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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "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_ElecRF_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_ElecRF_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     /* Avoid stupid compiler warnings */
 151     jnrA = jnrB = jnrC = jnrD = 0;
 152     j_coord_offsetA = 0;
 153     j_coord_offsetB = 0;
 154     j_coord_offsetC = 0;
 155     j_coord_offsetD = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 181                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 182
 183         fix1             = _mm256_setzero_pd();
 184         fiy1             = _mm256_setzero_pd();
 185         fiz1             = _mm256_setzero_pd();
 186         fix2             = _mm256_setzero_pd();
 187         fiy2             = _mm256_setzero_pd();
 188         fiz2             = _mm256_setzero_pd();
 189         fix3             = _mm256_setzero_pd();
 190         fiy3             = _mm256_setzero_pd();
 191         fiz3             = _mm256_setzero_pd();
 192
 193         /* Reset potential sums */
 194         velecsum         = _mm256_setzero_pd();
 195
 196         /* Start inner kernel loop */
 197         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 198         {
 199
 200             /* Get j neighbor index, and coordinate index */
 201             jnrA             = jjnr[jidx];
 202             jnrB             = jjnr[jidx+1];
 203             jnrC             = jjnr[jidx+2];
 204             jnrD             = jjnr[jidx+3];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209
 210             /* load j atom coordinates */
 211             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 212                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 213                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 214
 215             /* Calculate displacement vector */
 216             dx11             = _mm256_sub_pd(ix1,jx1);
 217             dy11             = _mm256_sub_pd(iy1,jy1);
 218             dz11             = _mm256_sub_pd(iz1,jz1);
 219             dx12             = _mm256_sub_pd(ix1,jx2);
 220             dy12             = _mm256_sub_pd(iy1,jy2);
 221             dz12             = _mm256_sub_pd(iz1,jz2);
 222             dx13             = _mm256_sub_pd(ix1,jx3);
 223             dy13             = _mm256_sub_pd(iy1,jy3);
 224             dz13             = _mm256_sub_pd(iz1,jz3);
 225             dx21             = _mm256_sub_pd(ix2,jx1);
 226             dy21             = _mm256_sub_pd(iy2,jy1);
 227             dz21             = _mm256_sub_pd(iz2,jz1);
 228             dx22             = _mm256_sub_pd(ix2,jx2);
 229             dy22             = _mm256_sub_pd(iy2,jy2);
 230             dz22             = _mm256_sub_pd(iz2,jz2);
 231             dx23             = _mm256_sub_pd(ix2,jx3);
 232             dy23             = _mm256_sub_pd(iy2,jy3);
 233             dz23             = _mm256_sub_pd(iz2,jz3);
 234             dx31             = _mm256_sub_pd(ix3,jx1);
 235             dy31             = _mm256_sub_pd(iy3,jy1);
 236             dz31             = _mm256_sub_pd(iz3,jz1);
 237             dx32             = _mm256_sub_pd(ix3,jx2);
 238             dy32             = _mm256_sub_pd(iy3,jy2);
 239             dz32             = _mm256_sub_pd(iz3,jz2);
 240             dx33             = _mm256_sub_pd(ix3,jx3);
 241             dy33             = _mm256_sub_pd(iy3,jy3);
 242             dz33             = _mm256_sub_pd(iz3,jz3);
 243
 244             /* Calculate squared distance and things based on it */
 245             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 246             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 247             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 248             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 249             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 250             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 251             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 252             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 253             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 254
 255             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 256             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 257             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 258             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 259             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 260             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 261             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 262             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 263             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 264
 265             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 266             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 267             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 268             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 269             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 270             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 271             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 272             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 273             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 274
 275             fjx1             = _mm256_setzero_pd();
 276             fjy1             = _mm256_setzero_pd();
 277             fjz1             = _mm256_setzero_pd();
 278             fjx2             = _mm256_setzero_pd();
 279             fjy2             = _mm256_setzero_pd();
 280             fjz2             = _mm256_setzero_pd();
 281             fjx3             = _mm256_setzero_pd();
 282             fjy3             = _mm256_setzero_pd();
 283             fjz3             = _mm256_setzero_pd();
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             /* REACTION-FIELD ELECTROSTATICS */
 290             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 291             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velecsum         = _mm256_add_pd(velecsum,velec);
 295
 296             fscal            = felec;
 297
 298             /* Calculate temporary vectorial force */
 299             tx               = _mm256_mul_pd(fscal,dx11);
 300             ty               = _mm256_mul_pd(fscal,dy11);
 301             tz               = _mm256_mul_pd(fscal,dz11);
 302
 303             /* Update vectorial force */
 304             fix1             = _mm256_add_pd(fix1,tx);
 305             fiy1             = _mm256_add_pd(fiy1,ty);
 306             fiz1             = _mm256_add_pd(fiz1,tz);
 307
 308             fjx1             = _mm256_add_pd(fjx1,tx);
 309             fjy1             = _mm256_add_pd(fjy1,ty);
 310             fjz1             = _mm256_add_pd(fjz1,tz);
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             /* REACTION-FIELD ELECTROSTATICS */
 317             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 318             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 319
 320             /* Update potential sum for this i atom from the interaction with this j atom. */
 321             velecsum         = _mm256_add_pd(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325             /* Calculate temporary vectorial force */
 326             tx               = _mm256_mul_pd(fscal,dx12);
 327             ty               = _mm256_mul_pd(fscal,dy12);
 328             tz               = _mm256_mul_pd(fscal,dz12);
 329
 330             /* Update vectorial force */
 331             fix1             = _mm256_add_pd(fix1,tx);
 332             fiy1             = _mm256_add_pd(fiy1,ty);
 333             fiz1             = _mm256_add_pd(fiz1,tz);
 334
 335             fjx2             = _mm256_add_pd(fjx2,tx);
 336             fjy2             = _mm256_add_pd(fjy2,ty);
 337             fjz2             = _mm256_add_pd(fjz2,tz);
 338
 339             /**************************
 340              * CALCULATE INTERACTIONS *
 341              **************************/
 342
 343             /* REACTION-FIELD ELECTROSTATICS */
 344             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
 345             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velecsum         = _mm256_add_pd(velecsum,velec);
 349
 350             fscal            = felec;
 351
 352             /* Calculate temporary vectorial force */
 353             tx               = _mm256_mul_pd(fscal,dx13);
 354             ty               = _mm256_mul_pd(fscal,dy13);
 355             tz               = _mm256_mul_pd(fscal,dz13);
 356
 357             /* Update vectorial force */
 358             fix1             = _mm256_add_pd(fix1,tx);
 359             fiy1             = _mm256_add_pd(fiy1,ty);
 360             fiz1             = _mm256_add_pd(fiz1,tz);
 361
 362             fjx3             = _mm256_add_pd(fjx3,tx);
 363             fjy3             = _mm256_add_pd(fjy3,ty);
 364             fjz3             = _mm256_add_pd(fjz3,tz);
 365
 366             /**************************
 367              * CALCULATE INTERACTIONS *
 368              **************************/
 369
 370             /* REACTION-FIELD ELECTROSTATICS */
 371             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 372             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velecsum         = _mm256_add_pd(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             /* Calculate temporary vectorial force */
 380             tx               = _mm256_mul_pd(fscal,dx21);
 381             ty               = _mm256_mul_pd(fscal,dy21);
 382             tz               = _mm256_mul_pd(fscal,dz21);
 383
 384             /* Update vectorial force */
 385             fix2             = _mm256_add_pd(fix2,tx);
 386             fiy2             = _mm256_add_pd(fiy2,ty);
 387             fiz2             = _mm256_add_pd(fiz2,tz);
 388
 389             fjx1             = _mm256_add_pd(fjx1,tx);
 390             fjy1             = _mm256_add_pd(fjy1,ty);
 391             fjz1             = _mm256_add_pd(fjz1,tz);
 392
 393             /**************************
 394              * CALCULATE INTERACTIONS *
 395              **************************/
 396
 397             /* REACTION-FIELD ELECTROSTATICS */
 398             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 399             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 400
 401             /* Update potential sum for this i atom from the interaction with this j atom. */
 402             velecsum         = _mm256_add_pd(velecsum,velec);
 403
 404             fscal            = felec;
 405
 406             /* Calculate temporary vectorial force */
 407             tx               = _mm256_mul_pd(fscal,dx22);
 408             ty               = _mm256_mul_pd(fscal,dy22);
 409             tz               = _mm256_mul_pd(fscal,dz22);
 410
 411             /* Update vectorial force */
 412             fix2             = _mm256_add_pd(fix2,tx);
 413             fiy2             = _mm256_add_pd(fiy2,ty);
 414             fiz2             = _mm256_add_pd(fiz2,tz);
 415
 416             fjx2             = _mm256_add_pd(fjx2,tx);
 417             fjy2             = _mm256_add_pd(fjy2,ty);
 418             fjz2             = _mm256_add_pd(fjz2,tz);
 419
 420             /**************************
 421              * CALCULATE INTERACTIONS *
 422              **************************/
 423
 424             /* REACTION-FIELD ELECTROSTATICS */
 425             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
 426             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
 427
 428             /* Update potential sum for this i atom from the interaction with this j atom. */
 429             velecsum         = _mm256_add_pd(velecsum,velec);
 430
 431             fscal            = felec;
 432
 433             /* Calculate temporary vectorial force */
 434             tx               = _mm256_mul_pd(fscal,dx23);
 435             ty               = _mm256_mul_pd(fscal,dy23);
 436             tz               = _mm256_mul_pd(fscal,dz23);
 437
 438             /* Update vectorial force */
 439             fix2             = _mm256_add_pd(fix2,tx);
 440             fiy2             = _mm256_add_pd(fiy2,ty);
 441             fiz2             = _mm256_add_pd(fiz2,tz);
 442
 443             fjx3             = _mm256_add_pd(fjx3,tx);
 444             fjy3             = _mm256_add_pd(fjy3,ty);
 445             fjz3             = _mm256_add_pd(fjz3,tz);
 446
 447             /**************************
 448              * CALCULATE INTERACTIONS *
 449              **************************/
 450
 451             /* REACTION-FIELD ELECTROSTATICS */
 452             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
 453             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
 454
 455             /* Update potential sum for this i atom from the interaction with this j atom. */
 456             velecsum         = _mm256_add_pd(velecsum,velec);
 457
 458             fscal            = felec;
 459
 460             /* Calculate temporary vectorial force */
 461             tx               = _mm256_mul_pd(fscal,dx31);
 462             ty               = _mm256_mul_pd(fscal,dy31);
 463             tz               = _mm256_mul_pd(fscal,dz31);
 464
 465             /* Update vectorial force */
 466             fix3             = _mm256_add_pd(fix3,tx);
 467             fiy3             = _mm256_add_pd(fiy3,ty);
 468             fiz3             = _mm256_add_pd(fiz3,tz);
 469
 470             fjx1             = _mm256_add_pd(fjx1,tx);
 471             fjy1             = _mm256_add_pd(fjy1,ty);
 472             fjz1             = _mm256_add_pd(fjz1,tz);
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             /* REACTION-FIELD ELECTROSTATICS */
 479             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
 480             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velecsum         = _mm256_add_pd(velecsum,velec);
 484
 485             fscal            = felec;
 486
 487             /* Calculate temporary vectorial force */
 488             tx               = _mm256_mul_pd(fscal,dx32);
 489             ty               = _mm256_mul_pd(fscal,dy32);
 490             tz               = _mm256_mul_pd(fscal,dz32);
 491
 492             /* Update vectorial force */
 493             fix3             = _mm256_add_pd(fix3,tx);
 494             fiy3             = _mm256_add_pd(fiy3,ty);
 495             fiz3             = _mm256_add_pd(fiz3,tz);
 496
 497             fjx2             = _mm256_add_pd(fjx2,tx);
 498             fjy2             = _mm256_add_pd(fjy2,ty);
 499             fjz2             = _mm256_add_pd(fjz2,tz);
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             /* REACTION-FIELD ELECTROSTATICS */
 506             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
 507             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
 508
 509             /* Update potential sum for this i atom from the interaction with this j atom. */
 510             velecsum         = _mm256_add_pd(velecsum,velec);
 511
 512             fscal            = felec;
 513
 514             /* Calculate temporary vectorial force */
 515             tx               = _mm256_mul_pd(fscal,dx33);
 516             ty               = _mm256_mul_pd(fscal,dy33);
 517             tz               = _mm256_mul_pd(fscal,dz33);
 518
 519             /* Update vectorial force */
 520             fix3             = _mm256_add_pd(fix3,tx);
 521             fiy3             = _mm256_add_pd(fiy3,ty);
 522             fiz3             = _mm256_add_pd(fiz3,tz);
 523
 524             fjx3             = _mm256_add_pd(fjx3,tx);
 525             fjy3             = _mm256_add_pd(fjy3,ty);
 526             fjz3             = _mm256_add_pd(fjz3,tz);
 527
 528             fjptrA             = f+j_coord_offsetA;
 529             fjptrB             = f+j_coord_offsetB;
 530             fjptrC             = f+j_coord_offsetC;
 531             fjptrD             = f+j_coord_offsetD;
 532
 533             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 534                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 535
 536             /* Inner loop uses 288 flops */
 537         }
 538
 539         if(jidx<j_index_end)
 540         {
 541
 542             /* Get j neighbor index, and coordinate index */
 543             jnrlistA         = jjnr[jidx];
 544             jnrlistB         = jjnr[jidx+1];
 545             jnrlistC         = jjnr[jidx+2];
 546             jnrlistD         = jjnr[jidx+3];
 547             /* Sign of each element will be negative for non-real atoms.
 548              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 549              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 550              */
 551             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 552
 553             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 554             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 555             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 556
 557             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 558             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 559             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 560             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 561             j_coord_offsetA  = DIM*jnrA;
 562             j_coord_offsetB  = DIM*jnrB;
 563             j_coord_offsetC  = DIM*jnrC;
 564             j_coord_offsetD  = DIM*jnrD;
 565
 566             /* load j atom coordinates */
 567             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 568                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 569                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 570
 571             /* Calculate displacement vector */
 572             dx11             = _mm256_sub_pd(ix1,jx1);
 573             dy11             = _mm256_sub_pd(iy1,jy1);
 574             dz11             = _mm256_sub_pd(iz1,jz1);
 575             dx12             = _mm256_sub_pd(ix1,jx2);
 576             dy12             = _mm256_sub_pd(iy1,jy2);
 577             dz12             = _mm256_sub_pd(iz1,jz2);
 578             dx13             = _mm256_sub_pd(ix1,jx3);
 579             dy13             = _mm256_sub_pd(iy1,jy3);
 580             dz13             = _mm256_sub_pd(iz1,jz3);
 581             dx21             = _mm256_sub_pd(ix2,jx1);
 582             dy21             = _mm256_sub_pd(iy2,jy1);
 583             dz21             = _mm256_sub_pd(iz2,jz1);
 584             dx22             = _mm256_sub_pd(ix2,jx2);
 585             dy22             = _mm256_sub_pd(iy2,jy2);
 586             dz22             = _mm256_sub_pd(iz2,jz2);
 587             dx23             = _mm256_sub_pd(ix2,jx3);
 588             dy23             = _mm256_sub_pd(iy2,jy3);
 589             dz23             = _mm256_sub_pd(iz2,jz3);
 590             dx31             = _mm256_sub_pd(ix3,jx1);
 591             dy31             = _mm256_sub_pd(iy3,jy1);
 592             dz31             = _mm256_sub_pd(iz3,jz1);
 593             dx32             = _mm256_sub_pd(ix3,jx2);
 594             dy32             = _mm256_sub_pd(iy3,jy2);
 595             dz32             = _mm256_sub_pd(iz3,jz2);
 596             dx33             = _mm256_sub_pd(ix3,jx3);
 597             dy33             = _mm256_sub_pd(iy3,jy3);
 598             dz33             = _mm256_sub_pd(iz3,jz3);
 599
 600             /* Calculate squared distance and things based on it */
 601             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 602             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 603             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 604             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 605             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 606             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 607             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 608             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 609             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 610
 611             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 612             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 613             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 614             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 615             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 616             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 617             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 618             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 619             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 620
 621             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 622             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 623             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 624             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 625             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 626             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 627             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 628             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 629             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 630
 631             fjx1             = _mm256_setzero_pd();
 632             fjy1             = _mm256_setzero_pd();
 633             fjz1             = _mm256_setzero_pd();
 634             fjx2             = _mm256_setzero_pd();
 635             fjy2             = _mm256_setzero_pd();
 636             fjz2             = _mm256_setzero_pd();
 637             fjx3             = _mm256_setzero_pd();
 638             fjy3             = _mm256_setzero_pd();
 639             fjz3             = _mm256_setzero_pd();
 640
 641             /**************************
 642              * CALCULATE INTERACTIONS *
 643              **************************/
 644
 645             /* REACTION-FIELD ELECTROSTATICS */
 646             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 647             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 648
 649             /* Update potential sum for this i atom from the interaction with this j atom. */
 650             velec            = _mm256_andnot_pd(dummy_mask,velec);
 651             velecsum         = _mm256_add_pd(velecsum,velec);
 652
 653             fscal            = felec;
 654
 655             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 656
 657             /* Calculate temporary vectorial force */
 658             tx               = _mm256_mul_pd(fscal,dx11);
 659             ty               = _mm256_mul_pd(fscal,dy11);
 660             tz               = _mm256_mul_pd(fscal,dz11);
 661
 662             /* Update vectorial force */
 663             fix1             = _mm256_add_pd(fix1,tx);
 664             fiy1             = _mm256_add_pd(fiy1,ty);
 665             fiz1             = _mm256_add_pd(fiz1,tz);
 666
 667             fjx1             = _mm256_add_pd(fjx1,tx);
 668             fjy1             = _mm256_add_pd(fjy1,ty);
 669             fjz1             = _mm256_add_pd(fjz1,tz);
 670
 671             /**************************
 672              * CALCULATE INTERACTIONS *
 673              **************************/
 674
 675             /* REACTION-FIELD ELECTROSTATICS */
 676             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 677             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 678
 679             /* Update potential sum for this i atom from the interaction with this j atom. */
 680             velec            = _mm256_andnot_pd(dummy_mask,velec);
 681             velecsum         = _mm256_add_pd(velecsum,velec);
 682
 683             fscal            = felec;
 684
 685             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 686
 687             /* Calculate temporary vectorial force */
 688             tx               = _mm256_mul_pd(fscal,dx12);
 689             ty               = _mm256_mul_pd(fscal,dy12);
 690             tz               = _mm256_mul_pd(fscal,dz12);
 691
 692             /* Update vectorial force */
 693             fix1             = _mm256_add_pd(fix1,tx);
 694             fiy1             = _mm256_add_pd(fiy1,ty);
 695             fiz1             = _mm256_add_pd(fiz1,tz);
 696
 697             fjx2             = _mm256_add_pd(fjx2,tx);
 698             fjy2             = _mm256_add_pd(fjy2,ty);
 699             fjz2             = _mm256_add_pd(fjz2,tz);
 700
 701             /**************************
 702              * CALCULATE INTERACTIONS *
 703              **************************/
 704
 705             /* REACTION-FIELD ELECTROSTATICS */
 706             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_add_pd(rinv13,_mm256_mul_pd(krf,rsq13)),crf));
 707             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
 708
 709             /* Update potential sum for this i atom from the interaction with this j atom. */
 710             velec            = _mm256_andnot_pd(dummy_mask,velec);
 711             velecsum         = _mm256_add_pd(velecsum,velec);
 712
 713             fscal            = felec;
 714
 715             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 716
 717             /* Calculate temporary vectorial force */
 718             tx               = _mm256_mul_pd(fscal,dx13);
 719             ty               = _mm256_mul_pd(fscal,dy13);
 720             tz               = _mm256_mul_pd(fscal,dz13);
 721
 722             /* Update vectorial force */
 723             fix1             = _mm256_add_pd(fix1,tx);
 724             fiy1             = _mm256_add_pd(fiy1,ty);
 725             fiz1             = _mm256_add_pd(fiz1,tz);
 726
 727             fjx3             = _mm256_add_pd(fjx3,tx);
 728             fjy3             = _mm256_add_pd(fjy3,ty);
 729             fjz3             = _mm256_add_pd(fjz3,tz);
 730
 731             /**************************
 732              * CALCULATE INTERACTIONS *
 733              **************************/
 734
 735             /* REACTION-FIELD ELECTROSTATICS */
 736             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 737             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 738
 739             /* Update potential sum for this i atom from the interaction with this j atom. */
 740             velec            = _mm256_andnot_pd(dummy_mask,velec);
 741             velecsum         = _mm256_add_pd(velecsum,velec);
 742
 743             fscal            = felec;
 744
 745             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 746
 747             /* Calculate temporary vectorial force */
 748             tx               = _mm256_mul_pd(fscal,dx21);
 749             ty               = _mm256_mul_pd(fscal,dy21);
 750             tz               = _mm256_mul_pd(fscal,dz21);
 751
 752             /* Update vectorial force */
 753             fix2             = _mm256_add_pd(fix2,tx);
 754             fiy2             = _mm256_add_pd(fiy2,ty);
 755             fiz2             = _mm256_add_pd(fiz2,tz);
 756
 757             fjx1             = _mm256_add_pd(fjx1,tx);
 758             fjy1             = _mm256_add_pd(fjy1,ty);
 759             fjz1             = _mm256_add_pd(fjz1,tz);
 760
 761             /**************************
 762              * CALCULATE INTERACTIONS *
 763              **************************/
 764
 765             /* REACTION-FIELD ELECTROSTATICS */
 766             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 767             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 768
 769             /* Update potential sum for this i atom from the interaction with this j atom. */
 770             velec            = _mm256_andnot_pd(dummy_mask,velec);
 771             velecsum         = _mm256_add_pd(velecsum,velec);
 772
 773             fscal            = felec;
 774
 775             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 776
 777             /* Calculate temporary vectorial force */
 778             tx               = _mm256_mul_pd(fscal,dx22);
 779             ty               = _mm256_mul_pd(fscal,dy22);
 780             tz               = _mm256_mul_pd(fscal,dz22);
 781
 782             /* Update vectorial force */
 783             fix2             = _mm256_add_pd(fix2,tx);
 784             fiy2             = _mm256_add_pd(fiy2,ty);
 785             fiz2             = _mm256_add_pd(fiz2,tz);
 786
 787             fjx2             = _mm256_add_pd(fjx2,tx);
 788             fjy2             = _mm256_add_pd(fjy2,ty);
 789             fjz2             = _mm256_add_pd(fjz2,tz);
 790
 791             /**************************
 792              * CALCULATE INTERACTIONS *
 793              **************************/
 794
 795             /* REACTION-FIELD ELECTROSTATICS */
 796             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_add_pd(rinv23,_mm256_mul_pd(krf,rsq23)),crf));
 797             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
 798
 799             /* Update potential sum for this i atom from the interaction with this j atom. */
 800             velec            = _mm256_andnot_pd(dummy_mask,velec);
 801             velecsum         = _mm256_add_pd(velecsum,velec);
 802
 803             fscal            = felec;
 804
 805             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 806
 807             /* Calculate temporary vectorial force */
 808             tx               = _mm256_mul_pd(fscal,dx23);
 809             ty               = _mm256_mul_pd(fscal,dy23);
 810             tz               = _mm256_mul_pd(fscal,dz23);
 811
 812             /* Update vectorial force */
 813             fix2             = _mm256_add_pd(fix2,tx);
 814             fiy2             = _mm256_add_pd(fiy2,ty);
 815             fiz2             = _mm256_add_pd(fiz2,tz);
 816
 817             fjx3             = _mm256_add_pd(fjx3,tx);
 818             fjy3             = _mm256_add_pd(fjy3,ty);
 819             fjz3             = _mm256_add_pd(fjz3,tz);
 820
 821             /**************************
 822              * CALCULATE INTERACTIONS *
 823              **************************/
 824
 825             /* REACTION-FIELD ELECTROSTATICS */
 826             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_add_pd(rinv31,_mm256_mul_pd(krf,rsq31)),crf));
 827             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
 828
 829             /* Update potential sum for this i atom from the interaction with this j atom. */
 830             velec            = _mm256_andnot_pd(dummy_mask,velec);
 831             velecsum         = _mm256_add_pd(velecsum,velec);
 832
 833             fscal            = felec;
 834
 835             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 836
 837             /* Calculate temporary vectorial force */
 838             tx               = _mm256_mul_pd(fscal,dx31);
 839             ty               = _mm256_mul_pd(fscal,dy31);
 840             tz               = _mm256_mul_pd(fscal,dz31);
 841
 842             /* Update vectorial force */
 843             fix3             = _mm256_add_pd(fix3,tx);
 844             fiy3             = _mm256_add_pd(fiy3,ty);
 845             fiz3             = _mm256_add_pd(fiz3,tz);
 846
 847             fjx1             = _mm256_add_pd(fjx1,tx);
 848             fjy1             = _mm256_add_pd(fjy1,ty);
 849             fjz1             = _mm256_add_pd(fjz1,tz);
 850
 851             /**************************
 852              * CALCULATE INTERACTIONS *
 853              **************************/
 854
 855             /* REACTION-FIELD ELECTROSTATICS */
 856             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_add_pd(rinv32,_mm256_mul_pd(krf,rsq32)),crf));
 857             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
 858
 859             /* Update potential sum for this i atom from the interaction with this j atom. */
 860             velec            = _mm256_andnot_pd(dummy_mask,velec);
 861             velecsum         = _mm256_add_pd(velecsum,velec);
 862
 863             fscal            = felec;
 864
 865             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 866
 867             /* Calculate temporary vectorial force */
 868             tx               = _mm256_mul_pd(fscal,dx32);
 869             ty               = _mm256_mul_pd(fscal,dy32);
 870             tz               = _mm256_mul_pd(fscal,dz32);
 871
 872             /* Update vectorial force */
 873             fix3             = _mm256_add_pd(fix3,tx);
 874             fiy3             = _mm256_add_pd(fiy3,ty);
 875             fiz3             = _mm256_add_pd(fiz3,tz);
 876
 877             fjx2             = _mm256_add_pd(fjx2,tx);
 878             fjy2             = _mm256_add_pd(fjy2,ty);
 879             fjz2             = _mm256_add_pd(fjz2,tz);
 880
 881             /**************************
 882              * CALCULATE INTERACTIONS *
 883              **************************/
 884
 885             /* REACTION-FIELD ELECTROSTATICS */
 886             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_add_pd(rinv33,_mm256_mul_pd(krf,rsq33)),crf));
 887             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
 888
 889             /* Update potential sum for this i atom from the interaction with this j atom. */
 890             velec            = _mm256_andnot_pd(dummy_mask,velec);
 891             velecsum         = _mm256_add_pd(velecsum,velec);
 892
 893             fscal            = felec;
 894
 895             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 896
 897             /* Calculate temporary vectorial force */
 898             tx               = _mm256_mul_pd(fscal,dx33);
 899             ty               = _mm256_mul_pd(fscal,dy33);
 900             tz               = _mm256_mul_pd(fscal,dz33);
 901
 902             /* Update vectorial force */
 903             fix3             = _mm256_add_pd(fix3,tx);
 904             fiy3             = _mm256_add_pd(fiy3,ty);
 905             fiz3             = _mm256_add_pd(fiz3,tz);
 906
 907             fjx3             = _mm256_add_pd(fjx3,tx);
 908             fjy3             = _mm256_add_pd(fjy3,ty);
 909             fjz3             = _mm256_add_pd(fjz3,tz);
 910
 911             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 912             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 913             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 914             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 915
 916             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 917                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 918
 919             /* Inner loop uses 288 flops */
 920         }
 921
 922         /* End of innermost loop */
 923
 924         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 925                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 926
 927         ggid                        = gid[iidx];
 928         /* Update potential energies */
 929         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 930
 931         /* Increment number of inner iterations */
 932         inneriter                  += j_index_end - j_index_start;
 933
 934         /* Outer loop uses 19 flops */
 935     }
 936
 937     /* Increment number of outer iterations */
 938     outeriter        += nri;
 939
 940     /* Update outer/inner flops */
 941
 942     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*288);
 943 }
 944 /*
 945  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4W4_F_avx_256_double
 946  * Electrostatics interaction: ReactionField
 947  * VdW interaction:            None
 948  * Geometry:                   Water4-Water4
 949  * Calculate force/pot:        Force
 950  */
 951 void
 952 nb_kernel_ElecRF_VdwNone_GeomW4W4_F_avx_256_double
 953                     (t_nblist                    * gmx_restrict       nlist,
 954                      rvec                        * gmx_restrict          xx,
 955                      rvec                        * gmx_restrict          ff,
 956                      t_forcerec                  * gmx_restrict          fr,
 957                      t_mdatoms                   * gmx_restrict     mdatoms,
 958                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 959                      t_nrnb                      * gmx_restrict        nrnb)
 960 {
 961     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 962      * just 0 for non-waters.
 963      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 964      * jnr indices corresponding to data put in the four positions in the SIMD register.
 965      */
 966     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 967     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 968     int              jnrA,jnrB,jnrC,jnrD;
 969     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 970     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 971     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 972     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 973     real             rcutoff_scalar;
 974     real             *shiftvec,*fshift,*x,*f;
 975     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 976     real             scratch[4*DIM];
 977     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 978     real *           vdwioffsetptr1;
 979     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 980     real *           vdwioffsetptr2;
 981     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 982     real *           vdwioffsetptr3;
 983     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 984     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
 985     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 986     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
 987     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 988     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
 989     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 990     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 991     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 992     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 993     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 994     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 995     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 996     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 997     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 998     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 999     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1000     real             *charge;
1001     __m256d          dummy_mask,cutoff_mask;
1002     __m128           tmpmask0,tmpmask1;
1003     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1004     __m256d          one     = _mm256_set1_pd(1.0);
1005     __m256d          two     = _mm256_set1_pd(2.0);
1006     x                = xx[0];
1007     f                = ff[0];
1008
1009     nri              = nlist->nri;
1010     iinr             = nlist->iinr;
1011     jindex           = nlist->jindex;
1012     jjnr             = nlist->jjnr;
1013     shiftidx         = nlist->shift;
1014     gid              = nlist->gid;
1015     shiftvec         = fr->shift_vec[0];
1016     fshift           = fr->fshift[0];
1017     facel            = _mm256_set1_pd(fr->epsfac);
1018     charge           = mdatoms->chargeA;
1019     krf              = _mm256_set1_pd(fr->ic->k_rf);
1020     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
1021     crf              = _mm256_set1_pd(fr->ic->c_rf);
1022
1023     /* Setup water-specific parameters */
1024     inr              = nlist->iinr[0];
1025     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1026     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1027     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1028
1029     jq1              = _mm256_set1_pd(charge[inr+1]);
1030     jq2              = _mm256_set1_pd(charge[inr+2]);
1031     jq3              = _mm256_set1_pd(charge[inr+3]);
1032     qq11             = _mm256_mul_pd(iq1,jq1);
1033     qq12             = _mm256_mul_pd(iq1,jq2);
1034     qq13             = _mm256_mul_pd(iq1,jq3);
1035     qq21             = _mm256_mul_pd(iq2,jq1);
1036     qq22             = _mm256_mul_pd(iq2,jq2);
1037     qq23             = _mm256_mul_pd(iq2,jq3);
1038     qq31             = _mm256_mul_pd(iq3,jq1);
1039     qq32             = _mm256_mul_pd(iq3,jq2);
1040     qq33             = _mm256_mul_pd(iq3,jq3);
1041
1042     /* Avoid stupid compiler warnings */
1043     jnrA = jnrB = jnrC = jnrD = 0;
1044     j_coord_offsetA = 0;
1045     j_coord_offsetB = 0;
1046     j_coord_offsetC = 0;
1047     j_coord_offsetD = 0;
1048
1049     outeriter        = 0;
1050     inneriter        = 0;
1051
1052     for(iidx=0;iidx<4*DIM;iidx++)
1053     {
1054         scratch[iidx] = 0.0;
1055     }
1056
1057     /* Start outer loop over neighborlists */
1058     for(iidx=0; iidx<nri; iidx++)
1059     {
1060         /* Load shift vector for this list */
1061         i_shift_offset   = DIM*shiftidx[iidx];
1062
1063         /* Load limits for loop over neighbors */
1064         j_index_start    = jindex[iidx];
1065         j_index_end      = jindex[iidx+1];
1066
1067         /* Get outer coordinate index */
1068         inr              = iinr[iidx];
1069         i_coord_offset   = DIM*inr;
1070
1071         /* Load i particle coords and add shift vector */
1072         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1073                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1074
1075         fix1             = _mm256_setzero_pd();
1076         fiy1             = _mm256_setzero_pd();
1077         fiz1             = _mm256_setzero_pd();
1078         fix2             = _mm256_setzero_pd();
1079         fiy2             = _mm256_setzero_pd();
1080         fiz2             = _mm256_setzero_pd();
1081         fix3             = _mm256_setzero_pd();
1082         fiy3             = _mm256_setzero_pd();
1083         fiz3             = _mm256_setzero_pd();
1084
1085         /* Start inner kernel loop */
1086         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1087         {
1088
1089             /* Get j neighbor index, and coordinate index */
1090             jnrA             = jjnr[jidx];
1091             jnrB             = jjnr[jidx+1];
1092             jnrC             = jjnr[jidx+2];
1093             jnrD             = jjnr[jidx+3];
1094             j_coord_offsetA  = DIM*jnrA;
1095             j_coord_offsetB  = DIM*jnrB;
1096             j_coord_offsetC  = DIM*jnrC;
1097             j_coord_offsetD  = DIM*jnrD;
1098
1099             /* load j atom coordinates */
1100             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1101                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1102                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1103
1104             /* Calculate displacement vector */
1105             dx11             = _mm256_sub_pd(ix1,jx1);
1106             dy11             = _mm256_sub_pd(iy1,jy1);
1107             dz11             = _mm256_sub_pd(iz1,jz1);
1108             dx12             = _mm256_sub_pd(ix1,jx2);
1109             dy12             = _mm256_sub_pd(iy1,jy2);
1110             dz12             = _mm256_sub_pd(iz1,jz2);
1111             dx13             = _mm256_sub_pd(ix1,jx3);
1112             dy13             = _mm256_sub_pd(iy1,jy3);
1113             dz13             = _mm256_sub_pd(iz1,jz3);
1114             dx21             = _mm256_sub_pd(ix2,jx1);
1115             dy21             = _mm256_sub_pd(iy2,jy1);
1116             dz21             = _mm256_sub_pd(iz2,jz1);
1117             dx22             = _mm256_sub_pd(ix2,jx2);
1118             dy22             = _mm256_sub_pd(iy2,jy2);
1119             dz22             = _mm256_sub_pd(iz2,jz2);
1120             dx23             = _mm256_sub_pd(ix2,jx3);
1121             dy23             = _mm256_sub_pd(iy2,jy3);
1122             dz23             = _mm256_sub_pd(iz2,jz3);
1123             dx31             = _mm256_sub_pd(ix3,jx1);
1124             dy31             = _mm256_sub_pd(iy3,jy1);
1125             dz31             = _mm256_sub_pd(iz3,jz1);
1126             dx32             = _mm256_sub_pd(ix3,jx2);
1127             dy32             = _mm256_sub_pd(iy3,jy2);
1128             dz32             = _mm256_sub_pd(iz3,jz2);
1129             dx33             = _mm256_sub_pd(ix3,jx3);
1130             dy33             = _mm256_sub_pd(iy3,jy3);
1131             dz33             = _mm256_sub_pd(iz3,jz3);
1132
1133             /* Calculate squared distance and things based on it */
1134             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1135             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1136             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1137             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1138             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1139             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1140             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1141             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1142             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1143
1144             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1145             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1146             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1147             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1148             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1149             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1150             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1151             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1152             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1153
1154             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1155             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1156             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1157             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1158             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1159             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1160             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1161             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1162             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1163
1164             fjx1             = _mm256_setzero_pd();
1165             fjy1             = _mm256_setzero_pd();
1166             fjz1             = _mm256_setzero_pd();
1167             fjx2             = _mm256_setzero_pd();
1168             fjy2             = _mm256_setzero_pd();
1169             fjz2             = _mm256_setzero_pd();
1170             fjx3             = _mm256_setzero_pd();
1171             fjy3             = _mm256_setzero_pd();
1172             fjz3             = _mm256_setzero_pd();
1173
1174             /**************************
1175              * CALCULATE INTERACTIONS *
1176              **************************/
1177
1178             /* REACTION-FIELD ELECTROSTATICS */
1179             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1180
1181             fscal            = felec;
1182
1183             /* Calculate temporary vectorial force */
1184             tx               = _mm256_mul_pd(fscal,dx11);
1185             ty               = _mm256_mul_pd(fscal,dy11);
1186             tz               = _mm256_mul_pd(fscal,dz11);
1187
1188             /* Update vectorial force */
1189             fix1             = _mm256_add_pd(fix1,tx);
1190             fiy1             = _mm256_add_pd(fiy1,ty);
1191             fiz1             = _mm256_add_pd(fiz1,tz);
1192
1193             fjx1             = _mm256_add_pd(fjx1,tx);
1194             fjy1             = _mm256_add_pd(fjy1,ty);
1195             fjz1             = _mm256_add_pd(fjz1,tz);
1196
1197             /**************************
1198              * CALCULATE INTERACTIONS *
1199              **************************/
1200
1201             /* REACTION-FIELD ELECTROSTATICS */
1202             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1203
1204             fscal            = felec;
1205
1206             /* Calculate temporary vectorial force */
1207             tx               = _mm256_mul_pd(fscal,dx12);
1208             ty               = _mm256_mul_pd(fscal,dy12);
1209             tz               = _mm256_mul_pd(fscal,dz12);
1210
1211             /* Update vectorial force */
1212             fix1             = _mm256_add_pd(fix1,tx);
1213             fiy1             = _mm256_add_pd(fiy1,ty);
1214             fiz1             = _mm256_add_pd(fiz1,tz);
1215
1216             fjx2             = _mm256_add_pd(fjx2,tx);
1217             fjy2             = _mm256_add_pd(fjy2,ty);
1218             fjz2             = _mm256_add_pd(fjz2,tz);
1219
1220             /**************************
1221              * CALCULATE INTERACTIONS *
1222              **************************/
1223
1224             /* REACTION-FIELD ELECTROSTATICS */
1225             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
1226
1227             fscal            = felec;
1228
1229             /* Calculate temporary vectorial force */
1230             tx               = _mm256_mul_pd(fscal,dx13);
1231             ty               = _mm256_mul_pd(fscal,dy13);
1232             tz               = _mm256_mul_pd(fscal,dz13);
1233
1234             /* Update vectorial force */
1235             fix1             = _mm256_add_pd(fix1,tx);
1236             fiy1             = _mm256_add_pd(fiy1,ty);
1237             fiz1             = _mm256_add_pd(fiz1,tz);
1238
1239             fjx3             = _mm256_add_pd(fjx3,tx);
1240             fjy3             = _mm256_add_pd(fjy3,ty);
1241             fjz3             = _mm256_add_pd(fjz3,tz);
1242
1243             /**************************
1244              * CALCULATE INTERACTIONS *
1245              **************************/
1246
1247             /* REACTION-FIELD ELECTROSTATICS */
1248             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1249
1250             fscal            = felec;
1251
1252             /* Calculate temporary vectorial force */
1253             tx               = _mm256_mul_pd(fscal,dx21);
1254             ty               = _mm256_mul_pd(fscal,dy21);
1255             tz               = _mm256_mul_pd(fscal,dz21);
1256
1257             /* Update vectorial force */
1258             fix2             = _mm256_add_pd(fix2,tx);
1259             fiy2             = _mm256_add_pd(fiy2,ty);
1260             fiz2             = _mm256_add_pd(fiz2,tz);
1261
1262             fjx1             = _mm256_add_pd(fjx1,tx);
1263             fjy1             = _mm256_add_pd(fjy1,ty);
1264             fjz1             = _mm256_add_pd(fjz1,tz);
1265
1266             /**************************
1267              * CALCULATE INTERACTIONS *
1268              **************************/
1269
1270             /* REACTION-FIELD ELECTROSTATICS */
1271             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1272
1273             fscal            = felec;
1274
1275             /* Calculate temporary vectorial force */
1276             tx               = _mm256_mul_pd(fscal,dx22);
1277             ty               = _mm256_mul_pd(fscal,dy22);
1278             tz               = _mm256_mul_pd(fscal,dz22);
1279
1280             /* Update vectorial force */
1281             fix2             = _mm256_add_pd(fix2,tx);
1282             fiy2             = _mm256_add_pd(fiy2,ty);
1283             fiz2             = _mm256_add_pd(fiz2,tz);
1284
1285             fjx2             = _mm256_add_pd(fjx2,tx);
1286             fjy2             = _mm256_add_pd(fjy2,ty);
1287             fjz2             = _mm256_add_pd(fjz2,tz);
1288
1289             /**************************
1290              * CALCULATE INTERACTIONS *
1291              **************************/
1292
1293             /* REACTION-FIELD ELECTROSTATICS */
1294             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1295
1296             fscal            = felec;
1297
1298             /* Calculate temporary vectorial force */
1299             tx               = _mm256_mul_pd(fscal,dx23);
1300             ty               = _mm256_mul_pd(fscal,dy23);
1301             tz               = _mm256_mul_pd(fscal,dz23);
1302
1303             /* Update vectorial force */
1304             fix2             = _mm256_add_pd(fix2,tx);
1305             fiy2             = _mm256_add_pd(fiy2,ty);
1306             fiz2             = _mm256_add_pd(fiz2,tz);
1307
1308             fjx3             = _mm256_add_pd(fjx3,tx);
1309             fjy3             = _mm256_add_pd(fjy3,ty);
1310             fjz3             = _mm256_add_pd(fjz3,tz);
1311
1312             /**************************
1313              * CALCULATE INTERACTIONS *
1314              **************************/
1315
1316             /* REACTION-FIELD ELECTROSTATICS */
1317             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1318
1319             fscal            = felec;
1320
1321             /* Calculate temporary vectorial force */
1322             tx               = _mm256_mul_pd(fscal,dx31);
1323             ty               = _mm256_mul_pd(fscal,dy31);
1324             tz               = _mm256_mul_pd(fscal,dz31);
1325
1326             /* Update vectorial force */
1327             fix3             = _mm256_add_pd(fix3,tx);
1328             fiy3             = _mm256_add_pd(fiy3,ty);
1329             fiz3             = _mm256_add_pd(fiz3,tz);
1330
1331             fjx1             = _mm256_add_pd(fjx1,tx);
1332             fjy1             = _mm256_add_pd(fjy1,ty);
1333             fjz1             = _mm256_add_pd(fjz1,tz);
1334
1335             /**************************
1336              * CALCULATE INTERACTIONS *
1337              **************************/
1338
1339             /* REACTION-FIELD ELECTROSTATICS */
1340             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1341
1342             fscal            = felec;
1343
1344             /* Calculate temporary vectorial force */
1345             tx               = _mm256_mul_pd(fscal,dx32);
1346             ty               = _mm256_mul_pd(fscal,dy32);
1347             tz               = _mm256_mul_pd(fscal,dz32);
1348
1349             /* Update vectorial force */
1350             fix3             = _mm256_add_pd(fix3,tx);
1351             fiy3             = _mm256_add_pd(fiy3,ty);
1352             fiz3             = _mm256_add_pd(fiz3,tz);
1353
1354             fjx2             = _mm256_add_pd(fjx2,tx);
1355             fjy2             = _mm256_add_pd(fjy2,ty);
1356             fjz2             = _mm256_add_pd(fjz2,tz);
1357
1358             /**************************
1359              * CALCULATE INTERACTIONS *
1360              **************************/
1361
1362             /* REACTION-FIELD ELECTROSTATICS */
1363             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1364
1365             fscal            = felec;
1366
1367             /* Calculate temporary vectorial force */
1368             tx               = _mm256_mul_pd(fscal,dx33);
1369             ty               = _mm256_mul_pd(fscal,dy33);
1370             tz               = _mm256_mul_pd(fscal,dz33);
1371
1372             /* Update vectorial force */
1373             fix3             = _mm256_add_pd(fix3,tx);
1374             fiy3             = _mm256_add_pd(fiy3,ty);
1375             fiz3             = _mm256_add_pd(fiz3,tz);
1376
1377             fjx3             = _mm256_add_pd(fjx3,tx);
1378             fjy3             = _mm256_add_pd(fjy3,ty);
1379             fjz3             = _mm256_add_pd(fjz3,tz);
1380
1381             fjptrA             = f+j_coord_offsetA;
1382             fjptrB             = f+j_coord_offsetB;
1383             fjptrC             = f+j_coord_offsetC;
1384             fjptrD             = f+j_coord_offsetD;
1385
1386             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1387                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1388
1389             /* Inner loop uses 243 flops */
1390         }
1391
1392         if(jidx<j_index_end)
1393         {
1394
1395             /* Get j neighbor index, and coordinate index */
1396             jnrlistA         = jjnr[jidx];
1397             jnrlistB         = jjnr[jidx+1];
1398             jnrlistC         = jjnr[jidx+2];
1399             jnrlistD         = jjnr[jidx+3];
1400             /* Sign of each element will be negative for non-real atoms.
1401              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1402              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1403              */
1404             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1405
1406             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1407             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1408             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1409
1410             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1411             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1412             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1413             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1414             j_coord_offsetA  = DIM*jnrA;
1415             j_coord_offsetB  = DIM*jnrB;
1416             j_coord_offsetC  = DIM*jnrC;
1417             j_coord_offsetD  = DIM*jnrD;
1418
1419             /* load j atom coordinates */
1420             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1421                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1422                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1423
1424             /* Calculate displacement vector */
1425             dx11             = _mm256_sub_pd(ix1,jx1);
1426             dy11             = _mm256_sub_pd(iy1,jy1);
1427             dz11             = _mm256_sub_pd(iz1,jz1);
1428             dx12             = _mm256_sub_pd(ix1,jx2);
1429             dy12             = _mm256_sub_pd(iy1,jy2);
1430             dz12             = _mm256_sub_pd(iz1,jz2);
1431             dx13             = _mm256_sub_pd(ix1,jx3);
1432             dy13             = _mm256_sub_pd(iy1,jy3);
1433             dz13             = _mm256_sub_pd(iz1,jz3);
1434             dx21             = _mm256_sub_pd(ix2,jx1);
1435             dy21             = _mm256_sub_pd(iy2,jy1);
1436             dz21             = _mm256_sub_pd(iz2,jz1);
1437             dx22             = _mm256_sub_pd(ix2,jx2);
1438             dy22             = _mm256_sub_pd(iy2,jy2);
1439             dz22             = _mm256_sub_pd(iz2,jz2);
1440             dx23             = _mm256_sub_pd(ix2,jx3);
1441             dy23             = _mm256_sub_pd(iy2,jy3);
1442             dz23             = _mm256_sub_pd(iz2,jz3);
1443             dx31             = _mm256_sub_pd(ix3,jx1);
1444             dy31             = _mm256_sub_pd(iy3,jy1);
1445             dz31             = _mm256_sub_pd(iz3,jz1);
1446             dx32             = _mm256_sub_pd(ix3,jx2);
1447             dy32             = _mm256_sub_pd(iy3,jy2);
1448             dz32             = _mm256_sub_pd(iz3,jz2);
1449             dx33             = _mm256_sub_pd(ix3,jx3);
1450             dy33             = _mm256_sub_pd(iy3,jy3);
1451             dz33             = _mm256_sub_pd(iz3,jz3);
1452
1453             /* Calculate squared distance and things based on it */
1454             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1455             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1456             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1457             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1458             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1459             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1460             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1461             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1462             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1463
1464             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1465             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1466             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1467             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1468             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1469             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1470             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1471             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1472             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1473
1474             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1475             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1476             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1477             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1478             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1479             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1480             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1481             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1482             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1483
1484             fjx1             = _mm256_setzero_pd();
1485             fjy1             = _mm256_setzero_pd();
1486             fjz1             = _mm256_setzero_pd();
1487             fjx2             = _mm256_setzero_pd();
1488             fjy2             = _mm256_setzero_pd();
1489             fjz2             = _mm256_setzero_pd();
1490             fjx3             = _mm256_setzero_pd();
1491             fjy3             = _mm256_setzero_pd();
1492             fjz3             = _mm256_setzero_pd();
1493
1494             /**************************
1495              * CALCULATE INTERACTIONS *
1496              **************************/
1497
1498             /* REACTION-FIELD ELECTROSTATICS */
1499             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1500
1501             fscal            = felec;
1502
1503             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1504
1505             /* Calculate temporary vectorial force */
1506             tx               = _mm256_mul_pd(fscal,dx11);
1507             ty               = _mm256_mul_pd(fscal,dy11);
1508             tz               = _mm256_mul_pd(fscal,dz11);
1509
1510             /* Update vectorial force */
1511             fix1             = _mm256_add_pd(fix1,tx);
1512             fiy1             = _mm256_add_pd(fiy1,ty);
1513             fiz1             = _mm256_add_pd(fiz1,tz);
1514
1515             fjx1             = _mm256_add_pd(fjx1,tx);
1516             fjy1             = _mm256_add_pd(fjy1,ty);
1517             fjz1             = _mm256_add_pd(fjz1,tz);
1518
1519             /**************************
1520              * CALCULATE INTERACTIONS *
1521              **************************/
1522
1523             /* REACTION-FIELD ELECTROSTATICS */
1524             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1525
1526             fscal            = felec;
1527
1528             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1529
1530             /* Calculate temporary vectorial force */
1531             tx               = _mm256_mul_pd(fscal,dx12);
1532             ty               = _mm256_mul_pd(fscal,dy12);
1533             tz               = _mm256_mul_pd(fscal,dz12);
1534
1535             /* Update vectorial force */
1536             fix1             = _mm256_add_pd(fix1,tx);
1537             fiy1             = _mm256_add_pd(fiy1,ty);
1538             fiz1             = _mm256_add_pd(fiz1,tz);
1539
1540             fjx2             = _mm256_add_pd(fjx2,tx);
1541             fjy2             = _mm256_add_pd(fjy2,ty);
1542             fjz2             = _mm256_add_pd(fjz2,tz);
1543
1544             /**************************
1545              * CALCULATE INTERACTIONS *
1546              **************************/
1547
1548             /* REACTION-FIELD ELECTROSTATICS */
1549             felec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_mul_pd(rinv13,rinvsq13),krf2));
1550
1551             fscal            = felec;
1552
1553             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1554
1555             /* Calculate temporary vectorial force */
1556             tx               = _mm256_mul_pd(fscal,dx13);
1557             ty               = _mm256_mul_pd(fscal,dy13);
1558             tz               = _mm256_mul_pd(fscal,dz13);
1559
1560             /* Update vectorial force */
1561             fix1             = _mm256_add_pd(fix1,tx);
1562             fiy1             = _mm256_add_pd(fiy1,ty);
1563             fiz1             = _mm256_add_pd(fiz1,tz);
1564
1565             fjx3             = _mm256_add_pd(fjx3,tx);
1566             fjy3             = _mm256_add_pd(fjy3,ty);
1567             fjz3             = _mm256_add_pd(fjz3,tz);
1568
1569             /**************************
1570              * CALCULATE INTERACTIONS *
1571              **************************/
1572
1573             /* REACTION-FIELD ELECTROSTATICS */
1574             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1575
1576             fscal            = felec;
1577
1578             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1579
1580             /* Calculate temporary vectorial force */
1581             tx               = _mm256_mul_pd(fscal,dx21);
1582             ty               = _mm256_mul_pd(fscal,dy21);
1583             tz               = _mm256_mul_pd(fscal,dz21);
1584
1585             /* Update vectorial force */
1586             fix2             = _mm256_add_pd(fix2,tx);
1587             fiy2             = _mm256_add_pd(fiy2,ty);
1588             fiz2             = _mm256_add_pd(fiz2,tz);
1589
1590             fjx1             = _mm256_add_pd(fjx1,tx);
1591             fjy1             = _mm256_add_pd(fjy1,ty);
1592             fjz1             = _mm256_add_pd(fjz1,tz);
1593
1594             /**************************
1595              * CALCULATE INTERACTIONS *
1596              **************************/
1597
1598             /* REACTION-FIELD ELECTROSTATICS */
1599             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1600
1601             fscal            = felec;
1602
1603             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1604
1605             /* Calculate temporary vectorial force */
1606             tx               = _mm256_mul_pd(fscal,dx22);
1607             ty               = _mm256_mul_pd(fscal,dy22);
1608             tz               = _mm256_mul_pd(fscal,dz22);
1609
1610             /* Update vectorial force */
1611             fix2             = _mm256_add_pd(fix2,tx);
1612             fiy2             = _mm256_add_pd(fiy2,ty);
1613             fiz2             = _mm256_add_pd(fiz2,tz);
1614
1615             fjx2             = _mm256_add_pd(fjx2,tx);
1616             fjy2             = _mm256_add_pd(fjy2,ty);
1617             fjz2             = _mm256_add_pd(fjz2,tz);
1618
1619             /**************************
1620              * CALCULATE INTERACTIONS *
1621              **************************/
1622
1623             /* REACTION-FIELD ELECTROSTATICS */
1624             felec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_mul_pd(rinv23,rinvsq23),krf2));
1625
1626             fscal            = felec;
1627
1628             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1629
1630             /* Calculate temporary vectorial force */
1631             tx               = _mm256_mul_pd(fscal,dx23);
1632             ty               = _mm256_mul_pd(fscal,dy23);
1633             tz               = _mm256_mul_pd(fscal,dz23);
1634
1635             /* Update vectorial force */
1636             fix2             = _mm256_add_pd(fix2,tx);
1637             fiy2             = _mm256_add_pd(fiy2,ty);
1638             fiz2             = _mm256_add_pd(fiz2,tz);
1639
1640             fjx3             = _mm256_add_pd(fjx3,tx);
1641             fjy3             = _mm256_add_pd(fjy3,ty);
1642             fjz3             = _mm256_add_pd(fjz3,tz);
1643
1644             /**************************
1645              * CALCULATE INTERACTIONS *
1646              **************************/
1647
1648             /* REACTION-FIELD ELECTROSTATICS */
1649             felec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_mul_pd(rinv31,rinvsq31),krf2));
1650
1651             fscal            = felec;
1652
1653             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1654
1655             /* Calculate temporary vectorial force */
1656             tx               = _mm256_mul_pd(fscal,dx31);
1657             ty               = _mm256_mul_pd(fscal,dy31);
1658             tz               = _mm256_mul_pd(fscal,dz31);
1659
1660             /* Update vectorial force */
1661             fix3             = _mm256_add_pd(fix3,tx);
1662             fiy3             = _mm256_add_pd(fiy3,ty);
1663             fiz3             = _mm256_add_pd(fiz3,tz);
1664
1665             fjx1             = _mm256_add_pd(fjx1,tx);
1666             fjy1             = _mm256_add_pd(fjy1,ty);
1667             fjz1             = _mm256_add_pd(fjz1,tz);
1668
1669             /**************************
1670              * CALCULATE INTERACTIONS *
1671              **************************/
1672
1673             /* REACTION-FIELD ELECTROSTATICS */
1674             felec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_mul_pd(rinv32,rinvsq32),krf2));
1675
1676             fscal            = felec;
1677
1678             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1679
1680             /* Calculate temporary vectorial force */
1681             tx               = _mm256_mul_pd(fscal,dx32);
1682             ty               = _mm256_mul_pd(fscal,dy32);
1683             tz               = _mm256_mul_pd(fscal,dz32);
1684
1685             /* Update vectorial force */
1686             fix3             = _mm256_add_pd(fix3,tx);
1687             fiy3             = _mm256_add_pd(fiy3,ty);
1688             fiz3             = _mm256_add_pd(fiz3,tz);
1689
1690             fjx2             = _mm256_add_pd(fjx2,tx);
1691             fjy2             = _mm256_add_pd(fjy2,ty);
1692             fjz2             = _mm256_add_pd(fjz2,tz);
1693
1694             /**************************
1695              * CALCULATE INTERACTIONS *
1696              **************************/
1697
1698             /* REACTION-FIELD ELECTROSTATICS */
1699             felec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_mul_pd(rinv33,rinvsq33),krf2));
1700
1701             fscal            = felec;
1702
1703             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1704
1705             /* Calculate temporary vectorial force */
1706             tx               = _mm256_mul_pd(fscal,dx33);
1707             ty               = _mm256_mul_pd(fscal,dy33);
1708             tz               = _mm256_mul_pd(fscal,dz33);
1709
1710             /* Update vectorial force */
1711             fix3             = _mm256_add_pd(fix3,tx);
1712             fiy3             = _mm256_add_pd(fiy3,ty);
1713             fiz3             = _mm256_add_pd(fiz3,tz);
1714
1715             fjx3             = _mm256_add_pd(fjx3,tx);
1716             fjy3             = _mm256_add_pd(fjy3,ty);
1717             fjz3             = _mm256_add_pd(fjz3,tz);
1718
1719             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1720             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1721             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1722             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1723
1724             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1725                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1726
1727             /* Inner loop uses 243 flops */
1728         }
1729
1730         /* End of innermost loop */
1731
1732         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1733                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1734
1735         /* Increment number of inner iterations */
1736         inneriter                  += j_index_end - j_index_start;
1737
1738         /* Outer loop uses 18 flops */
1739     }
1740
1741     /* Increment number of outer iterations */
1742     outeriter        += nri;
1743
1744     /* Update outer/inner flops */
1745
1746     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*243);
1747 }