src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwLJ_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_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_256_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_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     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     vftab            = kernel_data->table_elec->data;
 116     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 117
 118     /* Setup water-specific parameters */
 119     inr              = nlist->iinr[0];
 120     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 121     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 122     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 123     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 124
 125     /* Avoid stupid compiler warnings */
 126     jnrA = jnrB = jnrC = jnrD = 0;
 127     j_coord_offsetA = 0;
 128     j_coord_offsetB = 0;
 129     j_coord_offsetC = 0;
 130     j_coord_offsetD = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     for(iidx=0;iidx<4*DIM;iidx++)
 136     {
 137         scratch[iidx] = 0.0;
 138     }
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 156                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 157
 158         fix0             = _mm256_setzero_pd();
 159         fiy0             = _mm256_setzero_pd();
 160         fiz0             = _mm256_setzero_pd();
 161         fix1             = _mm256_setzero_pd();
 162         fiy1             = _mm256_setzero_pd();
 163         fiz1             = _mm256_setzero_pd();
 164         fix2             = _mm256_setzero_pd();
 165         fiy2             = _mm256_setzero_pd();
 166         fiz2             = _mm256_setzero_pd();
 167
 168         /* Reset potential sums */
 169         velecsum         = _mm256_setzero_pd();
 170         vvdwsum          = _mm256_setzero_pd();
 171
 172         /* Start inner kernel loop */
 173         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 174         {
 175
 176             /* Get j neighbor index, and coordinate index */
 177             jnrA             = jjnr[jidx];
 178             jnrB             = jjnr[jidx+1];
 179             jnrC             = jjnr[jidx+2];
 180             jnrD             = jjnr[jidx+3];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183             j_coord_offsetC  = DIM*jnrC;
 184             j_coord_offsetD  = DIM*jnrD;
 185
 186             /* load j atom coordinates */
 187             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 189                                                  &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx00             = _mm256_sub_pd(ix0,jx0);
 193             dy00             = _mm256_sub_pd(iy0,jy0);
 194             dz00             = _mm256_sub_pd(iz0,jz0);
 195             dx10             = _mm256_sub_pd(ix1,jx0);
 196             dy10             = _mm256_sub_pd(iy1,jy0);
 197             dz10             = _mm256_sub_pd(iz1,jz0);
 198             dx20             = _mm256_sub_pd(ix2,jx0);
 199             dy20             = _mm256_sub_pd(iy2,jy0);
 200             dz20             = _mm256_sub_pd(iz2,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 204             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 205             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 206
 207             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 208             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 209             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 210
 211             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 212
 213             /* Load parameters for j particles */
 214             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 215                                                                  charge+jnrC+0,charge+jnrD+0);
 216             vdwjidx0A        = 2*vdwtype[jnrA+0];
 217             vdwjidx0B        = 2*vdwtype[jnrB+0];
 218             vdwjidx0C        = 2*vdwtype[jnrC+0];
 219             vdwjidx0D        = 2*vdwtype[jnrD+0];
 220
 221             fjx0             = _mm256_setzero_pd();
 222             fjy0             = _mm256_setzero_pd();
 223             fjz0             = _mm256_setzero_pd();
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             r00              = _mm256_mul_pd(rsq00,rinv00);
 230
 231             /* Compute parameters for interactions between i and j atoms */
 232             qq00             = _mm256_mul_pd(iq0,jq0);
 233             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 234                                             vdwioffsetptr0+vdwjidx0B,
 235                                             vdwioffsetptr0+vdwjidx0C,
 236                                             vdwioffsetptr0+vdwjidx0D,
 237                                             &c6_00,&c12_00);
 238
 239             /* Calculate table index by multiplying r with table scale and truncate to integer */
 240             rt               = _mm256_mul_pd(r00,vftabscale);
 241             vfitab           = _mm256_cvttpd_epi32(rt);
 242             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 243             vfitab           = _mm_slli_epi32(vfitab,2);
 244
 245             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 246             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 247             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 248             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 249             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 250             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 251             Heps             = _mm256_mul_pd(vfeps,H);
 252             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 253             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 254             velec            = _mm256_mul_pd(qq00,VV);
 255             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 256             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 257
 258             /* LENNARD-JONES DISPERSION/REPULSION */
 259
 260             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 261             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 262             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 263             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 264             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 265
 266             /* Update potential sum for this i atom from the interaction with this j atom. */
 267             velecsum         = _mm256_add_pd(velecsum,velec);
 268             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 269
 270             fscal            = _mm256_add_pd(felec,fvdw);
 271
 272             /* Calculate temporary vectorial force */
 273             tx               = _mm256_mul_pd(fscal,dx00);
 274             ty               = _mm256_mul_pd(fscal,dy00);
 275             tz               = _mm256_mul_pd(fscal,dz00);
 276
 277             /* Update vectorial force */
 278             fix0             = _mm256_add_pd(fix0,tx);
 279             fiy0             = _mm256_add_pd(fiy0,ty);
 280             fiz0             = _mm256_add_pd(fiz0,tz);
 281
 282             fjx0             = _mm256_add_pd(fjx0,tx);
 283             fjy0             = _mm256_add_pd(fjy0,ty);
 284             fjz0             = _mm256_add_pd(fjz0,tz);
 285
 286             /**************************
 287              * CALCULATE INTERACTIONS *
 288              **************************/
 289
 290             r10              = _mm256_mul_pd(rsq10,rinv10);
 291
 292             /* Compute parameters for interactions between i and j atoms */
 293             qq10             = _mm256_mul_pd(iq1,jq0);
 294
 295             /* Calculate table index by multiplying r with table scale and truncate to integer */
 296             rt               = _mm256_mul_pd(r10,vftabscale);
 297             vfitab           = _mm256_cvttpd_epi32(rt);
 298             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 299             vfitab           = _mm_slli_epi32(vfitab,2);
 300
 301             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 302             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 303             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 304             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 305             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 306             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 307             Heps             = _mm256_mul_pd(vfeps,H);
 308             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 309             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 310             velec            = _mm256_mul_pd(qq10,VV);
 311             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 312             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 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             r20              = _mm256_mul_pd(rsq20,rinv20);
 338
 339             /* Compute parameters for interactions between i and j atoms */
 340             qq20             = _mm256_mul_pd(iq2,jq0);
 341
 342             /* Calculate table index by multiplying r with table scale and truncate to integer */
 343             rt               = _mm256_mul_pd(r20,vftabscale);
 344             vfitab           = _mm256_cvttpd_epi32(rt);
 345             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 346             vfitab           = _mm_slli_epi32(vfitab,2);
 347
 348             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 349             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 350             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 351             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 352             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 353             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 354             Heps             = _mm256_mul_pd(vfeps,H);
 355             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 356             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 357             velec            = _mm256_mul_pd(qq20,VV);
 358             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 359             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velecsum         = _mm256_add_pd(velecsum,velec);
 363
 364             fscal            = felec;
 365
 366             /* Calculate temporary vectorial force */
 367             tx               = _mm256_mul_pd(fscal,dx20);
 368             ty               = _mm256_mul_pd(fscal,dy20);
 369             tz               = _mm256_mul_pd(fscal,dz20);
 370
 371             /* Update vectorial force */
 372             fix2             = _mm256_add_pd(fix2,tx);
 373             fiy2             = _mm256_add_pd(fiy2,ty);
 374             fiz2             = _mm256_add_pd(fiz2,tz);
 375
 376             fjx0             = _mm256_add_pd(fjx0,tx);
 377             fjy0             = _mm256_add_pd(fjy0,ty);
 378             fjz0             = _mm256_add_pd(fjz0,tz);
 379
 380             fjptrA             = f+j_coord_offsetA;
 381             fjptrB             = f+j_coord_offsetB;
 382             fjptrC             = f+j_coord_offsetC;
 383             fjptrD             = f+j_coord_offsetD;
 384
 385             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 386
 387             /* Inner loop uses 145 flops */
 388         }
 389
 390         if(jidx<j_index_end)
 391         {
 392
 393             /* Get j neighbor index, and coordinate index */
 394             jnrlistA         = jjnr[jidx];
 395             jnrlistB         = jjnr[jidx+1];
 396             jnrlistC         = jjnr[jidx+2];
 397             jnrlistD         = jjnr[jidx+3];
 398             /* Sign of each element will be negative for non-real atoms.
 399              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 400              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 401              */
 402             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 403
 404             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 405             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 406             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 407
 408             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 409             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 410             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 411             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 412             j_coord_offsetA  = DIM*jnrA;
 413             j_coord_offsetB  = DIM*jnrB;
 414             j_coord_offsetC  = DIM*jnrC;
 415             j_coord_offsetD  = DIM*jnrD;
 416
 417             /* load j atom coordinates */
 418             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 419                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 420                                                  &jx0,&jy0,&jz0);
 421
 422             /* Calculate displacement vector */
 423             dx00             = _mm256_sub_pd(ix0,jx0);
 424             dy00             = _mm256_sub_pd(iy0,jy0);
 425             dz00             = _mm256_sub_pd(iz0,jz0);
 426             dx10             = _mm256_sub_pd(ix1,jx0);
 427             dy10             = _mm256_sub_pd(iy1,jy0);
 428             dz10             = _mm256_sub_pd(iz1,jz0);
 429             dx20             = _mm256_sub_pd(ix2,jx0);
 430             dy20             = _mm256_sub_pd(iy2,jy0);
 431             dz20             = _mm256_sub_pd(iz2,jz0);
 432
 433             /* Calculate squared distance and things based on it */
 434             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 435             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 436             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 437
 438             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 439             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 440             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 441
 442             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 443
 444             /* Load parameters for j particles */
 445             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 446                                                                  charge+jnrC+0,charge+jnrD+0);
 447             vdwjidx0A        = 2*vdwtype[jnrA+0];
 448             vdwjidx0B        = 2*vdwtype[jnrB+0];
 449             vdwjidx0C        = 2*vdwtype[jnrC+0];
 450             vdwjidx0D        = 2*vdwtype[jnrD+0];
 451
 452             fjx0             = _mm256_setzero_pd();
 453             fjy0             = _mm256_setzero_pd();
 454             fjz0             = _mm256_setzero_pd();
 455
 456             /**************************
 457              * CALCULATE INTERACTIONS *
 458              **************************/
 459
 460             r00              = _mm256_mul_pd(rsq00,rinv00);
 461             r00              = _mm256_andnot_pd(dummy_mask,r00);
 462
 463             /* Compute parameters for interactions between i and j atoms */
 464             qq00             = _mm256_mul_pd(iq0,jq0);
 465             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 466                                             vdwioffsetptr0+vdwjidx0B,
 467                                             vdwioffsetptr0+vdwjidx0C,
 468                                             vdwioffsetptr0+vdwjidx0D,
 469                                             &c6_00,&c12_00);
 470
 471             /* Calculate table index by multiplying r with table scale and truncate to integer */
 472             rt               = _mm256_mul_pd(r00,vftabscale);
 473             vfitab           = _mm256_cvttpd_epi32(rt);
 474             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 475             vfitab           = _mm_slli_epi32(vfitab,2);
 476
 477             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 478             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 479             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 480             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 481             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 482             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 483             Heps             = _mm256_mul_pd(vfeps,H);
 484             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 485             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 486             velec            = _mm256_mul_pd(qq00,VV);
 487             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 488             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 489
 490             /* LENNARD-JONES DISPERSION/REPULSION */
 491
 492             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 493             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 494             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 495             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 496             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 497
 498             /* Update potential sum for this i atom from the interaction with this j atom. */
 499             velec            = _mm256_andnot_pd(dummy_mask,velec);
 500             velecsum         = _mm256_add_pd(velecsum,velec);
 501             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 502             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 503
 504             fscal            = _mm256_add_pd(felec,fvdw);
 505
 506             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 507
 508             /* Calculate temporary vectorial force */
 509             tx               = _mm256_mul_pd(fscal,dx00);
 510             ty               = _mm256_mul_pd(fscal,dy00);
 511             tz               = _mm256_mul_pd(fscal,dz00);
 512
 513             /* Update vectorial force */
 514             fix0             = _mm256_add_pd(fix0,tx);
 515             fiy0             = _mm256_add_pd(fiy0,ty);
 516             fiz0             = _mm256_add_pd(fiz0,tz);
 517
 518             fjx0             = _mm256_add_pd(fjx0,tx);
 519             fjy0             = _mm256_add_pd(fjy0,ty);
 520             fjz0             = _mm256_add_pd(fjz0,tz);
 521
 522             /**************************
 523              * CALCULATE INTERACTIONS *
 524              **************************/
 525
 526             r10              = _mm256_mul_pd(rsq10,rinv10);
 527             r10              = _mm256_andnot_pd(dummy_mask,r10);
 528
 529             /* Compute parameters for interactions between i and j atoms */
 530             qq10             = _mm256_mul_pd(iq1,jq0);
 531
 532             /* Calculate table index by multiplying r with table scale and truncate to integer */
 533             rt               = _mm256_mul_pd(r10,vftabscale);
 534             vfitab           = _mm256_cvttpd_epi32(rt);
 535             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 536             vfitab           = _mm_slli_epi32(vfitab,2);
 537
 538             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 539             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 540             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 541             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 542             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 543             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 544             Heps             = _mm256_mul_pd(vfeps,H);
 545             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 546             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 547             velec            = _mm256_mul_pd(qq10,VV);
 548             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 549             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             velec            = _mm256_andnot_pd(dummy_mask,velec);
 553             velecsum         = _mm256_add_pd(velecsum,velec);
 554
 555             fscal            = felec;
 556
 557             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm256_mul_pd(fscal,dx10);
 561             ty               = _mm256_mul_pd(fscal,dy10);
 562             tz               = _mm256_mul_pd(fscal,dz10);
 563
 564             /* Update vectorial force */
 565             fix1             = _mm256_add_pd(fix1,tx);
 566             fiy1             = _mm256_add_pd(fiy1,ty);
 567             fiz1             = _mm256_add_pd(fiz1,tz);
 568
 569             fjx0             = _mm256_add_pd(fjx0,tx);
 570             fjy0             = _mm256_add_pd(fjy0,ty);
 571             fjz0             = _mm256_add_pd(fjz0,tz);
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             r20              = _mm256_mul_pd(rsq20,rinv20);
 578             r20              = _mm256_andnot_pd(dummy_mask,r20);
 579
 580             /* Compute parameters for interactions between i and j atoms */
 581             qq20             = _mm256_mul_pd(iq2,jq0);
 582
 583             /* Calculate table index by multiplying r with table scale and truncate to integer */
 584             rt               = _mm256_mul_pd(r20,vftabscale);
 585             vfitab           = _mm256_cvttpd_epi32(rt);
 586             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 587             vfitab           = _mm_slli_epi32(vfitab,2);
 588
 589             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 590             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 591             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 592             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 593             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 594             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 595             Heps             = _mm256_mul_pd(vfeps,H);
 596             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 597             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 598             velec            = _mm256_mul_pd(qq20,VV);
 599             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 600             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 601
 602             /* Update potential sum for this i atom from the interaction with this j atom. */
 603             velec            = _mm256_andnot_pd(dummy_mask,velec);
 604             velecsum         = _mm256_add_pd(velecsum,velec);
 605
 606             fscal            = felec;
 607
 608             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 609
 610             /* Calculate temporary vectorial force */
 611             tx               = _mm256_mul_pd(fscal,dx20);
 612             ty               = _mm256_mul_pd(fscal,dy20);
 613             tz               = _mm256_mul_pd(fscal,dz20);
 614
 615             /* Update vectorial force */
 616             fix2             = _mm256_add_pd(fix2,tx);
 617             fiy2             = _mm256_add_pd(fiy2,ty);
 618             fiz2             = _mm256_add_pd(fiz2,tz);
 619
 620             fjx0             = _mm256_add_pd(fjx0,tx);
 621             fjy0             = _mm256_add_pd(fjy0,ty);
 622             fjz0             = _mm256_add_pd(fjz0,tz);
 623
 624             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 625             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 626             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 627             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 628
 629             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 630
 631             /* Inner loop uses 148 flops */
 632         }
 633
 634         /* End of innermost loop */
 635
 636         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 637                                                  f+i_coord_offset,fshift+i_shift_offset);
 638
 639         ggid                        = gid[iidx];
 640         /* Update potential energies */
 641         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 642         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 643
 644         /* Increment number of inner iterations */
 645         inneriter                  += j_index_end - j_index_start;
 646
 647         /* Outer loop uses 20 flops */
 648     }
 649
 650     /* Increment number of outer iterations */
 651     outeriter        += nri;
 652
 653     /* Update outer/inner flops */
 654
 655     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*148);
 656 }
 657 /*
 658  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_256_double
 659  * Electrostatics interaction: CubicSplineTable
 660  * VdW interaction:            LennardJones
 661  * Geometry:                   Water3-Particle
 662  * Calculate force/pot:        Force
 663  */
 664 void
 665 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_256_double
 666                     (t_nblist * gmx_restrict                nlist,
 667                      rvec * gmx_restrict                    xx,
 668                      rvec * gmx_restrict                    ff,
 669                      t_forcerec * gmx_restrict              fr,
 670                      t_mdatoms * gmx_restrict               mdatoms,
 671                      nb_kernel_data_t * gmx_restrict        kernel_data,
 672                      t_nrnb * gmx_restrict                  nrnb)
 673 {
 674     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 675      * just 0 for non-waters.
 676      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 677      * jnr indices corresponding to data put in the four positions in the SIMD register.
 678      */
 679     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 680     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 681     int              jnrA,jnrB,jnrC,jnrD;
 682     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 683     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 684     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 685     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 686     real             rcutoff_scalar;
 687     real             *shiftvec,*fshift,*x,*f;
 688     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 689     real             scratch[4*DIM];
 690     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 691     real *           vdwioffsetptr0;
 692     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 693     real *           vdwioffsetptr1;
 694     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 695     real *           vdwioffsetptr2;
 696     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 697     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 698     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 699     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 700     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 701     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 702     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 703     real             *charge;
 704     int              nvdwtype;
 705     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 706     int              *vdwtype;
 707     real             *vdwparam;
 708     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 709     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 710     __m128i          vfitab;
 711     __m128i          ifour       = _mm_set1_epi32(4);
 712     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 713     real             *vftab;
 714     __m256d          dummy_mask,cutoff_mask;
 715     __m128           tmpmask0,tmpmask1;
 716     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 717     __m256d          one     = _mm256_set1_pd(1.0);
 718     __m256d          two     = _mm256_set1_pd(2.0);
 719     x                = xx[0];
 720     f                = ff[0];
 721
 722     nri              = nlist->nri;
 723     iinr             = nlist->iinr;
 724     jindex           = nlist->jindex;
 725     jjnr             = nlist->jjnr;
 726     shiftidx         = nlist->shift;
 727     gid              = nlist->gid;
 728     shiftvec         = fr->shift_vec[0];
 729     fshift           = fr->fshift[0];
 730     facel            = _mm256_set1_pd(fr->epsfac);
 731     charge           = mdatoms->chargeA;
 732     nvdwtype         = fr->ntype;
 733     vdwparam         = fr->nbfp;
 734     vdwtype          = mdatoms->typeA;
 735
 736     vftab            = kernel_data->table_elec->data;
 737     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 738
 739     /* Setup water-specific parameters */
 740     inr              = nlist->iinr[0];
 741     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 742     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 743     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 744     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 745
 746     /* Avoid stupid compiler warnings */
 747     jnrA = jnrB = jnrC = jnrD = 0;
 748     j_coord_offsetA = 0;
 749     j_coord_offsetB = 0;
 750     j_coord_offsetC = 0;
 751     j_coord_offsetD = 0;
 752
 753     outeriter        = 0;
 754     inneriter        = 0;
 755
 756     for(iidx=0;iidx<4*DIM;iidx++)
 757     {
 758         scratch[iidx] = 0.0;
 759     }
 760
 761     /* Start outer loop over neighborlists */
 762     for(iidx=0; iidx<nri; iidx++)
 763     {
 764         /* Load shift vector for this list */
 765         i_shift_offset   = DIM*shiftidx[iidx];
 766
 767         /* Load limits for loop over neighbors */
 768         j_index_start    = jindex[iidx];
 769         j_index_end      = jindex[iidx+1];
 770
 771         /* Get outer coordinate index */
 772         inr              = iinr[iidx];
 773         i_coord_offset   = DIM*inr;
 774
 775         /* Load i particle coords and add shift vector */
 776         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 777                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 778
 779         fix0             = _mm256_setzero_pd();
 780         fiy0             = _mm256_setzero_pd();
 781         fiz0             = _mm256_setzero_pd();
 782         fix1             = _mm256_setzero_pd();
 783         fiy1             = _mm256_setzero_pd();
 784         fiz1             = _mm256_setzero_pd();
 785         fix2             = _mm256_setzero_pd();
 786         fiy2             = _mm256_setzero_pd();
 787         fiz2             = _mm256_setzero_pd();
 788
 789         /* Start inner kernel loop */
 790         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 791         {
 792
 793             /* Get j neighbor index, and coordinate index */
 794             jnrA             = jjnr[jidx];
 795             jnrB             = jjnr[jidx+1];
 796             jnrC             = jjnr[jidx+2];
 797             jnrD             = jjnr[jidx+3];
 798             j_coord_offsetA  = DIM*jnrA;
 799             j_coord_offsetB  = DIM*jnrB;
 800             j_coord_offsetC  = DIM*jnrC;
 801             j_coord_offsetD  = DIM*jnrD;
 802
 803             /* load j atom coordinates */
 804             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 805                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 806                                                  &jx0,&jy0,&jz0);
 807
 808             /* Calculate displacement vector */
 809             dx00             = _mm256_sub_pd(ix0,jx0);
 810             dy00             = _mm256_sub_pd(iy0,jy0);
 811             dz00             = _mm256_sub_pd(iz0,jz0);
 812             dx10             = _mm256_sub_pd(ix1,jx0);
 813             dy10             = _mm256_sub_pd(iy1,jy0);
 814             dz10             = _mm256_sub_pd(iz1,jz0);
 815             dx20             = _mm256_sub_pd(ix2,jx0);
 816             dy20             = _mm256_sub_pd(iy2,jy0);
 817             dz20             = _mm256_sub_pd(iz2,jz0);
 818
 819             /* Calculate squared distance and things based on it */
 820             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 821             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 822             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 823
 824             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 825             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 826             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 827
 828             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 829
 830             /* Load parameters for j particles */
 831             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 832                                                                  charge+jnrC+0,charge+jnrD+0);
 833             vdwjidx0A        = 2*vdwtype[jnrA+0];
 834             vdwjidx0B        = 2*vdwtype[jnrB+0];
 835             vdwjidx0C        = 2*vdwtype[jnrC+0];
 836             vdwjidx0D        = 2*vdwtype[jnrD+0];
 837
 838             fjx0             = _mm256_setzero_pd();
 839             fjy0             = _mm256_setzero_pd();
 840             fjz0             = _mm256_setzero_pd();
 841
 842             /**************************
 843              * CALCULATE INTERACTIONS *
 844              **************************/
 845
 846             r00              = _mm256_mul_pd(rsq00,rinv00);
 847
 848             /* Compute parameters for interactions between i and j atoms */
 849             qq00             = _mm256_mul_pd(iq0,jq0);
 850             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 851                                             vdwioffsetptr0+vdwjidx0B,
 852                                             vdwioffsetptr0+vdwjidx0C,
 853                                             vdwioffsetptr0+vdwjidx0D,
 854                                             &c6_00,&c12_00);
 855
 856             /* Calculate table index by multiplying r with table scale and truncate to integer */
 857             rt               = _mm256_mul_pd(r00,vftabscale);
 858             vfitab           = _mm256_cvttpd_epi32(rt);
 859             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 860             vfitab           = _mm_slli_epi32(vfitab,2);
 861
 862             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 863             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 864             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 865             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 866             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 867             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 868             Heps             = _mm256_mul_pd(vfeps,H);
 869             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 870             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 871             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 872
 873             /* LENNARD-JONES DISPERSION/REPULSION */
 874
 875             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 876             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 877
 878             fscal            = _mm256_add_pd(felec,fvdw);
 879
 880             /* Calculate temporary vectorial force */
 881             tx               = _mm256_mul_pd(fscal,dx00);
 882             ty               = _mm256_mul_pd(fscal,dy00);
 883             tz               = _mm256_mul_pd(fscal,dz00);
 884
 885             /* Update vectorial force */
 886             fix0             = _mm256_add_pd(fix0,tx);
 887             fiy0             = _mm256_add_pd(fiy0,ty);
 888             fiz0             = _mm256_add_pd(fiz0,tz);
 889
 890             fjx0             = _mm256_add_pd(fjx0,tx);
 891             fjy0             = _mm256_add_pd(fjy0,ty);
 892             fjz0             = _mm256_add_pd(fjz0,tz);
 893
 894             /**************************
 895              * CALCULATE INTERACTIONS *
 896              **************************/
 897
 898             r10              = _mm256_mul_pd(rsq10,rinv10);
 899
 900             /* Compute parameters for interactions between i and j atoms */
 901             qq10             = _mm256_mul_pd(iq1,jq0);
 902
 903             /* Calculate table index by multiplying r with table scale and truncate to integer */
 904             rt               = _mm256_mul_pd(r10,vftabscale);
 905             vfitab           = _mm256_cvttpd_epi32(rt);
 906             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 907             vfitab           = _mm_slli_epi32(vfitab,2);
 908
 909             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 910             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 911             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 912             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 913             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 914             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 915             Heps             = _mm256_mul_pd(vfeps,H);
 916             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 917             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 918             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 919
 920             fscal            = felec;
 921
 922             /* Calculate temporary vectorial force */
 923             tx               = _mm256_mul_pd(fscal,dx10);
 924             ty               = _mm256_mul_pd(fscal,dy10);
 925             tz               = _mm256_mul_pd(fscal,dz10);
 926
 927             /* Update vectorial force */
 928             fix1             = _mm256_add_pd(fix1,tx);
 929             fiy1             = _mm256_add_pd(fiy1,ty);
 930             fiz1             = _mm256_add_pd(fiz1,tz);
 931
 932             fjx0             = _mm256_add_pd(fjx0,tx);
 933             fjy0             = _mm256_add_pd(fjy0,ty);
 934             fjz0             = _mm256_add_pd(fjz0,tz);
 935
 936             /**************************
 937              * CALCULATE INTERACTIONS *
 938              **************************/
 939
 940             r20              = _mm256_mul_pd(rsq20,rinv20);
 941
 942             /* Compute parameters for interactions between i and j atoms */
 943             qq20             = _mm256_mul_pd(iq2,jq0);
 944
 945             /* Calculate table index by multiplying r with table scale and truncate to integer */
 946             rt               = _mm256_mul_pd(r20,vftabscale);
 947             vfitab           = _mm256_cvttpd_epi32(rt);
 948             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 949             vfitab           = _mm_slli_epi32(vfitab,2);
 950
 951             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 952             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 953             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 954             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 955             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 956             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 957             Heps             = _mm256_mul_pd(vfeps,H);
 958             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 959             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 960             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 961
 962             fscal            = felec;
 963
 964             /* Calculate temporary vectorial force */
 965             tx               = _mm256_mul_pd(fscal,dx20);
 966             ty               = _mm256_mul_pd(fscal,dy20);
 967             tz               = _mm256_mul_pd(fscal,dz20);
 968
 969             /* Update vectorial force */
 970             fix2             = _mm256_add_pd(fix2,tx);
 971             fiy2             = _mm256_add_pd(fiy2,ty);
 972             fiz2             = _mm256_add_pd(fiz2,tz);
 973
 974             fjx0             = _mm256_add_pd(fjx0,tx);
 975             fjy0             = _mm256_add_pd(fjy0,ty);
 976             fjz0             = _mm256_add_pd(fjz0,tz);
 977
 978             fjptrA             = f+j_coord_offsetA;
 979             fjptrB             = f+j_coord_offsetB;
 980             fjptrC             = f+j_coord_offsetC;
 981             fjptrD             = f+j_coord_offsetD;
 982
 983             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 984
 985             /* Inner loop uses 128 flops */
 986         }
 987
 988         if(jidx<j_index_end)
 989         {
 990
 991             /* Get j neighbor index, and coordinate index */
 992             jnrlistA         = jjnr[jidx];
 993             jnrlistB         = jjnr[jidx+1];
 994             jnrlistC         = jjnr[jidx+2];
 995             jnrlistD         = jjnr[jidx+3];
 996             /* Sign of each element will be negative for non-real atoms.
 997              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 998              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 999              */
1000             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1001
1002             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1003             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1004             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1005
1006             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1007             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1008             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1009             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1010             j_coord_offsetA  = DIM*jnrA;
1011             j_coord_offsetB  = DIM*jnrB;
1012             j_coord_offsetC  = DIM*jnrC;
1013             j_coord_offsetD  = DIM*jnrD;
1014
1015             /* load j atom coordinates */
1016             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1017                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1018                                                  &jx0,&jy0,&jz0);
1019
1020             /* Calculate displacement vector */
1021             dx00             = _mm256_sub_pd(ix0,jx0);
1022             dy00             = _mm256_sub_pd(iy0,jy0);
1023             dz00             = _mm256_sub_pd(iz0,jz0);
1024             dx10             = _mm256_sub_pd(ix1,jx0);
1025             dy10             = _mm256_sub_pd(iy1,jy0);
1026             dz10             = _mm256_sub_pd(iz1,jz0);
1027             dx20             = _mm256_sub_pd(ix2,jx0);
1028             dy20             = _mm256_sub_pd(iy2,jy0);
1029             dz20             = _mm256_sub_pd(iz2,jz0);
1030
1031             /* Calculate squared distance and things based on it */
1032             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1033             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1034             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1035
1036             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1037             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1038             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1039
1040             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1041
1042             /* Load parameters for j particles */
1043             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1044                                                                  charge+jnrC+0,charge+jnrD+0);
1045             vdwjidx0A        = 2*vdwtype[jnrA+0];
1046             vdwjidx0B        = 2*vdwtype[jnrB+0];
1047             vdwjidx0C        = 2*vdwtype[jnrC+0];
1048             vdwjidx0D        = 2*vdwtype[jnrD+0];
1049
1050             fjx0             = _mm256_setzero_pd();
1051             fjy0             = _mm256_setzero_pd();
1052             fjz0             = _mm256_setzero_pd();
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             r00              = _mm256_mul_pd(rsq00,rinv00);
1059             r00              = _mm256_andnot_pd(dummy_mask,r00);
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             qq00             = _mm256_mul_pd(iq0,jq0);
1063             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1064                                             vdwioffsetptr0+vdwjidx0B,
1065                                             vdwioffsetptr0+vdwjidx0C,
1066                                             vdwioffsetptr0+vdwjidx0D,
1067                                             &c6_00,&c12_00);
1068
1069             /* Calculate table index by multiplying r with table scale and truncate to integer */
1070             rt               = _mm256_mul_pd(r00,vftabscale);
1071             vfitab           = _mm256_cvttpd_epi32(rt);
1072             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1073             vfitab           = _mm_slli_epi32(vfitab,2);
1074
1075             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1076             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1077             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1078             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1079             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1080             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1081             Heps             = _mm256_mul_pd(vfeps,H);
1082             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1083             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1084             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1085
1086             /* LENNARD-JONES DISPERSION/REPULSION */
1087
1088             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1089             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1090
1091             fscal            = _mm256_add_pd(felec,fvdw);
1092
1093             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1094
1095             /* Calculate temporary vectorial force */
1096             tx               = _mm256_mul_pd(fscal,dx00);
1097             ty               = _mm256_mul_pd(fscal,dy00);
1098             tz               = _mm256_mul_pd(fscal,dz00);
1099
1100             /* Update vectorial force */
1101             fix0             = _mm256_add_pd(fix0,tx);
1102             fiy0             = _mm256_add_pd(fiy0,ty);
1103             fiz0             = _mm256_add_pd(fiz0,tz);
1104
1105             fjx0             = _mm256_add_pd(fjx0,tx);
1106             fjy0             = _mm256_add_pd(fjy0,ty);
1107             fjz0             = _mm256_add_pd(fjz0,tz);
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             r10              = _mm256_mul_pd(rsq10,rinv10);
1114             r10              = _mm256_andnot_pd(dummy_mask,r10);
1115
1116             /* Compute parameters for interactions between i and j atoms */
1117             qq10             = _mm256_mul_pd(iq1,jq0);
1118
1119             /* Calculate table index by multiplying r with table scale and truncate to integer */
1120             rt               = _mm256_mul_pd(r10,vftabscale);
1121             vfitab           = _mm256_cvttpd_epi32(rt);
1122             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1123             vfitab           = _mm_slli_epi32(vfitab,2);
1124
1125             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1126             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1127             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1128             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1129             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1130             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1131             Heps             = _mm256_mul_pd(vfeps,H);
1132             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1133             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1134             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1135
1136             fscal            = felec;
1137
1138             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1139
1140             /* Calculate temporary vectorial force */
1141             tx               = _mm256_mul_pd(fscal,dx10);
1142             ty               = _mm256_mul_pd(fscal,dy10);
1143             tz               = _mm256_mul_pd(fscal,dz10);
1144
1145             /* Update vectorial force */
1146             fix1             = _mm256_add_pd(fix1,tx);
1147             fiy1             = _mm256_add_pd(fiy1,ty);
1148             fiz1             = _mm256_add_pd(fiz1,tz);
1149
1150             fjx0             = _mm256_add_pd(fjx0,tx);
1151             fjy0             = _mm256_add_pd(fjy0,ty);
1152             fjz0             = _mm256_add_pd(fjz0,tz);
1153
1154             /**************************
1155              * CALCULATE INTERACTIONS *
1156              **************************/
1157
1158             r20              = _mm256_mul_pd(rsq20,rinv20);
1159             r20              = _mm256_andnot_pd(dummy_mask,r20);
1160
1161             /* Compute parameters for interactions between i and j atoms */
1162             qq20             = _mm256_mul_pd(iq2,jq0);
1163
1164             /* Calculate table index by multiplying r with table scale and truncate to integer */
1165             rt               = _mm256_mul_pd(r20,vftabscale);
1166             vfitab           = _mm256_cvttpd_epi32(rt);
1167             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1168             vfitab           = _mm_slli_epi32(vfitab,2);
1169
1170             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1171             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1172             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1173             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1174             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1175             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1176             Heps             = _mm256_mul_pd(vfeps,H);
1177             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1178             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1179             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1180
1181             fscal            = felec;
1182
1183             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1184
1185             /* Calculate temporary vectorial force */
1186             tx               = _mm256_mul_pd(fscal,dx20);
1187             ty               = _mm256_mul_pd(fscal,dy20);
1188             tz               = _mm256_mul_pd(fscal,dz20);
1189
1190             /* Update vectorial force */
1191             fix2             = _mm256_add_pd(fix2,tx);
1192             fiy2             = _mm256_add_pd(fiy2,ty);
1193             fiz2             = _mm256_add_pd(fiz2,tz);
1194
1195             fjx0             = _mm256_add_pd(fjx0,tx);
1196             fjy0             = _mm256_add_pd(fjy0,ty);
1197             fjz0             = _mm256_add_pd(fjz0,tz);
1198
1199             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1200             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1201             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1202             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1203
1204             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1205
1206             /* Inner loop uses 131 flops */
1207         }
1208
1209         /* End of innermost loop */
1210
1211         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1212                                                  f+i_coord_offset,fshift+i_shift_offset);
1213
1214         /* Increment number of inner iterations */
1215         inneriter                  += j_index_end - j_index_start;
1216
1217         /* Outer loop uses 18 flops */
1218     }
1219
1220     /* Increment number of outer iterations */
1221     outeriter        += nri;
1222
1223     /* Update outer/inner flops */
1224
1225     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*131);
1226 }