src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_avx_256_double.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_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_VF_avx_256_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Water3
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_VF_avx_256_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     real *           vdwioffsetptr1;
  86     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     real *           vdwioffsetptr2;
  88     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  92     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  93     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  94     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  95     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  97     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  98     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  99     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 100     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 101     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 102     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 103     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 104     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 105     real             *charge;
 106     int              nvdwtype;
 107     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 108     int              *vdwtype;
 109     real             *vdwparam;
 110     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 111     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 112     __m128i          vfitab;
 113     __m128i          ifour       = _mm_set1_epi32(4);
 114     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 115     real             *vftab;
 116     __m256d          dummy_mask,cutoff_mask;
 117     __m128           tmpmask0,tmpmask1;
 118     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 119     __m256d          one     = _mm256_set1_pd(1.0);
 120     __m256d          two     = _mm256_set1_pd(2.0);
 121     x                = xx[0];
 122     f                = ff[0];
 123
 124     nri              = nlist->nri;
 125     iinr             = nlist->iinr;
 126     jindex           = nlist->jindex;
 127     jjnr             = nlist->jjnr;
 128     shiftidx         = nlist->shift;
 129     gid              = nlist->gid;
 130     shiftvec         = fr->shift_vec[0];
 131     fshift           = fr->fshift[0];
 132     facel            = _mm256_set1_pd(fr->ic->epsfac);
 133     charge           = mdatoms->chargeA;
 134     krf              = _mm256_set1_pd(fr->ic->k_rf);
 135     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 136     crf              = _mm256_set1_pd(fr->ic->c_rf);
 137     nvdwtype         = fr->ntype;
 138     vdwparam         = fr->nbfp;
 139     vdwtype          = mdatoms->typeA;
 140
 141     vftab            = kernel_data->table_vdw->data;
 142     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 143
 144     /* Setup water-specific parameters */
 145     inr              = nlist->iinr[0];
 146     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 147     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 148     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 149     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 150
 151     jq0              = _mm256_set1_pd(charge[inr+0]);
 152     jq1              = _mm256_set1_pd(charge[inr+1]);
 153     jq2              = _mm256_set1_pd(charge[inr+2]);
 154     vdwjidx0A        = 2*vdwtype[inr+0];
 155     qq00             = _mm256_mul_pd(iq0,jq0);
 156     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 157     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 158     qq01             = _mm256_mul_pd(iq0,jq1);
 159     qq02             = _mm256_mul_pd(iq0,jq2);
 160     qq10             = _mm256_mul_pd(iq1,jq0);
 161     qq11             = _mm256_mul_pd(iq1,jq1);
 162     qq12             = _mm256_mul_pd(iq1,jq2);
 163     qq20             = _mm256_mul_pd(iq2,jq0);
 164     qq21             = _mm256_mul_pd(iq2,jq1);
 165     qq22             = _mm256_mul_pd(iq2,jq2);
 166
 167     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 168     rcutoff_scalar   = fr->ic->rcoulomb;
 169     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 170     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 171
 172     /* Avoid stupid compiler warnings */
 173     jnrA = jnrB = jnrC = jnrD = 0;
 174     j_coord_offsetA = 0;
 175     j_coord_offsetB = 0;
 176     j_coord_offsetC = 0;
 177     j_coord_offsetD = 0;
 178
 179     outeriter        = 0;
 180     inneriter        = 0;
 181
 182     for(iidx=0;iidx<4*DIM;iidx++)
 183     {
 184         scratch[iidx] = 0.0;
 185     }
 186
 187     /* Start outer loop over neighborlists */
 188     for(iidx=0; iidx<nri; iidx++)
 189     {
 190         /* Load shift vector for this list */
 191         i_shift_offset   = DIM*shiftidx[iidx];
 192
 193         /* Load limits for loop over neighbors */
 194         j_index_start    = jindex[iidx];
 195         j_index_end      = jindex[iidx+1];
 196
 197         /* Get outer coordinate index */
 198         inr              = iinr[iidx];
 199         i_coord_offset   = DIM*inr;
 200
 201         /* Load i particle coords and add shift vector */
 202         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 203                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 204
 205         fix0             = _mm256_setzero_pd();
 206         fiy0             = _mm256_setzero_pd();
 207         fiz0             = _mm256_setzero_pd();
 208         fix1             = _mm256_setzero_pd();
 209         fiy1             = _mm256_setzero_pd();
 210         fiz1             = _mm256_setzero_pd();
 211         fix2             = _mm256_setzero_pd();
 212         fiy2             = _mm256_setzero_pd();
 213         fiz2             = _mm256_setzero_pd();
 214
 215         /* Reset potential sums */
 216         velecsum         = _mm256_setzero_pd();
 217         vvdwsum          = _mm256_setzero_pd();
 218
 219         /* Start inner kernel loop */
 220         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 221         {
 222
 223             /* Get j neighbor index, and coordinate index */
 224             jnrA             = jjnr[jidx];
 225             jnrB             = jjnr[jidx+1];
 226             jnrC             = jjnr[jidx+2];
 227             jnrD             = jjnr[jidx+3];
 228             j_coord_offsetA  = DIM*jnrA;
 229             j_coord_offsetB  = DIM*jnrB;
 230             j_coord_offsetC  = DIM*jnrC;
 231             j_coord_offsetD  = DIM*jnrD;
 232
 233             /* load j atom coordinates */
 234             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 235                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 236                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 237
 238             /* Calculate displacement vector */
 239             dx00             = _mm256_sub_pd(ix0,jx0);
 240             dy00             = _mm256_sub_pd(iy0,jy0);
 241             dz00             = _mm256_sub_pd(iz0,jz0);
 242             dx01             = _mm256_sub_pd(ix0,jx1);
 243             dy01             = _mm256_sub_pd(iy0,jy1);
 244             dz01             = _mm256_sub_pd(iz0,jz1);
 245             dx02             = _mm256_sub_pd(ix0,jx2);
 246             dy02             = _mm256_sub_pd(iy0,jy2);
 247             dz02             = _mm256_sub_pd(iz0,jz2);
 248             dx10             = _mm256_sub_pd(ix1,jx0);
 249             dy10             = _mm256_sub_pd(iy1,jy0);
 250             dz10             = _mm256_sub_pd(iz1,jz0);
 251             dx11             = _mm256_sub_pd(ix1,jx1);
 252             dy11             = _mm256_sub_pd(iy1,jy1);
 253             dz11             = _mm256_sub_pd(iz1,jz1);
 254             dx12             = _mm256_sub_pd(ix1,jx2);
 255             dy12             = _mm256_sub_pd(iy1,jy2);
 256             dz12             = _mm256_sub_pd(iz1,jz2);
 257             dx20             = _mm256_sub_pd(ix2,jx0);
 258             dy20             = _mm256_sub_pd(iy2,jy0);
 259             dz20             = _mm256_sub_pd(iz2,jz0);
 260             dx21             = _mm256_sub_pd(ix2,jx1);
 261             dy21             = _mm256_sub_pd(iy2,jy1);
 262             dz21             = _mm256_sub_pd(iz2,jz1);
 263             dx22             = _mm256_sub_pd(ix2,jx2);
 264             dy22             = _mm256_sub_pd(iy2,jy2);
 265             dz22             = _mm256_sub_pd(iz2,jz2);
 266
 267             /* Calculate squared distance and things based on it */
 268             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 269             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 270             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 271             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 272             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 273             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 274             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 275             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 276             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 277
 278             rinv00           = avx256_invsqrt_d(rsq00);
 279             rinv01           = avx256_invsqrt_d(rsq01);
 280             rinv02           = avx256_invsqrt_d(rsq02);
 281             rinv10           = avx256_invsqrt_d(rsq10);
 282             rinv11           = avx256_invsqrt_d(rsq11);
 283             rinv12           = avx256_invsqrt_d(rsq12);
 284             rinv20           = avx256_invsqrt_d(rsq20);
 285             rinv21           = avx256_invsqrt_d(rsq21);
 286             rinv22           = avx256_invsqrt_d(rsq22);
 287
 288             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 289             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 290             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 291             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 292             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 293             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 294             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 295             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 296             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 297
 298             fjx0             = _mm256_setzero_pd();
 299             fjy0             = _mm256_setzero_pd();
 300             fjz0             = _mm256_setzero_pd();
 301             fjx1             = _mm256_setzero_pd();
 302             fjy1             = _mm256_setzero_pd();
 303             fjz1             = _mm256_setzero_pd();
 304             fjx2             = _mm256_setzero_pd();
 305             fjy2             = _mm256_setzero_pd();
 306             fjz2             = _mm256_setzero_pd();
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 313             {
 314
 315             r00              = _mm256_mul_pd(rsq00,rinv00);
 316
 317             /* Calculate table index by multiplying r with table scale and truncate to integer */
 318             rt               = _mm256_mul_pd(r00,vftabscale);
 319             vfitab           = _mm256_cvttpd_epi32(rt);
 320             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 321             vfitab           = _mm_slli_epi32(vfitab,3);
 322
 323             /* REACTION-FIELD ELECTROSTATICS */
 324             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 325             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 326
 327             /* CUBIC SPLINE TABLE DISPERSION */
 328             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 329             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 330             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 331             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 332             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 333             Heps             = _mm256_mul_pd(vfeps,H);
 334             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 335             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 336             vvdw6            = _mm256_mul_pd(c6_00,VV);
 337             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 338             fvdw6            = _mm256_mul_pd(c6_00,FF);
 339
 340             /* CUBIC SPLINE TABLE REPULSION */
 341             vfitab           = _mm_add_epi32(vfitab,ifour);
 342             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 343             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 344             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 345             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 346             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 347             Heps             = _mm256_mul_pd(vfeps,H);
 348             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 349             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 350             vvdw12           = _mm256_mul_pd(c12_00,VV);
 351             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 352             fvdw12           = _mm256_mul_pd(c12_00,FF);
 353             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 354             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 355
 356             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velec            = _mm256_and_pd(velec,cutoff_mask);
 360             velecsum         = _mm256_add_pd(velecsum,velec);
 361             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 362             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 363
 364             fscal            = _mm256_add_pd(felec,fvdw);
 365
 366             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 367
 368             /* Calculate temporary vectorial force */
 369             tx               = _mm256_mul_pd(fscal,dx00);
 370             ty               = _mm256_mul_pd(fscal,dy00);
 371             tz               = _mm256_mul_pd(fscal,dz00);
 372
 373             /* Update vectorial force */
 374             fix0             = _mm256_add_pd(fix0,tx);
 375             fiy0             = _mm256_add_pd(fiy0,ty);
 376             fiz0             = _mm256_add_pd(fiz0,tz);
 377
 378             fjx0             = _mm256_add_pd(fjx0,tx);
 379             fjy0             = _mm256_add_pd(fjy0,ty);
 380             fjz0             = _mm256_add_pd(fjz0,tz);
 381
 382             }
 383
 384             /**************************
 385              * CALCULATE INTERACTIONS *
 386              **************************/
 387
 388             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 389             {
 390
 391             /* REACTION-FIELD ELECTROSTATICS */
 392             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
 393             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
 394
 395             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
 396
 397             /* Update potential sum for this i atom from the interaction with this j atom. */
 398             velec            = _mm256_and_pd(velec,cutoff_mask);
 399             velecsum         = _mm256_add_pd(velecsum,velec);
 400
 401             fscal            = felec;
 402
 403             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 404
 405             /* Calculate temporary vectorial force */
 406             tx               = _mm256_mul_pd(fscal,dx01);
 407             ty               = _mm256_mul_pd(fscal,dy01);
 408             tz               = _mm256_mul_pd(fscal,dz01);
 409
 410             /* Update vectorial force */
 411             fix0             = _mm256_add_pd(fix0,tx);
 412             fiy0             = _mm256_add_pd(fiy0,ty);
 413             fiz0             = _mm256_add_pd(fiz0,tz);
 414
 415             fjx1             = _mm256_add_pd(fjx1,tx);
 416             fjy1             = _mm256_add_pd(fjy1,ty);
 417             fjz1             = _mm256_add_pd(fjz1,tz);
 418
 419             }
 420
 421             /**************************
 422              * CALCULATE INTERACTIONS *
 423              **************************/
 424
 425             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 426             {
 427
 428             /* REACTION-FIELD ELECTROSTATICS */
 429             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
 430             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
 431
 432             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
 433
 434             /* Update potential sum for this i atom from the interaction with this j atom. */
 435             velec            = _mm256_and_pd(velec,cutoff_mask);
 436             velecsum         = _mm256_add_pd(velecsum,velec);
 437
 438             fscal            = felec;
 439
 440             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 441
 442             /* Calculate temporary vectorial force */
 443             tx               = _mm256_mul_pd(fscal,dx02);
 444             ty               = _mm256_mul_pd(fscal,dy02);
 445             tz               = _mm256_mul_pd(fscal,dz02);
 446
 447             /* Update vectorial force */
 448             fix0             = _mm256_add_pd(fix0,tx);
 449             fiy0             = _mm256_add_pd(fiy0,ty);
 450             fiz0             = _mm256_add_pd(fiz0,tz);
 451
 452             fjx2             = _mm256_add_pd(fjx2,tx);
 453             fjy2             = _mm256_add_pd(fjy2,ty);
 454             fjz2             = _mm256_add_pd(fjz2,tz);
 455
 456             }
 457
 458             /**************************
 459              * CALCULATE INTERACTIONS *
 460              **************************/
 461
 462             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 463             {
 464
 465             /* REACTION-FIELD ELECTROSTATICS */
 466             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 467             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 468
 469             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 470
 471             /* Update potential sum for this i atom from the interaction with this j atom. */
 472             velec            = _mm256_and_pd(velec,cutoff_mask);
 473             velecsum         = _mm256_add_pd(velecsum,velec);
 474
 475             fscal            = felec;
 476
 477             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 478
 479             /* Calculate temporary vectorial force */
 480             tx               = _mm256_mul_pd(fscal,dx10);
 481             ty               = _mm256_mul_pd(fscal,dy10);
 482             tz               = _mm256_mul_pd(fscal,dz10);
 483
 484             /* Update vectorial force */
 485             fix1             = _mm256_add_pd(fix1,tx);
 486             fiy1             = _mm256_add_pd(fiy1,ty);
 487             fiz1             = _mm256_add_pd(fiz1,tz);
 488
 489             fjx0             = _mm256_add_pd(fjx0,tx);
 490             fjy0             = _mm256_add_pd(fjy0,ty);
 491             fjz0             = _mm256_add_pd(fjz0,tz);
 492
 493             }
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 500             {
 501
 502             /* REACTION-FIELD ELECTROSTATICS */
 503             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 504             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 505
 506             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 507
 508             /* Update potential sum for this i atom from the interaction with this j atom. */
 509             velec            = _mm256_and_pd(velec,cutoff_mask);
 510             velecsum         = _mm256_add_pd(velecsum,velec);
 511
 512             fscal            = felec;
 513
 514             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 515
 516             /* Calculate temporary vectorial force */
 517             tx               = _mm256_mul_pd(fscal,dx11);
 518             ty               = _mm256_mul_pd(fscal,dy11);
 519             tz               = _mm256_mul_pd(fscal,dz11);
 520
 521             /* Update vectorial force */
 522             fix1             = _mm256_add_pd(fix1,tx);
 523             fiy1             = _mm256_add_pd(fiy1,ty);
 524             fiz1             = _mm256_add_pd(fiz1,tz);
 525
 526             fjx1             = _mm256_add_pd(fjx1,tx);
 527             fjy1             = _mm256_add_pd(fjy1,ty);
 528             fjz1             = _mm256_add_pd(fjz1,tz);
 529
 530             }
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 537             {
 538
 539             /* REACTION-FIELD ELECTROSTATICS */
 540             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 541             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 542
 543             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 544
 545             /* Update potential sum for this i atom from the interaction with this j atom. */
 546             velec            = _mm256_and_pd(velec,cutoff_mask);
 547             velecsum         = _mm256_add_pd(velecsum,velec);
 548
 549             fscal            = felec;
 550
 551             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 552
 553             /* Calculate temporary vectorial force */
 554             tx               = _mm256_mul_pd(fscal,dx12);
 555             ty               = _mm256_mul_pd(fscal,dy12);
 556             tz               = _mm256_mul_pd(fscal,dz12);
 557
 558             /* Update vectorial force */
 559             fix1             = _mm256_add_pd(fix1,tx);
 560             fiy1             = _mm256_add_pd(fiy1,ty);
 561             fiz1             = _mm256_add_pd(fiz1,tz);
 562
 563             fjx2             = _mm256_add_pd(fjx2,tx);
 564             fjy2             = _mm256_add_pd(fjy2,ty);
 565             fjz2             = _mm256_add_pd(fjz2,tz);
 566
 567             }
 568
 569             /**************************
 570              * CALCULATE INTERACTIONS *
 571              **************************/
 572
 573             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 574             {
 575
 576             /* REACTION-FIELD ELECTROSTATICS */
 577             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 578             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 579
 580             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 581
 582             /* Update potential sum for this i atom from the interaction with this j atom. */
 583             velec            = _mm256_and_pd(velec,cutoff_mask);
 584             velecsum         = _mm256_add_pd(velecsum,velec);
 585
 586             fscal            = felec;
 587
 588             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 589
 590             /* Calculate temporary vectorial force */
 591             tx               = _mm256_mul_pd(fscal,dx20);
 592             ty               = _mm256_mul_pd(fscal,dy20);
 593             tz               = _mm256_mul_pd(fscal,dz20);
 594
 595             /* Update vectorial force */
 596             fix2             = _mm256_add_pd(fix2,tx);
 597             fiy2             = _mm256_add_pd(fiy2,ty);
 598             fiz2             = _mm256_add_pd(fiz2,tz);
 599
 600             fjx0             = _mm256_add_pd(fjx0,tx);
 601             fjy0             = _mm256_add_pd(fjy0,ty);
 602             fjz0             = _mm256_add_pd(fjz0,tz);
 603
 604             }
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 611             {
 612
 613             /* REACTION-FIELD ELECTROSTATICS */
 614             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 615             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 616
 617             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
 618
 619             /* Update potential sum for this i atom from the interaction with this j atom. */
 620             velec            = _mm256_and_pd(velec,cutoff_mask);
 621             velecsum         = _mm256_add_pd(velecsum,velec);
 622
 623             fscal            = felec;
 624
 625             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 626
 627             /* Calculate temporary vectorial force */
 628             tx               = _mm256_mul_pd(fscal,dx21);
 629             ty               = _mm256_mul_pd(fscal,dy21);
 630             tz               = _mm256_mul_pd(fscal,dz21);
 631
 632             /* Update vectorial force */
 633             fix2             = _mm256_add_pd(fix2,tx);
 634             fiy2             = _mm256_add_pd(fiy2,ty);
 635             fiz2             = _mm256_add_pd(fiz2,tz);
 636
 637             fjx1             = _mm256_add_pd(fjx1,tx);
 638             fjy1             = _mm256_add_pd(fjy1,ty);
 639             fjz1             = _mm256_add_pd(fjz1,tz);
 640
 641             }
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 648             {
 649
 650             /* REACTION-FIELD ELECTROSTATICS */
 651             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 652             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 653
 654             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
 655
 656             /* Update potential sum for this i atom from the interaction with this j atom. */
 657             velec            = _mm256_and_pd(velec,cutoff_mask);
 658             velecsum         = _mm256_add_pd(velecsum,velec);
 659
 660             fscal            = felec;
 661
 662             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 663
 664             /* Calculate temporary vectorial force */
 665             tx               = _mm256_mul_pd(fscal,dx22);
 666             ty               = _mm256_mul_pd(fscal,dy22);
 667             tz               = _mm256_mul_pd(fscal,dz22);
 668
 669             /* Update vectorial force */
 670             fix2             = _mm256_add_pd(fix2,tx);
 671             fiy2             = _mm256_add_pd(fiy2,ty);
 672             fiz2             = _mm256_add_pd(fiz2,tz);
 673
 674             fjx2             = _mm256_add_pd(fjx2,tx);
 675             fjy2             = _mm256_add_pd(fjy2,ty);
 676             fjz2             = _mm256_add_pd(fjz2,tz);
 677
 678             }
 679
 680             fjptrA             = f+j_coord_offsetA;
 681             fjptrB             = f+j_coord_offsetB;
 682             fjptrC             = f+j_coord_offsetC;
 683             fjptrD             = f+j_coord_offsetD;
 684
 685             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 686                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 687
 688             /* Inner loop uses 360 flops */
 689         }
 690
 691         if(jidx<j_index_end)
 692         {
 693
 694             /* Get j neighbor index, and coordinate index */
 695             jnrlistA         = jjnr[jidx];
 696             jnrlistB         = jjnr[jidx+1];
 697             jnrlistC         = jjnr[jidx+2];
 698             jnrlistD         = jjnr[jidx+3];
 699             /* Sign of each element will be negative for non-real atoms.
 700              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 701              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 702              */
 703             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 704
 705             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 706             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 707             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 708
 709             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 710             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 711             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 712             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 713             j_coord_offsetA  = DIM*jnrA;
 714             j_coord_offsetB  = DIM*jnrB;
 715             j_coord_offsetC  = DIM*jnrC;
 716             j_coord_offsetD  = DIM*jnrD;
 717
 718             /* load j atom coordinates */
 719             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 720                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 721                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 722
 723             /* Calculate displacement vector */
 724             dx00             = _mm256_sub_pd(ix0,jx0);
 725             dy00             = _mm256_sub_pd(iy0,jy0);
 726             dz00             = _mm256_sub_pd(iz0,jz0);
 727             dx01             = _mm256_sub_pd(ix0,jx1);
 728             dy01             = _mm256_sub_pd(iy0,jy1);
 729             dz01             = _mm256_sub_pd(iz0,jz1);
 730             dx02             = _mm256_sub_pd(ix0,jx2);
 731             dy02             = _mm256_sub_pd(iy0,jy2);
 732             dz02             = _mm256_sub_pd(iz0,jz2);
 733             dx10             = _mm256_sub_pd(ix1,jx0);
 734             dy10             = _mm256_sub_pd(iy1,jy0);
 735             dz10             = _mm256_sub_pd(iz1,jz0);
 736             dx11             = _mm256_sub_pd(ix1,jx1);
 737             dy11             = _mm256_sub_pd(iy1,jy1);
 738             dz11             = _mm256_sub_pd(iz1,jz1);
 739             dx12             = _mm256_sub_pd(ix1,jx2);
 740             dy12             = _mm256_sub_pd(iy1,jy2);
 741             dz12             = _mm256_sub_pd(iz1,jz2);
 742             dx20             = _mm256_sub_pd(ix2,jx0);
 743             dy20             = _mm256_sub_pd(iy2,jy0);
 744             dz20             = _mm256_sub_pd(iz2,jz0);
 745             dx21             = _mm256_sub_pd(ix2,jx1);
 746             dy21             = _mm256_sub_pd(iy2,jy1);
 747             dz21             = _mm256_sub_pd(iz2,jz1);
 748             dx22             = _mm256_sub_pd(ix2,jx2);
 749             dy22             = _mm256_sub_pd(iy2,jy2);
 750             dz22             = _mm256_sub_pd(iz2,jz2);
 751
 752             /* Calculate squared distance and things based on it */
 753             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 754             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 755             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 756             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 757             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 758             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 759             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 760             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 761             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 762
 763             rinv00           = avx256_invsqrt_d(rsq00);
 764             rinv01           = avx256_invsqrt_d(rsq01);
 765             rinv02           = avx256_invsqrt_d(rsq02);
 766             rinv10           = avx256_invsqrt_d(rsq10);
 767             rinv11           = avx256_invsqrt_d(rsq11);
 768             rinv12           = avx256_invsqrt_d(rsq12);
 769             rinv20           = avx256_invsqrt_d(rsq20);
 770             rinv21           = avx256_invsqrt_d(rsq21);
 771             rinv22           = avx256_invsqrt_d(rsq22);
 772
 773             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 774             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 775             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 776             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 777             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 778             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 779             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 780             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 781             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 782
 783             fjx0             = _mm256_setzero_pd();
 784             fjy0             = _mm256_setzero_pd();
 785             fjz0             = _mm256_setzero_pd();
 786             fjx1             = _mm256_setzero_pd();
 787             fjy1             = _mm256_setzero_pd();
 788             fjz1             = _mm256_setzero_pd();
 789             fjx2             = _mm256_setzero_pd();
 790             fjy2             = _mm256_setzero_pd();
 791             fjz2             = _mm256_setzero_pd();
 792
 793             /**************************
 794              * CALCULATE INTERACTIONS *
 795              **************************/
 796
 797             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 798             {
 799
 800             r00              = _mm256_mul_pd(rsq00,rinv00);
 801             r00              = _mm256_andnot_pd(dummy_mask,r00);
 802
 803             /* Calculate table index by multiplying r with table scale and truncate to integer */
 804             rt               = _mm256_mul_pd(r00,vftabscale);
 805             vfitab           = _mm256_cvttpd_epi32(rt);
 806             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 807             vfitab           = _mm_slli_epi32(vfitab,3);
 808
 809             /* REACTION-FIELD ELECTROSTATICS */
 810             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 811             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 812
 813             /* CUBIC SPLINE TABLE DISPERSION */
 814             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 815             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 816             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 817             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 818             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 819             Heps             = _mm256_mul_pd(vfeps,H);
 820             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 821             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 822             vvdw6            = _mm256_mul_pd(c6_00,VV);
 823             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 824             fvdw6            = _mm256_mul_pd(c6_00,FF);
 825
 826             /* CUBIC SPLINE TABLE REPULSION */
 827             vfitab           = _mm_add_epi32(vfitab,ifour);
 828             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 829             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 830             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 831             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 832             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 833             Heps             = _mm256_mul_pd(vfeps,H);
 834             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 835             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 836             vvdw12           = _mm256_mul_pd(c12_00,VV);
 837             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 838             fvdw12           = _mm256_mul_pd(c12_00,FF);
 839             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 840             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 841
 842             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 843
 844             /* Update potential sum for this i atom from the interaction with this j atom. */
 845             velec            = _mm256_and_pd(velec,cutoff_mask);
 846             velec            = _mm256_andnot_pd(dummy_mask,velec);
 847             velecsum         = _mm256_add_pd(velecsum,velec);
 848             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 849             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 850             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 851
 852             fscal            = _mm256_add_pd(felec,fvdw);
 853
 854             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 855
 856             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 857
 858             /* Calculate temporary vectorial force */
 859             tx               = _mm256_mul_pd(fscal,dx00);
 860             ty               = _mm256_mul_pd(fscal,dy00);
 861             tz               = _mm256_mul_pd(fscal,dz00);
 862
 863             /* Update vectorial force */
 864             fix0             = _mm256_add_pd(fix0,tx);
 865             fiy0             = _mm256_add_pd(fiy0,ty);
 866             fiz0             = _mm256_add_pd(fiz0,tz);
 867
 868             fjx0             = _mm256_add_pd(fjx0,tx);
 869             fjy0             = _mm256_add_pd(fjy0,ty);
 870             fjz0             = _mm256_add_pd(fjz0,tz);
 871
 872             }
 873
 874             /**************************
 875              * CALCULATE INTERACTIONS *
 876              **************************/
 877
 878             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 879             {
 880
 881             /* REACTION-FIELD ELECTROSTATICS */
 882             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
 883             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
 884
 885             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
 886
 887             /* Update potential sum for this i atom from the interaction with this j atom. */
 888             velec            = _mm256_and_pd(velec,cutoff_mask);
 889             velec            = _mm256_andnot_pd(dummy_mask,velec);
 890             velecsum         = _mm256_add_pd(velecsum,velec);
 891
 892             fscal            = felec;
 893
 894             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 895
 896             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 897
 898             /* Calculate temporary vectorial force */
 899             tx               = _mm256_mul_pd(fscal,dx01);
 900             ty               = _mm256_mul_pd(fscal,dy01);
 901             tz               = _mm256_mul_pd(fscal,dz01);
 902
 903             /* Update vectorial force */
 904             fix0             = _mm256_add_pd(fix0,tx);
 905             fiy0             = _mm256_add_pd(fiy0,ty);
 906             fiz0             = _mm256_add_pd(fiz0,tz);
 907
 908             fjx1             = _mm256_add_pd(fjx1,tx);
 909             fjy1             = _mm256_add_pd(fjy1,ty);
 910             fjz1             = _mm256_add_pd(fjz1,tz);
 911
 912             }
 913
 914             /**************************
 915              * CALCULATE INTERACTIONS *
 916              **************************/
 917
 918             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 919             {
 920
 921             /* REACTION-FIELD ELECTROSTATICS */
 922             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
 923             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
 924
 925             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
 926
 927             /* Update potential sum for this i atom from the interaction with this j atom. */
 928             velec            = _mm256_and_pd(velec,cutoff_mask);
 929             velec            = _mm256_andnot_pd(dummy_mask,velec);
 930             velecsum         = _mm256_add_pd(velecsum,velec);
 931
 932             fscal            = felec;
 933
 934             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 935
 936             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 937
 938             /* Calculate temporary vectorial force */
 939             tx               = _mm256_mul_pd(fscal,dx02);
 940             ty               = _mm256_mul_pd(fscal,dy02);
 941             tz               = _mm256_mul_pd(fscal,dz02);
 942
 943             /* Update vectorial force */
 944             fix0             = _mm256_add_pd(fix0,tx);
 945             fiy0             = _mm256_add_pd(fiy0,ty);
 946             fiz0             = _mm256_add_pd(fiz0,tz);
 947
 948             fjx2             = _mm256_add_pd(fjx2,tx);
 949             fjy2             = _mm256_add_pd(fjy2,ty);
 950             fjz2             = _mm256_add_pd(fjz2,tz);
 951
 952             }
 953
 954             /**************************
 955              * CALCULATE INTERACTIONS *
 956              **************************/
 957
 958             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 959             {
 960
 961             /* REACTION-FIELD ELECTROSTATICS */
 962             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 963             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 964
 965             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 966
 967             /* Update potential sum for this i atom from the interaction with this j atom. */
 968             velec            = _mm256_and_pd(velec,cutoff_mask);
 969             velec            = _mm256_andnot_pd(dummy_mask,velec);
 970             velecsum         = _mm256_add_pd(velecsum,velec);
 971
 972             fscal            = felec;
 973
 974             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 975
 976             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 977
 978             /* Calculate temporary vectorial force */
 979             tx               = _mm256_mul_pd(fscal,dx10);
 980             ty               = _mm256_mul_pd(fscal,dy10);
 981             tz               = _mm256_mul_pd(fscal,dz10);
 982
 983             /* Update vectorial force */
 984             fix1             = _mm256_add_pd(fix1,tx);
 985             fiy1             = _mm256_add_pd(fiy1,ty);
 986             fiz1             = _mm256_add_pd(fiz1,tz);
 987
 988             fjx0             = _mm256_add_pd(fjx0,tx);
 989             fjy0             = _mm256_add_pd(fjy0,ty);
 990             fjz0             = _mm256_add_pd(fjz0,tz);
 991
 992             }
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 999             {
1000
1001             /* REACTION-FIELD ELECTROSTATICS */
1002             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
1003             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1004
1005             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1006
1007             /* Update potential sum for this i atom from the interaction with this j atom. */
1008             velec            = _mm256_and_pd(velec,cutoff_mask);
1009             velec            = _mm256_andnot_pd(dummy_mask,velec);
1010             velecsum         = _mm256_add_pd(velecsum,velec);
1011
1012             fscal            = felec;
1013
1014             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1015
1016             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1017
1018             /* Calculate temporary vectorial force */
1019             tx               = _mm256_mul_pd(fscal,dx11);
1020             ty               = _mm256_mul_pd(fscal,dy11);
1021             tz               = _mm256_mul_pd(fscal,dz11);
1022
1023             /* Update vectorial force */
1024             fix1             = _mm256_add_pd(fix1,tx);
1025             fiy1             = _mm256_add_pd(fiy1,ty);
1026             fiz1             = _mm256_add_pd(fiz1,tz);
1027
1028             fjx1             = _mm256_add_pd(fjx1,tx);
1029             fjy1             = _mm256_add_pd(fjy1,ty);
1030             fjz1             = _mm256_add_pd(fjz1,tz);
1031
1032             }
1033
1034             /**************************
1035              * CALCULATE INTERACTIONS *
1036              **************************/
1037
1038             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1039             {
1040
1041             /* REACTION-FIELD ELECTROSTATICS */
1042             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
1043             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1044
1045             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1046
1047             /* Update potential sum for this i atom from the interaction with this j atom. */
1048             velec            = _mm256_and_pd(velec,cutoff_mask);
1049             velec            = _mm256_andnot_pd(dummy_mask,velec);
1050             velecsum         = _mm256_add_pd(velecsum,velec);
1051
1052             fscal            = felec;
1053
1054             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1055
1056             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1057
1058             /* Calculate temporary vectorial force */
1059             tx               = _mm256_mul_pd(fscal,dx12);
1060             ty               = _mm256_mul_pd(fscal,dy12);
1061             tz               = _mm256_mul_pd(fscal,dz12);
1062
1063             /* Update vectorial force */
1064             fix1             = _mm256_add_pd(fix1,tx);
1065             fiy1             = _mm256_add_pd(fiy1,ty);
1066             fiz1             = _mm256_add_pd(fiz1,tz);
1067
1068             fjx2             = _mm256_add_pd(fjx2,tx);
1069             fjy2             = _mm256_add_pd(fjy2,ty);
1070             fjz2             = _mm256_add_pd(fjz2,tz);
1071
1072             }
1073
1074             /**************************
1075              * CALCULATE INTERACTIONS *
1076              **************************/
1077
1078             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1079             {
1080
1081             /* REACTION-FIELD ELECTROSTATICS */
1082             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
1083             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1084
1085             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1086
1087             /* Update potential sum for this i atom from the interaction with this j atom. */
1088             velec            = _mm256_and_pd(velec,cutoff_mask);
1089             velec            = _mm256_andnot_pd(dummy_mask,velec);
1090             velecsum         = _mm256_add_pd(velecsum,velec);
1091
1092             fscal            = felec;
1093
1094             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1095
1096             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1097
1098             /* Calculate temporary vectorial force */
1099             tx               = _mm256_mul_pd(fscal,dx20);
1100             ty               = _mm256_mul_pd(fscal,dy20);
1101             tz               = _mm256_mul_pd(fscal,dz20);
1102
1103             /* Update vectorial force */
1104             fix2             = _mm256_add_pd(fix2,tx);
1105             fiy2             = _mm256_add_pd(fiy2,ty);
1106             fiz2             = _mm256_add_pd(fiz2,tz);
1107
1108             fjx0             = _mm256_add_pd(fjx0,tx);
1109             fjy0             = _mm256_add_pd(fjy0,ty);
1110             fjz0             = _mm256_add_pd(fjz0,tz);
1111
1112             }
1113
1114             /**************************
1115              * CALCULATE INTERACTIONS *
1116              **************************/
1117
1118             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1119             {
1120
1121             /* REACTION-FIELD ELECTROSTATICS */
1122             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
1123             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1124
1125             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1126
1127             /* Update potential sum for this i atom from the interaction with this j atom. */
1128             velec            = _mm256_and_pd(velec,cutoff_mask);
1129             velec            = _mm256_andnot_pd(dummy_mask,velec);
1130             velecsum         = _mm256_add_pd(velecsum,velec);
1131
1132             fscal            = felec;
1133
1134             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1135
1136             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1137
1138             /* Calculate temporary vectorial force */
1139             tx               = _mm256_mul_pd(fscal,dx21);
1140             ty               = _mm256_mul_pd(fscal,dy21);
1141             tz               = _mm256_mul_pd(fscal,dz21);
1142
1143             /* Update vectorial force */
1144             fix2             = _mm256_add_pd(fix2,tx);
1145             fiy2             = _mm256_add_pd(fiy2,ty);
1146             fiz2             = _mm256_add_pd(fiz2,tz);
1147
1148             fjx1             = _mm256_add_pd(fjx1,tx);
1149             fjy1             = _mm256_add_pd(fjy1,ty);
1150             fjz1             = _mm256_add_pd(fjz1,tz);
1151
1152             }
1153
1154             /**************************
1155              * CALCULATE INTERACTIONS *
1156              **************************/
1157
1158             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1159             {
1160
1161             /* REACTION-FIELD ELECTROSTATICS */
1162             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
1163             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1164
1165             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1166
1167             /* Update potential sum for this i atom from the interaction with this j atom. */
1168             velec            = _mm256_and_pd(velec,cutoff_mask);
1169             velec            = _mm256_andnot_pd(dummy_mask,velec);
1170             velecsum         = _mm256_add_pd(velecsum,velec);
1171
1172             fscal            = felec;
1173
1174             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1175
1176             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1177
1178             /* Calculate temporary vectorial force */
1179             tx               = _mm256_mul_pd(fscal,dx22);
1180             ty               = _mm256_mul_pd(fscal,dy22);
1181             tz               = _mm256_mul_pd(fscal,dz22);
1182
1183             /* Update vectorial force */
1184             fix2             = _mm256_add_pd(fix2,tx);
1185             fiy2             = _mm256_add_pd(fiy2,ty);
1186             fiz2             = _mm256_add_pd(fiz2,tz);
1187
1188             fjx2             = _mm256_add_pd(fjx2,tx);
1189             fjy2             = _mm256_add_pd(fjy2,ty);
1190             fjz2             = _mm256_add_pd(fjz2,tz);
1191
1192             }
1193
1194             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1195             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1196             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1197             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1198
1199             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1200                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1201
1202             /* Inner loop uses 361 flops */
1203         }
1204
1205         /* End of innermost loop */
1206
1207         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1208                                                  f+i_coord_offset,fshift+i_shift_offset);
1209
1210         ggid                        = gid[iidx];
1211         /* Update potential energies */
1212         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1213         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1214
1215         /* Increment number of inner iterations */
1216         inneriter                  += j_index_end - j_index_start;
1217
1218         /* Outer loop uses 20 flops */
1219     }
1220
1221     /* Increment number of outer iterations */
1222     outeriter        += nri;
1223
1224     /* Update outer/inner flops */
1225
1226     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*361);
1227 }
1228 /*
1229  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_F_avx_256_double
1230  * Electrostatics interaction: ReactionField
1231  * VdW interaction:            CubicSplineTable
1232  * Geometry:                   Water3-Water3
1233  * Calculate force/pot:        Force
1234  */
1235 void
1236 nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_F_avx_256_double
1237                     (t_nblist                    * gmx_restrict       nlist,
1238                      rvec                        * gmx_restrict          xx,
1239                      rvec                        * gmx_restrict          ff,
1240                      struct t_forcerec           * gmx_restrict          fr,
1241                      t_mdatoms                   * gmx_restrict     mdatoms,
1242                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1243                      t_nrnb                      * gmx_restrict        nrnb)
1244 {
1245     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1246      * just 0 for non-waters.
1247      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1248      * jnr indices corresponding to data put in the four positions in the SIMD register.
1249      */
1250     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1251     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1252     int              jnrA,jnrB,jnrC,jnrD;
1253     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1254     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1255     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1256     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1257     real             rcutoff_scalar;
1258     real             *shiftvec,*fshift,*x,*f;
1259     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1260     real             scratch[4*DIM];
1261     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1262     real *           vdwioffsetptr0;
1263     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1264     real *           vdwioffsetptr1;
1265     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1266     real *           vdwioffsetptr2;
1267     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1268     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1269     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1270     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1271     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1272     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1273     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1274     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1275     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1276     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1277     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1278     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1279     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1280     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1281     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1282     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1283     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1284     real             *charge;
1285     int              nvdwtype;
1286     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1287     int              *vdwtype;
1288     real             *vdwparam;
1289     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1290     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1291     __m128i          vfitab;
1292     __m128i          ifour       = _mm_set1_epi32(4);
1293     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1294     real             *vftab;
1295     __m256d          dummy_mask,cutoff_mask;
1296     __m128           tmpmask0,tmpmask1;
1297     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1298     __m256d          one     = _mm256_set1_pd(1.0);
1299     __m256d          two     = _mm256_set1_pd(2.0);
1300     x                = xx[0];
1301     f                = ff[0];
1302
1303     nri              = nlist->nri;
1304     iinr             = nlist->iinr;
1305     jindex           = nlist->jindex;
1306     jjnr             = nlist->jjnr;
1307     shiftidx         = nlist->shift;
1308     gid              = nlist->gid;
1309     shiftvec         = fr->shift_vec[0];
1310     fshift           = fr->fshift[0];
1311     facel            = _mm256_set1_pd(fr->ic->epsfac);
1312     charge           = mdatoms->chargeA;
1313     krf              = _mm256_set1_pd(fr->ic->k_rf);
1314     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
1315     crf              = _mm256_set1_pd(fr->ic->c_rf);
1316     nvdwtype         = fr->ntype;
1317     vdwparam         = fr->nbfp;
1318     vdwtype          = mdatoms->typeA;
1319
1320     vftab            = kernel_data->table_vdw->data;
1321     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
1322
1323     /* Setup water-specific parameters */
1324     inr              = nlist->iinr[0];
1325     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1326     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1327     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1328     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1329
1330     jq0              = _mm256_set1_pd(charge[inr+0]);
1331     jq1              = _mm256_set1_pd(charge[inr+1]);
1332     jq2              = _mm256_set1_pd(charge[inr+2]);
1333     vdwjidx0A        = 2*vdwtype[inr+0];
1334     qq00             = _mm256_mul_pd(iq0,jq0);
1335     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1336     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1337     qq01             = _mm256_mul_pd(iq0,jq1);
1338     qq02             = _mm256_mul_pd(iq0,jq2);
1339     qq10             = _mm256_mul_pd(iq1,jq0);
1340     qq11             = _mm256_mul_pd(iq1,jq1);
1341     qq12             = _mm256_mul_pd(iq1,jq2);
1342     qq20             = _mm256_mul_pd(iq2,jq0);
1343     qq21             = _mm256_mul_pd(iq2,jq1);
1344     qq22             = _mm256_mul_pd(iq2,jq2);
1345
1346     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1347     rcutoff_scalar   = fr->ic->rcoulomb;
1348     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
1349     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
1350
1351     /* Avoid stupid compiler warnings */
1352     jnrA = jnrB = jnrC = jnrD = 0;
1353     j_coord_offsetA = 0;
1354     j_coord_offsetB = 0;
1355     j_coord_offsetC = 0;
1356     j_coord_offsetD = 0;
1357
1358     outeriter        = 0;
1359     inneriter        = 0;
1360
1361     for(iidx=0;iidx<4*DIM;iidx++)
1362     {
1363         scratch[iidx] = 0.0;
1364     }
1365
1366     /* Start outer loop over neighborlists */
1367     for(iidx=0; iidx<nri; iidx++)
1368     {
1369         /* Load shift vector for this list */
1370         i_shift_offset   = DIM*shiftidx[iidx];
1371
1372         /* Load limits for loop over neighbors */
1373         j_index_start    = jindex[iidx];
1374         j_index_end      = jindex[iidx+1];
1375
1376         /* Get outer coordinate index */
1377         inr              = iinr[iidx];
1378         i_coord_offset   = DIM*inr;
1379
1380         /* Load i particle coords and add shift vector */
1381         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1382                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1383
1384         fix0             = _mm256_setzero_pd();
1385         fiy0             = _mm256_setzero_pd();
1386         fiz0             = _mm256_setzero_pd();
1387         fix1             = _mm256_setzero_pd();
1388         fiy1             = _mm256_setzero_pd();
1389         fiz1             = _mm256_setzero_pd();
1390         fix2             = _mm256_setzero_pd();
1391         fiy2             = _mm256_setzero_pd();
1392         fiz2             = _mm256_setzero_pd();
1393
1394         /* Start inner kernel loop */
1395         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1396         {
1397
1398             /* Get j neighbor index, and coordinate index */
1399             jnrA             = jjnr[jidx];
1400             jnrB             = jjnr[jidx+1];
1401             jnrC             = jjnr[jidx+2];
1402             jnrD             = jjnr[jidx+3];
1403             j_coord_offsetA  = DIM*jnrA;
1404             j_coord_offsetB  = DIM*jnrB;
1405             j_coord_offsetC  = DIM*jnrC;
1406             j_coord_offsetD  = DIM*jnrD;
1407
1408             /* load j atom coordinates */
1409             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1410                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1411                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1412
1413             /* Calculate displacement vector */
1414             dx00             = _mm256_sub_pd(ix0,jx0);
1415             dy00             = _mm256_sub_pd(iy0,jy0);
1416             dz00             = _mm256_sub_pd(iz0,jz0);
1417             dx01             = _mm256_sub_pd(ix0,jx1);
1418             dy01             = _mm256_sub_pd(iy0,jy1);
1419             dz01             = _mm256_sub_pd(iz0,jz1);
1420             dx02             = _mm256_sub_pd(ix0,jx2);
1421             dy02             = _mm256_sub_pd(iy0,jy2);
1422             dz02             = _mm256_sub_pd(iz0,jz2);
1423             dx10             = _mm256_sub_pd(ix1,jx0);
1424             dy10             = _mm256_sub_pd(iy1,jy0);
1425             dz10             = _mm256_sub_pd(iz1,jz0);
1426             dx11             = _mm256_sub_pd(ix1,jx1);
1427             dy11             = _mm256_sub_pd(iy1,jy1);
1428             dz11             = _mm256_sub_pd(iz1,jz1);
1429             dx12             = _mm256_sub_pd(ix1,jx2);
1430             dy12             = _mm256_sub_pd(iy1,jy2);
1431             dz12             = _mm256_sub_pd(iz1,jz2);
1432             dx20             = _mm256_sub_pd(ix2,jx0);
1433             dy20             = _mm256_sub_pd(iy2,jy0);
1434             dz20             = _mm256_sub_pd(iz2,jz0);
1435             dx21             = _mm256_sub_pd(ix2,jx1);
1436             dy21             = _mm256_sub_pd(iy2,jy1);
1437             dz21             = _mm256_sub_pd(iz2,jz1);
1438             dx22             = _mm256_sub_pd(ix2,jx2);
1439             dy22             = _mm256_sub_pd(iy2,jy2);
1440             dz22             = _mm256_sub_pd(iz2,jz2);
1441
1442             /* Calculate squared distance and things based on it */
1443             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1444             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1445             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1446             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1447             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1448             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1449             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1450             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1451             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1452
1453             rinv00           = avx256_invsqrt_d(rsq00);
1454             rinv01           = avx256_invsqrt_d(rsq01);
1455             rinv02           = avx256_invsqrt_d(rsq02);
1456             rinv10           = avx256_invsqrt_d(rsq10);
1457             rinv11           = avx256_invsqrt_d(rsq11);
1458             rinv12           = avx256_invsqrt_d(rsq12);
1459             rinv20           = avx256_invsqrt_d(rsq20);
1460             rinv21           = avx256_invsqrt_d(rsq21);
1461             rinv22           = avx256_invsqrt_d(rsq22);
1462
1463             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1464             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1465             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1466             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1467             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1468             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1469             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1470             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1471             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1472
1473             fjx0             = _mm256_setzero_pd();
1474             fjy0             = _mm256_setzero_pd();
1475             fjz0             = _mm256_setzero_pd();
1476             fjx1             = _mm256_setzero_pd();
1477             fjy1             = _mm256_setzero_pd();
1478             fjz1             = _mm256_setzero_pd();
1479             fjx2             = _mm256_setzero_pd();
1480             fjy2             = _mm256_setzero_pd();
1481             fjz2             = _mm256_setzero_pd();
1482
1483             /**************************
1484              * CALCULATE INTERACTIONS *
1485              **************************/
1486
1487             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1488             {
1489
1490             r00              = _mm256_mul_pd(rsq00,rinv00);
1491
1492             /* Calculate table index by multiplying r with table scale and truncate to integer */
1493             rt               = _mm256_mul_pd(r00,vftabscale);
1494             vfitab           = _mm256_cvttpd_epi32(rt);
1495             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1496             vfitab           = _mm_slli_epi32(vfitab,3);
1497
1498             /* REACTION-FIELD ELECTROSTATICS */
1499             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1500
1501             /* CUBIC SPLINE TABLE DISPERSION */
1502             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1503             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1504             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1505             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1506             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1507             Heps             = _mm256_mul_pd(vfeps,H);
1508             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1509             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1510             fvdw6            = _mm256_mul_pd(c6_00,FF);
1511
1512             /* CUBIC SPLINE TABLE REPULSION */
1513             vfitab           = _mm_add_epi32(vfitab,ifour);
1514             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1515             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1516             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1517             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1518             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1519             Heps             = _mm256_mul_pd(vfeps,H);
1520             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1521             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1522             fvdw12           = _mm256_mul_pd(c12_00,FF);
1523             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1524
1525             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1526
1527             fscal            = _mm256_add_pd(felec,fvdw);
1528
1529             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1530
1531             /* Calculate temporary vectorial force */
1532             tx               = _mm256_mul_pd(fscal,dx00);
1533             ty               = _mm256_mul_pd(fscal,dy00);
1534             tz               = _mm256_mul_pd(fscal,dz00);
1535
1536             /* Update vectorial force */
1537             fix0             = _mm256_add_pd(fix0,tx);
1538             fiy0             = _mm256_add_pd(fiy0,ty);
1539             fiz0             = _mm256_add_pd(fiz0,tz);
1540
1541             fjx0             = _mm256_add_pd(fjx0,tx);
1542             fjy0             = _mm256_add_pd(fjy0,ty);
1543             fjz0             = _mm256_add_pd(fjz0,tz);
1544
1545             }
1546
1547             /**************************
1548              * CALCULATE INTERACTIONS *
1549              **************************/
1550
1551             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1552             {
1553
1554             /* REACTION-FIELD ELECTROSTATICS */
1555             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1556
1557             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
1558
1559             fscal            = felec;
1560
1561             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1562
1563             /* Calculate temporary vectorial force */
1564             tx               = _mm256_mul_pd(fscal,dx01);
1565             ty               = _mm256_mul_pd(fscal,dy01);
1566             tz               = _mm256_mul_pd(fscal,dz01);
1567
1568             /* Update vectorial force */
1569             fix0             = _mm256_add_pd(fix0,tx);
1570             fiy0             = _mm256_add_pd(fiy0,ty);
1571             fiz0             = _mm256_add_pd(fiz0,tz);
1572
1573             fjx1             = _mm256_add_pd(fjx1,tx);
1574             fjy1             = _mm256_add_pd(fjy1,ty);
1575             fjz1             = _mm256_add_pd(fjz1,tz);
1576
1577             }
1578
1579             /**************************
1580              * CALCULATE INTERACTIONS *
1581              **************************/
1582
1583             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1584             {
1585
1586             /* REACTION-FIELD ELECTROSTATICS */
1587             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1588
1589             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
1590
1591             fscal            = felec;
1592
1593             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1594
1595             /* Calculate temporary vectorial force */
1596             tx               = _mm256_mul_pd(fscal,dx02);
1597             ty               = _mm256_mul_pd(fscal,dy02);
1598             tz               = _mm256_mul_pd(fscal,dz02);
1599
1600             /* Update vectorial force */
1601             fix0             = _mm256_add_pd(fix0,tx);
1602             fiy0             = _mm256_add_pd(fiy0,ty);
1603             fiz0             = _mm256_add_pd(fiz0,tz);
1604
1605             fjx2             = _mm256_add_pd(fjx2,tx);
1606             fjy2             = _mm256_add_pd(fjy2,ty);
1607             fjz2             = _mm256_add_pd(fjz2,tz);
1608
1609             }
1610
1611             /**************************
1612              * CALCULATE INTERACTIONS *
1613              **************************/
1614
1615             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1616             {
1617
1618             /* REACTION-FIELD ELECTROSTATICS */
1619             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1620
1621             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1622
1623             fscal            = felec;
1624
1625             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1626
1627             /* Calculate temporary vectorial force */
1628             tx               = _mm256_mul_pd(fscal,dx10);
1629             ty               = _mm256_mul_pd(fscal,dy10);
1630             tz               = _mm256_mul_pd(fscal,dz10);
1631
1632             /* Update vectorial force */
1633             fix1             = _mm256_add_pd(fix1,tx);
1634             fiy1             = _mm256_add_pd(fiy1,ty);
1635             fiz1             = _mm256_add_pd(fiz1,tz);
1636
1637             fjx0             = _mm256_add_pd(fjx0,tx);
1638             fjy0             = _mm256_add_pd(fjy0,ty);
1639             fjz0             = _mm256_add_pd(fjz0,tz);
1640
1641             }
1642
1643             /**************************
1644              * CALCULATE INTERACTIONS *
1645              **************************/
1646
1647             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1648             {
1649
1650             /* REACTION-FIELD ELECTROSTATICS */
1651             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1652
1653             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1654
1655             fscal            = felec;
1656
1657             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1658
1659             /* Calculate temporary vectorial force */
1660             tx               = _mm256_mul_pd(fscal,dx11);
1661             ty               = _mm256_mul_pd(fscal,dy11);
1662             tz               = _mm256_mul_pd(fscal,dz11);
1663
1664             /* Update vectorial force */
1665             fix1             = _mm256_add_pd(fix1,tx);
1666             fiy1             = _mm256_add_pd(fiy1,ty);
1667             fiz1             = _mm256_add_pd(fiz1,tz);
1668
1669             fjx1             = _mm256_add_pd(fjx1,tx);
1670             fjy1             = _mm256_add_pd(fjy1,ty);
1671             fjz1             = _mm256_add_pd(fjz1,tz);
1672
1673             }
1674
1675             /**************************
1676              * CALCULATE INTERACTIONS *
1677              **************************/
1678
1679             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1680             {
1681
1682             /* REACTION-FIELD ELECTROSTATICS */
1683             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1684
1685             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1686
1687             fscal            = felec;
1688
1689             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1690
1691             /* Calculate temporary vectorial force */
1692             tx               = _mm256_mul_pd(fscal,dx12);
1693             ty               = _mm256_mul_pd(fscal,dy12);
1694             tz               = _mm256_mul_pd(fscal,dz12);
1695
1696             /* Update vectorial force */
1697             fix1             = _mm256_add_pd(fix1,tx);
1698             fiy1             = _mm256_add_pd(fiy1,ty);
1699             fiz1             = _mm256_add_pd(fiz1,tz);
1700
1701             fjx2             = _mm256_add_pd(fjx2,tx);
1702             fjy2             = _mm256_add_pd(fjy2,ty);
1703             fjz2             = _mm256_add_pd(fjz2,tz);
1704
1705             }
1706
1707             /**************************
1708              * CALCULATE INTERACTIONS *
1709              **************************/
1710
1711             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1712             {
1713
1714             /* REACTION-FIELD ELECTROSTATICS */
1715             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1716
1717             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1718
1719             fscal            = felec;
1720
1721             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1722
1723             /* Calculate temporary vectorial force */
1724             tx               = _mm256_mul_pd(fscal,dx20);
1725             ty               = _mm256_mul_pd(fscal,dy20);
1726             tz               = _mm256_mul_pd(fscal,dz20);
1727
1728             /* Update vectorial force */
1729             fix2             = _mm256_add_pd(fix2,tx);
1730             fiy2             = _mm256_add_pd(fiy2,ty);
1731             fiz2             = _mm256_add_pd(fiz2,tz);
1732
1733             fjx0             = _mm256_add_pd(fjx0,tx);
1734             fjy0             = _mm256_add_pd(fjy0,ty);
1735             fjz0             = _mm256_add_pd(fjz0,tz);
1736
1737             }
1738
1739             /**************************
1740              * CALCULATE INTERACTIONS *
1741              **************************/
1742
1743             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1744             {
1745
1746             /* REACTION-FIELD ELECTROSTATICS */
1747             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1748
1749             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1750
1751             fscal            = felec;
1752
1753             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1754
1755             /* Calculate temporary vectorial force */
1756             tx               = _mm256_mul_pd(fscal,dx21);
1757             ty               = _mm256_mul_pd(fscal,dy21);
1758             tz               = _mm256_mul_pd(fscal,dz21);
1759
1760             /* Update vectorial force */
1761             fix2             = _mm256_add_pd(fix2,tx);
1762             fiy2             = _mm256_add_pd(fiy2,ty);
1763             fiz2             = _mm256_add_pd(fiz2,tz);
1764
1765             fjx1             = _mm256_add_pd(fjx1,tx);
1766             fjy1             = _mm256_add_pd(fjy1,ty);
1767             fjz1             = _mm256_add_pd(fjz1,tz);
1768
1769             }
1770
1771             /**************************
1772              * CALCULATE INTERACTIONS *
1773              **************************/
1774
1775             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1776             {
1777
1778             /* REACTION-FIELD ELECTROSTATICS */
1779             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1780
1781             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1782
1783             fscal            = felec;
1784
1785             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1786
1787             /* Calculate temporary vectorial force */
1788             tx               = _mm256_mul_pd(fscal,dx22);
1789             ty               = _mm256_mul_pd(fscal,dy22);
1790             tz               = _mm256_mul_pd(fscal,dz22);
1791
1792             /* Update vectorial force */
1793             fix2             = _mm256_add_pd(fix2,tx);
1794             fiy2             = _mm256_add_pd(fiy2,ty);
1795             fiz2             = _mm256_add_pd(fiz2,tz);
1796
1797             fjx2             = _mm256_add_pd(fjx2,tx);
1798             fjy2             = _mm256_add_pd(fjy2,ty);
1799             fjz2             = _mm256_add_pd(fjz2,tz);
1800
1801             }
1802
1803             fjptrA             = f+j_coord_offsetA;
1804             fjptrB             = f+j_coord_offsetB;
1805             fjptrC             = f+j_coord_offsetC;
1806             fjptrD             = f+j_coord_offsetD;
1807
1808             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1809                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1810
1811             /* Inner loop uses 297 flops */
1812         }
1813
1814         if(jidx<j_index_end)
1815         {
1816
1817             /* Get j neighbor index, and coordinate index */
1818             jnrlistA         = jjnr[jidx];
1819             jnrlistB         = jjnr[jidx+1];
1820             jnrlistC         = jjnr[jidx+2];
1821             jnrlistD         = jjnr[jidx+3];
1822             /* Sign of each element will be negative for non-real atoms.
1823              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1824              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1825              */
1826             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1827
1828             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1829             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1830             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1831
1832             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1833             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1834             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1835             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1836             j_coord_offsetA  = DIM*jnrA;
1837             j_coord_offsetB  = DIM*jnrB;
1838             j_coord_offsetC  = DIM*jnrC;
1839             j_coord_offsetD  = DIM*jnrD;
1840
1841             /* load j atom coordinates */
1842             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1843                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1844                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1845
1846             /* Calculate displacement vector */
1847             dx00             = _mm256_sub_pd(ix0,jx0);
1848             dy00             = _mm256_sub_pd(iy0,jy0);
1849             dz00             = _mm256_sub_pd(iz0,jz0);
1850             dx01             = _mm256_sub_pd(ix0,jx1);
1851             dy01             = _mm256_sub_pd(iy0,jy1);
1852             dz01             = _mm256_sub_pd(iz0,jz1);
1853             dx02             = _mm256_sub_pd(ix0,jx2);
1854             dy02             = _mm256_sub_pd(iy0,jy2);
1855             dz02             = _mm256_sub_pd(iz0,jz2);
1856             dx10             = _mm256_sub_pd(ix1,jx0);
1857             dy10             = _mm256_sub_pd(iy1,jy0);
1858             dz10             = _mm256_sub_pd(iz1,jz0);
1859             dx11             = _mm256_sub_pd(ix1,jx1);
1860             dy11             = _mm256_sub_pd(iy1,jy1);
1861             dz11             = _mm256_sub_pd(iz1,jz1);
1862             dx12             = _mm256_sub_pd(ix1,jx2);
1863             dy12             = _mm256_sub_pd(iy1,jy2);
1864             dz12             = _mm256_sub_pd(iz1,jz2);
1865             dx20             = _mm256_sub_pd(ix2,jx0);
1866             dy20             = _mm256_sub_pd(iy2,jy0);
1867             dz20             = _mm256_sub_pd(iz2,jz0);
1868             dx21             = _mm256_sub_pd(ix2,jx1);
1869             dy21             = _mm256_sub_pd(iy2,jy1);
1870             dz21             = _mm256_sub_pd(iz2,jz1);
1871             dx22             = _mm256_sub_pd(ix2,jx2);
1872             dy22             = _mm256_sub_pd(iy2,jy2);
1873             dz22             = _mm256_sub_pd(iz2,jz2);
1874
1875             /* Calculate squared distance and things based on it */
1876             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1877             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1878             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1879             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1880             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1881             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1882             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1883             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1884             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1885
1886             rinv00           = avx256_invsqrt_d(rsq00);
1887             rinv01           = avx256_invsqrt_d(rsq01);
1888             rinv02           = avx256_invsqrt_d(rsq02);
1889             rinv10           = avx256_invsqrt_d(rsq10);
1890             rinv11           = avx256_invsqrt_d(rsq11);
1891             rinv12           = avx256_invsqrt_d(rsq12);
1892             rinv20           = avx256_invsqrt_d(rsq20);
1893             rinv21           = avx256_invsqrt_d(rsq21);
1894             rinv22           = avx256_invsqrt_d(rsq22);
1895
1896             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1897             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1898             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1899             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1900             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1901             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1902             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1903             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1904             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1905
1906             fjx0             = _mm256_setzero_pd();
1907             fjy0             = _mm256_setzero_pd();
1908             fjz0             = _mm256_setzero_pd();
1909             fjx1             = _mm256_setzero_pd();
1910             fjy1             = _mm256_setzero_pd();
1911             fjz1             = _mm256_setzero_pd();
1912             fjx2             = _mm256_setzero_pd();
1913             fjy2             = _mm256_setzero_pd();
1914             fjz2             = _mm256_setzero_pd();
1915
1916             /**************************
1917              * CALCULATE INTERACTIONS *
1918              **************************/
1919
1920             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1921             {
1922
1923             r00              = _mm256_mul_pd(rsq00,rinv00);
1924             r00              = _mm256_andnot_pd(dummy_mask,r00);
1925
1926             /* Calculate table index by multiplying r with table scale and truncate to integer */
1927             rt               = _mm256_mul_pd(r00,vftabscale);
1928             vfitab           = _mm256_cvttpd_epi32(rt);
1929             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1930             vfitab           = _mm_slli_epi32(vfitab,3);
1931
1932             /* REACTION-FIELD ELECTROSTATICS */
1933             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1934
1935             /* CUBIC SPLINE TABLE DISPERSION */
1936             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1937             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1938             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1939             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1940             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1941             Heps             = _mm256_mul_pd(vfeps,H);
1942             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1943             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1944             fvdw6            = _mm256_mul_pd(c6_00,FF);
1945
1946             /* CUBIC SPLINE TABLE REPULSION */
1947             vfitab           = _mm_add_epi32(vfitab,ifour);
1948             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1949             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1950             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1951             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1952             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1953             Heps             = _mm256_mul_pd(vfeps,H);
1954             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1955             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1956             fvdw12           = _mm256_mul_pd(c12_00,FF);
1957             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1958
1959             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1960
1961             fscal            = _mm256_add_pd(felec,fvdw);
1962
1963             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1964
1965             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1966
1967             /* Calculate temporary vectorial force */
1968             tx               = _mm256_mul_pd(fscal,dx00);
1969             ty               = _mm256_mul_pd(fscal,dy00);
1970             tz               = _mm256_mul_pd(fscal,dz00);
1971
1972             /* Update vectorial force */
1973             fix0             = _mm256_add_pd(fix0,tx);
1974             fiy0             = _mm256_add_pd(fiy0,ty);
1975             fiz0             = _mm256_add_pd(fiz0,tz);
1976
1977             fjx0             = _mm256_add_pd(fjx0,tx);
1978             fjy0             = _mm256_add_pd(fjy0,ty);
1979             fjz0             = _mm256_add_pd(fjz0,tz);
1980
1981             }
1982
1983             /**************************
1984              * CALCULATE INTERACTIONS *
1985              **************************/
1986
1987             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1988             {
1989
1990             /* REACTION-FIELD ELECTROSTATICS */
1991             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1992
1993             cutoff_mask      = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
1994
1995             fscal            = felec;
1996
1997             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1998
1999             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2000
2001             /* Calculate temporary vectorial force */
2002             tx               = _mm256_mul_pd(fscal,dx01);
2003             ty               = _mm256_mul_pd(fscal,dy01);
2004             tz               = _mm256_mul_pd(fscal,dz01);
2005
2006             /* Update vectorial force */
2007             fix0             = _mm256_add_pd(fix0,tx);
2008             fiy0             = _mm256_add_pd(fiy0,ty);
2009             fiz0             = _mm256_add_pd(fiz0,tz);
2010
2011             fjx1             = _mm256_add_pd(fjx1,tx);
2012             fjy1             = _mm256_add_pd(fjy1,ty);
2013             fjz1             = _mm256_add_pd(fjz1,tz);
2014
2015             }
2016
2017             /**************************
2018              * CALCULATE INTERACTIONS *
2019              **************************/
2020
2021             if (gmx_mm256_any_lt(rsq02,rcutoff2))
2022             {
2023
2024             /* REACTION-FIELD ELECTROSTATICS */
2025             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
2026
2027             cutoff_mask      = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
2028
2029             fscal            = felec;
2030
2031             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2032
2033             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2034
2035             /* Calculate temporary vectorial force */
2036             tx               = _mm256_mul_pd(fscal,dx02);
2037             ty               = _mm256_mul_pd(fscal,dy02);
2038             tz               = _mm256_mul_pd(fscal,dz02);
2039
2040             /* Update vectorial force */
2041             fix0             = _mm256_add_pd(fix0,tx);
2042             fiy0             = _mm256_add_pd(fiy0,ty);
2043             fiz0             = _mm256_add_pd(fiz0,tz);
2044
2045             fjx2             = _mm256_add_pd(fjx2,tx);
2046             fjy2             = _mm256_add_pd(fjy2,ty);
2047             fjz2             = _mm256_add_pd(fjz2,tz);
2048
2049             }
2050
2051             /**************************
2052              * CALCULATE INTERACTIONS *
2053              **************************/
2054
2055             if (gmx_mm256_any_lt(rsq10,rcutoff2))
2056             {
2057
2058             /* REACTION-FIELD ELECTROSTATICS */
2059             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
2060
2061             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
2062
2063             fscal            = felec;
2064
2065             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2066
2067             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2068
2069             /* Calculate temporary vectorial force */
2070             tx               = _mm256_mul_pd(fscal,dx10);
2071             ty               = _mm256_mul_pd(fscal,dy10);
2072             tz               = _mm256_mul_pd(fscal,dz10);
2073
2074             /* Update vectorial force */
2075             fix1             = _mm256_add_pd(fix1,tx);
2076             fiy1             = _mm256_add_pd(fiy1,ty);
2077             fiz1             = _mm256_add_pd(fiz1,tz);
2078
2079             fjx0             = _mm256_add_pd(fjx0,tx);
2080             fjy0             = _mm256_add_pd(fjy0,ty);
2081             fjz0             = _mm256_add_pd(fjz0,tz);
2082
2083             }
2084
2085             /**************************
2086              * CALCULATE INTERACTIONS *
2087              **************************/
2088
2089             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2090             {
2091
2092             /* REACTION-FIELD ELECTROSTATICS */
2093             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
2094
2095             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
2096
2097             fscal            = felec;
2098
2099             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2100
2101             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2102
2103             /* Calculate temporary vectorial force */
2104             tx               = _mm256_mul_pd(fscal,dx11);
2105             ty               = _mm256_mul_pd(fscal,dy11);
2106             tz               = _mm256_mul_pd(fscal,dz11);
2107
2108             /* Update vectorial force */
2109             fix1             = _mm256_add_pd(fix1,tx);
2110             fiy1             = _mm256_add_pd(fiy1,ty);
2111             fiz1             = _mm256_add_pd(fiz1,tz);
2112
2113             fjx1             = _mm256_add_pd(fjx1,tx);
2114             fjy1             = _mm256_add_pd(fjy1,ty);
2115             fjz1             = _mm256_add_pd(fjz1,tz);
2116
2117             }
2118
2119             /**************************
2120              * CALCULATE INTERACTIONS *
2121              **************************/
2122
2123             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2124             {
2125
2126             /* REACTION-FIELD ELECTROSTATICS */
2127             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
2128
2129             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
2130
2131             fscal            = felec;
2132
2133             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2134
2135             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2136
2137             /* Calculate temporary vectorial force */
2138             tx               = _mm256_mul_pd(fscal,dx12);
2139             ty               = _mm256_mul_pd(fscal,dy12);
2140             tz               = _mm256_mul_pd(fscal,dz12);
2141
2142             /* Update vectorial force */
2143             fix1             = _mm256_add_pd(fix1,tx);
2144             fiy1             = _mm256_add_pd(fiy1,ty);
2145             fiz1             = _mm256_add_pd(fiz1,tz);
2146
2147             fjx2             = _mm256_add_pd(fjx2,tx);
2148             fjy2             = _mm256_add_pd(fjy2,ty);
2149             fjz2             = _mm256_add_pd(fjz2,tz);
2150
2151             }
2152
2153             /**************************
2154              * CALCULATE INTERACTIONS *
2155              **************************/
2156
2157             if (gmx_mm256_any_lt(rsq20,rcutoff2))
2158             {
2159
2160             /* REACTION-FIELD ELECTROSTATICS */
2161             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
2162
2163             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
2164
2165             fscal            = felec;
2166
2167             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2168
2169             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2170
2171             /* Calculate temporary vectorial force */
2172             tx               = _mm256_mul_pd(fscal,dx20);
2173             ty               = _mm256_mul_pd(fscal,dy20);
2174             tz               = _mm256_mul_pd(fscal,dz20);
2175
2176             /* Update vectorial force */
2177             fix2             = _mm256_add_pd(fix2,tx);
2178             fiy2             = _mm256_add_pd(fiy2,ty);
2179             fiz2             = _mm256_add_pd(fiz2,tz);
2180
2181             fjx0             = _mm256_add_pd(fjx0,tx);
2182             fjy0             = _mm256_add_pd(fjy0,ty);
2183             fjz0             = _mm256_add_pd(fjz0,tz);
2184
2185             }
2186
2187             /**************************
2188              * CALCULATE INTERACTIONS *
2189              **************************/
2190
2191             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2192             {
2193
2194             /* REACTION-FIELD ELECTROSTATICS */
2195             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
2196
2197             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2198
2199             fscal            = felec;
2200
2201             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2202
2203             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2204
2205             /* Calculate temporary vectorial force */
2206             tx               = _mm256_mul_pd(fscal,dx21);
2207             ty               = _mm256_mul_pd(fscal,dy21);
2208             tz               = _mm256_mul_pd(fscal,dz21);
2209
2210             /* Update vectorial force */
2211             fix2             = _mm256_add_pd(fix2,tx);
2212             fiy2             = _mm256_add_pd(fiy2,ty);
2213             fiz2             = _mm256_add_pd(fiz2,tz);
2214
2215             fjx1             = _mm256_add_pd(fjx1,tx);
2216             fjy1             = _mm256_add_pd(fjy1,ty);
2217             fjz1             = _mm256_add_pd(fjz1,tz);
2218
2219             }
2220
2221             /**************************
2222              * CALCULATE INTERACTIONS *
2223              **************************/
2224
2225             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2226             {
2227
2228             /* REACTION-FIELD ELECTROSTATICS */
2229             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
2230
2231             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2232
2233             fscal            = felec;
2234
2235             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2236
2237             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2238
2239             /* Calculate temporary vectorial force */
2240             tx               = _mm256_mul_pd(fscal,dx22);
2241             ty               = _mm256_mul_pd(fscal,dy22);
2242             tz               = _mm256_mul_pd(fscal,dz22);
2243
2244             /* Update vectorial force */
2245             fix2             = _mm256_add_pd(fix2,tx);
2246             fiy2             = _mm256_add_pd(fiy2,ty);
2247             fiz2             = _mm256_add_pd(fiz2,tz);
2248
2249             fjx2             = _mm256_add_pd(fjx2,tx);
2250             fjy2             = _mm256_add_pd(fjy2,ty);
2251             fjz2             = _mm256_add_pd(fjz2,tz);
2252
2253             }
2254
2255             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2256             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2257             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2258             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2259
2260             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2261                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2262
2263             /* Inner loop uses 298 flops */
2264         }
2265
2266         /* End of innermost loop */
2267
2268         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2269                                                  f+i_coord_offset,fshift+i_shift_offset);
2270
2271         /* Increment number of inner iterations */
2272         inneriter                  += j_index_end - j_index_start;
2273
2274         /* Outer loop uses 18 flops */
2275     }
2276
2277     /* Increment number of outer iterations */
2278     outeriter        += nri;
2279
2280     /* Update outer/inner flops */
2281
2282     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*298);
2283 }