src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwNone_GeomW3W3_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3W3_VF_avx_256_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            None
  40  * Geometry:                   Water3-Water3
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwNone_GeomW3W3_VF_avx_256_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  77     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  78     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  79     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  80     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  81     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  82     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  83     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  84     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  85     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  86     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  87     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  88     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  89     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  90     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  91     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m256d          dummy_mask,cutoff_mask;
  94     __m128           tmpmask0,tmpmask1;
  95     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  96     __m256d          one     = _mm256_set1_pd(1.0);
  97     __m256d          two     = _mm256_set1_pd(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = _mm256_set1_pd(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111
 112     /* Setup water-specific parameters */
 113     inr              = nlist->iinr[0];
 114     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 115     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 116     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 117
 118     jq0              = _mm256_set1_pd(charge[inr+0]);
 119     jq1              = _mm256_set1_pd(charge[inr+1]);
 120     jq2              = _mm256_set1_pd(charge[inr+2]);
 121     qq00             = _mm256_mul_pd(iq0,jq0);
 122     qq01             = _mm256_mul_pd(iq0,jq1);
 123     qq02             = _mm256_mul_pd(iq0,jq2);
 124     qq10             = _mm256_mul_pd(iq1,jq0);
 125     qq11             = _mm256_mul_pd(iq1,jq1);
 126     qq12             = _mm256_mul_pd(iq1,jq2);
 127     qq20             = _mm256_mul_pd(iq2,jq0);
 128     qq21             = _mm256_mul_pd(iq2,jq1);
 129     qq22             = _mm256_mul_pd(iq2,jq2);
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     for(iidx=0;iidx<4*DIM;iidx++)
 142     {
 143         scratch[iidx] = 0.0;
 144     }
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 162                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 163
 164         fix0             = _mm256_setzero_pd();
 165         fiy0             = _mm256_setzero_pd();
 166         fiz0             = _mm256_setzero_pd();
 167         fix1             = _mm256_setzero_pd();
 168         fiy1             = _mm256_setzero_pd();
 169         fiz1             = _mm256_setzero_pd();
 170         fix2             = _mm256_setzero_pd();
 171         fiy2             = _mm256_setzero_pd();
 172         fiz2             = _mm256_setzero_pd();
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm256_setzero_pd();
 176
 177         /* Start inner kernel loop */
 178         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 179         {
 180
 181             /* Get j neighbor index, and coordinate index */
 182             jnrA             = jjnr[jidx];
 183             jnrB             = jjnr[jidx+1];
 184             jnrC             = jjnr[jidx+2];
 185             jnrD             = jjnr[jidx+3];
 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
 191             /* load j atom coordinates */
 192             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 193                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 194                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm256_sub_pd(ix0,jx0);
 198             dy00             = _mm256_sub_pd(iy0,jy0);
 199             dz00             = _mm256_sub_pd(iz0,jz0);
 200             dx01             = _mm256_sub_pd(ix0,jx1);
 201             dy01             = _mm256_sub_pd(iy0,jy1);
 202             dz01             = _mm256_sub_pd(iz0,jz1);
 203             dx02             = _mm256_sub_pd(ix0,jx2);
 204             dy02             = _mm256_sub_pd(iy0,jy2);
 205             dz02             = _mm256_sub_pd(iz0,jz2);
 206             dx10             = _mm256_sub_pd(ix1,jx0);
 207             dy10             = _mm256_sub_pd(iy1,jy0);
 208             dz10             = _mm256_sub_pd(iz1,jz0);
 209             dx11             = _mm256_sub_pd(ix1,jx1);
 210             dy11             = _mm256_sub_pd(iy1,jy1);
 211             dz11             = _mm256_sub_pd(iz1,jz1);
 212             dx12             = _mm256_sub_pd(ix1,jx2);
 213             dy12             = _mm256_sub_pd(iy1,jy2);
 214             dz12             = _mm256_sub_pd(iz1,jz2);
 215             dx20             = _mm256_sub_pd(ix2,jx0);
 216             dy20             = _mm256_sub_pd(iy2,jy0);
 217             dz20             = _mm256_sub_pd(iz2,jz0);
 218             dx21             = _mm256_sub_pd(ix2,jx1);
 219             dy21             = _mm256_sub_pd(iy2,jy1);
 220             dz21             = _mm256_sub_pd(iz2,jz1);
 221             dx22             = _mm256_sub_pd(ix2,jx2);
 222             dy22             = _mm256_sub_pd(iy2,jy2);
 223             dz22             = _mm256_sub_pd(iz2,jz2);
 224
 225             /* Calculate squared distance and things based on it */
 226             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 227             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 228             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 229             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 230             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 231             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 232             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 233             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 234             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 235
 236             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 237             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 238             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 239             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 240             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 241             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 242             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 243             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 244             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 245
 246             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 247             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 248             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 249             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 250             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 251             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 252             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 253             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 254             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 255
 256             fjx0             = _mm256_setzero_pd();
 257             fjy0             = _mm256_setzero_pd();
 258             fjz0             = _mm256_setzero_pd();
 259             fjx1             = _mm256_setzero_pd();
 260             fjy1             = _mm256_setzero_pd();
 261             fjz1             = _mm256_setzero_pd();
 262             fjx2             = _mm256_setzero_pd();
 263             fjy2             = _mm256_setzero_pd();
 264             fjz2             = _mm256_setzero_pd();
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             /* COULOMB ELECTROSTATICS */
 271             velec            = _mm256_mul_pd(qq00,rinv00);
 272             felec            = _mm256_mul_pd(velec,rinvsq00);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             velecsum         = _mm256_add_pd(velecsum,velec);
 276
 277             fscal            = felec;
 278
 279             /* Calculate temporary vectorial force */
 280             tx               = _mm256_mul_pd(fscal,dx00);
 281             ty               = _mm256_mul_pd(fscal,dy00);
 282             tz               = _mm256_mul_pd(fscal,dz00);
 283
 284             /* Update vectorial force */
 285             fix0             = _mm256_add_pd(fix0,tx);
 286             fiy0             = _mm256_add_pd(fiy0,ty);
 287             fiz0             = _mm256_add_pd(fiz0,tz);
 288
 289             fjx0             = _mm256_add_pd(fjx0,tx);
 290             fjy0             = _mm256_add_pd(fjy0,ty);
 291             fjz0             = _mm256_add_pd(fjz0,tz);
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             /* COULOMB ELECTROSTATICS */
 298             velec            = _mm256_mul_pd(qq01,rinv01);
 299             felec            = _mm256_mul_pd(velec,rinvsq01);
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             velecsum         = _mm256_add_pd(velecsum,velec);
 303
 304             fscal            = felec;
 305
 306             /* Calculate temporary vectorial force */
 307             tx               = _mm256_mul_pd(fscal,dx01);
 308             ty               = _mm256_mul_pd(fscal,dy01);
 309             tz               = _mm256_mul_pd(fscal,dz01);
 310
 311             /* Update vectorial force */
 312             fix0             = _mm256_add_pd(fix0,tx);
 313             fiy0             = _mm256_add_pd(fiy0,ty);
 314             fiz0             = _mm256_add_pd(fiz0,tz);
 315
 316             fjx1             = _mm256_add_pd(fjx1,tx);
 317             fjy1             = _mm256_add_pd(fjy1,ty);
 318             fjz1             = _mm256_add_pd(fjz1,tz);
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             /* COULOMB ELECTROSTATICS */
 325             velec            = _mm256_mul_pd(qq02,rinv02);
 326             felec            = _mm256_mul_pd(velec,rinvsq02);
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm256_add_pd(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm256_mul_pd(fscal,dx02);
 335             ty               = _mm256_mul_pd(fscal,dy02);
 336             tz               = _mm256_mul_pd(fscal,dz02);
 337
 338             /* Update vectorial force */
 339             fix0             = _mm256_add_pd(fix0,tx);
 340             fiy0             = _mm256_add_pd(fiy0,ty);
 341             fiz0             = _mm256_add_pd(fiz0,tz);
 342
 343             fjx2             = _mm256_add_pd(fjx2,tx);
 344             fjy2             = _mm256_add_pd(fjy2,ty);
 345             fjz2             = _mm256_add_pd(fjz2,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             /* COULOMB ELECTROSTATICS */
 352             velec            = _mm256_mul_pd(qq10,rinv10);
 353             felec            = _mm256_mul_pd(velec,rinvsq10);
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velecsum         = _mm256_add_pd(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm256_mul_pd(fscal,dx10);
 362             ty               = _mm256_mul_pd(fscal,dy10);
 363             tz               = _mm256_mul_pd(fscal,dz10);
 364
 365             /* Update vectorial force */
 366             fix1             = _mm256_add_pd(fix1,tx);
 367             fiy1             = _mm256_add_pd(fiy1,ty);
 368             fiz1             = _mm256_add_pd(fiz1,tz);
 369
 370             fjx0             = _mm256_add_pd(fjx0,tx);
 371             fjy0             = _mm256_add_pd(fjy0,ty);
 372             fjz0             = _mm256_add_pd(fjz0,tz);
 373
 374             /**************************
 375              * CALCULATE INTERACTIONS *
 376              **************************/
 377
 378             /* COULOMB ELECTROSTATICS */
 379             velec            = _mm256_mul_pd(qq11,rinv11);
 380             felec            = _mm256_mul_pd(velec,rinvsq11);
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             velecsum         = _mm256_add_pd(velecsum,velec);
 384
 385             fscal            = felec;
 386
 387             /* Calculate temporary vectorial force */
 388             tx               = _mm256_mul_pd(fscal,dx11);
 389             ty               = _mm256_mul_pd(fscal,dy11);
 390             tz               = _mm256_mul_pd(fscal,dz11);
 391
 392             /* Update vectorial force */
 393             fix1             = _mm256_add_pd(fix1,tx);
 394             fiy1             = _mm256_add_pd(fiy1,ty);
 395             fiz1             = _mm256_add_pd(fiz1,tz);
 396
 397             fjx1             = _mm256_add_pd(fjx1,tx);
 398             fjy1             = _mm256_add_pd(fjy1,ty);
 399             fjz1             = _mm256_add_pd(fjz1,tz);
 400
 401             /**************************
 402              * CALCULATE INTERACTIONS *
 403              **************************/
 404
 405             /* COULOMB ELECTROSTATICS */
 406             velec            = _mm256_mul_pd(qq12,rinv12);
 407             felec            = _mm256_mul_pd(velec,rinvsq12);
 408
 409             /* Update potential sum for this i atom from the interaction with this j atom. */
 410             velecsum         = _mm256_add_pd(velecsum,velec);
 411
 412             fscal            = felec;
 413
 414             /* Calculate temporary vectorial force */
 415             tx               = _mm256_mul_pd(fscal,dx12);
 416             ty               = _mm256_mul_pd(fscal,dy12);
 417             tz               = _mm256_mul_pd(fscal,dz12);
 418
 419             /* Update vectorial force */
 420             fix1             = _mm256_add_pd(fix1,tx);
 421             fiy1             = _mm256_add_pd(fiy1,ty);
 422             fiz1             = _mm256_add_pd(fiz1,tz);
 423
 424             fjx2             = _mm256_add_pd(fjx2,tx);
 425             fjy2             = _mm256_add_pd(fjy2,ty);
 426             fjz2             = _mm256_add_pd(fjz2,tz);
 427
 428             /**************************
 429              * CALCULATE INTERACTIONS *
 430              **************************/
 431
 432             /* COULOMB ELECTROSTATICS */
 433             velec            = _mm256_mul_pd(qq20,rinv20);
 434             felec            = _mm256_mul_pd(velec,rinvsq20);
 435
 436             /* Update potential sum for this i atom from the interaction with this j atom. */
 437             velecsum         = _mm256_add_pd(velecsum,velec);
 438
 439             fscal            = felec;
 440
 441             /* Calculate temporary vectorial force */
 442             tx               = _mm256_mul_pd(fscal,dx20);
 443             ty               = _mm256_mul_pd(fscal,dy20);
 444             tz               = _mm256_mul_pd(fscal,dz20);
 445
 446             /* Update vectorial force */
 447             fix2             = _mm256_add_pd(fix2,tx);
 448             fiy2             = _mm256_add_pd(fiy2,ty);
 449             fiz2             = _mm256_add_pd(fiz2,tz);
 450
 451             fjx0             = _mm256_add_pd(fjx0,tx);
 452             fjy0             = _mm256_add_pd(fjy0,ty);
 453             fjz0             = _mm256_add_pd(fjz0,tz);
 454
 455             /**************************
 456              * CALCULATE INTERACTIONS *
 457              **************************/
 458
 459             /* COULOMB ELECTROSTATICS */
 460             velec            = _mm256_mul_pd(qq21,rinv21);
 461             felec            = _mm256_mul_pd(velec,rinvsq21);
 462
 463             /* Update potential sum for this i atom from the interaction with this j atom. */
 464             velecsum         = _mm256_add_pd(velecsum,velec);
 465
 466             fscal            = felec;
 467
 468             /* Calculate temporary vectorial force */
 469             tx               = _mm256_mul_pd(fscal,dx21);
 470             ty               = _mm256_mul_pd(fscal,dy21);
 471             tz               = _mm256_mul_pd(fscal,dz21);
 472
 473             /* Update vectorial force */
 474             fix2             = _mm256_add_pd(fix2,tx);
 475             fiy2             = _mm256_add_pd(fiy2,ty);
 476             fiz2             = _mm256_add_pd(fiz2,tz);
 477
 478             fjx1             = _mm256_add_pd(fjx1,tx);
 479             fjy1             = _mm256_add_pd(fjy1,ty);
 480             fjz1             = _mm256_add_pd(fjz1,tz);
 481
 482             /**************************
 483              * CALCULATE INTERACTIONS *
 484              **************************/
 485
 486             /* COULOMB ELECTROSTATICS */
 487             velec            = _mm256_mul_pd(qq22,rinv22);
 488             felec            = _mm256_mul_pd(velec,rinvsq22);
 489
 490             /* Update potential sum for this i atom from the interaction with this j atom. */
 491             velecsum         = _mm256_add_pd(velecsum,velec);
 492
 493             fscal            = felec;
 494
 495             /* Calculate temporary vectorial force */
 496             tx               = _mm256_mul_pd(fscal,dx22);
 497             ty               = _mm256_mul_pd(fscal,dy22);
 498             tz               = _mm256_mul_pd(fscal,dz22);
 499
 500             /* Update vectorial force */
 501             fix2             = _mm256_add_pd(fix2,tx);
 502             fiy2             = _mm256_add_pd(fiy2,ty);
 503             fiz2             = _mm256_add_pd(fiz2,tz);
 504
 505             fjx2             = _mm256_add_pd(fjx2,tx);
 506             fjy2             = _mm256_add_pd(fjy2,ty);
 507             fjz2             = _mm256_add_pd(fjz2,tz);
 508
 509             fjptrA             = f+j_coord_offsetA;
 510             fjptrB             = f+j_coord_offsetB;
 511             fjptrC             = f+j_coord_offsetC;
 512             fjptrD             = f+j_coord_offsetD;
 513
 514             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 515                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 516
 517             /* Inner loop uses 243 flops */
 518         }
 519
 520         if(jidx<j_index_end)
 521         {
 522
 523             /* Get j neighbor index, and coordinate index */
 524             jnrlistA         = jjnr[jidx];
 525             jnrlistB         = jjnr[jidx+1];
 526             jnrlistC         = jjnr[jidx+2];
 527             jnrlistD         = jjnr[jidx+3];
 528             /* Sign of each element will be negative for non-real atoms.
 529              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 530              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 531              */
 532             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 533
 534             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 535             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 536             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 537
 538             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 539             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 540             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 541             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 542             j_coord_offsetA  = DIM*jnrA;
 543             j_coord_offsetB  = DIM*jnrB;
 544             j_coord_offsetC  = DIM*jnrC;
 545             j_coord_offsetD  = DIM*jnrD;
 546
 547             /* load j atom coordinates */
 548             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 549                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 550                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 551
 552             /* Calculate displacement vector */
 553             dx00             = _mm256_sub_pd(ix0,jx0);
 554             dy00             = _mm256_sub_pd(iy0,jy0);
 555             dz00             = _mm256_sub_pd(iz0,jz0);
 556             dx01             = _mm256_sub_pd(ix0,jx1);
 557             dy01             = _mm256_sub_pd(iy0,jy1);
 558             dz01             = _mm256_sub_pd(iz0,jz1);
 559             dx02             = _mm256_sub_pd(ix0,jx2);
 560             dy02             = _mm256_sub_pd(iy0,jy2);
 561             dz02             = _mm256_sub_pd(iz0,jz2);
 562             dx10             = _mm256_sub_pd(ix1,jx0);
 563             dy10             = _mm256_sub_pd(iy1,jy0);
 564             dz10             = _mm256_sub_pd(iz1,jz0);
 565             dx11             = _mm256_sub_pd(ix1,jx1);
 566             dy11             = _mm256_sub_pd(iy1,jy1);
 567             dz11             = _mm256_sub_pd(iz1,jz1);
 568             dx12             = _mm256_sub_pd(ix1,jx2);
 569             dy12             = _mm256_sub_pd(iy1,jy2);
 570             dz12             = _mm256_sub_pd(iz1,jz2);
 571             dx20             = _mm256_sub_pd(ix2,jx0);
 572             dy20             = _mm256_sub_pd(iy2,jy0);
 573             dz20             = _mm256_sub_pd(iz2,jz0);
 574             dx21             = _mm256_sub_pd(ix2,jx1);
 575             dy21             = _mm256_sub_pd(iy2,jy1);
 576             dz21             = _mm256_sub_pd(iz2,jz1);
 577             dx22             = _mm256_sub_pd(ix2,jx2);
 578             dy22             = _mm256_sub_pd(iy2,jy2);
 579             dz22             = _mm256_sub_pd(iz2,jz2);
 580
 581             /* Calculate squared distance and things based on it */
 582             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 583             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 584             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 585             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 586             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 587             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 588             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 589             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 590             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 591
 592             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 593             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 594             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 595             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 596             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 597             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 598             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 599             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 600             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 601
 602             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 603             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 604             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 605             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 606             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 607             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 608             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 609             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 610             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 611
 612             fjx0             = _mm256_setzero_pd();
 613             fjy0             = _mm256_setzero_pd();
 614             fjz0             = _mm256_setzero_pd();
 615             fjx1             = _mm256_setzero_pd();
 616             fjy1             = _mm256_setzero_pd();
 617             fjz1             = _mm256_setzero_pd();
 618             fjx2             = _mm256_setzero_pd();
 619             fjy2             = _mm256_setzero_pd();
 620             fjz2             = _mm256_setzero_pd();
 621
 622             /**************************
 623              * CALCULATE INTERACTIONS *
 624              **************************/
 625
 626             /* COULOMB ELECTROSTATICS */
 627             velec            = _mm256_mul_pd(qq00,rinv00);
 628             felec            = _mm256_mul_pd(velec,rinvsq00);
 629
 630             /* Update potential sum for this i atom from the interaction with this j atom. */
 631             velec            = _mm256_andnot_pd(dummy_mask,velec);
 632             velecsum         = _mm256_add_pd(velecsum,velec);
 633
 634             fscal            = felec;
 635
 636             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 637
 638             /* Calculate temporary vectorial force */
 639             tx               = _mm256_mul_pd(fscal,dx00);
 640             ty               = _mm256_mul_pd(fscal,dy00);
 641             tz               = _mm256_mul_pd(fscal,dz00);
 642
 643             /* Update vectorial force */
 644             fix0             = _mm256_add_pd(fix0,tx);
 645             fiy0             = _mm256_add_pd(fiy0,ty);
 646             fiz0             = _mm256_add_pd(fiz0,tz);
 647
 648             fjx0             = _mm256_add_pd(fjx0,tx);
 649             fjy0             = _mm256_add_pd(fjy0,ty);
 650             fjz0             = _mm256_add_pd(fjz0,tz);
 651
 652             /**************************
 653              * CALCULATE INTERACTIONS *
 654              **************************/
 655
 656             /* COULOMB ELECTROSTATICS */
 657             velec            = _mm256_mul_pd(qq01,rinv01);
 658             felec            = _mm256_mul_pd(velec,rinvsq01);
 659
 660             /* Update potential sum for this i atom from the interaction with this j atom. */
 661             velec            = _mm256_andnot_pd(dummy_mask,velec);
 662             velecsum         = _mm256_add_pd(velecsum,velec);
 663
 664             fscal            = felec;
 665
 666             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 667
 668             /* Calculate temporary vectorial force */
 669             tx               = _mm256_mul_pd(fscal,dx01);
 670             ty               = _mm256_mul_pd(fscal,dy01);
 671             tz               = _mm256_mul_pd(fscal,dz01);
 672
 673             /* Update vectorial force */
 674             fix0             = _mm256_add_pd(fix0,tx);
 675             fiy0             = _mm256_add_pd(fiy0,ty);
 676             fiz0             = _mm256_add_pd(fiz0,tz);
 677
 678             fjx1             = _mm256_add_pd(fjx1,tx);
 679             fjy1             = _mm256_add_pd(fjy1,ty);
 680             fjz1             = _mm256_add_pd(fjz1,tz);
 681
 682             /**************************
 683              * CALCULATE INTERACTIONS *
 684              **************************/
 685
 686             /* COULOMB ELECTROSTATICS */
 687             velec            = _mm256_mul_pd(qq02,rinv02);
 688             felec            = _mm256_mul_pd(velec,rinvsq02);
 689
 690             /* Update potential sum for this i atom from the interaction with this j atom. */
 691             velec            = _mm256_andnot_pd(dummy_mask,velec);
 692             velecsum         = _mm256_add_pd(velecsum,velec);
 693
 694             fscal            = felec;
 695
 696             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 697
 698             /* Calculate temporary vectorial force */
 699             tx               = _mm256_mul_pd(fscal,dx02);
 700             ty               = _mm256_mul_pd(fscal,dy02);
 701             tz               = _mm256_mul_pd(fscal,dz02);
 702
 703             /* Update vectorial force */
 704             fix0             = _mm256_add_pd(fix0,tx);
 705             fiy0             = _mm256_add_pd(fiy0,ty);
 706             fiz0             = _mm256_add_pd(fiz0,tz);
 707
 708             fjx2             = _mm256_add_pd(fjx2,tx);
 709             fjy2             = _mm256_add_pd(fjy2,ty);
 710             fjz2             = _mm256_add_pd(fjz2,tz);
 711
 712             /**************************
 713              * CALCULATE INTERACTIONS *
 714              **************************/
 715
 716             /* COULOMB ELECTROSTATICS */
 717             velec            = _mm256_mul_pd(qq10,rinv10);
 718             felec            = _mm256_mul_pd(velec,rinvsq10);
 719
 720             /* Update potential sum for this i atom from the interaction with this j atom. */
 721             velec            = _mm256_andnot_pd(dummy_mask,velec);
 722             velecsum         = _mm256_add_pd(velecsum,velec);
 723
 724             fscal            = felec;
 725
 726             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 727
 728             /* Calculate temporary vectorial force */
 729             tx               = _mm256_mul_pd(fscal,dx10);
 730             ty               = _mm256_mul_pd(fscal,dy10);
 731             tz               = _mm256_mul_pd(fscal,dz10);
 732
 733             /* Update vectorial force */
 734             fix1             = _mm256_add_pd(fix1,tx);
 735             fiy1             = _mm256_add_pd(fiy1,ty);
 736             fiz1             = _mm256_add_pd(fiz1,tz);
 737
 738             fjx0             = _mm256_add_pd(fjx0,tx);
 739             fjy0             = _mm256_add_pd(fjy0,ty);
 740             fjz0             = _mm256_add_pd(fjz0,tz);
 741
 742             /**************************
 743              * CALCULATE INTERACTIONS *
 744              **************************/
 745
 746             /* COULOMB ELECTROSTATICS */
 747             velec            = _mm256_mul_pd(qq11,rinv11);
 748             felec            = _mm256_mul_pd(velec,rinvsq11);
 749
 750             /* Update potential sum for this i atom from the interaction with this j atom. */
 751             velec            = _mm256_andnot_pd(dummy_mask,velec);
 752             velecsum         = _mm256_add_pd(velecsum,velec);
 753
 754             fscal            = felec;
 755
 756             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 757
 758             /* Calculate temporary vectorial force */
 759             tx               = _mm256_mul_pd(fscal,dx11);
 760             ty               = _mm256_mul_pd(fscal,dy11);
 761             tz               = _mm256_mul_pd(fscal,dz11);
 762
 763             /* Update vectorial force */
 764             fix1             = _mm256_add_pd(fix1,tx);
 765             fiy1             = _mm256_add_pd(fiy1,ty);
 766             fiz1             = _mm256_add_pd(fiz1,tz);
 767
 768             fjx1             = _mm256_add_pd(fjx1,tx);
 769             fjy1             = _mm256_add_pd(fjy1,ty);
 770             fjz1             = _mm256_add_pd(fjz1,tz);
 771
 772             /**************************
 773              * CALCULATE INTERACTIONS *
 774              **************************/
 775
 776             /* COULOMB ELECTROSTATICS */
 777             velec            = _mm256_mul_pd(qq12,rinv12);
 778             felec            = _mm256_mul_pd(velec,rinvsq12);
 779
 780             /* Update potential sum for this i atom from the interaction with this j atom. */
 781             velec            = _mm256_andnot_pd(dummy_mask,velec);
 782             velecsum         = _mm256_add_pd(velecsum,velec);
 783
 784             fscal            = felec;
 785
 786             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 787
 788             /* Calculate temporary vectorial force */
 789             tx               = _mm256_mul_pd(fscal,dx12);
 790             ty               = _mm256_mul_pd(fscal,dy12);
 791             tz               = _mm256_mul_pd(fscal,dz12);
 792
 793             /* Update vectorial force */
 794             fix1             = _mm256_add_pd(fix1,tx);
 795             fiy1             = _mm256_add_pd(fiy1,ty);
 796             fiz1             = _mm256_add_pd(fiz1,tz);
 797
 798             fjx2             = _mm256_add_pd(fjx2,tx);
 799             fjy2             = _mm256_add_pd(fjy2,ty);
 800             fjz2             = _mm256_add_pd(fjz2,tz);
 801
 802             /**************************
 803              * CALCULATE INTERACTIONS *
 804              **************************/
 805
 806             /* COULOMB ELECTROSTATICS */
 807             velec            = _mm256_mul_pd(qq20,rinv20);
 808             felec            = _mm256_mul_pd(velec,rinvsq20);
 809
 810             /* Update potential sum for this i atom from the interaction with this j atom. */
 811             velec            = _mm256_andnot_pd(dummy_mask,velec);
 812             velecsum         = _mm256_add_pd(velecsum,velec);
 813
 814             fscal            = felec;
 815
 816             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 817
 818             /* Calculate temporary vectorial force */
 819             tx               = _mm256_mul_pd(fscal,dx20);
 820             ty               = _mm256_mul_pd(fscal,dy20);
 821             tz               = _mm256_mul_pd(fscal,dz20);
 822
 823             /* Update vectorial force */
 824             fix2             = _mm256_add_pd(fix2,tx);
 825             fiy2             = _mm256_add_pd(fiy2,ty);
 826             fiz2             = _mm256_add_pd(fiz2,tz);
 827
 828             fjx0             = _mm256_add_pd(fjx0,tx);
 829             fjy0             = _mm256_add_pd(fjy0,ty);
 830             fjz0             = _mm256_add_pd(fjz0,tz);
 831
 832             /**************************
 833              * CALCULATE INTERACTIONS *
 834              **************************/
 835
 836             /* COULOMB ELECTROSTATICS */
 837             velec            = _mm256_mul_pd(qq21,rinv21);
 838             felec            = _mm256_mul_pd(velec,rinvsq21);
 839
 840             /* Update potential sum for this i atom from the interaction with this j atom. */
 841             velec            = _mm256_andnot_pd(dummy_mask,velec);
 842             velecsum         = _mm256_add_pd(velecsum,velec);
 843
 844             fscal            = felec;
 845
 846             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 847
 848             /* Calculate temporary vectorial force */
 849             tx               = _mm256_mul_pd(fscal,dx21);
 850             ty               = _mm256_mul_pd(fscal,dy21);
 851             tz               = _mm256_mul_pd(fscal,dz21);
 852
 853             /* Update vectorial force */
 854             fix2             = _mm256_add_pd(fix2,tx);
 855             fiy2             = _mm256_add_pd(fiy2,ty);
 856             fiz2             = _mm256_add_pd(fiz2,tz);
 857
 858             fjx1             = _mm256_add_pd(fjx1,tx);
 859             fjy1             = _mm256_add_pd(fjy1,ty);
 860             fjz1             = _mm256_add_pd(fjz1,tz);
 861
 862             /**************************
 863              * CALCULATE INTERACTIONS *
 864              **************************/
 865
 866             /* COULOMB ELECTROSTATICS */
 867             velec            = _mm256_mul_pd(qq22,rinv22);
 868             felec            = _mm256_mul_pd(velec,rinvsq22);
 869
 870             /* Update potential sum for this i atom from the interaction with this j atom. */
 871             velec            = _mm256_andnot_pd(dummy_mask,velec);
 872             velecsum         = _mm256_add_pd(velecsum,velec);
 873
 874             fscal            = felec;
 875
 876             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 877
 878             /* Calculate temporary vectorial force */
 879             tx               = _mm256_mul_pd(fscal,dx22);
 880             ty               = _mm256_mul_pd(fscal,dy22);
 881             tz               = _mm256_mul_pd(fscal,dz22);
 882
 883             /* Update vectorial force */
 884             fix2             = _mm256_add_pd(fix2,tx);
 885             fiy2             = _mm256_add_pd(fiy2,ty);
 886             fiz2             = _mm256_add_pd(fiz2,tz);
 887
 888             fjx2             = _mm256_add_pd(fjx2,tx);
 889             fjy2             = _mm256_add_pd(fjy2,ty);
 890             fjz2             = _mm256_add_pd(fjz2,tz);
 891
 892             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 893             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 894             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 895             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 896
 897             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 898                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 899
 900             /* Inner loop uses 243 flops */
 901         }
 902
 903         /* End of innermost loop */
 904
 905         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 906                                                  f+i_coord_offset,fshift+i_shift_offset);
 907
 908         ggid                        = gid[iidx];
 909         /* Update potential energies */
 910         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 911
 912         /* Increment number of inner iterations */
 913         inneriter                  += j_index_end - j_index_start;
 914
 915         /* Outer loop uses 19 flops */
 916     }
 917
 918     /* Increment number of outer iterations */
 919     outeriter        += nri;
 920
 921     /* Update outer/inner flops */
 922
 923     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*243);
 924 }
 925 /*
 926  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3W3_F_avx_256_double
 927  * Electrostatics interaction: Coulomb
 928  * VdW interaction:            None
 929  * Geometry:                   Water3-Water3
 930  * Calculate force/pot:        Force
 931  */
 932 void
 933 nb_kernel_ElecCoul_VdwNone_GeomW3W3_F_avx_256_double
 934                     (t_nblist * gmx_restrict                nlist,
 935                      rvec * gmx_restrict                    xx,
 936                      rvec * gmx_restrict                    ff,
 937                      t_forcerec * gmx_restrict              fr,
 938                      t_mdatoms * gmx_restrict               mdatoms,
 939                      nb_kernel_data_t * gmx_restrict        kernel_data,
 940                      t_nrnb * gmx_restrict                  nrnb)
 941 {
 942     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 943      * just 0 for non-waters.
 944      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 945      * jnr indices corresponding to data put in the four positions in the SIMD register.
 946      */
 947     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 948     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 949     int              jnrA,jnrB,jnrC,jnrD;
 950     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 951     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 952     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 953     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 954     real             rcutoff_scalar;
 955     real             *shiftvec,*fshift,*x,*f;
 956     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 957     real             scratch[4*DIM];
 958     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 959     real *           vdwioffsetptr0;
 960     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 961     real *           vdwioffsetptr1;
 962     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 963     real *           vdwioffsetptr2;
 964     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 965     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 966     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 967     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
 968     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 969     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
 970     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 971     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 972     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 973     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 974     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 975     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 976     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 977     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 978     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 979     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 980     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 981     real             *charge;
 982     __m256d          dummy_mask,cutoff_mask;
 983     __m128           tmpmask0,tmpmask1;
 984     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 985     __m256d          one     = _mm256_set1_pd(1.0);
 986     __m256d          two     = _mm256_set1_pd(2.0);
 987     x                = xx[0];
 988     f                = ff[0];
 989
 990     nri              = nlist->nri;
 991     iinr             = nlist->iinr;
 992     jindex           = nlist->jindex;
 993     jjnr             = nlist->jjnr;
 994     shiftidx         = nlist->shift;
 995     gid              = nlist->gid;
 996     shiftvec         = fr->shift_vec[0];
 997     fshift           = fr->fshift[0];
 998     facel            = _mm256_set1_pd(fr->epsfac);
 999     charge           = mdatoms->chargeA;
1000
1001     /* Setup water-specific parameters */
1002     inr              = nlist->iinr[0];
1003     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1004     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1005     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1006
1007     jq0              = _mm256_set1_pd(charge[inr+0]);
1008     jq1              = _mm256_set1_pd(charge[inr+1]);
1009     jq2              = _mm256_set1_pd(charge[inr+2]);
1010     qq00             = _mm256_mul_pd(iq0,jq0);
1011     qq01             = _mm256_mul_pd(iq0,jq1);
1012     qq02             = _mm256_mul_pd(iq0,jq2);
1013     qq10             = _mm256_mul_pd(iq1,jq0);
1014     qq11             = _mm256_mul_pd(iq1,jq1);
1015     qq12             = _mm256_mul_pd(iq1,jq2);
1016     qq20             = _mm256_mul_pd(iq2,jq0);
1017     qq21             = _mm256_mul_pd(iq2,jq1);
1018     qq22             = _mm256_mul_pd(iq2,jq2);
1019
1020     /* Avoid stupid compiler warnings */
1021     jnrA = jnrB = jnrC = jnrD = 0;
1022     j_coord_offsetA = 0;
1023     j_coord_offsetB = 0;
1024     j_coord_offsetC = 0;
1025     j_coord_offsetD = 0;
1026
1027     outeriter        = 0;
1028     inneriter        = 0;
1029
1030     for(iidx=0;iidx<4*DIM;iidx++)
1031     {
1032         scratch[iidx] = 0.0;
1033     }
1034
1035     /* Start outer loop over neighborlists */
1036     for(iidx=0; iidx<nri; iidx++)
1037     {
1038         /* Load shift vector for this list */
1039         i_shift_offset   = DIM*shiftidx[iidx];
1040
1041         /* Load limits for loop over neighbors */
1042         j_index_start    = jindex[iidx];
1043         j_index_end      = jindex[iidx+1];
1044
1045         /* Get outer coordinate index */
1046         inr              = iinr[iidx];
1047         i_coord_offset   = DIM*inr;
1048
1049         /* Load i particle coords and add shift vector */
1050         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1051                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1052
1053         fix0             = _mm256_setzero_pd();
1054         fiy0             = _mm256_setzero_pd();
1055         fiz0             = _mm256_setzero_pd();
1056         fix1             = _mm256_setzero_pd();
1057         fiy1             = _mm256_setzero_pd();
1058         fiz1             = _mm256_setzero_pd();
1059         fix2             = _mm256_setzero_pd();
1060         fiy2             = _mm256_setzero_pd();
1061         fiz2             = _mm256_setzero_pd();
1062
1063         /* Start inner kernel loop */
1064         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1065         {
1066
1067             /* Get j neighbor index, and coordinate index */
1068             jnrA             = jjnr[jidx];
1069             jnrB             = jjnr[jidx+1];
1070             jnrC             = jjnr[jidx+2];
1071             jnrD             = jjnr[jidx+3];
1072             j_coord_offsetA  = DIM*jnrA;
1073             j_coord_offsetB  = DIM*jnrB;
1074             j_coord_offsetC  = DIM*jnrC;
1075             j_coord_offsetD  = DIM*jnrD;
1076
1077             /* load j atom coordinates */
1078             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1079                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1080                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1081
1082             /* Calculate displacement vector */
1083             dx00             = _mm256_sub_pd(ix0,jx0);
1084             dy00             = _mm256_sub_pd(iy0,jy0);
1085             dz00             = _mm256_sub_pd(iz0,jz0);
1086             dx01             = _mm256_sub_pd(ix0,jx1);
1087             dy01             = _mm256_sub_pd(iy0,jy1);
1088             dz01             = _mm256_sub_pd(iz0,jz1);
1089             dx02             = _mm256_sub_pd(ix0,jx2);
1090             dy02             = _mm256_sub_pd(iy0,jy2);
1091             dz02             = _mm256_sub_pd(iz0,jz2);
1092             dx10             = _mm256_sub_pd(ix1,jx0);
1093             dy10             = _mm256_sub_pd(iy1,jy0);
1094             dz10             = _mm256_sub_pd(iz1,jz0);
1095             dx11             = _mm256_sub_pd(ix1,jx1);
1096             dy11             = _mm256_sub_pd(iy1,jy1);
1097             dz11             = _mm256_sub_pd(iz1,jz1);
1098             dx12             = _mm256_sub_pd(ix1,jx2);
1099             dy12             = _mm256_sub_pd(iy1,jy2);
1100             dz12             = _mm256_sub_pd(iz1,jz2);
1101             dx20             = _mm256_sub_pd(ix2,jx0);
1102             dy20             = _mm256_sub_pd(iy2,jy0);
1103             dz20             = _mm256_sub_pd(iz2,jz0);
1104             dx21             = _mm256_sub_pd(ix2,jx1);
1105             dy21             = _mm256_sub_pd(iy2,jy1);
1106             dz21             = _mm256_sub_pd(iz2,jz1);
1107             dx22             = _mm256_sub_pd(ix2,jx2);
1108             dy22             = _mm256_sub_pd(iy2,jy2);
1109             dz22             = _mm256_sub_pd(iz2,jz2);
1110
1111             /* Calculate squared distance and things based on it */
1112             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1113             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1114             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1115             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1116             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1117             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1118             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1119             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1120             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1121
1122             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1123             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1124             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1125             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1126             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1127             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1128             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1129             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1130             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1131
1132             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1133             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1134             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1135             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1136             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1137             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1138             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1139             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1140             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1141
1142             fjx0             = _mm256_setzero_pd();
1143             fjy0             = _mm256_setzero_pd();
1144             fjz0             = _mm256_setzero_pd();
1145             fjx1             = _mm256_setzero_pd();
1146             fjy1             = _mm256_setzero_pd();
1147             fjz1             = _mm256_setzero_pd();
1148             fjx2             = _mm256_setzero_pd();
1149             fjy2             = _mm256_setzero_pd();
1150             fjz2             = _mm256_setzero_pd();
1151
1152             /**************************
1153              * CALCULATE INTERACTIONS *
1154              **************************/
1155
1156             /* COULOMB ELECTROSTATICS */
1157             velec            = _mm256_mul_pd(qq00,rinv00);
1158             felec            = _mm256_mul_pd(velec,rinvsq00);
1159
1160             fscal            = felec;
1161
1162             /* Calculate temporary vectorial force */
1163             tx               = _mm256_mul_pd(fscal,dx00);
1164             ty               = _mm256_mul_pd(fscal,dy00);
1165             tz               = _mm256_mul_pd(fscal,dz00);
1166
1167             /* Update vectorial force */
1168             fix0             = _mm256_add_pd(fix0,tx);
1169             fiy0             = _mm256_add_pd(fiy0,ty);
1170             fiz0             = _mm256_add_pd(fiz0,tz);
1171
1172             fjx0             = _mm256_add_pd(fjx0,tx);
1173             fjy0             = _mm256_add_pd(fjy0,ty);
1174             fjz0             = _mm256_add_pd(fjz0,tz);
1175
1176             /**************************
1177              * CALCULATE INTERACTIONS *
1178              **************************/
1179
1180             /* COULOMB ELECTROSTATICS */
1181             velec            = _mm256_mul_pd(qq01,rinv01);
1182             felec            = _mm256_mul_pd(velec,rinvsq01);
1183
1184             fscal            = felec;
1185
1186             /* Calculate temporary vectorial force */
1187             tx               = _mm256_mul_pd(fscal,dx01);
1188             ty               = _mm256_mul_pd(fscal,dy01);
1189             tz               = _mm256_mul_pd(fscal,dz01);
1190
1191             /* Update vectorial force */
1192             fix0             = _mm256_add_pd(fix0,tx);
1193             fiy0             = _mm256_add_pd(fiy0,ty);
1194             fiz0             = _mm256_add_pd(fiz0,tz);
1195
1196             fjx1             = _mm256_add_pd(fjx1,tx);
1197             fjy1             = _mm256_add_pd(fjy1,ty);
1198             fjz1             = _mm256_add_pd(fjz1,tz);
1199
1200             /**************************
1201              * CALCULATE INTERACTIONS *
1202              **************************/
1203
1204             /* COULOMB ELECTROSTATICS */
1205             velec            = _mm256_mul_pd(qq02,rinv02);
1206             felec            = _mm256_mul_pd(velec,rinvsq02);
1207
1208             fscal            = felec;
1209
1210             /* Calculate temporary vectorial force */
1211             tx               = _mm256_mul_pd(fscal,dx02);
1212             ty               = _mm256_mul_pd(fscal,dy02);
1213             tz               = _mm256_mul_pd(fscal,dz02);
1214
1215             /* Update vectorial force */
1216             fix0             = _mm256_add_pd(fix0,tx);
1217             fiy0             = _mm256_add_pd(fiy0,ty);
1218             fiz0             = _mm256_add_pd(fiz0,tz);
1219
1220             fjx2             = _mm256_add_pd(fjx2,tx);
1221             fjy2             = _mm256_add_pd(fjy2,ty);
1222             fjz2             = _mm256_add_pd(fjz2,tz);
1223
1224             /**************************
1225              * CALCULATE INTERACTIONS *
1226              **************************/
1227
1228             /* COULOMB ELECTROSTATICS */
1229             velec            = _mm256_mul_pd(qq10,rinv10);
1230             felec            = _mm256_mul_pd(velec,rinvsq10);
1231
1232             fscal            = felec;
1233
1234             /* Calculate temporary vectorial force */
1235             tx               = _mm256_mul_pd(fscal,dx10);
1236             ty               = _mm256_mul_pd(fscal,dy10);
1237             tz               = _mm256_mul_pd(fscal,dz10);
1238
1239             /* Update vectorial force */
1240             fix1             = _mm256_add_pd(fix1,tx);
1241             fiy1             = _mm256_add_pd(fiy1,ty);
1242             fiz1             = _mm256_add_pd(fiz1,tz);
1243
1244             fjx0             = _mm256_add_pd(fjx0,tx);
1245             fjy0             = _mm256_add_pd(fjy0,ty);
1246             fjz0             = _mm256_add_pd(fjz0,tz);
1247
1248             /**************************
1249              * CALCULATE INTERACTIONS *
1250              **************************/
1251
1252             /* COULOMB ELECTROSTATICS */
1253             velec            = _mm256_mul_pd(qq11,rinv11);
1254             felec            = _mm256_mul_pd(velec,rinvsq11);
1255
1256             fscal            = felec;
1257
1258             /* Calculate temporary vectorial force */
1259             tx               = _mm256_mul_pd(fscal,dx11);
1260             ty               = _mm256_mul_pd(fscal,dy11);
1261             tz               = _mm256_mul_pd(fscal,dz11);
1262
1263             /* Update vectorial force */
1264             fix1             = _mm256_add_pd(fix1,tx);
1265             fiy1             = _mm256_add_pd(fiy1,ty);
1266             fiz1             = _mm256_add_pd(fiz1,tz);
1267
1268             fjx1             = _mm256_add_pd(fjx1,tx);
1269             fjy1             = _mm256_add_pd(fjy1,ty);
1270             fjz1             = _mm256_add_pd(fjz1,tz);
1271
1272             /**************************
1273              * CALCULATE INTERACTIONS *
1274              **************************/
1275
1276             /* COULOMB ELECTROSTATICS */
1277             velec            = _mm256_mul_pd(qq12,rinv12);
1278             felec            = _mm256_mul_pd(velec,rinvsq12);
1279
1280             fscal            = felec;
1281
1282             /* Calculate temporary vectorial force */
1283             tx               = _mm256_mul_pd(fscal,dx12);
1284             ty               = _mm256_mul_pd(fscal,dy12);
1285             tz               = _mm256_mul_pd(fscal,dz12);
1286
1287             /* Update vectorial force */
1288             fix1             = _mm256_add_pd(fix1,tx);
1289             fiy1             = _mm256_add_pd(fiy1,ty);
1290             fiz1             = _mm256_add_pd(fiz1,tz);
1291
1292             fjx2             = _mm256_add_pd(fjx2,tx);
1293             fjy2             = _mm256_add_pd(fjy2,ty);
1294             fjz2             = _mm256_add_pd(fjz2,tz);
1295
1296             /**************************
1297              * CALCULATE INTERACTIONS *
1298              **************************/
1299
1300             /* COULOMB ELECTROSTATICS */
1301             velec            = _mm256_mul_pd(qq20,rinv20);
1302             felec            = _mm256_mul_pd(velec,rinvsq20);
1303
1304             fscal            = felec;
1305
1306             /* Calculate temporary vectorial force */
1307             tx               = _mm256_mul_pd(fscal,dx20);
1308             ty               = _mm256_mul_pd(fscal,dy20);
1309             tz               = _mm256_mul_pd(fscal,dz20);
1310
1311             /* Update vectorial force */
1312             fix2             = _mm256_add_pd(fix2,tx);
1313             fiy2             = _mm256_add_pd(fiy2,ty);
1314             fiz2             = _mm256_add_pd(fiz2,tz);
1315
1316             fjx0             = _mm256_add_pd(fjx0,tx);
1317             fjy0             = _mm256_add_pd(fjy0,ty);
1318             fjz0             = _mm256_add_pd(fjz0,tz);
1319
1320             /**************************
1321              * CALCULATE INTERACTIONS *
1322              **************************/
1323
1324             /* COULOMB ELECTROSTATICS */
1325             velec            = _mm256_mul_pd(qq21,rinv21);
1326             felec            = _mm256_mul_pd(velec,rinvsq21);
1327
1328             fscal            = felec;
1329
1330             /* Calculate temporary vectorial force */
1331             tx               = _mm256_mul_pd(fscal,dx21);
1332             ty               = _mm256_mul_pd(fscal,dy21);
1333             tz               = _mm256_mul_pd(fscal,dz21);
1334
1335             /* Update vectorial force */
1336             fix2             = _mm256_add_pd(fix2,tx);
1337             fiy2             = _mm256_add_pd(fiy2,ty);
1338             fiz2             = _mm256_add_pd(fiz2,tz);
1339
1340             fjx1             = _mm256_add_pd(fjx1,tx);
1341             fjy1             = _mm256_add_pd(fjy1,ty);
1342             fjz1             = _mm256_add_pd(fjz1,tz);
1343
1344             /**************************
1345              * CALCULATE INTERACTIONS *
1346              **************************/
1347
1348             /* COULOMB ELECTROSTATICS */
1349             velec            = _mm256_mul_pd(qq22,rinv22);
1350             felec            = _mm256_mul_pd(velec,rinvsq22);
1351
1352             fscal            = felec;
1353
1354             /* Calculate temporary vectorial force */
1355             tx               = _mm256_mul_pd(fscal,dx22);
1356             ty               = _mm256_mul_pd(fscal,dy22);
1357             tz               = _mm256_mul_pd(fscal,dz22);
1358
1359             /* Update vectorial force */
1360             fix2             = _mm256_add_pd(fix2,tx);
1361             fiy2             = _mm256_add_pd(fiy2,ty);
1362             fiz2             = _mm256_add_pd(fiz2,tz);
1363
1364             fjx2             = _mm256_add_pd(fjx2,tx);
1365             fjy2             = _mm256_add_pd(fjy2,ty);
1366             fjz2             = _mm256_add_pd(fjz2,tz);
1367
1368             fjptrA             = f+j_coord_offsetA;
1369             fjptrB             = f+j_coord_offsetB;
1370             fjptrC             = f+j_coord_offsetC;
1371             fjptrD             = f+j_coord_offsetD;
1372
1373             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1374                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1375
1376             /* Inner loop uses 234 flops */
1377         }
1378
1379         if(jidx<j_index_end)
1380         {
1381
1382             /* Get j neighbor index, and coordinate index */
1383             jnrlistA         = jjnr[jidx];
1384             jnrlistB         = jjnr[jidx+1];
1385             jnrlistC         = jjnr[jidx+2];
1386             jnrlistD         = jjnr[jidx+3];
1387             /* Sign of each element will be negative for non-real atoms.
1388              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1389              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1390              */
1391             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1392
1393             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1394             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1395             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1396
1397             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1398             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1399             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1400             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1401             j_coord_offsetA  = DIM*jnrA;
1402             j_coord_offsetB  = DIM*jnrB;
1403             j_coord_offsetC  = DIM*jnrC;
1404             j_coord_offsetD  = DIM*jnrD;
1405
1406             /* load j atom coordinates */
1407             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1408                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1409                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1410
1411             /* Calculate displacement vector */
1412             dx00             = _mm256_sub_pd(ix0,jx0);
1413             dy00             = _mm256_sub_pd(iy0,jy0);
1414             dz00             = _mm256_sub_pd(iz0,jz0);
1415             dx01             = _mm256_sub_pd(ix0,jx1);
1416             dy01             = _mm256_sub_pd(iy0,jy1);
1417             dz01             = _mm256_sub_pd(iz0,jz1);
1418             dx02             = _mm256_sub_pd(ix0,jx2);
1419             dy02             = _mm256_sub_pd(iy0,jy2);
1420             dz02             = _mm256_sub_pd(iz0,jz2);
1421             dx10             = _mm256_sub_pd(ix1,jx0);
1422             dy10             = _mm256_sub_pd(iy1,jy0);
1423             dz10             = _mm256_sub_pd(iz1,jz0);
1424             dx11             = _mm256_sub_pd(ix1,jx1);
1425             dy11             = _mm256_sub_pd(iy1,jy1);
1426             dz11             = _mm256_sub_pd(iz1,jz1);
1427             dx12             = _mm256_sub_pd(ix1,jx2);
1428             dy12             = _mm256_sub_pd(iy1,jy2);
1429             dz12             = _mm256_sub_pd(iz1,jz2);
1430             dx20             = _mm256_sub_pd(ix2,jx0);
1431             dy20             = _mm256_sub_pd(iy2,jy0);
1432             dz20             = _mm256_sub_pd(iz2,jz0);
1433             dx21             = _mm256_sub_pd(ix2,jx1);
1434             dy21             = _mm256_sub_pd(iy2,jy1);
1435             dz21             = _mm256_sub_pd(iz2,jz1);
1436             dx22             = _mm256_sub_pd(ix2,jx2);
1437             dy22             = _mm256_sub_pd(iy2,jy2);
1438             dz22             = _mm256_sub_pd(iz2,jz2);
1439
1440             /* Calculate squared distance and things based on it */
1441             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1442             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1443             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1444             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1445             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1446             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1447             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1448             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1449             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1450
1451             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1452             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1453             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1454             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1455             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1456             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1457             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1458             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1459             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1460
1461             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1462             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1463             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1464             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1465             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1466             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1467             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1468             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1469             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1470
1471             fjx0             = _mm256_setzero_pd();
1472             fjy0             = _mm256_setzero_pd();
1473             fjz0             = _mm256_setzero_pd();
1474             fjx1             = _mm256_setzero_pd();
1475             fjy1             = _mm256_setzero_pd();
1476             fjz1             = _mm256_setzero_pd();
1477             fjx2             = _mm256_setzero_pd();
1478             fjy2             = _mm256_setzero_pd();
1479             fjz2             = _mm256_setzero_pd();
1480
1481             /**************************
1482              * CALCULATE INTERACTIONS *
1483              **************************/
1484
1485             /* COULOMB ELECTROSTATICS */
1486             velec            = _mm256_mul_pd(qq00,rinv00);
1487             felec            = _mm256_mul_pd(velec,rinvsq00);
1488
1489             fscal            = felec;
1490
1491             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1492
1493             /* Calculate temporary vectorial force */
1494             tx               = _mm256_mul_pd(fscal,dx00);
1495             ty               = _mm256_mul_pd(fscal,dy00);
1496             tz               = _mm256_mul_pd(fscal,dz00);
1497
1498             /* Update vectorial force */
1499             fix0             = _mm256_add_pd(fix0,tx);
1500             fiy0             = _mm256_add_pd(fiy0,ty);
1501             fiz0             = _mm256_add_pd(fiz0,tz);
1502
1503             fjx0             = _mm256_add_pd(fjx0,tx);
1504             fjy0             = _mm256_add_pd(fjy0,ty);
1505             fjz0             = _mm256_add_pd(fjz0,tz);
1506
1507             /**************************
1508              * CALCULATE INTERACTIONS *
1509              **************************/
1510
1511             /* COULOMB ELECTROSTATICS */
1512             velec            = _mm256_mul_pd(qq01,rinv01);
1513             felec            = _mm256_mul_pd(velec,rinvsq01);
1514
1515             fscal            = felec;
1516
1517             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1518
1519             /* Calculate temporary vectorial force */
1520             tx               = _mm256_mul_pd(fscal,dx01);
1521             ty               = _mm256_mul_pd(fscal,dy01);
1522             tz               = _mm256_mul_pd(fscal,dz01);
1523
1524             /* Update vectorial force */
1525             fix0             = _mm256_add_pd(fix0,tx);
1526             fiy0             = _mm256_add_pd(fiy0,ty);
1527             fiz0             = _mm256_add_pd(fiz0,tz);
1528
1529             fjx1             = _mm256_add_pd(fjx1,tx);
1530             fjy1             = _mm256_add_pd(fjy1,ty);
1531             fjz1             = _mm256_add_pd(fjz1,tz);
1532
1533             /**************************
1534              * CALCULATE INTERACTIONS *
1535              **************************/
1536
1537             /* COULOMB ELECTROSTATICS */
1538             velec            = _mm256_mul_pd(qq02,rinv02);
1539             felec            = _mm256_mul_pd(velec,rinvsq02);
1540
1541             fscal            = felec;
1542
1543             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1544
1545             /* Calculate temporary vectorial force */
1546             tx               = _mm256_mul_pd(fscal,dx02);
1547             ty               = _mm256_mul_pd(fscal,dy02);
1548             tz               = _mm256_mul_pd(fscal,dz02);
1549
1550             /* Update vectorial force */
1551             fix0             = _mm256_add_pd(fix0,tx);
1552             fiy0             = _mm256_add_pd(fiy0,ty);
1553             fiz0             = _mm256_add_pd(fiz0,tz);
1554
1555             fjx2             = _mm256_add_pd(fjx2,tx);
1556             fjy2             = _mm256_add_pd(fjy2,ty);
1557             fjz2             = _mm256_add_pd(fjz2,tz);
1558
1559             /**************************
1560              * CALCULATE INTERACTIONS *
1561              **************************/
1562
1563             /* COULOMB ELECTROSTATICS */
1564             velec            = _mm256_mul_pd(qq10,rinv10);
1565             felec            = _mm256_mul_pd(velec,rinvsq10);
1566
1567             fscal            = felec;
1568
1569             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1570
1571             /* Calculate temporary vectorial force */
1572             tx               = _mm256_mul_pd(fscal,dx10);
1573             ty               = _mm256_mul_pd(fscal,dy10);
1574             tz               = _mm256_mul_pd(fscal,dz10);
1575
1576             /* Update vectorial force */
1577             fix1             = _mm256_add_pd(fix1,tx);
1578             fiy1             = _mm256_add_pd(fiy1,ty);
1579             fiz1             = _mm256_add_pd(fiz1,tz);
1580
1581             fjx0             = _mm256_add_pd(fjx0,tx);
1582             fjy0             = _mm256_add_pd(fjy0,ty);
1583             fjz0             = _mm256_add_pd(fjz0,tz);
1584
1585             /**************************
1586              * CALCULATE INTERACTIONS *
1587              **************************/
1588
1589             /* COULOMB ELECTROSTATICS */
1590             velec            = _mm256_mul_pd(qq11,rinv11);
1591             felec            = _mm256_mul_pd(velec,rinvsq11);
1592
1593             fscal            = felec;
1594
1595             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1596
1597             /* Calculate temporary vectorial force */
1598             tx               = _mm256_mul_pd(fscal,dx11);
1599             ty               = _mm256_mul_pd(fscal,dy11);
1600             tz               = _mm256_mul_pd(fscal,dz11);
1601
1602             /* Update vectorial force */
1603             fix1             = _mm256_add_pd(fix1,tx);
1604             fiy1             = _mm256_add_pd(fiy1,ty);
1605             fiz1             = _mm256_add_pd(fiz1,tz);
1606
1607             fjx1             = _mm256_add_pd(fjx1,tx);
1608             fjy1             = _mm256_add_pd(fjy1,ty);
1609             fjz1             = _mm256_add_pd(fjz1,tz);
1610
1611             /**************************
1612              * CALCULATE INTERACTIONS *
1613              **************************/
1614
1615             /* COULOMB ELECTROSTATICS */
1616             velec            = _mm256_mul_pd(qq12,rinv12);
1617             felec            = _mm256_mul_pd(velec,rinvsq12);
1618
1619             fscal            = felec;
1620
1621             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1622
1623             /* Calculate temporary vectorial force */
1624             tx               = _mm256_mul_pd(fscal,dx12);
1625             ty               = _mm256_mul_pd(fscal,dy12);
1626             tz               = _mm256_mul_pd(fscal,dz12);
1627
1628             /* Update vectorial force */
1629             fix1             = _mm256_add_pd(fix1,tx);
1630             fiy1             = _mm256_add_pd(fiy1,ty);
1631             fiz1             = _mm256_add_pd(fiz1,tz);
1632
1633             fjx2             = _mm256_add_pd(fjx2,tx);
1634             fjy2             = _mm256_add_pd(fjy2,ty);
1635             fjz2             = _mm256_add_pd(fjz2,tz);
1636
1637             /**************************
1638              * CALCULATE INTERACTIONS *
1639              **************************/
1640
1641             /* COULOMB ELECTROSTATICS */
1642             velec            = _mm256_mul_pd(qq20,rinv20);
1643             felec            = _mm256_mul_pd(velec,rinvsq20);
1644
1645             fscal            = felec;
1646
1647             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1648
1649             /* Calculate temporary vectorial force */
1650             tx               = _mm256_mul_pd(fscal,dx20);
1651             ty               = _mm256_mul_pd(fscal,dy20);
1652             tz               = _mm256_mul_pd(fscal,dz20);
1653
1654             /* Update vectorial force */
1655             fix2             = _mm256_add_pd(fix2,tx);
1656             fiy2             = _mm256_add_pd(fiy2,ty);
1657             fiz2             = _mm256_add_pd(fiz2,tz);
1658
1659             fjx0             = _mm256_add_pd(fjx0,tx);
1660             fjy0             = _mm256_add_pd(fjy0,ty);
1661             fjz0             = _mm256_add_pd(fjz0,tz);
1662
1663             /**************************
1664              * CALCULATE INTERACTIONS *
1665              **************************/
1666
1667             /* COULOMB ELECTROSTATICS */
1668             velec            = _mm256_mul_pd(qq21,rinv21);
1669             felec            = _mm256_mul_pd(velec,rinvsq21);
1670
1671             fscal            = felec;
1672
1673             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1674
1675             /* Calculate temporary vectorial force */
1676             tx               = _mm256_mul_pd(fscal,dx21);
1677             ty               = _mm256_mul_pd(fscal,dy21);
1678             tz               = _mm256_mul_pd(fscal,dz21);
1679
1680             /* Update vectorial force */
1681             fix2             = _mm256_add_pd(fix2,tx);
1682             fiy2             = _mm256_add_pd(fiy2,ty);
1683             fiz2             = _mm256_add_pd(fiz2,tz);
1684
1685             fjx1             = _mm256_add_pd(fjx1,tx);
1686             fjy1             = _mm256_add_pd(fjy1,ty);
1687             fjz1             = _mm256_add_pd(fjz1,tz);
1688
1689             /**************************
1690              * CALCULATE INTERACTIONS *
1691              **************************/
1692
1693             /* COULOMB ELECTROSTATICS */
1694             velec            = _mm256_mul_pd(qq22,rinv22);
1695             felec            = _mm256_mul_pd(velec,rinvsq22);
1696
1697             fscal            = felec;
1698
1699             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1700
1701             /* Calculate temporary vectorial force */
1702             tx               = _mm256_mul_pd(fscal,dx22);
1703             ty               = _mm256_mul_pd(fscal,dy22);
1704             tz               = _mm256_mul_pd(fscal,dz22);
1705
1706             /* Update vectorial force */
1707             fix2             = _mm256_add_pd(fix2,tx);
1708             fiy2             = _mm256_add_pd(fiy2,ty);
1709             fiz2             = _mm256_add_pd(fiz2,tz);
1710
1711             fjx2             = _mm256_add_pd(fjx2,tx);
1712             fjy2             = _mm256_add_pd(fjy2,ty);
1713             fjz2             = _mm256_add_pd(fjz2,tz);
1714
1715             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1716             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1717             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1718             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1719
1720             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1721                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1722
1723             /* Inner loop uses 234 flops */
1724         }
1725
1726         /* End of innermost loop */
1727
1728         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1729                                                  f+i_coord_offset,fshift+i_shift_offset);
1730
1731         /* Increment number of inner iterations */
1732         inneriter                  += j_index_end - j_index_start;
1733
1734         /* Outer loop uses 18 flops */
1735     }
1736
1737     /* Increment number of outer iterations */
1738     outeriter        += nri;
1739
1740     /* Update outer/inner flops */
1741
1742     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*234);
1743 }