src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_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_ElecRF_VdwLJ_GeomW3P1_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_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     krf              = _mm256_set1_ps(fr->ic->k_rf);
 109     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 110     crf              = _mm256_set1_ps(fr->ic->c_rf);
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     /* Setup water-specific parameters */
 116     inr              = nlist->iinr[0];
 117     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 118     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 119     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 120     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126     j_coord_offsetC = 0;
 127     j_coord_offsetD = 0;
 128     j_coord_offsetE = 0;
 129     j_coord_offsetF = 0;
 130     j_coord_offsetG = 0;
 131     j_coord_offsetH = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     for(iidx=0;iidx<4*DIM;iidx++)
 137     {
 138         scratch[iidx] = 0.0;
 139     }
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 157                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 158
 159         fix0             = _mm256_setzero_ps();
 160         fiy0             = _mm256_setzero_ps();
 161         fiz0             = _mm256_setzero_ps();
 162         fix1             = _mm256_setzero_ps();
 163         fiy1             = _mm256_setzero_ps();
 164         fiz1             = _mm256_setzero_ps();
 165         fix2             = _mm256_setzero_ps();
 166         fiy2             = _mm256_setzero_ps();
 167         fiz2             = _mm256_setzero_ps();
 168
 169         /* Reset potential sums */
 170         velecsum         = _mm256_setzero_ps();
 171         vvdwsum          = _mm256_setzero_ps();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             jnrC             = jjnr[jidx+2];
 181             jnrD             = jjnr[jidx+3];
 182             jnrE             = jjnr[jidx+4];
 183             jnrF             = jjnr[jidx+5];
 184             jnrG             = jjnr[jidx+6];
 185             jnrH             = jjnr[jidx+7];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188             j_coord_offsetC  = DIM*jnrC;
 189             j_coord_offsetD  = DIM*jnrD;
 190             j_coord_offsetE  = DIM*jnrE;
 191             j_coord_offsetF  = DIM*jnrF;
 192             j_coord_offsetG  = DIM*jnrG;
 193             j_coord_offsetH  = DIM*jnrH;
 194
 195             /* load j atom coordinates */
 196             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 198                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 199                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 200                                                  &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm256_sub_ps(ix0,jx0);
 204             dy00             = _mm256_sub_ps(iy0,jy0);
 205             dz00             = _mm256_sub_ps(iz0,jz0);
 206             dx10             = _mm256_sub_ps(ix1,jx0);
 207             dy10             = _mm256_sub_ps(iy1,jy0);
 208             dz10             = _mm256_sub_ps(iz1,jz0);
 209             dx20             = _mm256_sub_ps(ix2,jx0);
 210             dy20             = _mm256_sub_ps(iy2,jy0);
 211             dz20             = _mm256_sub_ps(iz2,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 215             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 216             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 217
 218             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 219             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 220             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 221
 222             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 223             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 224             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 228                                                                  charge+jnrC+0,charge+jnrD+0,
 229                                                                  charge+jnrE+0,charge+jnrF+0,
 230                                                                  charge+jnrG+0,charge+jnrH+0);
 231             vdwjidx0A        = 2*vdwtype[jnrA+0];
 232             vdwjidx0B        = 2*vdwtype[jnrB+0];
 233             vdwjidx0C        = 2*vdwtype[jnrC+0];
 234             vdwjidx0D        = 2*vdwtype[jnrD+0];
 235             vdwjidx0E        = 2*vdwtype[jnrE+0];
 236             vdwjidx0F        = 2*vdwtype[jnrF+0];
 237             vdwjidx0G        = 2*vdwtype[jnrG+0];
 238             vdwjidx0H        = 2*vdwtype[jnrH+0];
 239
 240             fjx0             = _mm256_setzero_ps();
 241             fjy0             = _mm256_setzero_ps();
 242             fjz0             = _mm256_setzero_ps();
 243
 244             /**************************
 245              * CALCULATE INTERACTIONS *
 246              **************************/
 247
 248             /* Compute parameters for interactions between i and j atoms */
 249             qq00             = _mm256_mul_ps(iq0,jq0);
 250             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 251                                             vdwioffsetptr0+vdwjidx0B,
 252                                             vdwioffsetptr0+vdwjidx0C,
 253                                             vdwioffsetptr0+vdwjidx0D,
 254                                             vdwioffsetptr0+vdwjidx0E,
 255                                             vdwioffsetptr0+vdwjidx0F,
 256                                             vdwioffsetptr0+vdwjidx0G,
 257                                             vdwioffsetptr0+vdwjidx0H,
 258                                             &c6_00,&c12_00);
 259
 260             /* REACTION-FIELD ELECTROSTATICS */
 261             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 262             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 263
 264             /* LENNARD-JONES DISPERSION/REPULSION */
 265
 266             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 267             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 268             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 269             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 270             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             velecsum         = _mm256_add_ps(velecsum,velec);
 274             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 275
 276             fscal            = _mm256_add_ps(felec,fvdw);
 277
 278             /* Calculate temporary vectorial force */
 279             tx               = _mm256_mul_ps(fscal,dx00);
 280             ty               = _mm256_mul_ps(fscal,dy00);
 281             tz               = _mm256_mul_ps(fscal,dz00);
 282
 283             /* Update vectorial force */
 284             fix0             = _mm256_add_ps(fix0,tx);
 285             fiy0             = _mm256_add_ps(fiy0,ty);
 286             fiz0             = _mm256_add_ps(fiz0,tz);
 287
 288             fjx0             = _mm256_add_ps(fjx0,tx);
 289             fjy0             = _mm256_add_ps(fjy0,ty);
 290             fjz0             = _mm256_add_ps(fjz0,tz);
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             /* Compute parameters for interactions between i and j atoms */
 297             qq10             = _mm256_mul_ps(iq1,jq0);
 298
 299             /* REACTION-FIELD ELECTROSTATICS */
 300             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 301             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             velecsum         = _mm256_add_ps(velecsum,velec);
 305
 306             fscal            = felec;
 307
 308             /* Calculate temporary vectorial force */
 309             tx               = _mm256_mul_ps(fscal,dx10);
 310             ty               = _mm256_mul_ps(fscal,dy10);
 311             tz               = _mm256_mul_ps(fscal,dz10);
 312
 313             /* Update vectorial force */
 314             fix1             = _mm256_add_ps(fix1,tx);
 315             fiy1             = _mm256_add_ps(fiy1,ty);
 316             fiz1             = _mm256_add_ps(fiz1,tz);
 317
 318             fjx0             = _mm256_add_ps(fjx0,tx);
 319             fjy0             = _mm256_add_ps(fjy0,ty);
 320             fjz0             = _mm256_add_ps(fjz0,tz);
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq20             = _mm256_mul_ps(iq2,jq0);
 328
 329             /* REACTION-FIELD ELECTROSTATICS */
 330             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 331             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velecsum         = _mm256_add_ps(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338             /* Calculate temporary vectorial force */
 339             tx               = _mm256_mul_ps(fscal,dx20);
 340             ty               = _mm256_mul_ps(fscal,dy20);
 341             tz               = _mm256_mul_ps(fscal,dz20);
 342
 343             /* Update vectorial force */
 344             fix2             = _mm256_add_ps(fix2,tx);
 345             fiy2             = _mm256_add_ps(fiy2,ty);
 346             fiz2             = _mm256_add_ps(fiz2,tz);
 347
 348             fjx0             = _mm256_add_ps(fjx0,tx);
 349             fjy0             = _mm256_add_ps(fjy0,ty);
 350             fjz0             = _mm256_add_ps(fjz0,tz);
 351
 352             fjptrA             = f+j_coord_offsetA;
 353             fjptrB             = f+j_coord_offsetB;
 354             fjptrC             = f+j_coord_offsetC;
 355             fjptrD             = f+j_coord_offsetD;
 356             fjptrE             = f+j_coord_offsetE;
 357             fjptrF             = f+j_coord_offsetF;
 358             fjptrG             = f+j_coord_offsetG;
 359             fjptrH             = f+j_coord_offsetH;
 360
 361             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 362
 363             /* Inner loop uses 111 flops */
 364         }
 365
 366         if(jidx<j_index_end)
 367         {
 368
 369             /* Get j neighbor index, and coordinate index */
 370             jnrlistA         = jjnr[jidx];
 371             jnrlistB         = jjnr[jidx+1];
 372             jnrlistC         = jjnr[jidx+2];
 373             jnrlistD         = jjnr[jidx+3];
 374             jnrlistE         = jjnr[jidx+4];
 375             jnrlistF         = jjnr[jidx+5];
 376             jnrlistG         = jjnr[jidx+6];
 377             jnrlistH         = jjnr[jidx+7];
 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_ps(mask,val) to clear dummy entries.
 381              */
 382             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 383                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 384
 385             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 386             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 387             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 388             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 389             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 390             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 391             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 392             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 393             j_coord_offsetA  = DIM*jnrA;
 394             j_coord_offsetB  = DIM*jnrB;
 395             j_coord_offsetC  = DIM*jnrC;
 396             j_coord_offsetD  = DIM*jnrD;
 397             j_coord_offsetE  = DIM*jnrE;
 398             j_coord_offsetF  = DIM*jnrF;
 399             j_coord_offsetG  = DIM*jnrG;
 400             j_coord_offsetH  = DIM*jnrH;
 401
 402             /* load j atom coordinates */
 403             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 404                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 405                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 406                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 407                                                  &jx0,&jy0,&jz0);
 408
 409             /* Calculate displacement vector */
 410             dx00             = _mm256_sub_ps(ix0,jx0);
 411             dy00             = _mm256_sub_ps(iy0,jy0);
 412             dz00             = _mm256_sub_ps(iz0,jz0);
 413             dx10             = _mm256_sub_ps(ix1,jx0);
 414             dy10             = _mm256_sub_ps(iy1,jy0);
 415             dz10             = _mm256_sub_ps(iz1,jz0);
 416             dx20             = _mm256_sub_ps(ix2,jx0);
 417             dy20             = _mm256_sub_ps(iy2,jy0);
 418             dz20             = _mm256_sub_ps(iz2,jz0);
 419
 420             /* Calculate squared distance and things based on it */
 421             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 422             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 423             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 424
 425             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 426             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 427             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 428
 429             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 430             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 431             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 432
 433             /* Load parameters for j particles */
 434             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 435                                                                  charge+jnrC+0,charge+jnrD+0,
 436                                                                  charge+jnrE+0,charge+jnrF+0,
 437                                                                  charge+jnrG+0,charge+jnrH+0);
 438             vdwjidx0A        = 2*vdwtype[jnrA+0];
 439             vdwjidx0B        = 2*vdwtype[jnrB+0];
 440             vdwjidx0C        = 2*vdwtype[jnrC+0];
 441             vdwjidx0D        = 2*vdwtype[jnrD+0];
 442             vdwjidx0E        = 2*vdwtype[jnrE+0];
 443             vdwjidx0F        = 2*vdwtype[jnrF+0];
 444             vdwjidx0G        = 2*vdwtype[jnrG+0];
 445             vdwjidx0H        = 2*vdwtype[jnrH+0];
 446
 447             fjx0             = _mm256_setzero_ps();
 448             fjy0             = _mm256_setzero_ps();
 449             fjz0             = _mm256_setzero_ps();
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             /* Compute parameters for interactions between i and j atoms */
 456             qq00             = _mm256_mul_ps(iq0,jq0);
 457             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 458                                             vdwioffsetptr0+vdwjidx0B,
 459                                             vdwioffsetptr0+vdwjidx0C,
 460                                             vdwioffsetptr0+vdwjidx0D,
 461                                             vdwioffsetptr0+vdwjidx0E,
 462                                             vdwioffsetptr0+vdwjidx0F,
 463                                             vdwioffsetptr0+vdwjidx0G,
 464                                             vdwioffsetptr0+vdwjidx0H,
 465                                             &c6_00,&c12_00);
 466
 467             /* REACTION-FIELD ELECTROSTATICS */
 468             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 469             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 470
 471             /* LENNARD-JONES DISPERSION/REPULSION */
 472
 473             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 474             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 475             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 476             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 477             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 478
 479             /* Update potential sum for this i atom from the interaction with this j atom. */
 480             velec            = _mm256_andnot_ps(dummy_mask,velec);
 481             velecsum         = _mm256_add_ps(velecsum,velec);
 482             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 483             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 484
 485             fscal            = _mm256_add_ps(felec,fvdw);
 486
 487             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 488
 489             /* Calculate temporary vectorial force */
 490             tx               = _mm256_mul_ps(fscal,dx00);
 491             ty               = _mm256_mul_ps(fscal,dy00);
 492             tz               = _mm256_mul_ps(fscal,dz00);
 493
 494             /* Update vectorial force */
 495             fix0             = _mm256_add_ps(fix0,tx);
 496             fiy0             = _mm256_add_ps(fiy0,ty);
 497             fiz0             = _mm256_add_ps(fiz0,tz);
 498
 499             fjx0             = _mm256_add_ps(fjx0,tx);
 500             fjy0             = _mm256_add_ps(fjy0,ty);
 501             fjz0             = _mm256_add_ps(fjz0,tz);
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             /* Compute parameters for interactions between i and j atoms */
 508             qq10             = _mm256_mul_ps(iq1,jq0);
 509
 510             /* REACTION-FIELD ELECTROSTATICS */
 511             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 512             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 513
 514             /* Update potential sum for this i atom from the interaction with this j atom. */
 515             velec            = _mm256_andnot_ps(dummy_mask,velec);
 516             velecsum         = _mm256_add_ps(velecsum,velec);
 517
 518             fscal            = felec;
 519
 520             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 521
 522             /* Calculate temporary vectorial force */
 523             tx               = _mm256_mul_ps(fscal,dx10);
 524             ty               = _mm256_mul_ps(fscal,dy10);
 525             tz               = _mm256_mul_ps(fscal,dz10);
 526
 527             /* Update vectorial force */
 528             fix1             = _mm256_add_ps(fix1,tx);
 529             fiy1             = _mm256_add_ps(fiy1,ty);
 530             fiz1             = _mm256_add_ps(fiz1,tz);
 531
 532             fjx0             = _mm256_add_ps(fjx0,tx);
 533             fjy0             = _mm256_add_ps(fjy0,ty);
 534             fjz0             = _mm256_add_ps(fjz0,tz);
 535
 536             /**************************
 537              * CALCULATE INTERACTIONS *
 538              **************************/
 539
 540             /* Compute parameters for interactions between i and j atoms */
 541             qq20             = _mm256_mul_ps(iq2,jq0);
 542
 543             /* REACTION-FIELD ELECTROSTATICS */
 544             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 545             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 546
 547             /* Update potential sum for this i atom from the interaction with this j atom. */
 548             velec            = _mm256_andnot_ps(dummy_mask,velec);
 549             velecsum         = _mm256_add_ps(velecsum,velec);
 550
 551             fscal            = felec;
 552
 553             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 554
 555             /* Calculate temporary vectorial force */
 556             tx               = _mm256_mul_ps(fscal,dx20);
 557             ty               = _mm256_mul_ps(fscal,dy20);
 558             tz               = _mm256_mul_ps(fscal,dz20);
 559
 560             /* Update vectorial force */
 561             fix2             = _mm256_add_ps(fix2,tx);
 562             fiy2             = _mm256_add_ps(fiy2,ty);
 563             fiz2             = _mm256_add_ps(fiz2,tz);
 564
 565             fjx0             = _mm256_add_ps(fjx0,tx);
 566             fjy0             = _mm256_add_ps(fjy0,ty);
 567             fjz0             = _mm256_add_ps(fjz0,tz);
 568
 569             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 570             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 571             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 572             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 573             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 574             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 575             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 576             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 577
 578             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 579
 580             /* Inner loop uses 111 flops */
 581         }
 582
 583         /* End of innermost loop */
 584
 585         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 586                                                  f+i_coord_offset,fshift+i_shift_offset);
 587
 588         ggid                        = gid[iidx];
 589         /* Update potential energies */
 590         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 591         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 592
 593         /* Increment number of inner iterations */
 594         inneriter                  += j_index_end - j_index_start;
 595
 596         /* Outer loop uses 20 flops */
 597     }
 598
 599     /* Increment number of outer iterations */
 600     outeriter        += nri;
 601
 602     /* Update outer/inner flops */
 603
 604     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*111);
 605 }
 606 /*
 607  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_avx_256_single
 608  * Electrostatics interaction: ReactionField
 609  * VdW interaction:            LennardJones
 610  * Geometry:                   Water3-Particle
 611  * Calculate force/pot:        Force
 612  */
 613 void
 614 nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_avx_256_single
 615                     (t_nblist * gmx_restrict                nlist,
 616                      rvec * gmx_restrict                    xx,
 617                      rvec * gmx_restrict                    ff,
 618                      t_forcerec * gmx_restrict              fr,
 619                      t_mdatoms * gmx_restrict               mdatoms,
 620                      nb_kernel_data_t * gmx_restrict        kernel_data,
 621                      t_nrnb * gmx_restrict                  nrnb)
 622 {
 623     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 624      * just 0 for non-waters.
 625      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 626      * jnr indices corresponding to data put in the four positions in the SIMD register.
 627      */
 628     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 629     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 630     int              jnrA,jnrB,jnrC,jnrD;
 631     int              jnrE,jnrF,jnrG,jnrH;
 632     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 633     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 634     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 635     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 636     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 637     real             rcutoff_scalar;
 638     real             *shiftvec,*fshift,*x,*f;
 639     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 640     real             scratch[4*DIM];
 641     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 642     real *           vdwioffsetptr0;
 643     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 644     real *           vdwioffsetptr1;
 645     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 646     real *           vdwioffsetptr2;
 647     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 648     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 649     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 650     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 651     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 652     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 653     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 654     real             *charge;
 655     int              nvdwtype;
 656     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 657     int              *vdwtype;
 658     real             *vdwparam;
 659     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 660     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 661     __m256           dummy_mask,cutoff_mask;
 662     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 663     __m256           one     = _mm256_set1_ps(1.0);
 664     __m256           two     = _mm256_set1_ps(2.0);
 665     x                = xx[0];
 666     f                = ff[0];
 667
 668     nri              = nlist->nri;
 669     iinr             = nlist->iinr;
 670     jindex           = nlist->jindex;
 671     jjnr             = nlist->jjnr;
 672     shiftidx         = nlist->shift;
 673     gid              = nlist->gid;
 674     shiftvec         = fr->shift_vec[0];
 675     fshift           = fr->fshift[0];
 676     facel            = _mm256_set1_ps(fr->epsfac);
 677     charge           = mdatoms->chargeA;
 678     krf              = _mm256_set1_ps(fr->ic->k_rf);
 679     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 680     crf              = _mm256_set1_ps(fr->ic->c_rf);
 681     nvdwtype         = fr->ntype;
 682     vdwparam         = fr->nbfp;
 683     vdwtype          = mdatoms->typeA;
 684
 685     /* Setup water-specific parameters */
 686     inr              = nlist->iinr[0];
 687     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 688     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 689     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 690     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 691
 692     /* Avoid stupid compiler warnings */
 693     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 694     j_coord_offsetA = 0;
 695     j_coord_offsetB = 0;
 696     j_coord_offsetC = 0;
 697     j_coord_offsetD = 0;
 698     j_coord_offsetE = 0;
 699     j_coord_offsetF = 0;
 700     j_coord_offsetG = 0;
 701     j_coord_offsetH = 0;
 702
 703     outeriter        = 0;
 704     inneriter        = 0;
 705
 706     for(iidx=0;iidx<4*DIM;iidx++)
 707     {
 708         scratch[iidx] = 0.0;
 709     }
 710
 711     /* Start outer loop over neighborlists */
 712     for(iidx=0; iidx<nri; iidx++)
 713     {
 714         /* Load shift vector for this list */
 715         i_shift_offset   = DIM*shiftidx[iidx];
 716
 717         /* Load limits for loop over neighbors */
 718         j_index_start    = jindex[iidx];
 719         j_index_end      = jindex[iidx+1];
 720
 721         /* Get outer coordinate index */
 722         inr              = iinr[iidx];
 723         i_coord_offset   = DIM*inr;
 724
 725         /* Load i particle coords and add shift vector */
 726         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 727                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 728
 729         fix0             = _mm256_setzero_ps();
 730         fiy0             = _mm256_setzero_ps();
 731         fiz0             = _mm256_setzero_ps();
 732         fix1             = _mm256_setzero_ps();
 733         fiy1             = _mm256_setzero_ps();
 734         fiz1             = _mm256_setzero_ps();
 735         fix2             = _mm256_setzero_ps();
 736         fiy2             = _mm256_setzero_ps();
 737         fiz2             = _mm256_setzero_ps();
 738
 739         /* Start inner kernel loop */
 740         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 741         {
 742
 743             /* Get j neighbor index, and coordinate index */
 744             jnrA             = jjnr[jidx];
 745             jnrB             = jjnr[jidx+1];
 746             jnrC             = jjnr[jidx+2];
 747             jnrD             = jjnr[jidx+3];
 748             jnrE             = jjnr[jidx+4];
 749             jnrF             = jjnr[jidx+5];
 750             jnrG             = jjnr[jidx+6];
 751             jnrH             = jjnr[jidx+7];
 752             j_coord_offsetA  = DIM*jnrA;
 753             j_coord_offsetB  = DIM*jnrB;
 754             j_coord_offsetC  = DIM*jnrC;
 755             j_coord_offsetD  = DIM*jnrD;
 756             j_coord_offsetE  = DIM*jnrE;
 757             j_coord_offsetF  = DIM*jnrF;
 758             j_coord_offsetG  = DIM*jnrG;
 759             j_coord_offsetH  = DIM*jnrH;
 760
 761             /* load j atom coordinates */
 762             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 763                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 764                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 765                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 766                                                  &jx0,&jy0,&jz0);
 767
 768             /* Calculate displacement vector */
 769             dx00             = _mm256_sub_ps(ix0,jx0);
 770             dy00             = _mm256_sub_ps(iy0,jy0);
 771             dz00             = _mm256_sub_ps(iz0,jz0);
 772             dx10             = _mm256_sub_ps(ix1,jx0);
 773             dy10             = _mm256_sub_ps(iy1,jy0);
 774             dz10             = _mm256_sub_ps(iz1,jz0);
 775             dx20             = _mm256_sub_ps(ix2,jx0);
 776             dy20             = _mm256_sub_ps(iy2,jy0);
 777             dz20             = _mm256_sub_ps(iz2,jz0);
 778
 779             /* Calculate squared distance and things based on it */
 780             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 781             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 782             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 783
 784             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 785             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 786             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 787
 788             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 789             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 790             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 791
 792             /* Load parameters for j particles */
 793             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 794                                                                  charge+jnrC+0,charge+jnrD+0,
 795                                                                  charge+jnrE+0,charge+jnrF+0,
 796                                                                  charge+jnrG+0,charge+jnrH+0);
 797             vdwjidx0A        = 2*vdwtype[jnrA+0];
 798             vdwjidx0B        = 2*vdwtype[jnrB+0];
 799             vdwjidx0C        = 2*vdwtype[jnrC+0];
 800             vdwjidx0D        = 2*vdwtype[jnrD+0];
 801             vdwjidx0E        = 2*vdwtype[jnrE+0];
 802             vdwjidx0F        = 2*vdwtype[jnrF+0];
 803             vdwjidx0G        = 2*vdwtype[jnrG+0];
 804             vdwjidx0H        = 2*vdwtype[jnrH+0];
 805
 806             fjx0             = _mm256_setzero_ps();
 807             fjy0             = _mm256_setzero_ps();
 808             fjz0             = _mm256_setzero_ps();
 809
 810             /**************************
 811              * CALCULATE INTERACTIONS *
 812              **************************/
 813
 814             /* Compute parameters for interactions between i and j atoms */
 815             qq00             = _mm256_mul_ps(iq0,jq0);
 816             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 817                                             vdwioffsetptr0+vdwjidx0B,
 818                                             vdwioffsetptr0+vdwjidx0C,
 819                                             vdwioffsetptr0+vdwjidx0D,
 820                                             vdwioffsetptr0+vdwjidx0E,
 821                                             vdwioffsetptr0+vdwjidx0F,
 822                                             vdwioffsetptr0+vdwjidx0G,
 823                                             vdwioffsetptr0+vdwjidx0H,
 824                                             &c6_00,&c12_00);
 825
 826             /* REACTION-FIELD ELECTROSTATICS */
 827             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 828
 829             /* LENNARD-JONES DISPERSION/REPULSION */
 830
 831             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 832             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 833
 834             fscal            = _mm256_add_ps(felec,fvdw);
 835
 836             /* Calculate temporary vectorial force */
 837             tx               = _mm256_mul_ps(fscal,dx00);
 838             ty               = _mm256_mul_ps(fscal,dy00);
 839             tz               = _mm256_mul_ps(fscal,dz00);
 840
 841             /* Update vectorial force */
 842             fix0             = _mm256_add_ps(fix0,tx);
 843             fiy0             = _mm256_add_ps(fiy0,ty);
 844             fiz0             = _mm256_add_ps(fiz0,tz);
 845
 846             fjx0             = _mm256_add_ps(fjx0,tx);
 847             fjy0             = _mm256_add_ps(fjy0,ty);
 848             fjz0             = _mm256_add_ps(fjz0,tz);
 849
 850             /**************************
 851              * CALCULATE INTERACTIONS *
 852              **************************/
 853
 854             /* Compute parameters for interactions between i and j atoms */
 855             qq10             = _mm256_mul_ps(iq1,jq0);
 856
 857             /* REACTION-FIELD ELECTROSTATICS */
 858             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 859
 860             fscal            = felec;
 861
 862             /* Calculate temporary vectorial force */
 863             tx               = _mm256_mul_ps(fscal,dx10);
 864             ty               = _mm256_mul_ps(fscal,dy10);
 865             tz               = _mm256_mul_ps(fscal,dz10);
 866
 867             /* Update vectorial force */
 868             fix1             = _mm256_add_ps(fix1,tx);
 869             fiy1             = _mm256_add_ps(fiy1,ty);
 870             fiz1             = _mm256_add_ps(fiz1,tz);
 871
 872             fjx0             = _mm256_add_ps(fjx0,tx);
 873             fjy0             = _mm256_add_ps(fjy0,ty);
 874             fjz0             = _mm256_add_ps(fjz0,tz);
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             /* Compute parameters for interactions between i and j atoms */
 881             qq20             = _mm256_mul_ps(iq2,jq0);
 882
 883             /* REACTION-FIELD ELECTROSTATICS */
 884             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 885
 886             fscal            = felec;
 887
 888             /* Calculate temporary vectorial force */
 889             tx               = _mm256_mul_ps(fscal,dx20);
 890             ty               = _mm256_mul_ps(fscal,dy20);
 891             tz               = _mm256_mul_ps(fscal,dz20);
 892
 893             /* Update vectorial force */
 894             fix2             = _mm256_add_ps(fix2,tx);
 895             fiy2             = _mm256_add_ps(fiy2,ty);
 896             fiz2             = _mm256_add_ps(fiz2,tz);
 897
 898             fjx0             = _mm256_add_ps(fjx0,tx);
 899             fjy0             = _mm256_add_ps(fjy0,ty);
 900             fjz0             = _mm256_add_ps(fjz0,tz);
 901
 902             fjptrA             = f+j_coord_offsetA;
 903             fjptrB             = f+j_coord_offsetB;
 904             fjptrC             = f+j_coord_offsetC;
 905             fjptrD             = f+j_coord_offsetD;
 906             fjptrE             = f+j_coord_offsetE;
 907             fjptrF             = f+j_coord_offsetF;
 908             fjptrG             = f+j_coord_offsetG;
 909             fjptrH             = f+j_coord_offsetH;
 910
 911             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 912
 913             /* Inner loop uses 91 flops */
 914         }
 915
 916         if(jidx<j_index_end)
 917         {
 918
 919             /* Get j neighbor index, and coordinate index */
 920             jnrlistA         = jjnr[jidx];
 921             jnrlistB         = jjnr[jidx+1];
 922             jnrlistC         = jjnr[jidx+2];
 923             jnrlistD         = jjnr[jidx+3];
 924             jnrlistE         = jjnr[jidx+4];
 925             jnrlistF         = jjnr[jidx+5];
 926             jnrlistG         = jjnr[jidx+6];
 927             jnrlistH         = jjnr[jidx+7];
 928             /* Sign of each element will be negative for non-real atoms.
 929              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 930              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 931              */
 932             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 933                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 934
 935             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 936             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 937             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 938             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 939             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 940             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 941             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 942             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 943             j_coord_offsetA  = DIM*jnrA;
 944             j_coord_offsetB  = DIM*jnrB;
 945             j_coord_offsetC  = DIM*jnrC;
 946             j_coord_offsetD  = DIM*jnrD;
 947             j_coord_offsetE  = DIM*jnrE;
 948             j_coord_offsetF  = DIM*jnrF;
 949             j_coord_offsetG  = DIM*jnrG;
 950             j_coord_offsetH  = DIM*jnrH;
 951
 952             /* load j atom coordinates */
 953             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 954                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 955                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 956                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 957                                                  &jx0,&jy0,&jz0);
 958
 959             /* Calculate displacement vector */
 960             dx00             = _mm256_sub_ps(ix0,jx0);
 961             dy00             = _mm256_sub_ps(iy0,jy0);
 962             dz00             = _mm256_sub_ps(iz0,jz0);
 963             dx10             = _mm256_sub_ps(ix1,jx0);
 964             dy10             = _mm256_sub_ps(iy1,jy0);
 965             dz10             = _mm256_sub_ps(iz1,jz0);
 966             dx20             = _mm256_sub_ps(ix2,jx0);
 967             dy20             = _mm256_sub_ps(iy2,jy0);
 968             dz20             = _mm256_sub_ps(iz2,jz0);
 969
 970             /* Calculate squared distance and things based on it */
 971             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 972             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 973             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 974
 975             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 976             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 977             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 978
 979             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 980             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 981             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 982
 983             /* Load parameters for j particles */
 984             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 985                                                                  charge+jnrC+0,charge+jnrD+0,
 986                                                                  charge+jnrE+0,charge+jnrF+0,
 987                                                                  charge+jnrG+0,charge+jnrH+0);
 988             vdwjidx0A        = 2*vdwtype[jnrA+0];
 989             vdwjidx0B        = 2*vdwtype[jnrB+0];
 990             vdwjidx0C        = 2*vdwtype[jnrC+0];
 991             vdwjidx0D        = 2*vdwtype[jnrD+0];
 992             vdwjidx0E        = 2*vdwtype[jnrE+0];
 993             vdwjidx0F        = 2*vdwtype[jnrF+0];
 994             vdwjidx0G        = 2*vdwtype[jnrG+0];
 995             vdwjidx0H        = 2*vdwtype[jnrH+0];
 996
 997             fjx0             = _mm256_setzero_ps();
 998             fjy0             = _mm256_setzero_ps();
 999             fjz0             = _mm256_setzero_ps();
1000
1001             /**************************
1002              * CALCULATE INTERACTIONS *
1003              **************************/
1004
1005             /* Compute parameters for interactions between i and j atoms */
1006             qq00             = _mm256_mul_ps(iq0,jq0);
1007             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1008                                             vdwioffsetptr0+vdwjidx0B,
1009                                             vdwioffsetptr0+vdwjidx0C,
1010                                             vdwioffsetptr0+vdwjidx0D,
1011                                             vdwioffsetptr0+vdwjidx0E,
1012                                             vdwioffsetptr0+vdwjidx0F,
1013                                             vdwioffsetptr0+vdwjidx0G,
1014                                             vdwioffsetptr0+vdwjidx0H,
1015                                             &c6_00,&c12_00);
1016
1017             /* REACTION-FIELD ELECTROSTATICS */
1018             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1019
1020             /* LENNARD-JONES DISPERSION/REPULSION */
1021
1022             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1023             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1024
1025             fscal            = _mm256_add_ps(felec,fvdw);
1026
1027             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1028
1029             /* Calculate temporary vectorial force */
1030             tx               = _mm256_mul_ps(fscal,dx00);
1031             ty               = _mm256_mul_ps(fscal,dy00);
1032             tz               = _mm256_mul_ps(fscal,dz00);
1033
1034             /* Update vectorial force */
1035             fix0             = _mm256_add_ps(fix0,tx);
1036             fiy0             = _mm256_add_ps(fiy0,ty);
1037             fiz0             = _mm256_add_ps(fiz0,tz);
1038
1039             fjx0             = _mm256_add_ps(fjx0,tx);
1040             fjy0             = _mm256_add_ps(fjy0,ty);
1041             fjz0             = _mm256_add_ps(fjz0,tz);
1042
1043             /**************************
1044              * CALCULATE INTERACTIONS *
1045              **************************/
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             qq10             = _mm256_mul_ps(iq1,jq0);
1049
1050             /* REACTION-FIELD ELECTROSTATICS */
1051             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1052
1053             fscal            = felec;
1054
1055             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1056
1057             /* Calculate temporary vectorial force */
1058             tx               = _mm256_mul_ps(fscal,dx10);
1059             ty               = _mm256_mul_ps(fscal,dy10);
1060             tz               = _mm256_mul_ps(fscal,dz10);
1061
1062             /* Update vectorial force */
1063             fix1             = _mm256_add_ps(fix1,tx);
1064             fiy1             = _mm256_add_ps(fiy1,ty);
1065             fiz1             = _mm256_add_ps(fiz1,tz);
1066
1067             fjx0             = _mm256_add_ps(fjx0,tx);
1068             fjy0             = _mm256_add_ps(fjy0,ty);
1069             fjz0             = _mm256_add_ps(fjz0,tz);
1070
1071             /**************************
1072              * CALCULATE INTERACTIONS *
1073              **************************/
1074
1075             /* Compute parameters for interactions between i and j atoms */
1076             qq20             = _mm256_mul_ps(iq2,jq0);
1077
1078             /* REACTION-FIELD ELECTROSTATICS */
1079             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
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 91 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*91);
1130 }