src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwNone_GeomW4W4_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW4W4_VF_avx_256_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwNone_GeomW4W4_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  93     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  94     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  95     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  96     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  97     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  98     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  99     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 100     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 101     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 102     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 103     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 104     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 105     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 106     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 107     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     __m256d          dummy_mask,cutoff_mask;
 110     __m128           tmpmask0,tmpmask1;
 111     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 112     __m256d          one     = _mm256_set1_pd(1.0);
 113     __m256d          two     = _mm256_set1_pd(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_pd(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 131     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 132     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 133
 134     jq1              = _mm256_set1_pd(charge[inr+1]);
 135     jq2              = _mm256_set1_pd(charge[inr+2]);
 136     jq3              = _mm256_set1_pd(charge[inr+3]);
 137     qq11             = _mm256_mul_pd(iq1,jq1);
 138     qq12             = _mm256_mul_pd(iq1,jq2);
 139     qq13             = _mm256_mul_pd(iq1,jq3);
 140     qq21             = _mm256_mul_pd(iq2,jq1);
 141     qq22             = _mm256_mul_pd(iq2,jq2);
 142     qq23             = _mm256_mul_pd(iq2,jq3);
 143     qq31             = _mm256_mul_pd(iq3,jq1);
 144     qq32             = _mm256_mul_pd(iq3,jq2);
 145     qq33             = _mm256_mul_pd(iq3,jq3);
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = jnrC = jnrD = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151     j_coord_offsetC = 0;
 152     j_coord_offsetD = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 178                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 179
 180         fix1             = _mm256_setzero_pd();
 181         fiy1             = _mm256_setzero_pd();
 182         fiz1             = _mm256_setzero_pd();
 183         fix2             = _mm256_setzero_pd();
 184         fiy2             = _mm256_setzero_pd();
 185         fiz2             = _mm256_setzero_pd();
 186         fix3             = _mm256_setzero_pd();
 187         fiy3             = _mm256_setzero_pd();
 188         fiz3             = _mm256_setzero_pd();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm256_setzero_pd();
 192
 193         /* Start inner kernel loop */
 194         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 195         {
 196
 197             /* Get j neighbor index, and coordinate index */
 198             jnrA             = jjnr[jidx];
 199             jnrB             = jjnr[jidx+1];
 200             jnrC             = jjnr[jidx+2];
 201             jnrD             = jjnr[jidx+3];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204             j_coord_offsetC  = DIM*jnrC;
 205             j_coord_offsetD  = DIM*jnrD;
 206
 207             /* load j atom coordinates */
 208             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 209                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 210                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 211
 212             /* Calculate displacement vector */
 213             dx11             = _mm256_sub_pd(ix1,jx1);
 214             dy11             = _mm256_sub_pd(iy1,jy1);
 215             dz11             = _mm256_sub_pd(iz1,jz1);
 216             dx12             = _mm256_sub_pd(ix1,jx2);
 217             dy12             = _mm256_sub_pd(iy1,jy2);
 218             dz12             = _mm256_sub_pd(iz1,jz2);
 219             dx13             = _mm256_sub_pd(ix1,jx3);
 220             dy13             = _mm256_sub_pd(iy1,jy3);
 221             dz13             = _mm256_sub_pd(iz1,jz3);
 222             dx21             = _mm256_sub_pd(ix2,jx1);
 223             dy21             = _mm256_sub_pd(iy2,jy1);
 224             dz21             = _mm256_sub_pd(iz2,jz1);
 225             dx22             = _mm256_sub_pd(ix2,jx2);
 226             dy22             = _mm256_sub_pd(iy2,jy2);
 227             dz22             = _mm256_sub_pd(iz2,jz2);
 228             dx23             = _mm256_sub_pd(ix2,jx3);
 229             dy23             = _mm256_sub_pd(iy2,jy3);
 230             dz23             = _mm256_sub_pd(iz2,jz3);
 231             dx31             = _mm256_sub_pd(ix3,jx1);
 232             dy31             = _mm256_sub_pd(iy3,jy1);
 233             dz31             = _mm256_sub_pd(iz3,jz1);
 234             dx32             = _mm256_sub_pd(ix3,jx2);
 235             dy32             = _mm256_sub_pd(iy3,jy2);
 236             dz32             = _mm256_sub_pd(iz3,jz2);
 237             dx33             = _mm256_sub_pd(ix3,jx3);
 238             dy33             = _mm256_sub_pd(iy3,jy3);
 239             dz33             = _mm256_sub_pd(iz3,jz3);
 240
 241             /* Calculate squared distance and things based on it */
 242             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 243             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 244             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 245             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 246             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 247             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 248             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 249             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 250             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 251
 252             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 253             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 254             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 255             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 256             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 257             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 258             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 259             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 260             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 261
 262             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 263             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 264             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 265             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 266             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 267             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 268             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 269             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 270             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 271
 272             fjx1             = _mm256_setzero_pd();
 273             fjy1             = _mm256_setzero_pd();
 274             fjz1             = _mm256_setzero_pd();
 275             fjx2             = _mm256_setzero_pd();
 276             fjy2             = _mm256_setzero_pd();
 277             fjz2             = _mm256_setzero_pd();
 278             fjx3             = _mm256_setzero_pd();
 279             fjy3             = _mm256_setzero_pd();
 280             fjz3             = _mm256_setzero_pd();
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             /* COULOMB ELECTROSTATICS */
 287             velec            = _mm256_mul_pd(qq11,rinv11);
 288             felec            = _mm256_mul_pd(velec,rinvsq11);
 289
 290             /* Update potential sum for this i atom from the interaction with this j atom. */
 291             velecsum         = _mm256_add_pd(velecsum,velec);
 292
 293             fscal            = felec;
 294
 295             /* Calculate temporary vectorial force */
 296             tx               = _mm256_mul_pd(fscal,dx11);
 297             ty               = _mm256_mul_pd(fscal,dy11);
 298             tz               = _mm256_mul_pd(fscal,dz11);
 299
 300             /* Update vectorial force */
 301             fix1             = _mm256_add_pd(fix1,tx);
 302             fiy1             = _mm256_add_pd(fiy1,ty);
 303             fiz1             = _mm256_add_pd(fiz1,tz);
 304
 305             fjx1             = _mm256_add_pd(fjx1,tx);
 306             fjy1             = _mm256_add_pd(fjy1,ty);
 307             fjz1             = _mm256_add_pd(fjz1,tz);
 308
 309             /**************************
 310              * CALCULATE INTERACTIONS *
 311              **************************/
 312
 313             /* COULOMB ELECTROSTATICS */
 314             velec            = _mm256_mul_pd(qq12,rinv12);
 315             felec            = _mm256_mul_pd(velec,rinvsq12);
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velecsum         = _mm256_add_pd(velecsum,velec);
 319
 320             fscal            = felec;
 321
 322             /* Calculate temporary vectorial force */
 323             tx               = _mm256_mul_pd(fscal,dx12);
 324             ty               = _mm256_mul_pd(fscal,dy12);
 325             tz               = _mm256_mul_pd(fscal,dz12);
 326
 327             /* Update vectorial force */
 328             fix1             = _mm256_add_pd(fix1,tx);
 329             fiy1             = _mm256_add_pd(fiy1,ty);
 330             fiz1             = _mm256_add_pd(fiz1,tz);
 331
 332             fjx2             = _mm256_add_pd(fjx2,tx);
 333             fjy2             = _mm256_add_pd(fjy2,ty);
 334             fjz2             = _mm256_add_pd(fjz2,tz);
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             /* COULOMB ELECTROSTATICS */
 341             velec            = _mm256_mul_pd(qq13,rinv13);
 342             felec            = _mm256_mul_pd(velec,rinvsq13);
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm256_add_pd(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm256_mul_pd(fscal,dx13);
 351             ty               = _mm256_mul_pd(fscal,dy13);
 352             tz               = _mm256_mul_pd(fscal,dz13);
 353
 354             /* Update vectorial force */
 355             fix1             = _mm256_add_pd(fix1,tx);
 356             fiy1             = _mm256_add_pd(fiy1,ty);
 357             fiz1             = _mm256_add_pd(fiz1,tz);
 358
 359             fjx3             = _mm256_add_pd(fjx3,tx);
 360             fjy3             = _mm256_add_pd(fjy3,ty);
 361             fjz3             = _mm256_add_pd(fjz3,tz);
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             /* COULOMB ELECTROSTATICS */
 368             velec            = _mm256_mul_pd(qq21,rinv21);
 369             felec            = _mm256_mul_pd(velec,rinvsq21);
 370
 371             /* Update potential sum for this i atom from the interaction with this j atom. */
 372             velecsum         = _mm256_add_pd(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm256_mul_pd(fscal,dx21);
 378             ty               = _mm256_mul_pd(fscal,dy21);
 379             tz               = _mm256_mul_pd(fscal,dz21);
 380
 381             /* Update vectorial force */
 382             fix2             = _mm256_add_pd(fix2,tx);
 383             fiy2             = _mm256_add_pd(fiy2,ty);
 384             fiz2             = _mm256_add_pd(fiz2,tz);
 385
 386             fjx1             = _mm256_add_pd(fjx1,tx);
 387             fjy1             = _mm256_add_pd(fjy1,ty);
 388             fjz1             = _mm256_add_pd(fjz1,tz);
 389
 390             /**************************
 391              * CALCULATE INTERACTIONS *
 392              **************************/
 393
 394             /* COULOMB ELECTROSTATICS */
 395             velec            = _mm256_mul_pd(qq22,rinv22);
 396             felec            = _mm256_mul_pd(velec,rinvsq22);
 397
 398             /* Update potential sum for this i atom from the interaction with this j atom. */
 399             velecsum         = _mm256_add_pd(velecsum,velec);
 400
 401             fscal            = felec;
 402
 403             /* Calculate temporary vectorial force */
 404             tx               = _mm256_mul_pd(fscal,dx22);
 405             ty               = _mm256_mul_pd(fscal,dy22);
 406             tz               = _mm256_mul_pd(fscal,dz22);
 407
 408             /* Update vectorial force */
 409             fix2             = _mm256_add_pd(fix2,tx);
 410             fiy2             = _mm256_add_pd(fiy2,ty);
 411             fiz2             = _mm256_add_pd(fiz2,tz);
 412
 413             fjx2             = _mm256_add_pd(fjx2,tx);
 414             fjy2             = _mm256_add_pd(fjy2,ty);
 415             fjz2             = _mm256_add_pd(fjz2,tz);
 416
 417             /**************************
 418              * CALCULATE INTERACTIONS *
 419              **************************/
 420
 421             /* COULOMB ELECTROSTATICS */
 422             velec            = _mm256_mul_pd(qq23,rinv23);
 423             felec            = _mm256_mul_pd(velec,rinvsq23);
 424
 425             /* Update potential sum for this i atom from the interaction with this j atom. */
 426             velecsum         = _mm256_add_pd(velecsum,velec);
 427
 428             fscal            = felec;
 429
 430             /* Calculate temporary vectorial force */
 431             tx               = _mm256_mul_pd(fscal,dx23);
 432             ty               = _mm256_mul_pd(fscal,dy23);
 433             tz               = _mm256_mul_pd(fscal,dz23);
 434
 435             /* Update vectorial force */
 436             fix2             = _mm256_add_pd(fix2,tx);
 437             fiy2             = _mm256_add_pd(fiy2,ty);
 438             fiz2             = _mm256_add_pd(fiz2,tz);
 439
 440             fjx3             = _mm256_add_pd(fjx3,tx);
 441             fjy3             = _mm256_add_pd(fjy3,ty);
 442             fjz3             = _mm256_add_pd(fjz3,tz);
 443
 444             /**************************
 445              * CALCULATE INTERACTIONS *
 446              **************************/
 447
 448             /* COULOMB ELECTROSTATICS */
 449             velec            = _mm256_mul_pd(qq31,rinv31);
 450             felec            = _mm256_mul_pd(velec,rinvsq31);
 451
 452             /* Update potential sum for this i atom from the interaction with this j atom. */
 453             velecsum         = _mm256_add_pd(velecsum,velec);
 454
 455             fscal            = felec;
 456
 457             /* Calculate temporary vectorial force */
 458             tx               = _mm256_mul_pd(fscal,dx31);
 459             ty               = _mm256_mul_pd(fscal,dy31);
 460             tz               = _mm256_mul_pd(fscal,dz31);
 461
 462             /* Update vectorial force */
 463             fix3             = _mm256_add_pd(fix3,tx);
 464             fiy3             = _mm256_add_pd(fiy3,ty);
 465             fiz3             = _mm256_add_pd(fiz3,tz);
 466
 467             fjx1             = _mm256_add_pd(fjx1,tx);
 468             fjy1             = _mm256_add_pd(fjy1,ty);
 469             fjz1             = _mm256_add_pd(fjz1,tz);
 470
 471             /**************************
 472              * CALCULATE INTERACTIONS *
 473              **************************/
 474
 475             /* COULOMB ELECTROSTATICS */
 476             velec            = _mm256_mul_pd(qq32,rinv32);
 477             felec            = _mm256_mul_pd(velec,rinvsq32);
 478
 479             /* Update potential sum for this i atom from the interaction with this j atom. */
 480             velecsum         = _mm256_add_pd(velecsum,velec);
 481
 482             fscal            = felec;
 483
 484             /* Calculate temporary vectorial force */
 485             tx               = _mm256_mul_pd(fscal,dx32);
 486             ty               = _mm256_mul_pd(fscal,dy32);
 487             tz               = _mm256_mul_pd(fscal,dz32);
 488
 489             /* Update vectorial force */
 490             fix3             = _mm256_add_pd(fix3,tx);
 491             fiy3             = _mm256_add_pd(fiy3,ty);
 492             fiz3             = _mm256_add_pd(fiz3,tz);
 493
 494             fjx2             = _mm256_add_pd(fjx2,tx);
 495             fjy2             = _mm256_add_pd(fjy2,ty);
 496             fjz2             = _mm256_add_pd(fjz2,tz);
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             /* COULOMB ELECTROSTATICS */
 503             velec            = _mm256_mul_pd(qq33,rinv33);
 504             felec            = _mm256_mul_pd(velec,rinvsq33);
 505
 506             /* Update potential sum for this i atom from the interaction with this j atom. */
 507             velecsum         = _mm256_add_pd(velecsum,velec);
 508
 509             fscal            = felec;
 510
 511             /* Calculate temporary vectorial force */
 512             tx               = _mm256_mul_pd(fscal,dx33);
 513             ty               = _mm256_mul_pd(fscal,dy33);
 514             tz               = _mm256_mul_pd(fscal,dz33);
 515
 516             /* Update vectorial force */
 517             fix3             = _mm256_add_pd(fix3,tx);
 518             fiy3             = _mm256_add_pd(fiy3,ty);
 519             fiz3             = _mm256_add_pd(fiz3,tz);
 520
 521             fjx3             = _mm256_add_pd(fjx3,tx);
 522             fjy3             = _mm256_add_pd(fjy3,ty);
 523             fjz3             = _mm256_add_pd(fjz3,tz);
 524
 525             fjptrA             = f+j_coord_offsetA;
 526             fjptrB             = f+j_coord_offsetB;
 527             fjptrC             = f+j_coord_offsetC;
 528             fjptrD             = f+j_coord_offsetD;
 529
 530             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 531                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 532
 533             /* Inner loop uses 243 flops */
 534         }
 535
 536         if(jidx<j_index_end)
 537         {
 538
 539             /* Get j neighbor index, and coordinate index */
 540             jnrlistA         = jjnr[jidx];
 541             jnrlistB         = jjnr[jidx+1];
 542             jnrlistC         = jjnr[jidx+2];
 543             jnrlistD         = jjnr[jidx+3];
 544             /* Sign of each element will be negative for non-real atoms.
 545              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 546              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 547              */
 548             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 549
 550             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 551             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 552             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 553
 554             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 555             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 556             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 557             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 558             j_coord_offsetA  = DIM*jnrA;
 559             j_coord_offsetB  = DIM*jnrB;
 560             j_coord_offsetC  = DIM*jnrC;
 561             j_coord_offsetD  = DIM*jnrD;
 562
 563             /* load j atom coordinates */
 564             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 565                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 566                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 567
 568             /* Calculate displacement vector */
 569             dx11             = _mm256_sub_pd(ix1,jx1);
 570             dy11             = _mm256_sub_pd(iy1,jy1);
 571             dz11             = _mm256_sub_pd(iz1,jz1);
 572             dx12             = _mm256_sub_pd(ix1,jx2);
 573             dy12             = _mm256_sub_pd(iy1,jy2);
 574             dz12             = _mm256_sub_pd(iz1,jz2);
 575             dx13             = _mm256_sub_pd(ix1,jx3);
 576             dy13             = _mm256_sub_pd(iy1,jy3);
 577             dz13             = _mm256_sub_pd(iz1,jz3);
 578             dx21             = _mm256_sub_pd(ix2,jx1);
 579             dy21             = _mm256_sub_pd(iy2,jy1);
 580             dz21             = _mm256_sub_pd(iz2,jz1);
 581             dx22             = _mm256_sub_pd(ix2,jx2);
 582             dy22             = _mm256_sub_pd(iy2,jy2);
 583             dz22             = _mm256_sub_pd(iz2,jz2);
 584             dx23             = _mm256_sub_pd(ix2,jx3);
 585             dy23             = _mm256_sub_pd(iy2,jy3);
 586             dz23             = _mm256_sub_pd(iz2,jz3);
 587             dx31             = _mm256_sub_pd(ix3,jx1);
 588             dy31             = _mm256_sub_pd(iy3,jy1);
 589             dz31             = _mm256_sub_pd(iz3,jz1);
 590             dx32             = _mm256_sub_pd(ix3,jx2);
 591             dy32             = _mm256_sub_pd(iy3,jy2);
 592             dz32             = _mm256_sub_pd(iz3,jz2);
 593             dx33             = _mm256_sub_pd(ix3,jx3);
 594             dy33             = _mm256_sub_pd(iy3,jy3);
 595             dz33             = _mm256_sub_pd(iz3,jz3);
 596
 597             /* Calculate squared distance and things based on it */
 598             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 599             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 600             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 601             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 602             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 603             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 604             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 605             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 606             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 607
 608             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 609             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 610             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 611             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 612             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 613             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 614             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 615             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 616             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 617
 618             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 619             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 620             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 621             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 622             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 623             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 624             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 625             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 626             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 627
 628             fjx1             = _mm256_setzero_pd();
 629             fjy1             = _mm256_setzero_pd();
 630             fjz1             = _mm256_setzero_pd();
 631             fjx2             = _mm256_setzero_pd();
 632             fjy2             = _mm256_setzero_pd();
 633             fjz2             = _mm256_setzero_pd();
 634             fjx3             = _mm256_setzero_pd();
 635             fjy3             = _mm256_setzero_pd();
 636             fjz3             = _mm256_setzero_pd();
 637
 638             /**************************
 639              * CALCULATE INTERACTIONS *
 640              **************************/
 641
 642             /* COULOMB ELECTROSTATICS */
 643             velec            = _mm256_mul_pd(qq11,rinv11);
 644             felec            = _mm256_mul_pd(velec,rinvsq11);
 645
 646             /* Update potential sum for this i atom from the interaction with this j atom. */
 647             velec            = _mm256_andnot_pd(dummy_mask,velec);
 648             velecsum         = _mm256_add_pd(velecsum,velec);
 649
 650             fscal            = felec;
 651
 652             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 653
 654             /* Calculate temporary vectorial force */
 655             tx               = _mm256_mul_pd(fscal,dx11);
 656             ty               = _mm256_mul_pd(fscal,dy11);
 657             tz               = _mm256_mul_pd(fscal,dz11);
 658
 659             /* Update vectorial force */
 660             fix1             = _mm256_add_pd(fix1,tx);
 661             fiy1             = _mm256_add_pd(fiy1,ty);
 662             fiz1             = _mm256_add_pd(fiz1,tz);
 663
 664             fjx1             = _mm256_add_pd(fjx1,tx);
 665             fjy1             = _mm256_add_pd(fjy1,ty);
 666             fjz1             = _mm256_add_pd(fjz1,tz);
 667
 668             /**************************
 669              * CALCULATE INTERACTIONS *
 670              **************************/
 671
 672             /* COULOMB ELECTROSTATICS */
 673             velec            = _mm256_mul_pd(qq12,rinv12);
 674             felec            = _mm256_mul_pd(velec,rinvsq12);
 675
 676             /* Update potential sum for this i atom from the interaction with this j atom. */
 677             velec            = _mm256_andnot_pd(dummy_mask,velec);
 678             velecsum         = _mm256_add_pd(velecsum,velec);
 679
 680             fscal            = felec;
 681
 682             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 683
 684             /* Calculate temporary vectorial force */
 685             tx               = _mm256_mul_pd(fscal,dx12);
 686             ty               = _mm256_mul_pd(fscal,dy12);
 687             tz               = _mm256_mul_pd(fscal,dz12);
 688
 689             /* Update vectorial force */
 690             fix1             = _mm256_add_pd(fix1,tx);
 691             fiy1             = _mm256_add_pd(fiy1,ty);
 692             fiz1             = _mm256_add_pd(fiz1,tz);
 693
 694             fjx2             = _mm256_add_pd(fjx2,tx);
 695             fjy2             = _mm256_add_pd(fjy2,ty);
 696             fjz2             = _mm256_add_pd(fjz2,tz);
 697
 698             /**************************
 699              * CALCULATE INTERACTIONS *
 700              **************************/
 701
 702             /* COULOMB ELECTROSTATICS */
 703             velec            = _mm256_mul_pd(qq13,rinv13);
 704             felec            = _mm256_mul_pd(velec,rinvsq13);
 705
 706             /* Update potential sum for this i atom from the interaction with this j atom. */
 707             velec            = _mm256_andnot_pd(dummy_mask,velec);
 708             velecsum         = _mm256_add_pd(velecsum,velec);
 709
 710             fscal            = felec;
 711
 712             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 713
 714             /* Calculate temporary vectorial force */
 715             tx               = _mm256_mul_pd(fscal,dx13);
 716             ty               = _mm256_mul_pd(fscal,dy13);
 717             tz               = _mm256_mul_pd(fscal,dz13);
 718
 719             /* Update vectorial force */
 720             fix1             = _mm256_add_pd(fix1,tx);
 721             fiy1             = _mm256_add_pd(fiy1,ty);
 722             fiz1             = _mm256_add_pd(fiz1,tz);
 723
 724             fjx3             = _mm256_add_pd(fjx3,tx);
 725             fjy3             = _mm256_add_pd(fjy3,ty);
 726             fjz3             = _mm256_add_pd(fjz3,tz);
 727
 728             /**************************
 729              * CALCULATE INTERACTIONS *
 730              **************************/
 731
 732             /* COULOMB ELECTROSTATICS */
 733             velec            = _mm256_mul_pd(qq21,rinv21);
 734             felec            = _mm256_mul_pd(velec,rinvsq21);
 735
 736             /* Update potential sum for this i atom from the interaction with this j atom. */
 737             velec            = _mm256_andnot_pd(dummy_mask,velec);
 738             velecsum         = _mm256_add_pd(velecsum,velec);
 739
 740             fscal            = felec;
 741
 742             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 743
 744             /* Calculate temporary vectorial force */
 745             tx               = _mm256_mul_pd(fscal,dx21);
 746             ty               = _mm256_mul_pd(fscal,dy21);
 747             tz               = _mm256_mul_pd(fscal,dz21);
 748
 749             /* Update vectorial force */
 750             fix2             = _mm256_add_pd(fix2,tx);
 751             fiy2             = _mm256_add_pd(fiy2,ty);
 752             fiz2             = _mm256_add_pd(fiz2,tz);
 753
 754             fjx1             = _mm256_add_pd(fjx1,tx);
 755             fjy1             = _mm256_add_pd(fjy1,ty);
 756             fjz1             = _mm256_add_pd(fjz1,tz);
 757
 758             /**************************
 759              * CALCULATE INTERACTIONS *
 760              **************************/
 761
 762             /* COULOMB ELECTROSTATICS */
 763             velec            = _mm256_mul_pd(qq22,rinv22);
 764             felec            = _mm256_mul_pd(velec,rinvsq22);
 765
 766             /* Update potential sum for this i atom from the interaction with this j atom. */
 767             velec            = _mm256_andnot_pd(dummy_mask,velec);
 768             velecsum         = _mm256_add_pd(velecsum,velec);
 769
 770             fscal            = felec;
 771
 772             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 773
 774             /* Calculate temporary vectorial force */
 775             tx               = _mm256_mul_pd(fscal,dx22);
 776             ty               = _mm256_mul_pd(fscal,dy22);
 777             tz               = _mm256_mul_pd(fscal,dz22);
 778
 779             /* Update vectorial force */
 780             fix2             = _mm256_add_pd(fix2,tx);
 781             fiy2             = _mm256_add_pd(fiy2,ty);
 782             fiz2             = _mm256_add_pd(fiz2,tz);
 783
 784             fjx2             = _mm256_add_pd(fjx2,tx);
 785             fjy2             = _mm256_add_pd(fjy2,ty);
 786             fjz2             = _mm256_add_pd(fjz2,tz);
 787
 788             /**************************
 789              * CALCULATE INTERACTIONS *
 790              **************************/
 791
 792             /* COULOMB ELECTROSTATICS */
 793             velec            = _mm256_mul_pd(qq23,rinv23);
 794             felec            = _mm256_mul_pd(velec,rinvsq23);
 795
 796             /* Update potential sum for this i atom from the interaction with this j atom. */
 797             velec            = _mm256_andnot_pd(dummy_mask,velec);
 798             velecsum         = _mm256_add_pd(velecsum,velec);
 799
 800             fscal            = felec;
 801
 802             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 803
 804             /* Calculate temporary vectorial force */
 805             tx               = _mm256_mul_pd(fscal,dx23);
 806             ty               = _mm256_mul_pd(fscal,dy23);
 807             tz               = _mm256_mul_pd(fscal,dz23);
 808
 809             /* Update vectorial force */
 810             fix2             = _mm256_add_pd(fix2,tx);
 811             fiy2             = _mm256_add_pd(fiy2,ty);
 812             fiz2             = _mm256_add_pd(fiz2,tz);
 813
 814             fjx3             = _mm256_add_pd(fjx3,tx);
 815             fjy3             = _mm256_add_pd(fjy3,ty);
 816             fjz3             = _mm256_add_pd(fjz3,tz);
 817
 818             /**************************
 819              * CALCULATE INTERACTIONS *
 820              **************************/
 821
 822             /* COULOMB ELECTROSTATICS */
 823             velec            = _mm256_mul_pd(qq31,rinv31);
 824             felec            = _mm256_mul_pd(velec,rinvsq31);
 825
 826             /* Update potential sum for this i atom from the interaction with this j atom. */
 827             velec            = _mm256_andnot_pd(dummy_mask,velec);
 828             velecsum         = _mm256_add_pd(velecsum,velec);
 829
 830             fscal            = felec;
 831
 832             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 833
 834             /* Calculate temporary vectorial force */
 835             tx               = _mm256_mul_pd(fscal,dx31);
 836             ty               = _mm256_mul_pd(fscal,dy31);
 837             tz               = _mm256_mul_pd(fscal,dz31);
 838
 839             /* Update vectorial force */
 840             fix3             = _mm256_add_pd(fix3,tx);
 841             fiy3             = _mm256_add_pd(fiy3,ty);
 842             fiz3             = _mm256_add_pd(fiz3,tz);
 843
 844             fjx1             = _mm256_add_pd(fjx1,tx);
 845             fjy1             = _mm256_add_pd(fjy1,ty);
 846             fjz1             = _mm256_add_pd(fjz1,tz);
 847
 848             /**************************
 849              * CALCULATE INTERACTIONS *
 850              **************************/
 851
 852             /* COULOMB ELECTROSTATICS */
 853             velec            = _mm256_mul_pd(qq32,rinv32);
 854             felec            = _mm256_mul_pd(velec,rinvsq32);
 855
 856             /* Update potential sum for this i atom from the interaction with this j atom. */
 857             velec            = _mm256_andnot_pd(dummy_mask,velec);
 858             velecsum         = _mm256_add_pd(velecsum,velec);
 859
 860             fscal            = felec;
 861
 862             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 863
 864             /* Calculate temporary vectorial force */
 865             tx               = _mm256_mul_pd(fscal,dx32);
 866             ty               = _mm256_mul_pd(fscal,dy32);
 867             tz               = _mm256_mul_pd(fscal,dz32);
 868
 869             /* Update vectorial force */
 870             fix3             = _mm256_add_pd(fix3,tx);
 871             fiy3             = _mm256_add_pd(fiy3,ty);
 872             fiz3             = _mm256_add_pd(fiz3,tz);
 873
 874             fjx2             = _mm256_add_pd(fjx2,tx);
 875             fjy2             = _mm256_add_pd(fjy2,ty);
 876             fjz2             = _mm256_add_pd(fjz2,tz);
 877
 878             /**************************
 879              * CALCULATE INTERACTIONS *
 880              **************************/
 881
 882             /* COULOMB ELECTROSTATICS */
 883             velec            = _mm256_mul_pd(qq33,rinv33);
 884             felec            = _mm256_mul_pd(velec,rinvsq33);
 885
 886             /* Update potential sum for this i atom from the interaction with this j atom. */
 887             velec            = _mm256_andnot_pd(dummy_mask,velec);
 888             velecsum         = _mm256_add_pd(velecsum,velec);
 889
 890             fscal            = felec;
 891
 892             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 893
 894             /* Calculate temporary vectorial force */
 895             tx               = _mm256_mul_pd(fscal,dx33);
 896             ty               = _mm256_mul_pd(fscal,dy33);
 897             tz               = _mm256_mul_pd(fscal,dz33);
 898
 899             /* Update vectorial force */
 900             fix3             = _mm256_add_pd(fix3,tx);
 901             fiy3             = _mm256_add_pd(fiy3,ty);
 902             fiz3             = _mm256_add_pd(fiz3,tz);
 903
 904             fjx3             = _mm256_add_pd(fjx3,tx);
 905             fjy3             = _mm256_add_pd(fjy3,ty);
 906             fjz3             = _mm256_add_pd(fjz3,tz);
 907
 908             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 909             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 910             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 911             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 912
 913             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 914                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 915
 916             /* Inner loop uses 243 flops */
 917         }
 918
 919         /* End of innermost loop */
 920
 921         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 922                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 923
 924         ggid                        = gid[iidx];
 925         /* Update potential energies */
 926         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 927
 928         /* Increment number of inner iterations */
 929         inneriter                  += j_index_end - j_index_start;
 930
 931         /* Outer loop uses 19 flops */
 932     }
 933
 934     /* Increment number of outer iterations */
 935     outeriter        += nri;
 936
 937     /* Update outer/inner flops */
 938
 939     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*243);
 940 }
 941 /*
 942  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW4W4_F_avx_256_double
 943  * Electrostatics interaction: Coulomb
 944  * VdW interaction:            None
 945  * Geometry:                   Water4-Water4
 946  * Calculate force/pot:        Force
 947  */
 948 void
 949 nb_kernel_ElecCoul_VdwNone_GeomW4W4_F_avx_256_double
 950                     (t_nblist                    * gmx_restrict       nlist,
 951                      rvec                        * gmx_restrict          xx,
 952                      rvec                        * gmx_restrict          ff,
 953                      t_forcerec                  * gmx_restrict          fr,
 954                      t_mdatoms                   * gmx_restrict     mdatoms,
 955                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 956                      t_nrnb                      * gmx_restrict        nrnb)
 957 {
 958     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 959      * just 0 for non-waters.
 960      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 961      * jnr indices corresponding to data put in the four positions in the SIMD register.
 962      */
 963     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 964     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 965     int              jnrA,jnrB,jnrC,jnrD;
 966     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 967     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 968     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 969     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 970     real             rcutoff_scalar;
 971     real             *shiftvec,*fshift,*x,*f;
 972     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 973     real             scratch[4*DIM];
 974     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 975     real *           vdwioffsetptr1;
 976     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 977     real *           vdwioffsetptr2;
 978     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 979     real *           vdwioffsetptr3;
 980     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 981     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
 982     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 983     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
 984     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 985     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
 986     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 987     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 988     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 989     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 990     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 991     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 992     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 993     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 994     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 995     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 996     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 997     real             *charge;
 998     __m256d          dummy_mask,cutoff_mask;
 999     __m128           tmpmask0,tmpmask1;
1000     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1001     __m256d          one     = _mm256_set1_pd(1.0);
1002     __m256d          two     = _mm256_set1_pd(2.0);
1003     x                = xx[0];
1004     f                = ff[0];
1005
1006     nri              = nlist->nri;
1007     iinr             = nlist->iinr;
1008     jindex           = nlist->jindex;
1009     jjnr             = nlist->jjnr;
1010     shiftidx         = nlist->shift;
1011     gid              = nlist->gid;
1012     shiftvec         = fr->shift_vec[0];
1013     fshift           = fr->fshift[0];
1014     facel            = _mm256_set1_pd(fr->epsfac);
1015     charge           = mdatoms->chargeA;
1016
1017     /* Setup water-specific parameters */
1018     inr              = nlist->iinr[0];
1019     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1020     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1021     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1022
1023     jq1              = _mm256_set1_pd(charge[inr+1]);
1024     jq2              = _mm256_set1_pd(charge[inr+2]);
1025     jq3              = _mm256_set1_pd(charge[inr+3]);
1026     qq11             = _mm256_mul_pd(iq1,jq1);
1027     qq12             = _mm256_mul_pd(iq1,jq2);
1028     qq13             = _mm256_mul_pd(iq1,jq3);
1029     qq21             = _mm256_mul_pd(iq2,jq1);
1030     qq22             = _mm256_mul_pd(iq2,jq2);
1031     qq23             = _mm256_mul_pd(iq2,jq3);
1032     qq31             = _mm256_mul_pd(iq3,jq1);
1033     qq32             = _mm256_mul_pd(iq3,jq2);
1034     qq33             = _mm256_mul_pd(iq3,jq3);
1035
1036     /* Avoid stupid compiler warnings */
1037     jnrA = jnrB = jnrC = jnrD = 0;
1038     j_coord_offsetA = 0;
1039     j_coord_offsetB = 0;
1040     j_coord_offsetC = 0;
1041     j_coord_offsetD = 0;
1042
1043     outeriter        = 0;
1044     inneriter        = 0;
1045
1046     for(iidx=0;iidx<4*DIM;iidx++)
1047     {
1048         scratch[iidx] = 0.0;
1049     }
1050
1051     /* Start outer loop over neighborlists */
1052     for(iidx=0; iidx<nri; iidx++)
1053     {
1054         /* Load shift vector for this list */
1055         i_shift_offset   = DIM*shiftidx[iidx];
1056
1057         /* Load limits for loop over neighbors */
1058         j_index_start    = jindex[iidx];
1059         j_index_end      = jindex[iidx+1];
1060
1061         /* Get outer coordinate index */
1062         inr              = iinr[iidx];
1063         i_coord_offset   = DIM*inr;
1064
1065         /* Load i particle coords and add shift vector */
1066         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1067                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1068
1069         fix1             = _mm256_setzero_pd();
1070         fiy1             = _mm256_setzero_pd();
1071         fiz1             = _mm256_setzero_pd();
1072         fix2             = _mm256_setzero_pd();
1073         fiy2             = _mm256_setzero_pd();
1074         fiz2             = _mm256_setzero_pd();
1075         fix3             = _mm256_setzero_pd();
1076         fiy3             = _mm256_setzero_pd();
1077         fiz3             = _mm256_setzero_pd();
1078
1079         /* Start inner kernel loop */
1080         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1081         {
1082
1083             /* Get j neighbor index, and coordinate index */
1084             jnrA             = jjnr[jidx];
1085             jnrB             = jjnr[jidx+1];
1086             jnrC             = jjnr[jidx+2];
1087             jnrD             = jjnr[jidx+3];
1088             j_coord_offsetA  = DIM*jnrA;
1089             j_coord_offsetB  = DIM*jnrB;
1090             j_coord_offsetC  = DIM*jnrC;
1091             j_coord_offsetD  = DIM*jnrD;
1092
1093             /* load j atom coordinates */
1094             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1095                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1096                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1097
1098             /* Calculate displacement vector */
1099             dx11             = _mm256_sub_pd(ix1,jx1);
1100             dy11             = _mm256_sub_pd(iy1,jy1);
1101             dz11             = _mm256_sub_pd(iz1,jz1);
1102             dx12             = _mm256_sub_pd(ix1,jx2);
1103             dy12             = _mm256_sub_pd(iy1,jy2);
1104             dz12             = _mm256_sub_pd(iz1,jz2);
1105             dx13             = _mm256_sub_pd(ix1,jx3);
1106             dy13             = _mm256_sub_pd(iy1,jy3);
1107             dz13             = _mm256_sub_pd(iz1,jz3);
1108             dx21             = _mm256_sub_pd(ix2,jx1);
1109             dy21             = _mm256_sub_pd(iy2,jy1);
1110             dz21             = _mm256_sub_pd(iz2,jz1);
1111             dx22             = _mm256_sub_pd(ix2,jx2);
1112             dy22             = _mm256_sub_pd(iy2,jy2);
1113             dz22             = _mm256_sub_pd(iz2,jz2);
1114             dx23             = _mm256_sub_pd(ix2,jx3);
1115             dy23             = _mm256_sub_pd(iy2,jy3);
1116             dz23             = _mm256_sub_pd(iz2,jz3);
1117             dx31             = _mm256_sub_pd(ix3,jx1);
1118             dy31             = _mm256_sub_pd(iy3,jy1);
1119             dz31             = _mm256_sub_pd(iz3,jz1);
1120             dx32             = _mm256_sub_pd(ix3,jx2);
1121             dy32             = _mm256_sub_pd(iy3,jy2);
1122             dz32             = _mm256_sub_pd(iz3,jz2);
1123             dx33             = _mm256_sub_pd(ix3,jx3);
1124             dy33             = _mm256_sub_pd(iy3,jy3);
1125             dz33             = _mm256_sub_pd(iz3,jz3);
1126
1127             /* Calculate squared distance and things based on it */
1128             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1129             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1130             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1131             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1132             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1133             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1134             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1135             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1136             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1137
1138             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1139             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1140             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1141             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1142             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1143             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1144             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1145             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1146             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1147
1148             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1149             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1150             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1151             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1152             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1153             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1154             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1155             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1156             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1157
1158             fjx1             = _mm256_setzero_pd();
1159             fjy1             = _mm256_setzero_pd();
1160             fjz1             = _mm256_setzero_pd();
1161             fjx2             = _mm256_setzero_pd();
1162             fjy2             = _mm256_setzero_pd();
1163             fjz2             = _mm256_setzero_pd();
1164             fjx3             = _mm256_setzero_pd();
1165             fjy3             = _mm256_setzero_pd();
1166             fjz3             = _mm256_setzero_pd();
1167
1168             /**************************
1169              * CALCULATE INTERACTIONS *
1170              **************************/
1171
1172             /* COULOMB ELECTROSTATICS */
1173             velec            = _mm256_mul_pd(qq11,rinv11);
1174             felec            = _mm256_mul_pd(velec,rinvsq11);
1175
1176             fscal            = felec;
1177
1178             /* Calculate temporary vectorial force */
1179             tx               = _mm256_mul_pd(fscal,dx11);
1180             ty               = _mm256_mul_pd(fscal,dy11);
1181             tz               = _mm256_mul_pd(fscal,dz11);
1182
1183             /* Update vectorial force */
1184             fix1             = _mm256_add_pd(fix1,tx);
1185             fiy1             = _mm256_add_pd(fiy1,ty);
1186             fiz1             = _mm256_add_pd(fiz1,tz);
1187
1188             fjx1             = _mm256_add_pd(fjx1,tx);
1189             fjy1             = _mm256_add_pd(fjy1,ty);
1190             fjz1             = _mm256_add_pd(fjz1,tz);
1191
1192             /**************************
1193              * CALCULATE INTERACTIONS *
1194              **************************/
1195
1196             /* COULOMB ELECTROSTATICS */
1197             velec            = _mm256_mul_pd(qq12,rinv12);
1198             felec            = _mm256_mul_pd(velec,rinvsq12);
1199
1200             fscal            = felec;
1201
1202             /* Calculate temporary vectorial force */
1203             tx               = _mm256_mul_pd(fscal,dx12);
1204             ty               = _mm256_mul_pd(fscal,dy12);
1205             tz               = _mm256_mul_pd(fscal,dz12);
1206
1207             /* Update vectorial force */
1208             fix1             = _mm256_add_pd(fix1,tx);
1209             fiy1             = _mm256_add_pd(fiy1,ty);
1210             fiz1             = _mm256_add_pd(fiz1,tz);
1211
1212             fjx2             = _mm256_add_pd(fjx2,tx);
1213             fjy2             = _mm256_add_pd(fjy2,ty);
1214             fjz2             = _mm256_add_pd(fjz2,tz);
1215
1216             /**************************
1217              * CALCULATE INTERACTIONS *
1218              **************************/
1219
1220             /* COULOMB ELECTROSTATICS */
1221             velec            = _mm256_mul_pd(qq13,rinv13);
1222             felec            = _mm256_mul_pd(velec,rinvsq13);
1223
1224             fscal            = felec;
1225
1226             /* Calculate temporary vectorial force */
1227             tx               = _mm256_mul_pd(fscal,dx13);
1228             ty               = _mm256_mul_pd(fscal,dy13);
1229             tz               = _mm256_mul_pd(fscal,dz13);
1230
1231             /* Update vectorial force */
1232             fix1             = _mm256_add_pd(fix1,tx);
1233             fiy1             = _mm256_add_pd(fiy1,ty);
1234             fiz1             = _mm256_add_pd(fiz1,tz);
1235
1236             fjx3             = _mm256_add_pd(fjx3,tx);
1237             fjy3             = _mm256_add_pd(fjy3,ty);
1238             fjz3             = _mm256_add_pd(fjz3,tz);
1239
1240             /**************************
1241              * CALCULATE INTERACTIONS *
1242              **************************/
1243
1244             /* COULOMB ELECTROSTATICS */
1245             velec            = _mm256_mul_pd(qq21,rinv21);
1246             felec            = _mm256_mul_pd(velec,rinvsq21);
1247
1248             fscal            = felec;
1249
1250             /* Calculate temporary vectorial force */
1251             tx               = _mm256_mul_pd(fscal,dx21);
1252             ty               = _mm256_mul_pd(fscal,dy21);
1253             tz               = _mm256_mul_pd(fscal,dz21);
1254
1255             /* Update vectorial force */
1256             fix2             = _mm256_add_pd(fix2,tx);
1257             fiy2             = _mm256_add_pd(fiy2,ty);
1258             fiz2             = _mm256_add_pd(fiz2,tz);
1259
1260             fjx1             = _mm256_add_pd(fjx1,tx);
1261             fjy1             = _mm256_add_pd(fjy1,ty);
1262             fjz1             = _mm256_add_pd(fjz1,tz);
1263
1264             /**************************
1265              * CALCULATE INTERACTIONS *
1266              **************************/
1267
1268             /* COULOMB ELECTROSTATICS */
1269             velec            = _mm256_mul_pd(qq22,rinv22);
1270             felec            = _mm256_mul_pd(velec,rinvsq22);
1271
1272             fscal            = felec;
1273
1274             /* Calculate temporary vectorial force */
1275             tx               = _mm256_mul_pd(fscal,dx22);
1276             ty               = _mm256_mul_pd(fscal,dy22);
1277             tz               = _mm256_mul_pd(fscal,dz22);
1278
1279             /* Update vectorial force */
1280             fix2             = _mm256_add_pd(fix2,tx);
1281             fiy2             = _mm256_add_pd(fiy2,ty);
1282             fiz2             = _mm256_add_pd(fiz2,tz);
1283
1284             fjx2             = _mm256_add_pd(fjx2,tx);
1285             fjy2             = _mm256_add_pd(fjy2,ty);
1286             fjz2             = _mm256_add_pd(fjz2,tz);
1287
1288             /**************************
1289              * CALCULATE INTERACTIONS *
1290              **************************/
1291
1292             /* COULOMB ELECTROSTATICS */
1293             velec            = _mm256_mul_pd(qq23,rinv23);
1294             felec            = _mm256_mul_pd(velec,rinvsq23);
1295
1296             fscal            = felec;
1297
1298             /* Calculate temporary vectorial force */
1299             tx               = _mm256_mul_pd(fscal,dx23);
1300             ty               = _mm256_mul_pd(fscal,dy23);
1301             tz               = _mm256_mul_pd(fscal,dz23);
1302
1303             /* Update vectorial force */
1304             fix2             = _mm256_add_pd(fix2,tx);
1305             fiy2             = _mm256_add_pd(fiy2,ty);
1306             fiz2             = _mm256_add_pd(fiz2,tz);
1307
1308             fjx3             = _mm256_add_pd(fjx3,tx);
1309             fjy3             = _mm256_add_pd(fjy3,ty);
1310             fjz3             = _mm256_add_pd(fjz3,tz);
1311
1312             /**************************
1313              * CALCULATE INTERACTIONS *
1314              **************************/
1315
1316             /* COULOMB ELECTROSTATICS */
1317             velec            = _mm256_mul_pd(qq31,rinv31);
1318             felec            = _mm256_mul_pd(velec,rinvsq31);
1319
1320             fscal            = felec;
1321
1322             /* Calculate temporary vectorial force */
1323             tx               = _mm256_mul_pd(fscal,dx31);
1324             ty               = _mm256_mul_pd(fscal,dy31);
1325             tz               = _mm256_mul_pd(fscal,dz31);
1326
1327             /* Update vectorial force */
1328             fix3             = _mm256_add_pd(fix3,tx);
1329             fiy3             = _mm256_add_pd(fiy3,ty);
1330             fiz3             = _mm256_add_pd(fiz3,tz);
1331
1332             fjx1             = _mm256_add_pd(fjx1,tx);
1333             fjy1             = _mm256_add_pd(fjy1,ty);
1334             fjz1             = _mm256_add_pd(fjz1,tz);
1335
1336             /**************************
1337              * CALCULATE INTERACTIONS *
1338              **************************/
1339
1340             /* COULOMB ELECTROSTATICS */
1341             velec            = _mm256_mul_pd(qq32,rinv32);
1342             felec            = _mm256_mul_pd(velec,rinvsq32);
1343
1344             fscal            = felec;
1345
1346             /* Calculate temporary vectorial force */
1347             tx               = _mm256_mul_pd(fscal,dx32);
1348             ty               = _mm256_mul_pd(fscal,dy32);
1349             tz               = _mm256_mul_pd(fscal,dz32);
1350
1351             /* Update vectorial force */
1352             fix3             = _mm256_add_pd(fix3,tx);
1353             fiy3             = _mm256_add_pd(fiy3,ty);
1354             fiz3             = _mm256_add_pd(fiz3,tz);
1355
1356             fjx2             = _mm256_add_pd(fjx2,tx);
1357             fjy2             = _mm256_add_pd(fjy2,ty);
1358             fjz2             = _mm256_add_pd(fjz2,tz);
1359
1360             /**************************
1361              * CALCULATE INTERACTIONS *
1362              **************************/
1363
1364             /* COULOMB ELECTROSTATICS */
1365             velec            = _mm256_mul_pd(qq33,rinv33);
1366             felec            = _mm256_mul_pd(velec,rinvsq33);
1367
1368             fscal            = felec;
1369
1370             /* Calculate temporary vectorial force */
1371             tx               = _mm256_mul_pd(fscal,dx33);
1372             ty               = _mm256_mul_pd(fscal,dy33);
1373             tz               = _mm256_mul_pd(fscal,dz33);
1374
1375             /* Update vectorial force */
1376             fix3             = _mm256_add_pd(fix3,tx);
1377             fiy3             = _mm256_add_pd(fiy3,ty);
1378             fiz3             = _mm256_add_pd(fiz3,tz);
1379
1380             fjx3             = _mm256_add_pd(fjx3,tx);
1381             fjy3             = _mm256_add_pd(fjy3,ty);
1382             fjz3             = _mm256_add_pd(fjz3,tz);
1383
1384             fjptrA             = f+j_coord_offsetA;
1385             fjptrB             = f+j_coord_offsetB;
1386             fjptrC             = f+j_coord_offsetC;
1387             fjptrD             = f+j_coord_offsetD;
1388
1389             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1390                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1391
1392             /* Inner loop uses 234 flops */
1393         }
1394
1395         if(jidx<j_index_end)
1396         {
1397
1398             /* Get j neighbor index, and coordinate index */
1399             jnrlistA         = jjnr[jidx];
1400             jnrlistB         = jjnr[jidx+1];
1401             jnrlistC         = jjnr[jidx+2];
1402             jnrlistD         = jjnr[jidx+3];
1403             /* Sign of each element will be negative for non-real atoms.
1404              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1405              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1406              */
1407             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1408
1409             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1410             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1411             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1412
1413             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1414             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1415             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1416             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1417             j_coord_offsetA  = DIM*jnrA;
1418             j_coord_offsetB  = DIM*jnrB;
1419             j_coord_offsetC  = DIM*jnrC;
1420             j_coord_offsetD  = DIM*jnrD;
1421
1422             /* load j atom coordinates */
1423             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1424                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1425                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1426
1427             /* Calculate displacement vector */
1428             dx11             = _mm256_sub_pd(ix1,jx1);
1429             dy11             = _mm256_sub_pd(iy1,jy1);
1430             dz11             = _mm256_sub_pd(iz1,jz1);
1431             dx12             = _mm256_sub_pd(ix1,jx2);
1432             dy12             = _mm256_sub_pd(iy1,jy2);
1433             dz12             = _mm256_sub_pd(iz1,jz2);
1434             dx13             = _mm256_sub_pd(ix1,jx3);
1435             dy13             = _mm256_sub_pd(iy1,jy3);
1436             dz13             = _mm256_sub_pd(iz1,jz3);
1437             dx21             = _mm256_sub_pd(ix2,jx1);
1438             dy21             = _mm256_sub_pd(iy2,jy1);
1439             dz21             = _mm256_sub_pd(iz2,jz1);
1440             dx22             = _mm256_sub_pd(ix2,jx2);
1441             dy22             = _mm256_sub_pd(iy2,jy2);
1442             dz22             = _mm256_sub_pd(iz2,jz2);
1443             dx23             = _mm256_sub_pd(ix2,jx3);
1444             dy23             = _mm256_sub_pd(iy2,jy3);
1445             dz23             = _mm256_sub_pd(iz2,jz3);
1446             dx31             = _mm256_sub_pd(ix3,jx1);
1447             dy31             = _mm256_sub_pd(iy3,jy1);
1448             dz31             = _mm256_sub_pd(iz3,jz1);
1449             dx32             = _mm256_sub_pd(ix3,jx2);
1450             dy32             = _mm256_sub_pd(iy3,jy2);
1451             dz32             = _mm256_sub_pd(iz3,jz2);
1452             dx33             = _mm256_sub_pd(ix3,jx3);
1453             dy33             = _mm256_sub_pd(iy3,jy3);
1454             dz33             = _mm256_sub_pd(iz3,jz3);
1455
1456             /* Calculate squared distance and things based on it */
1457             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1458             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1459             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1460             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1461             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1462             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1463             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1464             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1465             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1466
1467             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1468             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1469             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1470             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1471             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1472             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1473             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1474             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1475             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1476
1477             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1478             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1479             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1480             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1481             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1482             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1483             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1484             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1485             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1486
1487             fjx1             = _mm256_setzero_pd();
1488             fjy1             = _mm256_setzero_pd();
1489             fjz1             = _mm256_setzero_pd();
1490             fjx2             = _mm256_setzero_pd();
1491             fjy2             = _mm256_setzero_pd();
1492             fjz2             = _mm256_setzero_pd();
1493             fjx3             = _mm256_setzero_pd();
1494             fjy3             = _mm256_setzero_pd();
1495             fjz3             = _mm256_setzero_pd();
1496
1497             /**************************
1498              * CALCULATE INTERACTIONS *
1499              **************************/
1500
1501             /* COULOMB ELECTROSTATICS */
1502             velec            = _mm256_mul_pd(qq11,rinv11);
1503             felec            = _mm256_mul_pd(velec,rinvsq11);
1504
1505             fscal            = felec;
1506
1507             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1508
1509             /* Calculate temporary vectorial force */
1510             tx               = _mm256_mul_pd(fscal,dx11);
1511             ty               = _mm256_mul_pd(fscal,dy11);
1512             tz               = _mm256_mul_pd(fscal,dz11);
1513
1514             /* Update vectorial force */
1515             fix1             = _mm256_add_pd(fix1,tx);
1516             fiy1             = _mm256_add_pd(fiy1,ty);
1517             fiz1             = _mm256_add_pd(fiz1,tz);
1518
1519             fjx1             = _mm256_add_pd(fjx1,tx);
1520             fjy1             = _mm256_add_pd(fjy1,ty);
1521             fjz1             = _mm256_add_pd(fjz1,tz);
1522
1523             /**************************
1524              * CALCULATE INTERACTIONS *
1525              **************************/
1526
1527             /* COULOMB ELECTROSTATICS */
1528             velec            = _mm256_mul_pd(qq12,rinv12);
1529             felec            = _mm256_mul_pd(velec,rinvsq12);
1530
1531             fscal            = felec;
1532
1533             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1534
1535             /* Calculate temporary vectorial force */
1536             tx               = _mm256_mul_pd(fscal,dx12);
1537             ty               = _mm256_mul_pd(fscal,dy12);
1538             tz               = _mm256_mul_pd(fscal,dz12);
1539
1540             /* Update vectorial force */
1541             fix1             = _mm256_add_pd(fix1,tx);
1542             fiy1             = _mm256_add_pd(fiy1,ty);
1543             fiz1             = _mm256_add_pd(fiz1,tz);
1544
1545             fjx2             = _mm256_add_pd(fjx2,tx);
1546             fjy2             = _mm256_add_pd(fjy2,ty);
1547             fjz2             = _mm256_add_pd(fjz2,tz);
1548
1549             /**************************
1550              * CALCULATE INTERACTIONS *
1551              **************************/
1552
1553             /* COULOMB ELECTROSTATICS */
1554             velec            = _mm256_mul_pd(qq13,rinv13);
1555             felec            = _mm256_mul_pd(velec,rinvsq13);
1556
1557             fscal            = felec;
1558
1559             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1560
1561             /* Calculate temporary vectorial force */
1562             tx               = _mm256_mul_pd(fscal,dx13);
1563             ty               = _mm256_mul_pd(fscal,dy13);
1564             tz               = _mm256_mul_pd(fscal,dz13);
1565
1566             /* Update vectorial force */
1567             fix1             = _mm256_add_pd(fix1,tx);
1568             fiy1             = _mm256_add_pd(fiy1,ty);
1569             fiz1             = _mm256_add_pd(fiz1,tz);
1570
1571             fjx3             = _mm256_add_pd(fjx3,tx);
1572             fjy3             = _mm256_add_pd(fjy3,ty);
1573             fjz3             = _mm256_add_pd(fjz3,tz);
1574
1575             /**************************
1576              * CALCULATE INTERACTIONS *
1577              **************************/
1578
1579             /* COULOMB ELECTROSTATICS */
1580             velec            = _mm256_mul_pd(qq21,rinv21);
1581             felec            = _mm256_mul_pd(velec,rinvsq21);
1582
1583             fscal            = felec;
1584
1585             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1586
1587             /* Calculate temporary vectorial force */
1588             tx               = _mm256_mul_pd(fscal,dx21);
1589             ty               = _mm256_mul_pd(fscal,dy21);
1590             tz               = _mm256_mul_pd(fscal,dz21);
1591
1592             /* Update vectorial force */
1593             fix2             = _mm256_add_pd(fix2,tx);
1594             fiy2             = _mm256_add_pd(fiy2,ty);
1595             fiz2             = _mm256_add_pd(fiz2,tz);
1596
1597             fjx1             = _mm256_add_pd(fjx1,tx);
1598             fjy1             = _mm256_add_pd(fjy1,ty);
1599             fjz1             = _mm256_add_pd(fjz1,tz);
1600
1601             /**************************
1602              * CALCULATE INTERACTIONS *
1603              **************************/
1604
1605             /* COULOMB ELECTROSTATICS */
1606             velec            = _mm256_mul_pd(qq22,rinv22);
1607             felec            = _mm256_mul_pd(velec,rinvsq22);
1608
1609             fscal            = felec;
1610
1611             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1612
1613             /* Calculate temporary vectorial force */
1614             tx               = _mm256_mul_pd(fscal,dx22);
1615             ty               = _mm256_mul_pd(fscal,dy22);
1616             tz               = _mm256_mul_pd(fscal,dz22);
1617
1618             /* Update vectorial force */
1619             fix2             = _mm256_add_pd(fix2,tx);
1620             fiy2             = _mm256_add_pd(fiy2,ty);
1621             fiz2             = _mm256_add_pd(fiz2,tz);
1622
1623             fjx2             = _mm256_add_pd(fjx2,tx);
1624             fjy2             = _mm256_add_pd(fjy2,ty);
1625             fjz2             = _mm256_add_pd(fjz2,tz);
1626
1627             /**************************
1628              * CALCULATE INTERACTIONS *
1629              **************************/
1630
1631             /* COULOMB ELECTROSTATICS */
1632             velec            = _mm256_mul_pd(qq23,rinv23);
1633             felec            = _mm256_mul_pd(velec,rinvsq23);
1634
1635             fscal            = felec;
1636
1637             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1638
1639             /* Calculate temporary vectorial force */
1640             tx               = _mm256_mul_pd(fscal,dx23);
1641             ty               = _mm256_mul_pd(fscal,dy23);
1642             tz               = _mm256_mul_pd(fscal,dz23);
1643
1644             /* Update vectorial force */
1645             fix2             = _mm256_add_pd(fix2,tx);
1646             fiy2             = _mm256_add_pd(fiy2,ty);
1647             fiz2             = _mm256_add_pd(fiz2,tz);
1648
1649             fjx3             = _mm256_add_pd(fjx3,tx);
1650             fjy3             = _mm256_add_pd(fjy3,ty);
1651             fjz3             = _mm256_add_pd(fjz3,tz);
1652
1653             /**************************
1654              * CALCULATE INTERACTIONS *
1655              **************************/
1656
1657             /* COULOMB ELECTROSTATICS */
1658             velec            = _mm256_mul_pd(qq31,rinv31);
1659             felec            = _mm256_mul_pd(velec,rinvsq31);
1660
1661             fscal            = felec;
1662
1663             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1664
1665             /* Calculate temporary vectorial force */
1666             tx               = _mm256_mul_pd(fscal,dx31);
1667             ty               = _mm256_mul_pd(fscal,dy31);
1668             tz               = _mm256_mul_pd(fscal,dz31);
1669
1670             /* Update vectorial force */
1671             fix3             = _mm256_add_pd(fix3,tx);
1672             fiy3             = _mm256_add_pd(fiy3,ty);
1673             fiz3             = _mm256_add_pd(fiz3,tz);
1674
1675             fjx1             = _mm256_add_pd(fjx1,tx);
1676             fjy1             = _mm256_add_pd(fjy1,ty);
1677             fjz1             = _mm256_add_pd(fjz1,tz);
1678
1679             /**************************
1680              * CALCULATE INTERACTIONS *
1681              **************************/
1682
1683             /* COULOMB ELECTROSTATICS */
1684             velec            = _mm256_mul_pd(qq32,rinv32);
1685             felec            = _mm256_mul_pd(velec,rinvsq32);
1686
1687             fscal            = felec;
1688
1689             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1690
1691             /* Calculate temporary vectorial force */
1692             tx               = _mm256_mul_pd(fscal,dx32);
1693             ty               = _mm256_mul_pd(fscal,dy32);
1694             tz               = _mm256_mul_pd(fscal,dz32);
1695
1696             /* Update vectorial force */
1697             fix3             = _mm256_add_pd(fix3,tx);
1698             fiy3             = _mm256_add_pd(fiy3,ty);
1699             fiz3             = _mm256_add_pd(fiz3,tz);
1700
1701             fjx2             = _mm256_add_pd(fjx2,tx);
1702             fjy2             = _mm256_add_pd(fjy2,ty);
1703             fjz2             = _mm256_add_pd(fjz2,tz);
1704
1705             /**************************
1706              * CALCULATE INTERACTIONS *
1707              **************************/
1708
1709             /* COULOMB ELECTROSTATICS */
1710             velec            = _mm256_mul_pd(qq33,rinv33);
1711             felec            = _mm256_mul_pd(velec,rinvsq33);
1712
1713             fscal            = felec;
1714
1715             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1716
1717             /* Calculate temporary vectorial force */
1718             tx               = _mm256_mul_pd(fscal,dx33);
1719             ty               = _mm256_mul_pd(fscal,dy33);
1720             tz               = _mm256_mul_pd(fscal,dz33);
1721
1722             /* Update vectorial force */
1723             fix3             = _mm256_add_pd(fix3,tx);
1724             fiy3             = _mm256_add_pd(fiy3,ty);
1725             fiz3             = _mm256_add_pd(fiz3,tz);
1726
1727             fjx3             = _mm256_add_pd(fjx3,tx);
1728             fjy3             = _mm256_add_pd(fjy3,ty);
1729             fjz3             = _mm256_add_pd(fjz3,tz);
1730
1731             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1732             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1733             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1734             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1735
1736             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1737                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1738
1739             /* Inner loop uses 234 flops */
1740         }
1741
1742         /* End of innermost loop */
1743
1744         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1745                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1746
1747         /* Increment number of inner iterations */
1748         inneriter                  += j_index_end - j_index_start;
1749
1750         /* Outer loop uses 18 flops */
1751     }
1752
1753     /* Increment number of outer iterations */
1754     outeriter        += nri;
1755
1756     /* Update outer/inner flops */
1757
1758     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*234);
1759 }