src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwLJ_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_ElecEw_VdwLJ_GeomW3W3_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_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     int              nvdwtype;
 110     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 111     int              *vdwtype;
 112     real             *vdwparam;
 113     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 114     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 115     __m256i          ewitab;
 116     __m128i          ewitab_lo,ewitab_hi;
 117     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 118     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 119     real             *ewtab;
 120     __m256           dummy_mask,cutoff_mask;
 121     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 122     __m256           one     = _mm256_set1_ps(1.0);
 123     __m256           two     = _mm256_set1_ps(2.0);
 124     x                = xx[0];
 125     f                = ff[0];
 126
 127     nri              = nlist->nri;
 128     iinr             = nlist->iinr;
 129     jindex           = nlist->jindex;
 130     jjnr             = nlist->jjnr;
 131     shiftidx         = nlist->shift;
 132     gid              = nlist->gid;
 133     shiftvec         = fr->shift_vec[0];
 134     fshift           = fr->fshift[0];
 135     facel            = _mm256_set1_ps(fr->epsfac);
 136     charge           = mdatoms->chargeA;
 137     nvdwtype         = fr->ntype;
 138     vdwparam         = fr->nbfp;
 139     vdwtype          = mdatoms->typeA;
 140
 141     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 142     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 143     beta2            = _mm256_mul_ps(beta,beta);
 144     beta3            = _mm256_mul_ps(beta,beta2);
 145
 146     ewtab            = fr->ic->tabq_coul_FDV0;
 147     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 148     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 149
 150     /* Setup water-specific parameters */
 151     inr              = nlist->iinr[0];
 152     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 153     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 154     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 155     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 156
 157     jq0              = _mm256_set1_ps(charge[inr+0]);
 158     jq1              = _mm256_set1_ps(charge[inr+1]);
 159     jq2              = _mm256_set1_ps(charge[inr+2]);
 160     vdwjidx0A        = 2*vdwtype[inr+0];
 161     qq00             = _mm256_mul_ps(iq0,jq0);
 162     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 163     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 164     qq01             = _mm256_mul_ps(iq0,jq1);
 165     qq02             = _mm256_mul_ps(iq0,jq2);
 166     qq10             = _mm256_mul_ps(iq1,jq0);
 167     qq11             = _mm256_mul_ps(iq1,jq1);
 168     qq12             = _mm256_mul_ps(iq1,jq2);
 169     qq20             = _mm256_mul_ps(iq2,jq0);
 170     qq21             = _mm256_mul_ps(iq2,jq1);
 171     qq22             = _mm256_mul_ps(iq2,jq2);
 172
 173     /* Avoid stupid compiler warnings */
 174     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 175     j_coord_offsetA = 0;
 176     j_coord_offsetB = 0;
 177     j_coord_offsetC = 0;
 178     j_coord_offsetD = 0;
 179     j_coord_offsetE = 0;
 180     j_coord_offsetF = 0;
 181     j_coord_offsetG = 0;
 182     j_coord_offsetH = 0;
 183
 184     outeriter        = 0;
 185     inneriter        = 0;
 186
 187     for(iidx=0;iidx<4*DIM;iidx++)
 188     {
 189         scratch[iidx] = 0.0;
 190     }
 191
 192     /* Start outer loop over neighborlists */
 193     for(iidx=0; iidx<nri; iidx++)
 194     {
 195         /* Load shift vector for this list */
 196         i_shift_offset   = DIM*shiftidx[iidx];
 197
 198         /* Load limits for loop over neighbors */
 199         j_index_start    = jindex[iidx];
 200         j_index_end      = jindex[iidx+1];
 201
 202         /* Get outer coordinate index */
 203         inr              = iinr[iidx];
 204         i_coord_offset   = DIM*inr;
 205
 206         /* Load i particle coords and add shift vector */
 207         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 208                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 209
 210         fix0             = _mm256_setzero_ps();
 211         fiy0             = _mm256_setzero_ps();
 212         fiz0             = _mm256_setzero_ps();
 213         fix1             = _mm256_setzero_ps();
 214         fiy1             = _mm256_setzero_ps();
 215         fiz1             = _mm256_setzero_ps();
 216         fix2             = _mm256_setzero_ps();
 217         fiy2             = _mm256_setzero_ps();
 218         fiz2             = _mm256_setzero_ps();
 219
 220         /* Reset potential sums */
 221         velecsum         = _mm256_setzero_ps();
 222         vvdwsum          = _mm256_setzero_ps();
 223
 224         /* Start inner kernel loop */
 225         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 226         {
 227
 228             /* Get j neighbor index, and coordinate index */
 229             jnrA             = jjnr[jidx];
 230             jnrB             = jjnr[jidx+1];
 231             jnrC             = jjnr[jidx+2];
 232             jnrD             = jjnr[jidx+3];
 233             jnrE             = jjnr[jidx+4];
 234             jnrF             = jjnr[jidx+5];
 235             jnrG             = jjnr[jidx+6];
 236             jnrH             = jjnr[jidx+7];
 237             j_coord_offsetA  = DIM*jnrA;
 238             j_coord_offsetB  = DIM*jnrB;
 239             j_coord_offsetC  = DIM*jnrC;
 240             j_coord_offsetD  = DIM*jnrD;
 241             j_coord_offsetE  = DIM*jnrE;
 242             j_coord_offsetF  = DIM*jnrF;
 243             j_coord_offsetG  = DIM*jnrG;
 244             j_coord_offsetH  = DIM*jnrH;
 245
 246             /* load j atom coordinates */
 247             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 248                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 249                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 250                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 251                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 252
 253             /* Calculate displacement vector */
 254             dx00             = _mm256_sub_ps(ix0,jx0);
 255             dy00             = _mm256_sub_ps(iy0,jy0);
 256             dz00             = _mm256_sub_ps(iz0,jz0);
 257             dx01             = _mm256_sub_ps(ix0,jx1);
 258             dy01             = _mm256_sub_ps(iy0,jy1);
 259             dz01             = _mm256_sub_ps(iz0,jz1);
 260             dx02             = _mm256_sub_ps(ix0,jx2);
 261             dy02             = _mm256_sub_ps(iy0,jy2);
 262             dz02             = _mm256_sub_ps(iz0,jz2);
 263             dx10             = _mm256_sub_ps(ix1,jx0);
 264             dy10             = _mm256_sub_ps(iy1,jy0);
 265             dz10             = _mm256_sub_ps(iz1,jz0);
 266             dx11             = _mm256_sub_ps(ix1,jx1);
 267             dy11             = _mm256_sub_ps(iy1,jy1);
 268             dz11             = _mm256_sub_ps(iz1,jz1);
 269             dx12             = _mm256_sub_ps(ix1,jx2);
 270             dy12             = _mm256_sub_ps(iy1,jy2);
 271             dz12             = _mm256_sub_ps(iz1,jz2);
 272             dx20             = _mm256_sub_ps(ix2,jx0);
 273             dy20             = _mm256_sub_ps(iy2,jy0);
 274             dz20             = _mm256_sub_ps(iz2,jz0);
 275             dx21             = _mm256_sub_ps(ix2,jx1);
 276             dy21             = _mm256_sub_ps(iy2,jy1);
 277             dz21             = _mm256_sub_ps(iz2,jz1);
 278             dx22             = _mm256_sub_ps(ix2,jx2);
 279             dy22             = _mm256_sub_ps(iy2,jy2);
 280             dz22             = _mm256_sub_ps(iz2,jz2);
 281
 282             /* Calculate squared distance and things based on it */
 283             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 284             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 285             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 286             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 287             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 288             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 289             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 290             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 291             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 292
 293             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 294             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 295             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 296             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 297             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 298             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 299             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 300             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 301             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 302
 303             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 304             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 305             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 306             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 307             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 308             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 309             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 310             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 311             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 312
 313             fjx0             = _mm256_setzero_ps();
 314             fjy0             = _mm256_setzero_ps();
 315             fjz0             = _mm256_setzero_ps();
 316             fjx1             = _mm256_setzero_ps();
 317             fjy1             = _mm256_setzero_ps();
 318             fjz1             = _mm256_setzero_ps();
 319             fjx2             = _mm256_setzero_ps();
 320             fjy2             = _mm256_setzero_ps();
 321             fjz2             = _mm256_setzero_ps();
 322
 323             /**************************
 324              * CALCULATE INTERACTIONS *
 325              **************************/
 326
 327             r00              = _mm256_mul_ps(rsq00,rinv00);
 328
 329             /* EWALD ELECTROSTATICS */
 330
 331             /* Analytical PME correction */
 332             zeta2            = _mm256_mul_ps(beta2,rsq00);
 333             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 334             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 335             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 336             felec            = _mm256_mul_ps(qq00,felec);
 337             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 338             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 339             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 340             velec            = _mm256_mul_ps(qq00,velec);
 341
 342             /* LENNARD-JONES DISPERSION/REPULSION */
 343
 344             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 345             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 346             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 347             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 348             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 349
 350             /* Update potential sum for this i atom from the interaction with this j atom. */
 351             velecsum         = _mm256_add_ps(velecsum,velec);
 352             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 353
 354             fscal            = _mm256_add_ps(felec,fvdw);
 355
 356             /* Calculate temporary vectorial force */
 357             tx               = _mm256_mul_ps(fscal,dx00);
 358             ty               = _mm256_mul_ps(fscal,dy00);
 359             tz               = _mm256_mul_ps(fscal,dz00);
 360
 361             /* Update vectorial force */
 362             fix0             = _mm256_add_ps(fix0,tx);
 363             fiy0             = _mm256_add_ps(fiy0,ty);
 364             fiz0             = _mm256_add_ps(fiz0,tz);
 365
 366             fjx0             = _mm256_add_ps(fjx0,tx);
 367             fjy0             = _mm256_add_ps(fjy0,ty);
 368             fjz0             = _mm256_add_ps(fjz0,tz);
 369
 370             /**************************
 371              * CALCULATE INTERACTIONS *
 372              **************************/
 373
 374             r01              = _mm256_mul_ps(rsq01,rinv01);
 375
 376             /* EWALD ELECTROSTATICS */
 377
 378             /* Analytical PME correction */
 379             zeta2            = _mm256_mul_ps(beta2,rsq01);
 380             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 381             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 382             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 383             felec            = _mm256_mul_ps(qq01,felec);
 384             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 385             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 386             velec            = _mm256_sub_ps(rinv01,pmecorrV);
 387             velec            = _mm256_mul_ps(qq01,velec);
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velecsum         = _mm256_add_ps(velecsum,velec);
 391
 392             fscal            = felec;
 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              * CALCULATE INTERACTIONS *
 410              **************************/
 411
 412             r02              = _mm256_mul_ps(rsq02,rinv02);
 413
 414             /* EWALD ELECTROSTATICS */
 415
 416             /* Analytical PME correction */
 417             zeta2            = _mm256_mul_ps(beta2,rsq02);
 418             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 419             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 420             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 421             felec            = _mm256_mul_ps(qq02,felec);
 422             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 423             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 424             velec            = _mm256_sub_ps(rinv02,pmecorrV);
 425             velec            = _mm256_mul_ps(qq02,velec);
 426
 427             /* Update potential sum for this i atom from the interaction with this j atom. */
 428             velecsum         = _mm256_add_ps(velecsum,velec);
 429
 430             fscal            = felec;
 431
 432             /* Calculate temporary vectorial force */
 433             tx               = _mm256_mul_ps(fscal,dx02);
 434             ty               = _mm256_mul_ps(fscal,dy02);
 435             tz               = _mm256_mul_ps(fscal,dz02);
 436
 437             /* Update vectorial force */
 438             fix0             = _mm256_add_ps(fix0,tx);
 439             fiy0             = _mm256_add_ps(fiy0,ty);
 440             fiz0             = _mm256_add_ps(fiz0,tz);
 441
 442             fjx2             = _mm256_add_ps(fjx2,tx);
 443             fjy2             = _mm256_add_ps(fjy2,ty);
 444             fjz2             = _mm256_add_ps(fjz2,tz);
 445
 446             /**************************
 447              * CALCULATE INTERACTIONS *
 448              **************************/
 449
 450             r10              = _mm256_mul_ps(rsq10,rinv10);
 451
 452             /* EWALD ELECTROSTATICS */
 453
 454             /* Analytical PME correction */
 455             zeta2            = _mm256_mul_ps(beta2,rsq10);
 456             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 457             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 458             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 459             felec            = _mm256_mul_ps(qq10,felec);
 460             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 461             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 462             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 463             velec            = _mm256_mul_ps(qq10,velec);
 464
 465             /* Update potential sum for this i atom from the interaction with this j atom. */
 466             velecsum         = _mm256_add_ps(velecsum,velec);
 467
 468             fscal            = felec;
 469
 470             /* Calculate temporary vectorial force */
 471             tx               = _mm256_mul_ps(fscal,dx10);
 472             ty               = _mm256_mul_ps(fscal,dy10);
 473             tz               = _mm256_mul_ps(fscal,dz10);
 474
 475             /* Update vectorial force */
 476             fix1             = _mm256_add_ps(fix1,tx);
 477             fiy1             = _mm256_add_ps(fiy1,ty);
 478             fiz1             = _mm256_add_ps(fiz1,tz);
 479
 480             fjx0             = _mm256_add_ps(fjx0,tx);
 481             fjy0             = _mm256_add_ps(fjy0,ty);
 482             fjz0             = _mm256_add_ps(fjz0,tz);
 483
 484             /**************************
 485              * CALCULATE INTERACTIONS *
 486              **************************/
 487
 488             r11              = _mm256_mul_ps(rsq11,rinv11);
 489
 490             /* EWALD ELECTROSTATICS */
 491
 492             /* Analytical PME correction */
 493             zeta2            = _mm256_mul_ps(beta2,rsq11);
 494             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 495             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 496             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 497             felec            = _mm256_mul_ps(qq11,felec);
 498             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 499             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 500             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 501             velec            = _mm256_mul_ps(qq11,velec);
 502
 503             /* Update potential sum for this i atom from the interaction with this j atom. */
 504             velecsum         = _mm256_add_ps(velecsum,velec);
 505
 506             fscal            = felec;
 507
 508             /* Calculate temporary vectorial force */
 509             tx               = _mm256_mul_ps(fscal,dx11);
 510             ty               = _mm256_mul_ps(fscal,dy11);
 511             tz               = _mm256_mul_ps(fscal,dz11);
 512
 513             /* Update vectorial force */
 514             fix1             = _mm256_add_ps(fix1,tx);
 515             fiy1             = _mm256_add_ps(fiy1,ty);
 516             fiz1             = _mm256_add_ps(fiz1,tz);
 517
 518             fjx1             = _mm256_add_ps(fjx1,tx);
 519             fjy1             = _mm256_add_ps(fjy1,ty);
 520             fjz1             = _mm256_add_ps(fjz1,tz);
 521
 522             /**************************
 523              * CALCULATE INTERACTIONS *
 524              **************************/
 525
 526             r12              = _mm256_mul_ps(rsq12,rinv12);
 527
 528             /* EWALD ELECTROSTATICS */
 529
 530             /* Analytical PME correction */
 531             zeta2            = _mm256_mul_ps(beta2,rsq12);
 532             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 533             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 534             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 535             felec            = _mm256_mul_ps(qq12,felec);
 536             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 537             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 538             velec            = _mm256_sub_ps(rinv12,pmecorrV);
 539             velec            = _mm256_mul_ps(qq12,velec);
 540
 541             /* Update potential sum for this i atom from the interaction with this j atom. */
 542             velecsum         = _mm256_add_ps(velecsum,velec);
 543
 544             fscal            = felec;
 545
 546             /* Calculate temporary vectorial force */
 547             tx               = _mm256_mul_ps(fscal,dx12);
 548             ty               = _mm256_mul_ps(fscal,dy12);
 549             tz               = _mm256_mul_ps(fscal,dz12);
 550
 551             /* Update vectorial force */
 552             fix1             = _mm256_add_ps(fix1,tx);
 553             fiy1             = _mm256_add_ps(fiy1,ty);
 554             fiz1             = _mm256_add_ps(fiz1,tz);
 555
 556             fjx2             = _mm256_add_ps(fjx2,tx);
 557             fjy2             = _mm256_add_ps(fjy2,ty);
 558             fjz2             = _mm256_add_ps(fjz2,tz);
 559
 560             /**************************
 561              * CALCULATE INTERACTIONS *
 562              **************************/
 563
 564             r20              = _mm256_mul_ps(rsq20,rinv20);
 565
 566             /* EWALD ELECTROSTATICS */
 567
 568             /* Analytical PME correction */
 569             zeta2            = _mm256_mul_ps(beta2,rsq20);
 570             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 571             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 572             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 573             felec            = _mm256_mul_ps(qq20,felec);
 574             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 575             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 576             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 577             velec            = _mm256_mul_ps(qq20,velec);
 578
 579             /* Update potential sum for this i atom from the interaction with this j atom. */
 580             velecsum         = _mm256_add_ps(velecsum,velec);
 581
 582             fscal            = felec;
 583
 584             /* Calculate temporary vectorial force */
 585             tx               = _mm256_mul_ps(fscal,dx20);
 586             ty               = _mm256_mul_ps(fscal,dy20);
 587             tz               = _mm256_mul_ps(fscal,dz20);
 588
 589             /* Update vectorial force */
 590             fix2             = _mm256_add_ps(fix2,tx);
 591             fiy2             = _mm256_add_ps(fiy2,ty);
 592             fiz2             = _mm256_add_ps(fiz2,tz);
 593
 594             fjx0             = _mm256_add_ps(fjx0,tx);
 595             fjy0             = _mm256_add_ps(fjy0,ty);
 596             fjz0             = _mm256_add_ps(fjz0,tz);
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             r21              = _mm256_mul_ps(rsq21,rinv21);
 603
 604             /* EWALD ELECTROSTATICS */
 605
 606             /* Analytical PME correction */
 607             zeta2            = _mm256_mul_ps(beta2,rsq21);
 608             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 609             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 610             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 611             felec            = _mm256_mul_ps(qq21,felec);
 612             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 613             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 614             velec            = _mm256_sub_ps(rinv21,pmecorrV);
 615             velec            = _mm256_mul_ps(qq21,velec);
 616
 617             /* Update potential sum for this i atom from the interaction with this j atom. */
 618             velecsum         = _mm256_add_ps(velecsum,velec);
 619
 620             fscal            = felec;
 621
 622             /* Calculate temporary vectorial force */
 623             tx               = _mm256_mul_ps(fscal,dx21);
 624             ty               = _mm256_mul_ps(fscal,dy21);
 625             tz               = _mm256_mul_ps(fscal,dz21);
 626
 627             /* Update vectorial force */
 628             fix2             = _mm256_add_ps(fix2,tx);
 629             fiy2             = _mm256_add_ps(fiy2,ty);
 630             fiz2             = _mm256_add_ps(fiz2,tz);
 631
 632             fjx1             = _mm256_add_ps(fjx1,tx);
 633             fjy1             = _mm256_add_ps(fjy1,ty);
 634             fjz1             = _mm256_add_ps(fjz1,tz);
 635
 636             /**************************
 637              * CALCULATE INTERACTIONS *
 638              **************************/
 639
 640             r22              = _mm256_mul_ps(rsq22,rinv22);
 641
 642             /* EWALD ELECTROSTATICS */
 643
 644             /* Analytical PME correction */
 645             zeta2            = _mm256_mul_ps(beta2,rsq22);
 646             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 647             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 648             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 649             felec            = _mm256_mul_ps(qq22,felec);
 650             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 651             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 652             velec            = _mm256_sub_ps(rinv22,pmecorrV);
 653             velec            = _mm256_mul_ps(qq22,velec);
 654
 655             /* Update potential sum for this i atom from the interaction with this j atom. */
 656             velecsum         = _mm256_add_ps(velecsum,velec);
 657
 658             fscal            = felec;
 659
 660             /* Calculate temporary vectorial force */
 661             tx               = _mm256_mul_ps(fscal,dx22);
 662             ty               = _mm256_mul_ps(fscal,dy22);
 663             tz               = _mm256_mul_ps(fscal,dz22);
 664
 665             /* Update vectorial force */
 666             fix2             = _mm256_add_ps(fix2,tx);
 667             fiy2             = _mm256_add_ps(fiy2,ty);
 668             fiz2             = _mm256_add_ps(fiz2,tz);
 669
 670             fjx2             = _mm256_add_ps(fjx2,tx);
 671             fjy2             = _mm256_add_ps(fjy2,ty);
 672             fjz2             = _mm256_add_ps(fjz2,tz);
 673
 674             fjptrA             = f+j_coord_offsetA;
 675             fjptrB             = f+j_coord_offsetB;
 676             fjptrC             = f+j_coord_offsetC;
 677             fjptrD             = f+j_coord_offsetD;
 678             fjptrE             = f+j_coord_offsetE;
 679             fjptrF             = f+j_coord_offsetF;
 680             fjptrG             = f+j_coord_offsetG;
 681             fjptrH             = f+j_coord_offsetH;
 682
 683             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 684                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 685
 686             /* Inner loop uses 768 flops */
 687         }
 688
 689         if(jidx<j_index_end)
 690         {
 691
 692             /* Get j neighbor index, and coordinate index */
 693             jnrlistA         = jjnr[jidx];
 694             jnrlistB         = jjnr[jidx+1];
 695             jnrlistC         = jjnr[jidx+2];
 696             jnrlistD         = jjnr[jidx+3];
 697             jnrlistE         = jjnr[jidx+4];
 698             jnrlistF         = jjnr[jidx+5];
 699             jnrlistG         = jjnr[jidx+6];
 700             jnrlistH         = jjnr[jidx+7];
 701             /* Sign of each element will be negative for non-real atoms.
 702              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 703              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 704              */
 705             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 706                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 707
 708             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 709             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 710             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 711             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 712             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 713             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 714             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 715             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 716             j_coord_offsetA  = DIM*jnrA;
 717             j_coord_offsetB  = DIM*jnrB;
 718             j_coord_offsetC  = DIM*jnrC;
 719             j_coord_offsetD  = DIM*jnrD;
 720             j_coord_offsetE  = DIM*jnrE;
 721             j_coord_offsetF  = DIM*jnrF;
 722             j_coord_offsetG  = DIM*jnrG;
 723             j_coord_offsetH  = DIM*jnrH;
 724
 725             /* load j atom coordinates */
 726             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 727                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 728                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 729                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 730                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 731
 732             /* Calculate displacement vector */
 733             dx00             = _mm256_sub_ps(ix0,jx0);
 734             dy00             = _mm256_sub_ps(iy0,jy0);
 735             dz00             = _mm256_sub_ps(iz0,jz0);
 736             dx01             = _mm256_sub_ps(ix0,jx1);
 737             dy01             = _mm256_sub_ps(iy0,jy1);
 738             dz01             = _mm256_sub_ps(iz0,jz1);
 739             dx02             = _mm256_sub_ps(ix0,jx2);
 740             dy02             = _mm256_sub_ps(iy0,jy2);
 741             dz02             = _mm256_sub_ps(iz0,jz2);
 742             dx10             = _mm256_sub_ps(ix1,jx0);
 743             dy10             = _mm256_sub_ps(iy1,jy0);
 744             dz10             = _mm256_sub_ps(iz1,jz0);
 745             dx11             = _mm256_sub_ps(ix1,jx1);
 746             dy11             = _mm256_sub_ps(iy1,jy1);
 747             dz11             = _mm256_sub_ps(iz1,jz1);
 748             dx12             = _mm256_sub_ps(ix1,jx2);
 749             dy12             = _mm256_sub_ps(iy1,jy2);
 750             dz12             = _mm256_sub_ps(iz1,jz2);
 751             dx20             = _mm256_sub_ps(ix2,jx0);
 752             dy20             = _mm256_sub_ps(iy2,jy0);
 753             dz20             = _mm256_sub_ps(iz2,jz0);
 754             dx21             = _mm256_sub_ps(ix2,jx1);
 755             dy21             = _mm256_sub_ps(iy2,jy1);
 756             dz21             = _mm256_sub_ps(iz2,jz1);
 757             dx22             = _mm256_sub_ps(ix2,jx2);
 758             dy22             = _mm256_sub_ps(iy2,jy2);
 759             dz22             = _mm256_sub_ps(iz2,jz2);
 760
 761             /* Calculate squared distance and things based on it */
 762             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 763             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 764             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 765             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 766             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 767             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 768             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 769             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 770             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 771
 772             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 773             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 774             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 775             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 776             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 777             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 778             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 779             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 780             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 781
 782             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 783             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 784             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 785             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 786             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 787             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 788             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 789             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 790             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 791
 792             fjx0             = _mm256_setzero_ps();
 793             fjy0             = _mm256_setzero_ps();
 794             fjz0             = _mm256_setzero_ps();
 795             fjx1             = _mm256_setzero_ps();
 796             fjy1             = _mm256_setzero_ps();
 797             fjz1             = _mm256_setzero_ps();
 798             fjx2             = _mm256_setzero_ps();
 799             fjy2             = _mm256_setzero_ps();
 800             fjz2             = _mm256_setzero_ps();
 801
 802             /**************************
 803              * CALCULATE INTERACTIONS *
 804              **************************/
 805
 806             r00              = _mm256_mul_ps(rsq00,rinv00);
 807             r00              = _mm256_andnot_ps(dummy_mask,r00);
 808
 809             /* EWALD ELECTROSTATICS */
 810
 811             /* Analytical PME correction */
 812             zeta2            = _mm256_mul_ps(beta2,rsq00);
 813             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 814             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 815             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 816             felec            = _mm256_mul_ps(qq00,felec);
 817             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 818             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 819             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 820             velec            = _mm256_mul_ps(qq00,velec);
 821
 822             /* LENNARD-JONES DISPERSION/REPULSION */
 823
 824             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 825             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 826             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 827             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 828             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 829
 830             /* Update potential sum for this i atom from the interaction with this j atom. */
 831             velec            = _mm256_andnot_ps(dummy_mask,velec);
 832             velecsum         = _mm256_add_ps(velecsum,velec);
 833             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 834             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 835
 836             fscal            = _mm256_add_ps(felec,fvdw);
 837
 838             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 839
 840             /* Calculate temporary vectorial force */
 841             tx               = _mm256_mul_ps(fscal,dx00);
 842             ty               = _mm256_mul_ps(fscal,dy00);
 843             tz               = _mm256_mul_ps(fscal,dz00);
 844
 845             /* Update vectorial force */
 846             fix0             = _mm256_add_ps(fix0,tx);
 847             fiy0             = _mm256_add_ps(fiy0,ty);
 848             fiz0             = _mm256_add_ps(fiz0,tz);
 849
 850             fjx0             = _mm256_add_ps(fjx0,tx);
 851             fjy0             = _mm256_add_ps(fjy0,ty);
 852             fjz0             = _mm256_add_ps(fjz0,tz);
 853
 854             /**************************
 855              * CALCULATE INTERACTIONS *
 856              **************************/
 857
 858             r01              = _mm256_mul_ps(rsq01,rinv01);
 859             r01              = _mm256_andnot_ps(dummy_mask,r01);
 860
 861             /* EWALD ELECTROSTATICS */
 862
 863             /* Analytical PME correction */
 864             zeta2            = _mm256_mul_ps(beta2,rsq01);
 865             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 866             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 867             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 868             felec            = _mm256_mul_ps(qq01,felec);
 869             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 870             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 871             velec            = _mm256_sub_ps(rinv01,pmecorrV);
 872             velec            = _mm256_mul_ps(qq01,velec);
 873
 874             /* Update potential sum for this i atom from the interaction with this j atom. */
 875             velec            = _mm256_andnot_ps(dummy_mask,velec);
 876             velecsum         = _mm256_add_ps(velecsum,velec);
 877
 878             fscal            = felec;
 879
 880             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 881
 882             /* Calculate temporary vectorial force */
 883             tx               = _mm256_mul_ps(fscal,dx01);
 884             ty               = _mm256_mul_ps(fscal,dy01);
 885             tz               = _mm256_mul_ps(fscal,dz01);
 886
 887             /* Update vectorial force */
 888             fix0             = _mm256_add_ps(fix0,tx);
 889             fiy0             = _mm256_add_ps(fiy0,ty);
 890             fiz0             = _mm256_add_ps(fiz0,tz);
 891
 892             fjx1             = _mm256_add_ps(fjx1,tx);
 893             fjy1             = _mm256_add_ps(fjy1,ty);
 894             fjz1             = _mm256_add_ps(fjz1,tz);
 895
 896             /**************************
 897              * CALCULATE INTERACTIONS *
 898              **************************/
 899
 900             r02              = _mm256_mul_ps(rsq02,rinv02);
 901             r02              = _mm256_andnot_ps(dummy_mask,r02);
 902
 903             /* EWALD ELECTROSTATICS */
 904
 905             /* Analytical PME correction */
 906             zeta2            = _mm256_mul_ps(beta2,rsq02);
 907             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 908             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 909             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 910             felec            = _mm256_mul_ps(qq02,felec);
 911             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 912             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 913             velec            = _mm256_sub_ps(rinv02,pmecorrV);
 914             velec            = _mm256_mul_ps(qq02,velec);
 915
 916             /* Update potential sum for this i atom from the interaction with this j atom. */
 917             velec            = _mm256_andnot_ps(dummy_mask,velec);
 918             velecsum         = _mm256_add_ps(velecsum,velec);
 919
 920             fscal            = felec;
 921
 922             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 923
 924             /* Calculate temporary vectorial force */
 925             tx               = _mm256_mul_ps(fscal,dx02);
 926             ty               = _mm256_mul_ps(fscal,dy02);
 927             tz               = _mm256_mul_ps(fscal,dz02);
 928
 929             /* Update vectorial force */
 930             fix0             = _mm256_add_ps(fix0,tx);
 931             fiy0             = _mm256_add_ps(fiy0,ty);
 932             fiz0             = _mm256_add_ps(fiz0,tz);
 933
 934             fjx2             = _mm256_add_ps(fjx2,tx);
 935             fjy2             = _mm256_add_ps(fjy2,ty);
 936             fjz2             = _mm256_add_ps(fjz2,tz);
 937
 938             /**************************
 939              * CALCULATE INTERACTIONS *
 940              **************************/
 941
 942             r10              = _mm256_mul_ps(rsq10,rinv10);
 943             r10              = _mm256_andnot_ps(dummy_mask,r10);
 944
 945             /* EWALD ELECTROSTATICS */
 946
 947             /* Analytical PME correction */
 948             zeta2            = _mm256_mul_ps(beta2,rsq10);
 949             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 950             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 951             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 952             felec            = _mm256_mul_ps(qq10,felec);
 953             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 954             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 955             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 956             velec            = _mm256_mul_ps(qq10,velec);
 957
 958             /* Update potential sum for this i atom from the interaction with this j atom. */
 959             velec            = _mm256_andnot_ps(dummy_mask,velec);
 960             velecsum         = _mm256_add_ps(velecsum,velec);
 961
 962             fscal            = felec;
 963
 964             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 965
 966             /* Calculate temporary vectorial force */
 967             tx               = _mm256_mul_ps(fscal,dx10);
 968             ty               = _mm256_mul_ps(fscal,dy10);
 969             tz               = _mm256_mul_ps(fscal,dz10);
 970
 971             /* Update vectorial force */
 972             fix1             = _mm256_add_ps(fix1,tx);
 973             fiy1             = _mm256_add_ps(fiy1,ty);
 974             fiz1             = _mm256_add_ps(fiz1,tz);
 975
 976             fjx0             = _mm256_add_ps(fjx0,tx);
 977             fjy0             = _mm256_add_ps(fjy0,ty);
 978             fjz0             = _mm256_add_ps(fjz0,tz);
 979
 980             /**************************
 981              * CALCULATE INTERACTIONS *
 982              **************************/
 983
 984             r11              = _mm256_mul_ps(rsq11,rinv11);
 985             r11              = _mm256_andnot_ps(dummy_mask,r11);
 986
 987             /* EWALD ELECTROSTATICS */
 988
 989             /* Analytical PME correction */
 990             zeta2            = _mm256_mul_ps(beta2,rsq11);
 991             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 992             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 993             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 994             felec            = _mm256_mul_ps(qq11,felec);
 995             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 996             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 997             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 998             velec            = _mm256_mul_ps(qq11,velec);
 999
1000             /* Update potential sum for this i atom from the interaction with this j atom. */
1001             velec            = _mm256_andnot_ps(dummy_mask,velec);
1002             velecsum         = _mm256_add_ps(velecsum,velec);
1003
1004             fscal            = felec;
1005
1006             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1007
1008             /* Calculate temporary vectorial force */
1009             tx               = _mm256_mul_ps(fscal,dx11);
1010             ty               = _mm256_mul_ps(fscal,dy11);
1011             tz               = _mm256_mul_ps(fscal,dz11);
1012
1013             /* Update vectorial force */
1014             fix1             = _mm256_add_ps(fix1,tx);
1015             fiy1             = _mm256_add_ps(fiy1,ty);
1016             fiz1             = _mm256_add_ps(fiz1,tz);
1017
1018             fjx1             = _mm256_add_ps(fjx1,tx);
1019             fjy1             = _mm256_add_ps(fjy1,ty);
1020             fjz1             = _mm256_add_ps(fjz1,tz);
1021
1022             /**************************
1023              * CALCULATE INTERACTIONS *
1024              **************************/
1025
1026             r12              = _mm256_mul_ps(rsq12,rinv12);
1027             r12              = _mm256_andnot_ps(dummy_mask,r12);
1028
1029             /* EWALD ELECTROSTATICS */
1030
1031             /* Analytical PME correction */
1032             zeta2            = _mm256_mul_ps(beta2,rsq12);
1033             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1034             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1035             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1036             felec            = _mm256_mul_ps(qq12,felec);
1037             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1038             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1039             velec            = _mm256_sub_ps(rinv12,pmecorrV);
1040             velec            = _mm256_mul_ps(qq12,velec);
1041
1042             /* Update potential sum for this i atom from the interaction with this j atom. */
1043             velec            = _mm256_andnot_ps(dummy_mask,velec);
1044             velecsum         = _mm256_add_ps(velecsum,velec);
1045
1046             fscal            = felec;
1047
1048             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1049
1050             /* Calculate temporary vectorial force */
1051             tx               = _mm256_mul_ps(fscal,dx12);
1052             ty               = _mm256_mul_ps(fscal,dy12);
1053             tz               = _mm256_mul_ps(fscal,dz12);
1054
1055             /* Update vectorial force */
1056             fix1             = _mm256_add_ps(fix1,tx);
1057             fiy1             = _mm256_add_ps(fiy1,ty);
1058             fiz1             = _mm256_add_ps(fiz1,tz);
1059
1060             fjx2             = _mm256_add_ps(fjx2,tx);
1061             fjy2             = _mm256_add_ps(fjy2,ty);
1062             fjz2             = _mm256_add_ps(fjz2,tz);
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             r20              = _mm256_mul_ps(rsq20,rinv20);
1069             r20              = _mm256_andnot_ps(dummy_mask,r20);
1070
1071             /* EWALD ELECTROSTATICS */
1072
1073             /* Analytical PME correction */
1074             zeta2            = _mm256_mul_ps(beta2,rsq20);
1075             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1076             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1077             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1078             felec            = _mm256_mul_ps(qq20,felec);
1079             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1080             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1081             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1082             velec            = _mm256_mul_ps(qq20,velec);
1083
1084             /* Update potential sum for this i atom from the interaction with this j atom. */
1085             velec            = _mm256_andnot_ps(dummy_mask,velec);
1086             velecsum         = _mm256_add_ps(velecsum,velec);
1087
1088             fscal            = felec;
1089
1090             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1091
1092             /* Calculate temporary vectorial force */
1093             tx               = _mm256_mul_ps(fscal,dx20);
1094             ty               = _mm256_mul_ps(fscal,dy20);
1095             tz               = _mm256_mul_ps(fscal,dz20);
1096
1097             /* Update vectorial force */
1098             fix2             = _mm256_add_ps(fix2,tx);
1099             fiy2             = _mm256_add_ps(fiy2,ty);
1100             fiz2             = _mm256_add_ps(fiz2,tz);
1101
1102             fjx0             = _mm256_add_ps(fjx0,tx);
1103             fjy0             = _mm256_add_ps(fjy0,ty);
1104             fjz0             = _mm256_add_ps(fjz0,tz);
1105
1106             /**************************
1107              * CALCULATE INTERACTIONS *
1108              **************************/
1109
1110             r21              = _mm256_mul_ps(rsq21,rinv21);
1111             r21              = _mm256_andnot_ps(dummy_mask,r21);
1112
1113             /* EWALD ELECTROSTATICS */
1114
1115             /* Analytical PME correction */
1116             zeta2            = _mm256_mul_ps(beta2,rsq21);
1117             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1118             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1119             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1120             felec            = _mm256_mul_ps(qq21,felec);
1121             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1122             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1123             velec            = _mm256_sub_ps(rinv21,pmecorrV);
1124             velec            = _mm256_mul_ps(qq21,velec);
1125
1126             /* Update potential sum for this i atom from the interaction with this j atom. */
1127             velec            = _mm256_andnot_ps(dummy_mask,velec);
1128             velecsum         = _mm256_add_ps(velecsum,velec);
1129
1130             fscal            = felec;
1131
1132             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1133
1134             /* Calculate temporary vectorial force */
1135             tx               = _mm256_mul_ps(fscal,dx21);
1136             ty               = _mm256_mul_ps(fscal,dy21);
1137             tz               = _mm256_mul_ps(fscal,dz21);
1138
1139             /* Update vectorial force */
1140             fix2             = _mm256_add_ps(fix2,tx);
1141             fiy2             = _mm256_add_ps(fiy2,ty);
1142             fiz2             = _mm256_add_ps(fiz2,tz);
1143
1144             fjx1             = _mm256_add_ps(fjx1,tx);
1145             fjy1             = _mm256_add_ps(fjy1,ty);
1146             fjz1             = _mm256_add_ps(fjz1,tz);
1147
1148             /**************************
1149              * CALCULATE INTERACTIONS *
1150              **************************/
1151
1152             r22              = _mm256_mul_ps(rsq22,rinv22);
1153             r22              = _mm256_andnot_ps(dummy_mask,r22);
1154
1155             /* EWALD ELECTROSTATICS */
1156
1157             /* Analytical PME correction */
1158             zeta2            = _mm256_mul_ps(beta2,rsq22);
1159             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1160             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1161             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1162             felec            = _mm256_mul_ps(qq22,felec);
1163             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1164             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1165             velec            = _mm256_sub_ps(rinv22,pmecorrV);
1166             velec            = _mm256_mul_ps(qq22,velec);
1167
1168             /* Update potential sum for this i atom from the interaction with this j atom. */
1169             velec            = _mm256_andnot_ps(dummy_mask,velec);
1170             velecsum         = _mm256_add_ps(velecsum,velec);
1171
1172             fscal            = felec;
1173
1174             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1175
1176             /* Calculate temporary vectorial force */
1177             tx               = _mm256_mul_ps(fscal,dx22);
1178             ty               = _mm256_mul_ps(fscal,dy22);
1179             tz               = _mm256_mul_ps(fscal,dz22);
1180
1181             /* Update vectorial force */
1182             fix2             = _mm256_add_ps(fix2,tx);
1183             fiy2             = _mm256_add_ps(fiy2,ty);
1184             fiz2             = _mm256_add_ps(fiz2,tz);
1185
1186             fjx2             = _mm256_add_ps(fjx2,tx);
1187             fjy2             = _mm256_add_ps(fjy2,ty);
1188             fjz2             = _mm256_add_ps(fjz2,tz);
1189
1190             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1191             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1192             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1193             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1194             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1195             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1196             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1197             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1198
1199             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1200                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1201
1202             /* Inner loop uses 777 flops */
1203         }
1204
1205         /* End of innermost loop */
1206
1207         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1208                                                  f+i_coord_offset,fshift+i_shift_offset);
1209
1210         ggid                        = gid[iidx];
1211         /* Update potential energies */
1212         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1213         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1214
1215         /* Increment number of inner iterations */
1216         inneriter                  += j_index_end - j_index_start;
1217
1218         /* Outer loop uses 20 flops */
1219     }
1220
1221     /* Increment number of outer iterations */
1222     outeriter        += nri;
1223
1224     /* Update outer/inner flops */
1225
1226     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*777);
1227 }
1228 /*
1229  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3W3_F_avx_256_single
1230  * Electrostatics interaction: Ewald
1231  * VdW interaction:            LennardJones
1232  * Geometry:                   Water3-Water3
1233  * Calculate force/pot:        Force
1234  */
1235 void
1236 nb_kernel_ElecEw_VdwLJ_GeomW3W3_F_avx_256_single
1237                     (t_nblist                    * gmx_restrict       nlist,
1238                      rvec                        * gmx_restrict          xx,
1239                      rvec                        * gmx_restrict          ff,
1240                      t_forcerec                  * gmx_restrict          fr,
1241                      t_mdatoms                   * gmx_restrict     mdatoms,
1242                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1243                      t_nrnb                      * gmx_restrict        nrnb)
1244 {
1245     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1246      * just 0 for non-waters.
1247      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1248      * jnr indices corresponding to data put in the four positions in the SIMD register.
1249      */
1250     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1251     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1252     int              jnrA,jnrB,jnrC,jnrD;
1253     int              jnrE,jnrF,jnrG,jnrH;
1254     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1255     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1256     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1257     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1258     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1259     real             rcutoff_scalar;
1260     real             *shiftvec,*fshift,*x,*f;
1261     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1262     real             scratch[4*DIM];
1263     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1264     real *           vdwioffsetptr0;
1265     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1266     real *           vdwioffsetptr1;
1267     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1268     real *           vdwioffsetptr2;
1269     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1270     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1271     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1272     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1273     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1274     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1275     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1276     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1277     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1278     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1279     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1280     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1281     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1282     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1283     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1284     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1285     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1286     real             *charge;
1287     int              nvdwtype;
1288     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1289     int              *vdwtype;
1290     real             *vdwparam;
1291     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1292     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1293     __m256i          ewitab;
1294     __m128i          ewitab_lo,ewitab_hi;
1295     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1296     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1297     real             *ewtab;
1298     __m256           dummy_mask,cutoff_mask;
1299     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1300     __m256           one     = _mm256_set1_ps(1.0);
1301     __m256           two     = _mm256_set1_ps(2.0);
1302     x                = xx[0];
1303     f                = ff[0];
1304
1305     nri              = nlist->nri;
1306     iinr             = nlist->iinr;
1307     jindex           = nlist->jindex;
1308     jjnr             = nlist->jjnr;
1309     shiftidx         = nlist->shift;
1310     gid              = nlist->gid;
1311     shiftvec         = fr->shift_vec[0];
1312     fshift           = fr->fshift[0];
1313     facel            = _mm256_set1_ps(fr->epsfac);
1314     charge           = mdatoms->chargeA;
1315     nvdwtype         = fr->ntype;
1316     vdwparam         = fr->nbfp;
1317     vdwtype          = mdatoms->typeA;
1318
1319     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1320     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1321     beta2            = _mm256_mul_ps(beta,beta);
1322     beta3            = _mm256_mul_ps(beta,beta2);
1323
1324     ewtab            = fr->ic->tabq_coul_F;
1325     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1326     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1327
1328     /* Setup water-specific parameters */
1329     inr              = nlist->iinr[0];
1330     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1331     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1332     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1333     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1334
1335     jq0              = _mm256_set1_ps(charge[inr+0]);
1336     jq1              = _mm256_set1_ps(charge[inr+1]);
1337     jq2              = _mm256_set1_ps(charge[inr+2]);
1338     vdwjidx0A        = 2*vdwtype[inr+0];
1339     qq00             = _mm256_mul_ps(iq0,jq0);
1340     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1341     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1342     qq01             = _mm256_mul_ps(iq0,jq1);
1343     qq02             = _mm256_mul_ps(iq0,jq2);
1344     qq10             = _mm256_mul_ps(iq1,jq0);
1345     qq11             = _mm256_mul_ps(iq1,jq1);
1346     qq12             = _mm256_mul_ps(iq1,jq2);
1347     qq20             = _mm256_mul_ps(iq2,jq0);
1348     qq21             = _mm256_mul_ps(iq2,jq1);
1349     qq22             = _mm256_mul_ps(iq2,jq2);
1350
1351     /* Avoid stupid compiler warnings */
1352     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1353     j_coord_offsetA = 0;
1354     j_coord_offsetB = 0;
1355     j_coord_offsetC = 0;
1356     j_coord_offsetD = 0;
1357     j_coord_offsetE = 0;
1358     j_coord_offsetF = 0;
1359     j_coord_offsetG = 0;
1360     j_coord_offsetH = 0;
1361
1362     outeriter        = 0;
1363     inneriter        = 0;
1364
1365     for(iidx=0;iidx<4*DIM;iidx++)
1366     {
1367         scratch[iidx] = 0.0;
1368     }
1369
1370     /* Start outer loop over neighborlists */
1371     for(iidx=0; iidx<nri; iidx++)
1372     {
1373         /* Load shift vector for this list */
1374         i_shift_offset   = DIM*shiftidx[iidx];
1375
1376         /* Load limits for loop over neighbors */
1377         j_index_start    = jindex[iidx];
1378         j_index_end      = jindex[iidx+1];
1379
1380         /* Get outer coordinate index */
1381         inr              = iinr[iidx];
1382         i_coord_offset   = DIM*inr;
1383
1384         /* Load i particle coords and add shift vector */
1385         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1386                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1387
1388         fix0             = _mm256_setzero_ps();
1389         fiy0             = _mm256_setzero_ps();
1390         fiz0             = _mm256_setzero_ps();
1391         fix1             = _mm256_setzero_ps();
1392         fiy1             = _mm256_setzero_ps();
1393         fiz1             = _mm256_setzero_ps();
1394         fix2             = _mm256_setzero_ps();
1395         fiy2             = _mm256_setzero_ps();
1396         fiz2             = _mm256_setzero_ps();
1397
1398         /* Start inner kernel loop */
1399         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1400         {
1401
1402             /* Get j neighbor index, and coordinate index */
1403             jnrA             = jjnr[jidx];
1404             jnrB             = jjnr[jidx+1];
1405             jnrC             = jjnr[jidx+2];
1406             jnrD             = jjnr[jidx+3];
1407             jnrE             = jjnr[jidx+4];
1408             jnrF             = jjnr[jidx+5];
1409             jnrG             = jjnr[jidx+6];
1410             jnrH             = jjnr[jidx+7];
1411             j_coord_offsetA  = DIM*jnrA;
1412             j_coord_offsetB  = DIM*jnrB;
1413             j_coord_offsetC  = DIM*jnrC;
1414             j_coord_offsetD  = DIM*jnrD;
1415             j_coord_offsetE  = DIM*jnrE;
1416             j_coord_offsetF  = DIM*jnrF;
1417             j_coord_offsetG  = DIM*jnrG;
1418             j_coord_offsetH  = DIM*jnrH;
1419
1420             /* load j atom coordinates */
1421             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1422                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1423                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1424                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1425                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1426
1427             /* Calculate displacement vector */
1428             dx00             = _mm256_sub_ps(ix0,jx0);
1429             dy00             = _mm256_sub_ps(iy0,jy0);
1430             dz00             = _mm256_sub_ps(iz0,jz0);
1431             dx01             = _mm256_sub_ps(ix0,jx1);
1432             dy01             = _mm256_sub_ps(iy0,jy1);
1433             dz01             = _mm256_sub_ps(iz0,jz1);
1434             dx02             = _mm256_sub_ps(ix0,jx2);
1435             dy02             = _mm256_sub_ps(iy0,jy2);
1436             dz02             = _mm256_sub_ps(iz0,jz2);
1437             dx10             = _mm256_sub_ps(ix1,jx0);
1438             dy10             = _mm256_sub_ps(iy1,jy0);
1439             dz10             = _mm256_sub_ps(iz1,jz0);
1440             dx11             = _mm256_sub_ps(ix1,jx1);
1441             dy11             = _mm256_sub_ps(iy1,jy1);
1442             dz11             = _mm256_sub_ps(iz1,jz1);
1443             dx12             = _mm256_sub_ps(ix1,jx2);
1444             dy12             = _mm256_sub_ps(iy1,jy2);
1445             dz12             = _mm256_sub_ps(iz1,jz2);
1446             dx20             = _mm256_sub_ps(ix2,jx0);
1447             dy20             = _mm256_sub_ps(iy2,jy0);
1448             dz20             = _mm256_sub_ps(iz2,jz0);
1449             dx21             = _mm256_sub_ps(ix2,jx1);
1450             dy21             = _mm256_sub_ps(iy2,jy1);
1451             dz21             = _mm256_sub_ps(iz2,jz1);
1452             dx22             = _mm256_sub_ps(ix2,jx2);
1453             dy22             = _mm256_sub_ps(iy2,jy2);
1454             dz22             = _mm256_sub_ps(iz2,jz2);
1455
1456             /* Calculate squared distance and things based on it */
1457             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1458             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1459             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1460             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1461             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1462             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1463             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1464             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1465             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1466
1467             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1468             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1469             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1470             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1471             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1472             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1473             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1474             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1475             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1476
1477             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1478             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1479             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1480             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1481             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1482             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1483             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1484             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1485             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1486
1487             fjx0             = _mm256_setzero_ps();
1488             fjy0             = _mm256_setzero_ps();
1489             fjz0             = _mm256_setzero_ps();
1490             fjx1             = _mm256_setzero_ps();
1491             fjy1             = _mm256_setzero_ps();
1492             fjz1             = _mm256_setzero_ps();
1493             fjx2             = _mm256_setzero_ps();
1494             fjy2             = _mm256_setzero_ps();
1495             fjz2             = _mm256_setzero_ps();
1496
1497             /**************************
1498              * CALCULATE INTERACTIONS *
1499              **************************/
1500
1501             r00              = _mm256_mul_ps(rsq00,rinv00);
1502
1503             /* EWALD ELECTROSTATICS */
1504
1505             /* Analytical PME correction */
1506             zeta2            = _mm256_mul_ps(beta2,rsq00);
1507             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1508             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1509             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1510             felec            = _mm256_mul_ps(qq00,felec);
1511
1512             /* LENNARD-JONES DISPERSION/REPULSION */
1513
1514             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1515             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1516
1517             fscal            = _mm256_add_ps(felec,fvdw);
1518
1519             /* Calculate temporary vectorial force */
1520             tx               = _mm256_mul_ps(fscal,dx00);
1521             ty               = _mm256_mul_ps(fscal,dy00);
1522             tz               = _mm256_mul_ps(fscal,dz00);
1523
1524             /* Update vectorial force */
1525             fix0             = _mm256_add_ps(fix0,tx);
1526             fiy0             = _mm256_add_ps(fiy0,ty);
1527             fiz0             = _mm256_add_ps(fiz0,tz);
1528
1529             fjx0             = _mm256_add_ps(fjx0,tx);
1530             fjy0             = _mm256_add_ps(fjy0,ty);
1531             fjz0             = _mm256_add_ps(fjz0,tz);
1532
1533             /**************************
1534              * CALCULATE INTERACTIONS *
1535              **************************/
1536
1537             r01              = _mm256_mul_ps(rsq01,rinv01);
1538
1539             /* EWALD ELECTROSTATICS */
1540
1541             /* Analytical PME correction */
1542             zeta2            = _mm256_mul_ps(beta2,rsq01);
1543             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
1544             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1545             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1546             felec            = _mm256_mul_ps(qq01,felec);
1547
1548             fscal            = felec;
1549
1550             /* Calculate temporary vectorial force */
1551             tx               = _mm256_mul_ps(fscal,dx01);
1552             ty               = _mm256_mul_ps(fscal,dy01);
1553             tz               = _mm256_mul_ps(fscal,dz01);
1554
1555             /* Update vectorial force */
1556             fix0             = _mm256_add_ps(fix0,tx);
1557             fiy0             = _mm256_add_ps(fiy0,ty);
1558             fiz0             = _mm256_add_ps(fiz0,tz);
1559
1560             fjx1             = _mm256_add_ps(fjx1,tx);
1561             fjy1             = _mm256_add_ps(fjy1,ty);
1562             fjz1             = _mm256_add_ps(fjz1,tz);
1563
1564             /**************************
1565              * CALCULATE INTERACTIONS *
1566              **************************/
1567
1568             r02              = _mm256_mul_ps(rsq02,rinv02);
1569
1570             /* EWALD ELECTROSTATICS */
1571
1572             /* Analytical PME correction */
1573             zeta2            = _mm256_mul_ps(beta2,rsq02);
1574             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
1575             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1576             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1577             felec            = _mm256_mul_ps(qq02,felec);
1578
1579             fscal            = felec;
1580
1581             /* Calculate temporary vectorial force */
1582             tx               = _mm256_mul_ps(fscal,dx02);
1583             ty               = _mm256_mul_ps(fscal,dy02);
1584             tz               = _mm256_mul_ps(fscal,dz02);
1585
1586             /* Update vectorial force */
1587             fix0             = _mm256_add_ps(fix0,tx);
1588             fiy0             = _mm256_add_ps(fiy0,ty);
1589             fiz0             = _mm256_add_ps(fiz0,tz);
1590
1591             fjx2             = _mm256_add_ps(fjx2,tx);
1592             fjy2             = _mm256_add_ps(fjy2,ty);
1593             fjz2             = _mm256_add_ps(fjz2,tz);
1594
1595             /**************************
1596              * CALCULATE INTERACTIONS *
1597              **************************/
1598
1599             r10              = _mm256_mul_ps(rsq10,rinv10);
1600
1601             /* EWALD ELECTROSTATICS */
1602
1603             /* Analytical PME correction */
1604             zeta2            = _mm256_mul_ps(beta2,rsq10);
1605             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1606             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1607             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1608             felec            = _mm256_mul_ps(qq10,felec);
1609
1610             fscal            = felec;
1611
1612             /* Calculate temporary vectorial force */
1613             tx               = _mm256_mul_ps(fscal,dx10);
1614             ty               = _mm256_mul_ps(fscal,dy10);
1615             tz               = _mm256_mul_ps(fscal,dz10);
1616
1617             /* Update vectorial force */
1618             fix1             = _mm256_add_ps(fix1,tx);
1619             fiy1             = _mm256_add_ps(fiy1,ty);
1620             fiz1             = _mm256_add_ps(fiz1,tz);
1621
1622             fjx0             = _mm256_add_ps(fjx0,tx);
1623             fjy0             = _mm256_add_ps(fjy0,ty);
1624             fjz0             = _mm256_add_ps(fjz0,tz);
1625
1626             /**************************
1627              * CALCULATE INTERACTIONS *
1628              **************************/
1629
1630             r11              = _mm256_mul_ps(rsq11,rinv11);
1631
1632             /* EWALD ELECTROSTATICS */
1633
1634             /* Analytical PME correction */
1635             zeta2            = _mm256_mul_ps(beta2,rsq11);
1636             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1637             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1638             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1639             felec            = _mm256_mul_ps(qq11,felec);
1640
1641             fscal            = felec;
1642
1643             /* Calculate temporary vectorial force */
1644             tx               = _mm256_mul_ps(fscal,dx11);
1645             ty               = _mm256_mul_ps(fscal,dy11);
1646             tz               = _mm256_mul_ps(fscal,dz11);
1647
1648             /* Update vectorial force */
1649             fix1             = _mm256_add_ps(fix1,tx);
1650             fiy1             = _mm256_add_ps(fiy1,ty);
1651             fiz1             = _mm256_add_ps(fiz1,tz);
1652
1653             fjx1             = _mm256_add_ps(fjx1,tx);
1654             fjy1             = _mm256_add_ps(fjy1,ty);
1655             fjz1             = _mm256_add_ps(fjz1,tz);
1656
1657             /**************************
1658              * CALCULATE INTERACTIONS *
1659              **************************/
1660
1661             r12              = _mm256_mul_ps(rsq12,rinv12);
1662
1663             /* EWALD ELECTROSTATICS */
1664
1665             /* Analytical PME correction */
1666             zeta2            = _mm256_mul_ps(beta2,rsq12);
1667             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1668             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1669             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1670             felec            = _mm256_mul_ps(qq12,felec);
1671
1672             fscal            = felec;
1673
1674             /* Calculate temporary vectorial force */
1675             tx               = _mm256_mul_ps(fscal,dx12);
1676             ty               = _mm256_mul_ps(fscal,dy12);
1677             tz               = _mm256_mul_ps(fscal,dz12);
1678
1679             /* Update vectorial force */
1680             fix1             = _mm256_add_ps(fix1,tx);
1681             fiy1             = _mm256_add_ps(fiy1,ty);
1682             fiz1             = _mm256_add_ps(fiz1,tz);
1683
1684             fjx2             = _mm256_add_ps(fjx2,tx);
1685             fjy2             = _mm256_add_ps(fjy2,ty);
1686             fjz2             = _mm256_add_ps(fjz2,tz);
1687
1688             /**************************
1689              * CALCULATE INTERACTIONS *
1690              **************************/
1691
1692             r20              = _mm256_mul_ps(rsq20,rinv20);
1693
1694             /* EWALD ELECTROSTATICS */
1695
1696             /* Analytical PME correction */
1697             zeta2            = _mm256_mul_ps(beta2,rsq20);
1698             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1699             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1700             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1701             felec            = _mm256_mul_ps(qq20,felec);
1702
1703             fscal            = felec;
1704
1705             /* Calculate temporary vectorial force */
1706             tx               = _mm256_mul_ps(fscal,dx20);
1707             ty               = _mm256_mul_ps(fscal,dy20);
1708             tz               = _mm256_mul_ps(fscal,dz20);
1709
1710             /* Update vectorial force */
1711             fix2             = _mm256_add_ps(fix2,tx);
1712             fiy2             = _mm256_add_ps(fiy2,ty);
1713             fiz2             = _mm256_add_ps(fiz2,tz);
1714
1715             fjx0             = _mm256_add_ps(fjx0,tx);
1716             fjy0             = _mm256_add_ps(fjy0,ty);
1717             fjz0             = _mm256_add_ps(fjz0,tz);
1718
1719             /**************************
1720              * CALCULATE INTERACTIONS *
1721              **************************/
1722
1723             r21              = _mm256_mul_ps(rsq21,rinv21);
1724
1725             /* EWALD ELECTROSTATICS */
1726
1727             /* Analytical PME correction */
1728             zeta2            = _mm256_mul_ps(beta2,rsq21);
1729             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1730             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1731             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1732             felec            = _mm256_mul_ps(qq21,felec);
1733
1734             fscal            = felec;
1735
1736             /* Calculate temporary vectorial force */
1737             tx               = _mm256_mul_ps(fscal,dx21);
1738             ty               = _mm256_mul_ps(fscal,dy21);
1739             tz               = _mm256_mul_ps(fscal,dz21);
1740
1741             /* Update vectorial force */
1742             fix2             = _mm256_add_ps(fix2,tx);
1743             fiy2             = _mm256_add_ps(fiy2,ty);
1744             fiz2             = _mm256_add_ps(fiz2,tz);
1745
1746             fjx1             = _mm256_add_ps(fjx1,tx);
1747             fjy1             = _mm256_add_ps(fjy1,ty);
1748             fjz1             = _mm256_add_ps(fjz1,tz);
1749
1750             /**************************
1751              * CALCULATE INTERACTIONS *
1752              **************************/
1753
1754             r22              = _mm256_mul_ps(rsq22,rinv22);
1755
1756             /* EWALD ELECTROSTATICS */
1757
1758             /* Analytical PME correction */
1759             zeta2            = _mm256_mul_ps(beta2,rsq22);
1760             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1761             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1762             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1763             felec            = _mm256_mul_ps(qq22,felec);
1764
1765             fscal            = felec;
1766
1767             /* Calculate temporary vectorial force */
1768             tx               = _mm256_mul_ps(fscal,dx22);
1769             ty               = _mm256_mul_ps(fscal,dy22);
1770             tz               = _mm256_mul_ps(fscal,dz22);
1771
1772             /* Update vectorial force */
1773             fix2             = _mm256_add_ps(fix2,tx);
1774             fiy2             = _mm256_add_ps(fiy2,ty);
1775             fiz2             = _mm256_add_ps(fiz2,tz);
1776
1777             fjx2             = _mm256_add_ps(fjx2,tx);
1778             fjy2             = _mm256_add_ps(fjy2,ty);
1779             fjz2             = _mm256_add_ps(fjz2,tz);
1780
1781             fjptrA             = f+j_coord_offsetA;
1782             fjptrB             = f+j_coord_offsetB;
1783             fjptrC             = f+j_coord_offsetC;
1784             fjptrD             = f+j_coord_offsetD;
1785             fjptrE             = f+j_coord_offsetE;
1786             fjptrF             = f+j_coord_offsetF;
1787             fjptrG             = f+j_coord_offsetG;
1788             fjptrH             = f+j_coord_offsetH;
1789
1790             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1791                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1792
1793             /* Inner loop uses 511 flops */
1794         }
1795
1796         if(jidx<j_index_end)
1797         {
1798
1799             /* Get j neighbor index, and coordinate index */
1800             jnrlistA         = jjnr[jidx];
1801             jnrlistB         = jjnr[jidx+1];
1802             jnrlistC         = jjnr[jidx+2];
1803             jnrlistD         = jjnr[jidx+3];
1804             jnrlistE         = jjnr[jidx+4];
1805             jnrlistF         = jjnr[jidx+5];
1806             jnrlistG         = jjnr[jidx+6];
1807             jnrlistH         = jjnr[jidx+7];
1808             /* Sign of each element will be negative for non-real atoms.
1809              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1810              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1811              */
1812             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1813                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1814
1815             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1816             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1817             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1818             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1819             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1820             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1821             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1822             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1823             j_coord_offsetA  = DIM*jnrA;
1824             j_coord_offsetB  = DIM*jnrB;
1825             j_coord_offsetC  = DIM*jnrC;
1826             j_coord_offsetD  = DIM*jnrD;
1827             j_coord_offsetE  = DIM*jnrE;
1828             j_coord_offsetF  = DIM*jnrF;
1829             j_coord_offsetG  = DIM*jnrG;
1830             j_coord_offsetH  = DIM*jnrH;
1831
1832             /* load j atom coordinates */
1833             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1834                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1835                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1836                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1837                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1838
1839             /* Calculate displacement vector */
1840             dx00             = _mm256_sub_ps(ix0,jx0);
1841             dy00             = _mm256_sub_ps(iy0,jy0);
1842             dz00             = _mm256_sub_ps(iz0,jz0);
1843             dx01             = _mm256_sub_ps(ix0,jx1);
1844             dy01             = _mm256_sub_ps(iy0,jy1);
1845             dz01             = _mm256_sub_ps(iz0,jz1);
1846             dx02             = _mm256_sub_ps(ix0,jx2);
1847             dy02             = _mm256_sub_ps(iy0,jy2);
1848             dz02             = _mm256_sub_ps(iz0,jz2);
1849             dx10             = _mm256_sub_ps(ix1,jx0);
1850             dy10             = _mm256_sub_ps(iy1,jy0);
1851             dz10             = _mm256_sub_ps(iz1,jz0);
1852             dx11             = _mm256_sub_ps(ix1,jx1);
1853             dy11             = _mm256_sub_ps(iy1,jy1);
1854             dz11             = _mm256_sub_ps(iz1,jz1);
1855             dx12             = _mm256_sub_ps(ix1,jx2);
1856             dy12             = _mm256_sub_ps(iy1,jy2);
1857             dz12             = _mm256_sub_ps(iz1,jz2);
1858             dx20             = _mm256_sub_ps(ix2,jx0);
1859             dy20             = _mm256_sub_ps(iy2,jy0);
1860             dz20             = _mm256_sub_ps(iz2,jz0);
1861             dx21             = _mm256_sub_ps(ix2,jx1);
1862             dy21             = _mm256_sub_ps(iy2,jy1);
1863             dz21             = _mm256_sub_ps(iz2,jz1);
1864             dx22             = _mm256_sub_ps(ix2,jx2);
1865             dy22             = _mm256_sub_ps(iy2,jy2);
1866             dz22             = _mm256_sub_ps(iz2,jz2);
1867
1868             /* Calculate squared distance and things based on it */
1869             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1870             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1871             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1872             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1873             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1874             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1875             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1876             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1877             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1878
1879             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1880             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1881             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1882             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1883             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1884             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1885             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1886             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1887             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1888
1889             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1890             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1891             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1892             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1893             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1894             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1895             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1896             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1897             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1898
1899             fjx0             = _mm256_setzero_ps();
1900             fjy0             = _mm256_setzero_ps();
1901             fjz0             = _mm256_setzero_ps();
1902             fjx1             = _mm256_setzero_ps();
1903             fjy1             = _mm256_setzero_ps();
1904             fjz1             = _mm256_setzero_ps();
1905             fjx2             = _mm256_setzero_ps();
1906             fjy2             = _mm256_setzero_ps();
1907             fjz2             = _mm256_setzero_ps();
1908
1909             /**************************
1910              * CALCULATE INTERACTIONS *
1911              **************************/
1912
1913             r00              = _mm256_mul_ps(rsq00,rinv00);
1914             r00              = _mm256_andnot_ps(dummy_mask,r00);
1915
1916             /* EWALD ELECTROSTATICS */
1917
1918             /* Analytical PME correction */
1919             zeta2            = _mm256_mul_ps(beta2,rsq00);
1920             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1921             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1922             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1923             felec            = _mm256_mul_ps(qq00,felec);
1924
1925             /* LENNARD-JONES DISPERSION/REPULSION */
1926
1927             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1928             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1929
1930             fscal            = _mm256_add_ps(felec,fvdw);
1931
1932             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1933
1934             /* Calculate temporary vectorial force */
1935             tx               = _mm256_mul_ps(fscal,dx00);
1936             ty               = _mm256_mul_ps(fscal,dy00);
1937             tz               = _mm256_mul_ps(fscal,dz00);
1938
1939             /* Update vectorial force */
1940             fix0             = _mm256_add_ps(fix0,tx);
1941             fiy0             = _mm256_add_ps(fiy0,ty);
1942             fiz0             = _mm256_add_ps(fiz0,tz);
1943
1944             fjx0             = _mm256_add_ps(fjx0,tx);
1945             fjy0             = _mm256_add_ps(fjy0,ty);
1946             fjz0             = _mm256_add_ps(fjz0,tz);
1947
1948             /**************************
1949              * CALCULATE INTERACTIONS *
1950              **************************/
1951
1952             r01              = _mm256_mul_ps(rsq01,rinv01);
1953             r01              = _mm256_andnot_ps(dummy_mask,r01);
1954
1955             /* EWALD ELECTROSTATICS */
1956
1957             /* Analytical PME correction */
1958             zeta2            = _mm256_mul_ps(beta2,rsq01);
1959             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
1960             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1961             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1962             felec            = _mm256_mul_ps(qq01,felec);
1963
1964             fscal            = felec;
1965
1966             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1967
1968             /* Calculate temporary vectorial force */
1969             tx               = _mm256_mul_ps(fscal,dx01);
1970             ty               = _mm256_mul_ps(fscal,dy01);
1971             tz               = _mm256_mul_ps(fscal,dz01);
1972
1973             /* Update vectorial force */
1974             fix0             = _mm256_add_ps(fix0,tx);
1975             fiy0             = _mm256_add_ps(fiy0,ty);
1976             fiz0             = _mm256_add_ps(fiz0,tz);
1977
1978             fjx1             = _mm256_add_ps(fjx1,tx);
1979             fjy1             = _mm256_add_ps(fjy1,ty);
1980             fjz1             = _mm256_add_ps(fjz1,tz);
1981
1982             /**************************
1983              * CALCULATE INTERACTIONS *
1984              **************************/
1985
1986             r02              = _mm256_mul_ps(rsq02,rinv02);
1987             r02              = _mm256_andnot_ps(dummy_mask,r02);
1988
1989             /* EWALD ELECTROSTATICS */
1990
1991             /* Analytical PME correction */
1992             zeta2            = _mm256_mul_ps(beta2,rsq02);
1993             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
1994             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1995             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1996             felec            = _mm256_mul_ps(qq02,felec);
1997
1998             fscal            = felec;
1999
2000             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2001
2002             /* Calculate temporary vectorial force */
2003             tx               = _mm256_mul_ps(fscal,dx02);
2004             ty               = _mm256_mul_ps(fscal,dy02);
2005             tz               = _mm256_mul_ps(fscal,dz02);
2006
2007             /* Update vectorial force */
2008             fix0             = _mm256_add_ps(fix0,tx);
2009             fiy0             = _mm256_add_ps(fiy0,ty);
2010             fiz0             = _mm256_add_ps(fiz0,tz);
2011
2012             fjx2             = _mm256_add_ps(fjx2,tx);
2013             fjy2             = _mm256_add_ps(fjy2,ty);
2014             fjz2             = _mm256_add_ps(fjz2,tz);
2015
2016             /**************************
2017              * CALCULATE INTERACTIONS *
2018              **************************/
2019
2020             r10              = _mm256_mul_ps(rsq10,rinv10);
2021             r10              = _mm256_andnot_ps(dummy_mask,r10);
2022
2023             /* EWALD ELECTROSTATICS */
2024
2025             /* Analytical PME correction */
2026             zeta2            = _mm256_mul_ps(beta2,rsq10);
2027             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
2028             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2029             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2030             felec            = _mm256_mul_ps(qq10,felec);
2031
2032             fscal            = felec;
2033
2034             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2035
2036             /* Calculate temporary vectorial force */
2037             tx               = _mm256_mul_ps(fscal,dx10);
2038             ty               = _mm256_mul_ps(fscal,dy10);
2039             tz               = _mm256_mul_ps(fscal,dz10);
2040
2041             /* Update vectorial force */
2042             fix1             = _mm256_add_ps(fix1,tx);
2043             fiy1             = _mm256_add_ps(fiy1,ty);
2044             fiz1             = _mm256_add_ps(fiz1,tz);
2045
2046             fjx0             = _mm256_add_ps(fjx0,tx);
2047             fjy0             = _mm256_add_ps(fjy0,ty);
2048             fjz0             = _mm256_add_ps(fjz0,tz);
2049
2050             /**************************
2051              * CALCULATE INTERACTIONS *
2052              **************************/
2053
2054             r11              = _mm256_mul_ps(rsq11,rinv11);
2055             r11              = _mm256_andnot_ps(dummy_mask,r11);
2056
2057             /* EWALD ELECTROSTATICS */
2058
2059             /* Analytical PME correction */
2060             zeta2            = _mm256_mul_ps(beta2,rsq11);
2061             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
2062             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2063             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2064             felec            = _mm256_mul_ps(qq11,felec);
2065
2066             fscal            = felec;
2067
2068             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2069
2070             /* Calculate temporary vectorial force */
2071             tx               = _mm256_mul_ps(fscal,dx11);
2072             ty               = _mm256_mul_ps(fscal,dy11);
2073             tz               = _mm256_mul_ps(fscal,dz11);
2074
2075             /* Update vectorial force */
2076             fix1             = _mm256_add_ps(fix1,tx);
2077             fiy1             = _mm256_add_ps(fiy1,ty);
2078             fiz1             = _mm256_add_ps(fiz1,tz);
2079
2080             fjx1             = _mm256_add_ps(fjx1,tx);
2081             fjy1             = _mm256_add_ps(fjy1,ty);
2082             fjz1             = _mm256_add_ps(fjz1,tz);
2083
2084             /**************************
2085              * CALCULATE INTERACTIONS *
2086              **************************/
2087
2088             r12              = _mm256_mul_ps(rsq12,rinv12);
2089             r12              = _mm256_andnot_ps(dummy_mask,r12);
2090
2091             /* EWALD ELECTROSTATICS */
2092
2093             /* Analytical PME correction */
2094             zeta2            = _mm256_mul_ps(beta2,rsq12);
2095             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2096             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2097             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2098             felec            = _mm256_mul_ps(qq12,felec);
2099
2100             fscal            = felec;
2101
2102             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2103
2104             /* Calculate temporary vectorial force */
2105             tx               = _mm256_mul_ps(fscal,dx12);
2106             ty               = _mm256_mul_ps(fscal,dy12);
2107             tz               = _mm256_mul_ps(fscal,dz12);
2108
2109             /* Update vectorial force */
2110             fix1             = _mm256_add_ps(fix1,tx);
2111             fiy1             = _mm256_add_ps(fiy1,ty);
2112             fiz1             = _mm256_add_ps(fiz1,tz);
2113
2114             fjx2             = _mm256_add_ps(fjx2,tx);
2115             fjy2             = _mm256_add_ps(fjy2,ty);
2116             fjz2             = _mm256_add_ps(fjz2,tz);
2117
2118             /**************************
2119              * CALCULATE INTERACTIONS *
2120              **************************/
2121
2122             r20              = _mm256_mul_ps(rsq20,rinv20);
2123             r20              = _mm256_andnot_ps(dummy_mask,r20);
2124
2125             /* EWALD ELECTROSTATICS */
2126
2127             /* Analytical PME correction */
2128             zeta2            = _mm256_mul_ps(beta2,rsq20);
2129             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
2130             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2131             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2132             felec            = _mm256_mul_ps(qq20,felec);
2133
2134             fscal            = felec;
2135
2136             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2137
2138             /* Calculate temporary vectorial force */
2139             tx               = _mm256_mul_ps(fscal,dx20);
2140             ty               = _mm256_mul_ps(fscal,dy20);
2141             tz               = _mm256_mul_ps(fscal,dz20);
2142
2143             /* Update vectorial force */
2144             fix2             = _mm256_add_ps(fix2,tx);
2145             fiy2             = _mm256_add_ps(fiy2,ty);
2146             fiz2             = _mm256_add_ps(fiz2,tz);
2147
2148             fjx0             = _mm256_add_ps(fjx0,tx);
2149             fjy0             = _mm256_add_ps(fjy0,ty);
2150             fjz0             = _mm256_add_ps(fjz0,tz);
2151
2152             /**************************
2153              * CALCULATE INTERACTIONS *
2154              **************************/
2155
2156             r21              = _mm256_mul_ps(rsq21,rinv21);
2157             r21              = _mm256_andnot_ps(dummy_mask,r21);
2158
2159             /* EWALD ELECTROSTATICS */
2160
2161             /* Analytical PME correction */
2162             zeta2            = _mm256_mul_ps(beta2,rsq21);
2163             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2164             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2165             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2166             felec            = _mm256_mul_ps(qq21,felec);
2167
2168             fscal            = felec;
2169
2170             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2171
2172             /* Calculate temporary vectorial force */
2173             tx               = _mm256_mul_ps(fscal,dx21);
2174             ty               = _mm256_mul_ps(fscal,dy21);
2175             tz               = _mm256_mul_ps(fscal,dz21);
2176
2177             /* Update vectorial force */
2178             fix2             = _mm256_add_ps(fix2,tx);
2179             fiy2             = _mm256_add_ps(fiy2,ty);
2180             fiz2             = _mm256_add_ps(fiz2,tz);
2181
2182             fjx1             = _mm256_add_ps(fjx1,tx);
2183             fjy1             = _mm256_add_ps(fjy1,ty);
2184             fjz1             = _mm256_add_ps(fjz1,tz);
2185
2186             /**************************
2187              * CALCULATE INTERACTIONS *
2188              **************************/
2189
2190             r22              = _mm256_mul_ps(rsq22,rinv22);
2191             r22              = _mm256_andnot_ps(dummy_mask,r22);
2192
2193             /* EWALD ELECTROSTATICS */
2194
2195             /* Analytical PME correction */
2196             zeta2            = _mm256_mul_ps(beta2,rsq22);
2197             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2198             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2199             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2200             felec            = _mm256_mul_ps(qq22,felec);
2201
2202             fscal            = felec;
2203
2204             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2205
2206             /* Calculate temporary vectorial force */
2207             tx               = _mm256_mul_ps(fscal,dx22);
2208             ty               = _mm256_mul_ps(fscal,dy22);
2209             tz               = _mm256_mul_ps(fscal,dz22);
2210
2211             /* Update vectorial force */
2212             fix2             = _mm256_add_ps(fix2,tx);
2213             fiy2             = _mm256_add_ps(fiy2,ty);
2214             fiz2             = _mm256_add_ps(fiz2,tz);
2215
2216             fjx2             = _mm256_add_ps(fjx2,tx);
2217             fjy2             = _mm256_add_ps(fjy2,ty);
2218             fjz2             = _mm256_add_ps(fjz2,tz);
2219
2220             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2221             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2222             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2223             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2224             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2225             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2226             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2227             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2228
2229             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2230                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2231
2232             /* Inner loop uses 520 flops */
2233         }
2234
2235         /* End of innermost loop */
2236
2237         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2238                                                  f+i_coord_offset,fshift+i_shift_offset);
2239
2240         /* Increment number of inner iterations */
2241         inneriter                  += j_index_end - j_index_start;
2242
2243         /* Outer loop uses 18 flops */
2244     }
2245
2246     /* Increment number of outer iterations */
2247     outeriter        += nri;
2248
2249     /* Update outer/inner flops */
2250
2251     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*520);
2252 }