src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwNone_GeomW3W3_avx_256_single.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_single 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_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3W3_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwNone_GeomW3W3_VF_avx_256_single
  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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  97     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  98     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  99     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 100     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 101     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 102     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 103     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 104     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 105     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 106     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 107     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 108     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 109     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 110     real             *charge;
 111     __m256i          ewitab;
 112     __m128i          ewitab_lo,ewitab_hi;
 113     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 114     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 115     real             *ewtab;
 116     __m256           dummy_mask,cutoff_mask;
 117     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 118     __m256           one     = _mm256_set1_ps(1.0);
 119     __m256           two     = _mm256_set1_ps(2.0);
 120     x                = xx[0];
 121     f                = ff[0];
 122
 123     nri              = nlist->nri;
 124     iinr             = nlist->iinr;
 125     jindex           = nlist->jindex;
 126     jjnr             = nlist->jjnr;
 127     shiftidx         = nlist->shift;
 128     gid              = nlist->gid;
 129     shiftvec         = fr->shift_vec[0];
 130     fshift           = fr->fshift[0];
 131     facel            = _mm256_set1_ps(fr->epsfac);
 132     charge           = mdatoms->chargeA;
 133
 134     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 135     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 136     beta2            = _mm256_mul_ps(beta,beta);
 137     beta3            = _mm256_mul_ps(beta,beta2);
 138
 139     ewtab            = fr->ic->tabq_coul_FDV0;
 140     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 141     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 146     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 147     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 148
 149     jq0              = _mm256_set1_ps(charge[inr+0]);
 150     jq1              = _mm256_set1_ps(charge[inr+1]);
 151     jq2              = _mm256_set1_ps(charge[inr+2]);
 152     qq00             = _mm256_mul_ps(iq0,jq0);
 153     qq01             = _mm256_mul_ps(iq0,jq1);
 154     qq02             = _mm256_mul_ps(iq0,jq2);
 155     qq10             = _mm256_mul_ps(iq1,jq0);
 156     qq11             = _mm256_mul_ps(iq1,jq1);
 157     qq12             = _mm256_mul_ps(iq1,jq2);
 158     qq20             = _mm256_mul_ps(iq2,jq0);
 159     qq21             = _mm256_mul_ps(iq2,jq1);
 160     qq22             = _mm256_mul_ps(iq2,jq2);
 161
 162     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 163     rcutoff_scalar   = fr->rcoulomb;
 164     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 165     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 166
 167     /* Avoid stupid compiler warnings */
 168     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 169     j_coord_offsetA = 0;
 170     j_coord_offsetB = 0;
 171     j_coord_offsetC = 0;
 172     j_coord_offsetD = 0;
 173     j_coord_offsetE = 0;
 174     j_coord_offsetF = 0;
 175     j_coord_offsetG = 0;
 176     j_coord_offsetH = 0;
 177
 178     outeriter        = 0;
 179     inneriter        = 0;
 180
 181     for(iidx=0;iidx<4*DIM;iidx++)
 182     {
 183         scratch[iidx] = 0.0;
 184     }
 185
 186     /* Start outer loop over neighborlists */
 187     for(iidx=0; iidx<nri; iidx++)
 188     {
 189         /* Load shift vector for this list */
 190         i_shift_offset   = DIM*shiftidx[iidx];
 191
 192         /* Load limits for loop over neighbors */
 193         j_index_start    = jindex[iidx];
 194         j_index_end      = jindex[iidx+1];
 195
 196         /* Get outer coordinate index */
 197         inr              = iinr[iidx];
 198         i_coord_offset   = DIM*inr;
 199
 200         /* Load i particle coords and add shift vector */
 201         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 202                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 203
 204         fix0             = _mm256_setzero_ps();
 205         fiy0             = _mm256_setzero_ps();
 206         fiz0             = _mm256_setzero_ps();
 207         fix1             = _mm256_setzero_ps();
 208         fiy1             = _mm256_setzero_ps();
 209         fiz1             = _mm256_setzero_ps();
 210         fix2             = _mm256_setzero_ps();
 211         fiy2             = _mm256_setzero_ps();
 212         fiz2             = _mm256_setzero_ps();
 213
 214         /* Reset potential sums */
 215         velecsum         = _mm256_setzero_ps();
 216
 217         /* Start inner kernel loop */
 218         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 219         {
 220
 221             /* Get j neighbor index, and coordinate index */
 222             jnrA             = jjnr[jidx];
 223             jnrB             = jjnr[jidx+1];
 224             jnrC             = jjnr[jidx+2];
 225             jnrD             = jjnr[jidx+3];
 226             jnrE             = jjnr[jidx+4];
 227             jnrF             = jjnr[jidx+5];
 228             jnrG             = jjnr[jidx+6];
 229             jnrH             = jjnr[jidx+7];
 230             j_coord_offsetA  = DIM*jnrA;
 231             j_coord_offsetB  = DIM*jnrB;
 232             j_coord_offsetC  = DIM*jnrC;
 233             j_coord_offsetD  = DIM*jnrD;
 234             j_coord_offsetE  = DIM*jnrE;
 235             j_coord_offsetF  = DIM*jnrF;
 236             j_coord_offsetG  = DIM*jnrG;
 237             j_coord_offsetH  = DIM*jnrH;
 238
 239             /* load j atom coordinates */
 240             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 241                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 242                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 243                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 244                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 245
 246             /* Calculate displacement vector */
 247             dx00             = _mm256_sub_ps(ix0,jx0);
 248             dy00             = _mm256_sub_ps(iy0,jy0);
 249             dz00             = _mm256_sub_ps(iz0,jz0);
 250             dx01             = _mm256_sub_ps(ix0,jx1);
 251             dy01             = _mm256_sub_ps(iy0,jy1);
 252             dz01             = _mm256_sub_ps(iz0,jz1);
 253             dx02             = _mm256_sub_ps(ix0,jx2);
 254             dy02             = _mm256_sub_ps(iy0,jy2);
 255             dz02             = _mm256_sub_ps(iz0,jz2);
 256             dx10             = _mm256_sub_ps(ix1,jx0);
 257             dy10             = _mm256_sub_ps(iy1,jy0);
 258             dz10             = _mm256_sub_ps(iz1,jz0);
 259             dx11             = _mm256_sub_ps(ix1,jx1);
 260             dy11             = _mm256_sub_ps(iy1,jy1);
 261             dz11             = _mm256_sub_ps(iz1,jz1);
 262             dx12             = _mm256_sub_ps(ix1,jx2);
 263             dy12             = _mm256_sub_ps(iy1,jy2);
 264             dz12             = _mm256_sub_ps(iz1,jz2);
 265             dx20             = _mm256_sub_ps(ix2,jx0);
 266             dy20             = _mm256_sub_ps(iy2,jy0);
 267             dz20             = _mm256_sub_ps(iz2,jz0);
 268             dx21             = _mm256_sub_ps(ix2,jx1);
 269             dy21             = _mm256_sub_ps(iy2,jy1);
 270             dz21             = _mm256_sub_ps(iz2,jz1);
 271             dx22             = _mm256_sub_ps(ix2,jx2);
 272             dy22             = _mm256_sub_ps(iy2,jy2);
 273             dz22             = _mm256_sub_ps(iz2,jz2);
 274
 275             /* Calculate squared distance and things based on it */
 276             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 277             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 278             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 279             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 280             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 281             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 282             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 283             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 284             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 285
 286             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 287             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 288             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 289             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 290             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 291             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 292             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 293             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 294             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 295
 296             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 297             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 298             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 299             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 300             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 301             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 302             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 303             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 304             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 305
 306             fjx0             = _mm256_setzero_ps();
 307             fjy0             = _mm256_setzero_ps();
 308             fjz0             = _mm256_setzero_ps();
 309             fjx1             = _mm256_setzero_ps();
 310             fjy1             = _mm256_setzero_ps();
 311             fjz1             = _mm256_setzero_ps();
 312             fjx2             = _mm256_setzero_ps();
 313             fjy2             = _mm256_setzero_ps();
 314             fjz2             = _mm256_setzero_ps();
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 321             {
 322
 323             r00              = _mm256_mul_ps(rsq00,rinv00);
 324
 325             /* EWALD ELECTROSTATICS */
 326
 327             /* Analytical PME correction */
 328             zeta2            = _mm256_mul_ps(beta2,rsq00);
 329             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 330             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 331             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 332             felec            = _mm256_mul_ps(qq00,felec);
 333             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 334             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 335             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 336             velec            = _mm256_mul_ps(qq00,velec);
 337
 338             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velec            = _mm256_and_ps(velec,cutoff_mask);
 342             velecsum         = _mm256_add_ps(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm256_mul_ps(fscal,dx00);
 350             ty               = _mm256_mul_ps(fscal,dy00);
 351             tz               = _mm256_mul_ps(fscal,dz00);
 352
 353             /* Update vectorial force */
 354             fix0             = _mm256_add_ps(fix0,tx);
 355             fiy0             = _mm256_add_ps(fiy0,ty);
 356             fiz0             = _mm256_add_ps(fiz0,tz);
 357
 358             fjx0             = _mm256_add_ps(fjx0,tx);
 359             fjy0             = _mm256_add_ps(fjy0,ty);
 360             fjz0             = _mm256_add_ps(fjz0,tz);
 361
 362             }
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 369             {
 370
 371             r01              = _mm256_mul_ps(rsq01,rinv01);
 372
 373             /* EWALD ELECTROSTATICS */
 374
 375             /* Analytical PME correction */
 376             zeta2            = _mm256_mul_ps(beta2,rsq01);
 377             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 378             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 379             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 380             felec            = _mm256_mul_ps(qq01,felec);
 381             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 382             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 383             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv01,sh_ewald),pmecorrV);
 384             velec            = _mm256_mul_ps(qq01,velec);
 385
 386             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velec            = _mm256_and_ps(velec,cutoff_mask);
 390             velecsum         = _mm256_add_ps(velecsum,velec);
 391
 392             fscal            = felec;
 393
 394             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm256_mul_ps(fscal,dx01);
 398             ty               = _mm256_mul_ps(fscal,dy01);
 399             tz               = _mm256_mul_ps(fscal,dz01);
 400
 401             /* Update vectorial force */
 402             fix0             = _mm256_add_ps(fix0,tx);
 403             fiy0             = _mm256_add_ps(fiy0,ty);
 404             fiz0             = _mm256_add_ps(fiz0,tz);
 405
 406             fjx1             = _mm256_add_ps(fjx1,tx);
 407             fjy1             = _mm256_add_ps(fjy1,ty);
 408             fjz1             = _mm256_add_ps(fjz1,tz);
 409
 410             }
 411
 412             /**************************
 413              * CALCULATE INTERACTIONS *
 414              **************************/
 415
 416             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 417             {
 418
 419             r02              = _mm256_mul_ps(rsq02,rinv02);
 420
 421             /* EWALD ELECTROSTATICS */
 422
 423             /* Analytical PME correction */
 424             zeta2            = _mm256_mul_ps(beta2,rsq02);
 425             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 426             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 427             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 428             felec            = _mm256_mul_ps(qq02,felec);
 429             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 430             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 431             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv02,sh_ewald),pmecorrV);
 432             velec            = _mm256_mul_ps(qq02,velec);
 433
 434             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
 435
 436             /* Update potential sum for this i atom from the interaction with this j atom. */
 437             velec            = _mm256_and_ps(velec,cutoff_mask);
 438             velecsum         = _mm256_add_ps(velecsum,velec);
 439
 440             fscal            = felec;
 441
 442             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 443
 444             /* Calculate temporary vectorial force */
 445             tx               = _mm256_mul_ps(fscal,dx02);
 446             ty               = _mm256_mul_ps(fscal,dy02);
 447             tz               = _mm256_mul_ps(fscal,dz02);
 448
 449             /* Update vectorial force */
 450             fix0             = _mm256_add_ps(fix0,tx);
 451             fiy0             = _mm256_add_ps(fiy0,ty);
 452             fiz0             = _mm256_add_ps(fiz0,tz);
 453
 454             fjx2             = _mm256_add_ps(fjx2,tx);
 455             fjy2             = _mm256_add_ps(fjy2,ty);
 456             fjz2             = _mm256_add_ps(fjz2,tz);
 457
 458             }
 459
 460             /**************************
 461              * CALCULATE INTERACTIONS *
 462              **************************/
 463
 464             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 465             {
 466
 467             r10              = _mm256_mul_ps(rsq10,rinv10);
 468
 469             /* EWALD ELECTROSTATICS */
 470
 471             /* Analytical PME correction */
 472             zeta2            = _mm256_mul_ps(beta2,rsq10);
 473             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 474             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 475             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 476             felec            = _mm256_mul_ps(qq10,felec);
 477             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 478             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 479             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 480             velec            = _mm256_mul_ps(qq10,velec);
 481
 482             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 483
 484             /* Update potential sum for this i atom from the interaction with this j atom. */
 485             velec            = _mm256_and_ps(velec,cutoff_mask);
 486             velecsum         = _mm256_add_ps(velecsum,velec);
 487
 488             fscal            = felec;
 489
 490             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 491
 492             /* Calculate temporary vectorial force */
 493             tx               = _mm256_mul_ps(fscal,dx10);
 494             ty               = _mm256_mul_ps(fscal,dy10);
 495             tz               = _mm256_mul_ps(fscal,dz10);
 496
 497             /* Update vectorial force */
 498             fix1             = _mm256_add_ps(fix1,tx);
 499             fiy1             = _mm256_add_ps(fiy1,ty);
 500             fiz1             = _mm256_add_ps(fiz1,tz);
 501
 502             fjx0             = _mm256_add_ps(fjx0,tx);
 503             fjy0             = _mm256_add_ps(fjy0,ty);
 504             fjz0             = _mm256_add_ps(fjz0,tz);
 505
 506             }
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 513             {
 514
 515             r11              = _mm256_mul_ps(rsq11,rinv11);
 516
 517             /* EWALD ELECTROSTATICS */
 518
 519             /* Analytical PME correction */
 520             zeta2            = _mm256_mul_ps(beta2,rsq11);
 521             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 522             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 523             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 524             felec            = _mm256_mul_ps(qq11,felec);
 525             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 526             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 527             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
 528             velec            = _mm256_mul_ps(qq11,velec);
 529
 530             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 531
 532             /* Update potential sum for this i atom from the interaction with this j atom. */
 533             velec            = _mm256_and_ps(velec,cutoff_mask);
 534             velecsum         = _mm256_add_ps(velecsum,velec);
 535
 536             fscal            = felec;
 537
 538             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 539
 540             /* Calculate temporary vectorial force */
 541             tx               = _mm256_mul_ps(fscal,dx11);
 542             ty               = _mm256_mul_ps(fscal,dy11);
 543             tz               = _mm256_mul_ps(fscal,dz11);
 544
 545             /* Update vectorial force */
 546             fix1             = _mm256_add_ps(fix1,tx);
 547             fiy1             = _mm256_add_ps(fiy1,ty);
 548             fiz1             = _mm256_add_ps(fiz1,tz);
 549
 550             fjx1             = _mm256_add_ps(fjx1,tx);
 551             fjy1             = _mm256_add_ps(fjy1,ty);
 552             fjz1             = _mm256_add_ps(fjz1,tz);
 553
 554             }
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 561             {
 562
 563             r12              = _mm256_mul_ps(rsq12,rinv12);
 564
 565             /* EWALD ELECTROSTATICS */
 566
 567             /* Analytical PME correction */
 568             zeta2            = _mm256_mul_ps(beta2,rsq12);
 569             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 570             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 571             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 572             felec            = _mm256_mul_ps(qq12,felec);
 573             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 574             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 575             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
 576             velec            = _mm256_mul_ps(qq12,velec);
 577
 578             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 579
 580             /* Update potential sum for this i atom from the interaction with this j atom. */
 581             velec            = _mm256_and_ps(velec,cutoff_mask);
 582             velecsum         = _mm256_add_ps(velecsum,velec);
 583
 584             fscal            = felec;
 585
 586             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 587
 588             /* Calculate temporary vectorial force */
 589             tx               = _mm256_mul_ps(fscal,dx12);
 590             ty               = _mm256_mul_ps(fscal,dy12);
 591             tz               = _mm256_mul_ps(fscal,dz12);
 592
 593             /* Update vectorial force */
 594             fix1             = _mm256_add_ps(fix1,tx);
 595             fiy1             = _mm256_add_ps(fiy1,ty);
 596             fiz1             = _mm256_add_ps(fiz1,tz);
 597
 598             fjx2             = _mm256_add_ps(fjx2,tx);
 599             fjy2             = _mm256_add_ps(fjy2,ty);
 600             fjz2             = _mm256_add_ps(fjz2,tz);
 601
 602             }
 603
 604             /**************************
 605              * CALCULATE INTERACTIONS *
 606              **************************/
 607
 608             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 609             {
 610
 611             r20              = _mm256_mul_ps(rsq20,rinv20);
 612
 613             /* EWALD ELECTROSTATICS */
 614
 615             /* Analytical PME correction */
 616             zeta2            = _mm256_mul_ps(beta2,rsq20);
 617             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 618             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 619             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 620             felec            = _mm256_mul_ps(qq20,felec);
 621             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 622             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 623             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 624             velec            = _mm256_mul_ps(qq20,velec);
 625
 626             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 627
 628             /* Update potential sum for this i atom from the interaction with this j atom. */
 629             velec            = _mm256_and_ps(velec,cutoff_mask);
 630             velecsum         = _mm256_add_ps(velecsum,velec);
 631
 632             fscal            = felec;
 633
 634             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 635
 636             /* Calculate temporary vectorial force */
 637             tx               = _mm256_mul_ps(fscal,dx20);
 638             ty               = _mm256_mul_ps(fscal,dy20);
 639             tz               = _mm256_mul_ps(fscal,dz20);
 640
 641             /* Update vectorial force */
 642             fix2             = _mm256_add_ps(fix2,tx);
 643             fiy2             = _mm256_add_ps(fiy2,ty);
 644             fiz2             = _mm256_add_ps(fiz2,tz);
 645
 646             fjx0             = _mm256_add_ps(fjx0,tx);
 647             fjy0             = _mm256_add_ps(fjy0,ty);
 648             fjz0             = _mm256_add_ps(fjz0,tz);
 649
 650             }
 651
 652             /**************************
 653              * CALCULATE INTERACTIONS *
 654              **************************/
 655
 656             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 657             {
 658
 659             r21              = _mm256_mul_ps(rsq21,rinv21);
 660
 661             /* EWALD ELECTROSTATICS */
 662
 663             /* Analytical PME correction */
 664             zeta2            = _mm256_mul_ps(beta2,rsq21);
 665             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 666             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 667             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 668             felec            = _mm256_mul_ps(qq21,felec);
 669             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 670             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 671             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
 672             velec            = _mm256_mul_ps(qq21,velec);
 673
 674             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
 675
 676             /* Update potential sum for this i atom from the interaction with this j atom. */
 677             velec            = _mm256_and_ps(velec,cutoff_mask);
 678             velecsum         = _mm256_add_ps(velecsum,velec);
 679
 680             fscal            = felec;
 681
 682             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 683
 684             /* Calculate temporary vectorial force */
 685             tx               = _mm256_mul_ps(fscal,dx21);
 686             ty               = _mm256_mul_ps(fscal,dy21);
 687             tz               = _mm256_mul_ps(fscal,dz21);
 688
 689             /* Update vectorial force */
 690             fix2             = _mm256_add_ps(fix2,tx);
 691             fiy2             = _mm256_add_ps(fiy2,ty);
 692             fiz2             = _mm256_add_ps(fiz2,tz);
 693
 694             fjx1             = _mm256_add_ps(fjx1,tx);
 695             fjy1             = _mm256_add_ps(fjy1,ty);
 696             fjz1             = _mm256_add_ps(fjz1,tz);
 697
 698             }
 699
 700             /**************************
 701              * CALCULATE INTERACTIONS *
 702              **************************/
 703
 704             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 705             {
 706
 707             r22              = _mm256_mul_ps(rsq22,rinv22);
 708
 709             /* EWALD ELECTROSTATICS */
 710
 711             /* Analytical PME correction */
 712             zeta2            = _mm256_mul_ps(beta2,rsq22);
 713             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 714             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 715             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 716             felec            = _mm256_mul_ps(qq22,felec);
 717             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 718             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 719             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
 720             velec            = _mm256_mul_ps(qq22,velec);
 721
 722             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
 723
 724             /* Update potential sum for this i atom from the interaction with this j atom. */
 725             velec            = _mm256_and_ps(velec,cutoff_mask);
 726             velecsum         = _mm256_add_ps(velecsum,velec);
 727
 728             fscal            = felec;
 729
 730             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 731
 732             /* Calculate temporary vectorial force */
 733             tx               = _mm256_mul_ps(fscal,dx22);
 734             ty               = _mm256_mul_ps(fscal,dy22);
 735             tz               = _mm256_mul_ps(fscal,dz22);
 736
 737             /* Update vectorial force */
 738             fix2             = _mm256_add_ps(fix2,tx);
 739             fiy2             = _mm256_add_ps(fiy2,ty);
 740             fiz2             = _mm256_add_ps(fiz2,tz);
 741
 742             fjx2             = _mm256_add_ps(fjx2,tx);
 743             fjy2             = _mm256_add_ps(fjy2,ty);
 744             fjz2             = _mm256_add_ps(fjz2,tz);
 745
 746             }
 747
 748             fjptrA             = f+j_coord_offsetA;
 749             fjptrB             = f+j_coord_offsetB;
 750             fjptrC             = f+j_coord_offsetC;
 751             fjptrD             = f+j_coord_offsetD;
 752             fjptrE             = f+j_coord_offsetE;
 753             fjptrF             = f+j_coord_offsetF;
 754             fjptrG             = f+j_coord_offsetG;
 755             fjptrH             = f+j_coord_offsetH;
 756
 757             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 758                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 759
 760             /* Inner loop uses 981 flops */
 761         }
 762
 763         if(jidx<j_index_end)
 764         {
 765
 766             /* Get j neighbor index, and coordinate index */
 767             jnrlistA         = jjnr[jidx];
 768             jnrlistB         = jjnr[jidx+1];
 769             jnrlistC         = jjnr[jidx+2];
 770             jnrlistD         = jjnr[jidx+3];
 771             jnrlistE         = jjnr[jidx+4];
 772             jnrlistF         = jjnr[jidx+5];
 773             jnrlistG         = jjnr[jidx+6];
 774             jnrlistH         = jjnr[jidx+7];
 775             /* Sign of each element will be negative for non-real atoms.
 776              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 777              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 778              */
 779             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 780                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 781
 782             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 783             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 784             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 785             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 786             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 787             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 788             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 789             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 790             j_coord_offsetA  = DIM*jnrA;
 791             j_coord_offsetB  = DIM*jnrB;
 792             j_coord_offsetC  = DIM*jnrC;
 793             j_coord_offsetD  = DIM*jnrD;
 794             j_coord_offsetE  = DIM*jnrE;
 795             j_coord_offsetF  = DIM*jnrF;
 796             j_coord_offsetG  = DIM*jnrG;
 797             j_coord_offsetH  = DIM*jnrH;
 798
 799             /* load j atom coordinates */
 800             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 801                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 802                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 803                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 804                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 805
 806             /* Calculate displacement vector */
 807             dx00             = _mm256_sub_ps(ix0,jx0);
 808             dy00             = _mm256_sub_ps(iy0,jy0);
 809             dz00             = _mm256_sub_ps(iz0,jz0);
 810             dx01             = _mm256_sub_ps(ix0,jx1);
 811             dy01             = _mm256_sub_ps(iy0,jy1);
 812             dz01             = _mm256_sub_ps(iz0,jz1);
 813             dx02             = _mm256_sub_ps(ix0,jx2);
 814             dy02             = _mm256_sub_ps(iy0,jy2);
 815             dz02             = _mm256_sub_ps(iz0,jz2);
 816             dx10             = _mm256_sub_ps(ix1,jx0);
 817             dy10             = _mm256_sub_ps(iy1,jy0);
 818             dz10             = _mm256_sub_ps(iz1,jz0);
 819             dx11             = _mm256_sub_ps(ix1,jx1);
 820             dy11             = _mm256_sub_ps(iy1,jy1);
 821             dz11             = _mm256_sub_ps(iz1,jz1);
 822             dx12             = _mm256_sub_ps(ix1,jx2);
 823             dy12             = _mm256_sub_ps(iy1,jy2);
 824             dz12             = _mm256_sub_ps(iz1,jz2);
 825             dx20             = _mm256_sub_ps(ix2,jx0);
 826             dy20             = _mm256_sub_ps(iy2,jy0);
 827             dz20             = _mm256_sub_ps(iz2,jz0);
 828             dx21             = _mm256_sub_ps(ix2,jx1);
 829             dy21             = _mm256_sub_ps(iy2,jy1);
 830             dz21             = _mm256_sub_ps(iz2,jz1);
 831             dx22             = _mm256_sub_ps(ix2,jx2);
 832             dy22             = _mm256_sub_ps(iy2,jy2);
 833             dz22             = _mm256_sub_ps(iz2,jz2);
 834
 835             /* Calculate squared distance and things based on it */
 836             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 837             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 838             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 839             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 840             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 841             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 842             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 843             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 844             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 845
 846             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 847             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 848             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 849             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 850             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 851             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 852             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 853             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 854             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 855
 856             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 857             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 858             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 859             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 860             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 861             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 862             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 863             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 864             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 865
 866             fjx0             = _mm256_setzero_ps();
 867             fjy0             = _mm256_setzero_ps();
 868             fjz0             = _mm256_setzero_ps();
 869             fjx1             = _mm256_setzero_ps();
 870             fjy1             = _mm256_setzero_ps();
 871             fjz1             = _mm256_setzero_ps();
 872             fjx2             = _mm256_setzero_ps();
 873             fjy2             = _mm256_setzero_ps();
 874             fjz2             = _mm256_setzero_ps();
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 881             {
 882
 883             r00              = _mm256_mul_ps(rsq00,rinv00);
 884             r00              = _mm256_andnot_ps(dummy_mask,r00);
 885
 886             /* EWALD ELECTROSTATICS */
 887
 888             /* Analytical PME correction */
 889             zeta2            = _mm256_mul_ps(beta2,rsq00);
 890             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 891             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 892             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 893             felec            = _mm256_mul_ps(qq00,felec);
 894             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 895             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 896             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 897             velec            = _mm256_mul_ps(qq00,velec);
 898
 899             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 900
 901             /* Update potential sum for this i atom from the interaction with this j atom. */
 902             velec            = _mm256_and_ps(velec,cutoff_mask);
 903             velec            = _mm256_andnot_ps(dummy_mask,velec);
 904             velecsum         = _mm256_add_ps(velecsum,velec);
 905
 906             fscal            = felec;
 907
 908             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 909
 910             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 911
 912             /* Calculate temporary vectorial force */
 913             tx               = _mm256_mul_ps(fscal,dx00);
 914             ty               = _mm256_mul_ps(fscal,dy00);
 915             tz               = _mm256_mul_ps(fscal,dz00);
 916
 917             /* Update vectorial force */
 918             fix0             = _mm256_add_ps(fix0,tx);
 919             fiy0             = _mm256_add_ps(fiy0,ty);
 920             fiz0             = _mm256_add_ps(fiz0,tz);
 921
 922             fjx0             = _mm256_add_ps(fjx0,tx);
 923             fjy0             = _mm256_add_ps(fjy0,ty);
 924             fjz0             = _mm256_add_ps(fjz0,tz);
 925
 926             }
 927
 928             /**************************
 929              * CALCULATE INTERACTIONS *
 930              **************************/
 931
 932             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 933             {
 934
 935             r01              = _mm256_mul_ps(rsq01,rinv01);
 936             r01              = _mm256_andnot_ps(dummy_mask,r01);
 937
 938             /* EWALD ELECTROSTATICS */
 939
 940             /* Analytical PME correction */
 941             zeta2            = _mm256_mul_ps(beta2,rsq01);
 942             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 943             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 944             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 945             felec            = _mm256_mul_ps(qq01,felec);
 946             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 947             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 948             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv01,sh_ewald),pmecorrV);
 949             velec            = _mm256_mul_ps(qq01,velec);
 950
 951             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 952
 953             /* Update potential sum for this i atom from the interaction with this j atom. */
 954             velec            = _mm256_and_ps(velec,cutoff_mask);
 955             velec            = _mm256_andnot_ps(dummy_mask,velec);
 956             velecsum         = _mm256_add_ps(velecsum,velec);
 957
 958             fscal            = felec;
 959
 960             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 961
 962             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 963
 964             /* Calculate temporary vectorial force */
 965             tx               = _mm256_mul_ps(fscal,dx01);
 966             ty               = _mm256_mul_ps(fscal,dy01);
 967             tz               = _mm256_mul_ps(fscal,dz01);
 968
 969             /* Update vectorial force */
 970             fix0             = _mm256_add_ps(fix0,tx);
 971             fiy0             = _mm256_add_ps(fiy0,ty);
 972             fiz0             = _mm256_add_ps(fiz0,tz);
 973
 974             fjx1             = _mm256_add_ps(fjx1,tx);
 975             fjy1             = _mm256_add_ps(fjy1,ty);
 976             fjz1             = _mm256_add_ps(fjz1,tz);
 977
 978             }
 979
 980             /**************************
 981              * CALCULATE INTERACTIONS *
 982              **************************/
 983
 984             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 985             {
 986
 987             r02              = _mm256_mul_ps(rsq02,rinv02);
 988             r02              = _mm256_andnot_ps(dummy_mask,r02);
 989
 990             /* EWALD ELECTROSTATICS */
 991
 992             /* Analytical PME correction */
 993             zeta2            = _mm256_mul_ps(beta2,rsq02);
 994             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 995             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 996             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 997             felec            = _mm256_mul_ps(qq02,felec);
 998             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 999             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1000             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv02,sh_ewald),pmecorrV);
1001             velec            = _mm256_mul_ps(qq02,velec);
1002
1003             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1004
1005             /* Update potential sum for this i atom from the interaction with this j atom. */
1006             velec            = _mm256_and_ps(velec,cutoff_mask);
1007             velec            = _mm256_andnot_ps(dummy_mask,velec);
1008             velecsum         = _mm256_add_ps(velecsum,velec);
1009
1010             fscal            = felec;
1011
1012             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1013
1014             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1015
1016             /* Calculate temporary vectorial force */
1017             tx               = _mm256_mul_ps(fscal,dx02);
1018             ty               = _mm256_mul_ps(fscal,dy02);
1019             tz               = _mm256_mul_ps(fscal,dz02);
1020
1021             /* Update vectorial force */
1022             fix0             = _mm256_add_ps(fix0,tx);
1023             fiy0             = _mm256_add_ps(fiy0,ty);
1024             fiz0             = _mm256_add_ps(fiz0,tz);
1025
1026             fjx2             = _mm256_add_ps(fjx2,tx);
1027             fjy2             = _mm256_add_ps(fjy2,ty);
1028             fjz2             = _mm256_add_ps(fjz2,tz);
1029
1030             }
1031
1032             /**************************
1033              * CALCULATE INTERACTIONS *
1034              **************************/
1035
1036             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1037             {
1038
1039             r10              = _mm256_mul_ps(rsq10,rinv10);
1040             r10              = _mm256_andnot_ps(dummy_mask,r10);
1041
1042             /* EWALD ELECTROSTATICS */
1043
1044             /* Analytical PME correction */
1045             zeta2            = _mm256_mul_ps(beta2,rsq10);
1046             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1047             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1048             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1049             felec            = _mm256_mul_ps(qq10,felec);
1050             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1051             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1052             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
1053             velec            = _mm256_mul_ps(qq10,velec);
1054
1055             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1056
1057             /* Update potential sum for this i atom from the interaction with this j atom. */
1058             velec            = _mm256_and_ps(velec,cutoff_mask);
1059             velec            = _mm256_andnot_ps(dummy_mask,velec);
1060             velecsum         = _mm256_add_ps(velecsum,velec);
1061
1062             fscal            = felec;
1063
1064             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1065
1066             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1067
1068             /* Calculate temporary vectorial force */
1069             tx               = _mm256_mul_ps(fscal,dx10);
1070             ty               = _mm256_mul_ps(fscal,dy10);
1071             tz               = _mm256_mul_ps(fscal,dz10);
1072
1073             /* Update vectorial force */
1074             fix1             = _mm256_add_ps(fix1,tx);
1075             fiy1             = _mm256_add_ps(fiy1,ty);
1076             fiz1             = _mm256_add_ps(fiz1,tz);
1077
1078             fjx0             = _mm256_add_ps(fjx0,tx);
1079             fjy0             = _mm256_add_ps(fjy0,ty);
1080             fjz0             = _mm256_add_ps(fjz0,tz);
1081
1082             }
1083
1084             /**************************
1085              * CALCULATE INTERACTIONS *
1086              **************************/
1087
1088             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1089             {
1090
1091             r11              = _mm256_mul_ps(rsq11,rinv11);
1092             r11              = _mm256_andnot_ps(dummy_mask,r11);
1093
1094             /* EWALD ELECTROSTATICS */
1095
1096             /* Analytical PME correction */
1097             zeta2            = _mm256_mul_ps(beta2,rsq11);
1098             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1099             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1100             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1101             felec            = _mm256_mul_ps(qq11,felec);
1102             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1103             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1104             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
1105             velec            = _mm256_mul_ps(qq11,velec);
1106
1107             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1108
1109             /* Update potential sum for this i atom from the interaction with this j atom. */
1110             velec            = _mm256_and_ps(velec,cutoff_mask);
1111             velec            = _mm256_andnot_ps(dummy_mask,velec);
1112             velecsum         = _mm256_add_ps(velecsum,velec);
1113
1114             fscal            = felec;
1115
1116             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1117
1118             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1119
1120             /* Calculate temporary vectorial force */
1121             tx               = _mm256_mul_ps(fscal,dx11);
1122             ty               = _mm256_mul_ps(fscal,dy11);
1123             tz               = _mm256_mul_ps(fscal,dz11);
1124
1125             /* Update vectorial force */
1126             fix1             = _mm256_add_ps(fix1,tx);
1127             fiy1             = _mm256_add_ps(fiy1,ty);
1128             fiz1             = _mm256_add_ps(fiz1,tz);
1129
1130             fjx1             = _mm256_add_ps(fjx1,tx);
1131             fjy1             = _mm256_add_ps(fjy1,ty);
1132             fjz1             = _mm256_add_ps(fjz1,tz);
1133
1134             }
1135
1136             /**************************
1137              * CALCULATE INTERACTIONS *
1138              **************************/
1139
1140             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1141             {
1142
1143             r12              = _mm256_mul_ps(rsq12,rinv12);
1144             r12              = _mm256_andnot_ps(dummy_mask,r12);
1145
1146             /* EWALD ELECTROSTATICS */
1147
1148             /* Analytical PME correction */
1149             zeta2            = _mm256_mul_ps(beta2,rsq12);
1150             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1151             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1152             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1153             felec            = _mm256_mul_ps(qq12,felec);
1154             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1155             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1156             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
1157             velec            = _mm256_mul_ps(qq12,velec);
1158
1159             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1160
1161             /* Update potential sum for this i atom from the interaction with this j atom. */
1162             velec            = _mm256_and_ps(velec,cutoff_mask);
1163             velec            = _mm256_andnot_ps(dummy_mask,velec);
1164             velecsum         = _mm256_add_ps(velecsum,velec);
1165
1166             fscal            = felec;
1167
1168             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1169
1170             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1171
1172             /* Calculate temporary vectorial force */
1173             tx               = _mm256_mul_ps(fscal,dx12);
1174             ty               = _mm256_mul_ps(fscal,dy12);
1175             tz               = _mm256_mul_ps(fscal,dz12);
1176
1177             /* Update vectorial force */
1178             fix1             = _mm256_add_ps(fix1,tx);
1179             fiy1             = _mm256_add_ps(fiy1,ty);
1180             fiz1             = _mm256_add_ps(fiz1,tz);
1181
1182             fjx2             = _mm256_add_ps(fjx2,tx);
1183             fjy2             = _mm256_add_ps(fjy2,ty);
1184             fjz2             = _mm256_add_ps(fjz2,tz);
1185
1186             }
1187
1188             /**************************
1189              * CALCULATE INTERACTIONS *
1190              **************************/
1191
1192             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1193             {
1194
1195             r20              = _mm256_mul_ps(rsq20,rinv20);
1196             r20              = _mm256_andnot_ps(dummy_mask,r20);
1197
1198             /* EWALD ELECTROSTATICS */
1199
1200             /* Analytical PME correction */
1201             zeta2            = _mm256_mul_ps(beta2,rsq20);
1202             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1203             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1204             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1205             felec            = _mm256_mul_ps(qq20,felec);
1206             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1207             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1208             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
1209             velec            = _mm256_mul_ps(qq20,velec);
1210
1211             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1212
1213             /* Update potential sum for this i atom from the interaction with this j atom. */
1214             velec            = _mm256_and_ps(velec,cutoff_mask);
1215             velec            = _mm256_andnot_ps(dummy_mask,velec);
1216             velecsum         = _mm256_add_ps(velecsum,velec);
1217
1218             fscal            = felec;
1219
1220             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1221
1222             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1223
1224             /* Calculate temporary vectorial force */
1225             tx               = _mm256_mul_ps(fscal,dx20);
1226             ty               = _mm256_mul_ps(fscal,dy20);
1227             tz               = _mm256_mul_ps(fscal,dz20);
1228
1229             /* Update vectorial force */
1230             fix2             = _mm256_add_ps(fix2,tx);
1231             fiy2             = _mm256_add_ps(fiy2,ty);
1232             fiz2             = _mm256_add_ps(fiz2,tz);
1233
1234             fjx0             = _mm256_add_ps(fjx0,tx);
1235             fjy0             = _mm256_add_ps(fjy0,ty);
1236             fjz0             = _mm256_add_ps(fjz0,tz);
1237
1238             }
1239
1240             /**************************
1241              * CALCULATE INTERACTIONS *
1242              **************************/
1243
1244             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1245             {
1246
1247             r21              = _mm256_mul_ps(rsq21,rinv21);
1248             r21              = _mm256_andnot_ps(dummy_mask,r21);
1249
1250             /* EWALD ELECTROSTATICS */
1251
1252             /* Analytical PME correction */
1253             zeta2            = _mm256_mul_ps(beta2,rsq21);
1254             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1255             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1256             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1257             felec            = _mm256_mul_ps(qq21,felec);
1258             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1259             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1260             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
1261             velec            = _mm256_mul_ps(qq21,velec);
1262
1263             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1264
1265             /* Update potential sum for this i atom from the interaction with this j atom. */
1266             velec            = _mm256_and_ps(velec,cutoff_mask);
1267             velec            = _mm256_andnot_ps(dummy_mask,velec);
1268             velecsum         = _mm256_add_ps(velecsum,velec);
1269
1270             fscal            = felec;
1271
1272             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1273
1274             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1275
1276             /* Calculate temporary vectorial force */
1277             tx               = _mm256_mul_ps(fscal,dx21);
1278             ty               = _mm256_mul_ps(fscal,dy21);
1279             tz               = _mm256_mul_ps(fscal,dz21);
1280
1281             /* Update vectorial force */
1282             fix2             = _mm256_add_ps(fix2,tx);
1283             fiy2             = _mm256_add_ps(fiy2,ty);
1284             fiz2             = _mm256_add_ps(fiz2,tz);
1285
1286             fjx1             = _mm256_add_ps(fjx1,tx);
1287             fjy1             = _mm256_add_ps(fjy1,ty);
1288             fjz1             = _mm256_add_ps(fjz1,tz);
1289
1290             }
1291
1292             /**************************
1293              * CALCULATE INTERACTIONS *
1294              **************************/
1295
1296             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1297             {
1298
1299             r22              = _mm256_mul_ps(rsq22,rinv22);
1300             r22              = _mm256_andnot_ps(dummy_mask,r22);
1301
1302             /* EWALD ELECTROSTATICS */
1303
1304             /* Analytical PME correction */
1305             zeta2            = _mm256_mul_ps(beta2,rsq22);
1306             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1307             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1308             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1309             felec            = _mm256_mul_ps(qq22,felec);
1310             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1311             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1312             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
1313             velec            = _mm256_mul_ps(qq22,velec);
1314
1315             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1316
1317             /* Update potential sum for this i atom from the interaction with this j atom. */
1318             velec            = _mm256_and_ps(velec,cutoff_mask);
1319             velec            = _mm256_andnot_ps(dummy_mask,velec);
1320             velecsum         = _mm256_add_ps(velecsum,velec);
1321
1322             fscal            = felec;
1323
1324             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1325
1326             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1327
1328             /* Calculate temporary vectorial force */
1329             tx               = _mm256_mul_ps(fscal,dx22);
1330             ty               = _mm256_mul_ps(fscal,dy22);
1331             tz               = _mm256_mul_ps(fscal,dz22);
1332
1333             /* Update vectorial force */
1334             fix2             = _mm256_add_ps(fix2,tx);
1335             fiy2             = _mm256_add_ps(fiy2,ty);
1336             fiz2             = _mm256_add_ps(fiz2,tz);
1337
1338             fjx2             = _mm256_add_ps(fjx2,tx);
1339             fjy2             = _mm256_add_ps(fjy2,ty);
1340             fjz2             = _mm256_add_ps(fjz2,tz);
1341
1342             }
1343
1344             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1345             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1346             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1347             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1348             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1349             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1350             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1351             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1352
1353             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1354                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1355
1356             /* Inner loop uses 990 flops */
1357         }
1358
1359         /* End of innermost loop */
1360
1361         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1362                                                  f+i_coord_offset,fshift+i_shift_offset);
1363
1364         ggid                        = gid[iidx];
1365         /* Update potential energies */
1366         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1367
1368         /* Increment number of inner iterations */
1369         inneriter                  += j_index_end - j_index_start;
1370
1371         /* Outer loop uses 19 flops */
1372     }
1373
1374     /* Increment number of outer iterations */
1375     outeriter        += nri;
1376
1377     /* Update outer/inner flops */
1378
1379     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*990);
1380 }
1381 /*
1382  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3W3_F_avx_256_single
1383  * Electrostatics interaction: Ewald
1384  * VdW interaction:            None
1385  * Geometry:                   Water3-Water3
1386  * Calculate force/pot:        Force
1387  */
1388 void
1389 nb_kernel_ElecEwSh_VdwNone_GeomW3W3_F_avx_256_single
1390                     (t_nblist                    * gmx_restrict       nlist,
1391                      rvec                        * gmx_restrict          xx,
1392                      rvec                        * gmx_restrict          ff,
1393                      t_forcerec                  * gmx_restrict          fr,
1394                      t_mdatoms                   * gmx_restrict     mdatoms,
1395                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1396                      t_nrnb                      * gmx_restrict        nrnb)
1397 {
1398     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1399      * just 0 for non-waters.
1400      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1401      * jnr indices corresponding to data put in the four positions in the SIMD register.
1402      */
1403     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1404     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1405     int              jnrA,jnrB,jnrC,jnrD;
1406     int              jnrE,jnrF,jnrG,jnrH;
1407     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1408     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1409     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1410     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1411     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1412     real             rcutoff_scalar;
1413     real             *shiftvec,*fshift,*x,*f;
1414     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1415     real             scratch[4*DIM];
1416     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1417     real *           vdwioffsetptr0;
1418     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1419     real *           vdwioffsetptr1;
1420     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1421     real *           vdwioffsetptr2;
1422     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1423     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1424     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1425     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1426     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1427     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1428     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1429     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1430     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1431     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1432     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1433     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1434     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1435     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1436     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1437     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1438     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1439     real             *charge;
1440     __m256i          ewitab;
1441     __m128i          ewitab_lo,ewitab_hi;
1442     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1443     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1444     real             *ewtab;
1445     __m256           dummy_mask,cutoff_mask;
1446     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1447     __m256           one     = _mm256_set1_ps(1.0);
1448     __m256           two     = _mm256_set1_ps(2.0);
1449     x                = xx[0];
1450     f                = ff[0];
1451
1452     nri              = nlist->nri;
1453     iinr             = nlist->iinr;
1454     jindex           = nlist->jindex;
1455     jjnr             = nlist->jjnr;
1456     shiftidx         = nlist->shift;
1457     gid              = nlist->gid;
1458     shiftvec         = fr->shift_vec[0];
1459     fshift           = fr->fshift[0];
1460     facel            = _mm256_set1_ps(fr->epsfac);
1461     charge           = mdatoms->chargeA;
1462
1463     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1464     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1465     beta2            = _mm256_mul_ps(beta,beta);
1466     beta3            = _mm256_mul_ps(beta,beta2);
1467
1468     ewtab            = fr->ic->tabq_coul_F;
1469     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1470     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1471
1472     /* Setup water-specific parameters */
1473     inr              = nlist->iinr[0];
1474     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1475     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1476     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1477
1478     jq0              = _mm256_set1_ps(charge[inr+0]);
1479     jq1              = _mm256_set1_ps(charge[inr+1]);
1480     jq2              = _mm256_set1_ps(charge[inr+2]);
1481     qq00             = _mm256_mul_ps(iq0,jq0);
1482     qq01             = _mm256_mul_ps(iq0,jq1);
1483     qq02             = _mm256_mul_ps(iq0,jq2);
1484     qq10             = _mm256_mul_ps(iq1,jq0);
1485     qq11             = _mm256_mul_ps(iq1,jq1);
1486     qq12             = _mm256_mul_ps(iq1,jq2);
1487     qq20             = _mm256_mul_ps(iq2,jq0);
1488     qq21             = _mm256_mul_ps(iq2,jq1);
1489     qq22             = _mm256_mul_ps(iq2,jq2);
1490
1491     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1492     rcutoff_scalar   = fr->rcoulomb;
1493     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1494     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1495
1496     /* Avoid stupid compiler warnings */
1497     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1498     j_coord_offsetA = 0;
1499     j_coord_offsetB = 0;
1500     j_coord_offsetC = 0;
1501     j_coord_offsetD = 0;
1502     j_coord_offsetE = 0;
1503     j_coord_offsetF = 0;
1504     j_coord_offsetG = 0;
1505     j_coord_offsetH = 0;
1506
1507     outeriter        = 0;
1508     inneriter        = 0;
1509
1510     for(iidx=0;iidx<4*DIM;iidx++)
1511     {
1512         scratch[iidx] = 0.0;
1513     }
1514
1515     /* Start outer loop over neighborlists */
1516     for(iidx=0; iidx<nri; iidx++)
1517     {
1518         /* Load shift vector for this list */
1519         i_shift_offset   = DIM*shiftidx[iidx];
1520
1521         /* Load limits for loop over neighbors */
1522         j_index_start    = jindex[iidx];
1523         j_index_end      = jindex[iidx+1];
1524
1525         /* Get outer coordinate index */
1526         inr              = iinr[iidx];
1527         i_coord_offset   = DIM*inr;
1528
1529         /* Load i particle coords and add shift vector */
1530         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1531                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1532
1533         fix0             = _mm256_setzero_ps();
1534         fiy0             = _mm256_setzero_ps();
1535         fiz0             = _mm256_setzero_ps();
1536         fix1             = _mm256_setzero_ps();
1537         fiy1             = _mm256_setzero_ps();
1538         fiz1             = _mm256_setzero_ps();
1539         fix2             = _mm256_setzero_ps();
1540         fiy2             = _mm256_setzero_ps();
1541         fiz2             = _mm256_setzero_ps();
1542
1543         /* Start inner kernel loop */
1544         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1545         {
1546
1547             /* Get j neighbor index, and coordinate index */
1548             jnrA             = jjnr[jidx];
1549             jnrB             = jjnr[jidx+1];
1550             jnrC             = jjnr[jidx+2];
1551             jnrD             = jjnr[jidx+3];
1552             jnrE             = jjnr[jidx+4];
1553             jnrF             = jjnr[jidx+5];
1554             jnrG             = jjnr[jidx+6];
1555             jnrH             = jjnr[jidx+7];
1556             j_coord_offsetA  = DIM*jnrA;
1557             j_coord_offsetB  = DIM*jnrB;
1558             j_coord_offsetC  = DIM*jnrC;
1559             j_coord_offsetD  = DIM*jnrD;
1560             j_coord_offsetE  = DIM*jnrE;
1561             j_coord_offsetF  = DIM*jnrF;
1562             j_coord_offsetG  = DIM*jnrG;
1563             j_coord_offsetH  = DIM*jnrH;
1564
1565             /* load j atom coordinates */
1566             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1567                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1568                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1569                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1570                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1571
1572             /* Calculate displacement vector */
1573             dx00             = _mm256_sub_ps(ix0,jx0);
1574             dy00             = _mm256_sub_ps(iy0,jy0);
1575             dz00             = _mm256_sub_ps(iz0,jz0);
1576             dx01             = _mm256_sub_ps(ix0,jx1);
1577             dy01             = _mm256_sub_ps(iy0,jy1);
1578             dz01             = _mm256_sub_ps(iz0,jz1);
1579             dx02             = _mm256_sub_ps(ix0,jx2);
1580             dy02             = _mm256_sub_ps(iy0,jy2);
1581             dz02             = _mm256_sub_ps(iz0,jz2);
1582             dx10             = _mm256_sub_ps(ix1,jx0);
1583             dy10             = _mm256_sub_ps(iy1,jy0);
1584             dz10             = _mm256_sub_ps(iz1,jz0);
1585             dx11             = _mm256_sub_ps(ix1,jx1);
1586             dy11             = _mm256_sub_ps(iy1,jy1);
1587             dz11             = _mm256_sub_ps(iz1,jz1);
1588             dx12             = _mm256_sub_ps(ix1,jx2);
1589             dy12             = _mm256_sub_ps(iy1,jy2);
1590             dz12             = _mm256_sub_ps(iz1,jz2);
1591             dx20             = _mm256_sub_ps(ix2,jx0);
1592             dy20             = _mm256_sub_ps(iy2,jy0);
1593             dz20             = _mm256_sub_ps(iz2,jz0);
1594             dx21             = _mm256_sub_ps(ix2,jx1);
1595             dy21             = _mm256_sub_ps(iy2,jy1);
1596             dz21             = _mm256_sub_ps(iz2,jz1);
1597             dx22             = _mm256_sub_ps(ix2,jx2);
1598             dy22             = _mm256_sub_ps(iy2,jy2);
1599             dz22             = _mm256_sub_ps(iz2,jz2);
1600
1601             /* Calculate squared distance and things based on it */
1602             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1603             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1604             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1605             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1606             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1607             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1608             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1609             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1610             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1611
1612             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1613             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1614             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1615             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1616             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1617             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1618             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1619             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1620             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1621
1622             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1623             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1624             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1625             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1626             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1627             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1628             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1629             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1630             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1631
1632             fjx0             = _mm256_setzero_ps();
1633             fjy0             = _mm256_setzero_ps();
1634             fjz0             = _mm256_setzero_ps();
1635             fjx1             = _mm256_setzero_ps();
1636             fjy1             = _mm256_setzero_ps();
1637             fjz1             = _mm256_setzero_ps();
1638             fjx2             = _mm256_setzero_ps();
1639             fjy2             = _mm256_setzero_ps();
1640             fjz2             = _mm256_setzero_ps();
1641
1642             /**************************
1643              * CALCULATE INTERACTIONS *
1644              **************************/
1645
1646             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1647             {
1648
1649             r00              = _mm256_mul_ps(rsq00,rinv00);
1650
1651             /* EWALD ELECTROSTATICS */
1652
1653             /* Analytical PME correction */
1654             zeta2            = _mm256_mul_ps(beta2,rsq00);
1655             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1656             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1657             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1658             felec            = _mm256_mul_ps(qq00,felec);
1659
1660             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1661
1662             fscal            = felec;
1663
1664             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1665
1666             /* Calculate temporary vectorial force */
1667             tx               = _mm256_mul_ps(fscal,dx00);
1668             ty               = _mm256_mul_ps(fscal,dy00);
1669             tz               = _mm256_mul_ps(fscal,dz00);
1670
1671             /* Update vectorial force */
1672             fix0             = _mm256_add_ps(fix0,tx);
1673             fiy0             = _mm256_add_ps(fiy0,ty);
1674             fiz0             = _mm256_add_ps(fiz0,tz);
1675
1676             fjx0             = _mm256_add_ps(fjx0,tx);
1677             fjy0             = _mm256_add_ps(fjy0,ty);
1678             fjz0             = _mm256_add_ps(fjz0,tz);
1679
1680             }
1681
1682             /**************************
1683              * CALCULATE INTERACTIONS *
1684              **************************/
1685
1686             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1687             {
1688
1689             r01              = _mm256_mul_ps(rsq01,rinv01);
1690
1691             /* EWALD ELECTROSTATICS */
1692
1693             /* Analytical PME correction */
1694             zeta2            = _mm256_mul_ps(beta2,rsq01);
1695             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
1696             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1697             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1698             felec            = _mm256_mul_ps(qq01,felec);
1699
1700             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
1701
1702             fscal            = felec;
1703
1704             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1705
1706             /* Calculate temporary vectorial force */
1707             tx               = _mm256_mul_ps(fscal,dx01);
1708             ty               = _mm256_mul_ps(fscal,dy01);
1709             tz               = _mm256_mul_ps(fscal,dz01);
1710
1711             /* Update vectorial force */
1712             fix0             = _mm256_add_ps(fix0,tx);
1713             fiy0             = _mm256_add_ps(fiy0,ty);
1714             fiz0             = _mm256_add_ps(fiz0,tz);
1715
1716             fjx1             = _mm256_add_ps(fjx1,tx);
1717             fjy1             = _mm256_add_ps(fjy1,ty);
1718             fjz1             = _mm256_add_ps(fjz1,tz);
1719
1720             }
1721
1722             /**************************
1723              * CALCULATE INTERACTIONS *
1724              **************************/
1725
1726             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1727             {
1728
1729             r02              = _mm256_mul_ps(rsq02,rinv02);
1730
1731             /* EWALD ELECTROSTATICS */
1732
1733             /* Analytical PME correction */
1734             zeta2            = _mm256_mul_ps(beta2,rsq02);
1735             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
1736             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1737             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1738             felec            = _mm256_mul_ps(qq02,felec);
1739
1740             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1741
1742             fscal            = felec;
1743
1744             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1745
1746             /* Calculate temporary vectorial force */
1747             tx               = _mm256_mul_ps(fscal,dx02);
1748             ty               = _mm256_mul_ps(fscal,dy02);
1749             tz               = _mm256_mul_ps(fscal,dz02);
1750
1751             /* Update vectorial force */
1752             fix0             = _mm256_add_ps(fix0,tx);
1753             fiy0             = _mm256_add_ps(fiy0,ty);
1754             fiz0             = _mm256_add_ps(fiz0,tz);
1755
1756             fjx2             = _mm256_add_ps(fjx2,tx);
1757             fjy2             = _mm256_add_ps(fjy2,ty);
1758             fjz2             = _mm256_add_ps(fjz2,tz);
1759
1760             }
1761
1762             /**************************
1763              * CALCULATE INTERACTIONS *
1764              **************************/
1765
1766             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1767             {
1768
1769             r10              = _mm256_mul_ps(rsq10,rinv10);
1770
1771             /* EWALD ELECTROSTATICS */
1772
1773             /* Analytical PME correction */
1774             zeta2            = _mm256_mul_ps(beta2,rsq10);
1775             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1776             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1777             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1778             felec            = _mm256_mul_ps(qq10,felec);
1779
1780             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1781
1782             fscal            = felec;
1783
1784             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1785
1786             /* Calculate temporary vectorial force */
1787             tx               = _mm256_mul_ps(fscal,dx10);
1788             ty               = _mm256_mul_ps(fscal,dy10);
1789             tz               = _mm256_mul_ps(fscal,dz10);
1790
1791             /* Update vectorial force */
1792             fix1             = _mm256_add_ps(fix1,tx);
1793             fiy1             = _mm256_add_ps(fiy1,ty);
1794             fiz1             = _mm256_add_ps(fiz1,tz);
1795
1796             fjx0             = _mm256_add_ps(fjx0,tx);
1797             fjy0             = _mm256_add_ps(fjy0,ty);
1798             fjz0             = _mm256_add_ps(fjz0,tz);
1799
1800             }
1801
1802             /**************************
1803              * CALCULATE INTERACTIONS *
1804              **************************/
1805
1806             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1807             {
1808
1809             r11              = _mm256_mul_ps(rsq11,rinv11);
1810
1811             /* EWALD ELECTROSTATICS */
1812
1813             /* Analytical PME correction */
1814             zeta2            = _mm256_mul_ps(beta2,rsq11);
1815             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1816             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1817             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1818             felec            = _mm256_mul_ps(qq11,felec);
1819
1820             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1821
1822             fscal            = felec;
1823
1824             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1825
1826             /* Calculate temporary vectorial force */
1827             tx               = _mm256_mul_ps(fscal,dx11);
1828             ty               = _mm256_mul_ps(fscal,dy11);
1829             tz               = _mm256_mul_ps(fscal,dz11);
1830
1831             /* Update vectorial force */
1832             fix1             = _mm256_add_ps(fix1,tx);
1833             fiy1             = _mm256_add_ps(fiy1,ty);
1834             fiz1             = _mm256_add_ps(fiz1,tz);
1835
1836             fjx1             = _mm256_add_ps(fjx1,tx);
1837             fjy1             = _mm256_add_ps(fjy1,ty);
1838             fjz1             = _mm256_add_ps(fjz1,tz);
1839
1840             }
1841
1842             /**************************
1843              * CALCULATE INTERACTIONS *
1844              **************************/
1845
1846             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1847             {
1848
1849             r12              = _mm256_mul_ps(rsq12,rinv12);
1850
1851             /* EWALD ELECTROSTATICS */
1852
1853             /* Analytical PME correction */
1854             zeta2            = _mm256_mul_ps(beta2,rsq12);
1855             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1856             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1857             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1858             felec            = _mm256_mul_ps(qq12,felec);
1859
1860             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1861
1862             fscal            = felec;
1863
1864             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1865
1866             /* Calculate temporary vectorial force */
1867             tx               = _mm256_mul_ps(fscal,dx12);
1868             ty               = _mm256_mul_ps(fscal,dy12);
1869             tz               = _mm256_mul_ps(fscal,dz12);
1870
1871             /* Update vectorial force */
1872             fix1             = _mm256_add_ps(fix1,tx);
1873             fiy1             = _mm256_add_ps(fiy1,ty);
1874             fiz1             = _mm256_add_ps(fiz1,tz);
1875
1876             fjx2             = _mm256_add_ps(fjx2,tx);
1877             fjy2             = _mm256_add_ps(fjy2,ty);
1878             fjz2             = _mm256_add_ps(fjz2,tz);
1879
1880             }
1881
1882             /**************************
1883              * CALCULATE INTERACTIONS *
1884              **************************/
1885
1886             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1887             {
1888
1889             r20              = _mm256_mul_ps(rsq20,rinv20);
1890
1891             /* EWALD ELECTROSTATICS */
1892
1893             /* Analytical PME correction */
1894             zeta2            = _mm256_mul_ps(beta2,rsq20);
1895             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1896             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1897             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1898             felec            = _mm256_mul_ps(qq20,felec);
1899
1900             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1901
1902             fscal            = felec;
1903
1904             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1905
1906             /* Calculate temporary vectorial force */
1907             tx               = _mm256_mul_ps(fscal,dx20);
1908             ty               = _mm256_mul_ps(fscal,dy20);
1909             tz               = _mm256_mul_ps(fscal,dz20);
1910
1911             /* Update vectorial force */
1912             fix2             = _mm256_add_ps(fix2,tx);
1913             fiy2             = _mm256_add_ps(fiy2,ty);
1914             fiz2             = _mm256_add_ps(fiz2,tz);
1915
1916             fjx0             = _mm256_add_ps(fjx0,tx);
1917             fjy0             = _mm256_add_ps(fjy0,ty);
1918             fjz0             = _mm256_add_ps(fjz0,tz);
1919
1920             }
1921
1922             /**************************
1923              * CALCULATE INTERACTIONS *
1924              **************************/
1925
1926             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1927             {
1928
1929             r21              = _mm256_mul_ps(rsq21,rinv21);
1930
1931             /* EWALD ELECTROSTATICS */
1932
1933             /* Analytical PME correction */
1934             zeta2            = _mm256_mul_ps(beta2,rsq21);
1935             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1936             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1937             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1938             felec            = _mm256_mul_ps(qq21,felec);
1939
1940             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1941
1942             fscal            = felec;
1943
1944             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1945
1946             /* Calculate temporary vectorial force */
1947             tx               = _mm256_mul_ps(fscal,dx21);
1948             ty               = _mm256_mul_ps(fscal,dy21);
1949             tz               = _mm256_mul_ps(fscal,dz21);
1950
1951             /* Update vectorial force */
1952             fix2             = _mm256_add_ps(fix2,tx);
1953             fiy2             = _mm256_add_ps(fiy2,ty);
1954             fiz2             = _mm256_add_ps(fiz2,tz);
1955
1956             fjx1             = _mm256_add_ps(fjx1,tx);
1957             fjy1             = _mm256_add_ps(fjy1,ty);
1958             fjz1             = _mm256_add_ps(fjz1,tz);
1959
1960             }
1961
1962             /**************************
1963              * CALCULATE INTERACTIONS *
1964              **************************/
1965
1966             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1967             {
1968
1969             r22              = _mm256_mul_ps(rsq22,rinv22);
1970
1971             /* EWALD ELECTROSTATICS */
1972
1973             /* Analytical PME correction */
1974             zeta2            = _mm256_mul_ps(beta2,rsq22);
1975             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1976             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1977             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1978             felec            = _mm256_mul_ps(qq22,felec);
1979
1980             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1981
1982             fscal            = felec;
1983
1984             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1985
1986             /* Calculate temporary vectorial force */
1987             tx               = _mm256_mul_ps(fscal,dx22);
1988             ty               = _mm256_mul_ps(fscal,dy22);
1989             tz               = _mm256_mul_ps(fscal,dz22);
1990
1991             /* Update vectorial force */
1992             fix2             = _mm256_add_ps(fix2,tx);
1993             fiy2             = _mm256_add_ps(fiy2,ty);
1994             fiz2             = _mm256_add_ps(fiz2,tz);
1995
1996             fjx2             = _mm256_add_ps(fjx2,tx);
1997             fjy2             = _mm256_add_ps(fjy2,ty);
1998             fjz2             = _mm256_add_ps(fjz2,tz);
1999
2000             }
2001
2002             fjptrA             = f+j_coord_offsetA;
2003             fjptrB             = f+j_coord_offsetB;
2004             fjptrC             = f+j_coord_offsetC;
2005             fjptrD             = f+j_coord_offsetD;
2006             fjptrE             = f+j_coord_offsetE;
2007             fjptrF             = f+j_coord_offsetF;
2008             fjptrG             = f+j_coord_offsetG;
2009             fjptrH             = f+j_coord_offsetH;
2010
2011             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2012                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2013
2014             /* Inner loop uses 531 flops */
2015         }
2016
2017         if(jidx<j_index_end)
2018         {
2019
2020             /* Get j neighbor index, and coordinate index */
2021             jnrlistA         = jjnr[jidx];
2022             jnrlistB         = jjnr[jidx+1];
2023             jnrlistC         = jjnr[jidx+2];
2024             jnrlistD         = jjnr[jidx+3];
2025             jnrlistE         = jjnr[jidx+4];
2026             jnrlistF         = jjnr[jidx+5];
2027             jnrlistG         = jjnr[jidx+6];
2028             jnrlistH         = jjnr[jidx+7];
2029             /* Sign of each element will be negative for non-real atoms.
2030              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2031              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2032              */
2033             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2034                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2035
2036             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2037             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2038             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2039             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2040             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2041             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2042             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2043             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2044             j_coord_offsetA  = DIM*jnrA;
2045             j_coord_offsetB  = DIM*jnrB;
2046             j_coord_offsetC  = DIM*jnrC;
2047             j_coord_offsetD  = DIM*jnrD;
2048             j_coord_offsetE  = DIM*jnrE;
2049             j_coord_offsetF  = DIM*jnrF;
2050             j_coord_offsetG  = DIM*jnrG;
2051             j_coord_offsetH  = DIM*jnrH;
2052
2053             /* load j atom coordinates */
2054             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2055                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2056                                                  x+j_coord_offsetE,x+j_coord_offsetF,
2057                                                  x+j_coord_offsetG,x+j_coord_offsetH,
2058                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2059
2060             /* Calculate displacement vector */
2061             dx00             = _mm256_sub_ps(ix0,jx0);
2062             dy00             = _mm256_sub_ps(iy0,jy0);
2063             dz00             = _mm256_sub_ps(iz0,jz0);
2064             dx01             = _mm256_sub_ps(ix0,jx1);
2065             dy01             = _mm256_sub_ps(iy0,jy1);
2066             dz01             = _mm256_sub_ps(iz0,jz1);
2067             dx02             = _mm256_sub_ps(ix0,jx2);
2068             dy02             = _mm256_sub_ps(iy0,jy2);
2069             dz02             = _mm256_sub_ps(iz0,jz2);
2070             dx10             = _mm256_sub_ps(ix1,jx0);
2071             dy10             = _mm256_sub_ps(iy1,jy0);
2072             dz10             = _mm256_sub_ps(iz1,jz0);
2073             dx11             = _mm256_sub_ps(ix1,jx1);
2074             dy11             = _mm256_sub_ps(iy1,jy1);
2075             dz11             = _mm256_sub_ps(iz1,jz1);
2076             dx12             = _mm256_sub_ps(ix1,jx2);
2077             dy12             = _mm256_sub_ps(iy1,jy2);
2078             dz12             = _mm256_sub_ps(iz1,jz2);
2079             dx20             = _mm256_sub_ps(ix2,jx0);
2080             dy20             = _mm256_sub_ps(iy2,jy0);
2081             dz20             = _mm256_sub_ps(iz2,jz0);
2082             dx21             = _mm256_sub_ps(ix2,jx1);
2083             dy21             = _mm256_sub_ps(iy2,jy1);
2084             dz21             = _mm256_sub_ps(iz2,jz1);
2085             dx22             = _mm256_sub_ps(ix2,jx2);
2086             dy22             = _mm256_sub_ps(iy2,jy2);
2087             dz22             = _mm256_sub_ps(iz2,jz2);
2088
2089             /* Calculate squared distance and things based on it */
2090             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2091             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
2092             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
2093             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
2094             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2095             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2096             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
2097             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2098             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2099
2100             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
2101             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
2102             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
2103             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
2104             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2105             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2106             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
2107             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2108             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2109
2110             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
2111             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
2112             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
2113             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
2114             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2115             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2116             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
2117             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2118             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2119
2120             fjx0             = _mm256_setzero_ps();
2121             fjy0             = _mm256_setzero_ps();
2122             fjz0             = _mm256_setzero_ps();
2123             fjx1             = _mm256_setzero_ps();
2124             fjy1             = _mm256_setzero_ps();
2125             fjz1             = _mm256_setzero_ps();
2126             fjx2             = _mm256_setzero_ps();
2127             fjy2             = _mm256_setzero_ps();
2128             fjz2             = _mm256_setzero_ps();
2129
2130             /**************************
2131              * CALCULATE INTERACTIONS *
2132              **************************/
2133
2134             if (gmx_mm256_any_lt(rsq00,rcutoff2))
2135             {
2136
2137             r00              = _mm256_mul_ps(rsq00,rinv00);
2138             r00              = _mm256_andnot_ps(dummy_mask,r00);
2139
2140             /* EWALD ELECTROSTATICS */
2141
2142             /* Analytical PME correction */
2143             zeta2            = _mm256_mul_ps(beta2,rsq00);
2144             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
2145             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2146             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2147             felec            = _mm256_mul_ps(qq00,felec);
2148
2149             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
2150
2151             fscal            = felec;
2152
2153             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2154
2155             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2156
2157             /* Calculate temporary vectorial force */
2158             tx               = _mm256_mul_ps(fscal,dx00);
2159             ty               = _mm256_mul_ps(fscal,dy00);
2160             tz               = _mm256_mul_ps(fscal,dz00);
2161
2162             /* Update vectorial force */
2163             fix0             = _mm256_add_ps(fix0,tx);
2164             fiy0             = _mm256_add_ps(fiy0,ty);
2165             fiz0             = _mm256_add_ps(fiz0,tz);
2166
2167             fjx0             = _mm256_add_ps(fjx0,tx);
2168             fjy0             = _mm256_add_ps(fjy0,ty);
2169             fjz0             = _mm256_add_ps(fjz0,tz);
2170
2171             }
2172
2173             /**************************
2174              * CALCULATE INTERACTIONS *
2175              **************************/
2176
2177             if (gmx_mm256_any_lt(rsq01,rcutoff2))
2178             {
2179
2180             r01              = _mm256_mul_ps(rsq01,rinv01);
2181             r01              = _mm256_andnot_ps(dummy_mask,r01);
2182
2183             /* EWALD ELECTROSTATICS */
2184
2185             /* Analytical PME correction */
2186             zeta2            = _mm256_mul_ps(beta2,rsq01);
2187             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
2188             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2189             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2190             felec            = _mm256_mul_ps(qq01,felec);
2191
2192             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
2193
2194             fscal            = felec;
2195
2196             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2197
2198             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2199
2200             /* Calculate temporary vectorial force */
2201             tx               = _mm256_mul_ps(fscal,dx01);
2202             ty               = _mm256_mul_ps(fscal,dy01);
2203             tz               = _mm256_mul_ps(fscal,dz01);
2204
2205             /* Update vectorial force */
2206             fix0             = _mm256_add_ps(fix0,tx);
2207             fiy0             = _mm256_add_ps(fiy0,ty);
2208             fiz0             = _mm256_add_ps(fiz0,tz);
2209
2210             fjx1             = _mm256_add_ps(fjx1,tx);
2211             fjy1             = _mm256_add_ps(fjy1,ty);
2212             fjz1             = _mm256_add_ps(fjz1,tz);
2213
2214             }
2215
2216             /**************************
2217              * CALCULATE INTERACTIONS *
2218              **************************/
2219
2220             if (gmx_mm256_any_lt(rsq02,rcutoff2))
2221             {
2222
2223             r02              = _mm256_mul_ps(rsq02,rinv02);
2224             r02              = _mm256_andnot_ps(dummy_mask,r02);
2225
2226             /* EWALD ELECTROSTATICS */
2227
2228             /* Analytical PME correction */
2229             zeta2            = _mm256_mul_ps(beta2,rsq02);
2230             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
2231             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2232             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2233             felec            = _mm256_mul_ps(qq02,felec);
2234
2235             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
2236
2237             fscal            = felec;
2238
2239             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2240
2241             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2242
2243             /* Calculate temporary vectorial force */
2244             tx               = _mm256_mul_ps(fscal,dx02);
2245             ty               = _mm256_mul_ps(fscal,dy02);
2246             tz               = _mm256_mul_ps(fscal,dz02);
2247
2248             /* Update vectorial force */
2249             fix0             = _mm256_add_ps(fix0,tx);
2250             fiy0             = _mm256_add_ps(fiy0,ty);
2251             fiz0             = _mm256_add_ps(fiz0,tz);
2252
2253             fjx2             = _mm256_add_ps(fjx2,tx);
2254             fjy2             = _mm256_add_ps(fjy2,ty);
2255             fjz2             = _mm256_add_ps(fjz2,tz);
2256
2257             }
2258
2259             /**************************
2260              * CALCULATE INTERACTIONS *
2261              **************************/
2262
2263             if (gmx_mm256_any_lt(rsq10,rcutoff2))
2264             {
2265
2266             r10              = _mm256_mul_ps(rsq10,rinv10);
2267             r10              = _mm256_andnot_ps(dummy_mask,r10);
2268
2269             /* EWALD ELECTROSTATICS */
2270
2271             /* Analytical PME correction */
2272             zeta2            = _mm256_mul_ps(beta2,rsq10);
2273             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
2274             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2275             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2276             felec            = _mm256_mul_ps(qq10,felec);
2277
2278             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
2279
2280             fscal            = felec;
2281
2282             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2283
2284             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2285
2286             /* Calculate temporary vectorial force */
2287             tx               = _mm256_mul_ps(fscal,dx10);
2288             ty               = _mm256_mul_ps(fscal,dy10);
2289             tz               = _mm256_mul_ps(fscal,dz10);
2290
2291             /* Update vectorial force */
2292             fix1             = _mm256_add_ps(fix1,tx);
2293             fiy1             = _mm256_add_ps(fiy1,ty);
2294             fiz1             = _mm256_add_ps(fiz1,tz);
2295
2296             fjx0             = _mm256_add_ps(fjx0,tx);
2297             fjy0             = _mm256_add_ps(fjy0,ty);
2298             fjz0             = _mm256_add_ps(fjz0,tz);
2299
2300             }
2301
2302             /**************************
2303              * CALCULATE INTERACTIONS *
2304              **************************/
2305
2306             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2307             {
2308
2309             r11              = _mm256_mul_ps(rsq11,rinv11);
2310             r11              = _mm256_andnot_ps(dummy_mask,r11);
2311
2312             /* EWALD ELECTROSTATICS */
2313
2314             /* Analytical PME correction */
2315             zeta2            = _mm256_mul_ps(beta2,rsq11);
2316             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
2317             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2318             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2319             felec            = _mm256_mul_ps(qq11,felec);
2320
2321             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2322
2323             fscal            = felec;
2324
2325             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2326
2327             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2328
2329             /* Calculate temporary vectorial force */
2330             tx               = _mm256_mul_ps(fscal,dx11);
2331             ty               = _mm256_mul_ps(fscal,dy11);
2332             tz               = _mm256_mul_ps(fscal,dz11);
2333
2334             /* Update vectorial force */
2335             fix1             = _mm256_add_ps(fix1,tx);
2336             fiy1             = _mm256_add_ps(fiy1,ty);
2337             fiz1             = _mm256_add_ps(fiz1,tz);
2338
2339             fjx1             = _mm256_add_ps(fjx1,tx);
2340             fjy1             = _mm256_add_ps(fjy1,ty);
2341             fjz1             = _mm256_add_ps(fjz1,tz);
2342
2343             }
2344
2345             /**************************
2346              * CALCULATE INTERACTIONS *
2347              **************************/
2348
2349             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2350             {
2351
2352             r12              = _mm256_mul_ps(rsq12,rinv12);
2353             r12              = _mm256_andnot_ps(dummy_mask,r12);
2354
2355             /* EWALD ELECTROSTATICS */
2356
2357             /* Analytical PME correction */
2358             zeta2            = _mm256_mul_ps(beta2,rsq12);
2359             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2360             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2361             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2362             felec            = _mm256_mul_ps(qq12,felec);
2363
2364             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2365
2366             fscal            = felec;
2367
2368             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2369
2370             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2371
2372             /* Calculate temporary vectorial force */
2373             tx               = _mm256_mul_ps(fscal,dx12);
2374             ty               = _mm256_mul_ps(fscal,dy12);
2375             tz               = _mm256_mul_ps(fscal,dz12);
2376
2377             /* Update vectorial force */
2378             fix1             = _mm256_add_ps(fix1,tx);
2379             fiy1             = _mm256_add_ps(fiy1,ty);
2380             fiz1             = _mm256_add_ps(fiz1,tz);
2381
2382             fjx2             = _mm256_add_ps(fjx2,tx);
2383             fjy2             = _mm256_add_ps(fjy2,ty);
2384             fjz2             = _mm256_add_ps(fjz2,tz);
2385
2386             }
2387
2388             /**************************
2389              * CALCULATE INTERACTIONS *
2390              **************************/
2391
2392             if (gmx_mm256_any_lt(rsq20,rcutoff2))
2393             {
2394
2395             r20              = _mm256_mul_ps(rsq20,rinv20);
2396             r20              = _mm256_andnot_ps(dummy_mask,r20);
2397
2398             /* EWALD ELECTROSTATICS */
2399
2400             /* Analytical PME correction */
2401             zeta2            = _mm256_mul_ps(beta2,rsq20);
2402             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
2403             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2404             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2405             felec            = _mm256_mul_ps(qq20,felec);
2406
2407             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
2408
2409             fscal            = felec;
2410
2411             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2412
2413             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2414
2415             /* Calculate temporary vectorial force */
2416             tx               = _mm256_mul_ps(fscal,dx20);
2417             ty               = _mm256_mul_ps(fscal,dy20);
2418             tz               = _mm256_mul_ps(fscal,dz20);
2419
2420             /* Update vectorial force */
2421             fix2             = _mm256_add_ps(fix2,tx);
2422             fiy2             = _mm256_add_ps(fiy2,ty);
2423             fiz2             = _mm256_add_ps(fiz2,tz);
2424
2425             fjx0             = _mm256_add_ps(fjx0,tx);
2426             fjy0             = _mm256_add_ps(fjy0,ty);
2427             fjz0             = _mm256_add_ps(fjz0,tz);
2428
2429             }
2430
2431             /**************************
2432              * CALCULATE INTERACTIONS *
2433              **************************/
2434
2435             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2436             {
2437
2438             r21              = _mm256_mul_ps(rsq21,rinv21);
2439             r21              = _mm256_andnot_ps(dummy_mask,r21);
2440
2441             /* EWALD ELECTROSTATICS */
2442
2443             /* Analytical PME correction */
2444             zeta2            = _mm256_mul_ps(beta2,rsq21);
2445             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2446             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2447             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2448             felec            = _mm256_mul_ps(qq21,felec);
2449
2450             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2451
2452             fscal            = felec;
2453
2454             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2455
2456             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2457
2458             /* Calculate temporary vectorial force */
2459             tx               = _mm256_mul_ps(fscal,dx21);
2460             ty               = _mm256_mul_ps(fscal,dy21);
2461             tz               = _mm256_mul_ps(fscal,dz21);
2462
2463             /* Update vectorial force */
2464             fix2             = _mm256_add_ps(fix2,tx);
2465             fiy2             = _mm256_add_ps(fiy2,ty);
2466             fiz2             = _mm256_add_ps(fiz2,tz);
2467
2468             fjx1             = _mm256_add_ps(fjx1,tx);
2469             fjy1             = _mm256_add_ps(fjy1,ty);
2470             fjz1             = _mm256_add_ps(fjz1,tz);
2471
2472             }
2473
2474             /**************************
2475              * CALCULATE INTERACTIONS *
2476              **************************/
2477
2478             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2479             {
2480
2481             r22              = _mm256_mul_ps(rsq22,rinv22);
2482             r22              = _mm256_andnot_ps(dummy_mask,r22);
2483
2484             /* EWALD ELECTROSTATICS */
2485
2486             /* Analytical PME correction */
2487             zeta2            = _mm256_mul_ps(beta2,rsq22);
2488             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2489             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2490             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2491             felec            = _mm256_mul_ps(qq22,felec);
2492
2493             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2494
2495             fscal            = felec;
2496
2497             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2498
2499             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2500
2501             /* Calculate temporary vectorial force */
2502             tx               = _mm256_mul_ps(fscal,dx22);
2503             ty               = _mm256_mul_ps(fscal,dy22);
2504             tz               = _mm256_mul_ps(fscal,dz22);
2505
2506             /* Update vectorial force */
2507             fix2             = _mm256_add_ps(fix2,tx);
2508             fiy2             = _mm256_add_ps(fiy2,ty);
2509             fiz2             = _mm256_add_ps(fiz2,tz);
2510
2511             fjx2             = _mm256_add_ps(fjx2,tx);
2512             fjy2             = _mm256_add_ps(fjy2,ty);
2513             fjz2             = _mm256_add_ps(fjz2,tz);
2514
2515             }
2516
2517             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2518             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2519             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2520             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2521             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2522             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2523             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2524             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2525
2526             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2527                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2528
2529             /* Inner loop uses 540 flops */
2530         }
2531
2532         /* End of innermost loop */
2533
2534         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2535                                                  f+i_coord_offset,fshift+i_shift_offset);
2536
2537         /* Increment number of inner iterations */
2538         inneriter                  += j_index_end - j_index_start;
2539
2540         /* Outer loop uses 18 flops */
2541     }
2542
2543     /* Increment number of outer iterations */
2544     outeriter        += nri;
2545
2546     /* Update outer/inner flops */
2547
2548     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*540);
2549 }