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, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4W4_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Water4
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwNone_GeomW4W4_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr1;
  85     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     real *           vdwioffsetptr2;
  87     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     real *           vdwioffsetptr3;
  89     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  91     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  92     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  93     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  94     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  95     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  96     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  97     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  98     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  99     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 100     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 101     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 102     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 103     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 104     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 105     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 106     real             *charge;
 107     __m256d          dummy_mask,cutoff_mask;
 108     __m128           tmpmask0,tmpmask1;
 109     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 110     __m256d          one     = _mm256_set1_pd(1.0);
 111     __m256d          two     = _mm256_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm256_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     krf              = _mm256_set1_pd(fr->ic->k_rf);
 126     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 127     crf              = _mm256_set1_pd(fr->ic->c_rf);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 132     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 133     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 134
 135     jq1              = _mm256_set1_pd(charge[inr+1]);
 136     jq2              = _mm256_set1_pd(charge[inr+2]);
 137     jq3              = _mm256_set1_pd(charge[inr+3]);
 138     qq11             = _mm256_mul_pd(iq1,jq1);
 139     qq12             = _mm256_mul_pd(iq1,jq2);
 140     qq13             = _mm256_mul_pd(iq1,jq3);
 141     qq21             = _mm256_mul_pd(iq2,jq1);
 142     qq22             = _mm256_mul_pd(iq2,jq2);
 143     qq23             = _mm256_mul_pd(iq2,jq3);
 144     qq31             = _mm256_mul_pd(iq3,jq1);
 145     qq32             = _mm256_mul_pd(iq3,jq2);
 146     qq33             = _mm256_mul_pd(iq3,jq3);
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = jnrC = jnrD = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152     j_coord_offsetC = 0;
 153     j_coord_offsetD = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 179                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 180
 181         fix1             = _mm256_setzero_pd();
 182         fiy1             = _mm256_setzero_pd();
 183         fiz1             = _mm256_setzero_pd();
 184         fix2             = _mm256_setzero_pd();
 185         fiy2             = _mm256_setzero_pd();
 186         fiz2             = _mm256_setzero_pd();
 187         fix3             = _mm256_setzero_pd();
 188         fiy3             = _mm256_setzero_pd();
 189         fiz3             = _mm256_setzero_pd();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm256_setzero_pd();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207
 208             /* load j atom coordinates */
 209             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 210                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 211                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 212
 213             /* Calculate displacement vector */
 214             dx11             = _mm256_sub_pd(ix1,jx1);
 215             dy11             = _mm256_sub_pd(iy1,jy1);
 216             dz11             = _mm256_sub_pd(iz1,jz1);
 217             dx12             = _mm256_sub_pd(ix1,jx2);
 218             dy12             = _mm256_sub_pd(iy1,jy2);
 219             dz12             = _mm256_sub_pd(iz1,jz2);
 220             dx13             = _mm256_sub_pd(ix1,jx3);
 221             dy13             = _mm256_sub_pd(iy1,jy3);
 222             dz13             = _mm256_sub_pd(iz1,jz3);
 223             dx21             = _mm256_sub_pd(ix2,jx1);
 224             dy21             = _mm256_sub_pd(iy2,jy1);
 225             dz21             = _mm256_sub_pd(iz2,jz1);
 226             dx22             = _mm256_sub_pd(ix2,jx2);
 227             dy22             = _mm256_sub_pd(iy2,jy2);
 228             dz22             = _mm256_sub_pd(iz2,jz2);
 229             dx23             = _mm256_sub_pd(ix2,jx3);
 230             dy23             = _mm256_sub_pd(iy2,jy3);
 231             dz23             = _mm256_sub_pd(iz2,jz3);
 232             dx31             = _mm256_sub_pd(ix3,jx1);
 233             dy31             = _mm256_sub_pd(iy3,jy1);
 234             dz31             = _mm256_sub_pd(iz3,jz1);
 235             dx32             = _mm256_sub_pd(ix3,jx2);
 236             dy32             = _mm256_sub_pd(iy3,jy2);
 237             dz32             = _mm256_sub_pd(iz3,jz2);
 238             dx33             = _mm256_sub_pd(ix3,jx3);
 239             dy33             = _mm256_sub_pd(iy3,jy3);
 240             dz33             = _mm256_sub_pd(iz3,jz3);
 241
 242             /* Calculate squared distance and things based on it */
 243             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 244             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 245             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 246             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 247             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 248             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 249             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 250             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 251             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 252
 253             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 254             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 255             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 256             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 257             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 258             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 259             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 260             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 261             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 262
 263             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 264             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 265             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 266             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 267             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 268             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 269             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 270             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 271             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 272
 273             fjx1             = _mm256_setzero_pd();
 274             fjy1             = _mm256_setzero_pd();
 275             fjz1             = _mm256_setzero_pd();
 276             fjx2             = _mm256_setzero_pd();
 277             fjy2             = _mm256_setzero_pd();
 278             fjz2             = _mm256_setzero_pd();
 279             fjx3             = _mm256_setzero_pd();
 280             fjy3             = _mm256_setzero_pd();
 281             fjz3             = _mm256_setzero_pd();
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             /* REACTION-FIELD ELECTROSTATICS */
 288             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 289             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 290
 291             /* Update potential sum for this i atom from the interaction with this j atom. */
 292             velecsum         = _mm256_add_pd(velecsum,velec);
 293
 294             fscal            = felec;
 295
 296             /* Calculate temporary vectorial force */
 297             tx               = _mm256_mul_pd(fscal,dx11);
 298             ty               = _mm256_mul_pd(fscal,dy11);
 299             tz               = _mm256_mul_pd(fscal,dz11);
 300
 301             /* Update vectorial force */
 302             fix1             = _mm256_add_pd(fix1,tx);
 303             fiy1             = _mm256_add_pd(fiy1,ty);
 304             fiz1             = _mm256_add_pd(fiz1,tz);
 305
 306             fjx1             = _mm256_add_pd(fjx1,tx);
 307             fjy1             = _mm256_add_pd(fjy1,ty);
 308             fjz1             = _mm256_add_pd(fjz1,tz);
 309
 310             /**************************
 311              * CALCULATE INTERACTIONS *
 312              **************************/
 313
 314             /* REACTION-FIELD ELECTROSTATICS */
 315             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 316             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velecsum         = _mm256_add_pd(velecsum,velec);
 320
 321             fscal            = felec;
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm256_mul_pd(fscal,dx12);
 325             ty               = _mm256_mul_pd(fscal,dy12);
 326             tz               = _mm256_mul_pd(fscal,dz12);
 327
 328             /* Update vectorial force */
 329             fix1             = _mm256_add_pd(fix1,tx);
 330             fiy1             = _mm256_add_pd(fiy1,ty);
 331             fiz1             = _mm256_add_pd(fiz1,tz);
 332
 333             fjx2             = _mm256_add_pd(fjx2,tx);
 334             fjy2             = _mm256_add_pd(fjy2,ty);
 335             fjz2             = _mm256_add_pd(fjz2,tz);
 336
 337             /**************************
 338              * CALCULATE INTERACTIONS *
 339              **************************/
 340
 341             /* REACTION-FIELD ELECTROSTATICS */
 342             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
 343             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velecsum         = _mm256_add_pd(velecsum,velec);
 347
 348             fscal            = felec;
 349
 350             /* Calculate temporary vectorial force */
 351             tx               = _mm256_mul_pd(fscal,dx13);
 352             ty               = _mm256_mul_pd(fscal,dy13);
 353             tz               = _mm256_mul_pd(fscal,dz13);
 354
 355             /* Update vectorial force */
 356             fix1             = _mm256_add_pd(fix1,tx);
 357             fiy1             = _mm256_add_pd(fiy1,ty);
 358             fiz1             = _mm256_add_pd(fiz1,tz);
 359
 360             fjx3             = _mm256_add_pd(fjx3,tx);
 361             fjy3             = _mm256_add_pd(fjy3,ty);
 362             fjz3             = _mm256_add_pd(fjz3,tz);
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             /* REACTION-FIELD ELECTROSTATICS */
 369             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 370             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velecsum         = _mm256_add_pd(velecsum,velec);
 374
 375             fscal            = felec;
 376
 377             /* Calculate temporary vectorial force */
 378             tx               = _mm256_mul_pd(fscal,dx21);
 379             ty               = _mm256_mul_pd(fscal,dy21);
 380             tz               = _mm256_mul_pd(fscal,dz21);
 381
 382             /* Update vectorial force */
 383             fix2             = _mm256_add_pd(fix2,tx);
 384             fiy2             = _mm256_add_pd(fiy2,ty);
 385             fiz2             = _mm256_add_pd(fiz2,tz);
 386
 387             fjx1             = _mm256_add_pd(fjx1,tx);
 388             fjy1             = _mm256_add_pd(fjy1,ty);
 389             fjz1             = _mm256_add_pd(fjz1,tz);
 390
 391             /**************************
 392              * CALCULATE INTERACTIONS *
 393              **************************/
 394
 395             /* REACTION-FIELD ELECTROSTATICS */
 396             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 397             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 398
 399             /* Update potential sum for this i atom from the interaction with this j atom. */
 400             velecsum         = _mm256_add_pd(velecsum,velec);
 401
 402             fscal            = felec;
 403
 404             /* Calculate temporary vectorial force */
 405             tx               = _mm256_mul_pd(fscal,dx22);
 406             ty               = _mm256_mul_pd(fscal,dy22);
 407             tz               = _mm256_mul_pd(fscal,dz22);
 408
 409             /* Update vectorial force */
 410             fix2             = _mm256_add_pd(fix2,tx);
 411             fiy2             = _mm256_add_pd(fiy2,ty);
 412             fiz2             = _mm256_add_pd(fiz2,tz);
 413
 414             fjx2             = _mm256_add_pd(fjx2,tx);
 415             fjy2             = _mm256_add_pd(fjy2,ty);
 416             fjz2             = _mm256_add_pd(fjz2,tz);
 417
 418             /**************************
 419              * CALCULATE INTERACTIONS *
 420              **************************/
 421
 422             /* REACTION-FIELD ELECTROSTATICS */
 423             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
 424             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
 425
 426             /* Update potential sum for this i atom from the interaction with this j atom. */
 427             velecsum         = _mm256_add_pd(velecsum,velec);
 428
 429             fscal            = felec;
 430
 431             /* Calculate temporary vectorial force */
 432             tx               = _mm256_mul_pd(fscal,dx23);
 433             ty               = _mm256_mul_pd(fscal,dy23);
 434             tz               = _mm256_mul_pd(fscal,dz23);
 435
 436             /* Update vectorial force */
 437             fix2             = _mm256_add_pd(fix2,tx);
 438             fiy2             = _mm256_add_pd(fiy2,ty);
 439             fiz2             = _mm256_add_pd(fiz2,tz);
 440
 441             fjx3             = _mm256_add_pd(fjx3,tx);
 442             fjy3             = _mm256_add_pd(fjy3,ty);
 443             fjz3             = _mm256_add_pd(fjz3,tz);
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             /* REACTION-FIELD ELECTROSTATICS */
 450             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
 451             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
 452
 453             /* Update potential sum for this i atom from the interaction with this j atom. */
 454             velecsum         = _mm256_add_pd(velecsum,velec);
 455
 456             fscal            = felec;
 457
 458             /* Calculate temporary vectorial force */
 459             tx               = _mm256_mul_pd(fscal,dx31);
 460             ty               = _mm256_mul_pd(fscal,dy31);
 461             tz               = _mm256_mul_pd(fscal,dz31);
 462
 463             /* Update vectorial force */
 464             fix3             = _mm256_add_pd(fix3,tx);
 465             fiy3             = _mm256_add_pd(fiy3,ty);
 466             fiz3             = _mm256_add_pd(fiz3,tz);
 467
 468             fjx1             = _mm256_add_pd(fjx1,tx);
 469             fjy1             = _mm256_add_pd(fjy1,ty);
 470             fjz1             = _mm256_add_pd(fjz1,tz);
 471
 472             /**************************
 473              * CALCULATE INTERACTIONS *
 474              **************************/
 475
 476             /* REACTION-FIELD ELECTROSTATICS */
 477             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
 478             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
 479
 480             /* Update potential sum for this i atom from the interaction with this j atom. */
 481             velecsum         = _mm256_add_pd(velecsum,velec);
 482
 483             fscal            = felec;
 484
 485             /* Calculate temporary vectorial force */
 486             tx               = _mm256_mul_pd(fscal,dx32);
 487             ty               = _mm256_mul_pd(fscal,dy32);
 488             tz               = _mm256_mul_pd(fscal,dz32);
 489
 490             /* Update vectorial force */
 491             fix3             = _mm256_add_pd(fix3,tx);
 492             fiy3             = _mm256_add_pd(fiy3,ty);
 493             fiz3             = _mm256_add_pd(fiz3,tz);
 494
 495             fjx2             = _mm256_add_pd(fjx2,tx);
 496             fjy2             = _mm256_add_pd(fjy2,ty);
 497             fjz2             = _mm256_add_pd(fjz2,tz);
 498
 499             /**************************
 500              * CALCULATE INTERACTIONS *
 501              **************************/
 502
 503             /* REACTION-FIELD ELECTROSTATICS */
 504             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
 505             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
 506
 507             /* Update potential sum for this i atom from the interaction with this j atom. */
 508             velecsum         = _mm256_add_pd(velecsum,velec);
 509
 510             fscal            = felec;
 511
 512             /* Calculate temporary vectorial force */
 513             tx               = _mm256_mul_pd(fscal,dx33);
 514             ty               = _mm256_mul_pd(fscal,dy33);
 515             tz               = _mm256_mul_pd(fscal,dz33);
 516
 517             /* Update vectorial force */
 518             fix3             = _mm256_add_pd(fix3,tx);
 519             fiy3             = _mm256_add_pd(fiy3,ty);
 520             fiz3             = _mm256_add_pd(fiz3,tz);
 521
 522             fjx3             = _mm256_add_pd(fjx3,tx);
 523             fjy3             = _mm256_add_pd(fjy3,ty);
 524             fjz3             = _mm256_add_pd(fjz3,tz);
 525
 526             fjptrA             = f+j_coord_offsetA;
 527             fjptrB             = f+j_coord_offsetB;
 528             fjptrC             = f+j_coord_offsetC;
 529             fjptrD             = f+j_coord_offsetD;
 530
 531             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 532                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 533
 534             /* Inner loop uses 288 flops */
 535         }
 536
 537         if(jidx<j_index_end)
 538         {
 539
 540             /* Get j neighbor index, and coordinate index */
 541             jnrlistA         = jjnr[jidx];
 542             jnrlistB         = jjnr[jidx+1];
 543             jnrlistC         = jjnr[jidx+2];
 544             jnrlistD         = jjnr[jidx+3];
 545             /* Sign of each element will be negative for non-real atoms.
 546              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 547              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 548              */
 549             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 550
 551             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 552             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 553             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 554
 555             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 556             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 557             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 558             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 559             j_coord_offsetA  = DIM*jnrA;
 560             j_coord_offsetB  = DIM*jnrB;
 561             j_coord_offsetC  = DIM*jnrC;
 562             j_coord_offsetD  = DIM*jnrD;
 563
 564             /* load j atom coordinates */
 565             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 566                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 567                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 568
 569             /* Calculate displacement vector */
 570             dx11             = _mm256_sub_pd(ix1,jx1);
 571             dy11             = _mm256_sub_pd(iy1,jy1);
 572             dz11             = _mm256_sub_pd(iz1,jz1);
 573             dx12             = _mm256_sub_pd(ix1,jx2);
 574             dy12             = _mm256_sub_pd(iy1,jy2);
 575             dz12             = _mm256_sub_pd(iz1,jz2);
 576             dx13             = _mm256_sub_pd(ix1,jx3);
 577             dy13             = _mm256_sub_pd(iy1,jy3);
 578             dz13             = _mm256_sub_pd(iz1,jz3);
 579             dx21             = _mm256_sub_pd(ix2,jx1);
 580             dy21             = _mm256_sub_pd(iy2,jy1);
 581             dz21             = _mm256_sub_pd(iz2,jz1);
 582             dx22             = _mm256_sub_pd(ix2,jx2);
 583             dy22             = _mm256_sub_pd(iy2,jy2);
 584             dz22             = _mm256_sub_pd(iz2,jz2);
 585             dx23             = _mm256_sub_pd(ix2,jx3);
 586             dy23             = _mm256_sub_pd(iy2,jy3);
 587             dz23             = _mm256_sub_pd(iz2,jz3);
 588             dx31             = _mm256_sub_pd(ix3,jx1);
 589             dy31             = _mm256_sub_pd(iy3,jy1);
 590             dz31             = _mm256_sub_pd(iz3,jz1);
 591             dx32             = _mm256_sub_pd(ix3,jx2);
 592             dy32             = _mm256_sub_pd(iy3,jy2);
 593             dz32             = _mm256_sub_pd(iz3,jz2);
 594             dx33             = _mm256_sub_pd(ix3,jx3);
 595             dy33             = _mm256_sub_pd(iy3,jy3);
 596             dz33             = _mm256_sub_pd(iz3,jz3);
 597
 598             /* Calculate squared distance and things based on it */
 599             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 600             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 601             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 602             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 603             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 604             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 605             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 606             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 607             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 608
 609             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 610             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 611             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 612             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 613             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 614             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 615             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 616             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 617             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 618
 619             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 620             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 621             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 622             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 623             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 624             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 625             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 626             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 627             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 628
 629             fjx1             = _mm256_setzero_pd();
 630             fjy1             = _mm256_setzero_pd();
 631             fjz1             = _mm256_setzero_pd();
 632             fjx2             = _mm256_setzero_pd();
 633             fjy2             = _mm256_setzero_pd();
 634             fjz2             = _mm256_setzero_pd();
 635             fjx3             = _mm256_setzero_pd();
 636             fjy3             = _mm256_setzero_pd();
 637             fjz3             = _mm256_setzero_pd();
 638
 639             /**************************
 640              * CALCULATE INTERACTIONS *
 641              **************************/
 642
 643             /* REACTION-FIELD ELECTROSTATICS */
 644             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 645             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 646
 647             /* Update potential sum for this i atom from the interaction with this j atom. */
 648             velec            = _mm256_andnot_pd(dummy_mask,velec);
 649             velecsum         = _mm256_add_pd(velecsum,velec);
 650
 651             fscal            = felec;
 652
 653             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 654
 655             /* Calculate temporary vectorial force */
 656             tx               = _mm256_mul_pd(fscal,dx11);
 657             ty               = _mm256_mul_pd(fscal,dy11);
 658             tz               = _mm256_mul_pd(fscal,dz11);
 659
 660             /* Update vectorial force */
 661             fix1             = _mm256_add_pd(fix1,tx);
 662             fiy1             = _mm256_add_pd(fiy1,ty);
 663             fiz1             = _mm256_add_pd(fiz1,tz);
 664
 665             fjx1             = _mm256_add_pd(fjx1,tx);
 666             fjy1             = _mm256_add_pd(fjy1,ty);
 667             fjz1             = _mm256_add_pd(fjz1,tz);
 668
 669             /**************************
 670              * CALCULATE INTERACTIONS *
 671              **************************/
 672
 673             /* REACTION-FIELD ELECTROSTATICS */
 674             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 675             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 676
 677             /* Update potential sum for this i atom from the interaction with this j atom. */
 678             velec            = _mm256_andnot_pd(dummy_mask,velec);
 679             velecsum         = _mm256_add_pd(velecsum,velec);
 680
 681             fscal            = felec;
 682
 683             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 684
 685             /* Calculate temporary vectorial force */
 686             tx               = _mm256_mul_pd(fscal,dx12);
 687             ty               = _mm256_mul_pd(fscal,dy12);
 688             tz               = _mm256_mul_pd(fscal,dz12);
 689
 690             /* Update vectorial force */
 691             fix1             = _mm256_add_pd(fix1,tx);
 692             fiy1             = _mm256_add_pd(fiy1,ty);
 693             fiz1             = _mm256_add_pd(fiz1,tz);
 694
 695             fjx2             = _mm256_add_pd(fjx2,tx);
 696             fjy2             = _mm256_add_pd(fjy2,ty);
 697             fjz2             = _mm256_add_pd(fjz2,tz);
 698
 699             /**************************
 700              * CALCULATE INTERACTIONS *
 701              **************************/
 702
 703             /* REACTION-FIELD ELECTROSTATICS */
 704             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
 705             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
 706
 707             /* Update potential sum for this i atom from the interaction with this j atom. */
 708             velec            = _mm256_andnot_pd(dummy_mask,velec);
 709             velecsum         = _mm256_add_pd(velecsum,velec);
 710
 711             fscal            = felec;
 712
 713             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 714
 715             /* Calculate temporary vectorial force */
 716             tx               = _mm256_mul_pd(fscal,dx13);
 717             ty               = _mm256_mul_pd(fscal,dy13);
 718             tz               = _mm256_mul_pd(fscal,dz13);
 719
 720             /* Update vectorial force */
 721             fix1             = _mm256_add_pd(fix1,tx);
 722             fiy1             = _mm256_add_pd(fiy1,ty);
 723             fiz1             = _mm256_add_pd(fiz1,tz);
 724
 725             fjx3             = _mm256_add_pd(fjx3,tx);
 726             fjy3             = _mm256_add_pd(fjy3,ty);
 727             fjz3             = _mm256_add_pd(fjz3,tz);
 728
 729             /**************************
 730              * CALCULATE INTERACTIONS *
 731              **************************/
 732
 733             /* REACTION-FIELD ELECTROSTATICS */
 734             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 735             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 736
 737             /* Update potential sum for this i atom from the interaction with this j atom. */
 738             velec            = _mm256_andnot_pd(dummy_mask,velec);
 739             velecsum         = _mm256_add_pd(velecsum,velec);
 740
 741             fscal            = felec;
 742
 743             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 744
 745             /* Calculate temporary vectorial force */
 746             tx               = _mm256_mul_pd(fscal,dx21);
 747             ty               = _mm256_mul_pd(fscal,dy21);
 748             tz               = _mm256_mul_pd(fscal,dz21);
 749
 750             /* Update vectorial force */
 751             fix2             = _mm256_add_pd(fix2,tx);
 752             fiy2             = _mm256_add_pd(fiy2,ty);
 753             fiz2             = _mm256_add_pd(fiz2,tz);
 754
 755             fjx1             = _mm256_add_pd(fjx1,tx);
 756             fjy1             = _mm256_add_pd(fjy1,ty);
 757             fjz1             = _mm256_add_pd(fjz1,tz);
 758
 759             /**************************
 760              * CALCULATE INTERACTIONS *
 761              **************************/
 762
 763             /* REACTION-FIELD ELECTROSTATICS */
 764             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 765             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 766
 767             /* Update potential sum for this i atom from the interaction with this j atom. */
 768             velec            = _mm256_andnot_pd(dummy_mask,velec);
 769             velecsum         = _mm256_add_pd(velecsum,velec);
 770
 771             fscal            = felec;
 772
 773             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 774
 775             /* Calculate temporary vectorial force */
 776             tx               = _mm256_mul_pd(fscal,dx22);
 777             ty               = _mm256_mul_pd(fscal,dy22);
 778             tz               = _mm256_mul_pd(fscal,dz22);
 779
 780             /* Update vectorial force */
 781             fix2             = _mm256_add_pd(fix2,tx);
 782             fiy2             = _mm256_add_pd(fiy2,ty);
 783             fiz2             = _mm256_add_pd(fiz2,tz);
 784
 785             fjx2             = _mm256_add_pd(fjx2,tx);
 786             fjy2             = _mm256_add_pd(fjy2,ty);
 787             fjz2             = _mm256_add_pd(fjz2,tz);
 788
 789             /**************************
 790              * CALCULATE INTERACTIONS *
 791              **************************/
 792
 793             /* REACTION-FIELD ELECTROSTATICS */
 794             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
 795             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
 796
 797             /* Update potential sum for this i atom from the interaction with this j atom. */
 798             velec            = _mm256_andnot_pd(dummy_mask,velec);
 799             velecsum         = _mm256_add_pd(velecsum,velec);
 800
 801             fscal            = felec;
 802
 803             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 804
 805             /* Calculate temporary vectorial force */
 806             tx               = _mm256_mul_pd(fscal,dx23);
 807             ty               = _mm256_mul_pd(fscal,dy23);
 808             tz               = _mm256_mul_pd(fscal,dz23);
 809
 810             /* Update vectorial force */
 811             fix2             = _mm256_add_pd(fix2,tx);
 812             fiy2             = _mm256_add_pd(fiy2,ty);
 813             fiz2             = _mm256_add_pd(fiz2,tz);
 814
 815             fjx3             = _mm256_add_pd(fjx3,tx);
 816             fjy3             = _mm256_add_pd(fjy3,ty);
 817             fjz3             = _mm256_add_pd(fjz3,tz);
 818
 819             /**************************
 820              * CALCULATE INTERACTIONS *
 821              **************************/
 822
 823             /* REACTION-FIELD ELECTROSTATICS */
 824             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
 825             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
 826
 827             /* Update potential sum for this i atom from the interaction with this j atom. */
 828             velec            = _mm256_andnot_pd(dummy_mask,velec);
 829             velecsum         = _mm256_add_pd(velecsum,velec);
 830
 831             fscal            = felec;
 832
 833             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 834
 835             /* Calculate temporary vectorial force */
 836             tx               = _mm256_mul_pd(fscal,dx31);
 837             ty               = _mm256_mul_pd(fscal,dy31);
 838             tz               = _mm256_mul_pd(fscal,dz31);
 839
 840             /* Update vectorial force */
 841             fix3             = _mm256_add_pd(fix3,tx);
 842             fiy3             = _mm256_add_pd(fiy3,ty);
 843             fiz3             = _mm256_add_pd(fiz3,tz);
 844
 845             fjx1             = _mm256_add_pd(fjx1,tx);
 846             fjy1             = _mm256_add_pd(fjy1,ty);
 847             fjz1             = _mm256_add_pd(fjz1,tz);
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             /* REACTION-FIELD ELECTROSTATICS */
 854             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
 855             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
 856
 857             /* Update potential sum for this i atom from the interaction with this j atom. */
 858             velec            = _mm256_andnot_pd(dummy_mask,velec);
 859             velecsum         = _mm256_add_pd(velecsum,velec);
 860
 861             fscal            = felec;
 862
 863             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 864
 865             /* Calculate temporary vectorial force */
 866             tx               = _mm256_mul_pd(fscal,dx32);
 867             ty               = _mm256_mul_pd(fscal,dy32);
 868             tz               = _mm256_mul_pd(fscal,dz32);
 869
 870             /* Update vectorial force */
 871             fix3             = _mm256_add_pd(fix3,tx);
 872             fiy3             = _mm256_add_pd(fiy3,ty);
 873             fiz3             = _mm256_add_pd(fiz3,tz);
 874
 875             fjx2             = _mm256_add_pd(fjx2,tx);
 876             fjy2             = _mm256_add_pd(fjy2,ty);
 877             fjz2             = _mm256_add_pd(fjz2,tz);
 878
 879             /**************************
 880              * CALCULATE INTERACTIONS *
 881              **************************/
 882
 883             /* REACTION-FIELD ELECTROSTATICS */
 884             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
 885             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
 886
 887             /* Update potential sum for this i atom from the interaction with this j atom. */
 888             velec            = _mm256_andnot_pd(dummy_mask,velec);
 889             velecsum         = _mm256_add_pd(velecsum,velec);
 890
 891             fscal            = felec;
 892
 893             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 894
 895             /* Calculate temporary vectorial force */
 896             tx               = _mm256_mul_pd(fscal,dx33);
 897             ty               = _mm256_mul_pd(fscal,dy33);
 898             tz               = _mm256_mul_pd(fscal,dz33);
 899
 900             /* Update vectorial force */
 901             fix3             = _mm256_add_pd(fix3,tx);
 902             fiy3             = _mm256_add_pd(fiy3,ty);
 903             fiz3             = _mm256_add_pd(fiz3,tz);
 904
 905             fjx3             = _mm256_add_pd(fjx3,tx);
 906             fjy3             = _mm256_add_pd(fjy3,ty);
 907             fjz3             = _mm256_add_pd(fjz3,tz);
 908
 909             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 910             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 911             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 912             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 913
 914             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 915                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 916
 917             /* Inner loop uses 288 flops */
 918         }
 919
 920         /* End of innermost loop */
 921
 922         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 923                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 924
 925         ggid                        = gid[iidx];
 926         /* Update potential energies */
 927         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 928
 929         /* Increment number of inner iterations */
 930         inneriter                  += j_index_end - j_index_start;
 931
 932         /* Outer loop uses 19 flops */
 933     }
 934
 935     /* Increment number of outer iterations */
 936     outeriter        += nri;
 937
 938     /* Update outer/inner flops */
 939
 940     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*288);
 941 }
 942 /*
 943  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4W4_F_avx_256_double
 944  * Electrostatics interaction: ReactionField
 945  * VdW interaction:            None
 946  * Geometry:                   Water4-Water4
 947  * Calculate force/pot:        Force
 948  */
 949 void
 950 nb_kernel_ElecRF_VdwNone_GeomW4W4_F_avx_256_double
 951                     (t_nblist                    * gmx_restrict       nlist,
 952                      rvec                        * gmx_restrict          xx,
 953                      rvec                        * gmx_restrict          ff,
 954                      t_forcerec                  * gmx_restrict          fr,
 955                      t_mdatoms                   * gmx_restrict     mdatoms,
 956                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 957                      t_nrnb                      * gmx_restrict        nrnb)
 958 {
 959     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 960      * just 0 for non-waters.
 961      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 962      * jnr indices corresponding to data put in the four positions in the SIMD register.
 963      */
 964     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 965     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 966     int              jnrA,jnrB,jnrC,jnrD;
 967     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 968     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 969     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 970     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 971     real             rcutoff_scalar;
 972     real             *shiftvec,*fshift,*x,*f;
 973     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 974     real             scratch[4*DIM];
 975     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 976     real *           vdwioffsetptr1;
 977     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 978     real *           vdwioffsetptr2;
 979     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 980     real *           vdwioffsetptr3;
 981     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 982     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
 983     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 984     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
 985     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 986     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
 987     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 988     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 989     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 990     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 991     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 992     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 993     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 994     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 995     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 996     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 997     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 998     real             *charge;
 999     __m256d          dummy_mask,cutoff_mask;
1000     __m128           tmpmask0,tmpmask1;
1001     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1002     __m256d          one     = _mm256_set1_pd(1.0);
1003     __m256d          two     = _mm256_set1_pd(2.0);
1004     x                = xx[0];
1005     f                = ff[0];
1006
1007     nri              = nlist->nri;
1008     iinr             = nlist->iinr;
1009     jindex           = nlist->jindex;
1010     jjnr             = nlist->jjnr;
1011     shiftidx         = nlist->shift;
1012     gid              = nlist->gid;
1013     shiftvec         = fr->shift_vec[0];
1014     fshift           = fr->fshift[0];
1015     facel            = _mm256_set1_pd(fr->epsfac);
1016     charge           = mdatoms->chargeA;
1017     krf              = _mm256_set1_pd(fr->ic->k_rf);
1018     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
1019     crf              = _mm256_set1_pd(fr->ic->c_rf);
1020
1021     /* Setup water-specific parameters */
1022     inr              = nlist->iinr[0];
1023     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1024     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1025     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1026
1027     jq1              = _mm256_set1_pd(charge[inr+1]);
1028     jq2              = _mm256_set1_pd(charge[inr+2]);
1029     jq3              = _mm256_set1_pd(charge[inr+3]);
1030     qq11             = _mm256_mul_pd(iq1,jq1);
1031     qq12             = _mm256_mul_pd(iq1,jq2);
1032     qq13             = _mm256_mul_pd(iq1,jq3);
1033     qq21             = _mm256_mul_pd(iq2,jq1);
1034     qq22             = _mm256_mul_pd(iq2,jq2);
1035     qq23             = _mm256_mul_pd(iq2,jq3);
1036     qq31             = _mm256_mul_pd(iq3,jq1);
1037     qq32             = _mm256_mul_pd(iq3,jq2);
1038     qq33             = _mm256_mul_pd(iq3,jq3);
1039
1040     /* Avoid stupid compiler warnings */
1041     jnrA = jnrB = jnrC = jnrD = 0;
1042     j_coord_offsetA = 0;
1043     j_coord_offsetB = 0;
1044     j_coord_offsetC = 0;
1045     j_coord_offsetD = 0;
1046
1047     outeriter        = 0;
1048     inneriter        = 0;
1049
1050     for(iidx=0;iidx<4*DIM;iidx++)
1051     {
1052         scratch[iidx] = 0.0;
1053     }
1054
1055     /* Start outer loop over neighborlists */
1056     for(iidx=0; iidx<nri; iidx++)
1057     {
1058         /* Load shift vector for this list */
1059         i_shift_offset   = DIM*shiftidx[iidx];
1060
1061         /* Load limits for loop over neighbors */
1062         j_index_start    = jindex[iidx];
1063         j_index_end      = jindex[iidx+1];
1064
1065         /* Get outer coordinate index */
1066         inr              = iinr[iidx];
1067         i_coord_offset   = DIM*inr;
1068
1069         /* Load i particle coords and add shift vector */
1070         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1071                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1072
1073         fix1             = _mm256_setzero_pd();
1074         fiy1             = _mm256_setzero_pd();
1075         fiz1             = _mm256_setzero_pd();
1076         fix2             = _mm256_setzero_pd();
1077         fiy2             = _mm256_setzero_pd();
1078         fiz2             = _mm256_setzero_pd();
1079         fix3             = _mm256_setzero_pd();
1080         fiy3             = _mm256_setzero_pd();
1081         fiz3             = _mm256_setzero_pd();
1082
1083         /* Start inner kernel loop */
1084         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1085         {
1086
1087             /* Get j neighbor index, and coordinate index */
1088             jnrA             = jjnr[jidx];
1089             jnrB             = jjnr[jidx+1];
1090             jnrC             = jjnr[jidx+2];
1091             jnrD             = jjnr[jidx+3];
1092             j_coord_offsetA  = DIM*jnrA;
1093             j_coord_offsetB  = DIM*jnrB;
1094             j_coord_offsetC  = DIM*jnrC;
1095             j_coord_offsetD  = DIM*jnrD;
1096
1097             /* load j atom coordinates */
1098             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1099                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1100                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1101
1102             /* Calculate displacement vector */
1103             dx11             = _mm256_sub_pd(ix1,jx1);
1104             dy11             = _mm256_sub_pd(iy1,jy1);
1105             dz11             = _mm256_sub_pd(iz1,jz1);
1106             dx12             = _mm256_sub_pd(ix1,jx2);
1107             dy12             = _mm256_sub_pd(iy1,jy2);
1108             dz12             = _mm256_sub_pd(iz1,jz2);
1109             dx13             = _mm256_sub_pd(ix1,jx3);
1110             dy13             = _mm256_sub_pd(iy1,jy3);
1111             dz13             = _mm256_sub_pd(iz1,jz3);
1112             dx21             = _mm256_sub_pd(ix2,jx1);
1113             dy21             = _mm256_sub_pd(iy2,jy1);
1114             dz21             = _mm256_sub_pd(iz2,jz1);
1115             dx22             = _mm256_sub_pd(ix2,jx2);
1116             dy22             = _mm256_sub_pd(iy2,jy2);
1117             dz22             = _mm256_sub_pd(iz2,jz2);
1118             dx23             = _mm256_sub_pd(ix2,jx3);
1119             dy23             = _mm256_sub_pd(iy2,jy3);
1120             dz23             = _mm256_sub_pd(iz2,jz3);
1121             dx31             = _mm256_sub_pd(ix3,jx1);
1122             dy31             = _mm256_sub_pd(iy3,jy1);
1123             dz31             = _mm256_sub_pd(iz3,jz1);
1124             dx32             = _mm256_sub_pd(ix3,jx2);
1125             dy32             = _mm256_sub_pd(iy3,jy2);
1126             dz32             = _mm256_sub_pd(iz3,jz2);
1127             dx33             = _mm256_sub_pd(ix3,jx3);
1128             dy33             = _mm256_sub_pd(iy3,jy3);
1129             dz33             = _mm256_sub_pd(iz3,jz3);
1130
1131             /* Calculate squared distance and things based on it */
1132             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1133             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1134             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1135             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1136             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1137             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1138             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1139             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1140             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1141
1142             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1143             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1144             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1145             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1146             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1147             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1148             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1149             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1150             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1151
1152             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1153             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1154             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1155             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1156             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1157             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1158             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1159             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1160             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1161
1162             fjx1             = _mm256_setzero_pd();
1163             fjy1             = _mm256_setzero_pd();
1164             fjz1             = _mm256_setzero_pd();
1165             fjx2             = _mm256_setzero_pd();
1166             fjy2             = _mm256_setzero_pd();
1167             fjz2             = _mm256_setzero_pd();
1168             fjx3             = _mm256_setzero_pd();
1169             fjy3             = _mm256_setzero_pd();
1170             fjz3             = _mm256_setzero_pd();
1171
1172             /**************************
1173              * CALCULATE INTERACTIONS *
1174              **************************/
1175
1176             /* REACTION-FIELD ELECTROSTATICS */
1177             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1178
1179             fscal            = felec;
1180
1181             /* Calculate temporary vectorial force */
1182             tx               = _mm256_mul_pd(fscal,dx11);
1183             ty               = _mm256_mul_pd(fscal,dy11);
1184             tz               = _mm256_mul_pd(fscal,dz11);
1185
1186             /* Update vectorial force */
1187             fix1             = _mm256_add_pd(fix1,tx);
1188             fiy1             = _mm256_add_pd(fiy1,ty);
1189             fiz1             = _mm256_add_pd(fiz1,tz);
1190
1191             fjx1             = _mm256_add_pd(fjx1,tx);
1192             fjy1             = _mm256_add_pd(fjy1,ty);
1193             fjz1             = _mm256_add_pd(fjz1,tz);
1194
1195             /**************************
1196              * CALCULATE INTERACTIONS *
1197              **************************/
1198
1199             /* REACTION-FIELD ELECTROSTATICS */
1200             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1201
1202             fscal            = felec;
1203
1204             /* Calculate temporary vectorial force */
1205             tx               = _mm256_mul_pd(fscal,dx12);
1206             ty               = _mm256_mul_pd(fscal,dy12);
1207             tz               = _mm256_mul_pd(fscal,dz12);
1208
1209             /* Update vectorial force */
1210             fix1             = _mm256_add_pd(fix1,tx);
1211             fiy1             = _mm256_add_pd(fiy1,ty);
1212             fiz1             = _mm256_add_pd(fiz1,tz);
1213
1214             fjx2             = _mm256_add_pd(fjx2,tx);
1215             fjy2             = _mm256_add_pd(fjy2,ty);
1216             fjz2             = _mm256_add_pd(fjz2,tz);
1217
1218             /**************************
1219              * CALCULATE INTERACTIONS *
1220              **************************/
1221
1222             /* REACTION-FIELD ELECTROSTATICS */
1223             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
1224
1225             fscal            = felec;
1226
1227             /* Calculate temporary vectorial force */
1228             tx               = _mm256_mul_pd(fscal,dx13);
1229             ty               = _mm256_mul_pd(fscal,dy13);
1230             tz               = _mm256_mul_pd(fscal,dz13);
1231
1232             /* Update vectorial force */
1233             fix1             = _mm256_add_pd(fix1,tx);
1234             fiy1             = _mm256_add_pd(fiy1,ty);
1235             fiz1             = _mm256_add_pd(fiz1,tz);
1236
1237             fjx3             = _mm256_add_pd(fjx3,tx);
1238             fjy3             = _mm256_add_pd(fjy3,ty);
1239             fjz3             = _mm256_add_pd(fjz3,tz);
1240
1241             /**************************
1242              * CALCULATE INTERACTIONS *
1243              **************************/
1244
1245             /* REACTION-FIELD ELECTROSTATICS */
1246             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
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             /* REACTION-FIELD ELECTROSTATICS */
1269             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1270
1271             fscal            = felec;
1272
1273             /* Calculate temporary vectorial force */
1274             tx               = _mm256_mul_pd(fscal,dx22);
1275             ty               = _mm256_mul_pd(fscal,dy22);
1276             tz               = _mm256_mul_pd(fscal,dz22);
1277
1278             /* Update vectorial force */
1279             fix2             = _mm256_add_pd(fix2,tx);
1280             fiy2             = _mm256_add_pd(fiy2,ty);
1281             fiz2             = _mm256_add_pd(fiz2,tz);
1282
1283             fjx2             = _mm256_add_pd(fjx2,tx);
1284             fjy2             = _mm256_add_pd(fjy2,ty);
1285             fjz2             = _mm256_add_pd(fjz2,tz);
1286
1287             /**************************
1288              * CALCULATE INTERACTIONS *
1289              **************************/
1290
1291             /* REACTION-FIELD ELECTROSTATICS */
1292             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1293
1294             fscal            = felec;
1295
1296             /* Calculate temporary vectorial force */
1297             tx               = _mm256_mul_pd(fscal,dx23);
1298             ty               = _mm256_mul_pd(fscal,dy23);
1299             tz               = _mm256_mul_pd(fscal,dz23);
1300
1301             /* Update vectorial force */
1302             fix2             = _mm256_add_pd(fix2,tx);
1303             fiy2             = _mm256_add_pd(fiy2,ty);
1304             fiz2             = _mm256_add_pd(fiz2,tz);
1305
1306             fjx3             = _mm256_add_pd(fjx3,tx);
1307             fjy3             = _mm256_add_pd(fjy3,ty);
1308             fjz3             = _mm256_add_pd(fjz3,tz);
1309
1310             /**************************
1311              * CALCULATE INTERACTIONS *
1312              **************************/
1313
1314             /* REACTION-FIELD ELECTROSTATICS */
1315             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1316
1317             fscal            = felec;
1318
1319             /* Calculate temporary vectorial force */
1320             tx               = _mm256_mul_pd(fscal,dx31);
1321             ty               = _mm256_mul_pd(fscal,dy31);
1322             tz               = _mm256_mul_pd(fscal,dz31);
1323
1324             /* Update vectorial force */
1325             fix3             = _mm256_add_pd(fix3,tx);
1326             fiy3             = _mm256_add_pd(fiy3,ty);
1327             fiz3             = _mm256_add_pd(fiz3,tz);
1328
1329             fjx1             = _mm256_add_pd(fjx1,tx);
1330             fjy1             = _mm256_add_pd(fjy1,ty);
1331             fjz1             = _mm256_add_pd(fjz1,tz);
1332
1333             /**************************
1334              * CALCULATE INTERACTIONS *
1335              **************************/
1336
1337             /* REACTION-FIELD ELECTROSTATICS */
1338             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1339
1340             fscal            = felec;
1341
1342             /* Calculate temporary vectorial force */
1343             tx               = _mm256_mul_pd(fscal,dx32);
1344             ty               = _mm256_mul_pd(fscal,dy32);
1345             tz               = _mm256_mul_pd(fscal,dz32);
1346
1347             /* Update vectorial force */
1348             fix3             = _mm256_add_pd(fix3,tx);
1349             fiy3             = _mm256_add_pd(fiy3,ty);
1350             fiz3             = _mm256_add_pd(fiz3,tz);
1351
1352             fjx2             = _mm256_add_pd(fjx2,tx);
1353             fjy2             = _mm256_add_pd(fjy2,ty);
1354             fjz2             = _mm256_add_pd(fjz2,tz);
1355
1356             /**************************
1357              * CALCULATE INTERACTIONS *
1358              **************************/
1359
1360             /* REACTION-FIELD ELECTROSTATICS */
1361             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1362
1363             fscal            = felec;
1364
1365             /* Calculate temporary vectorial force */
1366             tx               = _mm256_mul_pd(fscal,dx33);
1367             ty               = _mm256_mul_pd(fscal,dy33);
1368             tz               = _mm256_mul_pd(fscal,dz33);
1369
1370             /* Update vectorial force */
1371             fix3             = _mm256_add_pd(fix3,tx);
1372             fiy3             = _mm256_add_pd(fiy3,ty);
1373             fiz3             = _mm256_add_pd(fiz3,tz);
1374
1375             fjx3             = _mm256_add_pd(fjx3,tx);
1376             fjy3             = _mm256_add_pd(fjy3,ty);
1377             fjz3             = _mm256_add_pd(fjz3,tz);
1378
1379             fjptrA             = f+j_coord_offsetA;
1380             fjptrB             = f+j_coord_offsetB;
1381             fjptrC             = f+j_coord_offsetC;
1382             fjptrD             = f+j_coord_offsetD;
1383
1384             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1385                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1386
1387             /* Inner loop uses 243 flops */
1388         }
1389
1390         if(jidx<j_index_end)
1391         {
1392
1393             /* Get j neighbor index, and coordinate index */
1394             jnrlistA         = jjnr[jidx];
1395             jnrlistB         = jjnr[jidx+1];
1396             jnrlistC         = jjnr[jidx+2];
1397             jnrlistD         = jjnr[jidx+3];
1398             /* Sign of each element will be negative for non-real atoms.
1399              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1400              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1401              */
1402             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1403
1404             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1405             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1406             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1407
1408             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1409             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1410             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1411             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1412             j_coord_offsetA  = DIM*jnrA;
1413             j_coord_offsetB  = DIM*jnrB;
1414             j_coord_offsetC  = DIM*jnrC;
1415             j_coord_offsetD  = DIM*jnrD;
1416
1417             /* load j atom coordinates */
1418             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1419                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1420                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1421
1422             /* Calculate displacement vector */
1423             dx11             = _mm256_sub_pd(ix1,jx1);
1424             dy11             = _mm256_sub_pd(iy1,jy1);
1425             dz11             = _mm256_sub_pd(iz1,jz1);
1426             dx12             = _mm256_sub_pd(ix1,jx2);
1427             dy12             = _mm256_sub_pd(iy1,jy2);
1428             dz12             = _mm256_sub_pd(iz1,jz2);
1429             dx13             = _mm256_sub_pd(ix1,jx3);
1430             dy13             = _mm256_sub_pd(iy1,jy3);
1431             dz13             = _mm256_sub_pd(iz1,jz3);
1432             dx21             = _mm256_sub_pd(ix2,jx1);
1433             dy21             = _mm256_sub_pd(iy2,jy1);
1434             dz21             = _mm256_sub_pd(iz2,jz1);
1435             dx22             = _mm256_sub_pd(ix2,jx2);
1436             dy22             = _mm256_sub_pd(iy2,jy2);
1437             dz22             = _mm256_sub_pd(iz2,jz2);
1438             dx23             = _mm256_sub_pd(ix2,jx3);
1439             dy23             = _mm256_sub_pd(iy2,jy3);
1440             dz23             = _mm256_sub_pd(iz2,jz3);
1441             dx31             = _mm256_sub_pd(ix3,jx1);
1442             dy31             = _mm256_sub_pd(iy3,jy1);
1443             dz31             = _mm256_sub_pd(iz3,jz1);
1444             dx32             = _mm256_sub_pd(ix3,jx2);
1445             dy32             = _mm256_sub_pd(iy3,jy2);
1446             dz32             = _mm256_sub_pd(iz3,jz2);
1447             dx33             = _mm256_sub_pd(ix3,jx3);
1448             dy33             = _mm256_sub_pd(iy3,jy3);
1449             dz33             = _mm256_sub_pd(iz3,jz3);
1450
1451             /* Calculate squared distance and things based on it */
1452             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1453             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1454             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1455             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1456             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1457             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1458             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1459             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1460             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1461
1462             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1463             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1464             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1465             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1466             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1467             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1468             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1469             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1470             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1471
1472             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1473             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1474             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1475             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1476             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1477             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1478             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1479             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1480             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1481
1482             fjx1             = _mm256_setzero_pd();
1483             fjy1             = _mm256_setzero_pd();
1484             fjz1             = _mm256_setzero_pd();
1485             fjx2             = _mm256_setzero_pd();
1486             fjy2             = _mm256_setzero_pd();
1487             fjz2             = _mm256_setzero_pd();
1488             fjx3             = _mm256_setzero_pd();
1489             fjy3             = _mm256_setzero_pd();
1490             fjz3             = _mm256_setzero_pd();
1491
1492             /**************************
1493              * CALCULATE INTERACTIONS *
1494              **************************/
1495
1496             /* REACTION-FIELD ELECTROSTATICS */
1497             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1498
1499             fscal            = felec;
1500
1501             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1502
1503             /* Calculate temporary vectorial force */
1504             tx               = _mm256_mul_pd(fscal,dx11);
1505             ty               = _mm256_mul_pd(fscal,dy11);
1506             tz               = _mm256_mul_pd(fscal,dz11);
1507
1508             /* Update vectorial force */
1509             fix1             = _mm256_add_pd(fix1,tx);
1510             fiy1             = _mm256_add_pd(fiy1,ty);
1511             fiz1             = _mm256_add_pd(fiz1,tz);
1512
1513             fjx1             = _mm256_add_pd(fjx1,tx);
1514             fjy1             = _mm256_add_pd(fjy1,ty);
1515             fjz1             = _mm256_add_pd(fjz1,tz);
1516
1517             /**************************
1518              * CALCULATE INTERACTIONS *
1519              **************************/
1520
1521             /* REACTION-FIELD ELECTROSTATICS */
1522             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1523
1524             fscal            = felec;
1525
1526             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1527
1528             /* Calculate temporary vectorial force */
1529             tx               = _mm256_mul_pd(fscal,dx12);
1530             ty               = _mm256_mul_pd(fscal,dy12);
1531             tz               = _mm256_mul_pd(fscal,dz12);
1532
1533             /* Update vectorial force */
1534             fix1             = _mm256_add_pd(fix1,tx);
1535             fiy1             = _mm256_add_pd(fiy1,ty);
1536             fiz1             = _mm256_add_pd(fiz1,tz);
1537
1538             fjx2             = _mm256_add_pd(fjx2,tx);
1539             fjy2             = _mm256_add_pd(fjy2,ty);
1540             fjz2             = _mm256_add_pd(fjz2,tz);
1541
1542             /**************************
1543              * CALCULATE INTERACTIONS *
1544              **************************/
1545
1546             /* REACTION-FIELD ELECTROSTATICS */
1547             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
1548
1549             fscal            = felec;
1550
1551             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1552
1553             /* Calculate temporary vectorial force */
1554             tx               = _mm256_mul_pd(fscal,dx13);
1555             ty               = _mm256_mul_pd(fscal,dy13);
1556             tz               = _mm256_mul_pd(fscal,dz13);
1557
1558             /* Update vectorial force */
1559             fix1             = _mm256_add_pd(fix1,tx);
1560             fiy1             = _mm256_add_pd(fiy1,ty);
1561             fiz1             = _mm256_add_pd(fiz1,tz);
1562
1563             fjx3             = _mm256_add_pd(fjx3,tx);
1564             fjy3             = _mm256_add_pd(fjy3,ty);
1565             fjz3             = _mm256_add_pd(fjz3,tz);
1566
1567             /**************************
1568              * CALCULATE INTERACTIONS *
1569              **************************/
1570
1571             /* REACTION-FIELD ELECTROSTATICS */
1572             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1573
1574             fscal            = felec;
1575
1576             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1577
1578             /* Calculate temporary vectorial force */
1579             tx               = _mm256_mul_pd(fscal,dx21);
1580             ty               = _mm256_mul_pd(fscal,dy21);
1581             tz               = _mm256_mul_pd(fscal,dz21);
1582
1583             /* Update vectorial force */
1584             fix2             = _mm256_add_pd(fix2,tx);
1585             fiy2             = _mm256_add_pd(fiy2,ty);
1586             fiz2             = _mm256_add_pd(fiz2,tz);
1587
1588             fjx1             = _mm256_add_pd(fjx1,tx);
1589             fjy1             = _mm256_add_pd(fjy1,ty);
1590             fjz1             = _mm256_add_pd(fjz1,tz);
1591
1592             /**************************
1593              * CALCULATE INTERACTIONS *
1594              **************************/
1595
1596             /* REACTION-FIELD ELECTROSTATICS */
1597             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1598
1599             fscal            = felec;
1600
1601             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1602
1603             /* Calculate temporary vectorial force */
1604             tx               = _mm256_mul_pd(fscal,dx22);
1605             ty               = _mm256_mul_pd(fscal,dy22);
1606             tz               = _mm256_mul_pd(fscal,dz22);
1607
1608             /* Update vectorial force */
1609             fix2             = _mm256_add_pd(fix2,tx);
1610             fiy2             = _mm256_add_pd(fiy2,ty);
1611             fiz2             = _mm256_add_pd(fiz2,tz);
1612
1613             fjx2             = _mm256_add_pd(fjx2,tx);
1614             fjy2             = _mm256_add_pd(fjy2,ty);
1615             fjz2             = _mm256_add_pd(fjz2,tz);
1616
1617             /**************************
1618              * CALCULATE INTERACTIONS *
1619              **************************/
1620
1621             /* REACTION-FIELD ELECTROSTATICS */
1622             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1623
1624             fscal            = felec;
1625
1626             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1627
1628             /* Calculate temporary vectorial force */
1629             tx               = _mm256_mul_pd(fscal,dx23);
1630             ty               = _mm256_mul_pd(fscal,dy23);
1631             tz               = _mm256_mul_pd(fscal,dz23);
1632
1633             /* Update vectorial force */
1634             fix2             = _mm256_add_pd(fix2,tx);
1635             fiy2             = _mm256_add_pd(fiy2,ty);
1636             fiz2             = _mm256_add_pd(fiz2,tz);
1637
1638             fjx3             = _mm256_add_pd(fjx3,tx);
1639             fjy3             = _mm256_add_pd(fjy3,ty);
1640             fjz3             = _mm256_add_pd(fjz3,tz);
1641
1642             /**************************
1643              * CALCULATE INTERACTIONS *
1644              **************************/
1645
1646             /* REACTION-FIELD ELECTROSTATICS */
1647             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1648
1649             fscal            = felec;
1650
1651             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1652
1653             /* Calculate temporary vectorial force */
1654             tx               = _mm256_mul_pd(fscal,dx31);
1655             ty               = _mm256_mul_pd(fscal,dy31);
1656             tz               = _mm256_mul_pd(fscal,dz31);
1657
1658             /* Update vectorial force */
1659             fix3             = _mm256_add_pd(fix3,tx);
1660             fiy3             = _mm256_add_pd(fiy3,ty);
1661             fiz3             = _mm256_add_pd(fiz3,tz);
1662
1663             fjx1             = _mm256_add_pd(fjx1,tx);
1664             fjy1             = _mm256_add_pd(fjy1,ty);
1665             fjz1             = _mm256_add_pd(fjz1,tz);
1666
1667             /**************************
1668              * CALCULATE INTERACTIONS *
1669              **************************/
1670
1671             /* REACTION-FIELD ELECTROSTATICS */
1672             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1673
1674             fscal            = felec;
1675
1676             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1677
1678             /* Calculate temporary vectorial force */
1679             tx               = _mm256_mul_pd(fscal,dx32);
1680             ty               = _mm256_mul_pd(fscal,dy32);
1681             tz               = _mm256_mul_pd(fscal,dz32);
1682
1683             /* Update vectorial force */
1684             fix3             = _mm256_add_pd(fix3,tx);
1685             fiy3             = _mm256_add_pd(fiy3,ty);
1686             fiz3             = _mm256_add_pd(fiz3,tz);
1687
1688             fjx2             = _mm256_add_pd(fjx2,tx);
1689             fjy2             = _mm256_add_pd(fjy2,ty);
1690             fjz2             = _mm256_add_pd(fjz2,tz);
1691
1692             /**************************
1693              * CALCULATE INTERACTIONS *
1694              **************************/
1695
1696             /* REACTION-FIELD ELECTROSTATICS */
1697             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1698
1699             fscal            = felec;
1700
1701             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1702
1703             /* Calculate temporary vectorial force */
1704             tx               = _mm256_mul_pd(fscal,dx33);
1705             ty               = _mm256_mul_pd(fscal,dy33);
1706             tz               = _mm256_mul_pd(fscal,dz33);
1707
1708             /* Update vectorial force */
1709             fix3             = _mm256_add_pd(fix3,tx);
1710             fiy3             = _mm256_add_pd(fiy3,ty);
1711             fiz3             = _mm256_add_pd(fiz3,tz);
1712
1713             fjx3             = _mm256_add_pd(fjx3,tx);
1714             fjy3             = _mm256_add_pd(fjy3,ty);
1715             fjz3             = _mm256_add_pd(fjz3,tz);
1716
1717             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1718             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1719             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1720             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1721
1722             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1723                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1724
1725             /* Inner loop uses 243 flops */
1726         }
1727
1728         /* End of innermost loop */
1729
1730         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1731                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1732
1733         /* Increment number of inner iterations */
1734         inneriter                  += j_index_end - j_index_start;
1735
1736         /* Outer loop uses 18 flops */
1737     }
1738
1739     /* Increment number of outer iterations */
1740     outeriter        += nri;
1741
1742     /* Update outer/inner flops */
1743
1744     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*243);
1745 }