575
 576             /* Update vectorial force */
 577             fix2             = _mm256_add_pd(fix2,tx);
 578             fiy2             = _mm256_add_pd(fiy2,ty);
 579             fiz2             = _mm256_add_pd(fiz2,tz);
 580
 581             fjx0             = _mm256_add_pd(fjx0,tx);
 582             fjy0             = _mm256_add_pd(fjy0,ty);
 583             fjz0             = _mm256_add_pd(fjz0,tz);
 584
 585             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 586             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 587             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 588             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 589
 590             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 591
 592             /* Inner loop uses 135 flops */
 593         }
 594
 595         /* End of innermost loop */
 596
 597         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 598                                                  f+i_coord_offset,fshift+i_shift_offset);
 599
 600         ggid                        = gid[iidx];
 601         /* Update potential energies */
 602         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 603         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 604
 605         /* Increment number of inner iterations */
 606         inneriter                  += j_index_end - j_index_start;
 607
 608         /* Outer loop uses 20 flops */
 609     }
 610
 611     /* Increment number of outer iterations */
 612     outeriter        += nri;
 613
 614     /* Update outer/inner flops */
 615
 616     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*135);
 617 }
 618 /*
 619  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_256_double
 620  * Electrostatics interaction: ReactionField
 621  * VdW interaction:            CubicSplineTable
 622  * Geometry:                   Water3-Particle
 623  * Calculate force/pot:        Force
 624  */
 625 void
 626 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_256_double
 627                     (t_nblist * gmx_restrict                nlist,
 628                      rvec * gmx_restrict                    xx,
 629                      rvec * gmx_restrict                    ff,
 630                      t_forcerec * gmx_restrict              fr,
 631                      t_mdatoms * gmx_restrict               mdatoms,
 632                      nb_kernel_data_t * gmx_restrict        kernel_data,
 633                      t_nrnb * gmx_restrict                  nrnb)
 634 {
 635     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 636      * just 0 for non-waters.
 637      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 638      * jnr indices corresponding to data put in the four positions in the SIMD register.
 639      */
 640     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 641     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 642     int              jnrA,jnrB,jnrC,jnrD;
 643     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 644     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 645     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 646     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 647     real             rcutoff_scalar;
 648     real             *shiftvec,*fshift,*x,*f;
 649     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 650     real             scratch[4*DIM];
 651     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 652     real *           vdwioffsetptr0;
 653     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 654     real *           vdwioffsetptr1;
 655     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 656     real *           vdwioffsetptr2;
 657     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 658     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 659     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 660     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 661     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 662     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 663     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 664     real             *charge;
 665     int              nvdwtype;
 666     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 667     int              *vdwtype;
 668     real             *vdwparam;
 669     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 670     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 671     __m128i          vfitab;
 672     __m128i          ifour       = _mm_set1_epi32(4);
 673     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 674     real             *vftab;
 675     __m256d          dummy_mask,cutoff_mask;
 676     __m128           tmpmask0,tmpmask1;
 677     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 678     __m256d          one     = _mm256_set1_pd(1.0);
 679     __m256d          two     = _mm256_set1_pd(2.0);
 680     x                = xx[0];
 681     f                = ff[0];
 682
 683     nri              = nlist->nri;
 684     iinr             = nlist->iinr;
 685     jindex           = nlist->jindex;
 686     jjnr             = nlist->jjnr;
 687     shiftidx         = nlist->shift;
 688     gid              = nlist->gid;
 689     shiftvec         = fr->shift_vec[0];
 690     fshift           = fr->fshift[0];
 691     facel            = _mm256_set1_pd(fr->epsfac);
 692     charge           = mdatoms->chargeA;
 693     krf              = _mm256_set1_pd(fr->ic->k_rf);
 694     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 695     crf              = _mm256_set1_pd(fr->ic->c_rf);
 696     nvdwtype         = fr->ntype;
 697     vdwparam         = fr->nbfp;
 698     vdwtype          = mdatoms->typeA;
 699
 700     vftab            = kernel_data->table_vdw->data;
 701     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 702
 703     /* Setup water-specific parameters */
 704     inr              = nlist->iinr[0];
 705     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 706     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 707     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 708     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 709
 710     /* Avoid stupid compiler warnings */
 711     jnrA = jnrB = jnrC = jnrD = 0;
 712     j_coord_offsetA = 0;
 713     j_coord_offsetB = 0;
 714     j_coord_offsetC = 0;
 715     j_coord_offsetD = 0;
 716
 717     outeriter        = 0;
 718     inneriter        = 0;
 719
 720     for(iidx=0;iidx<4*DIM;iidx++)
 721     {
 722         scratch[iidx] = 0.0;
 723     }
 724
 725     /* Start outer loop over neighborlists */
 726     for(iidx=0; iidx<nri; iidx++)
 727     {
 728         /* Load shift vector for this list */
 729         i_shift_offset   = DIM*shiftidx[iidx];
 730
 731         /* Load limits for loop over neighbors */
 732         j_index_start    = jindex[iidx];
 733         j_index_end      = jindex[iidx+1];
 734
 735         /* Get outer coordinate index */
 736         inr              = iinr[iidx];
 737         i_coord_offset   = DIM*inr;
 738
 739         /* Load i particle coords and add shift vector */
 740         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 741                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 742
 743         fix0             = _mm256_setzero_pd();
 744         fiy0             = _mm256_setzero_pd();
 745         fiz0             = _mm256_setzero_pd();
 746         fix1             = _mm256_setzero_pd();
 747         fiy1             = _mm256_setzero_pd();
 748         fiz1             = _mm256_setzero_pd();
 749         fix2             = _mm256_setzero_pd();
 750         fiy2             = _mm256_setzero_pd();
 751         fiz2             = _mm256_setzero_pd();
 752
 753         /* Start inner kernel loop */
 754         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 755         {
 756
 757             /* Get j neighbor index, and coordinate index */
 758             jnrA             = jjnr[jidx];
 759             jnrB             = jjnr[jidx+1];
 760             jnrC             = jjnr[jidx+2];
 761             jnrD             = jjnr[jidx+3];
 762             j_coord_offsetA  = DIM*jnrA;
 763             j_coord_offsetB  = DIM*jnrB;
 764             j_coord_offsetC  = DIM*jnrC;
 765             j_coord_offsetD  = DIM*jnrD;
 766
 767             /* load j atom coordinates */
 768             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 769                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 770                                                  &jx0,&jy0,&jz0);
 771
 772             /* Calculate displacement vector */
 773             dx00             = _mm256_sub_pd(ix0,jx0);
 774             dy00             = _mm256_sub_pd(iy0,jy0);
 775             dz00             = _mm256_sub_pd(iz0,jz0);
 776             dx10             = _mm256_sub_pd(ix1,jx0);
 777             dy10             = _mm256_sub_pd(iy1,jy0);
 778             dz10             = _mm256_sub_pd(iz1,jz0);
 779             dx20             = _mm256_sub_pd(ix2,jx0);
 780             dy20             = _mm256_sub_pd(iy2,jy0);
 781             dz20             = _mm256_sub_pd(iz2,jz0);
 782
 783             /* Calculate squared distance and things based on it */
 784             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 785             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 786             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 787
 788             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 789             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 790             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 791
 792             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 793             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 794             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 795
 796             /* Load parameters for j particles */
 797             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 798                                                                  charge+jnrC+0,charge+jnrD+0);
 799             vdwjidx0A        = 2*vdwtype[jnrA+0];
 800             vdwjidx0B        = 2*vdwtype[jnrB+0];
 801             vdwjidx0C        = 2*vdwtype[jnrC+0];
 802             vdwjidx0D        = 2*vdwtype[jnrD+0];
 803
 804             fjx0             = _mm256_setzero_pd();
 805             fjy0             = _mm256_setzero_pd();
 806             fjz0             = _mm256_setzero_pd();
 807
 808             /**************************
 809              * CALCULATE INTERACTIONS *
 810              **************************/
 811
 812             r00              = _mm256_mul_pd(rsq00,rinv00);
 813
 814             /* Compute parameters for interactions between i and j atoms */
 815             qq00             = _mm256_mul_pd(iq0,jq0);
 816             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 817                                             vdwioffsetptr0+vdwjidx0B,
 818                                             vdwioffsetptr0+vdwjidx0C,
 819                                             vdwioffsetptr0+vdwjidx0D,
 820                                             &c6_00,&c12_00);
 821
 822             /* Calculate table index by multiplying r with table scale and truncate to integer */
 823             rt               = _mm256_mul_pd(r00,vftabscale);
 824             vfitab           = _mm256_cvttpd_epi32(rt);
 825             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 826             vfitab           = _mm_slli_epi32(vfitab,3);
 827
 828             /* REACTION-FIELD ELECTROSTATICS */
 829             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 830
 831             /* CUBIC SPLINE TABLE DISPERSION */
 832             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 833             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 834             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 835             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 836             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 837             Heps             = _mm256_mul_pd(vfeps,H);
 838             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 839             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 840             fvdw6            = _mm256_mul_pd(c6_00,FF);
 841
 842             /* CUBIC SPLINE TABLE REPULSION */
 843             vfitab           = _mm_add_epi32(vfitab,ifour);
 844             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 845             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 846             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 847             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 848             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 849             Heps             = _mm256_mul_pd(vfeps,H);
 850             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 851             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 852             fvdw12           = _mm256_mul_pd(c12_00,FF);
 853             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 854
 855             fscal            = _mm256_add_pd(felec,fvdw);
 856
 857             /* Calculate temporary vectorial force */
 858             tx               = _mm256_mul_pd(fscal,dx00);
 859             ty               = _mm256_mul_pd(fscal,dy00);
 860             tz               = _mm256_mul_pd(fscal,dz00);
 861
 862             /* Update vectorial force */
 863             fix0             = _mm256_add_pd(fix0,tx);
 864             fiy0             = _mm256_add_pd(fiy0,ty);
 865             fiz0             = _mm256_add_pd(fiz0,tz);
 866
 867             fjx0             = _mm256_add_pd(fjx0,tx);
 868             fjy0             = _mm256_add_pd(fjy0,ty);
 869             fjz0             = _mm256_add_pd(fjz0,tz);
 870
 871             /**************************
 872              * CALCULATE INTERACTIONS *
 873              **************************/
 874
 875             /* Compute parameters for interactions between i and j atoms */
 876             qq10             = _mm256_mul_pd(iq1,jq0);
 877
 878             /* REACTION-FIELD ELECTROSTATICS */
 879             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 880
 881             fscal            = felec;
 882
 883             /* Calculate temporary vectorial force */
 884             tx               = _mm256_mul_pd(fscal,dx10);
 885             ty               = _mm256_mul_pd(fscal,dy10);
 886             tz               = _mm256_mul_pd(fscal,dz10);
 887
 888             /* Update vectorial force */
 889             fix1             = _mm256_add_pd(fix1,tx);
 890             fiy1             = _mm256_add_pd(fiy1,ty);
 891             fiz1             = _mm256_add_pd(fiz1,tz);
 892
 893             fjx0             = _mm256_add_pd(fjx0,tx);
 894             fjy0             = _mm256_add_pd(fjy0,ty);
 895             fjz0             = _mm256_add_pd(fjz0,tz);
 896
 897             /**************************
 898              * CALCULATE INTERACTIONS *
 899              **************************/
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             qq20             = _mm256_mul_pd(iq2,jq0);
 903
 904             /* REACTION-FIELD ELECTROSTATICS */
 905             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 906
 907             fscal            = felec;
 908
 909             /* Calculate temporary vectorial force */
 910             tx               = _mm256_mul_pd(fscal,dx20);
 911             ty               = _mm256_mul_pd(fscal,dy20);
 912             tz               = _mm256_mul_pd(fscal,dz20);
 913
 914             /* Update vectorial force */
 915             fix2             = _mm256_add_pd(fix2,tx);
 916             fiy2             = _mm256_add_pd(fiy2,ty);
 917             fiz2             = _mm256_add_pd(fiz2,tz);
 918
 919             fjx0             = _mm256_add_pd(fjx0,tx);
 920             fjy0             = _mm256_add_pd(fjy0,ty);
 921             fjz0             = _mm256_add_pd(fjz0,tz);
 922
 923             fjptrA             = f+j_coord_offsetA;
 924             fjptrB             = f+j_coord_offsetB;
 925             fjptrC             = f+j_coord_offsetC;
 926             fjptrD             = f+j_coord_offsetD;
 927
 928             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 929
 930             /* Inner loop uses 111 flops */
 931         }
 932
 933         if(jidx<j_index_end)
 934         {
 935
 936             /* Get j neighbor index, and coordinate index */
 937             jnrlistA         = jjnr[jidx];
 938             jnrlistB         = jjnr[jidx+1];
 939             jnrlistC         = jjnr[jidx+2];
 940             jnrlistD         = jjnr[jidx+3];
 941             /* Sign of each element will be negative for non-real atoms.
 942              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 943              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 944              */
 945             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 946
 947             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 948             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 949             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 950
 951             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 952             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 953             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 954             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 955             j_coord_offsetA  = DIM*jnrA;
 956             j_coord_offsetB  = DIM*jnrB;
 957             j_coord_offsetC  = DIM*jnrC;
 958             j_coord_offsetD  = DIM*jnrD;
 959
 960             /* load j atom coordinates */
 961             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 962                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 963                                                  &jx0,&jy0,&jz0);
 964
 965             /* Calculate displacement vector */
 966             dx00             = _mm256_sub_pd(ix0,jx0);
 967             dy00             = _mm256_sub_pd(iy0,jy0);
 968             dz00             = _mm256_sub_pd(iz0,jz0);
 969             dx10             = _mm256_sub_pd(ix1,jx0);
 970             dy10             = _mm256_sub_pd(iy1,jy0);
 971             dz10             = _mm256_sub_pd(iz1,jz0);
 972             dx20             = _mm256_sub_pd(ix2,jx0);
 973             dy20             = _mm256_sub_pd(iy2,jy0);
 974             dz20             = _mm256_sub_pd(iz2,jz0);
 975
 976             /* Calculate squared distance and things based on it */
 977             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 978             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 979             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 980
 981             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 982             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 983             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 984
 985             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 986             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 987             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 988
 989             /* Load parameters for j particles */
 990             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 991                                                                  charge+jnrC+0,charge+jnrD+0);
 992             vdwjidx0A        = 2*vdwtype[jnrA+0];
 993             vdwjidx0B        = 2*vdwtype[jnrB+0];
 994             vdwjidx0C        = 2*vdwtype[jnrC+0];
 995             vdwjidx0D        = 2*vdwtype[jnrD+0];
 996
 997             fjx0             = _mm256_setzero_pd();
 998             fjy0             = _mm256_setzero_pd();
 999             fjz0             = _mm256_setzero_pd();
1000
1001             /**************************
1002              * CALCULATE INTERACTIONS *
1003              **************************/
1004
1005             r00              = _mm256_mul_pd(rsq00,rinv00);
1006             r00              = _mm256_andnot_pd(dummy_mask,r00);
1007
1008             /* Compute parameters for interactions between i and j atoms */
1009             qq00             = _mm256_mul_pd(iq0,jq0);
1010             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1011                                             vdwioffsetptr0+vdwjidx0B,
1012                                             vdwioffsetptr0+vdwjidx0C,
1013                                             vdwioffsetptr0+vdwjidx0D,
1014                                             &c6_00,&c12_00);
1015
1016             /* Calculate table index by multiplying r with table scale and truncate to integer */
1017             rt               = _mm256_mul_pd(r00,vftabscale);
1018             vfitab           = _mm256_cvttpd_epi32(rt);
1019             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1020             vfitab           = _mm_slli_epi32(vfitab,3);
1021
1022             /* REACTION-FIELD ELECTROSTATICS */
1023             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1024
1025             /* CUBIC SPLINE TABLE DISPERSION */
1026             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1027             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1028             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1029             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1030             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1031             Heps             = _mm256_mul_pd(vfeps,H);
1032             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1033             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1034             fvdw6            = _mm256_mul_pd(c6_00,FF);
1035
1036             /* CUBIC SPLINE TABLE REPULSION */
1037             vfitab           = _mm_add_epi32(vfitab,ifour);
1038             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1039             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1040             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1041             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1042             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1043             Heps             = _mm256_mul_pd(vfeps,H);
1044             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1045             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1046             fvdw12           = _mm256_mul_pd(c12_00,FF);
1047             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1048
1049             fscal            = _mm256_add_pd(felec,fvdw);
1050
1051             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1052
1053             /* Calculate temporary vectorial force */
1054             tx               = _mm256_mul_pd(fscal,dx00);
1055             ty               = _mm256_mul_pd(fscal,dy00);
1056             tz               = _mm256_mul_pd(fscal,dz00);
1057
1058             /* Update vectorial force */
1059             fix0             = _mm256_add_pd(fix0,tx);
1060             fiy0             = _mm256_add_pd(fiy0,ty);
1061             fiz0             = _mm256_add_pd(fiz0,tz);
1062
1063             fjx0             = _mm256_add_pd(fjx0,tx);
1064             fjy0             = _mm256_add_pd(fjy0,ty);
1065             fjz0             = _mm256_add_pd(fjz0,tz);
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             /* Compute parameters for interactions between i and j atoms */
1072             qq10             = _mm256_mul_pd(iq1,jq0);
1073
1074             /* REACTION-FIELD ELECTROSTATICS */
1075             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1076
1077             fscal            = felec;
1078
1079             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1080
1081             /* Calculate temporary vectorial force */
1082             tx               = _mm256_mul_pd(fscal,dx10);
1083             ty               = _mm256_mul_pd(fscal,dy10);
1084             tz               = _mm256_mul_pd(fscal,dz10);
1085
1086             /* Update vectorial force */
1087             fix1             = _mm256_add_pd(fix1,tx);
1088             fiy1             = _mm256_add_pd(fiy1,ty);
1089             fiz1             = _mm256_add_pd(fiz1,tz);
1090
1091             fjx0             = _mm256_add_pd(fjx0,tx);
1092             fjy0             = _mm256_add_pd(fjy0,ty);
1093             fjz0             = _mm256_add_pd(fjz0,tz);
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             /* Compute parameters for interactions between i and j atoms */
1100             qq20             = _mm256_mul_pd(iq2,jq0);
1101
1102             /* REACTION-FIELD ELECTROSTATICS */
1103             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1104
1105             fscal            = felec;
1106
1107             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1108
1109             /* Calculate temporary vectorial force */
1110             tx               = _mm256_mul_pd(fscal,dx20);
1111             ty               = _mm256_mul_pd(fscal,dy20);
1112             tz               = _mm256_mul_pd(fscal,dz20);
1113
1114             /* Update vectorial force */
1115             fix2             = _mm256_add_pd(fix2,tx);
1116             fiy2             = _mm256_add_pd(fiy2,ty);
1117             fiz2             = _mm256_add_pd(fiz2,tz);
1118
1119             fjx0             = _mm256_add_pd(fjx0,tx);
1120             fjy0             = _mm256_add_pd(fjy0,ty);
1121             fjz0             = _mm256_add_pd(fjz0,tz);
1122
1123             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1124             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1125             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1126             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1127
1128             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1129
1130             /* Inner loop uses 112 flops */
1131         }
1132
1133         /* End of innermost loop */
1134
1135         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1136                                                  f+i_coord_offset,fshift+i_shift_offset);
1137
1138         /* Increment number of inner iterations */
1139         inneriter                  += j_index_end - j_index_start;
1140
1141         /* Outer loop uses 18 flops */
1142     }
1143
1144     /* Increment number of outer iterations */
1145     outeriter        += nri;
1146
1147     /* Update outer/inner flops */
1148
1149     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*112);
1150 }