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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3W3_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwNone_GeomW3W3_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  95     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  97     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 100     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 101     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 102     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 105     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     __m256i          ewitab;
 110     __m128i          ewitab_lo,ewitab_hi;
 111     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 112     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 113     real             *ewtab;
 114     __m256           dummy_mask,cutoff_mask;
 115     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 116     __m256           one     = _mm256_set1_ps(1.0);
 117     __m256           two     = _mm256_set1_ps(2.0);
 118     x                = xx[0];
 119     f                = ff[0];
 120
 121     nri              = nlist->nri;
 122     iinr             = nlist->iinr;
 123     jindex           = nlist->jindex;
 124     jjnr             = nlist->jjnr;
 125     shiftidx         = nlist->shift;
 126     gid              = nlist->gid;
 127     shiftvec         = fr->shift_vec[0];
 128     fshift           = fr->fshift[0];
 129     facel            = _mm256_set1_ps(fr->epsfac);
 130     charge           = mdatoms->chargeA;
 131
 132     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 133     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 134     beta2            = _mm256_mul_ps(beta,beta);
 135     beta3            = _mm256_mul_ps(beta,beta2);
 136
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 144     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 145     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 146
 147     jq0              = _mm256_set1_ps(charge[inr+0]);
 148     jq1              = _mm256_set1_ps(charge[inr+1]);
 149     jq2              = _mm256_set1_ps(charge[inr+2]);
 150     qq00             = _mm256_mul_ps(iq0,jq0);
 151     qq01             = _mm256_mul_ps(iq0,jq1);
 152     qq02             = _mm256_mul_ps(iq0,jq2);
 153     qq10             = _mm256_mul_ps(iq1,jq0);
 154     qq11             = _mm256_mul_ps(iq1,jq1);
 155     qq12             = _mm256_mul_ps(iq1,jq2);
 156     qq20             = _mm256_mul_ps(iq2,jq0);
 157     qq21             = _mm256_mul_ps(iq2,jq1);
 158     qq22             = _mm256_mul_ps(iq2,jq2);
 159
 160     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 161     rcutoff_scalar   = fr->rcoulomb;
 162     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 163     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 164
 165     /* Avoid stupid compiler warnings */
 166     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 167     j_coord_offsetA = 0;
 168     j_coord_offsetB = 0;
 169     j_coord_offsetC = 0;
 170     j_coord_offsetD = 0;
 171     j_coord_offsetE = 0;
 172     j_coord_offsetF = 0;
 173     j_coord_offsetG = 0;
 174     j_coord_offsetH = 0;
 175
 176     outeriter        = 0;
 177     inneriter        = 0;
 178
 179     for(iidx=0;iidx<4*DIM;iidx++)
 180     {
 181         scratch[iidx] = 0.0;
 182     }
 183
 184     /* Start outer loop over neighborlists */
 185     for(iidx=0; iidx<nri; iidx++)
 186     {
 187         /* Load shift vector for this list */
 188         i_shift_offset   = DIM*shiftidx[iidx];
 189
 190         /* Load limits for loop over neighbors */
 191         j_index_start    = jindex[iidx];
 192         j_index_end      = jindex[iidx+1];
 193
 194         /* Get outer coordinate index */
 195         inr              = iinr[iidx];
 196         i_coord_offset   = DIM*inr;
 197
 198         /* Load i particle coords and add shift vector */
 199         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 200                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 201
 202         fix0             = _mm256_setzero_ps();
 203         fiy0             = _mm256_setzero_ps();
 204         fiz0             = _mm256_setzero_ps();
 205         fix1             = _mm256_setzero_ps();
 206         fiy1             = _mm256_setzero_ps();
 207         fiz1             = _mm256_setzero_ps();
 208         fix2             = _mm256_setzero_ps();
 209         fiy2             = _mm256_setzero_ps();
 210         fiz2             = _mm256_setzero_ps();
 211
 212         /* Reset potential sums */
 213         velecsum         = _mm256_setzero_ps();
 214
 215         /* Start inner kernel loop */
 216         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 217         {
 218
 219             /* Get j neighbor index, and coordinate index */
 220             jnrA             = jjnr[jidx];
 221             jnrB             = jjnr[jidx+1];
 222             jnrC             = jjnr[jidx+2];
 223             jnrD             = jjnr[jidx+3];
 224             jnrE             = jjnr[jidx+4];
 225             jnrF             = jjnr[jidx+5];
 226             jnrG             = jjnr[jidx+6];
 227             jnrH             = jjnr[jidx+7];
 228             j_coord_offsetA  = DIM*jnrA;
 229             j_coord_offsetB  = DIM*jnrB;
 230             j_coord_offsetC  = DIM*jnrC;
 231             j_coord_offsetD  = DIM*jnrD;
 232             j_coord_offsetE  = DIM*jnrE;
 233             j_coord_offsetF  = DIM*jnrF;
 234             j_coord_offsetG  = DIM*jnrG;
 235             j_coord_offsetH  = DIM*jnrH;
 236
 237             /* load j atom coordinates */
 238             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 239                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 240                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 241                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 242                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 243
 244             /* Calculate displacement vector */
 245             dx00             = _mm256_sub_ps(ix0,jx0);
 246             dy00             = _mm256_sub_ps(iy0,jy0);
 247             dz00             = _mm256_sub_ps(iz0,jz0);
 248             dx01             = _mm256_sub_ps(ix0,jx1);
 249             dy01             = _mm256_sub_ps(iy0,jy1);
 250             dz01             = _mm256_sub_ps(iz0,jz1);
 251             dx02             = _mm256_sub_ps(ix0,jx2);
 252             dy02             = _mm256_sub_ps(iy0,jy2);
 253             dz02             = _mm256_sub_ps(iz0,jz2);
 254             dx10             = _mm256_sub_ps(ix1,jx0);
 255             dy10             = _mm256_sub_ps(iy1,jy0);
 256             dz10             = _mm256_sub_ps(iz1,jz0);
 257             dx11             = _mm256_sub_ps(ix1,jx1);
 258             dy11             = _mm256_sub_ps(iy1,jy1);
 259             dz11             = _mm256_sub_ps(iz1,jz1);
 260             dx12             = _mm256_sub_ps(ix1,jx2);
 261             dy12             = _mm256_sub_ps(iy1,jy2);
 262             dz12             = _mm256_sub_ps(iz1,jz2);
 263             dx20             = _mm256_sub_ps(ix2,jx0);
 264             dy20             = _mm256_sub_ps(iy2,jy0);
 265             dz20             = _mm256_sub_ps(iz2,jz0);
 266             dx21             = _mm256_sub_ps(ix2,jx1);
 267             dy21             = _mm256_sub_ps(iy2,jy1);
 268             dz21             = _mm256_sub_ps(iz2,jz1);
 269             dx22             = _mm256_sub_ps(ix2,jx2);
 270             dy22             = _mm256_sub_ps(iy2,jy2);
 271             dz22             = _mm256_sub_ps(iz2,jz2);
 272
 273             /* Calculate squared distance and things based on it */
 274             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 275             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 276             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 277             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 278             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 279             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 280             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 281             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 282             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 283
 284             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 285             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 286             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 287             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 288             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 289             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 290             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 291             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 292             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 293
 294             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 295             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 296             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 297             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 298             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 299             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 300             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 301             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 302             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 303
 304             fjx0             = _mm256_setzero_ps();
 305             fjy0             = _mm256_setzero_ps();
 306             fjz0             = _mm256_setzero_ps();
 307             fjx1             = _mm256_setzero_ps();
 308             fjy1             = _mm256_setzero_ps();
 309             fjz1             = _mm256_setzero_ps();
 310             fjx2             = _mm256_setzero_ps();
 311             fjy2             = _mm256_setzero_ps();
 312             fjz2             = _mm256_setzero_ps();
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 319             {
 320
 321             r00              = _mm256_mul_ps(rsq00,rinv00);
 322
 323             /* EWALD ELECTROSTATICS */
 324
 325             /* Analytical PME correction */
 326             zeta2            = _mm256_mul_ps(beta2,rsq00);
 327             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 328             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 329             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 330             felec            = _mm256_mul_ps(qq00,felec);
 331             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 332             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 333             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 334             velec            = _mm256_mul_ps(qq00,velec);
 335
 336             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm256_and_ps(velec,cutoff_mask);
 340             velecsum         = _mm256_add_ps(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm256_mul_ps(fscal,dx00);
 348             ty               = _mm256_mul_ps(fscal,dy00);
 349             tz               = _mm256_mul_ps(fscal,dz00);
 350
 351             /* Update vectorial force */
 352             fix0             = _mm256_add_ps(fix0,tx);
 353             fiy0             = _mm256_add_ps(fiy0,ty);
 354             fiz0             = _mm256_add_ps(fiz0,tz);
 355
 356             fjx0             = _mm256_add_ps(fjx0,tx);
 357             fjy0             = _mm256_add_ps(fjy0,ty);
 358             fjz0             = _mm256_add_ps(fjz0,tz);
 359
 360             }
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 367             {
 368
 369             r01              = _mm256_mul_ps(rsq01,rinv01);
 370
 371             /* EWALD ELECTROSTATICS */
 372
 373             /* Analytical PME correction */
 374             zeta2            = _mm256_mul_ps(beta2,rsq01);
 375             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 376             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 377             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 378             felec            = _mm256_mul_ps(qq01,felec);
 379             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 380             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 381             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv01,sh_ewald),pmecorrV);
 382             velec            = _mm256_mul_ps(qq01,velec);
 383
 384             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velec            = _mm256_and_ps(velec,cutoff_mask);
 388             velecsum         = _mm256_add_ps(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 393
 394             /* Calculate temporary vectorial force */
 395             tx               = _mm256_mul_ps(fscal,dx01);
 396             ty               = _mm256_mul_ps(fscal,dy01);
 397             tz               = _mm256_mul_ps(fscal,dz01);
 398
 399             /* Update vectorial force */
 400             fix0             = _mm256_add_ps(fix0,tx);
 401             fiy0             = _mm256_add_ps(fiy0,ty);
 402             fiz0             = _mm256_add_ps(fiz0,tz);
 403
 404             fjx1             = _mm256_add_ps(fjx1,tx);
 405             fjy1             = _mm256_add_ps(fjy1,ty);
 406             fjz1             = _mm256_add_ps(fjz1,tz);
 407
 408             }
 409
 410             /**************************
 411              * CALCULATE INTERACTIONS *
 412              **************************/
 413
 414             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 415             {
 416
 417             r02              = _mm256_mul_ps(rsq02,rinv02);
 418
 419             /* EWALD ELECTROSTATICS */
 420
 421             /* Analytical PME correction */
 422             zeta2            = _mm256_mul_ps(beta2,rsq02);
 423             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 424             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 425             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 426             felec            = _mm256_mul_ps(qq02,felec);
 427             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 428             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 429             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv02,sh_ewald),pmecorrV);
 430             velec            = _mm256_mul_ps(qq02,velec);
 431
 432             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
 433
 434             /* Update potential sum for this i atom from the interaction with this j atom. */
 435             velec            = _mm256_and_ps(velec,cutoff_mask);
 436             velecsum         = _mm256_add_ps(velecsum,velec);
 437
 438             fscal            = felec;
 439
 440             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 441
 442             /* Calculate temporary vectorial force */
 443             tx               = _mm256_mul_ps(fscal,dx02);
 444             ty               = _mm256_mul_ps(fscal,dy02);
 445             tz               = _mm256_mul_ps(fscal,dz02);
 446
 447             /* Update vectorial force */
 448             fix0             = _mm256_add_ps(fix0,tx);
 449             fiy0             = _mm256_add_ps(fiy0,ty);
 450             fiz0             = _mm256_add_ps(fiz0,tz);
 451
 452             fjx2             = _mm256_add_ps(fjx2,tx);
 453             fjy2             = _mm256_add_ps(fjy2,ty);
 454             fjz2             = _mm256_add_ps(fjz2,tz);
 455
 456             }
 457
 458             /**************************
 459              * CALCULATE INTERACTIONS *
 460              **************************/
 461
 462             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 463             {
 464
 465             r10              = _mm256_mul_ps(rsq10,rinv10);
 466
 467             /* EWALD ELECTROSTATICS */
 468
 469             /* Analytical PME correction */
 470             zeta2            = _mm256_mul_ps(beta2,rsq10);
 471             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 472             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 473             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 474             felec            = _mm256_mul_ps(qq10,felec);
 475             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 476             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 477             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 478             velec            = _mm256_mul_ps(qq10,velec);
 479
 480             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velec            = _mm256_and_ps(velec,cutoff_mask);
 484             velecsum         = _mm256_add_ps(velecsum,velec);
 485
 486             fscal            = felec;
 487
 488             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 489
 490             /* Calculate temporary vectorial force */
 491             tx               = _mm256_mul_ps(fscal,dx10);
 492             ty               = _mm256_mul_ps(fscal,dy10);
 493             tz               = _mm256_mul_ps(fscal,dz10);
 494
 495             /* Update vectorial force */
 496             fix1             = _mm256_add_ps(fix1,tx);
 497             fiy1             = _mm256_add_ps(fiy1,ty);
 498             fiz1             = _mm256_add_ps(fiz1,tz);
 499
 500             fjx0             = _mm256_add_ps(fjx0,tx);
 501             fjy0             = _mm256_add_ps(fjy0,ty);
 502             fjz0             = _mm256_add_ps(fjz0,tz);
 503
 504             }
 505
 506             /**************************
 507              * CALCULATE INTERACTIONS *
 508              **************************/
 509
 510             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 511             {
 512
 513             r11              = _mm256_mul_ps(rsq11,rinv11);
 514
 515             /* EWALD ELECTROSTATICS */
 516
 517             /* Analytical PME correction */
 518             zeta2            = _mm256_mul_ps(beta2,rsq11);
 519             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 520             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 521             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 522             felec            = _mm256_mul_ps(qq11,felec);
 523             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 524             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 525             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
 526             velec            = _mm256_mul_ps(qq11,velec);
 527
 528             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             velec            = _mm256_and_ps(velec,cutoff_mask);
 532             velecsum         = _mm256_add_ps(velecsum,velec);
 533
 534             fscal            = felec;
 535
 536             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 537
 538             /* Calculate temporary vectorial force */
 539             tx               = _mm256_mul_ps(fscal,dx11);
 540             ty               = _mm256_mul_ps(fscal,dy11);
 541             tz               = _mm256_mul_ps(fscal,dz11);
 542
 543             /* Update vectorial force */
 544             fix1             = _mm256_add_ps(fix1,tx);
 545             fiy1             = _mm256_add_ps(fiy1,ty);
 546             fiz1             = _mm256_add_ps(fiz1,tz);
 547
 548             fjx1             = _mm256_add_ps(fjx1,tx);
 549             fjy1             = _mm256_add_ps(fjy1,ty);
 550             fjz1             = _mm256_add_ps(fjz1,tz);
 551
 552             }
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 559             {
 560
 561             r12              = _mm256_mul_ps(rsq12,rinv12);
 562
 563             /* EWALD ELECTROSTATICS */
 564
 565             /* Analytical PME correction */
 566             zeta2            = _mm256_mul_ps(beta2,rsq12);
 567             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 568             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 569             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 570             felec            = _mm256_mul_ps(qq12,felec);
 571             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 572             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 573             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
 574             velec            = _mm256_mul_ps(qq12,velec);
 575
 576             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 577
 578             /* Update potential sum for this i atom from the interaction with this j atom. */
 579             velec            = _mm256_and_ps(velec,cutoff_mask);
 580             velecsum         = _mm256_add_ps(velecsum,velec);
 581
 582             fscal            = felec;
 583
 584             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 585
 586             /* Calculate temporary vectorial force */
 587             tx               = _mm256_mul_ps(fscal,dx12);
 588             ty               = _mm256_mul_ps(fscal,dy12);
 589             tz               = _mm256_mul_ps(fscal,dz12);
 590
 591             /* Update vectorial force */
 592             fix1             = _mm256_add_ps(fix1,tx);
 593             fiy1             = _mm256_add_ps(fiy1,ty);
 594             fiz1             = _mm256_add_ps(fiz1,tz);
 595
 596             fjx2             = _mm256_add_ps(fjx2,tx);
 597             fjy2             = _mm256_add_ps(fjy2,ty);
 598             fjz2             = _mm256_add_ps(fjz2,tz);
 599
 600             }
 601
 602             /**************************
 603              * CALCULATE INTERACTIONS *
 604              **************************/
 605
 606             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 607             {
 608
 609             r20              = _mm256_mul_ps(rsq20,rinv20);
 610
 611             /* EWALD ELECTROSTATICS */
 612
 613             /* Analytical PME correction */
 614             zeta2            = _mm256_mul_ps(beta2,rsq20);
 615             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 616             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 617             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 618             felec            = _mm256_mul_ps(qq20,felec);
 619             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 620             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 621             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 622             velec            = _mm256_mul_ps(qq20,velec);
 623
 624             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             velec            = _mm256_and_ps(velec,cutoff_mask);
 628             velecsum         = _mm256_add_ps(velecsum,velec);
 629
 630             fscal            = felec;
 631
 632             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 633
 634             /* Calculate temporary vectorial force */
 635             tx               = _mm256_mul_ps(fscal,dx20);
 636             ty               = _mm256_mul_ps(fscal,dy20);
 637             tz               = _mm256_mul_ps(fscal,dz20);
 638
 639             /* Update vectorial force */
 640             fix2             = _mm256_add_ps(fix2,tx);
 641             fiy2             = _mm256_add_ps(fiy2,ty);
 642             fiz2             = _mm256_add_ps(fiz2,tz);
 643
 644             fjx0             = _mm256_add_ps(fjx0,tx);
 645             fjy0             = _mm256_add_ps(fjy0,ty);
 646             fjz0             = _mm256_add_ps(fjz0,tz);
 647
 648             }
 649
 650             /**************************
 651              * CALCULATE INTERACTIONS *
 652              **************************/
 653
 654             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 655             {
 656
 657             r21              = _mm256_mul_ps(rsq21,rinv21);
 658
 659             /* EWALD ELECTROSTATICS */
 660
 661             /* Analytical PME correction */
 662             zeta2            = _mm256_mul_ps(beta2,rsq21);
 663             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 664             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 665             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 666             felec            = _mm256_mul_ps(qq21,felec);
 667             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 668             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 669             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
 670             velec            = _mm256_mul_ps(qq21,velec);
 671
 672             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
 673
 674             /* Update potential sum for this i atom from the interaction with this j atom. */
 675             velec            = _mm256_and_ps(velec,cutoff_mask);
 676             velecsum         = _mm256_add_ps(velecsum,velec);
 677
 678             fscal            = felec;
 679
 680             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 681
 682             /* Calculate temporary vectorial force */
 683             tx               = _mm256_mul_ps(fscal,dx21);
 684             ty               = _mm256_mul_ps(fscal,dy21);
 685             tz               = _mm256_mul_ps(fscal,dz21);
 686
 687             /* Update vectorial force */
 688             fix2             = _mm256_add_ps(fix2,tx);
 689             fiy2             = _mm256_add_ps(fiy2,ty);
 690             fiz2             = _mm256_add_ps(fiz2,tz);
 691
 692             fjx1             = _mm256_add_ps(fjx1,tx);
 693             fjy1             = _mm256_add_ps(fjy1,ty);
 694             fjz1             = _mm256_add_ps(fjz1,tz);
 695
 696             }
 697
 698             /**************************
 699              * CALCULATE INTERACTIONS *
 700              **************************/
 701
 702             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 703             {
 704
 705             r22              = _mm256_mul_ps(rsq22,rinv22);
 706
 707             /* EWALD ELECTROSTATICS */
 708
 709             /* Analytical PME correction */
 710             zeta2            = _mm256_mul_ps(beta2,rsq22);
 711             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 712             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 713             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 714             felec            = _mm256_mul_ps(qq22,felec);
 715             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 716             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 717             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
 718             velec            = _mm256_mul_ps(qq22,velec);
 719
 720             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
 721
 722             /* Update potential sum for this i atom from the interaction with this j atom. */
 723             velec            = _mm256_and_ps(velec,cutoff_mask);
 724             velecsum         = _mm256_add_ps(velecsum,velec);
 725
 726             fscal            = felec;
 727
 728             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 729
 730             /* Calculate temporary vectorial force */
 731             tx               = _mm256_mul_ps(fscal,dx22);
 732             ty               = _mm256_mul_ps(fscal,dy22);
 733             tz               = _mm256_mul_ps(fscal,dz22);
 734
 735             /* Update vectorial force */
 736             fix2             = _mm256_add_ps(fix2,tx);
 737             fiy2             = _mm256_add_ps(fiy2,ty);
 738             fiz2             = _mm256_add_ps(fiz2,tz);
 739
 740             fjx2             = _mm256_add_ps(fjx2,tx);
 741             fjy2             = _mm256_add_ps(fjy2,ty);
 742             fjz2             = _mm256_add_ps(fjz2,tz);
 743
 744             }
 745
 746             fjptrA             = f+j_coord_offsetA;
 747             fjptrB             = f+j_coord_offsetB;
 748             fjptrC             = f+j_coord_offsetC;
 749             fjptrD             = f+j_coord_offsetD;
 750             fjptrE             = f+j_coord_offsetE;
 751             fjptrF             = f+j_coord_offsetF;
 752             fjptrG             = f+j_coord_offsetG;
 753             fjptrH             = f+j_coord_offsetH;
 754
 755             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 756                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 757
 758             /* Inner loop uses 981 flops */
 759         }
 760
 761         if(jidx<j_index_end)
 762         {
 763
 764             /* Get j neighbor index, and coordinate index */
 765             jnrlistA         = jjnr[jidx];
 766             jnrlistB         = jjnr[jidx+1];
 767             jnrlistC         = jjnr[jidx+2];
 768             jnrlistD         = jjnr[jidx+3];
 769             jnrlistE         = jjnr[jidx+4];
 770             jnrlistF         = jjnr[jidx+5];
 771             jnrlistG         = jjnr[jidx+6];
 772             jnrlistH         = jjnr[jidx+7];
 773             /* Sign of each element will be negative for non-real atoms.
 774              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 775              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 776              */
 777             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 778                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 779
 780             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 781             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 782             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 783             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 784             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 785             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 786             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 787             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 788             j_coord_offsetA  = DIM*jnrA;
 789             j_coord_offsetB  = DIM*jnrB;
 790             j_coord_offsetC  = DIM*jnrC;
 791             j_coord_offsetD  = DIM*jnrD;
 792             j_coord_offsetE  = DIM*jnrE;
 793             j_coord_offsetF  = DIM*jnrF;
 794             j_coord_offsetG  = DIM*jnrG;
 795             j_coord_offsetH  = DIM*jnrH;
 796
 797             /* load j atom coordinates */
 798             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 799                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 800                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 801                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 802                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 803
 804             /* Calculate displacement vector */
 805             dx00             = _mm256_sub_ps(ix0,jx0);
 806             dy00             = _mm256_sub_ps(iy0,jy0);
 807             dz00             = _mm256_sub_ps(iz0,jz0);
 808             dx01             = _mm256_sub_ps(ix0,jx1);
 809             dy01             = _mm256_sub_ps(iy0,jy1);
 810             dz01             = _mm256_sub_ps(iz0,jz1);
 811             dx02             = _mm256_sub_ps(ix0,jx2);
 812             dy02             = _mm256_sub_ps(iy0,jy2);
 813             dz02             = _mm256_sub_ps(iz0,jz2);
 814             dx10             = _mm256_sub_ps(ix1,jx0);
 815             dy10             = _mm256_sub_ps(iy1,jy0);
 816             dz10             = _mm256_sub_ps(iz1,jz0);
 817             dx11             = _mm256_sub_ps(ix1,jx1);
 818             dy11             = _mm256_sub_ps(iy1,jy1);
 819             dz11             = _mm256_sub_ps(iz1,jz1);
 820             dx12             = _mm256_sub_ps(ix1,jx2);
 821             dy12             = _mm256_sub_ps(iy1,jy2);
 822             dz12             = _mm256_sub_ps(iz1,jz2);
 823             dx20             = _mm256_sub_ps(ix2,jx0);
 824             dy20             = _mm256_sub_ps(iy2,jy0);
 825             dz20             = _mm256_sub_ps(iz2,jz0);
 826             dx21             = _mm256_sub_ps(ix2,jx1);
 827             dy21             = _mm256_sub_ps(iy2,jy1);
 828             dz21             = _mm256_sub_ps(iz2,jz1);
 829             dx22             = _mm256_sub_ps(ix2,jx2);
 830             dy22             = _mm256_sub_ps(iy2,jy2);
 831             dz22             = _mm256_sub_ps(iz2,jz2);
 832
 833             /* Calculate squared distance and things based on it */
 834             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 835             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 836             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 837             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 838             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 839             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 840             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 841             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 842             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 843
 844             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 845             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 846             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 847             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 848             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 849             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 850             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 851             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 852             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 853
 854             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 855             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 856             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 857             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 858             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 859             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 860             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 861             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 862             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 863
 864             fjx0             = _mm256_setzero_ps();
 865             fjy0             = _mm256_setzero_ps();
 866             fjz0             = _mm256_setzero_ps();
 867             fjx1             = _mm256_setzero_ps();
 868             fjy1             = _mm256_setzero_ps();
 869             fjz1             = _mm256_setzero_ps();
 870             fjx2             = _mm256_setzero_ps();
 871             fjy2             = _mm256_setzero_ps();
 872             fjz2             = _mm256_setzero_ps();
 873
 874             /**************************
 875              * CALCULATE INTERACTIONS *
 876              **************************/
 877
 878             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 879             {
 880
 881             r00              = _mm256_mul_ps(rsq00,rinv00);
 882             r00              = _mm256_andnot_ps(dummy_mask,r00);
 883
 884             /* EWALD ELECTROSTATICS */
 885
 886             /* Analytical PME correction */
 887             zeta2            = _mm256_mul_ps(beta2,rsq00);
 888             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 889             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 890             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 891             felec            = _mm256_mul_ps(qq00,felec);
 892             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 893             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 894             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 895             velec            = _mm256_mul_ps(qq00,velec);
 896
 897             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 898
 899             /* Update potential sum for this i atom from the interaction with this j atom. */
 900             velec            = _mm256_and_ps(velec,cutoff_mask);
 901             velec            = _mm256_andnot_ps(dummy_mask,velec);
 902             velecsum         = _mm256_add_ps(velecsum,velec);
 903
 904             fscal            = felec;
 905
 906             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 907
 908             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 909
 910             /* Calculate temporary vectorial force */
 911             tx               = _mm256_mul_ps(fscal,dx00);
 912             ty               = _mm256_mul_ps(fscal,dy00);
 913             tz               = _mm256_mul_ps(fscal,dz00);
 914
 915             /* Update vectorial force */
 916             fix0             = _mm256_add_ps(fix0,tx);
 917             fiy0             = _mm256_add_ps(fiy0,ty);
 918             fiz0             = _mm256_add_ps(fiz0,tz);
 919
 920             fjx0             = _mm256_add_ps(fjx0,tx);
 921             fjy0             = _mm256_add_ps(fjy0,ty);
 922             fjz0             = _mm256_add_ps(fjz0,tz);
 923
 924             }
 925
 926             /**************************
 927              * CALCULATE INTERACTIONS *
 928              **************************/
 929
 930             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 931             {
 932
 933             r01              = _mm256_mul_ps(rsq01,rinv01);
 934             r01              = _mm256_andnot_ps(dummy_mask,r01);
 935
 936             /* EWALD ELECTROSTATICS */
 937
 938             /* Analytical PME correction */
 939             zeta2            = _mm256_mul_ps(beta2,rsq01);
 940             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 941             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 942             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 943             felec            = _mm256_mul_ps(qq01,felec);
 944             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 945             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 946             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv01,sh_ewald),pmecorrV);
 947             velec            = _mm256_mul_ps(qq01,velec);
 948
 949             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 950
 951             /* Update potential sum for this i atom from the interaction with this j atom. */
 952             velec            = _mm256_and_ps(velec,cutoff_mask);
 953             velec            = _mm256_andnot_ps(dummy_mask,velec);
 954             velecsum         = _mm256_add_ps(velecsum,velec);
 955
 956             fscal            = felec;
 957
 958             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 959
 960             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 961
 962             /* Calculate temporary vectorial force */
 963             tx               = _mm256_mul_ps(fscal,dx01);
 964             ty               = _mm256_mul_ps(fscal,dy01);
 965             tz               = _mm256_mul_ps(fscal,dz01);
 966
 967             /* Update vectorial force */
 968             fix0             = _mm256_add_ps(fix0,tx);
 969             fiy0             = _mm256_add_ps(fiy0,ty);
 970             fiz0             = _mm256_add_ps(fiz0,tz);
 971
 972             fjx1             = _mm256_add_ps(fjx1,tx);
 973             fjy1             = _mm256_add_ps(fjy1,ty);
 974             fjz1             = _mm256_add_ps(fjz1,tz);
 975
 976             }
 977
 978             /**************************
 979              * CALCULATE INTERACTIONS *
 980              **************************/
 981
 982             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 983             {
 984
 985             r02              = _mm256_mul_ps(rsq02,rinv02);
 986             r02              = _mm256_andnot_ps(dummy_mask,r02);
 987
 988             /* EWALD ELECTROSTATICS */
 989
 990             /* Analytical PME correction */
 991             zeta2            = _mm256_mul_ps(beta2,rsq02);
 992             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 993             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 994             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 995             felec            = _mm256_mul_ps(qq02,felec);
 996             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 997             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 998             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv02,sh_ewald),pmecorrV);
 999             velec            = _mm256_mul_ps(qq02,velec);
1000
1001             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1002
1003             /* Update potential sum for this i atom from the interaction with this j atom. */
1004             velec            = _mm256_and_ps(velec,cutoff_mask);
1005             velec            = _mm256_andnot_ps(dummy_mask,velec);
1006             velecsum         = _mm256_add_ps(velecsum,velec);
1007
1008             fscal            = felec;
1009
1010             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1011
1012             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1013
1014             /* Calculate temporary vectorial force */
1015             tx               = _mm256_mul_ps(fscal,dx02);
1016             ty               = _mm256_mul_ps(fscal,dy02);
1017             tz               = _mm256_mul_ps(fscal,dz02);
1018
1019             /* Update vectorial force */
1020             fix0             = _mm256_add_ps(fix0,tx);
1021             fiy0             = _mm256_add_ps(fiy0,ty);
1022             fiz0             = _mm256_add_ps(fiz0,tz);
1023
1024             fjx2             = _mm256_add_ps(fjx2,tx);
1025             fjy2             = _mm256_add_ps(fjy2,ty);
1026             fjz2             = _mm256_add_ps(fjz2,tz);
1027
1028             }
1029
1030             /**************************
1031              * CALCULATE INTERACTIONS *
1032              **************************/
1033
1034             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1035             {
1036
1037             r10              = _mm256_mul_ps(rsq10,rinv10);
1038             r10              = _mm256_andnot_ps(dummy_mask,r10);
1039
1040             /* EWALD ELECTROSTATICS */
1041
1042             /* Analytical PME correction */
1043             zeta2            = _mm256_mul_ps(beta2,rsq10);
1044             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1045             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1046             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1047             felec            = _mm256_mul_ps(qq10,felec);
1048             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1049             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1050             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
1051             velec            = _mm256_mul_ps(qq10,velec);
1052
1053             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1054
1055             /* Update potential sum for this i atom from the interaction with this j atom. */
1056             velec            = _mm256_and_ps(velec,cutoff_mask);
1057             velec            = _mm256_andnot_ps(dummy_mask,velec);
1058             velecsum         = _mm256_add_ps(velecsum,velec);
1059
1060             fscal            = felec;
1061
1062             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1063
1064             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1065
1066             /* Calculate temporary vectorial force */
1067             tx               = _mm256_mul_ps(fscal,dx10);
1068             ty               = _mm256_mul_ps(fscal,dy10);
1069             tz               = _mm256_mul_ps(fscal,dz10);
1070
1071             /* Update vectorial force */
1072             fix1             = _mm256_add_ps(fix1,tx);
1073             fiy1             = _mm256_add_ps(fiy1,ty);
1074             fiz1             = _mm256_add_ps(fiz1,tz);
1075
1076             fjx0             = _mm256_add_ps(fjx0,tx);
1077             fjy0             = _mm256_add_ps(fjy0,ty);
1078             fjz0             = _mm256_add_ps(fjz0,tz);
1079
1080             }
1081
1082             /**************************
1083              * CALCULATE INTERACTIONS *
1084              **************************/
1085
1086             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1087             {
1088
1089             r11              = _mm256_mul_ps(rsq11,rinv11);
1090             r11              = _mm256_andnot_ps(dummy_mask,r11);
1091
1092             /* EWALD ELECTROSTATICS */
1093
1094             /* Analytical PME correction */
1095             zeta2            = _mm256_mul_ps(beta2,rsq11);
1096             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1097             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1098             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1099             felec            = _mm256_mul_ps(qq11,felec);
1100             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1101             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1102             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
1103             velec            = _mm256_mul_ps(qq11,velec);
1104
1105             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1106
1107             /* Update potential sum for this i atom from the interaction with this j atom. */
1108             velec            = _mm256_and_ps(velec,cutoff_mask);
1109             velec            = _mm256_andnot_ps(dummy_mask,velec);
1110             velecsum         = _mm256_add_ps(velecsum,velec);
1111
1112             fscal            = felec;
1113
1114             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1115
1116             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1117
1118             /* Calculate temporary vectorial force */
1119             tx               = _mm256_mul_ps(fscal,dx11);
1120             ty               = _mm256_mul_ps(fscal,dy11);
1121             tz               = _mm256_mul_ps(fscal,dz11);
1122
1123             /* Update vectorial force */
1124             fix1             = _mm256_add_ps(fix1,tx);
1125             fiy1             = _mm256_add_ps(fiy1,ty);
1126             fiz1             = _mm256_add_ps(fiz1,tz);
1127
1128             fjx1             = _mm256_add_ps(fjx1,tx);
1129             fjy1             = _mm256_add_ps(fjy1,ty);
1130             fjz1             = _mm256_add_ps(fjz1,tz);
1131
1132             }
1133
1134             /**************************
1135              * CALCULATE INTERACTIONS *
1136              **************************/
1137
1138             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1139             {
1140
1141             r12              = _mm256_mul_ps(rsq12,rinv12);
1142             r12              = _mm256_andnot_ps(dummy_mask,r12);
1143
1144             /* EWALD ELECTROSTATICS */
1145
1146             /* Analytical PME correction */
1147             zeta2            = _mm256_mul_ps(beta2,rsq12);
1148             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1149             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1150             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1151             felec            = _mm256_mul_ps(qq12,felec);
1152             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1153             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1154             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
1155             velec            = _mm256_mul_ps(qq12,velec);
1156
1157             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1158
1159             /* Update potential sum for this i atom from the interaction with this j atom. */
1160             velec            = _mm256_and_ps(velec,cutoff_mask);
1161             velec            = _mm256_andnot_ps(dummy_mask,velec);
1162             velecsum         = _mm256_add_ps(velecsum,velec);
1163
1164             fscal            = felec;
1165
1166             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1167
1168             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1169
1170             /* Calculate temporary vectorial force */
1171             tx               = _mm256_mul_ps(fscal,dx12);
1172             ty               = _mm256_mul_ps(fscal,dy12);
1173             tz               = _mm256_mul_ps(fscal,dz12);
1174
1175             /* Update vectorial force */
1176             fix1             = _mm256_add_ps(fix1,tx);
1177             fiy1             = _mm256_add_ps(fiy1,ty);
1178             fiz1             = _mm256_add_ps(fiz1,tz);
1179
1180             fjx2             = _mm256_add_ps(fjx2,tx);
1181             fjy2             = _mm256_add_ps(fjy2,ty);
1182             fjz2             = _mm256_add_ps(fjz2,tz);
1183
1184             }
1185
1186             /**************************
1187              * CALCULATE INTERACTIONS *
1188              **************************/
1189
1190             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1191             {
1192
1193             r20              = _mm256_mul_ps(rsq20,rinv20);
1194             r20              = _mm256_andnot_ps(dummy_mask,r20);
1195
1196             /* EWALD ELECTROSTATICS */
1197
1198             /* Analytical PME correction */
1199             zeta2            = _mm256_mul_ps(beta2,rsq20);
1200             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1201             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1202             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1203             felec            = _mm256_mul_ps(qq20,felec);
1204             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1205             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1206             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
1207             velec            = _mm256_mul_ps(qq20,velec);
1208
1209             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1210
1211             /* Update potential sum for this i atom from the interaction with this j atom. */
1212             velec            = _mm256_and_ps(velec,cutoff_mask);
1213             velec            = _mm256_andnot_ps(dummy_mask,velec);
1214             velecsum         = _mm256_add_ps(velecsum,velec);
1215
1216             fscal            = felec;
1217
1218             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1219
1220             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1221
1222             /* Calculate temporary vectorial force */
1223             tx               = _mm256_mul_ps(fscal,dx20);
1224             ty               = _mm256_mul_ps(fscal,dy20);
1225             tz               = _mm256_mul_ps(fscal,dz20);
1226
1227             /* Update vectorial force */
1228             fix2             = _mm256_add_ps(fix2,tx);
1229             fiy2             = _mm256_add_ps(fiy2,ty);
1230             fiz2             = _mm256_add_ps(fiz2,tz);
1231
1232             fjx0             = _mm256_add_ps(fjx0,tx);
1233             fjy0             = _mm256_add_ps(fjy0,ty);
1234             fjz0             = _mm256_add_ps(fjz0,tz);
1235
1236             }
1237
1238             /**************************
1239              * CALCULATE INTERACTIONS *
1240              **************************/
1241
1242             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1243             {
1244
1245             r21              = _mm256_mul_ps(rsq21,rinv21);
1246             r21              = _mm256_andnot_ps(dummy_mask,r21);
1247
1248             /* EWALD ELECTROSTATICS */
1249
1250             /* Analytical PME correction */
1251             zeta2            = _mm256_mul_ps(beta2,rsq21);
1252             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1253             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1254             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1255             felec            = _mm256_mul_ps(qq21,felec);
1256             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1257             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1258             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
1259             velec            = _mm256_mul_ps(qq21,velec);
1260
1261             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1262
1263             /* Update potential sum for this i atom from the interaction with this j atom. */
1264             velec            = _mm256_and_ps(velec,cutoff_mask);
1265             velec            = _mm256_andnot_ps(dummy_mask,velec);
1266             velecsum         = _mm256_add_ps(velecsum,velec);
1267
1268             fscal            = felec;
1269
1270             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1271
1272             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1273
1274             /* Calculate temporary vectorial force */
1275             tx               = _mm256_mul_ps(fscal,dx21);
1276             ty               = _mm256_mul_ps(fscal,dy21);
1277             tz               = _mm256_mul_ps(fscal,dz21);
1278
1279             /* Update vectorial force */
1280             fix2             = _mm256_add_ps(fix2,tx);
1281             fiy2             = _mm256_add_ps(fiy2,ty);
1282             fiz2             = _mm256_add_ps(fiz2,tz);
1283
1284             fjx1             = _mm256_add_ps(fjx1,tx);
1285             fjy1             = _mm256_add_ps(fjy1,ty);
1286             fjz1             = _mm256_add_ps(fjz1,tz);
1287
1288             }
1289
1290             /**************************
1291              * CALCULATE INTERACTIONS *
1292              **************************/
1293
1294             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1295             {
1296
1297             r22              = _mm256_mul_ps(rsq22,rinv22);
1298             r22              = _mm256_andnot_ps(dummy_mask,r22);
1299
1300             /* EWALD ELECTROSTATICS */
1301
1302             /* Analytical PME correction */
1303             zeta2            = _mm256_mul_ps(beta2,rsq22);
1304             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1305             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1306             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1307             felec            = _mm256_mul_ps(qq22,felec);
1308             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1309             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1310             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
1311             velec            = _mm256_mul_ps(qq22,velec);
1312
1313             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1314
1315             /* Update potential sum for this i atom from the interaction with this j atom. */
1316             velec            = _mm256_and_ps(velec,cutoff_mask);
1317             velec            = _mm256_andnot_ps(dummy_mask,velec);
1318             velecsum         = _mm256_add_ps(velecsum,velec);
1319
1320             fscal            = felec;
1321
1322             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1323
1324             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1325
1326             /* Calculate temporary vectorial force */
1327             tx               = _mm256_mul_ps(fscal,dx22);
1328             ty               = _mm256_mul_ps(fscal,dy22);
1329             tz               = _mm256_mul_ps(fscal,dz22);
1330
1331             /* Update vectorial force */
1332             fix2             = _mm256_add_ps(fix2,tx);
1333             fiy2             = _mm256_add_ps(fiy2,ty);
1334             fiz2             = _mm256_add_ps(fiz2,tz);
1335
1336             fjx2             = _mm256_add_ps(fjx2,tx);
1337             fjy2             = _mm256_add_ps(fjy2,ty);
1338             fjz2             = _mm256_add_ps(fjz2,tz);
1339
1340             }
1341
1342             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1343             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1344             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1345             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1346             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1347             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1348             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1349             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1350
1351             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1352                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1353
1354             /* Inner loop uses 990 flops */
1355         }
1356
1357         /* End of innermost loop */
1358
1359         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1360                                                  f+i_coord_offset,fshift+i_shift_offset);
1361
1362         ggid                        = gid[iidx];
1363         /* Update potential energies */
1364         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1365
1366         /* Increment number of inner iterations */
1367         inneriter                  += j_index_end - j_index_start;
1368
1369         /* Outer loop uses 19 flops */
1370     }
1371
1372     /* Increment number of outer iterations */
1373     outeriter        += nri;
1374
1375     /* Update outer/inner flops */
1376
1377     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*990);
1378 }
1379 /*
1380  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3W3_F_avx_256_single
1381  * Electrostatics interaction: Ewald
1382  * VdW interaction:            None
1383  * Geometry:                   Water3-Water3
1384  * Calculate force/pot:        Force
1385  */
1386 void
1387 nb_kernel_ElecEwSh_VdwNone_GeomW3W3_F_avx_256_single
1388                     (t_nblist                    * gmx_restrict       nlist,
1389                      rvec                        * gmx_restrict          xx,
1390                      rvec                        * gmx_restrict          ff,
1391                      t_forcerec                  * gmx_restrict          fr,
1392                      t_mdatoms                   * gmx_restrict     mdatoms,
1393                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1394                      t_nrnb                      * gmx_restrict        nrnb)
1395 {
1396     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1397      * just 0 for non-waters.
1398      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1399      * jnr indices corresponding to data put in the four positions in the SIMD register.
1400      */
1401     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1402     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1403     int              jnrA,jnrB,jnrC,jnrD;
1404     int              jnrE,jnrF,jnrG,jnrH;
1405     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1406     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1407     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1408     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1409     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1410     real             rcutoff_scalar;
1411     real             *shiftvec,*fshift,*x,*f;
1412     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1413     real             scratch[4*DIM];
1414     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1415     real *           vdwioffsetptr0;
1416     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1417     real *           vdwioffsetptr1;
1418     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1419     real *           vdwioffsetptr2;
1420     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1421     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1422     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1423     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1424     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1425     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1426     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1427     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1428     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1429     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1430     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1431     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1432     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1433     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1434     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1435     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1436     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1437     real             *charge;
1438     __m256i          ewitab;
1439     __m128i          ewitab_lo,ewitab_hi;
1440     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1441     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1442     real             *ewtab;
1443     __m256           dummy_mask,cutoff_mask;
1444     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1445     __m256           one     = _mm256_set1_ps(1.0);
1446     __m256           two     = _mm256_set1_ps(2.0);
1447     x                = xx[0];
1448     f                = ff[0];
1449
1450     nri              = nlist->nri;
1451     iinr             = nlist->iinr;
1452     jindex           = nlist->jindex;
1453     jjnr             = nlist->jjnr;
1454     shiftidx         = nlist->shift;
1455     gid              = nlist->gid;
1456     shiftvec         = fr->shift_vec[0];
1457     fshift           = fr->fshift[0];
1458     facel            = _mm256_set1_ps(fr->epsfac);
1459     charge           = mdatoms->chargeA;
1460
1461     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1462     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1463     beta2            = _mm256_mul_ps(beta,beta);
1464     beta3            = _mm256_mul_ps(beta,beta2);
1465
1466     ewtab            = fr->ic->tabq_coul_F;
1467     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1468     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1469
1470     /* Setup water-specific parameters */
1471     inr              = nlist->iinr[0];
1472     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1473     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1474     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1475
1476     jq0              = _mm256_set1_ps(charge[inr+0]);
1477     jq1              = _mm256_set1_ps(charge[inr+1]);
1478     jq2              = _mm256_set1_ps(charge[inr+2]);
1479     qq00             = _mm256_mul_ps(iq0,jq0);
1480     qq01             = _mm256_mul_ps(iq0,jq1);
1481     qq02             = _mm256_mul_ps(iq0,jq2);
1482     qq10             = _mm256_mul_ps(iq1,jq0);
1483     qq11             = _mm256_mul_ps(iq1,jq1);
1484     qq12             = _mm256_mul_ps(iq1,jq2);
1485     qq20             = _mm256_mul_ps(iq2,jq0);
1486     qq21             = _mm256_mul_ps(iq2,jq1);
1487     qq22             = _mm256_mul_ps(iq2,jq2);
1488
1489     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1490     rcutoff_scalar   = fr->rcoulomb;
1491     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1492     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1493
1494     /* Avoid stupid compiler warnings */
1495     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1496     j_coord_offsetA = 0;
1497     j_coord_offsetB = 0;
1498     j_coord_offsetC = 0;
1499     j_coord_offsetD = 0;
1500     j_coord_offsetE = 0;
1501     j_coord_offsetF = 0;
1502     j_coord_offsetG = 0;
1503     j_coord_offsetH = 0;
1504
1505     outeriter        = 0;
1506     inneriter        = 0;
1507
1508     for(iidx=0;iidx<4*DIM;iidx++)
1509     {
1510         scratch[iidx] = 0.0;
1511     }
1512
1513     /* Start outer loop over neighborlists */
1514     for(iidx=0; iidx<nri; iidx++)
1515     {
1516         /* Load shift vector for this list */
1517         i_shift_offset   = DIM*shiftidx[iidx];
1518
1519         /* Load limits for loop over neighbors */
1520         j_index_start    = jindex[iidx];
1521         j_index_end      = jindex[iidx+1];
1522
1523         /* Get outer coordinate index */
1524         inr              = iinr[iidx];
1525         i_coord_offset   = DIM*inr;
1526
1527         /* Load i particle coords and add shift vector */
1528         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1529                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1530
1531         fix0             = _mm256_setzero_ps();
1532         fiy0             = _mm256_setzero_ps();
1533         fiz0             = _mm256_setzero_ps();
1534         fix1             = _mm256_setzero_ps();
1535         fiy1             = _mm256_setzero_ps();
1536         fiz1             = _mm256_setzero_ps();
1537         fix2             = _mm256_setzero_ps();
1538         fiy2             = _mm256_setzero_ps();
1539         fiz2             = _mm256_setzero_ps();
1540
1541         /* Start inner kernel loop */
1542         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1543         {
1544
1545             /* Get j neighbor index, and coordinate index */
1546             jnrA             = jjnr[jidx];
1547             jnrB             = jjnr[jidx+1];
1548             jnrC             = jjnr[jidx+2];
1549             jnrD             = jjnr[jidx+3];
1550             jnrE             = jjnr[jidx+4];
1551             jnrF             = jjnr[jidx+5];
1552             jnrG             = jjnr[jidx+6];
1553             jnrH             = jjnr[jidx+7];
1554             j_coord_offsetA  = DIM*jnrA;
1555             j_coord_offsetB  = DIM*jnrB;
1556             j_coord_offsetC  = DIM*jnrC;
1557             j_coord_offsetD  = DIM*jnrD;
1558             j_coord_offsetE  = DIM*jnrE;
1559             j_coord_offsetF  = DIM*jnrF;
1560             j_coord_offsetG  = DIM*jnrG;
1561             j_coord_offsetH  = DIM*jnrH;
1562
1563             /* load j atom coordinates */
1564             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1565                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1566                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1567                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1568                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1569
1570             /* Calculate displacement vector */
1571             dx00             = _mm256_sub_ps(ix0,jx0);
1572             dy00             = _mm256_sub_ps(iy0,jy0);
1573             dz00             = _mm256_sub_ps(iz0,jz0);
1574             dx01             = _mm256_sub_ps(ix0,jx1);
1575             dy01             = _mm256_sub_ps(iy0,jy1);
1576             dz01             = _mm256_sub_ps(iz0,jz1);
1577             dx02             = _mm256_sub_ps(ix0,jx2);
1578             dy02             = _mm256_sub_ps(iy0,jy2);
1579             dz02             = _mm256_sub_ps(iz0,jz2);
1580             dx10             = _mm256_sub_ps(ix1,jx0);
1581             dy10             = _mm256_sub_ps(iy1,jy0);
1582             dz10             = _mm256_sub_ps(iz1,jz0);
1583             dx11             = _mm256_sub_ps(ix1,jx1);
1584             dy11             = _mm256_sub_ps(iy1,jy1);
1585             dz11             = _mm256_sub_ps(iz1,jz1);
1586             dx12             = _mm256_sub_ps(ix1,jx2);
1587             dy12             = _mm256_sub_ps(iy1,jy2);
1588             dz12             = _mm256_sub_ps(iz1,jz2);
1589             dx20             = _mm256_sub_ps(ix2,jx0);
1590             dy20             = _mm256_sub_ps(iy2,jy0);
1591             dz20             = _mm256_sub_ps(iz2,jz0);
1592             dx21             = _mm256_sub_ps(ix2,jx1);
1593             dy21             = _mm256_sub_ps(iy2,jy1);
1594             dz21             = _mm256_sub_ps(iz2,jz1);
1595             dx22             = _mm256_sub_ps(ix2,jx2);
1596             dy22             = _mm256_sub_ps(iy2,jy2);
1597             dz22             = _mm256_sub_ps(iz2,jz2);
1598
1599             /* Calculate squared distance and things based on it */
1600             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1601             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1602             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1603             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1604             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1605             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1606             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1607             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1608             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1609
1610             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1611             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1612             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1613             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1614             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1615             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1616             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1617             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1618             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1619
1620             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1621             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1622             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1623             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1624             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1625             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1626             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1627             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1628             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1629
1630             fjx0             = _mm256_setzero_ps();
1631             fjy0             = _mm256_setzero_ps();
1632             fjz0             = _mm256_setzero_ps();
1633             fjx1             = _mm256_setzero_ps();
1634             fjy1             = _mm256_setzero_ps();
1635             fjz1             = _mm256_setzero_ps();
1636             fjx2             = _mm256_setzero_ps();
1637             fjy2             = _mm256_setzero_ps();
1638             fjz2             = _mm256_setzero_ps();
1639
1640             /**************************
1641              * CALCULATE INTERACTIONS *
1642              **************************/
1643
1644             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1645             {
1646
1647             r00              = _mm256_mul_ps(rsq00,rinv00);
1648
1649             /* EWALD ELECTROSTATICS */
1650
1651             /* Analytical PME correction */
1652             zeta2            = _mm256_mul_ps(beta2,rsq00);
1653             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1654             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1655             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1656             felec            = _mm256_mul_ps(qq00,felec);
1657
1658             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1659
1660             fscal            = felec;
1661
1662             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1663
1664             /* Calculate temporary vectorial force */
1665             tx               = _mm256_mul_ps(fscal,dx00);
1666             ty               = _mm256_mul_ps(fscal,dy00);
1667             tz               = _mm256_mul_ps(fscal,dz00);
1668
1669             /* Update vectorial force */
1670             fix0             = _mm256_add_ps(fix0,tx);
1671             fiy0             = _mm256_add_ps(fiy0,ty);
1672             fiz0             = _mm256_add_ps(fiz0,tz);
1673
1674             fjx0             = _mm256_add_ps(fjx0,tx);
1675             fjy0             = _mm256_add_ps(fjy0,ty);
1676             fjz0             = _mm256_add_ps(fjz0,tz);
1677
1678             }
1679
1680             /**************************
1681              * CALCULATE INTERACTIONS *
1682              **************************/
1683
1684             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1685             {
1686
1687             r01              = _mm256_mul_ps(rsq01,rinv01);
1688
1689             /* EWALD ELECTROSTATICS */
1690
1691             /* Analytical PME correction */
1692             zeta2            = _mm256_mul_ps(beta2,rsq01);
1693             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
1694             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1695             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1696             felec            = _mm256_mul_ps(qq01,felec);
1697
1698             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
1699
1700             fscal            = felec;
1701
1702             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1703
1704             /* Calculate temporary vectorial force */
1705             tx               = _mm256_mul_ps(fscal,dx01);
1706             ty               = _mm256_mul_ps(fscal,dy01);
1707             tz               = _mm256_mul_ps(fscal,dz01);
1708
1709             /* Update vectorial force */
1710             fix0             = _mm256_add_ps(fix0,tx);
1711             fiy0             = _mm256_add_ps(fiy0,ty);
1712             fiz0             = _mm256_add_ps(fiz0,tz);
1713
1714             fjx1             = _mm256_add_ps(fjx1,tx);
1715             fjy1             = _mm256_add_ps(fjy1,ty);
1716             fjz1             = _mm256_add_ps(fjz1,tz);
1717
1718             }
1719
1720             /**************************
1721              * CALCULATE INTERACTIONS *
1722              **************************/
1723
1724             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1725             {
1726
1727             r02              = _mm256_mul_ps(rsq02,rinv02);
1728
1729             /* EWALD ELECTROSTATICS */
1730
1731             /* Analytical PME correction */
1732             zeta2            = _mm256_mul_ps(beta2,rsq02);
1733             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
1734             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1735             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1736             felec            = _mm256_mul_ps(qq02,felec);
1737
1738             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1739
1740             fscal            = felec;
1741
1742             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1743
1744             /* Calculate temporary vectorial force */
1745             tx               = _mm256_mul_ps(fscal,dx02);
1746             ty               = _mm256_mul_ps(fscal,dy02);
1747             tz               = _mm256_mul_ps(fscal,dz02);
1748
1749             /* Update vectorial force */
1750             fix0             = _mm256_add_ps(fix0,tx);
1751             fiy0             = _mm256_add_ps(fiy0,ty);
1752             fiz0             = _mm256_add_ps(fiz0,tz);
1753
1754             fjx2             = _mm256_add_ps(fjx2,tx);
1755             fjy2             = _mm256_add_ps(fjy2,ty);
1756             fjz2             = _mm256_add_ps(fjz2,tz);
1757
1758             }
1759
1760             /**************************
1761              * CALCULATE INTERACTIONS *
1762              **************************/
1763
1764             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1765             {
1766
1767             r10              = _mm256_mul_ps(rsq10,rinv10);
1768
1769             /* EWALD ELECTROSTATICS */
1770
1771             /* Analytical PME correction */
1772             zeta2            = _mm256_mul_ps(beta2,rsq10);
1773             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1774             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1775             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1776             felec            = _mm256_mul_ps(qq10,felec);
1777
1778             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1779
1780             fscal            = felec;
1781
1782             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1783
1784             /* Calculate temporary vectorial force */
1785             tx               = _mm256_mul_ps(fscal,dx10);
1786             ty               = _mm256_mul_ps(fscal,dy10);
1787             tz               = _mm256_mul_ps(fscal,dz10);
1788
1789             /* Update vectorial force */
1790             fix1             = _mm256_add_ps(fix1,tx);
1791             fiy1             = _mm256_add_ps(fiy1,ty);
1792             fiz1             = _mm256_add_ps(fiz1,tz);
1793
1794             fjx0             = _mm256_add_ps(fjx0,tx);
1795             fjy0             = _mm256_add_ps(fjy0,ty);
1796             fjz0             = _mm256_add_ps(fjz0,tz);
1797
1798             }
1799
1800             /**************************
1801              * CALCULATE INTERACTIONS *
1802              **************************/
1803
1804             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1805             {
1806
1807             r11              = _mm256_mul_ps(rsq11,rinv11);
1808
1809             /* EWALD ELECTROSTATICS */
1810
1811             /* Analytical PME correction */
1812             zeta2            = _mm256_mul_ps(beta2,rsq11);
1813             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1814             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1815             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1816             felec            = _mm256_mul_ps(qq11,felec);
1817
1818             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1819
1820             fscal            = felec;
1821
1822             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1823
1824             /* Calculate temporary vectorial force */
1825             tx               = _mm256_mul_ps(fscal,dx11);
1826             ty               = _mm256_mul_ps(fscal,dy11);
1827             tz               = _mm256_mul_ps(fscal,dz11);
1828
1829             /* Update vectorial force */
1830             fix1             = _mm256_add_ps(fix1,tx);
1831             fiy1             = _mm256_add_ps(fiy1,ty);
1832             fiz1             = _mm256_add_ps(fiz1,tz);
1833
1834             fjx1             = _mm256_add_ps(fjx1,tx);
1835             fjy1             = _mm256_add_ps(fjy1,ty);
1836             fjz1             = _mm256_add_ps(fjz1,tz);
1837
1838             }
1839
1840             /**************************
1841              * CALCULATE INTERACTIONS *
1842              **************************/
1843
1844             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1845             {
1846
1847             r12              = _mm256_mul_ps(rsq12,rinv12);
1848
1849             /* EWALD ELECTROSTATICS */
1850
1851             /* Analytical PME correction */
1852             zeta2            = _mm256_mul_ps(beta2,rsq12);
1853             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1854             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1855             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1856             felec            = _mm256_mul_ps(qq12,felec);
1857
1858             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1859
1860             fscal            = felec;
1861
1862             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1863
1864             /* Calculate temporary vectorial force */
1865             tx               = _mm256_mul_ps(fscal,dx12);
1866             ty               = _mm256_mul_ps(fscal,dy12);
1867             tz               = _mm256_mul_ps(fscal,dz12);
1868
1869             /* Update vectorial force */
1870             fix1             = _mm256_add_ps(fix1,tx);
1871             fiy1             = _mm256_add_ps(fiy1,ty);
1872             fiz1             = _mm256_add_ps(fiz1,tz);
1873
1874             fjx2             = _mm256_add_ps(fjx2,tx);
1875             fjy2             = _mm256_add_ps(fjy2,ty);
1876             fjz2             = _mm256_add_ps(fjz2,tz);
1877
1878             }
1879
1880             /**************************
1881              * CALCULATE INTERACTIONS *
1882              **************************/
1883
1884             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1885             {
1886
1887             r20              = _mm256_mul_ps(rsq20,rinv20);
1888
1889             /* EWALD ELECTROSTATICS */
1890
1891             /* Analytical PME correction */
1892             zeta2            = _mm256_mul_ps(beta2,rsq20);
1893             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1894             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1895             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1896             felec            = _mm256_mul_ps(qq20,felec);
1897
1898             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1899
1900             fscal            = felec;
1901
1902             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1903
1904             /* Calculate temporary vectorial force */
1905             tx               = _mm256_mul_ps(fscal,dx20);
1906             ty               = _mm256_mul_ps(fscal,dy20);
1907             tz               = _mm256_mul_ps(fscal,dz20);
1908
1909             /* Update vectorial force */
1910             fix2             = _mm256_add_ps(fix2,tx);
1911             fiy2             = _mm256_add_ps(fiy2,ty);
1912             fiz2             = _mm256_add_ps(fiz2,tz);
1913
1914             fjx0             = _mm256_add_ps(fjx0,tx);
1915             fjy0             = _mm256_add_ps(fjy0,ty);
1916             fjz0             = _mm256_add_ps(fjz0,tz);
1917
1918             }
1919
1920             /**************************
1921              * CALCULATE INTERACTIONS *
1922              **************************/
1923
1924             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1925             {
1926
1927             r21              = _mm256_mul_ps(rsq21,rinv21);
1928
1929             /* EWALD ELECTROSTATICS */
1930
1931             /* Analytical PME correction */
1932             zeta2            = _mm256_mul_ps(beta2,rsq21);
1933             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1934             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1935             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1936             felec            = _mm256_mul_ps(qq21,felec);
1937
1938             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1939
1940             fscal            = felec;
1941
1942             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1943
1944             /* Calculate temporary vectorial force */
1945             tx               = _mm256_mul_ps(fscal,dx21);
1946             ty               = _mm256_mul_ps(fscal,dy21);
1947             tz               = _mm256_mul_ps(fscal,dz21);
1948
1949             /* Update vectorial force */
1950             fix2             = _mm256_add_ps(fix2,tx);
1951             fiy2             = _mm256_add_ps(fiy2,ty);
1952             fiz2             = _mm256_add_ps(fiz2,tz);
1953
1954             fjx1             = _mm256_add_ps(fjx1,tx);
1955             fjy1             = _mm256_add_ps(fjy1,ty);
1956             fjz1             = _mm256_add_ps(fjz1,tz);
1957
1958             }
1959
1960             /**************************
1961              * CALCULATE INTERACTIONS *
1962              **************************/
1963
1964             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1965             {
1966
1967             r22              = _mm256_mul_ps(rsq22,rinv22);
1968
1969             /* EWALD ELECTROSTATICS */
1970
1971             /* Analytical PME correction */
1972             zeta2            = _mm256_mul_ps(beta2,rsq22);
1973             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1974             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1975             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1976             felec            = _mm256_mul_ps(qq22,felec);
1977
1978             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1979
1980             fscal            = felec;
1981
1982             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1983
1984             /* Calculate temporary vectorial force */
1985             tx               = _mm256_mul_ps(fscal,dx22);
1986             ty               = _mm256_mul_ps(fscal,dy22);
1987             tz               = _mm256_mul_ps(fscal,dz22);
1988
1989             /* Update vectorial force */
1990             fix2             = _mm256_add_ps(fix2,tx);
1991             fiy2             = _mm256_add_ps(fiy2,ty);
1992             fiz2             = _mm256_add_ps(fiz2,tz);
1993
1994             fjx2             = _mm256_add_ps(fjx2,tx);
1995             fjy2             = _mm256_add_ps(fjy2,ty);
1996             fjz2             = _mm256_add_ps(fjz2,tz);
1997
1998             }
1999
2000             fjptrA             = f+j_coord_offsetA;
2001             fjptrB             = f+j_coord_offsetB;
2002             fjptrC             = f+j_coord_offsetC;
2003             fjptrD             = f+j_coord_offsetD;
2004             fjptrE             = f+j_coord_offsetE;
2005             fjptrF             = f+j_coord_offsetF;
2006             fjptrG             = f+j_coord_offsetG;
2007             fjptrH             = f+j_coord_offsetH;
2008
2009             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2010                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2011
2012             /* Inner loop uses 531 flops */
2013         }
2014
2015         if(jidx<j_index_end)
2016         {
2017
2018             /* Get j neighbor index, and coordinate index */
2019             jnrlistA         = jjnr[jidx];
2020             jnrlistB         = jjnr[jidx+1];
2021             jnrlistC         = jjnr[jidx+2];
2022             jnrlistD         = jjnr[jidx+3];
2023             jnrlistE         = jjnr[jidx+4];
2024             jnrlistF         = jjnr[jidx+5];
2025             jnrlistG         = jjnr[jidx+6];
2026             jnrlistH         = jjnr[jidx+7];
2027             /* Sign of each element will be negative for non-real atoms.
2028              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2029              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2030              */
2031             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2032                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2033
2034             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2035             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2036             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2037             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2038             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2039             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2040             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2041             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2042             j_coord_offsetA  = DIM*jnrA;
2043             j_coord_offsetB  = DIM*jnrB;
2044             j_coord_offsetC  = DIM*jnrC;
2045             j_coord_offsetD  = DIM*jnrD;
2046             j_coord_offsetE  = DIM*jnrE;
2047             j_coord_offsetF  = DIM*jnrF;
2048             j_coord_offsetG  = DIM*jnrG;
2049             j_coord_offsetH  = DIM*jnrH;
2050
2051             /* load j atom coordinates */
2052             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2053                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2054                                                  x+j_coord_offsetE,x+j_coord_offsetF,
2055                                                  x+j_coord_offsetG,x+j_coord_offsetH,
2056                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2057
2058             /* Calculate displacement vector */
2059             dx00             = _mm256_sub_ps(ix0,jx0);
2060             dy00             = _mm256_sub_ps(iy0,jy0);
2061             dz00             = _mm256_sub_ps(iz0,jz0);
2062             dx01             = _mm256_sub_ps(ix0,jx1);
2063             dy01             = _mm256_sub_ps(iy0,jy1);
2064             dz01             = _mm256_sub_ps(iz0,jz1);
2065             dx02             = _mm256_sub_ps(ix0,jx2);
2066             dy02             = _mm256_sub_ps(iy0,jy2);
2067             dz02             = _mm256_sub_ps(iz0,jz2);
2068             dx10             = _mm256_sub_ps(ix1,jx0);
2069             dy10             = _mm256_sub_ps(iy1,jy0);
2070             dz10             = _mm256_sub_ps(iz1,jz0);
2071             dx11             = _mm256_sub_ps(ix1,jx1);
2072             dy11             = _mm256_sub_ps(iy1,jy1);
2073             dz11             = _mm256_sub_ps(iz1,jz1);
2074             dx12             = _mm256_sub_ps(ix1,jx2);
2075             dy12             = _mm256_sub_ps(iy1,jy2);
2076             dz12             = _mm256_sub_ps(iz1,jz2);
2077             dx20             = _mm256_sub_ps(ix2,jx0);
2078             dy20             = _mm256_sub_ps(iy2,jy0);
2079             dz20             = _mm256_sub_ps(iz2,jz0);
2080             dx21             = _mm256_sub_ps(ix2,jx1);
2081             dy21             = _mm256_sub_ps(iy2,jy1);
2082             dz21             = _mm256_sub_ps(iz2,jz1);
2083             dx22             = _mm256_sub_ps(ix2,jx2);
2084             dy22             = _mm256_sub_ps(iy2,jy2);
2085             dz22             = _mm256_sub_ps(iz2,jz2);
2086
2087             /* Calculate squared distance and things based on it */
2088             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2089             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
2090             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
2091             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
2092             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2093             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2094             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
2095             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2096             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2097
2098             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
2099             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
2100             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
2101             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
2102             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2103             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2104             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
2105             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2106             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2107
2108             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
2109             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
2110             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
2111             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
2112             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2113             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2114             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
2115             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2116             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2117
2118             fjx0             = _mm256_setzero_ps();
2119             fjy0             = _mm256_setzero_ps();
2120             fjz0             = _mm256_setzero_ps();
2121             fjx1             = _mm256_setzero_ps();
2122             fjy1             = _mm256_setzero_ps();
2123             fjz1             = _mm256_setzero_ps();
2124             fjx2             = _mm256_setzero_ps();
2125             fjy2             = _mm256_setzero_ps();
2126             fjz2             = _mm256_setzero_ps();
2127
2128             /**************************
2129              * CALCULATE INTERACTIONS *
2130              **************************/
2131
2132             if (gmx_mm256_any_lt(rsq00,rcutoff2))
2133             {
2134
2135             r00              = _mm256_mul_ps(rsq00,rinv00);
2136             r00              = _mm256_andnot_ps(dummy_mask,r00);
2137
2138             /* EWALD ELECTROSTATICS */
2139
2140             /* Analytical PME correction */
2141             zeta2            = _mm256_mul_ps(beta2,rsq00);
2142             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
2143             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2144             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2145             felec            = _mm256_mul_ps(qq00,felec);
2146
2147             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
2148
2149             fscal            = felec;
2150
2151             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2152
2153             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2154
2155             /* Calculate temporary vectorial force */
2156             tx               = _mm256_mul_ps(fscal,dx00);
2157             ty               = _mm256_mul_ps(fscal,dy00);
2158             tz               = _mm256_mul_ps(fscal,dz00);
2159
2160             /* Update vectorial force */
2161             fix0             = _mm256_add_ps(fix0,tx);
2162             fiy0             = _mm256_add_ps(fiy0,ty);
2163             fiz0             = _mm256_add_ps(fiz0,tz);
2164
2165             fjx0             = _mm256_add_ps(fjx0,tx);
2166             fjy0             = _mm256_add_ps(fjy0,ty);
2167             fjz0             = _mm256_add_ps(fjz0,tz);
2168
2169             }
2170
2171             /**************************
2172              * CALCULATE INTERACTIONS *
2173              **************************/
2174
2175             if (gmx_mm256_any_lt(rsq01,rcutoff2))
2176             {
2177
2178             r01              = _mm256_mul_ps(rsq01,rinv01);
2179             r01              = _mm256_andnot_ps(dummy_mask,r01);
2180
2181             /* EWALD ELECTROSTATICS */
2182
2183             /* Analytical PME correction */
2184             zeta2            = _mm256_mul_ps(beta2,rsq01);
2185             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
2186             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2187             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2188             felec            = _mm256_mul_ps(qq01,felec);
2189
2190             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
2191
2192             fscal            = felec;
2193
2194             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2195
2196             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2197
2198             /* Calculate temporary vectorial force */
2199             tx               = _mm256_mul_ps(fscal,dx01);
2200             ty               = _mm256_mul_ps(fscal,dy01);
2201             tz               = _mm256_mul_ps(fscal,dz01);
2202
2203             /* Update vectorial force */
2204             fix0             = _mm256_add_ps(fix0,tx);
2205             fiy0             = _mm256_add_ps(fiy0,ty);
2206             fiz0             = _mm256_add_ps(fiz0,tz);
2207
2208             fjx1             = _mm256_add_ps(fjx1,tx);
2209             fjy1             = _mm256_add_ps(fjy1,ty);
2210             fjz1             = _mm256_add_ps(fjz1,tz);
2211
2212             }
2213
2214             /**************************
2215              * CALCULATE INTERACTIONS *
2216              **************************/
2217
2218             if (gmx_mm256_any_lt(rsq02,rcutoff2))
2219             {
2220
2221             r02              = _mm256_mul_ps(rsq02,rinv02);
2222             r02              = _mm256_andnot_ps(dummy_mask,r02);
2223
2224             /* EWALD ELECTROSTATICS */
2225
2226             /* Analytical PME correction */
2227             zeta2            = _mm256_mul_ps(beta2,rsq02);
2228             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
2229             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2230             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2231             felec            = _mm256_mul_ps(qq02,felec);
2232
2233             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
2234
2235             fscal            = felec;
2236
2237             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2238
2239             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2240
2241             /* Calculate temporary vectorial force */
2242             tx               = _mm256_mul_ps(fscal,dx02);
2243             ty               = _mm256_mul_ps(fscal,dy02);
2244             tz               = _mm256_mul_ps(fscal,dz02);
2245
2246             /* Update vectorial force */
2247             fix0             = _mm256_add_ps(fix0,tx);
2248             fiy0             = _mm256_add_ps(fiy0,ty);
2249             fiz0             = _mm256_add_ps(fiz0,tz);
2250
2251             fjx2             = _mm256_add_ps(fjx2,tx);
2252             fjy2             = _mm256_add_ps(fjy2,ty);
2253             fjz2             = _mm256_add_ps(fjz2,tz);
2254
2255             }
2256
2257             /**************************
2258              * CALCULATE INTERACTIONS *
2259              **************************/
2260
2261             if (gmx_mm256_any_lt(rsq10,rcutoff2))
2262             {
2263
2264             r10              = _mm256_mul_ps(rsq10,rinv10);
2265             r10              = _mm256_andnot_ps(dummy_mask,r10);
2266
2267             /* EWALD ELECTROSTATICS */
2268
2269             /* Analytical PME correction */
2270             zeta2            = _mm256_mul_ps(beta2,rsq10);
2271             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
2272             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2273             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2274             felec            = _mm256_mul_ps(qq10,felec);
2275
2276             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
2277
2278             fscal            = felec;
2279
2280             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2281
2282             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2283
2284             /* Calculate temporary vectorial force */
2285             tx               = _mm256_mul_ps(fscal,dx10);
2286             ty               = _mm256_mul_ps(fscal,dy10);
2287             tz               = _mm256_mul_ps(fscal,dz10);
2288
2289             /* Update vectorial force */
2290             fix1             = _mm256_add_ps(fix1,tx);
2291             fiy1             = _mm256_add_ps(fiy1,ty);
2292             fiz1             = _mm256_add_ps(fiz1,tz);
2293
2294             fjx0             = _mm256_add_ps(fjx0,tx);
2295             fjy0             = _mm256_add_ps(fjy0,ty);
2296             fjz0             = _mm256_add_ps(fjz0,tz);
2297
2298             }
2299
2300             /**************************
2301              * CALCULATE INTERACTIONS *
2302              **************************/
2303
2304             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2305             {
2306
2307             r11              = _mm256_mul_ps(rsq11,rinv11);
2308             r11              = _mm256_andnot_ps(dummy_mask,r11);
2309
2310             /* EWALD ELECTROSTATICS */
2311
2312             /* Analytical PME correction */
2313             zeta2            = _mm256_mul_ps(beta2,rsq11);
2314             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
2315             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2316             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2317             felec            = _mm256_mul_ps(qq11,felec);
2318
2319             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2320
2321             fscal            = felec;
2322
2323             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2324
2325             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2326
2327             /* Calculate temporary vectorial force */
2328             tx               = _mm256_mul_ps(fscal,dx11);
2329             ty               = _mm256_mul_ps(fscal,dy11);
2330             tz               = _mm256_mul_ps(fscal,dz11);
2331
2332             /* Update vectorial force */
2333             fix1             = _mm256_add_ps(fix1,tx);
2334             fiy1             = _mm256_add_ps(fiy1,ty);
2335             fiz1             = _mm256_add_ps(fiz1,tz);
2336
2337             fjx1             = _mm256_add_ps(fjx1,tx);
2338             fjy1             = _mm256_add_ps(fjy1,ty);
2339             fjz1             = _mm256_add_ps(fjz1,tz);
2340
2341             }
2342
2343             /**************************
2344              * CALCULATE INTERACTIONS *
2345              **************************/
2346
2347             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2348             {
2349
2350             r12              = _mm256_mul_ps(rsq12,rinv12);
2351             r12              = _mm256_andnot_ps(dummy_mask,r12);
2352
2353             /* EWALD ELECTROSTATICS */
2354
2355             /* Analytical PME correction */
2356             zeta2            = _mm256_mul_ps(beta2,rsq12);
2357             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2358             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2359             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2360             felec            = _mm256_mul_ps(qq12,felec);
2361
2362             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2363
2364             fscal            = felec;
2365
2366             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2367
2368             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2369
2370             /* Calculate temporary vectorial force */
2371             tx               = _mm256_mul_ps(fscal,dx12);
2372             ty               = _mm256_mul_ps(fscal,dy12);
2373             tz               = _mm256_mul_ps(fscal,dz12);
2374
2375             /* Update vectorial force */
2376             fix1             = _mm256_add_ps(fix1,tx);
2377             fiy1             = _mm256_add_ps(fiy1,ty);
2378             fiz1             = _mm256_add_ps(fiz1,tz);
2379
2380             fjx2             = _mm256_add_ps(fjx2,tx);
2381             fjy2             = _mm256_add_ps(fjy2,ty);
2382             fjz2             = _mm256_add_ps(fjz2,tz);
2383
2384             }
2385
2386             /**************************
2387              * CALCULATE INTERACTIONS *
2388              **************************/
2389
2390             if (gmx_mm256_any_lt(rsq20,rcutoff2))
2391             {
2392
2393             r20              = _mm256_mul_ps(rsq20,rinv20);
2394             r20              = _mm256_andnot_ps(dummy_mask,r20);
2395
2396             /* EWALD ELECTROSTATICS */
2397
2398             /* Analytical PME correction */
2399             zeta2            = _mm256_mul_ps(beta2,rsq20);
2400             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
2401             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2402             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2403             felec            = _mm256_mul_ps(qq20,felec);
2404
2405             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
2406
2407             fscal            = felec;
2408
2409             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2410
2411             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2412
2413             /* Calculate temporary vectorial force */
2414             tx               = _mm256_mul_ps(fscal,dx20);
2415             ty               = _mm256_mul_ps(fscal,dy20);
2416             tz               = _mm256_mul_ps(fscal,dz20);
2417
2418             /* Update vectorial force */
2419             fix2             = _mm256_add_ps(fix2,tx);
2420             fiy2             = _mm256_add_ps(fiy2,ty);
2421             fiz2             = _mm256_add_ps(fiz2,tz);
2422
2423             fjx0             = _mm256_add_ps(fjx0,tx);
2424             fjy0             = _mm256_add_ps(fjy0,ty);
2425             fjz0             = _mm256_add_ps(fjz0,tz);
2426
2427             }
2428
2429             /**************************
2430              * CALCULATE INTERACTIONS *
2431              **************************/
2432
2433             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2434             {
2435
2436             r21              = _mm256_mul_ps(rsq21,rinv21);
2437             r21              = _mm256_andnot_ps(dummy_mask,r21);
2438
2439             /* EWALD ELECTROSTATICS */
2440
2441             /* Analytical PME correction */
2442             zeta2            = _mm256_mul_ps(beta2,rsq21);
2443             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2444             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2445             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2446             felec            = _mm256_mul_ps(qq21,felec);
2447
2448             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2449
2450             fscal            = felec;
2451
2452             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2453
2454             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2455
2456             /* Calculate temporary vectorial force */
2457             tx               = _mm256_mul_ps(fscal,dx21);
2458             ty               = _mm256_mul_ps(fscal,dy21);
2459             tz               = _mm256_mul_ps(fscal,dz21);
2460
2461             /* Update vectorial force */
2462             fix2             = _mm256_add_ps(fix2,tx);
2463             fiy2             = _mm256_add_ps(fiy2,ty);
2464             fiz2             = _mm256_add_ps(fiz2,tz);
2465
2466             fjx1             = _mm256_add_ps(fjx1,tx);
2467             fjy1             = _mm256_add_ps(fjy1,ty);
2468             fjz1             = _mm256_add_ps(fjz1,tz);
2469
2470             }
2471
2472             /**************************
2473              * CALCULATE INTERACTIONS *
2474              **************************/
2475
2476             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2477             {
2478
2479             r22              = _mm256_mul_ps(rsq22,rinv22);
2480             r22              = _mm256_andnot_ps(dummy_mask,r22);
2481
2482             /* EWALD ELECTROSTATICS */
2483
2484             /* Analytical PME correction */
2485             zeta2            = _mm256_mul_ps(beta2,rsq22);
2486             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2487             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2488             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2489             felec            = _mm256_mul_ps(qq22,felec);
2490
2491             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2492
2493             fscal            = felec;
2494
2495             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2496
2497             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2498
2499             /* Calculate temporary vectorial force */
2500             tx               = _mm256_mul_ps(fscal,dx22);
2501             ty               = _mm256_mul_ps(fscal,dy22);
2502             tz               = _mm256_mul_ps(fscal,dz22);
2503
2504             /* Update vectorial force */
2505             fix2             = _mm256_add_ps(fix2,tx);
2506             fiy2             = _mm256_add_ps(fiy2,ty);
2507             fiz2             = _mm256_add_ps(fiz2,tz);
2508
2509             fjx2             = _mm256_add_ps(fjx2,tx);
2510             fjy2             = _mm256_add_ps(fjy2,ty);
2511             fjz2             = _mm256_add_ps(fjz2,tz);
2512
2513             }
2514
2515             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2516             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2517             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2518             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2519             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2520             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2521             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2522             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2523
2524             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2525                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2526
2527             /* Inner loop uses 540 flops */
2528         }
2529
2530         /* End of innermost loop */
2531
2532         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2533                                                  f+i_coord_offset,fshift+i_shift_offset);
2534
2535         /* Increment number of inner iterations */
2536         inneriter                  += j_index_end - j_index_start;
2537
2538         /* Outer loop uses 18 flops */
2539     }
2540
2541     /* Increment number of outer iterations */
2542     outeriter        += nri;
2543
2544     /* Update outer/inner flops */
2545
2546     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*540);
2547 }