src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_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_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_256_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_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     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     vftab            = kernel_data->table_vdw->data;
 116     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->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             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 213             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 214
 215             /* Load parameters for j particles */
 216             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 217                                                                  charge+jnrC+0,charge+jnrD+0);
 218             vdwjidx0A        = 2*vdwtype[jnrA+0];
 219             vdwjidx0B        = 2*vdwtype[jnrB+0];
 220             vdwjidx0C        = 2*vdwtype[jnrC+0];
 221             vdwjidx0D        = 2*vdwtype[jnrD+0];
 222
 223             fjx0             = _mm256_setzero_pd();
 224             fjy0             = _mm256_setzero_pd();
 225             fjz0             = _mm256_setzero_pd();
 226
 227             /**************************
 228              * CALCULATE INTERACTIONS *
 229              **************************/
 230
 231             r00              = _mm256_mul_pd(rsq00,rinv00);
 232
 233             /* Compute parameters for interactions between i and j atoms */
 234             qq00             = _mm256_mul_pd(iq0,jq0);
 235             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 236                                             vdwioffsetptr0+vdwjidx0B,
 237                                             vdwioffsetptr0+vdwjidx0C,
 238                                             vdwioffsetptr0+vdwjidx0D,
 239                                             &c6_00,&c12_00);
 240
 241             /* Calculate table index by multiplying r with table scale and truncate to integer */
 242             rt               = _mm256_mul_pd(r00,vftabscale);
 243             vfitab           = _mm256_cvttpd_epi32(rt);
 244             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 245             vfitab           = _mm_slli_epi32(vfitab,3);
 246
 247             /* COULOMB ELECTROSTATICS */
 248             velec            = _mm256_mul_pd(qq00,rinv00);
 249             felec            = _mm256_mul_pd(velec,rinvsq00);
 250
 251             /* CUBIC SPLINE TABLE DISPERSION */
 252             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 253             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 254             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 255             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 256             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 257             Heps             = _mm256_mul_pd(vfeps,H);
 258             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 259             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 260             vvdw6            = _mm256_mul_pd(c6_00,VV);
 261             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 262             fvdw6            = _mm256_mul_pd(c6_00,FF);
 263
 264             /* CUBIC SPLINE TABLE REPULSION */
 265             vfitab           = _mm_add_epi32(vfitab,ifour);
 266             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 267             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 268             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 269             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 270             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 271             Heps             = _mm256_mul_pd(vfeps,H);
 272             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 273             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 274             vvdw12           = _mm256_mul_pd(c12_00,VV);
 275             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 276             fvdw12           = _mm256_mul_pd(c12_00,FF);
 277             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 278             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _mm256_add_pd(velecsum,velec);
 282             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 283
 284             fscal            = _mm256_add_pd(felec,fvdw);
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm256_mul_pd(fscal,dx00);
 288             ty               = _mm256_mul_pd(fscal,dy00);
 289             tz               = _mm256_mul_pd(fscal,dz00);
 290
 291             /* Update vectorial force */
 292             fix0             = _mm256_add_pd(fix0,tx);
 293             fiy0             = _mm256_add_pd(fiy0,ty);
 294             fiz0             = _mm256_add_pd(fiz0,tz);
 295
 296             fjx0             = _mm256_add_pd(fjx0,tx);
 297             fjy0             = _mm256_add_pd(fjy0,ty);
 298             fjz0             = _mm256_add_pd(fjz0,tz);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq10             = _mm256_mul_pd(iq1,jq0);
 306
 307             /* COULOMB ELECTROSTATICS */
 308             velec            = _mm256_mul_pd(qq10,rinv10);
 309             felec            = _mm256_mul_pd(velec,rinvsq10);
 310
 311             /* Update potential sum for this i atom from the interaction with this j atom. */
 312             velecsum         = _mm256_add_pd(velecsum,velec);
 313
 314             fscal            = felec;
 315
 316             /* Calculate temporary vectorial force */
 317             tx               = _mm256_mul_pd(fscal,dx10);
 318             ty               = _mm256_mul_pd(fscal,dy10);
 319             tz               = _mm256_mul_pd(fscal,dz10);
 320
 321             /* Update vectorial force */
 322             fix1             = _mm256_add_pd(fix1,tx);
 323             fiy1             = _mm256_add_pd(fiy1,ty);
 324             fiz1             = _mm256_add_pd(fiz1,tz);
 325
 326             fjx0             = _mm256_add_pd(fjx0,tx);
 327             fjy0             = _mm256_add_pd(fjy0,ty);
 328             fjz0             = _mm256_add_pd(fjz0,tz);
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             /* Compute parameters for interactions between i and j atoms */
 335             qq20             = _mm256_mul_pd(iq2,jq0);
 336
 337             /* COULOMB ELECTROSTATICS */
 338             velec            = _mm256_mul_pd(qq20,rinv20);
 339             felec            = _mm256_mul_pd(velec,rinvsq20);
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm256_add_pd(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm256_mul_pd(fscal,dx20);
 348             ty               = _mm256_mul_pd(fscal,dy20);
 349             tz               = _mm256_mul_pd(fscal,dz20);
 350
 351             /* Update vectorial force */
 352             fix2             = _mm256_add_pd(fix2,tx);
 353             fiy2             = _mm256_add_pd(fiy2,ty);
 354             fiz2             = _mm256_add_pd(fiz2,tz);
 355
 356             fjx0             = _mm256_add_pd(fjx0,tx);
 357             fjy0             = _mm256_add_pd(fjy0,ty);
 358             fjz0             = _mm256_add_pd(fjz0,tz);
 359
 360             fjptrA             = f+j_coord_offsetA;
 361             fjptrB             = f+j_coord_offsetB;
 362             fjptrC             = f+j_coord_offsetC;
 363             fjptrD             = f+j_coord_offsetD;
 364
 365             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 366
 367             /* Inner loop uses 119 flops */
 368         }
 369
 370         if(jidx<j_index_end)
 371         {
 372
 373             /* Get j neighbor index, and coordinate index */
 374             jnrlistA         = jjnr[jidx];
 375             jnrlistB         = jjnr[jidx+1];
 376             jnrlistC         = jjnr[jidx+2];
 377             jnrlistD         = jjnr[jidx+3];
 378             /* Sign of each element will be negative for non-real atoms.
 379              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 380              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 381              */
 382             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 383
 384             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 385             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 386             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 387
 388             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 389             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 390             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 391             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 392             j_coord_offsetA  = DIM*jnrA;
 393             j_coord_offsetB  = DIM*jnrB;
 394             j_coord_offsetC  = DIM*jnrC;
 395             j_coord_offsetD  = DIM*jnrD;
 396
 397             /* load j atom coordinates */
 398             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 399                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 400                                                  &jx0,&jy0,&jz0);
 401
 402             /* Calculate displacement vector */
 403             dx00             = _mm256_sub_pd(ix0,jx0);
 404             dy00             = _mm256_sub_pd(iy0,jy0);
 405             dz00             = _mm256_sub_pd(iz0,jz0);
 406             dx10             = _mm256_sub_pd(ix1,jx0);
 407             dy10             = _mm256_sub_pd(iy1,jy0);
 408             dz10             = _mm256_sub_pd(iz1,jz0);
 409             dx20             = _mm256_sub_pd(ix2,jx0);
 410             dy20             = _mm256_sub_pd(iy2,jy0);
 411             dz20             = _mm256_sub_pd(iz2,jz0);
 412
 413             /* Calculate squared distance and things based on it */
 414             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 415             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 416             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 417
 418             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 419             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 420             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 421
 422             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 423             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 424             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 425
 426             /* Load parameters for j particles */
 427             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 428                                                                  charge+jnrC+0,charge+jnrD+0);
 429             vdwjidx0A        = 2*vdwtype[jnrA+0];
 430             vdwjidx0B        = 2*vdwtype[jnrB+0];
 431             vdwjidx0C        = 2*vdwtype[jnrC+0];
 432             vdwjidx0D        = 2*vdwtype[jnrD+0];
 433
 434             fjx0             = _mm256_setzero_pd();
 435             fjy0             = _mm256_setzero_pd();
 436             fjz0             = _mm256_setzero_pd();
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             r00              = _mm256_mul_pd(rsq00,rinv00);
 443             r00              = _mm256_andnot_pd(dummy_mask,r00);
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             qq00             = _mm256_mul_pd(iq0,jq0);
 447             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 448                                             vdwioffsetptr0+vdwjidx0B,
 449                                             vdwioffsetptr0+vdwjidx0C,
 450                                             vdwioffsetptr0+vdwjidx0D,
 451                                             &c6_00,&c12_00);
 452
 453             /* Calculate table index by multiplying r with table scale and truncate to integer */
 454             rt               = _mm256_mul_pd(r00,vftabscale);
 455             vfitab           = _mm256_cvttpd_epi32(rt);
 456             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 457             vfitab           = _mm_slli_epi32(vfitab,3);
 458
 459             /* COULOMB ELECTROSTATICS */
 460             velec            = _mm256_mul_pd(qq00,rinv00);
 461             felec            = _mm256_mul_pd(velec,rinvsq00);
 462
 463             /* CUBIC SPLINE TABLE DISPERSION */
 464             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 465             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 466             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 467             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 468             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 469             Heps             = _mm256_mul_pd(vfeps,H);
 470             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 471             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 472             vvdw6            = _mm256_mul_pd(c6_00,VV);
 473             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 474             fvdw6            = _mm256_mul_pd(c6_00,FF);
 475
 476             /* CUBIC SPLINE TABLE REPULSION */
 477             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdw12           = _mm256_mul_pd(c12_00,VV);
 487             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 488             fvdw12           = _mm256_mul_pd(c12_00,FF);
 489             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 490             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 491
 492             /* Update potential sum for this i atom from the interaction with this j atom. */
 493             velec            = _mm256_andnot_pd(dummy_mask,velec);
 494             velecsum         = _mm256_add_pd(velecsum,velec);
 495             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 496             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 497
 498             fscal            = _mm256_add_pd(felec,fvdw);
 499
 500             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 501
 502             /* Calculate temporary vectorial force */
 503             tx               = _mm256_mul_pd(fscal,dx00);
 504             ty               = _mm256_mul_pd(fscal,dy00);
 505             tz               = _mm256_mul_pd(fscal,dz00);
 506
 507             /* Update vectorial force */
 508             fix0             = _mm256_add_pd(fix0,tx);
 509             fiy0             = _mm256_add_pd(fiy0,ty);
 510             fiz0             = _mm256_add_pd(fiz0,tz);
 511
 512             fjx0             = _mm256_add_pd(fjx0,tx);
 513             fjy0             = _mm256_add_pd(fjy0,ty);
 514             fjz0             = _mm256_add_pd(fjz0,tz);
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             /* Compute parameters for interactions between i and j atoms */
 521             qq10             = _mm256_mul_pd(iq1,jq0);
 522
 523             /* COULOMB ELECTROSTATICS */
 524             velec            = _mm256_mul_pd(qq10,rinv10);
 525             felec            = _mm256_mul_pd(velec,rinvsq10);
 526
 527             /* Update potential sum for this i atom from the interaction with this j atom. */
 528             velec            = _mm256_andnot_pd(dummy_mask,velec);
 529             velecsum         = _mm256_add_pd(velecsum,velec);
 530
 531             fscal            = felec;
 532
 533             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 534
 535             /* Calculate temporary vectorial force */
 536             tx               = _mm256_mul_pd(fscal,dx10);
 537             ty               = _mm256_mul_pd(fscal,dy10);
 538             tz               = _mm256_mul_pd(fscal,dz10);
 539
 540             /* Update vectorial force */
 541             fix1             = _mm256_add_pd(fix1,tx);
 542             fiy1             = _mm256_add_pd(fiy1,ty);
 543             fiz1             = _mm256_add_pd(fiz1,tz);
 544
 545             fjx0             = _mm256_add_pd(fjx0,tx);
 546             fjy0             = _mm256_add_pd(fjy0,ty);
 547             fjz0             = _mm256_add_pd(fjz0,tz);
 548
 549             /**************************
 550              * CALCULATE INTERACTIONS *
 551              **************************/
 552
 553             /* Compute parameters for interactions between i and j atoms */
 554             qq20             = _mm256_mul_pd(iq2,jq0);
 555
 556             /* COULOMB ELECTROSTATICS */
 557             velec            = _mm256_mul_pd(qq20,rinv20);
 558             felec            = _mm256_mul_pd(velec,rinvsq20);
 559
 560             /* Update potential sum for this i atom from the interaction with this j atom. */
 561             velec            = _mm256_andnot_pd(dummy_mask,velec);
 562             velecsum         = _mm256_add_pd(velecsum,velec);
 563
 564             fscal            = felec;
 565
 566             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 567
 568             /* Calculate temporary vectorial force */
 569             tx               = _mm256_mul_pd(fscal,dx20);
 570             ty               = _mm256_mul_pd(fscal,dy20);
 571             tz               = _mm256_mul_pd(fscal,dz20);
 572
 573             /* Update vectorial force */
 574             fix2             = _mm256_add_pd(fix2,tx);
 575             fiy2             = _mm256_add_pd(fiy2,ty);
 576             fiz2             = _mm256_add_pd(fiz2,tz);
 577
 578             fjx0             = _mm256_add_pd(fjx0,tx);
 579             fjy0             = _mm256_add_pd(fjy0,ty);
 580             fjz0             = _mm256_add_pd(fjz0,tz);
 581
 582             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 583             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 584             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 585             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 586
 587             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 588
 589             /* Inner loop uses 120 flops */
 590         }
 591
 592         /* End of innermost loop */
 593
 594         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 595                                                  f+i_coord_offset,fshift+i_shift_offset);
 596
 597         ggid                        = gid[iidx];
 598         /* Update potential energies */
 599         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 600         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 601
 602         /* Increment number of inner iterations */
 603         inneriter                  += j_index_end - j_index_start;
 604
 605         /* Outer loop uses 20 flops */
 606     }
 607
 608     /* Increment number of outer iterations */
 609     outeriter        += nri;
 610
 611     /* Update outer/inner flops */
 612
 613     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*120);
 614 }
 615 /*
 616  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_256_double
 617  * Electrostatics interaction: Coulomb
 618  * VdW interaction:            CubicSplineTable
 619  * Geometry:                   Water3-Particle
 620  * Calculate force/pot:        Force
 621  */
 622 void
 623 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_256_double
 624                     (t_nblist * gmx_restrict                nlist,
 625                      rvec * gmx_restrict                    xx,
 626                      rvec * gmx_restrict                    ff,
 627                      t_forcerec * gmx_restrict              fr,
 628                      t_mdatoms * gmx_restrict               mdatoms,
 629                      nb_kernel_data_t * gmx_restrict        kernel_data,
 630                      t_nrnb * gmx_restrict                  nrnb)
 631 {
 632     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 633      * just 0 for non-waters.
 634      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 635      * jnr indices corresponding to data put in the four positions in the SIMD register.
 636      */
 637     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 638     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 639     int              jnrA,jnrB,jnrC,jnrD;
 640     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 641     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 642     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 643     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 644     real             rcutoff_scalar;
 645     real             *shiftvec,*fshift,*x,*f;
 646     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 647     real             scratch[4*DIM];
 648     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 649     real *           vdwioffsetptr0;
 650     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 651     real *           vdwioffsetptr1;
 652     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 653     real *           vdwioffsetptr2;
 654     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 655     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 656     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 657     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 658     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 659     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 660     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 661     real             *charge;
 662     int              nvdwtype;
 663     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 664     int              *vdwtype;
 665     real             *vdwparam;
 666     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 667     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 668     __m128i          vfitab;
 669     __m128i          ifour       = _mm_set1_epi32(4);
 670     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 671     real             *vftab;
 672     __m256d          dummy_mask,cutoff_mask;
 673     __m128           tmpmask0,tmpmask1;
 674     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 675     __m256d          one     = _mm256_set1_pd(1.0);
 676     __m256d          two     = _mm256_set1_pd(2.0);
 677     x                = xx[0];
 678     f                = ff[0];
 679
 680     nri              = nlist->nri;
 681     iinr             = nlist->iinr;
 682     jindex           = nlist->jindex;
 683     jjnr             = nlist->jjnr;
 684     shiftidx         = nlist->shift;
 685     gid              = nlist->gid;
 686     shiftvec         = fr->shift_vec[0];
 687     fshift           = fr->fshift[0];
 688     facel            = _mm256_set1_pd(fr->epsfac);
 689     charge           = mdatoms->chargeA;
 690     nvdwtype         = fr->ntype;
 691     vdwparam         = fr->nbfp;
 692     vdwtype          = mdatoms->typeA;
 693
 694     vftab            = kernel_data->table_vdw->data;
 695     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 696
 697     /* Setup water-specific parameters */
 698     inr              = nlist->iinr[0];
 699     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 700     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 701     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 702     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 703
 704     /* Avoid stupid compiler warnings */
 705     jnrA = jnrB = jnrC = jnrD = 0;
 706     j_coord_offsetA = 0;
 707     j_coord_offsetB = 0;
 708     j_coord_offsetC = 0;
 709     j_coord_offsetD = 0;
 710
 711     outeriter        = 0;
 712     inneriter        = 0;
 713
 714     for(iidx=0;iidx<4*DIM;iidx++)
 715     {
 716         scratch[iidx] = 0.0;
 717     }
 718
 719     /* Start outer loop over neighborlists */
 720     for(iidx=0; iidx<nri; iidx++)
 721     {
 722         /* Load shift vector for this list */
 723         i_shift_offset   = DIM*shiftidx[iidx];
 724
 725         /* Load limits for loop over neighbors */
 726         j_index_start    = jindex[iidx];
 727         j_index_end      = jindex[iidx+1];
 728
 729         /* Get outer coordinate index */
 730         inr              = iinr[iidx];
 731         i_coord_offset   = DIM*inr;
 732
 733         /* Load i particle coords and add shift vector */
 734         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 735                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 736
 737         fix0             = _mm256_setzero_pd();
 738         fiy0             = _mm256_setzero_pd();
 739         fiz0             = _mm256_setzero_pd();
 740         fix1             = _mm256_setzero_pd();
 741         fiy1             = _mm256_setzero_pd();
 742         fiz1             = _mm256_setzero_pd();
 743         fix2             = _mm256_setzero_pd();
 744         fiy2             = _mm256_setzero_pd();
 745         fiz2             = _mm256_setzero_pd();
 746
 747         /* Start inner kernel loop */
 748         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 749         {
 750
 751             /* Get j neighbor index, and coordinate index */
 752             jnrA             = jjnr[jidx];
 753             jnrB             = jjnr[jidx+1];
 754             jnrC             = jjnr[jidx+2];
 755             jnrD             = jjnr[jidx+3];
 756             j_coord_offsetA  = DIM*jnrA;
 757             j_coord_offsetB  = DIM*jnrB;
 758             j_coord_offsetC  = DIM*jnrC;
 759             j_coord_offsetD  = DIM*jnrD;
 760
 761             /* load j atom coordinates */
 762             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 763                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 764                                                  &jx0,&jy0,&jz0);
 765
 766             /* Calculate displacement vector */
 767             dx00             = _mm256_sub_pd(ix0,jx0);
 768             dy00             = _mm256_sub_pd(iy0,jy0);
 769             dz00             = _mm256_sub_pd(iz0,jz0);
 770             dx10             = _mm256_sub_pd(ix1,jx0);
 771             dy10             = _mm256_sub_pd(iy1,jy0);
 772             dz10             = _mm256_sub_pd(iz1,jz0);
 773             dx20             = _mm256_sub_pd(ix2,jx0);
 774             dy20             = _mm256_sub_pd(iy2,jy0);
 775             dz20             = _mm256_sub_pd(iz2,jz0);
 776
 777             /* Calculate squared distance and things based on it */
 778             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 779             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 780             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 781
 782             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 783             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 784             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 785
 786             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 787             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 788             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 789
 790             /* Load parameters for j particles */
 791             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 792                                                                  charge+jnrC+0,charge+jnrD+0);
 793             vdwjidx0A        = 2*vdwtype[jnrA+0];
 794             vdwjidx0B        = 2*vdwtype[jnrB+0];
 795             vdwjidx0C        = 2*vdwtype[jnrC+0];
 796             vdwjidx0D        = 2*vdwtype[jnrD+0];
 797
 798             fjx0             = _mm256_setzero_pd();
 799             fjy0             = _mm256_setzero_pd();
 800             fjz0             = _mm256_setzero_pd();
 801
 802             /**************************
 803              * CALCULATE INTERACTIONS *
 804              **************************/
 805
 806             r00              = _mm256_mul_pd(rsq00,rinv00);
 807
 808             /* Compute parameters for interactions between i and j atoms */
 809             qq00             = _mm256_mul_pd(iq0,jq0);
 810             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 811                                             vdwioffsetptr0+vdwjidx0B,
 812                                             vdwioffsetptr0+vdwjidx0C,
 813                                             vdwioffsetptr0+vdwjidx0D,
 814                                             &c6_00,&c12_00);
 815
 816             /* Calculate table index by multiplying r with table scale and truncate to integer */
 817             rt               = _mm256_mul_pd(r00,vftabscale);
 818             vfitab           = _mm256_cvttpd_epi32(rt);
 819             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 820             vfitab           = _mm_slli_epi32(vfitab,3);
 821
 822             /* COULOMB ELECTROSTATICS */
 823             velec            = _mm256_mul_pd(qq00,rinv00);
 824             felec            = _mm256_mul_pd(velec,rinvsq00);
 825
 826             /* CUBIC SPLINE TABLE DISPERSION */
 827             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 828             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 829             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 830             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 831             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 832             Heps             = _mm256_mul_pd(vfeps,H);
 833             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 834             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 835             fvdw6            = _mm256_mul_pd(c6_00,FF);
 836
 837             /* CUBIC SPLINE TABLE REPULSION */
 838             vfitab           = _mm_add_epi32(vfitab,ifour);
 839             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 840             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 841             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 842             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 843             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 844             Heps             = _mm256_mul_pd(vfeps,H);
 845             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 846             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 847             fvdw12           = _mm256_mul_pd(c12_00,FF);
 848             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 849
 850             fscal            = _mm256_add_pd(felec,fvdw);
 851
 852             /* Calculate temporary vectorial force */
 853             tx               = _mm256_mul_pd(fscal,dx00);
 854             ty               = _mm256_mul_pd(fscal,dy00);
 855             tz               = _mm256_mul_pd(fscal,dz00);
 856
 857             /* Update vectorial force */
 858             fix0             = _mm256_add_pd(fix0,tx);
 859             fiy0             = _mm256_add_pd(fiy0,ty);
 860             fiz0             = _mm256_add_pd(fiz0,tz);
 861
 862             fjx0             = _mm256_add_pd(fjx0,tx);
 863             fjy0             = _mm256_add_pd(fjy0,ty);
 864             fjz0             = _mm256_add_pd(fjz0,tz);
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             /* Compute parameters for interactions between i and j atoms */
 871             qq10             = _mm256_mul_pd(iq1,jq0);
 872
 873             /* COULOMB ELECTROSTATICS */
 874             velec            = _mm256_mul_pd(qq10,rinv10);
 875             felec            = _mm256_mul_pd(velec,rinvsq10);
 876
 877             fscal            = felec;
 878
 879             /* Calculate temporary vectorial force */
 880             tx               = _mm256_mul_pd(fscal,dx10);
 881             ty               = _mm256_mul_pd(fscal,dy10);
 882             tz               = _mm256_mul_pd(fscal,dz10);
 883
 884             /* Update vectorial force */
 885             fix1             = _mm256_add_pd(fix1,tx);
 886             fiy1             = _mm256_add_pd(fiy1,ty);
 887             fiz1             = _mm256_add_pd(fiz1,tz);
 888
 889             fjx0             = _mm256_add_pd(fjx0,tx);
 890             fjy0             = _mm256_add_pd(fjy0,ty);
 891             fjz0             = _mm256_add_pd(fjz0,tz);
 892
 893             /**************************
 894              * CALCULATE INTERACTIONS *
 895              **************************/
 896
 897             /* Compute parameters for interactions between i and j atoms */
 898             qq20             = _mm256_mul_pd(iq2,jq0);
 899
 900             /* COULOMB ELECTROSTATICS */
 901             velec            = _mm256_mul_pd(qq20,rinv20);
 902             felec            = _mm256_mul_pd(velec,rinvsq20);
 903
 904             fscal            = felec;
 905
 906             /* Calculate temporary vectorial force */
 907             tx               = _mm256_mul_pd(fscal,dx20);
 908             ty               = _mm256_mul_pd(fscal,dy20);
 909             tz               = _mm256_mul_pd(fscal,dz20);
 910
 911             /* Update vectorial force */
 912             fix2             = _mm256_add_pd(fix2,tx);
 913             fiy2             = _mm256_add_pd(fiy2,ty);
 914             fiz2             = _mm256_add_pd(fiz2,tz);
 915
 916             fjx0             = _mm256_add_pd(fjx0,tx);
 917             fjy0             = _mm256_add_pd(fjy0,ty);
 918             fjz0             = _mm256_add_pd(fjz0,tz);
 919
 920             fjptrA             = f+j_coord_offsetA;
 921             fjptrB             = f+j_coord_offsetB;
 922             fjptrC             = f+j_coord_offsetC;
 923             fjptrD             = f+j_coord_offsetD;
 924
 925             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 926
 927             /* Inner loop uses 108 flops */
 928         }
 929
 930         if(jidx<j_index_end)
 931         {
 932
 933             /* Get j neighbor index, and coordinate index */
 934             jnrlistA         = jjnr[jidx];
 935             jnrlistB         = jjnr[jidx+1];
 936             jnrlistC         = jjnr[jidx+2];
 937             jnrlistD         = jjnr[jidx+3];
 938             /* Sign of each element will be negative for non-real atoms.
 939              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 940              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 941              */
 942             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 943
 944             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 945             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 946             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 947
 948             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 949             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 950             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 951             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 952             j_coord_offsetA  = DIM*jnrA;
 953             j_coord_offsetB  = DIM*jnrB;
 954             j_coord_offsetC  = DIM*jnrC;
 955             j_coord_offsetD  = DIM*jnrD;
 956
 957             /* load j atom coordinates */
 958             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 959                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 960                                                  &jx0,&jy0,&jz0);
 961
 962             /* Calculate displacement vector */
 963             dx00             = _mm256_sub_pd(ix0,jx0);
 964             dy00             = _mm256_sub_pd(iy0,jy0);
 965             dz00             = _mm256_sub_pd(iz0,jz0);
 966             dx10             = _mm256_sub_pd(ix1,jx0);
 967             dy10             = _mm256_sub_pd(iy1,jy0);
 968             dz10             = _mm256_sub_pd(iz1,jz0);
 969             dx20             = _mm256_sub_pd(ix2,jx0);
 970             dy20             = _mm256_sub_pd(iy2,jy0);
 971             dz20             = _mm256_sub_pd(iz2,jz0);
 972
 973             /* Calculate squared distance and things based on it */
 974             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 975             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 976             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 977
 978             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 979             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 980             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 981
 982             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 983             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 984             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 985
 986             /* Load parameters for j particles */
 987             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 988                                                                  charge+jnrC+0,charge+jnrD+0);
 989             vdwjidx0A        = 2*vdwtype[jnrA+0];
 990             vdwjidx0B        = 2*vdwtype[jnrB+0];
 991             vdwjidx0C        = 2*vdwtype[jnrC+0];
 992             vdwjidx0D        = 2*vdwtype[jnrD+0];
 993
 994             fjx0             = _mm256_setzero_pd();
 995             fjy0             = _mm256_setzero_pd();
 996             fjz0             = _mm256_setzero_pd();
 997
 998             /**************************
 999              * CALCULATE INTERACTIONS *
1000              **************************/
1001
1002             r00              = _mm256_mul_pd(rsq00,rinv00);
1003             r00              = _mm256_andnot_pd(dummy_mask,r00);
1004
1005             /* Compute parameters for interactions between i and j atoms */
1006             qq00             = _mm256_mul_pd(iq0,jq0);
1007             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1008                                             vdwioffsetptr0+vdwjidx0B,
1009                                             vdwioffsetptr0+vdwjidx0C,
1010                                             vdwioffsetptr0+vdwjidx0D,
1011                                             &c6_00,&c12_00);
1012
1013             /* Calculate table index by multiplying r with table scale and truncate to integer */
1014             rt               = _mm256_mul_pd(r00,vftabscale);
1015             vfitab           = _mm256_cvttpd_epi32(rt);
1016             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1017             vfitab           = _mm_slli_epi32(vfitab,3);
1018
1019             /* COULOMB ELECTROSTATICS */
1020             velec            = _mm256_mul_pd(qq00,rinv00);
1021             felec            = _mm256_mul_pd(velec,rinvsq00);
1022
1023             /* CUBIC SPLINE TABLE DISPERSION */
1024             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1025             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1026             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1027             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1028             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1029             Heps             = _mm256_mul_pd(vfeps,H);
1030             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1031             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1032             fvdw6            = _mm256_mul_pd(c6_00,FF);
1033
1034             /* CUBIC SPLINE TABLE REPULSION */
1035             vfitab           = _mm_add_epi32(vfitab,ifour);
1036             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1037             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1038             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1039             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1040             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1041             Heps             = _mm256_mul_pd(vfeps,H);
1042             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1043             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1044             fvdw12           = _mm256_mul_pd(c12_00,FF);
1045             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1046
1047             fscal            = _mm256_add_pd(felec,fvdw);
1048
1049             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1050
1051             /* Calculate temporary vectorial force */
1052             tx               = _mm256_mul_pd(fscal,dx00);
1053             ty               = _mm256_mul_pd(fscal,dy00);
1054             tz               = _mm256_mul_pd(fscal,dz00);
1055
1056             /* Update vectorial force */
1057             fix0             = _mm256_add_pd(fix0,tx);
1058             fiy0             = _mm256_add_pd(fiy0,ty);
1059             fiz0             = _mm256_add_pd(fiz0,tz);
1060
1061             fjx0             = _mm256_add_pd(fjx0,tx);
1062             fjy0             = _mm256_add_pd(fjy0,ty);
1063             fjz0             = _mm256_add_pd(fjz0,tz);
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             /* Compute parameters for interactions between i and j atoms */
1070             qq10             = _mm256_mul_pd(iq1,jq0);
1071
1072             /* COULOMB ELECTROSTATICS */
1073             velec            = _mm256_mul_pd(qq10,rinv10);
1074             felec            = _mm256_mul_pd(velec,rinvsq10);
1075
1076             fscal            = felec;
1077
1078             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1079
1080             /* Calculate temporary vectorial force */
1081             tx               = _mm256_mul_pd(fscal,dx10);
1082             ty               = _mm256_mul_pd(fscal,dy10);
1083             tz               = _mm256_mul_pd(fscal,dz10);
1084
1085             /* Update vectorial force */
1086             fix1             = _mm256_add_pd(fix1,tx);
1087             fiy1             = _mm256_add_pd(fiy1,ty);
1088             fiz1             = _mm256_add_pd(fiz1,tz);
1089
1090             fjx0             = _mm256_add_pd(fjx0,tx);
1091             fjy0             = _mm256_add_pd(fjy0,ty);
1092             fjz0             = _mm256_add_pd(fjz0,tz);
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             /* Compute parameters for interactions between i and j atoms */
1099             qq20             = _mm256_mul_pd(iq2,jq0);
1100
1101             /* COULOMB ELECTROSTATICS */
1102             velec            = _mm256_mul_pd(qq20,rinv20);
1103             felec            = _mm256_mul_pd(velec,rinvsq20);
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 109 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*109);
1150 }