src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwNone_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_VdwNone_GeomW3P1_VF_avx_256_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            None
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwNone_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     __m256           dummy_mask,cutoff_mask;
  86     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  87     __m256           one     = _mm256_set1_ps(1.0);
  88     __m256           two     = _mm256_set1_ps(2.0);
  89     x                = xx[0];
  90     f                = ff[0];
  91
  92     nri              = nlist->nri;
  93     iinr             = nlist->iinr;
  94     jindex           = nlist->jindex;
  95     jjnr             = nlist->jjnr;
  96     shiftidx         = nlist->shift;
  97     gid              = nlist->gid;
  98     shiftvec         = fr->shift_vec[0];
  99     fshift           = fr->fshift[0];
 100     facel            = _mm256_set1_ps(fr->epsfac);
 101     charge           = mdatoms->chargeA;
 102
 103     /* Setup water-specific parameters */
 104     inr              = nlist->iinr[0];
 105     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 106     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 107     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 108
 109     /* Avoid stupid compiler warnings */
 110     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 111     j_coord_offsetA = 0;
 112     j_coord_offsetB = 0;
 113     j_coord_offsetC = 0;
 114     j_coord_offsetD = 0;
 115     j_coord_offsetE = 0;
 116     j_coord_offsetF = 0;
 117     j_coord_offsetG = 0;
 118     j_coord_offsetH = 0;
 119
 120     outeriter        = 0;
 121     inneriter        = 0;
 122
 123     for(iidx=0;iidx<4*DIM;iidx++)
 124     {
 125         scratch[iidx] = 0.0;
 126     }
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 144                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 145
 146         fix0             = _mm256_setzero_ps();
 147         fiy0             = _mm256_setzero_ps();
 148         fiz0             = _mm256_setzero_ps();
 149         fix1             = _mm256_setzero_ps();
 150         fiy1             = _mm256_setzero_ps();
 151         fiz1             = _mm256_setzero_ps();
 152         fix2             = _mm256_setzero_ps();
 153         fiy2             = _mm256_setzero_ps();
 154         fiz2             = _mm256_setzero_ps();
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm256_setzero_ps();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             jnrC             = jjnr[jidx+2];
 167             jnrD             = jjnr[jidx+3];
 168             jnrE             = jjnr[jidx+4];
 169             jnrF             = jjnr[jidx+5];
 170             jnrG             = jjnr[jidx+6];
 171             jnrH             = jjnr[jidx+7];
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174             j_coord_offsetC  = DIM*jnrC;
 175             j_coord_offsetD  = DIM*jnrD;
 176             j_coord_offsetE  = DIM*jnrE;
 177             j_coord_offsetF  = DIM*jnrF;
 178             j_coord_offsetG  = DIM*jnrG;
 179             j_coord_offsetH  = DIM*jnrH;
 180
 181             /* load j atom coordinates */
 182             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 183                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 184                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 185                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 186                                                  &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm256_sub_ps(ix0,jx0);
 190             dy00             = _mm256_sub_ps(iy0,jy0);
 191             dz00             = _mm256_sub_ps(iz0,jz0);
 192             dx10             = _mm256_sub_ps(ix1,jx0);
 193             dy10             = _mm256_sub_ps(iy1,jy0);
 194             dz10             = _mm256_sub_ps(iz1,jz0);
 195             dx20             = _mm256_sub_ps(ix2,jx0);
 196             dy20             = _mm256_sub_ps(iy2,jy0);
 197             dz20             = _mm256_sub_ps(iz2,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 201             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 202             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 203
 204             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 205             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 206             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 207
 208             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 209             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 210             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 211
 212             /* Load parameters for j particles */
 213             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 214                                                                  charge+jnrC+0,charge+jnrD+0,
 215                                                                  charge+jnrE+0,charge+jnrF+0,
 216                                                                  charge+jnrG+0,charge+jnrH+0);
 217
 218             fjx0             = _mm256_setzero_ps();
 219             fjy0             = _mm256_setzero_ps();
 220             fjz0             = _mm256_setzero_ps();
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             /* Compute parameters for interactions between i and j atoms */
 227             qq00             = _mm256_mul_ps(iq0,jq0);
 228
 229             /* COULOMB ELECTROSTATICS */
 230             velec            = _mm256_mul_ps(qq00,rinv00);
 231             felec            = _mm256_mul_ps(velec,rinvsq00);
 232
 233             /* Update potential sum for this i atom from the interaction with this j atom. */
 234             velecsum         = _mm256_add_ps(velecsum,velec);
 235
 236             fscal            = felec;
 237
 238             /* Calculate temporary vectorial force */
 239             tx               = _mm256_mul_ps(fscal,dx00);
 240             ty               = _mm256_mul_ps(fscal,dy00);
 241             tz               = _mm256_mul_ps(fscal,dz00);
 242
 243             /* Update vectorial force */
 244             fix0             = _mm256_add_ps(fix0,tx);
 245             fiy0             = _mm256_add_ps(fiy0,ty);
 246             fiz0             = _mm256_add_ps(fiz0,tz);
 247
 248             fjx0             = _mm256_add_ps(fjx0,tx);
 249             fjy0             = _mm256_add_ps(fjy0,ty);
 250             fjz0             = _mm256_add_ps(fjz0,tz);
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             qq10             = _mm256_mul_ps(iq1,jq0);
 258
 259             /* COULOMB ELECTROSTATICS */
 260             velec            = _mm256_mul_ps(qq10,rinv10);
 261             felec            = _mm256_mul_ps(velec,rinvsq10);
 262
 263             /* Update potential sum for this i atom from the interaction with this j atom. */
 264             velecsum         = _mm256_add_ps(velecsum,velec);
 265
 266             fscal            = felec;
 267
 268             /* Calculate temporary vectorial force */
 269             tx               = _mm256_mul_ps(fscal,dx10);
 270             ty               = _mm256_mul_ps(fscal,dy10);
 271             tz               = _mm256_mul_ps(fscal,dz10);
 272
 273             /* Update vectorial force */
 274             fix1             = _mm256_add_ps(fix1,tx);
 275             fiy1             = _mm256_add_ps(fiy1,ty);
 276             fiz1             = _mm256_add_ps(fiz1,tz);
 277
 278             fjx0             = _mm256_add_ps(fjx0,tx);
 279             fjy0             = _mm256_add_ps(fjy0,ty);
 280             fjz0             = _mm256_add_ps(fjz0,tz);
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             /* Compute parameters for interactions between i and j atoms */
 287             qq20             = _mm256_mul_ps(iq2,jq0);
 288
 289             /* COULOMB ELECTROSTATICS */
 290             velec            = _mm256_mul_ps(qq20,rinv20);
 291             felec            = _mm256_mul_ps(velec,rinvsq20);
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velecsum         = _mm256_add_ps(velecsum,velec);
 295
 296             fscal            = felec;
 297
 298             /* Calculate temporary vectorial force */
 299             tx               = _mm256_mul_ps(fscal,dx20);
 300             ty               = _mm256_mul_ps(fscal,dy20);
 301             tz               = _mm256_mul_ps(fscal,dz20);
 302
 303             /* Update vectorial force */
 304             fix2             = _mm256_add_ps(fix2,tx);
 305             fiy2             = _mm256_add_ps(fiy2,ty);
 306             fiz2             = _mm256_add_ps(fiz2,tz);
 307
 308             fjx0             = _mm256_add_ps(fjx0,tx);
 309             fjy0             = _mm256_add_ps(fjy0,ty);
 310             fjz0             = _mm256_add_ps(fjz0,tz);
 311
 312             fjptrA             = f+j_coord_offsetA;
 313             fjptrB             = f+j_coord_offsetB;
 314             fjptrC             = f+j_coord_offsetC;
 315             fjptrD             = f+j_coord_offsetD;
 316             fjptrE             = f+j_coord_offsetE;
 317             fjptrF             = f+j_coord_offsetF;
 318             fjptrG             = f+j_coord_offsetG;
 319             fjptrH             = f+j_coord_offsetH;
 320
 321             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 322
 323             /* Inner loop uses 84 flops */
 324         }
 325
 326         if(jidx<j_index_end)
 327         {
 328
 329             /* Get j neighbor index, and coordinate index */
 330             jnrlistA         = jjnr[jidx];
 331             jnrlistB         = jjnr[jidx+1];
 332             jnrlistC         = jjnr[jidx+2];
 333             jnrlistD         = jjnr[jidx+3];
 334             jnrlistE         = jjnr[jidx+4];
 335             jnrlistF         = jjnr[jidx+5];
 336             jnrlistG         = jjnr[jidx+6];
 337             jnrlistH         = jjnr[jidx+7];
 338             /* Sign of each element will be negative for non-real atoms.
 339              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 340              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 341              */
 342             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 343                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 344
 345             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 346             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 347             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 348             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 349             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 350             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 351             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 352             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 353             j_coord_offsetA  = DIM*jnrA;
 354             j_coord_offsetB  = DIM*jnrB;
 355             j_coord_offsetC  = DIM*jnrC;
 356             j_coord_offsetD  = DIM*jnrD;
 357             j_coord_offsetE  = DIM*jnrE;
 358             j_coord_offsetF  = DIM*jnrF;
 359             j_coord_offsetG  = DIM*jnrG;
 360             j_coord_offsetH  = DIM*jnrH;
 361
 362             /* load j atom coordinates */
 363             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 364                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 365                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 366                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 367                                                  &jx0,&jy0,&jz0);
 368
 369             /* Calculate displacement vector */
 370             dx00             = _mm256_sub_ps(ix0,jx0);
 371             dy00             = _mm256_sub_ps(iy0,jy0);
 372             dz00             = _mm256_sub_ps(iz0,jz0);
 373             dx10             = _mm256_sub_ps(ix1,jx0);
 374             dy10             = _mm256_sub_ps(iy1,jy0);
 375             dz10             = _mm256_sub_ps(iz1,jz0);
 376             dx20             = _mm256_sub_ps(ix2,jx0);
 377             dy20             = _mm256_sub_ps(iy2,jy0);
 378             dz20             = _mm256_sub_ps(iz2,jz0);
 379
 380             /* Calculate squared distance and things based on it */
 381             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 382             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 383             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 384
 385             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 386             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 387             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 388
 389             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 390             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 391             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 392
 393             /* Load parameters for j particles */
 394             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 395                                                                  charge+jnrC+0,charge+jnrD+0,
 396                                                                  charge+jnrE+0,charge+jnrF+0,
 397                                                                  charge+jnrG+0,charge+jnrH+0);
 398
 399             fjx0             = _mm256_setzero_ps();
 400             fjy0             = _mm256_setzero_ps();
 401             fjz0             = _mm256_setzero_ps();
 402
 403             /**************************
 404              * CALCULATE INTERACTIONS *
 405              **************************/
 406
 407             /* Compute parameters for interactions between i and j atoms */
 408             qq00             = _mm256_mul_ps(iq0,jq0);
 409
 410             /* COULOMB ELECTROSTATICS */
 411             velec            = _mm256_mul_ps(qq00,rinv00);
 412             felec            = _mm256_mul_ps(velec,rinvsq00);
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             velec            = _mm256_andnot_ps(dummy_mask,velec);
 416             velecsum         = _mm256_add_ps(velecsum,velec);
 417
 418             fscal            = felec;
 419
 420             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 421
 422             /* Calculate temporary vectorial force */
 423             tx               = _mm256_mul_ps(fscal,dx00);
 424             ty               = _mm256_mul_ps(fscal,dy00);
 425             tz               = _mm256_mul_ps(fscal,dz00);
 426
 427             /* Update vectorial force */
 428             fix0             = _mm256_add_ps(fix0,tx);
 429             fiy0             = _mm256_add_ps(fiy0,ty);
 430             fiz0             = _mm256_add_ps(fiz0,tz);
 431
 432             fjx0             = _mm256_add_ps(fjx0,tx);
 433             fjy0             = _mm256_add_ps(fjy0,ty);
 434             fjz0             = _mm256_add_ps(fjz0,tz);
 435
 436             /**************************
 437              * CALCULATE INTERACTIONS *
 438              **************************/
 439
 440             /* Compute parameters for interactions between i and j atoms */
 441             qq10             = _mm256_mul_ps(iq1,jq0);
 442
 443             /* COULOMB ELECTROSTATICS */
 444             velec            = _mm256_mul_ps(qq10,rinv10);
 445             felec            = _mm256_mul_ps(velec,rinvsq10);
 446
 447             /* Update potential sum for this i atom from the interaction with this j atom. */
 448             velec            = _mm256_andnot_ps(dummy_mask,velec);
 449             velecsum         = _mm256_add_ps(velecsum,velec);
 450
 451             fscal            = felec;
 452
 453             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 454
 455             /* Calculate temporary vectorial force */
 456             tx               = _mm256_mul_ps(fscal,dx10);
 457             ty               = _mm256_mul_ps(fscal,dy10);
 458             tz               = _mm256_mul_ps(fscal,dz10);
 459
 460             /* Update vectorial force */
 461             fix1             = _mm256_add_ps(fix1,tx);
 462             fiy1             = _mm256_add_ps(fiy1,ty);
 463             fiz1             = _mm256_add_ps(fiz1,tz);
 464
 465             fjx0             = _mm256_add_ps(fjx0,tx);
 466             fjy0             = _mm256_add_ps(fjy0,ty);
 467             fjz0             = _mm256_add_ps(fjz0,tz);
 468
 469             /**************************
 470              * CALCULATE INTERACTIONS *
 471              **************************/
 472
 473             /* Compute parameters for interactions between i and j atoms */
 474             qq20             = _mm256_mul_ps(iq2,jq0);
 475
 476             /* COULOMB ELECTROSTATICS */
 477             velec            = _mm256_mul_ps(qq20,rinv20);
 478             felec            = _mm256_mul_ps(velec,rinvsq20);
 479
 480             /* Update potential sum for this i atom from the interaction with this j atom. */
 481             velec            = _mm256_andnot_ps(dummy_mask,velec);
 482             velecsum         = _mm256_add_ps(velecsum,velec);
 483
 484             fscal            = felec;
 485
 486             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 487
 488             /* Calculate temporary vectorial force */
 489             tx               = _mm256_mul_ps(fscal,dx20);
 490             ty               = _mm256_mul_ps(fscal,dy20);
 491             tz               = _mm256_mul_ps(fscal,dz20);
 492
 493             /* Update vectorial force */
 494             fix2             = _mm256_add_ps(fix2,tx);
 495             fiy2             = _mm256_add_ps(fiy2,ty);
 496             fiz2             = _mm256_add_ps(fiz2,tz);
 497
 498             fjx0             = _mm256_add_ps(fjx0,tx);
 499             fjy0             = _mm256_add_ps(fjy0,ty);
 500             fjz0             = _mm256_add_ps(fjz0,tz);
 501
 502             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 503             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 504             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 505             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 506             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 507             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 508             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 509             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 510
 511             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 512
 513             /* Inner loop uses 84 flops */
 514         }
 515
 516         /* End of innermost loop */
 517
 518         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 519                                                  f+i_coord_offset,fshift+i_shift_offset);
 520
 521         ggid                        = gid[iidx];
 522         /* Update potential energies */
 523         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 524
 525         /* Increment number of inner iterations */
 526         inneriter                  += j_index_end - j_index_start;
 527
 528         /* Outer loop uses 19 flops */
 529     }
 530
 531     /* Increment number of outer iterations */
 532     outeriter        += nri;
 533
 534     /* Update outer/inner flops */
 535
 536     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*84);
 537 }
 538 /*
 539  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_avx_256_single
 540  * Electrostatics interaction: Coulomb
 541  * VdW interaction:            None
 542  * Geometry:                   Water3-Particle
 543  * Calculate force/pot:        Force
 544  */
 545 void
 546 nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_avx_256_single
 547                     (t_nblist * gmx_restrict                nlist,
 548                      rvec * gmx_restrict                    xx,
 549                      rvec * gmx_restrict                    ff,
 550                      t_forcerec * gmx_restrict              fr,
 551                      t_mdatoms * gmx_restrict               mdatoms,
 552                      nb_kernel_data_t * gmx_restrict        kernel_data,
 553                      t_nrnb * gmx_restrict                  nrnb)
 554 {
 555     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 556      * just 0 for non-waters.
 557      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 558      * jnr indices corresponding to data put in the four positions in the SIMD register.
 559      */
 560     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 561     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 562     int              jnrA,jnrB,jnrC,jnrD;
 563     int              jnrE,jnrF,jnrG,jnrH;
 564     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 565     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 566     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 567     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 568     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 569     real             rcutoff_scalar;
 570     real             *shiftvec,*fshift,*x,*f;
 571     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 572     real             scratch[4*DIM];
 573     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 574     real *           vdwioffsetptr0;
 575     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 576     real *           vdwioffsetptr1;
 577     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 578     real *           vdwioffsetptr2;
 579     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 580     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 581     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 582     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 583     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 584     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 585     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 586     real             *charge;
 587     __m256           dummy_mask,cutoff_mask;
 588     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 589     __m256           one     = _mm256_set1_ps(1.0);
 590     __m256           two     = _mm256_set1_ps(2.0);
 591     x                = xx[0];
 592     f                = ff[0];
 593
 594     nri              = nlist->nri;
 595     iinr             = nlist->iinr;
 596     jindex           = nlist->jindex;
 597     jjnr             = nlist->jjnr;
 598     shiftidx         = nlist->shift;
 599     gid              = nlist->gid;
 600     shiftvec         = fr->shift_vec[0];
 601     fshift           = fr->fshift[0];
 602     facel            = _mm256_set1_ps(fr->epsfac);
 603     charge           = mdatoms->chargeA;
 604
 605     /* Setup water-specific parameters */
 606     inr              = nlist->iinr[0];
 607     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 608     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 609     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 610
 611     /* Avoid stupid compiler warnings */
 612     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 613     j_coord_offsetA = 0;
 614     j_coord_offsetB = 0;
 615     j_coord_offsetC = 0;
 616     j_coord_offsetD = 0;
 617     j_coord_offsetE = 0;
 618     j_coord_offsetF = 0;
 619     j_coord_offsetG = 0;
 620     j_coord_offsetH = 0;
 621
 622     outeriter        = 0;
 623     inneriter        = 0;
 624
 625     for(iidx=0;iidx<4*DIM;iidx++)
 626     {
 627         scratch[iidx] = 0.0;
 628     }
 629
 630     /* Start outer loop over neighborlists */
 631     for(iidx=0; iidx<nri; iidx++)
 632     {
 633         /* Load shift vector for this list */
 634         i_shift_offset   = DIM*shiftidx[iidx];
 635
 636         /* Load limits for loop over neighbors */
 637         j_index_start    = jindex[iidx];
 638         j_index_end      = jindex[iidx+1];
 639
 640         /* Get outer coordinate index */
 641         inr              = iinr[iidx];
 642         i_coord_offset   = DIM*inr;
 643
 644         /* Load i particle coords and add shift vector */
 645         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 646                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 647
 648         fix0             = _mm256_setzero_ps();
 649         fiy0             = _mm256_setzero_ps();
 650         fiz0             = _mm256_setzero_ps();
 651         fix1             = _mm256_setzero_ps();
 652         fiy1             = _mm256_setzero_ps();
 653         fiz1             = _mm256_setzero_ps();
 654         fix2             = _mm256_setzero_ps();
 655         fiy2             = _mm256_setzero_ps();
 656         fiz2             = _mm256_setzero_ps();
 657
 658         /* Start inner kernel loop */
 659         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 660         {
 661
 662             /* Get j neighbor index, and coordinate index */
 663             jnrA             = jjnr[jidx];
 664             jnrB             = jjnr[jidx+1];
 665             jnrC             = jjnr[jidx+2];
 666             jnrD             = jjnr[jidx+3];
 667             jnrE             = jjnr[jidx+4];
 668             jnrF             = jjnr[jidx+5];
 669             jnrG             = jjnr[jidx+6];
 670             jnrH             = jjnr[jidx+7];
 671             j_coord_offsetA  = DIM*jnrA;
 672             j_coord_offsetB  = DIM*jnrB;
 673             j_coord_offsetC  = DIM*jnrC;
 674             j_coord_offsetD  = DIM*jnrD;
 675             j_coord_offsetE  = DIM*jnrE;
 676             j_coord_offsetF  = DIM*jnrF;
 677             j_coord_offsetG  = DIM*jnrG;
 678             j_coord_offsetH  = DIM*jnrH;
 679
 680             /* load j atom coordinates */
 681             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 682                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 683                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 684                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 685                                                  &jx0,&jy0,&jz0);
 686
 687             /* Calculate displacement vector */
 688             dx00             = _mm256_sub_ps(ix0,jx0);
 689             dy00             = _mm256_sub_ps(iy0,jy0);
 690             dz00             = _mm256_sub_ps(iz0,jz0);
 691             dx10             = _mm256_sub_ps(ix1,jx0);
 692             dy10             = _mm256_sub_ps(iy1,jy0);
 693             dz10             = _mm256_sub_ps(iz1,jz0);
 694             dx20             = _mm256_sub_ps(ix2,jx0);
 695             dy20             = _mm256_sub_ps(iy2,jy0);
 696             dz20             = _mm256_sub_ps(iz2,jz0);
 697
 698             /* Calculate squared distance and things based on it */
 699             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 700             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 701             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 702
 703             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 704             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 705             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 706
 707             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 708             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 709             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 710
 711             /* Load parameters for j particles */
 712             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 713                                                                  charge+jnrC+0,charge+jnrD+0,
 714                                                                  charge+jnrE+0,charge+jnrF+0,
 715                                                                  charge+jnrG+0,charge+jnrH+0);
 716
 717             fjx0             = _mm256_setzero_ps();
 718             fjy0             = _mm256_setzero_ps();
 719             fjz0             = _mm256_setzero_ps();
 720
 721             /**************************
 722              * CALCULATE INTERACTIONS *
 723              **************************/
 724
 725             /* Compute parameters for interactions between i and j atoms */
 726             qq00             = _mm256_mul_ps(iq0,jq0);
 727
 728             /* COULOMB ELECTROSTATICS */
 729             velec            = _mm256_mul_ps(qq00,rinv00);
 730             felec            = _mm256_mul_ps(velec,rinvsq00);
 731
 732             fscal            = felec;
 733
 734             /* Calculate temporary vectorial force */
 735             tx               = _mm256_mul_ps(fscal,dx00);
 736             ty               = _mm256_mul_ps(fscal,dy00);
 737             tz               = _mm256_mul_ps(fscal,dz00);
 738
 739             /* Update vectorial force */
 740             fix0             = _mm256_add_ps(fix0,tx);
 741             fiy0             = _mm256_add_ps(fiy0,ty);
 742             fiz0             = _mm256_add_ps(fiz0,tz);
 743
 744             fjx0             = _mm256_add_ps(fjx0,tx);
 745             fjy0             = _mm256_add_ps(fjy0,ty);
 746             fjz0             = _mm256_add_ps(fjz0,tz);
 747
 748             /**************************
 749              * CALCULATE INTERACTIONS *
 750              **************************/
 751
 752             /* Compute parameters for interactions between i and j atoms */
 753             qq10             = _mm256_mul_ps(iq1,jq0);
 754
 755             /* COULOMB ELECTROSTATICS */
 756             velec            = _mm256_mul_ps(qq10,rinv10);
 757             felec            = _mm256_mul_ps(velec,rinvsq10);
 758
 759             fscal            = felec;
 760
 761             /* Calculate temporary vectorial force */
 762             tx               = _mm256_mul_ps(fscal,dx10);
 763             ty               = _mm256_mul_ps(fscal,dy10);
 764             tz               = _mm256_mul_ps(fscal,dz10);
 765
 766             /* Update vectorial force */
 767             fix1             = _mm256_add_ps(fix1,tx);
 768             fiy1             = _mm256_add_ps(fiy1,ty);
 769             fiz1             = _mm256_add_ps(fiz1,tz);
 770
 771             fjx0             = _mm256_add_ps(fjx0,tx);
 772             fjy0             = _mm256_add_ps(fjy0,ty);
 773             fjz0             = _mm256_add_ps(fjz0,tz);
 774
 775             /**************************
 776              * CALCULATE INTERACTIONS *
 777              **************************/
 778
 779             /* Compute parameters for interactions between i and j atoms */
 780             qq20             = _mm256_mul_ps(iq2,jq0);
 781
 782             /* COULOMB ELECTROSTATICS */
 783             velec            = _mm256_mul_ps(qq20,rinv20);
 784             felec            = _mm256_mul_ps(velec,rinvsq20);
 785
 786             fscal            = felec;
 787
 788             /* Calculate temporary vectorial force */
 789             tx               = _mm256_mul_ps(fscal,dx20);
 790             ty               = _mm256_mul_ps(fscal,dy20);
 791             tz               = _mm256_mul_ps(fscal,dz20);
 792
 793             /* Update vectorial force */
 794             fix2             = _mm256_add_ps(fix2,tx);
 795             fiy2             = _mm256_add_ps(fiy2,ty);
 796             fiz2             = _mm256_add_ps(fiz2,tz);
 797
 798             fjx0             = _mm256_add_ps(fjx0,tx);
 799             fjy0             = _mm256_add_ps(fjy0,ty);
 800             fjz0             = _mm256_add_ps(fjz0,tz);
 801
 802             fjptrA             = f+j_coord_offsetA;
 803             fjptrB             = f+j_coord_offsetB;
 804             fjptrC             = f+j_coord_offsetC;
 805             fjptrD             = f+j_coord_offsetD;
 806             fjptrE             = f+j_coord_offsetE;
 807             fjptrF             = f+j_coord_offsetF;
 808             fjptrG             = f+j_coord_offsetG;
 809             fjptrH             = f+j_coord_offsetH;
 810
 811             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 812
 813             /* Inner loop uses 81 flops */
 814         }
 815
 816         if(jidx<j_index_end)
 817         {
 818
 819             /* Get j neighbor index, and coordinate index */
 820             jnrlistA         = jjnr[jidx];
 821             jnrlistB         = jjnr[jidx+1];
 822             jnrlistC         = jjnr[jidx+2];
 823             jnrlistD         = jjnr[jidx+3];
 824             jnrlistE         = jjnr[jidx+4];
 825             jnrlistF         = jjnr[jidx+5];
 826             jnrlistG         = jjnr[jidx+6];
 827             jnrlistH         = jjnr[jidx+7];
 828             /* Sign of each element will be negative for non-real atoms.
 829              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 830              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 831              */
 832             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 833                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 834
 835             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 836             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 837             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 838             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 839             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 840             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 841             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 842             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 843             j_coord_offsetA  = DIM*jnrA;
 844             j_coord_offsetB  = DIM*jnrB;
 845             j_coord_offsetC  = DIM*jnrC;
 846             j_coord_offsetD  = DIM*jnrD;
 847             j_coord_offsetE  = DIM*jnrE;
 848             j_coord_offsetF  = DIM*jnrF;
 849             j_coord_offsetG  = DIM*jnrG;
 850             j_coord_offsetH  = DIM*jnrH;
 851
 852             /* load j atom coordinates */
 853             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 854                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 855                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 856                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 857                                                  &jx0,&jy0,&jz0);
 858
 859             /* Calculate displacement vector */
 860             dx00             = _mm256_sub_ps(ix0,jx0);
 861             dy00             = _mm256_sub_ps(iy0,jy0);
 862             dz00             = _mm256_sub_ps(iz0,jz0);
 863             dx10             = _mm256_sub_ps(ix1,jx0);
 864             dy10             = _mm256_sub_ps(iy1,jy0);
 865             dz10             = _mm256_sub_ps(iz1,jz0);
 866             dx20             = _mm256_sub_ps(ix2,jx0);
 867             dy20             = _mm256_sub_ps(iy2,jy0);
 868             dz20             = _mm256_sub_ps(iz2,jz0);
 869
 870             /* Calculate squared distance and things based on it */
 871             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 872             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 873             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 874
 875             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 876             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 877             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 878
 879             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 880             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 881             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 882
 883             /* Load parameters for j particles */
 884             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 885                                                                  charge+jnrC+0,charge+jnrD+0,
 886                                                                  charge+jnrE+0,charge+jnrF+0,
 887                                                                  charge+jnrG+0,charge+jnrH+0);
 888
 889             fjx0             = _mm256_setzero_ps();
 890             fjy0             = _mm256_setzero_ps();
 891             fjz0             = _mm256_setzero_ps();
 892
 893             /**************************
 894              * CALCULATE INTERACTIONS *
 895              **************************/
 896
 897             /* Compute parameters for interactions between i and j atoms */
 898             qq00             = _mm256_mul_ps(iq0,jq0);
 899
 900             /* COULOMB ELECTROSTATICS */
 901             velec            = _mm256_mul_ps(qq00,rinv00);
 902             felec            = _mm256_mul_ps(velec,rinvsq00);
 903
 904             fscal            = felec;
 905
 906             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 907
 908             /* Calculate temporary vectorial force */
 909             tx               = _mm256_mul_ps(fscal,dx00);
 910             ty               = _mm256_mul_ps(fscal,dy00);
 911             tz               = _mm256_mul_ps(fscal,dz00);
 912
 913             /* Update vectorial force */
 914             fix0             = _mm256_add_ps(fix0,tx);
 915             fiy0             = _mm256_add_ps(fiy0,ty);
 916             fiz0             = _mm256_add_ps(fiz0,tz);
 917
 918             fjx0             = _mm256_add_ps(fjx0,tx);
 919             fjy0             = _mm256_add_ps(fjy0,ty);
 920             fjz0             = _mm256_add_ps(fjz0,tz);
 921
 922             /**************************
 923              * CALCULATE INTERACTIONS *
 924              **************************/
 925
 926             /* Compute parameters for interactions between i and j atoms */
 927             qq10             = _mm256_mul_ps(iq1,jq0);
 928
 929             /* COULOMB ELECTROSTATICS */
 930             velec            = _mm256_mul_ps(qq10,rinv10);
 931             felec            = _mm256_mul_ps(velec,rinvsq10);
 932
 933             fscal            = felec;
 934
 935             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 936
 937             /* Calculate temporary vectorial force */
 938             tx               = _mm256_mul_ps(fscal,dx10);
 939             ty               = _mm256_mul_ps(fscal,dy10);
 940             tz               = _mm256_mul_ps(fscal,dz10);
 941
 942             /* Update vectorial force */
 943             fix1             = _mm256_add_ps(fix1,tx);
 944             fiy1             = _mm256_add_ps(fiy1,ty);
 945             fiz1             = _mm256_add_ps(fiz1,tz);
 946
 947             fjx0             = _mm256_add_ps(fjx0,tx);
 948             fjy0             = _mm256_add_ps(fjy0,ty);
 949             fjz0             = _mm256_add_ps(fjz0,tz);
 950
 951             /**************************
 952              * CALCULATE INTERACTIONS *
 953              **************************/
 954
 955             /* Compute parameters for interactions between i and j atoms */
 956             qq20             = _mm256_mul_ps(iq2,jq0);
 957
 958             /* COULOMB ELECTROSTATICS */
 959             velec            = _mm256_mul_ps(qq20,rinv20);
 960             felec            = _mm256_mul_ps(velec,rinvsq20);
 961
 962             fscal            = felec;
 963
 964             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 965
 966             /* Calculate temporary vectorial force */
 967             tx               = _mm256_mul_ps(fscal,dx20);
 968             ty               = _mm256_mul_ps(fscal,dy20);
 969             tz               = _mm256_mul_ps(fscal,dz20);
 970
 971             /* Update vectorial force */
 972             fix2             = _mm256_add_ps(fix2,tx);
 973             fiy2             = _mm256_add_ps(fiy2,ty);
 974             fiz2             = _mm256_add_ps(fiz2,tz);
 975
 976             fjx0             = _mm256_add_ps(fjx0,tx);
 977             fjy0             = _mm256_add_ps(fjy0,ty);
 978             fjz0             = _mm256_add_ps(fjz0,tz);
 979
 980             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 981             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 982             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 983             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 984             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 985             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 986             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 987             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 988
 989             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 990
 991             /* Inner loop uses 81 flops */
 992         }
 993
 994         /* End of innermost loop */
 995
 996         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 997                                                  f+i_coord_offset,fshift+i_shift_offset);
 998
 999         /* Increment number of inner iterations */
1000         inneriter                  += j_index_end - j_index_start;
1001
1002         /* Outer loop uses 18 flops */
1003     }
1004
1005     /* Increment number of outer iterations */
1006     outeriter        += nri;
1007
1008     /* Update outer/inner flops */
1009
1010     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*81);
1011 }