src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwLJ_GeomW3P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_256_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  90     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  91     __m256           dummy_mask,cutoff_mask;
  92     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  93     __m256           one     = _mm256_set1_ps(1.0);
  94     __m256           two     = _mm256_set1_ps(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm256_set1_ps(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108     nvdwtype         = fr->ntype;
 109     vdwparam         = fr->nbfp;
 110     vdwtype          = mdatoms->typeA;
 111
 112     /* Setup water-specific parameters */
 113     inr              = nlist->iinr[0];
 114     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 115     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 116     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 117     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123     j_coord_offsetC = 0;
 124     j_coord_offsetD = 0;
 125     j_coord_offsetE = 0;
 126     j_coord_offsetF = 0;
 127     j_coord_offsetG = 0;
 128     j_coord_offsetH = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     for(iidx=0;iidx<4*DIM;iidx++)
 134     {
 135         scratch[iidx] = 0.0;
 136     }
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 154                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 155
 156         fix0             = _mm256_setzero_ps();
 157         fiy0             = _mm256_setzero_ps();
 158         fiz0             = _mm256_setzero_ps();
 159         fix1             = _mm256_setzero_ps();
 160         fiy1             = _mm256_setzero_ps();
 161         fiz1             = _mm256_setzero_ps();
 162         fix2             = _mm256_setzero_ps();
 163         fiy2             = _mm256_setzero_ps();
 164         fiz2             = _mm256_setzero_ps();
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm256_setzero_ps();
 168         vvdwsum          = _mm256_setzero_ps();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             jnrC             = jjnr[jidx+2];
 178             jnrD             = jjnr[jidx+3];
 179             jnrE             = jjnr[jidx+4];
 180             jnrF             = jjnr[jidx+5];
 181             jnrG             = jjnr[jidx+6];
 182             jnrH             = jjnr[jidx+7];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185             j_coord_offsetC  = DIM*jnrC;
 186             j_coord_offsetD  = DIM*jnrD;
 187             j_coord_offsetE  = DIM*jnrE;
 188             j_coord_offsetF  = DIM*jnrF;
 189             j_coord_offsetG  = DIM*jnrG;
 190             j_coord_offsetH  = DIM*jnrH;
 191
 192             /* load j atom coordinates */
 193             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 195                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 196                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 197                                                  &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm256_sub_ps(ix0,jx0);
 201             dy00             = _mm256_sub_ps(iy0,jy0);
 202             dz00             = _mm256_sub_ps(iz0,jz0);
 203             dx10             = _mm256_sub_ps(ix1,jx0);
 204             dy10             = _mm256_sub_ps(iy1,jy0);
 205             dz10             = _mm256_sub_ps(iz1,jz0);
 206             dx20             = _mm256_sub_ps(ix2,jx0);
 207             dy20             = _mm256_sub_ps(iy2,jy0);
 208             dz20             = _mm256_sub_ps(iz2,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 212             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 213             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 214
 215             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 216             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 217             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 218
 219             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 220             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 221             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 222
 223             /* Load parameters for j particles */
 224             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 225                                                                  charge+jnrC+0,charge+jnrD+0,
 226                                                                  charge+jnrE+0,charge+jnrF+0,
 227                                                                  charge+jnrG+0,charge+jnrH+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230             vdwjidx0C        = 2*vdwtype[jnrC+0];
 231             vdwjidx0D        = 2*vdwtype[jnrD+0];
 232             vdwjidx0E        = 2*vdwtype[jnrE+0];
 233             vdwjidx0F        = 2*vdwtype[jnrF+0];
 234             vdwjidx0G        = 2*vdwtype[jnrG+0];
 235             vdwjidx0H        = 2*vdwtype[jnrH+0];
 236
 237             fjx0             = _mm256_setzero_ps();
 238             fjy0             = _mm256_setzero_ps();
 239             fjz0             = _mm256_setzero_ps();
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq00             = _mm256_mul_ps(iq0,jq0);
 247             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 248                                             vdwioffsetptr0+vdwjidx0B,
 249                                             vdwioffsetptr0+vdwjidx0C,
 250                                             vdwioffsetptr0+vdwjidx0D,
 251                                             vdwioffsetptr0+vdwjidx0E,
 252                                             vdwioffsetptr0+vdwjidx0F,
 253                                             vdwioffsetptr0+vdwjidx0G,
 254                                             vdwioffsetptr0+vdwjidx0H,
 255                                             &c6_00,&c12_00);
 256
 257             /* COULOMB ELECTROSTATICS */
 258             velec            = _mm256_mul_ps(qq00,rinv00);
 259             felec            = _mm256_mul_ps(velec,rinvsq00);
 260
 261             /* LENNARD-JONES DISPERSION/REPULSION */
 262
 263             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 264             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 265             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 266             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 267             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 268
 269             /* Update potential sum for this i atom from the interaction with this j atom. */
 270             velecsum         = _mm256_add_ps(velecsum,velec);
 271             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 272
 273             fscal            = _mm256_add_ps(felec,fvdw);
 274
 275             /* Calculate temporary vectorial force */
 276             tx               = _mm256_mul_ps(fscal,dx00);
 277             ty               = _mm256_mul_ps(fscal,dy00);
 278             tz               = _mm256_mul_ps(fscal,dz00);
 279
 280             /* Update vectorial force */
 281             fix0             = _mm256_add_ps(fix0,tx);
 282             fiy0             = _mm256_add_ps(fiy0,ty);
 283             fiz0             = _mm256_add_ps(fiz0,tz);
 284
 285             fjx0             = _mm256_add_ps(fjx0,tx);
 286             fjy0             = _mm256_add_ps(fjy0,ty);
 287             fjz0             = _mm256_add_ps(fjz0,tz);
 288
 289             /**************************
 290              * CALCULATE INTERACTIONS *
 291              **************************/
 292
 293             /* Compute parameters for interactions between i and j atoms */
 294             qq10             = _mm256_mul_ps(iq1,jq0);
 295
 296             /* COULOMB ELECTROSTATICS */
 297             velec            = _mm256_mul_ps(qq10,rinv10);
 298             felec            = _mm256_mul_ps(velec,rinvsq10);
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velecsum         = _mm256_add_ps(velecsum,velec);
 302
 303             fscal            = felec;
 304
 305             /* Calculate temporary vectorial force */
 306             tx               = _mm256_mul_ps(fscal,dx10);
 307             ty               = _mm256_mul_ps(fscal,dy10);
 308             tz               = _mm256_mul_ps(fscal,dz10);
 309
 310             /* Update vectorial force */
 311             fix1             = _mm256_add_ps(fix1,tx);
 312             fiy1             = _mm256_add_ps(fiy1,ty);
 313             fiz1             = _mm256_add_ps(fiz1,tz);
 314
 315             fjx0             = _mm256_add_ps(fjx0,tx);
 316             fjy0             = _mm256_add_ps(fjy0,ty);
 317             fjz0             = _mm256_add_ps(fjz0,tz);
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq20             = _mm256_mul_ps(iq2,jq0);
 325
 326             /* COULOMB ELECTROSTATICS */
 327             velec            = _mm256_mul_ps(qq20,rinv20);
 328             felec            = _mm256_mul_ps(velec,rinvsq20);
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velecsum         = _mm256_add_ps(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm256_mul_ps(fscal,dx20);
 337             ty               = _mm256_mul_ps(fscal,dy20);
 338             tz               = _mm256_mul_ps(fscal,dz20);
 339
 340             /* Update vectorial force */
 341             fix2             = _mm256_add_ps(fix2,tx);
 342             fiy2             = _mm256_add_ps(fiy2,ty);
 343             fiz2             = _mm256_add_ps(fiz2,tz);
 344
 345             fjx0             = _mm256_add_ps(fjx0,tx);
 346             fjy0             = _mm256_add_ps(fjy0,ty);
 347             fjz0             = _mm256_add_ps(fjz0,tz);
 348
 349             fjptrA             = f+j_coord_offsetA;
 350             fjptrB             = f+j_coord_offsetB;
 351             fjptrC             = f+j_coord_offsetC;
 352             fjptrD             = f+j_coord_offsetD;
 353             fjptrE             = f+j_coord_offsetE;
 354             fjptrF             = f+j_coord_offsetF;
 355             fjptrG             = f+j_coord_offsetG;
 356             fjptrH             = f+j_coord_offsetH;
 357
 358             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 359
 360             /* Inner loop uses 96 flops */
 361         }
 362
 363         if(jidx<j_index_end)
 364         {
 365
 366             /* Get j neighbor index, and coordinate index */
 367             jnrlistA         = jjnr[jidx];
 368             jnrlistB         = jjnr[jidx+1];
 369             jnrlistC         = jjnr[jidx+2];
 370             jnrlistD         = jjnr[jidx+3];
 371             jnrlistE         = jjnr[jidx+4];
 372             jnrlistF         = jjnr[jidx+5];
 373             jnrlistG         = jjnr[jidx+6];
 374             jnrlistH         = jjnr[jidx+7];
 375             /* Sign of each element will be negative for non-real atoms.
 376              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 377              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 378              */
 379             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 380                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 381
 382             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 383             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 384             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 385             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 386             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 387             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 388             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 389             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 390             j_coord_offsetA  = DIM*jnrA;
 391             j_coord_offsetB  = DIM*jnrB;
 392             j_coord_offsetC  = DIM*jnrC;
 393             j_coord_offsetD  = DIM*jnrD;
 394             j_coord_offsetE  = DIM*jnrE;
 395             j_coord_offsetF  = DIM*jnrF;
 396             j_coord_offsetG  = DIM*jnrG;
 397             j_coord_offsetH  = DIM*jnrH;
 398
 399             /* load j atom coordinates */
 400             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 401                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 402                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 403                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 404                                                  &jx0,&jy0,&jz0);
 405
 406             /* Calculate displacement vector */
 407             dx00             = _mm256_sub_ps(ix0,jx0);
 408             dy00             = _mm256_sub_ps(iy0,jy0);
 409             dz00             = _mm256_sub_ps(iz0,jz0);
 410             dx10             = _mm256_sub_ps(ix1,jx0);
 411             dy10             = _mm256_sub_ps(iy1,jy0);
 412             dz10             = _mm256_sub_ps(iz1,jz0);
 413             dx20             = _mm256_sub_ps(ix2,jx0);
 414             dy20             = _mm256_sub_ps(iy2,jy0);
 415             dz20             = _mm256_sub_ps(iz2,jz0);
 416
 417             /* Calculate squared distance and things based on it */
 418             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 419             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 420             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 421
 422             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 423             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 424             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 425
 426             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 427             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 428             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 429
 430             /* Load parameters for j particles */
 431             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 432                                                                  charge+jnrC+0,charge+jnrD+0,
 433                                                                  charge+jnrE+0,charge+jnrF+0,
 434                                                                  charge+jnrG+0,charge+jnrH+0);
 435             vdwjidx0A        = 2*vdwtype[jnrA+0];
 436             vdwjidx0B        = 2*vdwtype[jnrB+0];
 437             vdwjidx0C        = 2*vdwtype[jnrC+0];
 438             vdwjidx0D        = 2*vdwtype[jnrD+0];
 439             vdwjidx0E        = 2*vdwtype[jnrE+0];
 440             vdwjidx0F        = 2*vdwtype[jnrF+0];
 441             vdwjidx0G        = 2*vdwtype[jnrG+0];
 442             vdwjidx0H        = 2*vdwtype[jnrH+0];
 443
 444             fjx0             = _mm256_setzero_ps();
 445             fjy0             = _mm256_setzero_ps();
 446             fjz0             = _mm256_setzero_ps();
 447
 448             /**************************
 449              * CALCULATE INTERACTIONS *
 450              **************************/
 451
 452             /* Compute parameters for interactions between i and j atoms */
 453             qq00             = _mm256_mul_ps(iq0,jq0);
 454             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 455                                             vdwioffsetptr0+vdwjidx0B,
 456                                             vdwioffsetptr0+vdwjidx0C,
 457                                             vdwioffsetptr0+vdwjidx0D,
 458                                             vdwioffsetptr0+vdwjidx0E,
 459                                             vdwioffsetptr0+vdwjidx0F,
 460                                             vdwioffsetptr0+vdwjidx0G,
 461                                             vdwioffsetptr0+vdwjidx0H,
 462                                             &c6_00,&c12_00);
 463
 464             /* COULOMB ELECTROSTATICS */
 465             velec            = _mm256_mul_ps(qq00,rinv00);
 466             felec            = _mm256_mul_ps(velec,rinvsq00);
 467
 468             /* LENNARD-JONES DISPERSION/REPULSION */
 469
 470             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 471             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 472             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 473             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 474             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 475
 476             /* Update potential sum for this i atom from the interaction with this j atom. */
 477             velec            = _mm256_andnot_ps(dummy_mask,velec);
 478             velecsum         = _mm256_add_ps(velecsum,velec);
 479             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 480             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 481
 482             fscal            = _mm256_add_ps(felec,fvdw);
 483
 484             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 485
 486             /* Calculate temporary vectorial force */
 487             tx               = _mm256_mul_ps(fscal,dx00);
 488             ty               = _mm256_mul_ps(fscal,dy00);
 489             tz               = _mm256_mul_ps(fscal,dz00);
 490
 491             /* Update vectorial force */
 492             fix0             = _mm256_add_ps(fix0,tx);
 493             fiy0             = _mm256_add_ps(fiy0,ty);
 494             fiz0             = _mm256_add_ps(fiz0,tz);
 495
 496             fjx0             = _mm256_add_ps(fjx0,tx);
 497             fjy0             = _mm256_add_ps(fjy0,ty);
 498             fjz0             = _mm256_add_ps(fjz0,tz);
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             /* Compute parameters for interactions between i and j atoms */
 505             qq10             = _mm256_mul_ps(iq1,jq0);
 506
 507             /* COULOMB ELECTROSTATICS */
 508             velec            = _mm256_mul_ps(qq10,rinv10);
 509             felec            = _mm256_mul_ps(velec,rinvsq10);
 510
 511             /* Update potential sum for this i atom from the interaction with this j atom. */
 512             velec            = _mm256_andnot_ps(dummy_mask,velec);
 513             velecsum         = _mm256_add_ps(velecsum,velec);
 514
 515             fscal            = felec;
 516
 517             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 518
 519             /* Calculate temporary vectorial force */
 520             tx               = _mm256_mul_ps(fscal,dx10);
 521             ty               = _mm256_mul_ps(fscal,dy10);
 522             tz               = _mm256_mul_ps(fscal,dz10);
 523
 524             /* Update vectorial force */
 525             fix1             = _mm256_add_ps(fix1,tx);
 526             fiy1             = _mm256_add_ps(fiy1,ty);
 527             fiz1             = _mm256_add_ps(fiz1,tz);
 528
 529             fjx0             = _mm256_add_ps(fjx0,tx);
 530             fjy0             = _mm256_add_ps(fjy0,ty);
 531             fjz0             = _mm256_add_ps(fjz0,tz);
 532
 533             /**************************
 534              * CALCULATE INTERACTIONS *
 535              **************************/
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq20             = _mm256_mul_ps(iq2,jq0);
 539
 540             /* COULOMB ELECTROSTATICS */
 541             velec            = _mm256_mul_ps(qq20,rinv20);
 542             felec            = _mm256_mul_ps(velec,rinvsq20);
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             velec            = _mm256_andnot_ps(dummy_mask,velec);
 546             velecsum         = _mm256_add_ps(velecsum,velec);
 547
 548             fscal            = felec;
 549
 550             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm256_mul_ps(fscal,dx20);
 554             ty               = _mm256_mul_ps(fscal,dy20);
 555             tz               = _mm256_mul_ps(fscal,dz20);
 556
 557             /* Update vectorial force */
 558             fix2             = _mm256_add_ps(fix2,tx);
 559             fiy2             = _mm256_add_ps(fiy2,ty);
 560             fiz2             = _mm256_add_ps(fiz2,tz);
 561
 562             fjx0             = _mm256_add_ps(fjx0,tx);
 563             fjy0             = _mm256_add_ps(fjy0,ty);
 564             fjz0             = _mm256_add_ps(fjz0,tz);
 565
 566             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 567             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 568             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 569             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 570             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 571             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 572             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 573             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 574
 575             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 576
 577             /* Inner loop uses 96 flops */
 578         }
 579
 580         /* End of innermost loop */
 581
 582         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 583                                                  f+i_coord_offset,fshift+i_shift_offset);
 584
 585         ggid                        = gid[iidx];
 586         /* Update potential energies */
 587         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 588         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 589
 590         /* Increment number of inner iterations */
 591         inneriter                  += j_index_end - j_index_start;
 592
 593         /* Outer loop uses 20 flops */
 594     }
 595
 596     /* Increment number of outer iterations */
 597     outeriter        += nri;
 598
 599     /* Update outer/inner flops */
 600
 601     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*96);
 602 }
 603 /*
 604  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_single
 605  * Electrostatics interaction: Coulomb
 606  * VdW interaction:            LennardJones
 607  * Geometry:                   Water3-Particle
 608  * Calculate force/pot:        Force
 609  */
 610 void
 611 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_single
 612                     (t_nblist * gmx_restrict                nlist,
 613                      rvec * gmx_restrict                    xx,
 614                      rvec * gmx_restrict                    ff,
 615                      t_forcerec * gmx_restrict              fr,
 616                      t_mdatoms * gmx_restrict               mdatoms,
 617                      nb_kernel_data_t * gmx_restrict        kernel_data,
 618                      t_nrnb * gmx_restrict                  nrnb)
 619 {
 620     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 621      * just 0 for non-waters.
 622      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 623      * jnr indices corresponding to data put in the four positions in the SIMD register.
 624      */
 625     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 626     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 627     int              jnrA,jnrB,jnrC,jnrD;
 628     int              jnrE,jnrF,jnrG,jnrH;
 629     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 630     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 631     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 632     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 633     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 634     real             rcutoff_scalar;
 635     real             *shiftvec,*fshift,*x,*f;
 636     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 637     real             scratch[4*DIM];
 638     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 639     real *           vdwioffsetptr0;
 640     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 641     real *           vdwioffsetptr1;
 642     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 643     real *           vdwioffsetptr2;
 644     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 645     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 646     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 647     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 648     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 649     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 650     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 651     real             *charge;
 652     int              nvdwtype;
 653     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 654     int              *vdwtype;
 655     real             *vdwparam;
 656     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 657     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 658     __m256           dummy_mask,cutoff_mask;
 659     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 660     __m256           one     = _mm256_set1_ps(1.0);
 661     __m256           two     = _mm256_set1_ps(2.0);
 662     x                = xx[0];
 663     f                = ff[0];
 664
 665     nri              = nlist->nri;
 666     iinr             = nlist->iinr;
 667     jindex           = nlist->jindex;
 668     jjnr             = nlist->jjnr;
 669     shiftidx         = nlist->shift;
 670     gid              = nlist->gid;
 671     shiftvec         = fr->shift_vec[0];
 672     fshift           = fr->fshift[0];
 673     facel            = _mm256_set1_ps(fr->epsfac);
 674     charge           = mdatoms->chargeA;
 675     nvdwtype         = fr->ntype;
 676     vdwparam         = fr->nbfp;
 677     vdwtype          = mdatoms->typeA;
 678
 679     /* Setup water-specific parameters */
 680     inr              = nlist->iinr[0];
 681     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 682     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 683     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 684     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 685
 686     /* Avoid stupid compiler warnings */
 687     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 688     j_coord_offsetA = 0;
 689     j_coord_offsetB = 0;
 690     j_coord_offsetC = 0;
 691     j_coord_offsetD = 0;
 692     j_coord_offsetE = 0;
 693     j_coord_offsetF = 0;
 694     j_coord_offsetG = 0;
 695     j_coord_offsetH = 0;
 696
 697     outeriter        = 0;
 698     inneriter        = 0;
 699
 700     for(iidx=0;iidx<4*DIM;iidx++)
 701     {
 702         scratch[iidx] = 0.0;
 703     }
 704
 705     /* Start outer loop over neighborlists */
 706     for(iidx=0; iidx<nri; iidx++)
 707     {
 708         /* Load shift vector for this list */
 709         i_shift_offset   = DIM*shiftidx[iidx];
 710
 711         /* Load limits for loop over neighbors */
 712         j_index_start    = jindex[iidx];
 713         j_index_end      = jindex[iidx+1];
 714
 715         /* Get outer coordinate index */
 716         inr              = iinr[iidx];
 717         i_coord_offset   = DIM*inr;
 718
 719         /* Load i particle coords and add shift vector */
 720         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 721                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 722
 723         fix0             = _mm256_setzero_ps();
 724         fiy0             = _mm256_setzero_ps();
 725         fiz0             = _mm256_setzero_ps();
 726         fix1             = _mm256_setzero_ps();
 727         fiy1             = _mm256_setzero_ps();
 728         fiz1             = _mm256_setzero_ps();
 729         fix2             = _mm256_setzero_ps();
 730         fiy2             = _mm256_setzero_ps();
 731         fiz2             = _mm256_setzero_ps();
 732
 733         /* Start inner kernel loop */
 734         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 735         {
 736
 737             /* Get j neighbor index, and coordinate index */
 738             jnrA             = jjnr[jidx];
 739             jnrB             = jjnr[jidx+1];
 740             jnrC             = jjnr[jidx+2];
 741             jnrD             = jjnr[jidx+3];
 742             jnrE             = jjnr[jidx+4];
 743             jnrF             = jjnr[jidx+5];
 744             jnrG             = jjnr[jidx+6];
 745             jnrH             = jjnr[jidx+7];
 746             j_coord_offsetA  = DIM*jnrA;
 747             j_coord_offsetB  = DIM*jnrB;
 748             j_coord_offsetC  = DIM*jnrC;
 749             j_coord_offsetD  = DIM*jnrD;
 750             j_coord_offsetE  = DIM*jnrE;
 751             j_coord_offsetF  = DIM*jnrF;
 752             j_coord_offsetG  = DIM*jnrG;
 753             j_coord_offsetH  = DIM*jnrH;
 754
 755             /* load j atom coordinates */
 756             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 757                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 758                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 759                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 760                                                  &jx0,&jy0,&jz0);
 761
 762             /* Calculate displacement vector */
 763             dx00             = _mm256_sub_ps(ix0,jx0);
 764             dy00             = _mm256_sub_ps(iy0,jy0);
 765             dz00             = _mm256_sub_ps(iz0,jz0);
 766             dx10             = _mm256_sub_ps(ix1,jx0);
 767             dy10             = _mm256_sub_ps(iy1,jy0);
 768             dz10             = _mm256_sub_ps(iz1,jz0);
 769             dx20             = _mm256_sub_ps(ix2,jx0);
 770             dy20             = _mm256_sub_ps(iy2,jy0);
 771             dz20             = _mm256_sub_ps(iz2,jz0);
 772
 773             /* Calculate squared distance and things based on it */
 774             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 775             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 776             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 777
 778             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 779             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 780             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 781
 782             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 783             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 784             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 785
 786             /* Load parameters for j particles */
 787             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 788                                                                  charge+jnrC+0,charge+jnrD+0,
 789                                                                  charge+jnrE+0,charge+jnrF+0,
 790                                                                  charge+jnrG+0,charge+jnrH+0);
 791             vdwjidx0A        = 2*vdwtype[jnrA+0];
 792             vdwjidx0B        = 2*vdwtype[jnrB+0];
 793             vdwjidx0C        = 2*vdwtype[jnrC+0];
 794             vdwjidx0D        = 2*vdwtype[jnrD+0];
 795             vdwjidx0E        = 2*vdwtype[jnrE+0];
 796             vdwjidx0F        = 2*vdwtype[jnrF+0];
 797             vdwjidx0G        = 2*vdwtype[jnrG+0];
 798             vdwjidx0H        = 2*vdwtype[jnrH+0];
 799
 800             fjx0             = _mm256_setzero_ps();
 801             fjy0             = _mm256_setzero_ps();
 802             fjz0             = _mm256_setzero_ps();
 803
 804             /**************************
 805              * CALCULATE INTERACTIONS *
 806              **************************/
 807
 808             /* Compute parameters for interactions between i and j atoms */
 809             qq00             = _mm256_mul_ps(iq0,jq0);
 810             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 811                                             vdwioffsetptr0+vdwjidx0B,
 812                                             vdwioffsetptr0+vdwjidx0C,
 813                                             vdwioffsetptr0+vdwjidx0D,
 814                                             vdwioffsetptr0+vdwjidx0E,
 815                                             vdwioffsetptr0+vdwjidx0F,
 816                                             vdwioffsetptr0+vdwjidx0G,
 817                                             vdwioffsetptr0+vdwjidx0H,
 818                                             &c6_00,&c12_00);
 819
 820             /* COULOMB ELECTROSTATICS */
 821             velec            = _mm256_mul_ps(qq00,rinv00);
 822             felec            = _mm256_mul_ps(velec,rinvsq00);
 823
 824             /* LENNARD-JONES DISPERSION/REPULSION */
 825
 826             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 827             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 828
 829             fscal            = _mm256_add_ps(felec,fvdw);
 830
 831             /* Calculate temporary vectorial force */
 832             tx               = _mm256_mul_ps(fscal,dx00);
 833             ty               = _mm256_mul_ps(fscal,dy00);
 834             tz               = _mm256_mul_ps(fscal,dz00);
 835
 836             /* Update vectorial force */
 837             fix0             = _mm256_add_ps(fix0,tx);
 838             fiy0             = _mm256_add_ps(fiy0,ty);
 839             fiz0             = _mm256_add_ps(fiz0,tz);
 840
 841             fjx0             = _mm256_add_ps(fjx0,tx);
 842             fjy0             = _mm256_add_ps(fjy0,ty);
 843             fjz0             = _mm256_add_ps(fjz0,tz);
 844
 845             /**************************
 846              * CALCULATE INTERACTIONS *
 847              **************************/
 848
 849             /* Compute parameters for interactions between i and j atoms */
 850             qq10             = _mm256_mul_ps(iq1,jq0);
 851
 852             /* COULOMB ELECTROSTATICS */
 853             velec            = _mm256_mul_ps(qq10,rinv10);
 854             felec            = _mm256_mul_ps(velec,rinvsq10);
 855
 856             fscal            = felec;
 857
 858             /* Calculate temporary vectorial force */
 859             tx               = _mm256_mul_ps(fscal,dx10);
 860             ty               = _mm256_mul_ps(fscal,dy10);
 861             tz               = _mm256_mul_ps(fscal,dz10);
 862
 863             /* Update vectorial force */
 864             fix1             = _mm256_add_ps(fix1,tx);
 865             fiy1             = _mm256_add_ps(fiy1,ty);
 866             fiz1             = _mm256_add_ps(fiz1,tz);
 867
 868             fjx0             = _mm256_add_ps(fjx0,tx);
 869             fjy0             = _mm256_add_ps(fjy0,ty);
 870             fjz0             = _mm256_add_ps(fjz0,tz);
 871
 872             /**************************
 873              * CALCULATE INTERACTIONS *
 874              **************************/
 875
 876             /* Compute parameters for interactions between i and j atoms */
 877             qq20             = _mm256_mul_ps(iq2,jq0);
 878
 879             /* COULOMB ELECTROSTATICS */
 880             velec            = _mm256_mul_ps(qq20,rinv20);
 881             felec            = _mm256_mul_ps(velec,rinvsq20);
 882
 883             fscal            = felec;
 884
 885             /* Calculate temporary vectorial force */
 886             tx               = _mm256_mul_ps(fscal,dx20);
 887             ty               = _mm256_mul_ps(fscal,dy20);
 888             tz               = _mm256_mul_ps(fscal,dz20);
 889
 890             /* Update vectorial force */
 891             fix2             = _mm256_add_ps(fix2,tx);
 892             fiy2             = _mm256_add_ps(fiy2,ty);
 893             fiz2             = _mm256_add_ps(fiz2,tz);
 894
 895             fjx0             = _mm256_add_ps(fjx0,tx);
 896             fjy0             = _mm256_add_ps(fjy0,ty);
 897             fjz0             = _mm256_add_ps(fjz0,tz);
 898
 899             fjptrA             = f+j_coord_offsetA;
 900             fjptrB             = f+j_coord_offsetB;
 901             fjptrC             = f+j_coord_offsetC;
 902             fjptrD             = f+j_coord_offsetD;
 903             fjptrE             = f+j_coord_offsetE;
 904             fjptrF             = f+j_coord_offsetF;
 905             fjptrG             = f+j_coord_offsetG;
 906             fjptrH             = f+j_coord_offsetH;
 907
 908             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 909
 910             /* Inner loop uses 88 flops */
 911         }
 912
 913         if(jidx<j_index_end)
 914         {
 915
 916             /* Get j neighbor index, and coordinate index */
 917             jnrlistA         = jjnr[jidx];
 918             jnrlistB         = jjnr[jidx+1];
 919             jnrlistC         = jjnr[jidx+2];
 920             jnrlistD         = jjnr[jidx+3];
 921             jnrlistE         = jjnr[jidx+4];
 922             jnrlistF         = jjnr[jidx+5];
 923             jnrlistG         = jjnr[jidx+6];
 924             jnrlistH         = jjnr[jidx+7];
 925             /* Sign of each element will be negative for non-real atoms.
 926              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 927              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 928              */
 929             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 930                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 931
 932             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 933             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 934             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 935             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 936             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 937             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 938             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 939             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 940             j_coord_offsetA  = DIM*jnrA;
 941             j_coord_offsetB  = DIM*jnrB;
 942             j_coord_offsetC  = DIM*jnrC;
 943             j_coord_offsetD  = DIM*jnrD;
 944             j_coord_offsetE  = DIM*jnrE;
 945             j_coord_offsetF  = DIM*jnrF;
 946             j_coord_offsetG  = DIM*jnrG;
 947             j_coord_offsetH  = DIM*jnrH;
 948
 949             /* load j atom coordinates */
 950             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 951                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 952                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 953                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 954                                                  &jx0,&jy0,&jz0);
 955
 956             /* Calculate displacement vector */
 957             dx00             = _mm256_sub_ps(ix0,jx0);
 958             dy00             = _mm256_sub_ps(iy0,jy0);
 959             dz00             = _mm256_sub_ps(iz0,jz0);
 960             dx10             = _mm256_sub_ps(ix1,jx0);
 961             dy10             = _mm256_sub_ps(iy1,jy0);
 962             dz10             = _mm256_sub_ps(iz1,jz0);
 963             dx20             = _mm256_sub_ps(ix2,jx0);
 964             dy20             = _mm256_sub_ps(iy2,jy0);
 965             dz20             = _mm256_sub_ps(iz2,jz0);
 966
 967             /* Calculate squared distance and things based on it */
 968             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 969             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 970             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 971
 972             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 973             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 974             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 975
 976             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 977             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 978             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 979
 980             /* Load parameters for j particles */
 981             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 982                                                                  charge+jnrC+0,charge+jnrD+0,
 983                                                                  charge+jnrE+0,charge+jnrF+0,
 984                                                                  charge+jnrG+0,charge+jnrH+0);
 985             vdwjidx0A        = 2*vdwtype[jnrA+0];
 986             vdwjidx0B        = 2*vdwtype[jnrB+0];
 987             vdwjidx0C        = 2*vdwtype[jnrC+0];
 988             vdwjidx0D        = 2*vdwtype[jnrD+0];
 989             vdwjidx0E        = 2*vdwtype[jnrE+0];
 990             vdwjidx0F        = 2*vdwtype[jnrF+0];
 991             vdwjidx0G        = 2*vdwtype[jnrG+0];
 992             vdwjidx0H        = 2*vdwtype[jnrH+0];
 993
 994             fjx0             = _mm256_setzero_ps();
 995             fjy0             = _mm256_setzero_ps();
 996             fjz0             = _mm256_setzero_ps();
 997
 998             /**************************
 999              * CALCULATE INTERACTIONS *
1000              **************************/
1001
1002             /* Compute parameters for interactions between i and j atoms */
1003             qq00             = _mm256_mul_ps(iq0,jq0);
1004             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1005                                             vdwioffsetptr0+vdwjidx0B,
1006                                             vdwioffsetptr0+vdwjidx0C,
1007                                             vdwioffsetptr0+vdwjidx0D,
1008                                             vdwioffsetptr0+vdwjidx0E,
1009                                             vdwioffsetptr0+vdwjidx0F,
1010                                             vdwioffsetptr0+vdwjidx0G,
1011                                             vdwioffsetptr0+vdwjidx0H,
1012                                             &c6_00,&c12_00);
1013
1014             /* COULOMB ELECTROSTATICS */
1015             velec            = _mm256_mul_ps(qq00,rinv00);
1016             felec            = _mm256_mul_ps(velec,rinvsq00);
1017
1018             /* LENNARD-JONES DISPERSION/REPULSION */
1019
1020             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1021             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1022
1023             fscal            = _mm256_add_ps(felec,fvdw);
1024
1025             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1026
1027             /* Calculate temporary vectorial force */
1028             tx               = _mm256_mul_ps(fscal,dx00);
1029             ty               = _mm256_mul_ps(fscal,dy00);
1030             tz               = _mm256_mul_ps(fscal,dz00);
1031
1032             /* Update vectorial force */
1033             fix0             = _mm256_add_ps(fix0,tx);
1034             fiy0             = _mm256_add_ps(fiy0,ty);
1035             fiz0             = _mm256_add_ps(fiz0,tz);
1036
1037             fjx0             = _mm256_add_ps(fjx0,tx);
1038             fjy0             = _mm256_add_ps(fjy0,ty);
1039             fjz0             = _mm256_add_ps(fjz0,tz);
1040
1041             /**************************
1042              * CALCULATE INTERACTIONS *
1043              **************************/
1044
1045             /* Compute parameters for interactions between i and j atoms */
1046             qq10             = _mm256_mul_ps(iq1,jq0);
1047
1048             /* COULOMB ELECTROSTATICS */
1049             velec            = _mm256_mul_ps(qq10,rinv10);
1050             felec            = _mm256_mul_ps(velec,rinvsq10);
1051
1052             fscal            = felec;
1053
1054             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1055
1056             /* Calculate temporary vectorial force */
1057             tx               = _mm256_mul_ps(fscal,dx10);
1058             ty               = _mm256_mul_ps(fscal,dy10);
1059             tz               = _mm256_mul_ps(fscal,dz10);
1060
1061             /* Update vectorial force */
1062             fix1             = _mm256_add_ps(fix1,tx);
1063             fiy1             = _mm256_add_ps(fiy1,ty);
1064             fiz1             = _mm256_add_ps(fiz1,tz);
1065
1066             fjx0             = _mm256_add_ps(fjx0,tx);
1067             fjy0             = _mm256_add_ps(fjy0,ty);
1068             fjz0             = _mm256_add_ps(fjz0,tz);
1069
1070             /**************************
1071              * CALCULATE INTERACTIONS *
1072              **************************/
1073
1074             /* Compute parameters for interactions between i and j atoms */
1075             qq20             = _mm256_mul_ps(iq2,jq0);
1076
1077             /* COULOMB ELECTROSTATICS */
1078             velec            = _mm256_mul_ps(qq20,rinv20);
1079             felec            = _mm256_mul_ps(velec,rinvsq20);
1080
1081             fscal            = felec;
1082
1083             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1084
1085             /* Calculate temporary vectorial force */
1086             tx               = _mm256_mul_ps(fscal,dx20);
1087             ty               = _mm256_mul_ps(fscal,dy20);
1088             tz               = _mm256_mul_ps(fscal,dz20);
1089
1090             /* Update vectorial force */
1091             fix2             = _mm256_add_ps(fix2,tx);
1092             fiy2             = _mm256_add_ps(fiy2,ty);
1093             fiz2             = _mm256_add_ps(fiz2,tz);
1094
1095             fjx0             = _mm256_add_ps(fjx0,tx);
1096             fjy0             = _mm256_add_ps(fjy0,ty);
1097             fjz0             = _mm256_add_ps(fjz0,tz);
1098
1099             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1100             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1101             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1102             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1103             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1104             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1105             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1106             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1107
1108             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1109
1110             /* Inner loop uses 88 flops */
1111         }
1112
1113         /* End of innermost loop */
1114
1115         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1116                                                  f+i_coord_offset,fshift+i_shift_offset);
1117
1118         /* Increment number of inner iterations */
1119         inneriter                  += j_index_end - j_index_start;
1120
1121         /* Outer loop uses 18 flops */
1122     }
1123
1124     /* Increment number of outer iterations */
1125     outeriter        += nri;
1126
1127     /* Update outer/inner flops */
1128
1129     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);
1130 }