src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_VdwNone_GeomW4W4_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4W4_VF_avx_256_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            None
  53  * Geometry:                   Water4-Water4
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_VdwNone_GeomW4W4_VF_avx_256_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr1;
  84     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  85     real *           vdwioffsetptr2;
  86     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  87     real *           vdwioffsetptr3;
  88     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  89     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  90     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  91     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  92     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  93     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  94     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  95     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  96     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  97     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  98     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  99     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 100     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 101     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 102     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 103     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 104     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 105     real             *charge;
 106     __m256d          dummy_mask,cutoff_mask;
 107     __m128           tmpmask0,tmpmask1;
 108     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 109     __m256d          one     = _mm256_set1_pd(1.0);
 110     __m256d          two     = _mm256_set1_pd(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm256_set1_pd(fr->ic->epsfac);
 123     charge           = mdatoms->chargeA;
 124     krf              = _mm256_set1_pd(fr->ic->k_rf);
 125     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 126     crf              = _mm256_set1_pd(fr->ic->c_rf);
 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           = avx256_invsqrt_d(rsq11);
 253             rinv12           = avx256_invsqrt_d(rsq12);
 254             rinv13           = avx256_invsqrt_d(rsq13);
 255             rinv21           = avx256_invsqrt_d(rsq21);
 256             rinv22           = avx256_invsqrt_d(rsq22);
 257             rinv23           = avx256_invsqrt_d(rsq23);
 258             rinv31           = avx256_invsqrt_d(rsq31);
 259             rinv32           = avx256_invsqrt_d(rsq32);
 260             rinv33           = avx256_invsqrt_d(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             /* REACTION-FIELD ELECTROSTATICS */
 287             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 288             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 314             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 315             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 341             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
 342             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 368             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 369             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 395             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 396             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 422             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
 423             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 449             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
 450             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 476             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
 477             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 503             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
 504             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
 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 288 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           = avx256_invsqrt_d(rsq11);
 609             rinv12           = avx256_invsqrt_d(rsq12);
 610             rinv13           = avx256_invsqrt_d(rsq13);
 611             rinv21           = avx256_invsqrt_d(rsq21);
 612             rinv22           = avx256_invsqrt_d(rsq22);
 613             rinv23           = avx256_invsqrt_d(rsq23);
 614             rinv31           = avx256_invsqrt_d(rsq31);
 615             rinv32           = avx256_invsqrt_d(rsq32);
 616             rinv33           = avx256_invsqrt_d(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             /* REACTION-FIELD ELECTROSTATICS */
 643             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 644             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 673             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 674             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 703             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
 704             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 733             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 734             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 763             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 764             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 793             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
 794             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 823             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
 824             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 853             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
 854             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 883             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
 884             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
 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 288 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*288);
 940 }
 941 /*
 942  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4W4_F_avx_256_double
 943  * Electrostatics interaction: ReactionField
 944  * VdW interaction:            None
 945  * Geometry:                   Water4-Water4
 946  * Calculate force/pot:        Force
 947  */
 948 void
 949 nb_kernel_ElecRF_VdwNone_GeomW4W4_F_avx_256_double
 950                     (t_nblist                    * gmx_restrict       nlist,
 951                      rvec                        * gmx_restrict          xx,
 952                      rvec                        * gmx_restrict          ff,
 953                      struct 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->ic->epsfac);
1015     charge           = mdatoms->chargeA;
1016     krf              = _mm256_set1_pd(fr->ic->k_rf);
1017     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
1018     crf              = _mm256_set1_pd(fr->ic->c_rf);
1019
1020     /* Setup water-specific parameters */
1021     inr              = nlist->iinr[0];
1022     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1023     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1024     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1025
1026     jq1              = _mm256_set1_pd(charge[inr+1]);
1027     jq2              = _mm256_set1_pd(charge[inr+2]);
1028     jq3              = _mm256_set1_pd(charge[inr+3]);
1029     qq11             = _mm256_mul_pd(iq1,jq1);
1030     qq12             = _mm256_mul_pd(iq1,jq2);
1031     qq13             = _mm256_mul_pd(iq1,jq3);
1032     qq21             = _mm256_mul_pd(iq2,jq1);
1033     qq22             = _mm256_mul_pd(iq2,jq2);
1034     qq23             = _mm256_mul_pd(iq2,jq3);
1035     qq31             = _mm256_mul_pd(iq3,jq1);
1036     qq32             = _mm256_mul_pd(iq3,jq2);
1037     qq33             = _mm256_mul_pd(iq3,jq3);
1038
1039     /* Avoid stupid compiler warnings */
1040     jnrA = jnrB = jnrC = jnrD = 0;
1041     j_coord_offsetA = 0;
1042     j_coord_offsetB = 0;
1043     j_coord_offsetC = 0;
1044     j_coord_offsetD = 0;
1045
1046     outeriter        = 0;
1047     inneriter        = 0;
1048
1049     for(iidx=0;iidx<4*DIM;iidx++)
1050     {
1051         scratch[iidx] = 0.0;
1052     }
1053
1054     /* Start outer loop over neighborlists */
1055     for(iidx=0; iidx<nri; iidx++)
1056     {
1057         /* Load shift vector for this list */
1058         i_shift_offset   = DIM*shiftidx[iidx];
1059
1060         /* Load limits for loop over neighbors */
1061         j_index_start    = jindex[iidx];
1062         j_index_end      = jindex[iidx+1];
1063
1064         /* Get outer coordinate index */
1065         inr              = iinr[iidx];
1066         i_coord_offset   = DIM*inr;
1067
1068         /* Load i particle coords and add shift vector */
1069         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1070                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1071
1072         fix1             = _mm256_setzero_pd();
1073         fiy1             = _mm256_setzero_pd();
1074         fiz1             = _mm256_setzero_pd();
1075         fix2             = _mm256_setzero_pd();
1076         fiy2             = _mm256_setzero_pd();
1077         fiz2             = _mm256_setzero_pd();
1078         fix3             = _mm256_setzero_pd();
1079         fiy3             = _mm256_setzero_pd();
1080         fiz3             = _mm256_setzero_pd();
1081
1082         /* Start inner kernel loop */
1083         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1084         {
1085
1086             /* Get j neighbor index, and coordinate index */
1087             jnrA             = jjnr[jidx];
1088             jnrB             = jjnr[jidx+1];
1089             jnrC             = jjnr[jidx+2];
1090             jnrD             = jjnr[jidx+3];
1091             j_coord_offsetA  = DIM*jnrA;
1092             j_coord_offsetB  = DIM*jnrB;
1093             j_coord_offsetC  = DIM*jnrC;
1094             j_coord_offsetD  = DIM*jnrD;
1095
1096             /* load j atom coordinates */
1097             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1098                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1099                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1100
1101             /* Calculate displacement vector */
1102             dx11             = _mm256_sub_pd(ix1,jx1);
1103             dy11             = _mm256_sub_pd(iy1,jy1);
1104             dz11             = _mm256_sub_pd(iz1,jz1);
1105             dx12             = _mm256_sub_pd(ix1,jx2);
1106             dy12             = _mm256_sub_pd(iy1,jy2);
1107             dz12             = _mm256_sub_pd(iz1,jz2);
1108             dx13             = _mm256_sub_pd(ix1,jx3);
1109             dy13             = _mm256_sub_pd(iy1,jy3);
1110             dz13             = _mm256_sub_pd(iz1,jz3);
1111             dx21             = _mm256_sub_pd(ix2,jx1);
1112             dy21             = _mm256_sub_pd(iy2,jy1);
1113             dz21             = _mm256_sub_pd(iz2,jz1);
1114             dx22             = _mm256_sub_pd(ix2,jx2);
1115             dy22             = _mm256_sub_pd(iy2,jy2);
1116             dz22             = _mm256_sub_pd(iz2,jz2);
1117             dx23             = _mm256_sub_pd(ix2,jx3);
1118             dy23             = _mm256_sub_pd(iy2,jy3);
1119             dz23             = _mm256_sub_pd(iz2,jz3);
1120             dx31             = _mm256_sub_pd(ix3,jx1);
1121             dy31             = _mm256_sub_pd(iy3,jy1);
1122             dz31             = _mm256_sub_pd(iz3,jz1);
1123             dx32             = _mm256_sub_pd(ix3,jx2);
1124             dy32             = _mm256_sub_pd(iy3,jy2);
1125             dz32             = _mm256_sub_pd(iz3,jz2);
1126             dx33             = _mm256_sub_pd(ix3,jx3);
1127             dy33             = _mm256_sub_pd(iy3,jy3);
1128             dz33             = _mm256_sub_pd(iz3,jz3);
1129
1130             /* Calculate squared distance and things based on it */
1131             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1132             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1133             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1134             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1135             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1136             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1137             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1138             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1139             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1140
1141             rinv11           = avx256_invsqrt_d(rsq11);
1142             rinv12           = avx256_invsqrt_d(rsq12);
1143             rinv13           = avx256_invsqrt_d(rsq13);
1144             rinv21           = avx256_invsqrt_d(rsq21);
1145             rinv22           = avx256_invsqrt_d(rsq22);
1146             rinv23           = avx256_invsqrt_d(rsq23);
1147             rinv31           = avx256_invsqrt_d(rsq31);
1148             rinv32           = avx256_invsqrt_d(rsq32);
1149             rinv33           = avx256_invsqrt_d(rsq33);
1150
1151             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1152             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1153             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1154             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1155             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1156             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1157             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1158             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1159             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1160
1161             fjx1             = _mm256_setzero_pd();
1162             fjy1             = _mm256_setzero_pd();
1163             fjz1             = _mm256_setzero_pd();
1164             fjx2             = _mm256_setzero_pd();
1165             fjy2             = _mm256_setzero_pd();
1166             fjz2             = _mm256_setzero_pd();
1167             fjx3             = _mm256_setzero_pd();
1168             fjy3             = _mm256_setzero_pd();
1169             fjz3             = _mm256_setzero_pd();
1170
1171             /**************************
1172              * CALCULATE INTERACTIONS *
1173              **************************/
1174
1175             /* REACTION-FIELD ELECTROSTATICS */
1176             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1177
1178             fscal            = felec;
1179
1180             /* Calculate temporary vectorial force */
1181             tx               = _mm256_mul_pd(fscal,dx11);
1182             ty               = _mm256_mul_pd(fscal,dy11);
1183             tz               = _mm256_mul_pd(fscal,dz11);
1184
1185             /* Update vectorial force */
1186             fix1             = _mm256_add_pd(fix1,tx);
1187             fiy1             = _mm256_add_pd(fiy1,ty);
1188             fiz1             = _mm256_add_pd(fiz1,tz);
1189
1190             fjx1             = _mm256_add_pd(fjx1,tx);
1191             fjy1             = _mm256_add_pd(fjy1,ty);
1192             fjz1             = _mm256_add_pd(fjz1,tz);
1193
1194             /**************************
1195              * CALCULATE INTERACTIONS *
1196              **************************/
1197
1198             /* REACTION-FIELD ELECTROSTATICS */
1199             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1200
1201             fscal            = felec;
1202
1203             /* Calculate temporary vectorial force */
1204             tx               = _mm256_mul_pd(fscal,dx12);
1205             ty               = _mm256_mul_pd(fscal,dy12);
1206             tz               = _mm256_mul_pd(fscal,dz12);
1207
1208             /* Update vectorial force */
1209             fix1             = _mm256_add_pd(fix1,tx);
1210             fiy1             = _mm256_add_pd(fiy1,ty);
1211             fiz1             = _mm256_add_pd(fiz1,tz);
1212
1213             fjx2             = _mm256_add_pd(fjx2,tx);
1214             fjy2             = _mm256_add_pd(fjy2,ty);
1215             fjz2             = _mm256_add_pd(fjz2,tz);
1216
1217             /**************************
1218              * CALCULATE INTERACTIONS *
1219              **************************/
1220
1221             /* REACTION-FIELD ELECTROSTATICS */
1222             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
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             /* REACTION-FIELD ELECTROSTATICS */
1245             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1246
1247             fscal            = felec;
1248
1249             /* Calculate temporary vectorial force */
1250             tx               = _mm256_mul_pd(fscal,dx21);
1251             ty               = _mm256_mul_pd(fscal,dy21);
1252             tz               = _mm256_mul_pd(fscal,dz21);
1253
1254             /* Update vectorial force */
1255             fix2             = _mm256_add_pd(fix2,tx);
1256             fiy2             = _mm256_add_pd(fiy2,ty);
1257             fiz2             = _mm256_add_pd(fiz2,tz);
1258
1259             fjx1             = _mm256_add_pd(fjx1,tx);
1260             fjy1             = _mm256_add_pd(fjy1,ty);
1261             fjz1             = _mm256_add_pd(fjz1,tz);
1262
1263             /**************************
1264              * CALCULATE INTERACTIONS *
1265              **************************/
1266
1267             /* REACTION-FIELD ELECTROSTATICS */
1268             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1269
1270             fscal            = felec;
1271
1272             /* Calculate temporary vectorial force */
1273             tx               = _mm256_mul_pd(fscal,dx22);
1274             ty               = _mm256_mul_pd(fscal,dy22);
1275             tz               = _mm256_mul_pd(fscal,dz22);
1276
1277             /* Update vectorial force */
1278             fix2             = _mm256_add_pd(fix2,tx);
1279             fiy2             = _mm256_add_pd(fiy2,ty);
1280             fiz2             = _mm256_add_pd(fiz2,tz);
1281
1282             fjx2             = _mm256_add_pd(fjx2,tx);
1283             fjy2             = _mm256_add_pd(fjy2,ty);
1284             fjz2             = _mm256_add_pd(fjz2,tz);
1285
1286             /**************************
1287              * CALCULATE INTERACTIONS *
1288              **************************/
1289
1290             /* REACTION-FIELD ELECTROSTATICS */
1291             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1292
1293             fscal            = felec;
1294
1295             /* Calculate temporary vectorial force */
1296             tx               = _mm256_mul_pd(fscal,dx23);
1297             ty               = _mm256_mul_pd(fscal,dy23);
1298             tz               = _mm256_mul_pd(fscal,dz23);
1299
1300             /* Update vectorial force */
1301             fix2             = _mm256_add_pd(fix2,tx);
1302             fiy2             = _mm256_add_pd(fiy2,ty);
1303             fiz2             = _mm256_add_pd(fiz2,tz);
1304
1305             fjx3             = _mm256_add_pd(fjx3,tx);
1306             fjy3             = _mm256_add_pd(fjy3,ty);
1307             fjz3             = _mm256_add_pd(fjz3,tz);
1308
1309             /**************************
1310              * CALCULATE INTERACTIONS *
1311              **************************/
1312
1313             /* REACTION-FIELD ELECTROSTATICS */
1314             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1315
1316             fscal            = felec;
1317
1318             /* Calculate temporary vectorial force */
1319             tx               = _mm256_mul_pd(fscal,dx31);
1320             ty               = _mm256_mul_pd(fscal,dy31);
1321             tz               = _mm256_mul_pd(fscal,dz31);
1322
1323             /* Update vectorial force */
1324             fix3             = _mm256_add_pd(fix3,tx);
1325             fiy3             = _mm256_add_pd(fiy3,ty);
1326             fiz3             = _mm256_add_pd(fiz3,tz);
1327
1328             fjx1             = _mm256_add_pd(fjx1,tx);
1329             fjy1             = _mm256_add_pd(fjy1,ty);
1330             fjz1             = _mm256_add_pd(fjz1,tz);
1331
1332             /**************************
1333              * CALCULATE INTERACTIONS *
1334              **************************/
1335
1336             /* REACTION-FIELD ELECTROSTATICS */
1337             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1338
1339             fscal            = felec;
1340
1341             /* Calculate temporary vectorial force */
1342             tx               = _mm256_mul_pd(fscal,dx32);
1343             ty               = _mm256_mul_pd(fscal,dy32);
1344             tz               = _mm256_mul_pd(fscal,dz32);
1345
1346             /* Update vectorial force */
1347             fix3             = _mm256_add_pd(fix3,tx);
1348             fiy3             = _mm256_add_pd(fiy3,ty);
1349             fiz3             = _mm256_add_pd(fiz3,tz);
1350
1351             fjx2             = _mm256_add_pd(fjx2,tx);
1352             fjy2             = _mm256_add_pd(fjy2,ty);
1353             fjz2             = _mm256_add_pd(fjz2,tz);
1354
1355             /**************************
1356              * CALCULATE INTERACTIONS *
1357              **************************/
1358
1359             /* REACTION-FIELD ELECTROSTATICS */
1360             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1361
1362             fscal            = felec;
1363
1364             /* Calculate temporary vectorial force */
1365             tx               = _mm256_mul_pd(fscal,dx33);
1366             ty               = _mm256_mul_pd(fscal,dy33);
1367             tz               = _mm256_mul_pd(fscal,dz33);
1368
1369             /* Update vectorial force */
1370             fix3             = _mm256_add_pd(fix3,tx);
1371             fiy3             = _mm256_add_pd(fiy3,ty);
1372             fiz3             = _mm256_add_pd(fiz3,tz);
1373
1374             fjx3             = _mm256_add_pd(fjx3,tx);
1375             fjy3             = _mm256_add_pd(fjy3,ty);
1376             fjz3             = _mm256_add_pd(fjz3,tz);
1377
1378             fjptrA             = f+j_coord_offsetA;
1379             fjptrB             = f+j_coord_offsetB;
1380             fjptrC             = f+j_coord_offsetC;
1381             fjptrD             = f+j_coord_offsetD;
1382
1383             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1384                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1385
1386             /* Inner loop uses 243 flops */
1387         }
1388
1389         if(jidx<j_index_end)
1390         {
1391
1392             /* Get j neighbor index, and coordinate index */
1393             jnrlistA         = jjnr[jidx];
1394             jnrlistB         = jjnr[jidx+1];
1395             jnrlistC         = jjnr[jidx+2];
1396             jnrlistD         = jjnr[jidx+3];
1397             /* Sign of each element will be negative for non-real atoms.
1398              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1399              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1400              */
1401             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1402
1403             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1404             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1405             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1406
1407             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1408             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1409             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1410             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1411             j_coord_offsetA  = DIM*jnrA;
1412             j_coord_offsetB  = DIM*jnrB;
1413             j_coord_offsetC  = DIM*jnrC;
1414             j_coord_offsetD  = DIM*jnrD;
1415
1416             /* load j atom coordinates */
1417             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1418                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1419                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1420
1421             /* Calculate displacement vector */
1422             dx11             = _mm256_sub_pd(ix1,jx1);
1423             dy11             = _mm256_sub_pd(iy1,jy1);
1424             dz11             = _mm256_sub_pd(iz1,jz1);
1425             dx12             = _mm256_sub_pd(ix1,jx2);
1426             dy12             = _mm256_sub_pd(iy1,jy2);
1427             dz12             = _mm256_sub_pd(iz1,jz2);
1428             dx13             = _mm256_sub_pd(ix1,jx3);
1429             dy13             = _mm256_sub_pd(iy1,jy3);
1430             dz13             = _mm256_sub_pd(iz1,jz3);
1431             dx21             = _mm256_sub_pd(ix2,jx1);
1432             dy21             = _mm256_sub_pd(iy2,jy1);
1433             dz21             = _mm256_sub_pd(iz2,jz1);
1434             dx22             = _mm256_sub_pd(ix2,jx2);
1435             dy22             = _mm256_sub_pd(iy2,jy2);
1436             dz22             = _mm256_sub_pd(iz2,jz2);
1437             dx23             = _mm256_sub_pd(ix2,jx3);
1438             dy23             = _mm256_sub_pd(iy2,jy3);
1439             dz23             = _mm256_sub_pd(iz2,jz3);
1440             dx31             = _mm256_sub_pd(ix3,jx1);
1441             dy31             = _mm256_sub_pd(iy3,jy1);
1442             dz31             = _mm256_sub_pd(iz3,jz1);
1443             dx32             = _mm256_sub_pd(ix3,jx2);
1444             dy32             = _mm256_sub_pd(iy3,jy2);
1445             dz32             = _mm256_sub_pd(iz3,jz2);
1446             dx33             = _mm256_sub_pd(ix3,jx3);
1447             dy33             = _mm256_sub_pd(iy3,jy3);
1448             dz33             = _mm256_sub_pd(iz3,jz3);
1449
1450             /* Calculate squared distance and things based on it */
1451             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1452             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1453             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1454             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1455             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1456             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1457             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1458             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1459             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1460
1461             rinv11           = avx256_invsqrt_d(rsq11);
1462             rinv12           = avx256_invsqrt_d(rsq12);
1463             rinv13           = avx256_invsqrt_d(rsq13);
1464             rinv21           = avx256_invsqrt_d(rsq21);
1465             rinv22           = avx256_invsqrt_d(rsq22);
1466             rinv23           = avx256_invsqrt_d(rsq23);
1467             rinv31           = avx256_invsqrt_d(rsq31);
1468             rinv32           = avx256_invsqrt_d(rsq32);
1469             rinv33           = avx256_invsqrt_d(rsq33);
1470
1471             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1472             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1473             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1474             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1475             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1476             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1477             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1478             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1479             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1480
1481             fjx1             = _mm256_setzero_pd();
1482             fjy1             = _mm256_setzero_pd();
1483             fjz1             = _mm256_setzero_pd();
1484             fjx2             = _mm256_setzero_pd();
1485             fjy2             = _mm256_setzero_pd();
1486             fjz2             = _mm256_setzero_pd();
1487             fjx3             = _mm256_setzero_pd();
1488             fjy3             = _mm256_setzero_pd();
1489             fjz3             = _mm256_setzero_pd();
1490
1491             /**************************
1492              * CALCULATE INTERACTIONS *
1493              **************************/
1494
1495             /* REACTION-FIELD ELECTROSTATICS */
1496             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1497
1498             fscal            = felec;
1499
1500             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1501
1502             /* Calculate temporary vectorial force */
1503             tx               = _mm256_mul_pd(fscal,dx11);
1504             ty               = _mm256_mul_pd(fscal,dy11);
1505             tz               = _mm256_mul_pd(fscal,dz11);
1506
1507             /* Update vectorial force */
1508             fix1             = _mm256_add_pd(fix1,tx);
1509             fiy1             = _mm256_add_pd(fiy1,ty);
1510             fiz1             = _mm256_add_pd(fiz1,tz);
1511
1512             fjx1             = _mm256_add_pd(fjx1,tx);
1513             fjy1             = _mm256_add_pd(fjy1,ty);
1514             fjz1             = _mm256_add_pd(fjz1,tz);
1515
1516             /**************************
1517              * CALCULATE INTERACTIONS *
1518              **************************/
1519
1520             /* REACTION-FIELD ELECTROSTATICS */
1521             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1522
1523             fscal            = felec;
1524
1525             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1526
1527             /* Calculate temporary vectorial force */
1528             tx               = _mm256_mul_pd(fscal,dx12);
1529             ty               = _mm256_mul_pd(fscal,dy12);
1530             tz               = _mm256_mul_pd(fscal,dz12);
1531
1532             /* Update vectorial force */
1533             fix1             = _mm256_add_pd(fix1,tx);
1534             fiy1             = _mm256_add_pd(fiy1,ty);
1535             fiz1             = _mm256_add_pd(fiz1,tz);
1536
1537             fjx2             = _mm256_add_pd(fjx2,tx);
1538             fjy2             = _mm256_add_pd(fjy2,ty);
1539             fjz2             = _mm256_add_pd(fjz2,tz);
1540
1541             /**************************
1542              * CALCULATE INTERACTIONS *
1543              **************************/
1544
1545             /* REACTION-FIELD ELECTROSTATICS */
1546             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
1547
1548             fscal            = felec;
1549
1550             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1551
1552             /* Calculate temporary vectorial force */
1553             tx               = _mm256_mul_pd(fscal,dx13);
1554             ty               = _mm256_mul_pd(fscal,dy13);
1555             tz               = _mm256_mul_pd(fscal,dz13);
1556
1557             /* Update vectorial force */
1558             fix1             = _mm256_add_pd(fix1,tx);
1559             fiy1             = _mm256_add_pd(fiy1,ty);
1560             fiz1             = _mm256_add_pd(fiz1,tz);
1561
1562             fjx3             = _mm256_add_pd(fjx3,tx);
1563             fjy3             = _mm256_add_pd(fjy3,ty);
1564             fjz3             = _mm256_add_pd(fjz3,tz);
1565
1566             /**************************
1567              * CALCULATE INTERACTIONS *
1568              **************************/
1569
1570             /* REACTION-FIELD ELECTROSTATICS */
1571             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1572
1573             fscal            = felec;
1574
1575             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1576
1577             /* Calculate temporary vectorial force */
1578             tx               = _mm256_mul_pd(fscal,dx21);
1579             ty               = _mm256_mul_pd(fscal,dy21);
1580             tz               = _mm256_mul_pd(fscal,dz21);
1581
1582             /* Update vectorial force */
1583             fix2             = _mm256_add_pd(fix2,tx);
1584             fiy2             = _mm256_add_pd(fiy2,ty);
1585             fiz2             = _mm256_add_pd(fiz2,tz);
1586
1587             fjx1             = _mm256_add_pd(fjx1,tx);
1588             fjy1             = _mm256_add_pd(fjy1,ty);
1589             fjz1             = _mm256_add_pd(fjz1,tz);
1590
1591             /**************************
1592              * CALCULATE INTERACTIONS *
1593              **************************/
1594
1595             /* REACTION-FIELD ELECTROSTATICS */
1596             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1597
1598             fscal            = felec;
1599
1600             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1601
1602             /* Calculate temporary vectorial force */
1603             tx               = _mm256_mul_pd(fscal,dx22);
1604             ty               = _mm256_mul_pd(fscal,dy22);
1605             tz               = _mm256_mul_pd(fscal,dz22);
1606
1607             /* Update vectorial force */
1608             fix2             = _mm256_add_pd(fix2,tx);
1609             fiy2             = _mm256_add_pd(fiy2,ty);
1610             fiz2             = _mm256_add_pd(fiz2,tz);
1611
1612             fjx2             = _mm256_add_pd(fjx2,tx);
1613             fjy2             = _mm256_add_pd(fjy2,ty);
1614             fjz2             = _mm256_add_pd(fjz2,tz);
1615
1616             /**************************
1617              * CALCULATE INTERACTIONS *
1618              **************************/
1619
1620             /* REACTION-FIELD ELECTROSTATICS */
1621             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1622
1623             fscal            = felec;
1624
1625             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1626
1627             /* Calculate temporary vectorial force */
1628             tx               = _mm256_mul_pd(fscal,dx23);
1629             ty               = _mm256_mul_pd(fscal,dy23);
1630             tz               = _mm256_mul_pd(fscal,dz23);
1631
1632             /* Update vectorial force */
1633             fix2             = _mm256_add_pd(fix2,tx);
1634             fiy2             = _mm256_add_pd(fiy2,ty);
1635             fiz2             = _mm256_add_pd(fiz2,tz);
1636
1637             fjx3             = _mm256_add_pd(fjx3,tx);
1638             fjy3             = _mm256_add_pd(fjy3,ty);
1639             fjz3             = _mm256_add_pd(fjz3,tz);
1640
1641             /**************************
1642              * CALCULATE INTERACTIONS *
1643              **************************/
1644
1645             /* REACTION-FIELD ELECTROSTATICS */
1646             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1647
1648             fscal            = felec;
1649
1650             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1651
1652             /* Calculate temporary vectorial force */
1653             tx               = _mm256_mul_pd(fscal,dx31);
1654             ty               = _mm256_mul_pd(fscal,dy31);
1655             tz               = _mm256_mul_pd(fscal,dz31);
1656
1657             /* Update vectorial force */
1658             fix3             = _mm256_add_pd(fix3,tx);
1659             fiy3             = _mm256_add_pd(fiy3,ty);
1660             fiz3             = _mm256_add_pd(fiz3,tz);
1661
1662             fjx1             = _mm256_add_pd(fjx1,tx);
1663             fjy1             = _mm256_add_pd(fjy1,ty);
1664             fjz1             = _mm256_add_pd(fjz1,tz);
1665
1666             /**************************
1667              * CALCULATE INTERACTIONS *
1668              **************************/
1669
1670             /* REACTION-FIELD ELECTROSTATICS */
1671             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1672
1673             fscal            = felec;
1674
1675             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1676
1677             /* Calculate temporary vectorial force */
1678             tx               = _mm256_mul_pd(fscal,dx32);
1679             ty               = _mm256_mul_pd(fscal,dy32);
1680             tz               = _mm256_mul_pd(fscal,dz32);
1681
1682             /* Update vectorial force */
1683             fix3             = _mm256_add_pd(fix3,tx);
1684             fiy3             = _mm256_add_pd(fiy3,ty);
1685             fiz3             = _mm256_add_pd(fiz3,tz);
1686
1687             fjx2             = _mm256_add_pd(fjx2,tx);
1688             fjy2             = _mm256_add_pd(fjy2,ty);
1689             fjz2             = _mm256_add_pd(fjz2,tz);
1690
1691             /**************************
1692              * CALCULATE INTERACTIONS *
1693              **************************/
1694
1695             /* REACTION-FIELD ELECTROSTATICS */
1696             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1697
1698             fscal            = felec;
1699
1700             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1701
1702             /* Calculate temporary vectorial force */
1703             tx               = _mm256_mul_pd(fscal,dx33);
1704             ty               = _mm256_mul_pd(fscal,dy33);
1705             tz               = _mm256_mul_pd(fscal,dz33);
1706
1707             /* Update vectorial force */
1708             fix3             = _mm256_add_pd(fix3,tx);
1709             fiy3             = _mm256_add_pd(fiy3,ty);
1710             fiz3             = _mm256_add_pd(fiz3,tz);
1711
1712             fjx3             = _mm256_add_pd(fjx3,tx);
1713             fjy3             = _mm256_add_pd(fjy3,ty);
1714             fjz3             = _mm256_add_pd(fjz3,tz);
1715
1716             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1717             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1718             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1719             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1720
1721             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1722                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1723
1724             /* Inner loop uses 243 flops */
1725         }
1726
1727         /* End of innermost loop */
1728
1729         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1730                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1731
1732         /* Increment number of inner iterations */
1733         inneriter                  += j_index_end - j_index_start;
1734
1735         /* Outer loop uses 18 flops */
1736     }
1737
1738     /* Increment number of outer iterations */
1739     outeriter        += nri;
1740
1741     /* Update outer/inner flops */
1742
1743     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*243);
1744 }