src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4W4_VF_avx_256_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwNone_GeomW4W4_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  93     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  94     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  95     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  96     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  97     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  98     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  99     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 100     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 101     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 102     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 103     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 104     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 105     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 106     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 107     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     __m256d          dummy_mask,cutoff_mask;
 110     __m128           tmpmask0,tmpmask1;
 111     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 112     __m256d          one     = _mm256_set1_pd(1.0);
 113     __m256d          two     = _mm256_set1_pd(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_pd(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     krf              = _mm256_set1_pd(fr->ic->k_rf);
 128     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 129     crf              = _mm256_set1_pd(fr->ic->c_rf);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 134     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 135     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 136
 137     jq1              = _mm256_set1_pd(charge[inr+1]);
 138     jq2              = _mm256_set1_pd(charge[inr+2]);
 139     jq3              = _mm256_set1_pd(charge[inr+3]);
 140     qq11             = _mm256_mul_pd(iq1,jq1);
 141     qq12             = _mm256_mul_pd(iq1,jq2);
 142     qq13             = _mm256_mul_pd(iq1,jq3);
 143     qq21             = _mm256_mul_pd(iq2,jq1);
 144     qq22             = _mm256_mul_pd(iq2,jq2);
 145     qq23             = _mm256_mul_pd(iq2,jq3);
 146     qq31             = _mm256_mul_pd(iq3,jq1);
 147     qq32             = _mm256_mul_pd(iq3,jq2);
 148     qq33             = _mm256_mul_pd(iq3,jq3);
 149
 150     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 151     rcutoff_scalar   = fr->rcoulomb;
 152     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 153     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 154
 155     /* Avoid stupid compiler warnings */
 156     jnrA = jnrB = jnrC = jnrD = 0;
 157     j_coord_offsetA = 0;
 158     j_coord_offsetB = 0;
 159     j_coord_offsetC = 0;
 160     j_coord_offsetD = 0;
 161
 162     outeriter        = 0;
 163     inneriter        = 0;
 164
 165     for(iidx=0;iidx<4*DIM;iidx++)
 166     {
 167         scratch[iidx] = 0.0;
 168     }
 169
 170     /* Start outer loop over neighborlists */
 171     for(iidx=0; iidx<nri; iidx++)
 172     {
 173         /* Load shift vector for this list */
 174         i_shift_offset   = DIM*shiftidx[iidx];
 175
 176         /* Load limits for loop over neighbors */
 177         j_index_start    = jindex[iidx];
 178         j_index_end      = jindex[iidx+1];
 179
 180         /* Get outer coordinate index */
 181         inr              = iinr[iidx];
 182         i_coord_offset   = DIM*inr;
 183
 184         /* Load i particle coords and add shift vector */
 185         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 186                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 187
 188         fix1             = _mm256_setzero_pd();
 189         fiy1             = _mm256_setzero_pd();
 190         fiz1             = _mm256_setzero_pd();
 191         fix2             = _mm256_setzero_pd();
 192         fiy2             = _mm256_setzero_pd();
 193         fiz2             = _mm256_setzero_pd();
 194         fix3             = _mm256_setzero_pd();
 195         fiy3             = _mm256_setzero_pd();
 196         fiz3             = _mm256_setzero_pd();
 197
 198         /* Reset potential sums */
 199         velecsum         = _mm256_setzero_pd();
 200
 201         /* Start inner kernel loop */
 202         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 203         {
 204
 205             /* Get j neighbor index, and coordinate index */
 206             jnrA             = jjnr[jidx];
 207             jnrB             = jjnr[jidx+1];
 208             jnrC             = jjnr[jidx+2];
 209             jnrD             = jjnr[jidx+3];
 210             j_coord_offsetA  = DIM*jnrA;
 211             j_coord_offsetB  = DIM*jnrB;
 212             j_coord_offsetC  = DIM*jnrC;
 213             j_coord_offsetD  = DIM*jnrD;
 214
 215             /* load j atom coordinates */
 216             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 217                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 218                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 219
 220             /* Calculate displacement vector */
 221             dx11             = _mm256_sub_pd(ix1,jx1);
 222             dy11             = _mm256_sub_pd(iy1,jy1);
 223             dz11             = _mm256_sub_pd(iz1,jz1);
 224             dx12             = _mm256_sub_pd(ix1,jx2);
 225             dy12             = _mm256_sub_pd(iy1,jy2);
 226             dz12             = _mm256_sub_pd(iz1,jz2);
 227             dx13             = _mm256_sub_pd(ix1,jx3);
 228             dy13             = _mm256_sub_pd(iy1,jy3);
 229             dz13             = _mm256_sub_pd(iz1,jz3);
 230             dx21             = _mm256_sub_pd(ix2,jx1);
 231             dy21             = _mm256_sub_pd(iy2,jy1);
 232             dz21             = _mm256_sub_pd(iz2,jz1);
 233             dx22             = _mm256_sub_pd(ix2,jx2);
 234             dy22             = _mm256_sub_pd(iy2,jy2);
 235             dz22             = _mm256_sub_pd(iz2,jz2);
 236             dx23             = _mm256_sub_pd(ix2,jx3);
 237             dy23             = _mm256_sub_pd(iy2,jy3);
 238             dz23             = _mm256_sub_pd(iz2,jz3);
 239             dx31             = _mm256_sub_pd(ix3,jx1);
 240             dy31             = _mm256_sub_pd(iy3,jy1);
 241             dz31             = _mm256_sub_pd(iz3,jz1);
 242             dx32             = _mm256_sub_pd(ix3,jx2);
 243             dy32             = _mm256_sub_pd(iy3,jy2);
 244             dz32             = _mm256_sub_pd(iz3,jz2);
 245             dx33             = _mm256_sub_pd(ix3,jx3);
 246             dy33             = _mm256_sub_pd(iy3,jy3);
 247             dz33             = _mm256_sub_pd(iz3,jz3);
 248
 249             /* Calculate squared distance and things based on it */
 250             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 251             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 252             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 253             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 254             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 255             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 256             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 257             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 258             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 259
 260             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 261             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 262             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 263             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 264             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 265             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 266             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 267             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 268             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 269
 270             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 271             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 272             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 273             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 274             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 275             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 276             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 277             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 278             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 279
 280             fjx1             = _mm256_setzero_pd();
 281             fjy1             = _mm256_setzero_pd();
 282             fjz1             = _mm256_setzero_pd();
 283             fjx2             = _mm256_setzero_pd();
 284             fjy2             = _mm256_setzero_pd();
 285             fjz2             = _mm256_setzero_pd();
 286             fjx3             = _mm256_setzero_pd();
 287             fjy3             = _mm256_setzero_pd();
 288             fjz3             = _mm256_setzero_pd();
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 295             {
 296
 297             /* REACTION-FIELD ELECTROSTATICS */
 298             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 299             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 300
 301             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             velec            = _mm256_and_pd(velec,cutoff_mask);
 305             velecsum         = _mm256_add_pd(velecsum,velec);
 306
 307             fscal            = felec;
 308
 309             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 310
 311             /* Calculate temporary vectorial force */
 312             tx               = _mm256_mul_pd(fscal,dx11);
 313             ty               = _mm256_mul_pd(fscal,dy11);
 314             tz               = _mm256_mul_pd(fscal,dz11);
 315
 316             /* Update vectorial force */
 317             fix1             = _mm256_add_pd(fix1,tx);
 318             fiy1             = _mm256_add_pd(fiy1,ty);
 319             fiz1             = _mm256_add_pd(fiz1,tz);
 320
 321             fjx1             = _mm256_add_pd(fjx1,tx);
 322             fjy1             = _mm256_add_pd(fjy1,ty);
 323             fjz1             = _mm256_add_pd(fjz1,tz);
 324
 325             }
 326
 327             /**************************
 328              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 332             {
 333
 334             /* REACTION-FIELD ELECTROSTATICS */
 335             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 336             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 337
 338             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velec            = _mm256_and_pd(velec,cutoff_mask);
 342             velecsum         = _mm256_add_pd(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm256_mul_pd(fscal,dx12);
 350             ty               = _mm256_mul_pd(fscal,dy12);
 351             tz               = _mm256_mul_pd(fscal,dz12);
 352
 353             /* Update vectorial force */
 354             fix1             = _mm256_add_pd(fix1,tx);
 355             fiy1             = _mm256_add_pd(fiy1,ty);
 356             fiz1             = _mm256_add_pd(fiz1,tz);
 357
 358             fjx2             = _mm256_add_pd(fjx2,tx);
 359             fjy2             = _mm256_add_pd(fjy2,ty);
 360             fjz2             = _mm256_add_pd(fjz2,tz);
 361
 362             }
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 369             {
 370
 371             /* REACTION-FIELD ELECTROSTATICS */
 372             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
 373             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
 374
 375             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velec            = _mm256_and_pd(velec,cutoff_mask);
 379             velecsum         = _mm256_add_pd(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 384
 385             /* Calculate temporary vectorial force */
 386             tx               = _mm256_mul_pd(fscal,dx13);
 387             ty               = _mm256_mul_pd(fscal,dy13);
 388             tz               = _mm256_mul_pd(fscal,dz13);
 389
 390             /* Update vectorial force */
 391             fix1             = _mm256_add_pd(fix1,tx);
 392             fiy1             = _mm256_add_pd(fiy1,ty);
 393             fiz1             = _mm256_add_pd(fiz1,tz);
 394
 395             fjx3             = _mm256_add_pd(fjx3,tx);
 396             fjy3             = _mm256_add_pd(fjy3,ty);
 397             fjz3             = _mm256_add_pd(fjz3,tz);
 398
 399             }
 400
 401             /**************************
 402              * CALCULATE INTERACTIONS *
 403              **************************/
 404
 405             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 406             {
 407
 408             /* REACTION-FIELD ELECTROSTATICS */
 409             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 410             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 411
 412             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             velec            = _mm256_and_pd(velec,cutoff_mask);
 416             velecsum         = _mm256_add_pd(velecsum,velec);
 417
 418             fscal            = felec;
 419
 420             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 421
 422             /* Calculate temporary vectorial force */
 423             tx               = _mm256_mul_pd(fscal,dx21);
 424             ty               = _mm256_mul_pd(fscal,dy21);
 425             tz               = _mm256_mul_pd(fscal,dz21);
 426
 427             /* Update vectorial force */
 428             fix2             = _mm256_add_pd(fix2,tx);
 429             fiy2             = _mm256_add_pd(fiy2,ty);
 430             fiz2             = _mm256_add_pd(fiz2,tz);
 431
 432             fjx1             = _mm256_add_pd(fjx1,tx);
 433             fjy1             = _mm256_add_pd(fjy1,ty);
 434             fjz1             = _mm256_add_pd(fjz1,tz);
 435
 436             }
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 443             {
 444
 445             /* REACTION-FIELD ELECTROSTATICS */
 446             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 447             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 448
 449             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
 450
 451             /* Update potential sum for this i atom from the interaction with this j atom. */
 452             velec            = _mm256_and_pd(velec,cutoff_mask);
 453             velecsum         = _mm256_add_pd(velecsum,velec);
 454
 455             fscal            = felec;
 456
 457             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 458
 459             /* Calculate temporary vectorial force */
 460             tx               = _mm256_mul_pd(fscal,dx22);
 461             ty               = _mm256_mul_pd(fscal,dy22);
 462             tz               = _mm256_mul_pd(fscal,dz22);
 463
 464             /* Update vectorial force */
 465             fix2             = _mm256_add_pd(fix2,tx);
 466             fiy2             = _mm256_add_pd(fiy2,ty);
 467             fiz2             = _mm256_add_pd(fiz2,tz);
 468
 469             fjx2             = _mm256_add_pd(fjx2,tx);
 470             fjy2             = _mm256_add_pd(fjy2,ty);
 471             fjz2             = _mm256_add_pd(fjz2,tz);
 472
 473             }
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 480             {
 481
 482             /* REACTION-FIELD ELECTROSTATICS */
 483             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
 484             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
 485
 486             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
 487
 488             /* Update potential sum for this i atom from the interaction with this j atom. */
 489             velec            = _mm256_and_pd(velec,cutoff_mask);
 490             velecsum         = _mm256_add_pd(velecsum,velec);
 491
 492             fscal            = felec;
 493
 494             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 495
 496             /* Calculate temporary vectorial force */
 497             tx               = _mm256_mul_pd(fscal,dx23);
 498             ty               = _mm256_mul_pd(fscal,dy23);
 499             tz               = _mm256_mul_pd(fscal,dz23);
 500
 501             /* Update vectorial force */
 502             fix2             = _mm256_add_pd(fix2,tx);
 503             fiy2             = _mm256_add_pd(fiy2,ty);
 504             fiz2             = _mm256_add_pd(fiz2,tz);
 505
 506             fjx3             = _mm256_add_pd(fjx3,tx);
 507             fjy3             = _mm256_add_pd(fjy3,ty);
 508             fjz3             = _mm256_add_pd(fjz3,tz);
 509
 510             }
 511
 512             /**************************
 513              * CALCULATE INTERACTIONS *
 514              **************************/
 515
 516             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 517             {
 518
 519             /* REACTION-FIELD ELECTROSTATICS */
 520             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
 521             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
 522
 523             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
 524
 525             /* Update potential sum for this i atom from the interaction with this j atom. */
 526             velec            = _mm256_and_pd(velec,cutoff_mask);
 527             velecsum         = _mm256_add_pd(velecsum,velec);
 528
 529             fscal            = felec;
 530
 531             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 532
 533             /* Calculate temporary vectorial force */
 534             tx               = _mm256_mul_pd(fscal,dx31);
 535             ty               = _mm256_mul_pd(fscal,dy31);
 536             tz               = _mm256_mul_pd(fscal,dz31);
 537
 538             /* Update vectorial force */
 539             fix3             = _mm256_add_pd(fix3,tx);
 540             fiy3             = _mm256_add_pd(fiy3,ty);
 541             fiz3             = _mm256_add_pd(fiz3,tz);
 542
 543             fjx1             = _mm256_add_pd(fjx1,tx);
 544             fjy1             = _mm256_add_pd(fjy1,ty);
 545             fjz1             = _mm256_add_pd(fjz1,tz);
 546
 547             }
 548
 549             /**************************
 550              * CALCULATE INTERACTIONS *
 551              **************************/
 552
 553             if (gmx_mm256_any_lt(rsq32,rcutoff2))
 554             {
 555
 556             /* REACTION-FIELD ELECTROSTATICS */
 557             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
 558             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
 559
 560             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             velec            = _mm256_and_pd(velec,cutoff_mask);
 564             velecsum         = _mm256_add_pd(velecsum,velec);
 565
 566             fscal            = felec;
 567
 568             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 569
 570             /* Calculate temporary vectorial force */
 571             tx               = _mm256_mul_pd(fscal,dx32);
 572             ty               = _mm256_mul_pd(fscal,dy32);
 573             tz               = _mm256_mul_pd(fscal,dz32);
 574
 575             /* Update vectorial force */
 576             fix3             = _mm256_add_pd(fix3,tx);
 577             fiy3             = _mm256_add_pd(fiy3,ty);
 578             fiz3             = _mm256_add_pd(fiz3,tz);
 579
 580             fjx2             = _mm256_add_pd(fjx2,tx);
 581             fjy2             = _mm256_add_pd(fjy2,ty);
 582             fjz2             = _mm256_add_pd(fjz2,tz);
 583
 584             }
 585
 586             /**************************
 587              * CALCULATE INTERACTIONS *
 588              **************************/
 589
 590             if (gmx_mm256_any_lt(rsq33,rcutoff2))
 591             {
 592
 593             /* REACTION-FIELD ELECTROSTATICS */
 594             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
 595             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
 596
 597             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
 598
 599             /* Update potential sum for this i atom from the interaction with this j atom. */
 600             velec            = _mm256_and_pd(velec,cutoff_mask);
 601             velecsum         = _mm256_add_pd(velecsum,velec);
 602
 603             fscal            = felec;
 604
 605             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 606
 607             /* Calculate temporary vectorial force */
 608             tx               = _mm256_mul_pd(fscal,dx33);
 609             ty               = _mm256_mul_pd(fscal,dy33);
 610             tz               = _mm256_mul_pd(fscal,dz33);
 611
 612             /* Update vectorial force */
 613             fix3             = _mm256_add_pd(fix3,tx);
 614             fiy3             = _mm256_add_pd(fiy3,ty);
 615             fiz3             = _mm256_add_pd(fiz3,tz);
 616
 617             fjx3             = _mm256_add_pd(fjx3,tx);
 618             fjy3             = _mm256_add_pd(fjy3,ty);
 619             fjz3             = _mm256_add_pd(fjz3,tz);
 620
 621             }
 622
 623             fjptrA             = f+j_coord_offsetA;
 624             fjptrB             = f+j_coord_offsetB;
 625             fjptrC             = f+j_coord_offsetC;
 626             fjptrD             = f+j_coord_offsetD;
 627
 628             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 629                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 630
 631             /* Inner loop uses 324 flops */
 632         }
 633
 634         if(jidx<j_index_end)
 635         {
 636
 637             /* Get j neighbor index, and coordinate index */
 638             jnrlistA         = jjnr[jidx];
 639             jnrlistB         = jjnr[jidx+1];
 640             jnrlistC         = jjnr[jidx+2];
 641             jnrlistD         = jjnr[jidx+3];
 642             /* Sign of each element will be negative for non-real atoms.
 643              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 644              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 645              */
 646             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 647
 648             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 649             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 650             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 651
 652             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 653             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 654             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 655             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 656             j_coord_offsetA  = DIM*jnrA;
 657             j_coord_offsetB  = DIM*jnrB;
 658             j_coord_offsetC  = DIM*jnrC;
 659             j_coord_offsetD  = DIM*jnrD;
 660
 661             /* load j atom coordinates */
 662             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 663                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 664                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 665
 666             /* Calculate displacement vector */
 667             dx11             = _mm256_sub_pd(ix1,jx1);
 668             dy11             = _mm256_sub_pd(iy1,jy1);
 669             dz11             = _mm256_sub_pd(iz1,jz1);
 670             dx12             = _mm256_sub_pd(ix1,jx2);
 671             dy12             = _mm256_sub_pd(iy1,jy2);
 672             dz12             = _mm256_sub_pd(iz1,jz2);
 673             dx13             = _mm256_sub_pd(ix1,jx3);
 674             dy13             = _mm256_sub_pd(iy1,jy3);
 675             dz13             = _mm256_sub_pd(iz1,jz3);
 676             dx21             = _mm256_sub_pd(ix2,jx1);
 677             dy21             = _mm256_sub_pd(iy2,jy1);
 678             dz21             = _mm256_sub_pd(iz2,jz1);
 679             dx22             = _mm256_sub_pd(ix2,jx2);
 680             dy22             = _mm256_sub_pd(iy2,jy2);
 681             dz22             = _mm256_sub_pd(iz2,jz2);
 682             dx23             = _mm256_sub_pd(ix2,jx3);
 683             dy23             = _mm256_sub_pd(iy2,jy3);
 684             dz23             = _mm256_sub_pd(iz2,jz3);
 685             dx31             = _mm256_sub_pd(ix3,jx1);
 686             dy31             = _mm256_sub_pd(iy3,jy1);
 687             dz31             = _mm256_sub_pd(iz3,jz1);
 688             dx32             = _mm256_sub_pd(ix3,jx2);
 689             dy32             = _mm256_sub_pd(iy3,jy2);
 690             dz32             = _mm256_sub_pd(iz3,jz2);
 691             dx33             = _mm256_sub_pd(ix3,jx3);
 692             dy33             = _mm256_sub_pd(iy3,jy3);
 693             dz33             = _mm256_sub_pd(iz3,jz3);
 694
 695             /* Calculate squared distance and things based on it */
 696             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 697             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 698             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 699             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 700             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 701             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 702             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 703             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 704             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 705
 706             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 707             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 708             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 709             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 710             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 711             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 712             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 713             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 714             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 715
 716             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 717             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 718             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 719             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 720             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 721             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 722             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 723             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 724             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 725
 726             fjx1             = _mm256_setzero_pd();
 727             fjy1             = _mm256_setzero_pd();
 728             fjz1             = _mm256_setzero_pd();
 729             fjx2             = _mm256_setzero_pd();
 730             fjy2             = _mm256_setzero_pd();
 731             fjz2             = _mm256_setzero_pd();
 732             fjx3             = _mm256_setzero_pd();
 733             fjy3             = _mm256_setzero_pd();
 734             fjz3             = _mm256_setzero_pd();
 735
 736             /**************************
 737              * CALCULATE INTERACTIONS *
 738              **************************/
 739
 740             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 741             {
 742
 743             /* REACTION-FIELD ELECTROSTATICS */
 744             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 745             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 746
 747             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 748
 749             /* Update potential sum for this i atom from the interaction with this j atom. */
 750             velec            = _mm256_and_pd(velec,cutoff_mask);
 751             velec            = _mm256_andnot_pd(dummy_mask,velec);
 752             velecsum         = _mm256_add_pd(velecsum,velec);
 753
 754             fscal            = felec;
 755
 756             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 757
 758             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 759
 760             /* Calculate temporary vectorial force */
 761             tx               = _mm256_mul_pd(fscal,dx11);
 762             ty               = _mm256_mul_pd(fscal,dy11);
 763             tz               = _mm256_mul_pd(fscal,dz11);
 764
 765             /* Update vectorial force */
 766             fix1             = _mm256_add_pd(fix1,tx);
 767             fiy1             = _mm256_add_pd(fiy1,ty);
 768             fiz1             = _mm256_add_pd(fiz1,tz);
 769
 770             fjx1             = _mm256_add_pd(fjx1,tx);
 771             fjy1             = _mm256_add_pd(fjy1,ty);
 772             fjz1             = _mm256_add_pd(fjz1,tz);
 773
 774             }
 775
 776             /**************************
 777              * CALCULATE INTERACTIONS *
 778              **************************/
 779
 780             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 781             {
 782
 783             /* REACTION-FIELD ELECTROSTATICS */
 784             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 785             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 786
 787             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 788
 789             /* Update potential sum for this i atom from the interaction with this j atom. */
 790             velec            = _mm256_and_pd(velec,cutoff_mask);
 791             velec            = _mm256_andnot_pd(dummy_mask,velec);
 792             velecsum         = _mm256_add_pd(velecsum,velec);
 793
 794             fscal            = felec;
 795
 796             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 797
 798             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 799
 800             /* Calculate temporary vectorial force */
 801             tx               = _mm256_mul_pd(fscal,dx12);
 802             ty               = _mm256_mul_pd(fscal,dy12);
 803             tz               = _mm256_mul_pd(fscal,dz12);
 804
 805             /* Update vectorial force */
 806             fix1             = _mm256_add_pd(fix1,tx);
 807             fiy1             = _mm256_add_pd(fiy1,ty);
 808             fiz1             = _mm256_add_pd(fiz1,tz);
 809
 810             fjx2             = _mm256_add_pd(fjx2,tx);
 811             fjy2             = _mm256_add_pd(fjy2,ty);
 812             fjz2             = _mm256_add_pd(fjz2,tz);
 813
 814             }
 815
 816             /**************************
 817              * CALCULATE INTERACTIONS *
 818              **************************/
 819
 820             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 821             {
 822
 823             /* REACTION-FIELD ELECTROSTATICS */
 824             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
 825             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
 826
 827             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
 828
 829             /* Update potential sum for this i atom from the interaction with this j atom. */
 830             velec            = _mm256_and_pd(velec,cutoff_mask);
 831             velec            = _mm256_andnot_pd(dummy_mask,velec);
 832             velecsum         = _mm256_add_pd(velecsum,velec);
 833
 834             fscal            = felec;
 835
 836             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 837
 838             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 839
 840             /* Calculate temporary vectorial force */
 841             tx               = _mm256_mul_pd(fscal,dx13);
 842             ty               = _mm256_mul_pd(fscal,dy13);
 843             tz               = _mm256_mul_pd(fscal,dz13);
 844
 845             /* Update vectorial force */
 846             fix1             = _mm256_add_pd(fix1,tx);
 847             fiy1             = _mm256_add_pd(fiy1,ty);
 848             fiz1             = _mm256_add_pd(fiz1,tz);
 849
 850             fjx3             = _mm256_add_pd(fjx3,tx);
 851             fjy3             = _mm256_add_pd(fjy3,ty);
 852             fjz3             = _mm256_add_pd(fjz3,tz);
 853
 854             }
 855
 856             /**************************
 857              * CALCULATE INTERACTIONS *
 858              **************************/
 859
 860             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 861             {
 862
 863             /* REACTION-FIELD ELECTROSTATICS */
 864             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 865             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 866
 867             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
 868
 869             /* Update potential sum for this i atom from the interaction with this j atom. */
 870             velec            = _mm256_and_pd(velec,cutoff_mask);
 871             velec            = _mm256_andnot_pd(dummy_mask,velec);
 872             velecsum         = _mm256_add_pd(velecsum,velec);
 873
 874             fscal            = felec;
 875
 876             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 877
 878             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 879
 880             /* Calculate temporary vectorial force */
 881             tx               = _mm256_mul_pd(fscal,dx21);
 882             ty               = _mm256_mul_pd(fscal,dy21);
 883             tz               = _mm256_mul_pd(fscal,dz21);
 884
 885             /* Update vectorial force */
 886             fix2             = _mm256_add_pd(fix2,tx);
 887             fiy2             = _mm256_add_pd(fiy2,ty);
 888             fiz2             = _mm256_add_pd(fiz2,tz);
 889
 890             fjx1             = _mm256_add_pd(fjx1,tx);
 891             fjy1             = _mm256_add_pd(fjy1,ty);
 892             fjz1             = _mm256_add_pd(fjz1,tz);
 893
 894             }
 895
 896             /**************************
 897              * CALCULATE INTERACTIONS *
 898              **************************/
 899
 900             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 901             {
 902
 903             /* REACTION-FIELD ELECTROSTATICS */
 904             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 905             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 906
 907             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
 908
 909             /* Update potential sum for this i atom from the interaction with this j atom. */
 910             velec            = _mm256_and_pd(velec,cutoff_mask);
 911             velec            = _mm256_andnot_pd(dummy_mask,velec);
 912             velecsum         = _mm256_add_pd(velecsum,velec);
 913
 914             fscal            = felec;
 915
 916             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 917
 918             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 919
 920             /* Calculate temporary vectorial force */
 921             tx               = _mm256_mul_pd(fscal,dx22);
 922             ty               = _mm256_mul_pd(fscal,dy22);
 923             tz               = _mm256_mul_pd(fscal,dz22);
 924
 925             /* Update vectorial force */
 926             fix2             = _mm256_add_pd(fix2,tx);
 927             fiy2             = _mm256_add_pd(fiy2,ty);
 928             fiz2             = _mm256_add_pd(fiz2,tz);
 929
 930             fjx2             = _mm256_add_pd(fjx2,tx);
 931             fjy2             = _mm256_add_pd(fjy2,ty);
 932             fjz2             = _mm256_add_pd(fjz2,tz);
 933
 934             }
 935
 936             /**************************
 937              * CALCULATE INTERACTIONS *
 938              **************************/
 939
 940             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 941             {
 942
 943             /* REACTION-FIELD ELECTROSTATICS */
 944             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
 945             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
 946
 947             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
 948
 949             /* Update potential sum for this i atom from the interaction with this j atom. */
 950             velec            = _mm256_and_pd(velec,cutoff_mask);
 951             velec            = _mm256_andnot_pd(dummy_mask,velec);
 952             velecsum         = _mm256_add_pd(velecsum,velec);
 953
 954             fscal            = felec;
 955
 956             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 957
 958             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 959
 960             /* Calculate temporary vectorial force */
 961             tx               = _mm256_mul_pd(fscal,dx23);
 962             ty               = _mm256_mul_pd(fscal,dy23);
 963             tz               = _mm256_mul_pd(fscal,dz23);
 964
 965             /* Update vectorial force */
 966             fix2             = _mm256_add_pd(fix2,tx);
 967             fiy2             = _mm256_add_pd(fiy2,ty);
 968             fiz2             = _mm256_add_pd(fiz2,tz);
 969
 970             fjx3             = _mm256_add_pd(fjx3,tx);
 971             fjy3             = _mm256_add_pd(fjy3,ty);
 972             fjz3             = _mm256_add_pd(fjz3,tz);
 973
 974             }
 975
 976             /**************************
 977              * CALCULATE INTERACTIONS *
 978              **************************/
 979
 980             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 981             {
 982
 983             /* REACTION-FIELD ELECTROSTATICS */
 984             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
 985             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
 986
 987             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
 988
 989             /* Update potential sum for this i atom from the interaction with this j atom. */
 990             velec            = _mm256_and_pd(velec,cutoff_mask);
 991             velec            = _mm256_andnot_pd(dummy_mask,velec);
 992             velecsum         = _mm256_add_pd(velecsum,velec);
 993
 994             fscal            = felec;
 995
 996             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 997
 998             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 999
1000             /* Calculate temporary vectorial force */
1001             tx               = _mm256_mul_pd(fscal,dx31);
1002             ty               = _mm256_mul_pd(fscal,dy31);
1003             tz               = _mm256_mul_pd(fscal,dz31);
1004
1005             /* Update vectorial force */
1006             fix3             = _mm256_add_pd(fix3,tx);
1007             fiy3             = _mm256_add_pd(fiy3,ty);
1008             fiz3             = _mm256_add_pd(fiz3,tz);
1009
1010             fjx1             = _mm256_add_pd(fjx1,tx);
1011             fjy1             = _mm256_add_pd(fjy1,ty);
1012             fjz1             = _mm256_add_pd(fjz1,tz);
1013
1014             }
1015
1016             /**************************
1017              * CALCULATE INTERACTIONS *
1018              **************************/
1019
1020             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1021             {
1022
1023             /* REACTION-FIELD ELECTROSTATICS */
1024             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
1025             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1026
1027             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
1028
1029             /* Update potential sum for this i atom from the interaction with this j atom. */
1030             velec            = _mm256_and_pd(velec,cutoff_mask);
1031             velec            = _mm256_andnot_pd(dummy_mask,velec);
1032             velecsum         = _mm256_add_pd(velecsum,velec);
1033
1034             fscal            = felec;
1035
1036             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1037
1038             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1039
1040             /* Calculate temporary vectorial force */
1041             tx               = _mm256_mul_pd(fscal,dx32);
1042             ty               = _mm256_mul_pd(fscal,dy32);
1043             tz               = _mm256_mul_pd(fscal,dz32);
1044
1045             /* Update vectorial force */
1046             fix3             = _mm256_add_pd(fix3,tx);
1047             fiy3             = _mm256_add_pd(fiy3,ty);
1048             fiz3             = _mm256_add_pd(fiz3,tz);
1049
1050             fjx2             = _mm256_add_pd(fjx2,tx);
1051             fjy2             = _mm256_add_pd(fjy2,ty);
1052             fjz2             = _mm256_add_pd(fjz2,tz);
1053
1054             }
1055
1056             /**************************
1057              * CALCULATE INTERACTIONS *
1058              **************************/
1059
1060             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1061             {
1062
1063             /* REACTION-FIELD ELECTROSTATICS */
1064             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
1065             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1066
1067             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
1068
1069             /* Update potential sum for this i atom from the interaction with this j atom. */
1070             velec            = _mm256_and_pd(velec,cutoff_mask);
1071             velec            = _mm256_andnot_pd(dummy_mask,velec);
1072             velecsum         = _mm256_add_pd(velecsum,velec);
1073
1074             fscal            = felec;
1075
1076             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1077
1078             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1079
1080             /* Calculate temporary vectorial force */
1081             tx               = _mm256_mul_pd(fscal,dx33);
1082             ty               = _mm256_mul_pd(fscal,dy33);
1083             tz               = _mm256_mul_pd(fscal,dz33);
1084
1085             /* Update vectorial force */
1086             fix3             = _mm256_add_pd(fix3,tx);
1087             fiy3             = _mm256_add_pd(fiy3,ty);
1088             fiz3             = _mm256_add_pd(fiz3,tz);
1089
1090             fjx3             = _mm256_add_pd(fjx3,tx);
1091             fjy3             = _mm256_add_pd(fjy3,ty);
1092             fjz3             = _mm256_add_pd(fjz3,tz);
1093
1094             }
1095
1096             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1097             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1098             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1099             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1100
1101             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1102                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1103
1104             /* Inner loop uses 324 flops */
1105         }
1106
1107         /* End of innermost loop */
1108
1109         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1110                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1111
1112         ggid                        = gid[iidx];
1113         /* Update potential energies */
1114         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1115
1116         /* Increment number of inner iterations */
1117         inneriter                  += j_index_end - j_index_start;
1118
1119         /* Outer loop uses 19 flops */
1120     }
1121
1122     /* Increment number of outer iterations */
1123     outeriter        += nri;
1124
1125     /* Update outer/inner flops */
1126
1127     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*324);
1128 }
1129 /*
1130  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4W4_F_avx_256_double
1131  * Electrostatics interaction: ReactionField
1132  * VdW interaction:            None
1133  * Geometry:                   Water4-Water4
1134  * Calculate force/pot:        Force
1135  */
1136 void
1137 nb_kernel_ElecRFCut_VdwNone_GeomW4W4_F_avx_256_double
1138                     (t_nblist                    * gmx_restrict       nlist,
1139                      rvec                        * gmx_restrict          xx,
1140                      rvec                        * gmx_restrict          ff,
1141                      t_forcerec                  * gmx_restrict          fr,
1142                      t_mdatoms                   * gmx_restrict     mdatoms,
1143                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1144                      t_nrnb                      * gmx_restrict        nrnb)
1145 {
1146     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1147      * just 0 for non-waters.
1148      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1149      * jnr indices corresponding to data put in the four positions in the SIMD register.
1150      */
1151     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1152     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1153     int              jnrA,jnrB,jnrC,jnrD;
1154     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1155     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1156     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1157     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1158     real             rcutoff_scalar;
1159     real             *shiftvec,*fshift,*x,*f;
1160     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1161     real             scratch[4*DIM];
1162     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1163     real *           vdwioffsetptr1;
1164     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1165     real *           vdwioffsetptr2;
1166     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1167     real *           vdwioffsetptr3;
1168     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1169     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1170     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1171     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1172     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1173     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1174     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1175     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1176     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1177     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1178     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1179     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1180     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1181     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1182     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1183     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1184     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1185     real             *charge;
1186     __m256d          dummy_mask,cutoff_mask;
1187     __m128           tmpmask0,tmpmask1;
1188     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1189     __m256d          one     = _mm256_set1_pd(1.0);
1190     __m256d          two     = _mm256_set1_pd(2.0);
1191     x                = xx[0];
1192     f                = ff[0];
1193
1194     nri              = nlist->nri;
1195     iinr             = nlist->iinr;
1196     jindex           = nlist->jindex;
1197     jjnr             = nlist->jjnr;
1198     shiftidx         = nlist->shift;
1199     gid              = nlist->gid;
1200     shiftvec         = fr->shift_vec[0];
1201     fshift           = fr->fshift[0];
1202     facel            = _mm256_set1_pd(fr->epsfac);
1203     charge           = mdatoms->chargeA;
1204     krf              = _mm256_set1_pd(fr->ic->k_rf);
1205     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
1206     crf              = _mm256_set1_pd(fr->ic->c_rf);
1207
1208     /* Setup water-specific parameters */
1209     inr              = nlist->iinr[0];
1210     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1211     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1212     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1213
1214     jq1              = _mm256_set1_pd(charge[inr+1]);
1215     jq2              = _mm256_set1_pd(charge[inr+2]);
1216     jq3              = _mm256_set1_pd(charge[inr+3]);
1217     qq11             = _mm256_mul_pd(iq1,jq1);
1218     qq12             = _mm256_mul_pd(iq1,jq2);
1219     qq13             = _mm256_mul_pd(iq1,jq3);
1220     qq21             = _mm256_mul_pd(iq2,jq1);
1221     qq22             = _mm256_mul_pd(iq2,jq2);
1222     qq23             = _mm256_mul_pd(iq2,jq3);
1223     qq31             = _mm256_mul_pd(iq3,jq1);
1224     qq32             = _mm256_mul_pd(iq3,jq2);
1225     qq33             = _mm256_mul_pd(iq3,jq3);
1226
1227     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1228     rcutoff_scalar   = fr->rcoulomb;
1229     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
1230     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
1231
1232     /* Avoid stupid compiler warnings */
1233     jnrA = jnrB = jnrC = jnrD = 0;
1234     j_coord_offsetA = 0;
1235     j_coord_offsetB = 0;
1236     j_coord_offsetC = 0;
1237     j_coord_offsetD = 0;
1238
1239     outeriter        = 0;
1240     inneriter        = 0;
1241
1242     for(iidx=0;iidx<4*DIM;iidx++)
1243     {
1244         scratch[iidx] = 0.0;
1245     }
1246
1247     /* Start outer loop over neighborlists */
1248     for(iidx=0; iidx<nri; iidx++)
1249     {
1250         /* Load shift vector for this list */
1251         i_shift_offset   = DIM*shiftidx[iidx];
1252
1253         /* Load limits for loop over neighbors */
1254         j_index_start    = jindex[iidx];
1255         j_index_end      = jindex[iidx+1];
1256
1257         /* Get outer coordinate index */
1258         inr              = iinr[iidx];
1259         i_coord_offset   = DIM*inr;
1260
1261         /* Load i particle coords and add shift vector */
1262         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1263                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1264
1265         fix1             = _mm256_setzero_pd();
1266         fiy1             = _mm256_setzero_pd();
1267         fiz1             = _mm256_setzero_pd();
1268         fix2             = _mm256_setzero_pd();
1269         fiy2             = _mm256_setzero_pd();
1270         fiz2             = _mm256_setzero_pd();
1271         fix3             = _mm256_setzero_pd();
1272         fiy3             = _mm256_setzero_pd();
1273         fiz3             = _mm256_setzero_pd();
1274
1275         /* Start inner kernel loop */
1276         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1277         {
1278
1279             /* Get j neighbor index, and coordinate index */
1280             jnrA             = jjnr[jidx];
1281             jnrB             = jjnr[jidx+1];
1282             jnrC             = jjnr[jidx+2];
1283             jnrD             = jjnr[jidx+3];
1284             j_coord_offsetA  = DIM*jnrA;
1285             j_coord_offsetB  = DIM*jnrB;
1286             j_coord_offsetC  = DIM*jnrC;
1287             j_coord_offsetD  = DIM*jnrD;
1288
1289             /* load j atom coordinates */
1290             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1291                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1292                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1293
1294             /* Calculate displacement vector */
1295             dx11             = _mm256_sub_pd(ix1,jx1);
1296             dy11             = _mm256_sub_pd(iy1,jy1);
1297             dz11             = _mm256_sub_pd(iz1,jz1);
1298             dx12             = _mm256_sub_pd(ix1,jx2);
1299             dy12             = _mm256_sub_pd(iy1,jy2);
1300             dz12             = _mm256_sub_pd(iz1,jz2);
1301             dx13             = _mm256_sub_pd(ix1,jx3);
1302             dy13             = _mm256_sub_pd(iy1,jy3);
1303             dz13             = _mm256_sub_pd(iz1,jz3);
1304             dx21             = _mm256_sub_pd(ix2,jx1);
1305             dy21             = _mm256_sub_pd(iy2,jy1);
1306             dz21             = _mm256_sub_pd(iz2,jz1);
1307             dx22             = _mm256_sub_pd(ix2,jx2);
1308             dy22             = _mm256_sub_pd(iy2,jy2);
1309             dz22             = _mm256_sub_pd(iz2,jz2);
1310             dx23             = _mm256_sub_pd(ix2,jx3);
1311             dy23             = _mm256_sub_pd(iy2,jy3);
1312             dz23             = _mm256_sub_pd(iz2,jz3);
1313             dx31             = _mm256_sub_pd(ix3,jx1);
1314             dy31             = _mm256_sub_pd(iy3,jy1);
1315             dz31             = _mm256_sub_pd(iz3,jz1);
1316             dx32             = _mm256_sub_pd(ix3,jx2);
1317             dy32             = _mm256_sub_pd(iy3,jy2);
1318             dz32             = _mm256_sub_pd(iz3,jz2);
1319             dx33             = _mm256_sub_pd(ix3,jx3);
1320             dy33             = _mm256_sub_pd(iy3,jy3);
1321             dz33             = _mm256_sub_pd(iz3,jz3);
1322
1323             /* Calculate squared distance and things based on it */
1324             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1325             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1326             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1327             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1328             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1329             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1330             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1331             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1332             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1333
1334             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1335             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1336             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1337             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1338             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1339             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1340             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1341             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1342             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1343
1344             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1345             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1346             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1347             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1348             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1349             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1350             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1351             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1352             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1353
1354             fjx1             = _mm256_setzero_pd();
1355             fjy1             = _mm256_setzero_pd();
1356             fjz1             = _mm256_setzero_pd();
1357             fjx2             = _mm256_setzero_pd();
1358             fjy2             = _mm256_setzero_pd();
1359             fjz2             = _mm256_setzero_pd();
1360             fjx3             = _mm256_setzero_pd();
1361             fjy3             = _mm256_setzero_pd();
1362             fjz3             = _mm256_setzero_pd();
1363
1364             /**************************
1365              * CALCULATE INTERACTIONS *
1366              **************************/
1367
1368             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1369             {
1370
1371             /* REACTION-FIELD ELECTROSTATICS */
1372             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1373
1374             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1375
1376             fscal            = felec;
1377
1378             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1379
1380             /* Calculate temporary vectorial force */
1381             tx               = _mm256_mul_pd(fscal,dx11);
1382             ty               = _mm256_mul_pd(fscal,dy11);
1383             tz               = _mm256_mul_pd(fscal,dz11);
1384
1385             /* Update vectorial force */
1386             fix1             = _mm256_add_pd(fix1,tx);
1387             fiy1             = _mm256_add_pd(fiy1,ty);
1388             fiz1             = _mm256_add_pd(fiz1,tz);
1389
1390             fjx1             = _mm256_add_pd(fjx1,tx);
1391             fjy1             = _mm256_add_pd(fjy1,ty);
1392             fjz1             = _mm256_add_pd(fjz1,tz);
1393
1394             }
1395
1396             /**************************
1397              * CALCULATE INTERACTIONS *
1398              **************************/
1399
1400             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1401             {
1402
1403             /* REACTION-FIELD ELECTROSTATICS */
1404             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1405
1406             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1407
1408             fscal            = felec;
1409
1410             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1411
1412             /* Calculate temporary vectorial force */
1413             tx               = _mm256_mul_pd(fscal,dx12);
1414             ty               = _mm256_mul_pd(fscal,dy12);
1415             tz               = _mm256_mul_pd(fscal,dz12);
1416
1417             /* Update vectorial force */
1418             fix1             = _mm256_add_pd(fix1,tx);
1419             fiy1             = _mm256_add_pd(fiy1,ty);
1420             fiz1             = _mm256_add_pd(fiz1,tz);
1421
1422             fjx2             = _mm256_add_pd(fjx2,tx);
1423             fjy2             = _mm256_add_pd(fjy2,ty);
1424             fjz2             = _mm256_add_pd(fjz2,tz);
1425
1426             }
1427
1428             /**************************
1429              * CALCULATE INTERACTIONS *
1430              **************************/
1431
1432             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1433             {
1434
1435             /* REACTION-FIELD ELECTROSTATICS */
1436             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
1437
1438             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
1439
1440             fscal            = felec;
1441
1442             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1443
1444             /* Calculate temporary vectorial force */
1445             tx               = _mm256_mul_pd(fscal,dx13);
1446             ty               = _mm256_mul_pd(fscal,dy13);
1447             tz               = _mm256_mul_pd(fscal,dz13);
1448
1449             /* Update vectorial force */
1450             fix1             = _mm256_add_pd(fix1,tx);
1451             fiy1             = _mm256_add_pd(fiy1,ty);
1452             fiz1             = _mm256_add_pd(fiz1,tz);
1453
1454             fjx3             = _mm256_add_pd(fjx3,tx);
1455             fjy3             = _mm256_add_pd(fjy3,ty);
1456             fjz3             = _mm256_add_pd(fjz3,tz);
1457
1458             }
1459
1460             /**************************
1461              * CALCULATE INTERACTIONS *
1462              **************************/
1463
1464             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1465             {
1466
1467             /* REACTION-FIELD ELECTROSTATICS */
1468             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1469
1470             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1471
1472             fscal            = felec;
1473
1474             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1475
1476             /* Calculate temporary vectorial force */
1477             tx               = _mm256_mul_pd(fscal,dx21);
1478             ty               = _mm256_mul_pd(fscal,dy21);
1479             tz               = _mm256_mul_pd(fscal,dz21);
1480
1481             /* Update vectorial force */
1482             fix2             = _mm256_add_pd(fix2,tx);
1483             fiy2             = _mm256_add_pd(fiy2,ty);
1484             fiz2             = _mm256_add_pd(fiz2,tz);
1485
1486             fjx1             = _mm256_add_pd(fjx1,tx);
1487             fjy1             = _mm256_add_pd(fjy1,ty);
1488             fjz1             = _mm256_add_pd(fjz1,tz);
1489
1490             }
1491
1492             /**************************
1493              * CALCULATE INTERACTIONS *
1494              **************************/
1495
1496             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1497             {
1498
1499             /* REACTION-FIELD ELECTROSTATICS */
1500             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1501
1502             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1503
1504             fscal            = felec;
1505
1506             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1507
1508             /* Calculate temporary vectorial force */
1509             tx               = _mm256_mul_pd(fscal,dx22);
1510             ty               = _mm256_mul_pd(fscal,dy22);
1511             tz               = _mm256_mul_pd(fscal,dz22);
1512
1513             /* Update vectorial force */
1514             fix2             = _mm256_add_pd(fix2,tx);
1515             fiy2             = _mm256_add_pd(fiy2,ty);
1516             fiz2             = _mm256_add_pd(fiz2,tz);
1517
1518             fjx2             = _mm256_add_pd(fjx2,tx);
1519             fjy2             = _mm256_add_pd(fjy2,ty);
1520             fjz2             = _mm256_add_pd(fjz2,tz);
1521
1522             }
1523
1524             /**************************
1525              * CALCULATE INTERACTIONS *
1526              **************************/
1527
1528             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1529             {
1530
1531             /* REACTION-FIELD ELECTROSTATICS */
1532             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1533
1534             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
1535
1536             fscal            = felec;
1537
1538             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1539
1540             /* Calculate temporary vectorial force */
1541             tx               = _mm256_mul_pd(fscal,dx23);
1542             ty               = _mm256_mul_pd(fscal,dy23);
1543             tz               = _mm256_mul_pd(fscal,dz23);
1544
1545             /* Update vectorial force */
1546             fix2             = _mm256_add_pd(fix2,tx);
1547             fiy2             = _mm256_add_pd(fiy2,ty);
1548             fiz2             = _mm256_add_pd(fiz2,tz);
1549
1550             fjx3             = _mm256_add_pd(fjx3,tx);
1551             fjy3             = _mm256_add_pd(fjy3,ty);
1552             fjz3             = _mm256_add_pd(fjz3,tz);
1553
1554             }
1555
1556             /**************************
1557              * CALCULATE INTERACTIONS *
1558              **************************/
1559
1560             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1561             {
1562
1563             /* REACTION-FIELD ELECTROSTATICS */
1564             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1565
1566             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
1567
1568             fscal            = felec;
1569
1570             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1571
1572             /* Calculate temporary vectorial force */
1573             tx               = _mm256_mul_pd(fscal,dx31);
1574             ty               = _mm256_mul_pd(fscal,dy31);
1575             tz               = _mm256_mul_pd(fscal,dz31);
1576
1577             /* Update vectorial force */
1578             fix3             = _mm256_add_pd(fix3,tx);
1579             fiy3             = _mm256_add_pd(fiy3,ty);
1580             fiz3             = _mm256_add_pd(fiz3,tz);
1581
1582             fjx1             = _mm256_add_pd(fjx1,tx);
1583             fjy1             = _mm256_add_pd(fjy1,ty);
1584             fjz1             = _mm256_add_pd(fjz1,tz);
1585
1586             }
1587
1588             /**************************
1589              * CALCULATE INTERACTIONS *
1590              **************************/
1591
1592             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1593             {
1594
1595             /* REACTION-FIELD ELECTROSTATICS */
1596             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1597
1598             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
1599
1600             fscal            = felec;
1601
1602             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1603
1604             /* Calculate temporary vectorial force */
1605             tx               = _mm256_mul_pd(fscal,dx32);
1606             ty               = _mm256_mul_pd(fscal,dy32);
1607             tz               = _mm256_mul_pd(fscal,dz32);
1608
1609             /* Update vectorial force */
1610             fix3             = _mm256_add_pd(fix3,tx);
1611             fiy3             = _mm256_add_pd(fiy3,ty);
1612             fiz3             = _mm256_add_pd(fiz3,tz);
1613
1614             fjx2             = _mm256_add_pd(fjx2,tx);
1615             fjy2             = _mm256_add_pd(fjy2,ty);
1616             fjz2             = _mm256_add_pd(fjz2,tz);
1617
1618             }
1619
1620             /**************************
1621              * CALCULATE INTERACTIONS *
1622              **************************/
1623
1624             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1625             {
1626
1627             /* REACTION-FIELD ELECTROSTATICS */
1628             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1629
1630             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
1631
1632             fscal            = felec;
1633
1634             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1635
1636             /* Calculate temporary vectorial force */
1637             tx               = _mm256_mul_pd(fscal,dx33);
1638             ty               = _mm256_mul_pd(fscal,dy33);
1639             tz               = _mm256_mul_pd(fscal,dz33);
1640
1641             /* Update vectorial force */
1642             fix3             = _mm256_add_pd(fix3,tx);
1643             fiy3             = _mm256_add_pd(fiy3,ty);
1644             fiz3             = _mm256_add_pd(fiz3,tz);
1645
1646             fjx3             = _mm256_add_pd(fjx3,tx);
1647             fjy3             = _mm256_add_pd(fjy3,ty);
1648             fjz3             = _mm256_add_pd(fjz3,tz);
1649
1650             }
1651
1652             fjptrA             = f+j_coord_offsetA;
1653             fjptrB             = f+j_coord_offsetB;
1654             fjptrC             = f+j_coord_offsetC;
1655             fjptrD             = f+j_coord_offsetD;
1656
1657             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1658                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1659
1660             /* Inner loop uses 270 flops */
1661         }
1662
1663         if(jidx<j_index_end)
1664         {
1665
1666             /* Get j neighbor index, and coordinate index */
1667             jnrlistA         = jjnr[jidx];
1668             jnrlistB         = jjnr[jidx+1];
1669             jnrlistC         = jjnr[jidx+2];
1670             jnrlistD         = jjnr[jidx+3];
1671             /* Sign of each element will be negative for non-real atoms.
1672              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1673              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1674              */
1675             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1676
1677             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1678             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1679             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1680
1681             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1682             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1683             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1684             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1685             j_coord_offsetA  = DIM*jnrA;
1686             j_coord_offsetB  = DIM*jnrB;
1687             j_coord_offsetC  = DIM*jnrC;
1688             j_coord_offsetD  = DIM*jnrD;
1689
1690             /* load j atom coordinates */
1691             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1692                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1693                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1694
1695             /* Calculate displacement vector */
1696             dx11             = _mm256_sub_pd(ix1,jx1);
1697             dy11             = _mm256_sub_pd(iy1,jy1);
1698             dz11             = _mm256_sub_pd(iz1,jz1);
1699             dx12             = _mm256_sub_pd(ix1,jx2);
1700             dy12             = _mm256_sub_pd(iy1,jy2);
1701             dz12             = _mm256_sub_pd(iz1,jz2);
1702             dx13             = _mm256_sub_pd(ix1,jx3);
1703             dy13             = _mm256_sub_pd(iy1,jy3);
1704             dz13             = _mm256_sub_pd(iz1,jz3);
1705             dx21             = _mm256_sub_pd(ix2,jx1);
1706             dy21             = _mm256_sub_pd(iy2,jy1);
1707             dz21             = _mm256_sub_pd(iz2,jz1);
1708             dx22             = _mm256_sub_pd(ix2,jx2);
1709             dy22             = _mm256_sub_pd(iy2,jy2);
1710             dz22             = _mm256_sub_pd(iz2,jz2);
1711             dx23             = _mm256_sub_pd(ix2,jx3);
1712             dy23             = _mm256_sub_pd(iy2,jy3);
1713             dz23             = _mm256_sub_pd(iz2,jz3);
1714             dx31             = _mm256_sub_pd(ix3,jx1);
1715             dy31             = _mm256_sub_pd(iy3,jy1);
1716             dz31             = _mm256_sub_pd(iz3,jz1);
1717             dx32             = _mm256_sub_pd(ix3,jx2);
1718             dy32             = _mm256_sub_pd(iy3,jy2);
1719             dz32             = _mm256_sub_pd(iz3,jz2);
1720             dx33             = _mm256_sub_pd(ix3,jx3);
1721             dy33             = _mm256_sub_pd(iy3,jy3);
1722             dz33             = _mm256_sub_pd(iz3,jz3);
1723
1724             /* Calculate squared distance and things based on it */
1725             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1726             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1727             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1728             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1729             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1730             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1731             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1732             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1733             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1734
1735             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1736             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1737             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1738             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1739             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1740             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1741             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1742             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1743             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1744
1745             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1746             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1747             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1748             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1749             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1750             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1751             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1752             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1753             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1754
1755             fjx1             = _mm256_setzero_pd();
1756             fjy1             = _mm256_setzero_pd();
1757             fjz1             = _mm256_setzero_pd();
1758             fjx2             = _mm256_setzero_pd();
1759             fjy2             = _mm256_setzero_pd();
1760             fjz2             = _mm256_setzero_pd();
1761             fjx3             = _mm256_setzero_pd();
1762             fjy3             = _mm256_setzero_pd();
1763             fjz3             = _mm256_setzero_pd();
1764
1765             /**************************
1766              * CALCULATE INTERACTIONS *
1767              **************************/
1768
1769             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1770             {
1771
1772             /* REACTION-FIELD ELECTROSTATICS */
1773             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1774
1775             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1776
1777             fscal            = felec;
1778
1779             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1780
1781             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1782
1783             /* Calculate temporary vectorial force */
1784             tx               = _mm256_mul_pd(fscal,dx11);
1785             ty               = _mm256_mul_pd(fscal,dy11);
1786             tz               = _mm256_mul_pd(fscal,dz11);
1787
1788             /* Update vectorial force */
1789             fix1             = _mm256_add_pd(fix1,tx);
1790             fiy1             = _mm256_add_pd(fiy1,ty);
1791             fiz1             = _mm256_add_pd(fiz1,tz);
1792
1793             fjx1             = _mm256_add_pd(fjx1,tx);
1794             fjy1             = _mm256_add_pd(fjy1,ty);
1795             fjz1             = _mm256_add_pd(fjz1,tz);
1796
1797             }
1798
1799             /**************************
1800              * CALCULATE INTERACTIONS *
1801              **************************/
1802
1803             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1804             {
1805
1806             /* REACTION-FIELD ELECTROSTATICS */
1807             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1808
1809             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1810
1811             fscal            = felec;
1812
1813             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1814
1815             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1816
1817             /* Calculate temporary vectorial force */
1818             tx               = _mm256_mul_pd(fscal,dx12);
1819             ty               = _mm256_mul_pd(fscal,dy12);
1820             tz               = _mm256_mul_pd(fscal,dz12);
1821
1822             /* Update vectorial force */
1823             fix1             = _mm256_add_pd(fix1,tx);
1824             fiy1             = _mm256_add_pd(fiy1,ty);
1825             fiz1             = _mm256_add_pd(fiz1,tz);
1826
1827             fjx2             = _mm256_add_pd(fjx2,tx);
1828             fjy2             = _mm256_add_pd(fjy2,ty);
1829             fjz2             = _mm256_add_pd(fjz2,tz);
1830
1831             }
1832
1833             /**************************
1834              * CALCULATE INTERACTIONS *
1835              **************************/
1836
1837             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1838             {
1839
1840             /* REACTION-FIELD ELECTROSTATICS */
1841             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
1842
1843             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
1844
1845             fscal            = felec;
1846
1847             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1848
1849             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1850
1851             /* Calculate temporary vectorial force */
1852             tx               = _mm256_mul_pd(fscal,dx13);
1853             ty               = _mm256_mul_pd(fscal,dy13);
1854             tz               = _mm256_mul_pd(fscal,dz13);
1855
1856             /* Update vectorial force */
1857             fix1             = _mm256_add_pd(fix1,tx);
1858             fiy1             = _mm256_add_pd(fiy1,ty);
1859             fiz1             = _mm256_add_pd(fiz1,tz);
1860
1861             fjx3             = _mm256_add_pd(fjx3,tx);
1862             fjy3             = _mm256_add_pd(fjy3,ty);
1863             fjz3             = _mm256_add_pd(fjz3,tz);
1864
1865             }
1866
1867             /**************************
1868              * CALCULATE INTERACTIONS *
1869              **************************/
1870
1871             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1872             {
1873
1874             /* REACTION-FIELD ELECTROSTATICS */
1875             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1876
1877             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1878
1879             fscal            = felec;
1880
1881             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1882
1883             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1884
1885             /* Calculate temporary vectorial force */
1886             tx               = _mm256_mul_pd(fscal,dx21);
1887             ty               = _mm256_mul_pd(fscal,dy21);
1888             tz               = _mm256_mul_pd(fscal,dz21);
1889
1890             /* Update vectorial force */
1891             fix2             = _mm256_add_pd(fix2,tx);
1892             fiy2             = _mm256_add_pd(fiy2,ty);
1893             fiz2             = _mm256_add_pd(fiz2,tz);
1894
1895             fjx1             = _mm256_add_pd(fjx1,tx);
1896             fjy1             = _mm256_add_pd(fjy1,ty);
1897             fjz1             = _mm256_add_pd(fjz1,tz);
1898
1899             }
1900
1901             /**************************
1902              * CALCULATE INTERACTIONS *
1903              **************************/
1904
1905             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1906             {
1907
1908             /* REACTION-FIELD ELECTROSTATICS */
1909             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1910
1911             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1912
1913             fscal            = felec;
1914
1915             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1916
1917             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1918
1919             /* Calculate temporary vectorial force */
1920             tx               = _mm256_mul_pd(fscal,dx22);
1921             ty               = _mm256_mul_pd(fscal,dy22);
1922             tz               = _mm256_mul_pd(fscal,dz22);
1923
1924             /* Update vectorial force */
1925             fix2             = _mm256_add_pd(fix2,tx);
1926             fiy2             = _mm256_add_pd(fiy2,ty);
1927             fiz2             = _mm256_add_pd(fiz2,tz);
1928
1929             fjx2             = _mm256_add_pd(fjx2,tx);
1930             fjy2             = _mm256_add_pd(fjy2,ty);
1931             fjz2             = _mm256_add_pd(fjz2,tz);
1932
1933             }
1934
1935             /**************************
1936              * CALCULATE INTERACTIONS *
1937              **************************/
1938
1939             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1940             {
1941
1942             /* REACTION-FIELD ELECTROSTATICS */
1943             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1944
1945             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
1946
1947             fscal            = felec;
1948
1949             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1950
1951             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1952
1953             /* Calculate temporary vectorial force */
1954             tx               = _mm256_mul_pd(fscal,dx23);
1955             ty               = _mm256_mul_pd(fscal,dy23);
1956             tz               = _mm256_mul_pd(fscal,dz23);
1957
1958             /* Update vectorial force */
1959             fix2             = _mm256_add_pd(fix2,tx);
1960             fiy2             = _mm256_add_pd(fiy2,ty);
1961             fiz2             = _mm256_add_pd(fiz2,tz);
1962
1963             fjx3             = _mm256_add_pd(fjx3,tx);
1964             fjy3             = _mm256_add_pd(fjy3,ty);
1965             fjz3             = _mm256_add_pd(fjz3,tz);
1966
1967             }
1968
1969             /**************************
1970              * CALCULATE INTERACTIONS *
1971              **************************/
1972
1973             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1974             {
1975
1976             /* REACTION-FIELD ELECTROSTATICS */
1977             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1978
1979             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
1980
1981             fscal            = felec;
1982
1983             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1984
1985             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1986
1987             /* Calculate temporary vectorial force */
1988             tx               = _mm256_mul_pd(fscal,dx31);
1989             ty               = _mm256_mul_pd(fscal,dy31);
1990             tz               = _mm256_mul_pd(fscal,dz31);
1991
1992             /* Update vectorial force */
1993             fix3             = _mm256_add_pd(fix3,tx);
1994             fiy3             = _mm256_add_pd(fiy3,ty);
1995             fiz3             = _mm256_add_pd(fiz3,tz);
1996
1997             fjx1             = _mm256_add_pd(fjx1,tx);
1998             fjy1             = _mm256_add_pd(fjy1,ty);
1999             fjz1             = _mm256_add_pd(fjz1,tz);
2000
2001             }
2002
2003             /**************************
2004              * CALCULATE INTERACTIONS *
2005              **************************/
2006
2007             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2008             {
2009
2010             /* REACTION-FIELD ELECTROSTATICS */
2011             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
2012
2013             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
2014
2015             fscal            = felec;
2016
2017             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2018
2019             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2020
2021             /* Calculate temporary vectorial force */
2022             tx               = _mm256_mul_pd(fscal,dx32);
2023             ty               = _mm256_mul_pd(fscal,dy32);
2024             tz               = _mm256_mul_pd(fscal,dz32);
2025
2026             /* Update vectorial force */
2027             fix3             = _mm256_add_pd(fix3,tx);
2028             fiy3             = _mm256_add_pd(fiy3,ty);
2029             fiz3             = _mm256_add_pd(fiz3,tz);
2030
2031             fjx2             = _mm256_add_pd(fjx2,tx);
2032             fjy2             = _mm256_add_pd(fjy2,ty);
2033             fjz2             = _mm256_add_pd(fjz2,tz);
2034
2035             }
2036
2037             /**************************
2038              * CALCULATE INTERACTIONS *
2039              **************************/
2040
2041             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2042             {
2043
2044             /* REACTION-FIELD ELECTROSTATICS */
2045             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
2046
2047             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
2048
2049             fscal            = felec;
2050
2051             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2052
2053             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2054
2055             /* Calculate temporary vectorial force */
2056             tx               = _mm256_mul_pd(fscal,dx33);
2057             ty               = _mm256_mul_pd(fscal,dy33);
2058             tz               = _mm256_mul_pd(fscal,dz33);
2059
2060             /* Update vectorial force */
2061             fix3             = _mm256_add_pd(fix3,tx);
2062             fiy3             = _mm256_add_pd(fiy3,ty);
2063             fiz3             = _mm256_add_pd(fiz3,tz);
2064
2065             fjx3             = _mm256_add_pd(fjx3,tx);
2066             fjy3             = _mm256_add_pd(fjy3,ty);
2067             fjz3             = _mm256_add_pd(fjz3,tz);
2068
2069             }
2070
2071             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2072             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2073             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2074             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2075
2076             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2077                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2078
2079             /* Inner loop uses 270 flops */
2080         }
2081
2082         /* End of innermost loop */
2083
2084         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2085                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
2086
2087         /* Increment number of inner iterations */
2088         inneriter                  += j_index_end - j_index_start;
2089
2090         /* Outer loop uses 18 flops */
2091     }
2092
2093     /* Increment number of outer iterations */
2094     outeriter        += nri;
2095
2096     /* Update outer/inner flops */
2097
2098     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*270);
2099 }