src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwNone_GeomW3W3_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3W3_VF_avx_256_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Water3-Water3
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwNone_GeomW3W3_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  92     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  94     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  95     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  96     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  97     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  98     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  99     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 100     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 101     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 102     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 103     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 104     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 105     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 106     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 107     real             *charge;
 108     __m256i          ewitab;
 109     __m128i          ewitab_lo,ewitab_hi;
 110     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 112     real             *ewtab;
 113     __m256           dummy_mask,cutoff_mask;
 114     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 115     __m256           one     = _mm256_set1_ps(1.0);
 116     __m256           two     = _mm256_set1_ps(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_ps(fr->ic->epsfac);
 129     charge           = mdatoms->chargeA;
 130
 131     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 132     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 133     beta2            = _mm256_mul_ps(beta,beta);
 134     beta3            = _mm256_mul_ps(beta,beta2);
 135
 136     ewtab            = fr->ic->tabq_coul_FDV0;
 137     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 138     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 139
 140     /* Setup water-specific parameters */
 141     inr              = nlist->iinr[0];
 142     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 143     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 144     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 145
 146     jq0              = _mm256_set1_ps(charge[inr+0]);
 147     jq1              = _mm256_set1_ps(charge[inr+1]);
 148     jq2              = _mm256_set1_ps(charge[inr+2]);
 149     qq00             = _mm256_mul_ps(iq0,jq0);
 150     qq01             = _mm256_mul_ps(iq0,jq1);
 151     qq02             = _mm256_mul_ps(iq0,jq2);
 152     qq10             = _mm256_mul_ps(iq1,jq0);
 153     qq11             = _mm256_mul_ps(iq1,jq1);
 154     qq12             = _mm256_mul_ps(iq1,jq2);
 155     qq20             = _mm256_mul_ps(iq2,jq0);
 156     qq21             = _mm256_mul_ps(iq2,jq1);
 157     qq22             = _mm256_mul_ps(iq2,jq2);
 158
 159     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 160     rcutoff_scalar   = fr->ic->rcoulomb;
 161     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 162     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 163
 164     /* Avoid stupid compiler warnings */
 165     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 166     j_coord_offsetA = 0;
 167     j_coord_offsetB = 0;
 168     j_coord_offsetC = 0;
 169     j_coord_offsetD = 0;
 170     j_coord_offsetE = 0;
 171     j_coord_offsetF = 0;
 172     j_coord_offsetG = 0;
 173     j_coord_offsetH = 0;
 174
 175     outeriter        = 0;
 176     inneriter        = 0;
 177
 178     for(iidx=0;iidx<4*DIM;iidx++)
 179     {
 180         scratch[iidx] = 0.0;
 181     }
 182
 183     /* Start outer loop over neighborlists */
 184     for(iidx=0; iidx<nri; iidx++)
 185     {
 186         /* Load shift vector for this list */
 187         i_shift_offset   = DIM*shiftidx[iidx];
 188
 189         /* Load limits for loop over neighbors */
 190         j_index_start    = jindex[iidx];
 191         j_index_end      = jindex[iidx+1];
 192
 193         /* Get outer coordinate index */
 194         inr              = iinr[iidx];
 195         i_coord_offset   = DIM*inr;
 196
 197         /* Load i particle coords and add shift vector */
 198         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 199                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 200
 201         fix0             = _mm256_setzero_ps();
 202         fiy0             = _mm256_setzero_ps();
 203         fiz0             = _mm256_setzero_ps();
 204         fix1             = _mm256_setzero_ps();
 205         fiy1             = _mm256_setzero_ps();
 206         fiz1             = _mm256_setzero_ps();
 207         fix2             = _mm256_setzero_ps();
 208         fiy2             = _mm256_setzero_ps();
 209         fiz2             = _mm256_setzero_ps();
 210
 211         /* Reset potential sums */
 212         velecsum         = _mm256_setzero_ps();
 213
 214         /* Start inner kernel loop */
 215         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 216         {
 217
 218             /* Get j neighbor index, and coordinate index */
 219             jnrA             = jjnr[jidx];
 220             jnrB             = jjnr[jidx+1];
 221             jnrC             = jjnr[jidx+2];
 222             jnrD             = jjnr[jidx+3];
 223             jnrE             = jjnr[jidx+4];
 224             jnrF             = jjnr[jidx+5];
 225             jnrG             = jjnr[jidx+6];
 226             jnrH             = jjnr[jidx+7];
 227             j_coord_offsetA  = DIM*jnrA;
 228             j_coord_offsetB  = DIM*jnrB;
 229             j_coord_offsetC  = DIM*jnrC;
 230             j_coord_offsetD  = DIM*jnrD;
 231             j_coord_offsetE  = DIM*jnrE;
 232             j_coord_offsetF  = DIM*jnrF;
 233             j_coord_offsetG  = DIM*jnrG;
 234             j_coord_offsetH  = DIM*jnrH;
 235
 236             /* load j atom coordinates */
 237             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 238                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 239                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 240                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 241                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 242
 243             /* Calculate displacement vector */
 244             dx00             = _mm256_sub_ps(ix0,jx0);
 245             dy00             = _mm256_sub_ps(iy0,jy0);
 246             dz00             = _mm256_sub_ps(iz0,jz0);
 247             dx01             = _mm256_sub_ps(ix0,jx1);
 248             dy01             = _mm256_sub_ps(iy0,jy1);
 249             dz01             = _mm256_sub_ps(iz0,jz1);
 250             dx02             = _mm256_sub_ps(ix0,jx2);
 251             dy02             = _mm256_sub_ps(iy0,jy2);
 252             dz02             = _mm256_sub_ps(iz0,jz2);
 253             dx10             = _mm256_sub_ps(ix1,jx0);
 254             dy10             = _mm256_sub_ps(iy1,jy0);
 255             dz10             = _mm256_sub_ps(iz1,jz0);
 256             dx11             = _mm256_sub_ps(ix1,jx1);
 257             dy11             = _mm256_sub_ps(iy1,jy1);
 258             dz11             = _mm256_sub_ps(iz1,jz1);
 259             dx12             = _mm256_sub_ps(ix1,jx2);
 260             dy12             = _mm256_sub_ps(iy1,jy2);
 261             dz12             = _mm256_sub_ps(iz1,jz2);
 262             dx20             = _mm256_sub_ps(ix2,jx0);
 263             dy20             = _mm256_sub_ps(iy2,jy0);
 264             dz20             = _mm256_sub_ps(iz2,jz0);
 265             dx21             = _mm256_sub_ps(ix2,jx1);
 266             dy21             = _mm256_sub_ps(iy2,jy1);
 267             dz21             = _mm256_sub_ps(iz2,jz1);
 268             dx22             = _mm256_sub_ps(ix2,jx2);
 269             dy22             = _mm256_sub_ps(iy2,jy2);
 270             dz22             = _mm256_sub_ps(iz2,jz2);
 271
 272             /* Calculate squared distance and things based on it */
 273             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 274             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 275             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 276             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 277             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 278             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 279             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 280             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 281             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 282
 283             rinv00           = avx256_invsqrt_f(rsq00);
 284             rinv01           = avx256_invsqrt_f(rsq01);
 285             rinv02           = avx256_invsqrt_f(rsq02);
 286             rinv10           = avx256_invsqrt_f(rsq10);
 287             rinv11           = avx256_invsqrt_f(rsq11);
 288             rinv12           = avx256_invsqrt_f(rsq12);
 289             rinv20           = avx256_invsqrt_f(rsq20);
 290             rinv21           = avx256_invsqrt_f(rsq21);
 291             rinv22           = avx256_invsqrt_f(rsq22);
 292
 293             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 294             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 295             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 296             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 297             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 298             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 299             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 300             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 301             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 302
 303             fjx0             = _mm256_setzero_ps();
 304             fjy0             = _mm256_setzero_ps();
 305             fjz0             = _mm256_setzero_ps();
 306             fjx1             = _mm256_setzero_ps();
 307             fjy1             = _mm256_setzero_ps();
 308             fjz1             = _mm256_setzero_ps();
 309             fjx2             = _mm256_setzero_ps();
 310             fjy2             = _mm256_setzero_ps();
 311             fjz2             = _mm256_setzero_ps();
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 318             {
 319
 320             r00              = _mm256_mul_ps(rsq00,rinv00);
 321
 322             /* EWALD ELECTROSTATICS */
 323
 324             /* Analytical PME correction */
 325             zeta2            = _mm256_mul_ps(beta2,rsq00);
 326             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 327             pmecorrF         = avx256_pmecorrF_f(zeta2);
 328             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 329             felec            = _mm256_mul_ps(qq00,felec);
 330             pmecorrV         = avx256_pmecorrV_f(zeta2);
 331             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 332             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 333             velec            = _mm256_mul_ps(qq00,velec);
 334
 335             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velec            = _mm256_and_ps(velec,cutoff_mask);
 339             velecsum         = _mm256_add_ps(velecsum,velec);
 340
 341             fscal            = felec;
 342
 343             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 344
 345             /* Calculate temporary vectorial force */
 346             tx               = _mm256_mul_ps(fscal,dx00);
 347             ty               = _mm256_mul_ps(fscal,dy00);
 348             tz               = _mm256_mul_ps(fscal,dz00);
 349
 350             /* Update vectorial force */
 351             fix0             = _mm256_add_ps(fix0,tx);
 352             fiy0             = _mm256_add_ps(fiy0,ty);
 353             fiz0             = _mm256_add_ps(fiz0,tz);
 354
 355             fjx0             = _mm256_add_ps(fjx0,tx);
 356             fjy0             = _mm256_add_ps(fjy0,ty);
 357             fjz0             = _mm256_add_ps(fjz0,tz);
 358
 359             }
 360
 361             /**************************
 362              * CALCULATE INTERACTIONS *
 363              **************************/
 364
 365             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 366             {
 367
 368             r01              = _mm256_mul_ps(rsq01,rinv01);
 369
 370             /* EWALD ELECTROSTATICS */
 371
 372             /* Analytical PME correction */
 373             zeta2            = _mm256_mul_ps(beta2,rsq01);
 374             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 375             pmecorrF         = avx256_pmecorrF_f(zeta2);
 376             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 377             felec            = _mm256_mul_ps(qq01,felec);
 378             pmecorrV         = avx256_pmecorrV_f(zeta2);
 379             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 380             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv01,sh_ewald),pmecorrV);
 381             velec            = _mm256_mul_ps(qq01,velec);
 382
 383             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 384
 385             /* Update potential sum for this i atom from the interaction with this j atom. */
 386             velec            = _mm256_and_ps(velec,cutoff_mask);
 387             velecsum         = _mm256_add_ps(velecsum,velec);
 388
 389             fscal            = felec;
 390
 391             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 392
 393             /* Calculate temporary vectorial force */
 394             tx               = _mm256_mul_ps(fscal,dx01);
 395             ty               = _mm256_mul_ps(fscal,dy01);
 396             tz               = _mm256_mul_ps(fscal,dz01);
 397
 398             /* Update vectorial force */
 399             fix0             = _mm256_add_ps(fix0,tx);
 400             fiy0             = _mm256_add_ps(fiy0,ty);
 401             fiz0             = _mm256_add_ps(fiz0,tz);
 402
 403             fjx1             = _mm256_add_ps(fjx1,tx);
 404             fjy1             = _mm256_add_ps(fjy1,ty);
 405             fjz1             = _mm256_add_ps(fjz1,tz);
 406
 407             }
 408
 409             /**************************
 410              * CALCULATE INTERACTIONS *
 411              **************************/
 412
 413             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 414             {
 415
 416             r02              = _mm256_mul_ps(rsq02,rinv02);
 417
 418             /* EWALD ELECTROSTATICS */
 419
 420             /* Analytical PME correction */
 421             zeta2            = _mm256_mul_ps(beta2,rsq02);
 422             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 423             pmecorrF         = avx256_pmecorrF_f(zeta2);
 424             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 425             felec            = _mm256_mul_ps(qq02,felec);
 426             pmecorrV         = avx256_pmecorrV_f(zeta2);
 427             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 428             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv02,sh_ewald),pmecorrV);
 429             velec            = _mm256_mul_ps(qq02,velec);
 430
 431             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
 432
 433             /* Update potential sum for this i atom from the interaction with this j atom. */
 434             velec            = _mm256_and_ps(velec,cutoff_mask);
 435             velecsum         = _mm256_add_ps(velecsum,velec);
 436
 437             fscal            = felec;
 438
 439             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 440
 441             /* Calculate temporary vectorial force */
 442             tx               = _mm256_mul_ps(fscal,dx02);
 443             ty               = _mm256_mul_ps(fscal,dy02);
 444             tz               = _mm256_mul_ps(fscal,dz02);
 445
 446             /* Update vectorial force */
 447             fix0             = _mm256_add_ps(fix0,tx);
 448             fiy0             = _mm256_add_ps(fiy0,ty);
 449             fiz0             = _mm256_add_ps(fiz0,tz);
 450
 451             fjx2             = _mm256_add_ps(fjx2,tx);
 452             fjy2             = _mm256_add_ps(fjy2,ty);
 453             fjz2             = _mm256_add_ps(fjz2,tz);
 454
 455             }
 456
 457             /**************************
 458              * CALCULATE INTERACTIONS *
 459              **************************/
 460
 461             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 462             {
 463
 464             r10              = _mm256_mul_ps(rsq10,rinv10);
 465
 466             /* EWALD ELECTROSTATICS */
 467
 468             /* Analytical PME correction */
 469             zeta2            = _mm256_mul_ps(beta2,rsq10);
 470             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 471             pmecorrF         = avx256_pmecorrF_f(zeta2);
 472             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 473             felec            = _mm256_mul_ps(qq10,felec);
 474             pmecorrV         = avx256_pmecorrV_f(zeta2);
 475             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 476             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 477             velec            = _mm256_mul_ps(qq10,velec);
 478
 479             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 480
 481             /* Update potential sum for this i atom from the interaction with this j atom. */
 482             velec            = _mm256_and_ps(velec,cutoff_mask);
 483             velecsum         = _mm256_add_ps(velecsum,velec);
 484
 485             fscal            = felec;
 486
 487             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 488
 489             /* Calculate temporary vectorial force */
 490             tx               = _mm256_mul_ps(fscal,dx10);
 491             ty               = _mm256_mul_ps(fscal,dy10);
 492             tz               = _mm256_mul_ps(fscal,dz10);
 493
 494             /* Update vectorial force */
 495             fix1             = _mm256_add_ps(fix1,tx);
 496             fiy1             = _mm256_add_ps(fiy1,ty);
 497             fiz1             = _mm256_add_ps(fiz1,tz);
 498
 499             fjx0             = _mm256_add_ps(fjx0,tx);
 500             fjy0             = _mm256_add_ps(fjy0,ty);
 501             fjz0             = _mm256_add_ps(fjz0,tz);
 502
 503             }
 504
 505             /**************************
 506              * CALCULATE INTERACTIONS *
 507              **************************/
 508
 509             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 510             {
 511
 512             r11              = _mm256_mul_ps(rsq11,rinv11);
 513
 514             /* EWALD ELECTROSTATICS */
 515
 516             /* Analytical PME correction */
 517             zeta2            = _mm256_mul_ps(beta2,rsq11);
 518             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 519             pmecorrF         = avx256_pmecorrF_f(zeta2);
 520             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 521             felec            = _mm256_mul_ps(qq11,felec);
 522             pmecorrV         = avx256_pmecorrV_f(zeta2);
 523             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 524             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
 525             velec            = _mm256_mul_ps(qq11,velec);
 526
 527             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             velec            = _mm256_and_ps(velec,cutoff_mask);
 531             velecsum         = _mm256_add_ps(velecsum,velec);
 532
 533             fscal            = felec;
 534
 535             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 536
 537             /* Calculate temporary vectorial force */
 538             tx               = _mm256_mul_ps(fscal,dx11);
 539             ty               = _mm256_mul_ps(fscal,dy11);
 540             tz               = _mm256_mul_ps(fscal,dz11);
 541
 542             /* Update vectorial force */
 543             fix1             = _mm256_add_ps(fix1,tx);
 544             fiy1             = _mm256_add_ps(fiy1,ty);
 545             fiz1             = _mm256_add_ps(fiz1,tz);
 546
 547             fjx1             = _mm256_add_ps(fjx1,tx);
 548             fjy1             = _mm256_add_ps(fjy1,ty);
 549             fjz1             = _mm256_add_ps(fjz1,tz);
 550
 551             }
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 558             {
 559
 560             r12              = _mm256_mul_ps(rsq12,rinv12);
 561
 562             /* EWALD ELECTROSTATICS */
 563
 564             /* Analytical PME correction */
 565             zeta2            = _mm256_mul_ps(beta2,rsq12);
 566             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 567             pmecorrF         = avx256_pmecorrF_f(zeta2);
 568             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 569             felec            = _mm256_mul_ps(qq12,felec);
 570             pmecorrV         = avx256_pmecorrV_f(zeta2);
 571             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 572             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
 573             velec            = _mm256_mul_ps(qq12,velec);
 574
 575             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm256_and_ps(velec,cutoff_mask);
 579             velecsum         = _mm256_add_ps(velecsum,velec);
 580
 581             fscal            = felec;
 582
 583             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 584
 585             /* Calculate temporary vectorial force */
 586             tx               = _mm256_mul_ps(fscal,dx12);
 587             ty               = _mm256_mul_ps(fscal,dy12);
 588             tz               = _mm256_mul_ps(fscal,dz12);
 589
 590             /* Update vectorial force */
 591             fix1             = _mm256_add_ps(fix1,tx);
 592             fiy1             = _mm256_add_ps(fiy1,ty);
 593             fiz1             = _mm256_add_ps(fiz1,tz);
 594
 595             fjx2             = _mm256_add_ps(fjx2,tx);
 596             fjy2             = _mm256_add_ps(fjy2,ty);
 597             fjz2             = _mm256_add_ps(fjz2,tz);
 598
 599             }
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 606             {
 607
 608             r20              = _mm256_mul_ps(rsq20,rinv20);
 609
 610             /* EWALD ELECTROSTATICS */
 611
 612             /* Analytical PME correction */
 613             zeta2            = _mm256_mul_ps(beta2,rsq20);
 614             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 615             pmecorrF         = avx256_pmecorrF_f(zeta2);
 616             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 617             felec            = _mm256_mul_ps(qq20,felec);
 618             pmecorrV         = avx256_pmecorrV_f(zeta2);
 619             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 620             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 621             velec            = _mm256_mul_ps(qq20,velec);
 622
 623             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 624
 625             /* Update potential sum for this i atom from the interaction with this j atom. */
 626             velec            = _mm256_and_ps(velec,cutoff_mask);
 627             velecsum         = _mm256_add_ps(velecsum,velec);
 628
 629             fscal            = felec;
 630
 631             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 632
 633             /* Calculate temporary vectorial force */
 634             tx               = _mm256_mul_ps(fscal,dx20);
 635             ty               = _mm256_mul_ps(fscal,dy20);
 636             tz               = _mm256_mul_ps(fscal,dz20);
 637
 638             /* Update vectorial force */
 639             fix2             = _mm256_add_ps(fix2,tx);
 640             fiy2             = _mm256_add_ps(fiy2,ty);
 641             fiz2             = _mm256_add_ps(fiz2,tz);
 642
 643             fjx0             = _mm256_add_ps(fjx0,tx);
 644             fjy0             = _mm256_add_ps(fjy0,ty);
 645             fjz0             = _mm256_add_ps(fjz0,tz);
 646
 647             }
 648
 649             /**************************
 650              * CALCULATE INTERACTIONS *
 651              **************************/
 652
 653             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 654             {
 655
 656             r21              = _mm256_mul_ps(rsq21,rinv21);
 657
 658             /* EWALD ELECTROSTATICS */
 659
 660             /* Analytical PME correction */
 661             zeta2            = _mm256_mul_ps(beta2,rsq21);
 662             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 663             pmecorrF         = avx256_pmecorrF_f(zeta2);
 664             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 665             felec            = _mm256_mul_ps(qq21,felec);
 666             pmecorrV         = avx256_pmecorrV_f(zeta2);
 667             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 668             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
 669             velec            = _mm256_mul_ps(qq21,velec);
 670
 671             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
 672
 673             /* Update potential sum for this i atom from the interaction with this j atom. */
 674             velec            = _mm256_and_ps(velec,cutoff_mask);
 675             velecsum         = _mm256_add_ps(velecsum,velec);
 676
 677             fscal            = felec;
 678
 679             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 680
 681             /* Calculate temporary vectorial force */
 682             tx               = _mm256_mul_ps(fscal,dx21);
 683             ty               = _mm256_mul_ps(fscal,dy21);
 684             tz               = _mm256_mul_ps(fscal,dz21);
 685
 686             /* Update vectorial force */
 687             fix2             = _mm256_add_ps(fix2,tx);
 688             fiy2             = _mm256_add_ps(fiy2,ty);
 689             fiz2             = _mm256_add_ps(fiz2,tz);
 690
 691             fjx1             = _mm256_add_ps(fjx1,tx);
 692             fjy1             = _mm256_add_ps(fjy1,ty);
 693             fjz1             = _mm256_add_ps(fjz1,tz);
 694
 695             }
 696
 697             /**************************
 698              * CALCULATE INTERACTIONS *
 699              **************************/
 700
 701             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 702             {
 703
 704             r22              = _mm256_mul_ps(rsq22,rinv22);
 705
 706             /* EWALD ELECTROSTATICS */
 707
 708             /* Analytical PME correction */
 709             zeta2            = _mm256_mul_ps(beta2,rsq22);
 710             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 711             pmecorrF         = avx256_pmecorrF_f(zeta2);
 712             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 713             felec            = _mm256_mul_ps(qq22,felec);
 714             pmecorrV         = avx256_pmecorrV_f(zeta2);
 715             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 716             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
 717             velec            = _mm256_mul_ps(qq22,velec);
 718
 719             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
 720
 721             /* Update potential sum for this i atom from the interaction with this j atom. */
 722             velec            = _mm256_and_ps(velec,cutoff_mask);
 723             velecsum         = _mm256_add_ps(velecsum,velec);
 724
 725             fscal            = felec;
 726
 727             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 728
 729             /* Calculate temporary vectorial force */
 730             tx               = _mm256_mul_ps(fscal,dx22);
 731             ty               = _mm256_mul_ps(fscal,dy22);
 732             tz               = _mm256_mul_ps(fscal,dz22);
 733
 734             /* Update vectorial force */
 735             fix2             = _mm256_add_ps(fix2,tx);
 736             fiy2             = _mm256_add_ps(fiy2,ty);
 737             fiz2             = _mm256_add_ps(fiz2,tz);
 738
 739             fjx2             = _mm256_add_ps(fjx2,tx);
 740             fjy2             = _mm256_add_ps(fjy2,ty);
 741             fjz2             = _mm256_add_ps(fjz2,tz);
 742
 743             }
 744
 745             fjptrA             = f+j_coord_offsetA;
 746             fjptrB             = f+j_coord_offsetB;
 747             fjptrC             = f+j_coord_offsetC;
 748             fjptrD             = f+j_coord_offsetD;
 749             fjptrE             = f+j_coord_offsetE;
 750             fjptrF             = f+j_coord_offsetF;
 751             fjptrG             = f+j_coord_offsetG;
 752             fjptrH             = f+j_coord_offsetH;
 753
 754             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 755                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 756
 757             /* Inner loop uses 981 flops */
 758         }
 759
 760         if(jidx<j_index_end)
 761         {
 762
 763             /* Get j neighbor index, and coordinate index */
 764             jnrlistA         = jjnr[jidx];
 765             jnrlistB         = jjnr[jidx+1];
 766             jnrlistC         = jjnr[jidx+2];
 767             jnrlistD         = jjnr[jidx+3];
 768             jnrlistE         = jjnr[jidx+4];
 769             jnrlistF         = jjnr[jidx+5];
 770             jnrlistG         = jjnr[jidx+6];
 771             jnrlistH         = jjnr[jidx+7];
 772             /* Sign of each element will be negative for non-real atoms.
 773              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 774              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 775              */
 776             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 777                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 778
 779             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 780             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 781             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 782             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 783             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 784             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 785             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 786             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 787             j_coord_offsetA  = DIM*jnrA;
 788             j_coord_offsetB  = DIM*jnrB;
 789             j_coord_offsetC  = DIM*jnrC;
 790             j_coord_offsetD  = DIM*jnrD;
 791             j_coord_offsetE  = DIM*jnrE;
 792             j_coord_offsetF  = DIM*jnrF;
 793             j_coord_offsetG  = DIM*jnrG;
 794             j_coord_offsetH  = DIM*jnrH;
 795
 796             /* load j atom coordinates */
 797             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 798                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 799                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 800                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 801                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 802
 803             /* Calculate displacement vector */
 804             dx00             = _mm256_sub_ps(ix0,jx0);
 805             dy00             = _mm256_sub_ps(iy0,jy0);
 806             dz00             = _mm256_sub_ps(iz0,jz0);
 807             dx01             = _mm256_sub_ps(ix0,jx1);
 808             dy01             = _mm256_sub_ps(iy0,jy1);
 809             dz01             = _mm256_sub_ps(iz0,jz1);
 810             dx02             = _mm256_sub_ps(ix0,jx2);
 811             dy02             = _mm256_sub_ps(iy0,jy2);
 812             dz02             = _mm256_sub_ps(iz0,jz2);
 813             dx10             = _mm256_sub_ps(ix1,jx0);
 814             dy10             = _mm256_sub_ps(iy1,jy0);
 815             dz10             = _mm256_sub_ps(iz1,jz0);
 816             dx11             = _mm256_sub_ps(ix1,jx1);
 817             dy11             = _mm256_sub_ps(iy1,jy1);
 818             dz11             = _mm256_sub_ps(iz1,jz1);
 819             dx12             = _mm256_sub_ps(ix1,jx2);
 820             dy12             = _mm256_sub_ps(iy1,jy2);
 821             dz12             = _mm256_sub_ps(iz1,jz2);
 822             dx20             = _mm256_sub_ps(ix2,jx0);
 823             dy20             = _mm256_sub_ps(iy2,jy0);
 824             dz20             = _mm256_sub_ps(iz2,jz0);
 825             dx21             = _mm256_sub_ps(ix2,jx1);
 826             dy21             = _mm256_sub_ps(iy2,jy1);
 827             dz21             = _mm256_sub_ps(iz2,jz1);
 828             dx22             = _mm256_sub_ps(ix2,jx2);
 829             dy22             = _mm256_sub_ps(iy2,jy2);
 830             dz22             = _mm256_sub_ps(iz2,jz2);
 831
 832             /* Calculate squared distance and things based on it */
 833             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 834             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 835             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 836             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 837             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 838             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 839             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 840             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 841             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 842
 843             rinv00           = avx256_invsqrt_f(rsq00);
 844             rinv01           = avx256_invsqrt_f(rsq01);
 845             rinv02           = avx256_invsqrt_f(rsq02);
 846             rinv10           = avx256_invsqrt_f(rsq10);
 847             rinv11           = avx256_invsqrt_f(rsq11);
 848             rinv12           = avx256_invsqrt_f(rsq12);
 849             rinv20           = avx256_invsqrt_f(rsq20);
 850             rinv21           = avx256_invsqrt_f(rsq21);
 851             rinv22           = avx256_invsqrt_f(rsq22);
 852
 853             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 854             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 855             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 856             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 857             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 858             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 859             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 860             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 861             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 862
 863             fjx0             = _mm256_setzero_ps();
 864             fjy0             = _mm256_setzero_ps();
 865             fjz0             = _mm256_setzero_ps();
 866             fjx1             = _mm256_setzero_ps();
 867             fjy1             = _mm256_setzero_ps();
 868             fjz1             = _mm256_setzero_ps();
 869             fjx2             = _mm256_setzero_ps();
 870             fjy2             = _mm256_setzero_ps();
 871             fjz2             = _mm256_setzero_ps();
 872
 873             /**************************
 874              * CALCULATE INTERACTIONS *
 875              **************************/
 876
 877             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 878             {
 879
 880             r00              = _mm256_mul_ps(rsq00,rinv00);
 881             r00              = _mm256_andnot_ps(dummy_mask,r00);
 882
 883             /* EWALD ELECTROSTATICS */
 884
 885             /* Analytical PME correction */
 886             zeta2            = _mm256_mul_ps(beta2,rsq00);
 887             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 888             pmecorrF         = avx256_pmecorrF_f(zeta2);
 889             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 890             felec            = _mm256_mul_ps(qq00,felec);
 891             pmecorrV         = avx256_pmecorrV_f(zeta2);
 892             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 893             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 894             velec            = _mm256_mul_ps(qq00,velec);
 895
 896             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 897
 898             /* Update potential sum for this i atom from the interaction with this j atom. */
 899             velec            = _mm256_and_ps(velec,cutoff_mask);
 900             velec            = _mm256_andnot_ps(dummy_mask,velec);
 901             velecsum         = _mm256_add_ps(velecsum,velec);
 902
 903             fscal            = felec;
 904
 905             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 906
 907             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 908
 909             /* Calculate temporary vectorial force */
 910             tx               = _mm256_mul_ps(fscal,dx00);
 911             ty               = _mm256_mul_ps(fscal,dy00);
 912             tz               = _mm256_mul_ps(fscal,dz00);
 913
 914             /* Update vectorial force */
 915             fix0             = _mm256_add_ps(fix0,tx);
 916             fiy0             = _mm256_add_ps(fiy0,ty);
 917             fiz0             = _mm256_add_ps(fiz0,tz);
 918
 919             fjx0             = _mm256_add_ps(fjx0,tx);
 920             fjy0             = _mm256_add_ps(fjy0,ty);
 921             fjz0             = _mm256_add_ps(fjz0,tz);
 922
 923             }
 924
 925             /**************************
 926              * CALCULATE INTERACTIONS *
 927              **************************/
 928
 929             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 930             {
 931
 932             r01              = _mm256_mul_ps(rsq01,rinv01);
 933             r01              = _mm256_andnot_ps(dummy_mask,r01);
 934
 935             /* EWALD ELECTROSTATICS */
 936
 937             /* Analytical PME correction */
 938             zeta2            = _mm256_mul_ps(beta2,rsq01);
 939             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 940             pmecorrF         = avx256_pmecorrF_f(zeta2);
 941             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 942             felec            = _mm256_mul_ps(qq01,felec);
 943             pmecorrV         = avx256_pmecorrV_f(zeta2);
 944             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 945             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv01,sh_ewald),pmecorrV);
 946             velec            = _mm256_mul_ps(qq01,velec);
 947
 948             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 949
 950             /* Update potential sum for this i atom from the interaction with this j atom. */
 951             velec            = _mm256_and_ps(velec,cutoff_mask);
 952             velec            = _mm256_andnot_ps(dummy_mask,velec);
 953             velecsum         = _mm256_add_ps(velecsum,velec);
 954
 955             fscal            = felec;
 956
 957             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 958
 959             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 960
 961             /* Calculate temporary vectorial force */
 962             tx               = _mm256_mul_ps(fscal,dx01);
 963             ty               = _mm256_mul_ps(fscal,dy01);
 964             tz               = _mm256_mul_ps(fscal,dz01);
 965
 966             /* Update vectorial force */
 967             fix0             = _mm256_add_ps(fix0,tx);
 968             fiy0             = _mm256_add_ps(fiy0,ty);
 969             fiz0             = _mm256_add_ps(fiz0,tz);
 970
 971             fjx1             = _mm256_add_ps(fjx1,tx);
 972             fjy1             = _mm256_add_ps(fjy1,ty);
 973             fjz1             = _mm256_add_ps(fjz1,tz);
 974
 975             }
 976
 977             /**************************
 978              * CALCULATE INTERACTIONS *
 979              **************************/
 980
 981             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 982             {
 983
 984             r02              = _mm256_mul_ps(rsq02,rinv02);
 985             r02              = _mm256_andnot_ps(dummy_mask,r02);
 986
 987             /* EWALD ELECTROSTATICS */
 988
 989             /* Analytical PME correction */
 990             zeta2            = _mm256_mul_ps(beta2,rsq02);
 991             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 992             pmecorrF         = avx256_pmecorrF_f(zeta2);
 993             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 994             felec            = _mm256_mul_ps(qq02,felec);
 995             pmecorrV         = avx256_pmecorrV_f(zeta2);
 996             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 997             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv02,sh_ewald),pmecorrV);
 998             velec            = _mm256_mul_ps(qq02,velec);
 999
1000             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1001
1002             /* Update potential sum for this i atom from the interaction with this j atom. */
1003             velec            = _mm256_and_ps(velec,cutoff_mask);
1004             velec            = _mm256_andnot_ps(dummy_mask,velec);
1005             velecsum         = _mm256_add_ps(velecsum,velec);
1006
1007             fscal            = felec;
1008
1009             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1010
1011             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1012
1013             /* Calculate temporary vectorial force */
1014             tx               = _mm256_mul_ps(fscal,dx02);
1015             ty               = _mm256_mul_ps(fscal,dy02);
1016             tz               = _mm256_mul_ps(fscal,dz02);
1017
1018             /* Update vectorial force */
1019             fix0             = _mm256_add_ps(fix0,tx);
1020             fiy0             = _mm256_add_ps(fiy0,ty);
1021             fiz0             = _mm256_add_ps(fiz0,tz);
1022
1023             fjx2             = _mm256_add_ps(fjx2,tx);
1024             fjy2             = _mm256_add_ps(fjy2,ty);
1025             fjz2             = _mm256_add_ps(fjz2,tz);
1026
1027             }
1028
1029             /**************************
1030              * CALCULATE INTERACTIONS *
1031              **************************/
1032
1033             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1034             {
1035
1036             r10              = _mm256_mul_ps(rsq10,rinv10);
1037             r10              = _mm256_andnot_ps(dummy_mask,r10);
1038
1039             /* EWALD ELECTROSTATICS */
1040
1041             /* Analytical PME correction */
1042             zeta2            = _mm256_mul_ps(beta2,rsq10);
1043             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1044             pmecorrF         = avx256_pmecorrF_f(zeta2);
1045             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1046             felec            = _mm256_mul_ps(qq10,felec);
1047             pmecorrV         = avx256_pmecorrV_f(zeta2);
1048             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1049             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
1050             velec            = _mm256_mul_ps(qq10,velec);
1051
1052             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1053
1054             /* Update potential sum for this i atom from the interaction with this j atom. */
1055             velec            = _mm256_and_ps(velec,cutoff_mask);
1056             velec            = _mm256_andnot_ps(dummy_mask,velec);
1057             velecsum         = _mm256_add_ps(velecsum,velec);
1058
1059             fscal            = felec;
1060
1061             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1062
1063             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1064
1065             /* Calculate temporary vectorial force */
1066             tx               = _mm256_mul_ps(fscal,dx10);
1067             ty               = _mm256_mul_ps(fscal,dy10);
1068             tz               = _mm256_mul_ps(fscal,dz10);
1069
1070             /* Update vectorial force */
1071             fix1             = _mm256_add_ps(fix1,tx);
1072             fiy1             = _mm256_add_ps(fiy1,ty);
1073             fiz1             = _mm256_add_ps(fiz1,tz);
1074
1075             fjx0             = _mm256_add_ps(fjx0,tx);
1076             fjy0             = _mm256_add_ps(fjy0,ty);
1077             fjz0             = _mm256_add_ps(fjz0,tz);
1078
1079             }
1080
1081             /**************************
1082              * CALCULATE INTERACTIONS *
1083              **************************/
1084
1085             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1086             {
1087
1088             r11              = _mm256_mul_ps(rsq11,rinv11);
1089             r11              = _mm256_andnot_ps(dummy_mask,r11);
1090
1091             /* EWALD ELECTROSTATICS */
1092
1093             /* Analytical PME correction */
1094             zeta2            = _mm256_mul_ps(beta2,rsq11);
1095             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1096             pmecorrF         = avx256_pmecorrF_f(zeta2);
1097             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1098             felec            = _mm256_mul_ps(qq11,felec);
1099             pmecorrV         = avx256_pmecorrV_f(zeta2);
1100             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1101             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
1102             velec            = _mm256_mul_ps(qq11,velec);
1103
1104             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1105
1106             /* Update potential sum for this i atom from the interaction with this j atom. */
1107             velec            = _mm256_and_ps(velec,cutoff_mask);
1108             velec            = _mm256_andnot_ps(dummy_mask,velec);
1109             velecsum         = _mm256_add_ps(velecsum,velec);
1110
1111             fscal            = felec;
1112
1113             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1114
1115             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1116
1117             /* Calculate temporary vectorial force */
1118             tx               = _mm256_mul_ps(fscal,dx11);
1119             ty               = _mm256_mul_ps(fscal,dy11);
1120             tz               = _mm256_mul_ps(fscal,dz11);
1121
1122             /* Update vectorial force */
1123             fix1             = _mm256_add_ps(fix1,tx);
1124             fiy1             = _mm256_add_ps(fiy1,ty);
1125             fiz1             = _mm256_add_ps(fiz1,tz);
1126
1127             fjx1             = _mm256_add_ps(fjx1,tx);
1128             fjy1             = _mm256_add_ps(fjy1,ty);
1129             fjz1             = _mm256_add_ps(fjz1,tz);
1130
1131             }
1132
1133             /**************************
1134              * CALCULATE INTERACTIONS *
1135              **************************/
1136
1137             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1138             {
1139
1140             r12              = _mm256_mul_ps(rsq12,rinv12);
1141             r12              = _mm256_andnot_ps(dummy_mask,r12);
1142
1143             /* EWALD ELECTROSTATICS */
1144
1145             /* Analytical PME correction */
1146             zeta2            = _mm256_mul_ps(beta2,rsq12);
1147             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1148             pmecorrF         = avx256_pmecorrF_f(zeta2);
1149             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1150             felec            = _mm256_mul_ps(qq12,felec);
1151             pmecorrV         = avx256_pmecorrV_f(zeta2);
1152             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1153             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
1154             velec            = _mm256_mul_ps(qq12,velec);
1155
1156             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1157
1158             /* Update potential sum for this i atom from the interaction with this j atom. */
1159             velec            = _mm256_and_ps(velec,cutoff_mask);
1160             velec            = _mm256_andnot_ps(dummy_mask,velec);
1161             velecsum         = _mm256_add_ps(velecsum,velec);
1162
1163             fscal            = felec;
1164
1165             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1166
1167             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1168
1169             /* Calculate temporary vectorial force */
1170             tx               = _mm256_mul_ps(fscal,dx12);
1171             ty               = _mm256_mul_ps(fscal,dy12);
1172             tz               = _mm256_mul_ps(fscal,dz12);
1173
1174             /* Update vectorial force */
1175             fix1             = _mm256_add_ps(fix1,tx);
1176             fiy1             = _mm256_add_ps(fiy1,ty);
1177             fiz1             = _mm256_add_ps(fiz1,tz);
1178
1179             fjx2             = _mm256_add_ps(fjx2,tx);
1180             fjy2             = _mm256_add_ps(fjy2,ty);
1181             fjz2             = _mm256_add_ps(fjz2,tz);
1182
1183             }
1184
1185             /**************************
1186              * CALCULATE INTERACTIONS *
1187              **************************/
1188
1189             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1190             {
1191
1192             r20              = _mm256_mul_ps(rsq20,rinv20);
1193             r20              = _mm256_andnot_ps(dummy_mask,r20);
1194
1195             /* EWALD ELECTROSTATICS */
1196
1197             /* Analytical PME correction */
1198             zeta2            = _mm256_mul_ps(beta2,rsq20);
1199             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1200             pmecorrF         = avx256_pmecorrF_f(zeta2);
1201             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1202             felec            = _mm256_mul_ps(qq20,felec);
1203             pmecorrV         = avx256_pmecorrV_f(zeta2);
1204             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1205             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
1206             velec            = _mm256_mul_ps(qq20,velec);
1207
1208             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1209
1210             /* Update potential sum for this i atom from the interaction with this j atom. */
1211             velec            = _mm256_and_ps(velec,cutoff_mask);
1212             velec            = _mm256_andnot_ps(dummy_mask,velec);
1213             velecsum         = _mm256_add_ps(velecsum,velec);
1214
1215             fscal            = felec;
1216
1217             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1218
1219             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1220
1221             /* Calculate temporary vectorial force */
1222             tx               = _mm256_mul_ps(fscal,dx20);
1223             ty               = _mm256_mul_ps(fscal,dy20);
1224             tz               = _mm256_mul_ps(fscal,dz20);
1225
1226             /* Update vectorial force */
1227             fix2             = _mm256_add_ps(fix2,tx);
1228             fiy2             = _mm256_add_ps(fiy2,ty);
1229             fiz2             = _mm256_add_ps(fiz2,tz);
1230
1231             fjx0             = _mm256_add_ps(fjx0,tx);
1232             fjy0             = _mm256_add_ps(fjy0,ty);
1233             fjz0             = _mm256_add_ps(fjz0,tz);
1234
1235             }
1236
1237             /**************************
1238              * CALCULATE INTERACTIONS *
1239              **************************/
1240
1241             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1242             {
1243
1244             r21              = _mm256_mul_ps(rsq21,rinv21);
1245             r21              = _mm256_andnot_ps(dummy_mask,r21);
1246
1247             /* EWALD ELECTROSTATICS */
1248
1249             /* Analytical PME correction */
1250             zeta2            = _mm256_mul_ps(beta2,rsq21);
1251             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1252             pmecorrF         = avx256_pmecorrF_f(zeta2);
1253             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1254             felec            = _mm256_mul_ps(qq21,felec);
1255             pmecorrV         = avx256_pmecorrV_f(zeta2);
1256             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1257             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
1258             velec            = _mm256_mul_ps(qq21,velec);
1259
1260             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1261
1262             /* Update potential sum for this i atom from the interaction with this j atom. */
1263             velec            = _mm256_and_ps(velec,cutoff_mask);
1264             velec            = _mm256_andnot_ps(dummy_mask,velec);
1265             velecsum         = _mm256_add_ps(velecsum,velec);
1266
1267             fscal            = felec;
1268
1269             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1270
1271             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1272
1273             /* Calculate temporary vectorial force */
1274             tx               = _mm256_mul_ps(fscal,dx21);
1275             ty               = _mm256_mul_ps(fscal,dy21);
1276             tz               = _mm256_mul_ps(fscal,dz21);
1277
1278             /* Update vectorial force */
1279             fix2             = _mm256_add_ps(fix2,tx);
1280             fiy2             = _mm256_add_ps(fiy2,ty);
1281             fiz2             = _mm256_add_ps(fiz2,tz);
1282
1283             fjx1             = _mm256_add_ps(fjx1,tx);
1284             fjy1             = _mm256_add_ps(fjy1,ty);
1285             fjz1             = _mm256_add_ps(fjz1,tz);
1286
1287             }
1288
1289             /**************************
1290              * CALCULATE INTERACTIONS *
1291              **************************/
1292
1293             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1294             {
1295
1296             r22              = _mm256_mul_ps(rsq22,rinv22);
1297             r22              = _mm256_andnot_ps(dummy_mask,r22);
1298
1299             /* EWALD ELECTROSTATICS */
1300
1301             /* Analytical PME correction */
1302             zeta2            = _mm256_mul_ps(beta2,rsq22);
1303             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1304             pmecorrF         = avx256_pmecorrF_f(zeta2);
1305             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1306             felec            = _mm256_mul_ps(qq22,felec);
1307             pmecorrV         = avx256_pmecorrV_f(zeta2);
1308             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1309             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
1310             velec            = _mm256_mul_ps(qq22,velec);
1311
1312             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1313
1314             /* Update potential sum for this i atom from the interaction with this j atom. */
1315             velec            = _mm256_and_ps(velec,cutoff_mask);
1316             velec            = _mm256_andnot_ps(dummy_mask,velec);
1317             velecsum         = _mm256_add_ps(velecsum,velec);
1318
1319             fscal            = felec;
1320
1321             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1322
1323             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1324
1325             /* Calculate temporary vectorial force */
1326             tx               = _mm256_mul_ps(fscal,dx22);
1327             ty               = _mm256_mul_ps(fscal,dy22);
1328             tz               = _mm256_mul_ps(fscal,dz22);
1329
1330             /* Update vectorial force */
1331             fix2             = _mm256_add_ps(fix2,tx);
1332             fiy2             = _mm256_add_ps(fiy2,ty);
1333             fiz2             = _mm256_add_ps(fiz2,tz);
1334
1335             fjx2             = _mm256_add_ps(fjx2,tx);
1336             fjy2             = _mm256_add_ps(fjy2,ty);
1337             fjz2             = _mm256_add_ps(fjz2,tz);
1338
1339             }
1340
1341             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1342             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1343             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1344             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1345             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1346             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1347             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1348             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1349
1350             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1351                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1352
1353             /* Inner loop uses 990 flops */
1354         }
1355
1356         /* End of innermost loop */
1357
1358         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1359                                                  f+i_coord_offset,fshift+i_shift_offset);
1360
1361         ggid                        = gid[iidx];
1362         /* Update potential energies */
1363         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1364
1365         /* Increment number of inner iterations */
1366         inneriter                  += j_index_end - j_index_start;
1367
1368         /* Outer loop uses 19 flops */
1369     }
1370
1371     /* Increment number of outer iterations */
1372     outeriter        += nri;
1373
1374     /* Update outer/inner flops */
1375
1376     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*990);
1377 }
1378 /*
1379  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3W3_F_avx_256_single
1380  * Electrostatics interaction: Ewald
1381  * VdW interaction:            None
1382  * Geometry:                   Water3-Water3
1383  * Calculate force/pot:        Force
1384  */
1385 void
1386 nb_kernel_ElecEwSh_VdwNone_GeomW3W3_F_avx_256_single
1387                     (t_nblist                    * gmx_restrict       nlist,
1388                      rvec                        * gmx_restrict          xx,
1389                      rvec                        * gmx_restrict          ff,
1390                      struct t_forcerec           * gmx_restrict          fr,
1391                      t_mdatoms                   * gmx_restrict     mdatoms,
1392                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1393                      t_nrnb                      * gmx_restrict        nrnb)
1394 {
1395     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1396      * just 0 for non-waters.
1397      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1398      * jnr indices corresponding to data put in the four positions in the SIMD register.
1399      */
1400     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1401     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1402     int              jnrA,jnrB,jnrC,jnrD;
1403     int              jnrE,jnrF,jnrG,jnrH;
1404     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1405     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1406     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1407     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1408     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1409     real             rcutoff_scalar;
1410     real             *shiftvec,*fshift,*x,*f;
1411     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1412     real             scratch[4*DIM];
1413     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1414     real *           vdwioffsetptr0;
1415     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1416     real *           vdwioffsetptr1;
1417     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1418     real *           vdwioffsetptr2;
1419     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1420     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1421     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1422     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1423     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1424     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1425     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1426     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1427     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1428     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1429     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1430     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1431     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1432     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1433     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1434     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1435     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1436     real             *charge;
1437     __m256i          ewitab;
1438     __m128i          ewitab_lo,ewitab_hi;
1439     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1440     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1441     real             *ewtab;
1442     __m256           dummy_mask,cutoff_mask;
1443     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1444     __m256           one     = _mm256_set1_ps(1.0);
1445     __m256           two     = _mm256_set1_ps(2.0);
1446     x                = xx[0];
1447     f                = ff[0];
1448
1449     nri              = nlist->nri;
1450     iinr             = nlist->iinr;
1451     jindex           = nlist->jindex;
1452     jjnr             = nlist->jjnr;
1453     shiftidx         = nlist->shift;
1454     gid              = nlist->gid;
1455     shiftvec         = fr->shift_vec[0];
1456     fshift           = fr->fshift[0];
1457     facel            = _mm256_set1_ps(fr->ic->epsfac);
1458     charge           = mdatoms->chargeA;
1459
1460     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1461     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1462     beta2            = _mm256_mul_ps(beta,beta);
1463     beta3            = _mm256_mul_ps(beta,beta2);
1464
1465     ewtab            = fr->ic->tabq_coul_F;
1466     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1467     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1468
1469     /* Setup water-specific parameters */
1470     inr              = nlist->iinr[0];
1471     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1472     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1473     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1474
1475     jq0              = _mm256_set1_ps(charge[inr+0]);
1476     jq1              = _mm256_set1_ps(charge[inr+1]);
1477     jq2              = _mm256_set1_ps(charge[inr+2]);
1478     qq00             = _mm256_mul_ps(iq0,jq0);
1479     qq01             = _mm256_mul_ps(iq0,jq1);
1480     qq02             = _mm256_mul_ps(iq0,jq2);
1481     qq10             = _mm256_mul_ps(iq1,jq0);
1482     qq11             = _mm256_mul_ps(iq1,jq1);
1483     qq12             = _mm256_mul_ps(iq1,jq2);
1484     qq20             = _mm256_mul_ps(iq2,jq0);
1485     qq21             = _mm256_mul_ps(iq2,jq1);
1486     qq22             = _mm256_mul_ps(iq2,jq2);
1487
1488     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1489     rcutoff_scalar   = fr->ic->rcoulomb;
1490     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1491     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1492
1493     /* Avoid stupid compiler warnings */
1494     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1495     j_coord_offsetA = 0;
1496     j_coord_offsetB = 0;
1497     j_coord_offsetC = 0;
1498     j_coord_offsetD = 0;
1499     j_coord_offsetE = 0;
1500     j_coord_offsetF = 0;
1501     j_coord_offsetG = 0;
1502     j_coord_offsetH = 0;
1503
1504     outeriter        = 0;
1505     inneriter        = 0;
1506
1507     for(iidx=0;iidx<4*DIM;iidx++)
1508     {
1509         scratch[iidx] = 0.0;
1510     }
1511
1512     /* Start outer loop over neighborlists */
1513     for(iidx=0; iidx<nri; iidx++)
1514     {
1515         /* Load shift vector for this list */
1516         i_shift_offset   = DIM*shiftidx[iidx];
1517
1518         /* Load limits for loop over neighbors */
1519         j_index_start    = jindex[iidx];
1520         j_index_end      = jindex[iidx+1];
1521
1522         /* Get outer coordinate index */
1523         inr              = iinr[iidx];
1524         i_coord_offset   = DIM*inr;
1525
1526         /* Load i particle coords and add shift vector */
1527         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1528                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1529
1530         fix0             = _mm256_setzero_ps();
1531         fiy0             = _mm256_setzero_ps();
1532         fiz0             = _mm256_setzero_ps();
1533         fix1             = _mm256_setzero_ps();
1534         fiy1             = _mm256_setzero_ps();
1535         fiz1             = _mm256_setzero_ps();
1536         fix2             = _mm256_setzero_ps();
1537         fiy2             = _mm256_setzero_ps();
1538         fiz2             = _mm256_setzero_ps();
1539
1540         /* Start inner kernel loop */
1541         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1542         {
1543
1544             /* Get j neighbor index, and coordinate index */
1545             jnrA             = jjnr[jidx];
1546             jnrB             = jjnr[jidx+1];
1547             jnrC             = jjnr[jidx+2];
1548             jnrD             = jjnr[jidx+3];
1549             jnrE             = jjnr[jidx+4];
1550             jnrF             = jjnr[jidx+5];
1551             jnrG             = jjnr[jidx+6];
1552             jnrH             = jjnr[jidx+7];
1553             j_coord_offsetA  = DIM*jnrA;
1554             j_coord_offsetB  = DIM*jnrB;
1555             j_coord_offsetC  = DIM*jnrC;
1556             j_coord_offsetD  = DIM*jnrD;
1557             j_coord_offsetE  = DIM*jnrE;
1558             j_coord_offsetF  = DIM*jnrF;
1559             j_coord_offsetG  = DIM*jnrG;
1560             j_coord_offsetH  = DIM*jnrH;
1561
1562             /* load j atom coordinates */
1563             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1564                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1565                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1566                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1567                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1568
1569             /* Calculate displacement vector */
1570             dx00             = _mm256_sub_ps(ix0,jx0);
1571             dy00             = _mm256_sub_ps(iy0,jy0);
1572             dz00             = _mm256_sub_ps(iz0,jz0);
1573             dx01             = _mm256_sub_ps(ix0,jx1);
1574             dy01             = _mm256_sub_ps(iy0,jy1);
1575             dz01             = _mm256_sub_ps(iz0,jz1);
1576             dx02             = _mm256_sub_ps(ix0,jx2);
1577             dy02             = _mm256_sub_ps(iy0,jy2);
1578             dz02             = _mm256_sub_ps(iz0,jz2);
1579             dx10             = _mm256_sub_ps(ix1,jx0);
1580             dy10             = _mm256_sub_ps(iy1,jy0);
1581             dz10             = _mm256_sub_ps(iz1,jz0);
1582             dx11             = _mm256_sub_ps(ix1,jx1);
1583             dy11             = _mm256_sub_ps(iy1,jy1);
1584             dz11             = _mm256_sub_ps(iz1,jz1);
1585             dx12             = _mm256_sub_ps(ix1,jx2);
1586             dy12             = _mm256_sub_ps(iy1,jy2);
1587             dz12             = _mm256_sub_ps(iz1,jz2);
1588             dx20             = _mm256_sub_ps(ix2,jx0);
1589             dy20             = _mm256_sub_ps(iy2,jy0);
1590             dz20             = _mm256_sub_ps(iz2,jz0);
1591             dx21             = _mm256_sub_ps(ix2,jx1);
1592             dy21             = _mm256_sub_ps(iy2,jy1);
1593             dz21             = _mm256_sub_ps(iz2,jz1);
1594             dx22             = _mm256_sub_ps(ix2,jx2);
1595             dy22             = _mm256_sub_ps(iy2,jy2);
1596             dz22             = _mm256_sub_ps(iz2,jz2);
1597
1598             /* Calculate squared distance and things based on it */
1599             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1600             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1601             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1602             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1603             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1604             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1605             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1606             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1607             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1608
1609             rinv00           = avx256_invsqrt_f(rsq00);
1610             rinv01           = avx256_invsqrt_f(rsq01);
1611             rinv02           = avx256_invsqrt_f(rsq02);
1612             rinv10           = avx256_invsqrt_f(rsq10);
1613             rinv11           = avx256_invsqrt_f(rsq11);
1614             rinv12           = avx256_invsqrt_f(rsq12);
1615             rinv20           = avx256_invsqrt_f(rsq20);
1616             rinv21           = avx256_invsqrt_f(rsq21);
1617             rinv22           = avx256_invsqrt_f(rsq22);
1618
1619             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1620             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1621             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1622             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1623             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1624             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1625             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1626             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1627             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1628
1629             fjx0             = _mm256_setzero_ps();
1630             fjy0             = _mm256_setzero_ps();
1631             fjz0             = _mm256_setzero_ps();
1632             fjx1             = _mm256_setzero_ps();
1633             fjy1             = _mm256_setzero_ps();
1634             fjz1             = _mm256_setzero_ps();
1635             fjx2             = _mm256_setzero_ps();
1636             fjy2             = _mm256_setzero_ps();
1637             fjz2             = _mm256_setzero_ps();
1638
1639             /**************************
1640              * CALCULATE INTERACTIONS *
1641              **************************/
1642
1643             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1644             {
1645
1646             r00              = _mm256_mul_ps(rsq00,rinv00);
1647
1648             /* EWALD ELECTROSTATICS */
1649
1650             /* Analytical PME correction */
1651             zeta2            = _mm256_mul_ps(beta2,rsq00);
1652             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1653             pmecorrF         = avx256_pmecorrF_f(zeta2);
1654             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1655             felec            = _mm256_mul_ps(qq00,felec);
1656
1657             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1658
1659             fscal            = felec;
1660
1661             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1662
1663             /* Calculate temporary vectorial force */
1664             tx               = _mm256_mul_ps(fscal,dx00);
1665             ty               = _mm256_mul_ps(fscal,dy00);
1666             tz               = _mm256_mul_ps(fscal,dz00);
1667
1668             /* Update vectorial force */
1669             fix0             = _mm256_add_ps(fix0,tx);
1670             fiy0             = _mm256_add_ps(fiy0,ty);
1671             fiz0             = _mm256_add_ps(fiz0,tz);
1672
1673             fjx0             = _mm256_add_ps(fjx0,tx);
1674             fjy0             = _mm256_add_ps(fjy0,ty);
1675             fjz0             = _mm256_add_ps(fjz0,tz);
1676
1677             }
1678
1679             /**************************
1680              * CALCULATE INTERACTIONS *
1681              **************************/
1682
1683             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1684             {
1685
1686             r01              = _mm256_mul_ps(rsq01,rinv01);
1687
1688             /* EWALD ELECTROSTATICS */
1689
1690             /* Analytical PME correction */
1691             zeta2            = _mm256_mul_ps(beta2,rsq01);
1692             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
1693             pmecorrF         = avx256_pmecorrF_f(zeta2);
1694             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1695             felec            = _mm256_mul_ps(qq01,felec);
1696
1697             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
1698
1699             fscal            = felec;
1700
1701             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1702
1703             /* Calculate temporary vectorial force */
1704             tx               = _mm256_mul_ps(fscal,dx01);
1705             ty               = _mm256_mul_ps(fscal,dy01);
1706             tz               = _mm256_mul_ps(fscal,dz01);
1707
1708             /* Update vectorial force */
1709             fix0             = _mm256_add_ps(fix0,tx);
1710             fiy0             = _mm256_add_ps(fiy0,ty);
1711             fiz0             = _mm256_add_ps(fiz0,tz);
1712
1713             fjx1             = _mm256_add_ps(fjx1,tx);
1714             fjy1             = _mm256_add_ps(fjy1,ty);
1715             fjz1             = _mm256_add_ps(fjz1,tz);
1716
1717             }
1718
1719             /**************************
1720              * CALCULATE INTERACTIONS *
1721              **************************/
1722
1723             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1724             {
1725
1726             r02              = _mm256_mul_ps(rsq02,rinv02);
1727
1728             /* EWALD ELECTROSTATICS */
1729
1730             /* Analytical PME correction */
1731             zeta2            = _mm256_mul_ps(beta2,rsq02);
1732             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
1733             pmecorrF         = avx256_pmecorrF_f(zeta2);
1734             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1735             felec            = _mm256_mul_ps(qq02,felec);
1736
1737             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1738
1739             fscal            = felec;
1740
1741             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1742
1743             /* Calculate temporary vectorial force */
1744             tx               = _mm256_mul_ps(fscal,dx02);
1745             ty               = _mm256_mul_ps(fscal,dy02);
1746             tz               = _mm256_mul_ps(fscal,dz02);
1747
1748             /* Update vectorial force */
1749             fix0             = _mm256_add_ps(fix0,tx);
1750             fiy0             = _mm256_add_ps(fiy0,ty);
1751             fiz0             = _mm256_add_ps(fiz0,tz);
1752
1753             fjx2             = _mm256_add_ps(fjx2,tx);
1754             fjy2             = _mm256_add_ps(fjy2,ty);
1755             fjz2             = _mm256_add_ps(fjz2,tz);
1756
1757             }
1758
1759             /**************************
1760              * CALCULATE INTERACTIONS *
1761              **************************/
1762
1763             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1764             {
1765
1766             r10              = _mm256_mul_ps(rsq10,rinv10);
1767
1768             /* EWALD ELECTROSTATICS */
1769
1770             /* Analytical PME correction */
1771             zeta2            = _mm256_mul_ps(beta2,rsq10);
1772             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1773             pmecorrF         = avx256_pmecorrF_f(zeta2);
1774             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1775             felec            = _mm256_mul_ps(qq10,felec);
1776
1777             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1778
1779             fscal            = felec;
1780
1781             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1782
1783             /* Calculate temporary vectorial force */
1784             tx               = _mm256_mul_ps(fscal,dx10);
1785             ty               = _mm256_mul_ps(fscal,dy10);
1786             tz               = _mm256_mul_ps(fscal,dz10);
1787
1788             /* Update vectorial force */
1789             fix1             = _mm256_add_ps(fix1,tx);
1790             fiy1             = _mm256_add_ps(fiy1,ty);
1791             fiz1             = _mm256_add_ps(fiz1,tz);
1792
1793             fjx0             = _mm256_add_ps(fjx0,tx);
1794             fjy0             = _mm256_add_ps(fjy0,ty);
1795             fjz0             = _mm256_add_ps(fjz0,tz);
1796
1797             }
1798
1799             /**************************
1800              * CALCULATE INTERACTIONS *
1801              **************************/
1802
1803             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1804             {
1805
1806             r11              = _mm256_mul_ps(rsq11,rinv11);
1807
1808             /* EWALD ELECTROSTATICS */
1809
1810             /* Analytical PME correction */
1811             zeta2            = _mm256_mul_ps(beta2,rsq11);
1812             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1813             pmecorrF         = avx256_pmecorrF_f(zeta2);
1814             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1815             felec            = _mm256_mul_ps(qq11,felec);
1816
1817             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1818
1819             fscal            = felec;
1820
1821             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1822
1823             /* Calculate temporary vectorial force */
1824             tx               = _mm256_mul_ps(fscal,dx11);
1825             ty               = _mm256_mul_ps(fscal,dy11);
1826             tz               = _mm256_mul_ps(fscal,dz11);
1827
1828             /* Update vectorial force */
1829             fix1             = _mm256_add_ps(fix1,tx);
1830             fiy1             = _mm256_add_ps(fiy1,ty);
1831             fiz1             = _mm256_add_ps(fiz1,tz);
1832
1833             fjx1             = _mm256_add_ps(fjx1,tx);
1834             fjy1             = _mm256_add_ps(fjy1,ty);
1835             fjz1             = _mm256_add_ps(fjz1,tz);
1836
1837             }
1838
1839             /**************************
1840              * CALCULATE INTERACTIONS *
1841              **************************/
1842
1843             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1844             {
1845
1846             r12              = _mm256_mul_ps(rsq12,rinv12);
1847
1848             /* EWALD ELECTROSTATICS */
1849
1850             /* Analytical PME correction */
1851             zeta2            = _mm256_mul_ps(beta2,rsq12);
1852             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1853             pmecorrF         = avx256_pmecorrF_f(zeta2);
1854             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1855             felec            = _mm256_mul_ps(qq12,felec);
1856
1857             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1858
1859             fscal            = felec;
1860
1861             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1862
1863             /* Calculate temporary vectorial force */
1864             tx               = _mm256_mul_ps(fscal,dx12);
1865             ty               = _mm256_mul_ps(fscal,dy12);
1866             tz               = _mm256_mul_ps(fscal,dz12);
1867
1868             /* Update vectorial force */
1869             fix1             = _mm256_add_ps(fix1,tx);
1870             fiy1             = _mm256_add_ps(fiy1,ty);
1871             fiz1             = _mm256_add_ps(fiz1,tz);
1872
1873             fjx2             = _mm256_add_ps(fjx2,tx);
1874             fjy2             = _mm256_add_ps(fjy2,ty);
1875             fjz2             = _mm256_add_ps(fjz2,tz);
1876
1877             }
1878
1879             /**************************
1880              * CALCULATE INTERACTIONS *
1881              **************************/
1882
1883             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1884             {
1885
1886             r20              = _mm256_mul_ps(rsq20,rinv20);
1887
1888             /* EWALD ELECTROSTATICS */
1889
1890             /* Analytical PME correction */
1891             zeta2            = _mm256_mul_ps(beta2,rsq20);
1892             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1893             pmecorrF         = avx256_pmecorrF_f(zeta2);
1894             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1895             felec            = _mm256_mul_ps(qq20,felec);
1896
1897             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1898
1899             fscal            = felec;
1900
1901             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1902
1903             /* Calculate temporary vectorial force */
1904             tx               = _mm256_mul_ps(fscal,dx20);
1905             ty               = _mm256_mul_ps(fscal,dy20);
1906             tz               = _mm256_mul_ps(fscal,dz20);
1907
1908             /* Update vectorial force */
1909             fix2             = _mm256_add_ps(fix2,tx);
1910             fiy2             = _mm256_add_ps(fiy2,ty);
1911             fiz2             = _mm256_add_ps(fiz2,tz);
1912
1913             fjx0             = _mm256_add_ps(fjx0,tx);
1914             fjy0             = _mm256_add_ps(fjy0,ty);
1915             fjz0             = _mm256_add_ps(fjz0,tz);
1916
1917             }
1918
1919             /**************************
1920              * CALCULATE INTERACTIONS *
1921              **************************/
1922
1923             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1924             {
1925
1926             r21              = _mm256_mul_ps(rsq21,rinv21);
1927
1928             /* EWALD ELECTROSTATICS */
1929
1930             /* Analytical PME correction */
1931             zeta2            = _mm256_mul_ps(beta2,rsq21);
1932             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1933             pmecorrF         = avx256_pmecorrF_f(zeta2);
1934             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1935             felec            = _mm256_mul_ps(qq21,felec);
1936
1937             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1938
1939             fscal            = felec;
1940
1941             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1942
1943             /* Calculate temporary vectorial force */
1944             tx               = _mm256_mul_ps(fscal,dx21);
1945             ty               = _mm256_mul_ps(fscal,dy21);
1946             tz               = _mm256_mul_ps(fscal,dz21);
1947
1948             /* Update vectorial force */
1949             fix2             = _mm256_add_ps(fix2,tx);
1950             fiy2             = _mm256_add_ps(fiy2,ty);
1951             fiz2             = _mm256_add_ps(fiz2,tz);
1952
1953             fjx1             = _mm256_add_ps(fjx1,tx);
1954             fjy1             = _mm256_add_ps(fjy1,ty);
1955             fjz1             = _mm256_add_ps(fjz1,tz);
1956
1957             }
1958
1959             /**************************
1960              * CALCULATE INTERACTIONS *
1961              **************************/
1962
1963             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1964             {
1965
1966             r22              = _mm256_mul_ps(rsq22,rinv22);
1967
1968             /* EWALD ELECTROSTATICS */
1969
1970             /* Analytical PME correction */
1971             zeta2            = _mm256_mul_ps(beta2,rsq22);
1972             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1973             pmecorrF         = avx256_pmecorrF_f(zeta2);
1974             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1975             felec            = _mm256_mul_ps(qq22,felec);
1976
1977             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1978
1979             fscal            = felec;
1980
1981             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1982
1983             /* Calculate temporary vectorial force */
1984             tx               = _mm256_mul_ps(fscal,dx22);
1985             ty               = _mm256_mul_ps(fscal,dy22);
1986             tz               = _mm256_mul_ps(fscal,dz22);
1987
1988             /* Update vectorial force */
1989             fix2             = _mm256_add_ps(fix2,tx);
1990             fiy2             = _mm256_add_ps(fiy2,ty);
1991             fiz2             = _mm256_add_ps(fiz2,tz);
1992
1993             fjx2             = _mm256_add_ps(fjx2,tx);
1994             fjy2             = _mm256_add_ps(fjy2,ty);
1995             fjz2             = _mm256_add_ps(fjz2,tz);
1996
1997             }
1998
1999             fjptrA             = f+j_coord_offsetA;
2000             fjptrB             = f+j_coord_offsetB;
2001             fjptrC             = f+j_coord_offsetC;
2002             fjptrD             = f+j_coord_offsetD;
2003             fjptrE             = f+j_coord_offsetE;
2004             fjptrF             = f+j_coord_offsetF;
2005             fjptrG             = f+j_coord_offsetG;
2006             fjptrH             = f+j_coord_offsetH;
2007
2008             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2009                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2010
2011             /* Inner loop uses 531 flops */
2012         }
2013
2014         if(jidx<j_index_end)
2015         {
2016
2017             /* Get j neighbor index, and coordinate index */
2018             jnrlistA         = jjnr[jidx];
2019             jnrlistB         = jjnr[jidx+1];
2020             jnrlistC         = jjnr[jidx+2];
2021             jnrlistD         = jjnr[jidx+3];
2022             jnrlistE         = jjnr[jidx+4];
2023             jnrlistF         = jjnr[jidx+5];
2024             jnrlistG         = jjnr[jidx+6];
2025             jnrlistH         = jjnr[jidx+7];
2026             /* Sign of each element will be negative for non-real atoms.
2027              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2028              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2029              */
2030             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2031                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2032
2033             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2034             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2035             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2036             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2037             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2038             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2039             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2040             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2041             j_coord_offsetA  = DIM*jnrA;
2042             j_coord_offsetB  = DIM*jnrB;
2043             j_coord_offsetC  = DIM*jnrC;
2044             j_coord_offsetD  = DIM*jnrD;
2045             j_coord_offsetE  = DIM*jnrE;
2046             j_coord_offsetF  = DIM*jnrF;
2047             j_coord_offsetG  = DIM*jnrG;
2048             j_coord_offsetH  = DIM*jnrH;
2049
2050             /* load j atom coordinates */
2051             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2052                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2053                                                  x+j_coord_offsetE,x+j_coord_offsetF,
2054                                                  x+j_coord_offsetG,x+j_coord_offsetH,
2055                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2056
2057             /* Calculate displacement vector */
2058             dx00             = _mm256_sub_ps(ix0,jx0);
2059             dy00             = _mm256_sub_ps(iy0,jy0);
2060             dz00             = _mm256_sub_ps(iz0,jz0);
2061             dx01             = _mm256_sub_ps(ix0,jx1);
2062             dy01             = _mm256_sub_ps(iy0,jy1);
2063             dz01             = _mm256_sub_ps(iz0,jz1);
2064             dx02             = _mm256_sub_ps(ix0,jx2);
2065             dy02             = _mm256_sub_ps(iy0,jy2);
2066             dz02             = _mm256_sub_ps(iz0,jz2);
2067             dx10             = _mm256_sub_ps(ix1,jx0);
2068             dy10             = _mm256_sub_ps(iy1,jy0);
2069             dz10             = _mm256_sub_ps(iz1,jz0);
2070             dx11             = _mm256_sub_ps(ix1,jx1);
2071             dy11             = _mm256_sub_ps(iy1,jy1);
2072             dz11             = _mm256_sub_ps(iz1,jz1);
2073             dx12             = _mm256_sub_ps(ix1,jx2);
2074             dy12             = _mm256_sub_ps(iy1,jy2);
2075             dz12             = _mm256_sub_ps(iz1,jz2);
2076             dx20             = _mm256_sub_ps(ix2,jx0);
2077             dy20             = _mm256_sub_ps(iy2,jy0);
2078             dz20             = _mm256_sub_ps(iz2,jz0);
2079             dx21             = _mm256_sub_ps(ix2,jx1);
2080             dy21             = _mm256_sub_ps(iy2,jy1);
2081             dz21             = _mm256_sub_ps(iz2,jz1);
2082             dx22             = _mm256_sub_ps(ix2,jx2);
2083             dy22             = _mm256_sub_ps(iy2,jy2);
2084             dz22             = _mm256_sub_ps(iz2,jz2);
2085
2086             /* Calculate squared distance and things based on it */
2087             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2088             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
2089             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
2090             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
2091             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2092             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2093             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
2094             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2095             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2096
2097             rinv00           = avx256_invsqrt_f(rsq00);
2098             rinv01           = avx256_invsqrt_f(rsq01);
2099             rinv02           = avx256_invsqrt_f(rsq02);
2100             rinv10           = avx256_invsqrt_f(rsq10);
2101             rinv11           = avx256_invsqrt_f(rsq11);
2102             rinv12           = avx256_invsqrt_f(rsq12);
2103             rinv20           = avx256_invsqrt_f(rsq20);
2104             rinv21           = avx256_invsqrt_f(rsq21);
2105             rinv22           = avx256_invsqrt_f(rsq22);
2106
2107             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
2108             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
2109             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
2110             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
2111             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
2112             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
2113             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
2114             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
2115             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
2116
2117             fjx0             = _mm256_setzero_ps();
2118             fjy0             = _mm256_setzero_ps();
2119             fjz0             = _mm256_setzero_ps();
2120             fjx1             = _mm256_setzero_ps();
2121             fjy1             = _mm256_setzero_ps();
2122             fjz1             = _mm256_setzero_ps();
2123             fjx2             = _mm256_setzero_ps();
2124             fjy2             = _mm256_setzero_ps();
2125             fjz2             = _mm256_setzero_ps();
2126
2127             /**************************
2128              * CALCULATE INTERACTIONS *
2129              **************************/
2130
2131             if (gmx_mm256_any_lt(rsq00,rcutoff2))
2132             {
2133
2134             r00              = _mm256_mul_ps(rsq00,rinv00);
2135             r00              = _mm256_andnot_ps(dummy_mask,r00);
2136
2137             /* EWALD ELECTROSTATICS */
2138
2139             /* Analytical PME correction */
2140             zeta2            = _mm256_mul_ps(beta2,rsq00);
2141             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
2142             pmecorrF         = avx256_pmecorrF_f(zeta2);
2143             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2144             felec            = _mm256_mul_ps(qq00,felec);
2145
2146             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
2147
2148             fscal            = felec;
2149
2150             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2151
2152             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2153
2154             /* Calculate temporary vectorial force */
2155             tx               = _mm256_mul_ps(fscal,dx00);
2156             ty               = _mm256_mul_ps(fscal,dy00);
2157             tz               = _mm256_mul_ps(fscal,dz00);
2158
2159             /* Update vectorial force */
2160             fix0             = _mm256_add_ps(fix0,tx);
2161             fiy0             = _mm256_add_ps(fiy0,ty);
2162             fiz0             = _mm256_add_ps(fiz0,tz);
2163
2164             fjx0             = _mm256_add_ps(fjx0,tx);
2165             fjy0             = _mm256_add_ps(fjy0,ty);
2166             fjz0             = _mm256_add_ps(fjz0,tz);
2167
2168             }
2169
2170             /**************************
2171              * CALCULATE INTERACTIONS *
2172              **************************/
2173
2174             if (gmx_mm256_any_lt(rsq01,rcutoff2))
2175             {
2176
2177             r01              = _mm256_mul_ps(rsq01,rinv01);
2178             r01              = _mm256_andnot_ps(dummy_mask,r01);
2179
2180             /* EWALD ELECTROSTATICS */
2181
2182             /* Analytical PME correction */
2183             zeta2            = _mm256_mul_ps(beta2,rsq01);
2184             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
2185             pmecorrF         = avx256_pmecorrF_f(zeta2);
2186             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2187             felec            = _mm256_mul_ps(qq01,felec);
2188
2189             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
2190
2191             fscal            = felec;
2192
2193             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2194
2195             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2196
2197             /* Calculate temporary vectorial force */
2198             tx               = _mm256_mul_ps(fscal,dx01);
2199             ty               = _mm256_mul_ps(fscal,dy01);
2200             tz               = _mm256_mul_ps(fscal,dz01);
2201
2202             /* Update vectorial force */
2203             fix0             = _mm256_add_ps(fix0,tx);
2204             fiy0             = _mm256_add_ps(fiy0,ty);
2205             fiz0             = _mm256_add_ps(fiz0,tz);
2206
2207             fjx1             = _mm256_add_ps(fjx1,tx);
2208             fjy1             = _mm256_add_ps(fjy1,ty);
2209             fjz1             = _mm256_add_ps(fjz1,tz);
2210
2211             }
2212
2213             /**************************
2214              * CALCULATE INTERACTIONS *
2215              **************************/
2216
2217             if (gmx_mm256_any_lt(rsq02,rcutoff2))
2218             {
2219
2220             r02              = _mm256_mul_ps(rsq02,rinv02);
2221             r02              = _mm256_andnot_ps(dummy_mask,r02);
2222
2223             /* EWALD ELECTROSTATICS */
2224
2225             /* Analytical PME correction */
2226             zeta2            = _mm256_mul_ps(beta2,rsq02);
2227             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
2228             pmecorrF         = avx256_pmecorrF_f(zeta2);
2229             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2230             felec            = _mm256_mul_ps(qq02,felec);
2231
2232             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
2233
2234             fscal            = felec;
2235
2236             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2237
2238             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2239
2240             /* Calculate temporary vectorial force */
2241             tx               = _mm256_mul_ps(fscal,dx02);
2242             ty               = _mm256_mul_ps(fscal,dy02);
2243             tz               = _mm256_mul_ps(fscal,dz02);
2244
2245             /* Update vectorial force */
2246             fix0             = _mm256_add_ps(fix0,tx);
2247             fiy0             = _mm256_add_ps(fiy0,ty);
2248             fiz0             = _mm256_add_ps(fiz0,tz);
2249
2250             fjx2             = _mm256_add_ps(fjx2,tx);
2251             fjy2             = _mm256_add_ps(fjy2,ty);
2252             fjz2             = _mm256_add_ps(fjz2,tz);
2253
2254             }
2255
2256             /**************************
2257              * CALCULATE INTERACTIONS *
2258              **************************/
2259
2260             if (gmx_mm256_any_lt(rsq10,rcutoff2))
2261             {
2262
2263             r10              = _mm256_mul_ps(rsq10,rinv10);
2264             r10              = _mm256_andnot_ps(dummy_mask,r10);
2265
2266             /* EWALD ELECTROSTATICS */
2267
2268             /* Analytical PME correction */
2269             zeta2            = _mm256_mul_ps(beta2,rsq10);
2270             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
2271             pmecorrF         = avx256_pmecorrF_f(zeta2);
2272             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2273             felec            = _mm256_mul_ps(qq10,felec);
2274
2275             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
2276
2277             fscal            = felec;
2278
2279             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2280
2281             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2282
2283             /* Calculate temporary vectorial force */
2284             tx               = _mm256_mul_ps(fscal,dx10);
2285             ty               = _mm256_mul_ps(fscal,dy10);
2286             tz               = _mm256_mul_ps(fscal,dz10);
2287
2288             /* Update vectorial force */
2289             fix1             = _mm256_add_ps(fix1,tx);
2290             fiy1             = _mm256_add_ps(fiy1,ty);
2291             fiz1             = _mm256_add_ps(fiz1,tz);
2292
2293             fjx0             = _mm256_add_ps(fjx0,tx);
2294             fjy0             = _mm256_add_ps(fjy0,ty);
2295             fjz0             = _mm256_add_ps(fjz0,tz);
2296
2297             }
2298
2299             /**************************
2300              * CALCULATE INTERACTIONS *
2301              **************************/
2302
2303             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2304             {
2305
2306             r11              = _mm256_mul_ps(rsq11,rinv11);
2307             r11              = _mm256_andnot_ps(dummy_mask,r11);
2308
2309             /* EWALD ELECTROSTATICS */
2310
2311             /* Analytical PME correction */
2312             zeta2            = _mm256_mul_ps(beta2,rsq11);
2313             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
2314             pmecorrF         = avx256_pmecorrF_f(zeta2);
2315             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2316             felec            = _mm256_mul_ps(qq11,felec);
2317
2318             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2319
2320             fscal            = felec;
2321
2322             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2323
2324             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2325
2326             /* Calculate temporary vectorial force */
2327             tx               = _mm256_mul_ps(fscal,dx11);
2328             ty               = _mm256_mul_ps(fscal,dy11);
2329             tz               = _mm256_mul_ps(fscal,dz11);
2330
2331             /* Update vectorial force */
2332             fix1             = _mm256_add_ps(fix1,tx);
2333             fiy1             = _mm256_add_ps(fiy1,ty);
2334             fiz1             = _mm256_add_ps(fiz1,tz);
2335
2336             fjx1             = _mm256_add_ps(fjx1,tx);
2337             fjy1             = _mm256_add_ps(fjy1,ty);
2338             fjz1             = _mm256_add_ps(fjz1,tz);
2339
2340             }
2341
2342             /**************************
2343              * CALCULATE INTERACTIONS *
2344              **************************/
2345
2346             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2347             {
2348
2349             r12              = _mm256_mul_ps(rsq12,rinv12);
2350             r12              = _mm256_andnot_ps(dummy_mask,r12);
2351
2352             /* EWALD ELECTROSTATICS */
2353
2354             /* Analytical PME correction */
2355             zeta2            = _mm256_mul_ps(beta2,rsq12);
2356             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2357             pmecorrF         = avx256_pmecorrF_f(zeta2);
2358             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2359             felec            = _mm256_mul_ps(qq12,felec);
2360
2361             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2362
2363             fscal            = felec;
2364
2365             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2366
2367             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2368
2369             /* Calculate temporary vectorial force */
2370             tx               = _mm256_mul_ps(fscal,dx12);
2371             ty               = _mm256_mul_ps(fscal,dy12);
2372             tz               = _mm256_mul_ps(fscal,dz12);
2373
2374             /* Update vectorial force */
2375             fix1             = _mm256_add_ps(fix1,tx);
2376             fiy1             = _mm256_add_ps(fiy1,ty);
2377             fiz1             = _mm256_add_ps(fiz1,tz);
2378
2379             fjx2             = _mm256_add_ps(fjx2,tx);
2380             fjy2             = _mm256_add_ps(fjy2,ty);
2381             fjz2             = _mm256_add_ps(fjz2,tz);
2382
2383             }
2384
2385             /**************************
2386              * CALCULATE INTERACTIONS *
2387              **************************/
2388
2389             if (gmx_mm256_any_lt(rsq20,rcutoff2))
2390             {
2391
2392             r20              = _mm256_mul_ps(rsq20,rinv20);
2393             r20              = _mm256_andnot_ps(dummy_mask,r20);
2394
2395             /* EWALD ELECTROSTATICS */
2396
2397             /* Analytical PME correction */
2398             zeta2            = _mm256_mul_ps(beta2,rsq20);
2399             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
2400             pmecorrF         = avx256_pmecorrF_f(zeta2);
2401             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2402             felec            = _mm256_mul_ps(qq20,felec);
2403
2404             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
2405
2406             fscal            = felec;
2407
2408             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2409
2410             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2411
2412             /* Calculate temporary vectorial force */
2413             tx               = _mm256_mul_ps(fscal,dx20);
2414             ty               = _mm256_mul_ps(fscal,dy20);
2415             tz               = _mm256_mul_ps(fscal,dz20);
2416
2417             /* Update vectorial force */
2418             fix2             = _mm256_add_ps(fix2,tx);
2419             fiy2             = _mm256_add_ps(fiy2,ty);
2420             fiz2             = _mm256_add_ps(fiz2,tz);
2421
2422             fjx0             = _mm256_add_ps(fjx0,tx);
2423             fjy0             = _mm256_add_ps(fjy0,ty);
2424             fjz0             = _mm256_add_ps(fjz0,tz);
2425
2426             }
2427
2428             /**************************
2429              * CALCULATE INTERACTIONS *
2430              **************************/
2431
2432             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2433             {
2434
2435             r21              = _mm256_mul_ps(rsq21,rinv21);
2436             r21              = _mm256_andnot_ps(dummy_mask,r21);
2437
2438             /* EWALD ELECTROSTATICS */
2439
2440             /* Analytical PME correction */
2441             zeta2            = _mm256_mul_ps(beta2,rsq21);
2442             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2443             pmecorrF         = avx256_pmecorrF_f(zeta2);
2444             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2445             felec            = _mm256_mul_ps(qq21,felec);
2446
2447             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2448
2449             fscal            = felec;
2450
2451             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2452
2453             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2454
2455             /* Calculate temporary vectorial force */
2456             tx               = _mm256_mul_ps(fscal,dx21);
2457             ty               = _mm256_mul_ps(fscal,dy21);
2458             tz               = _mm256_mul_ps(fscal,dz21);
2459
2460             /* Update vectorial force */
2461             fix2             = _mm256_add_ps(fix2,tx);
2462             fiy2             = _mm256_add_ps(fiy2,ty);
2463             fiz2             = _mm256_add_ps(fiz2,tz);
2464
2465             fjx1             = _mm256_add_ps(fjx1,tx);
2466             fjy1             = _mm256_add_ps(fjy1,ty);
2467             fjz1             = _mm256_add_ps(fjz1,tz);
2468
2469             }
2470
2471             /**************************
2472              * CALCULATE INTERACTIONS *
2473              **************************/
2474
2475             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2476             {
2477
2478             r22              = _mm256_mul_ps(rsq22,rinv22);
2479             r22              = _mm256_andnot_ps(dummy_mask,r22);
2480
2481             /* EWALD ELECTROSTATICS */
2482
2483             /* Analytical PME correction */
2484             zeta2            = _mm256_mul_ps(beta2,rsq22);
2485             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2486             pmecorrF         = avx256_pmecorrF_f(zeta2);
2487             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2488             felec            = _mm256_mul_ps(qq22,felec);
2489
2490             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2491
2492             fscal            = felec;
2493
2494             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2495
2496             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2497
2498             /* Calculate temporary vectorial force */
2499             tx               = _mm256_mul_ps(fscal,dx22);
2500             ty               = _mm256_mul_ps(fscal,dy22);
2501             tz               = _mm256_mul_ps(fscal,dz22);
2502
2503             /* Update vectorial force */
2504             fix2             = _mm256_add_ps(fix2,tx);
2505             fiy2             = _mm256_add_ps(fiy2,ty);
2506             fiz2             = _mm256_add_ps(fiz2,tz);
2507
2508             fjx2             = _mm256_add_ps(fjx2,tx);
2509             fjy2             = _mm256_add_ps(fjy2,ty);
2510             fjz2             = _mm256_add_ps(fjz2,tz);
2511
2512             }
2513
2514             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2515             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2516             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2517             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2518             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2519             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2520             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2521             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2522
2523             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2524                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2525
2526             /* Inner loop uses 540 flops */
2527         }
2528
2529         /* End of innermost loop */
2530
2531         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2532                                                  f+i_coord_offset,fshift+i_shift_offset);
2533
2534         /* Increment number of inner iterations */
2535         inneriter                  += j_index_end - j_index_start;
2536
2537         /* Outer loop uses 18 flops */
2538     }
2539
2540     /* Increment number of outer iterations */
2541     outeriter        += nri;
2542
2543     /* Update outer/inner flops */
2544
2545     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*540);
2546 }