src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwCSTab_GeomW3W3_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3W3_VF_avx_256_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomW3W3_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  97     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  98     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  99     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 100     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 101     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 102     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 103     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 104     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 105     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 106     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 107     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 108     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 109     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 110     real             *charge;
 111     int              nvdwtype;
 112     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 113     int              *vdwtype;
 114     real             *vdwparam;
 115     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 116     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 117     __m256i          vfitab;
 118     __m128i          vfitab_lo,vfitab_hi;
 119     __m128i          ifour       = _mm_set1_epi32(4);
 120     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 121     real             *vftab;
 122     __m256           dummy_mask,cutoff_mask;
 123     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 124     __m256           one     = _mm256_set1_ps(1.0);
 125     __m256           two     = _mm256_set1_ps(2.0);
 126     x                = xx[0];
 127     f                = ff[0];
 128
 129     nri              = nlist->nri;
 130     iinr             = nlist->iinr;
 131     jindex           = nlist->jindex;
 132     jjnr             = nlist->jjnr;
 133     shiftidx         = nlist->shift;
 134     gid              = nlist->gid;
 135     shiftvec         = fr->shift_vec[0];
 136     fshift           = fr->fshift[0];
 137     facel            = _mm256_set1_ps(fr->epsfac);
 138     charge           = mdatoms->chargeA;
 139     krf              = _mm256_set1_ps(fr->ic->k_rf);
 140     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 141     crf              = _mm256_set1_ps(fr->ic->c_rf);
 142     nvdwtype         = fr->ntype;
 143     vdwparam         = fr->nbfp;
 144     vdwtype          = mdatoms->typeA;
 145
 146     vftab            = kernel_data->table_vdw->data;
 147     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 148
 149     /* Setup water-specific parameters */
 150     inr              = nlist->iinr[0];
 151     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 152     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 153     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 154     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 155
 156     jq0              = _mm256_set1_ps(charge[inr+0]);
 157     jq1              = _mm256_set1_ps(charge[inr+1]);
 158     jq2              = _mm256_set1_ps(charge[inr+2]);
 159     vdwjidx0A        = 2*vdwtype[inr+0];
 160     qq00             = _mm256_mul_ps(iq0,jq0);
 161     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 162     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 163     qq01             = _mm256_mul_ps(iq0,jq1);
 164     qq02             = _mm256_mul_ps(iq0,jq2);
 165     qq10             = _mm256_mul_ps(iq1,jq0);
 166     qq11             = _mm256_mul_ps(iq1,jq1);
 167     qq12             = _mm256_mul_ps(iq1,jq2);
 168     qq20             = _mm256_mul_ps(iq2,jq0);
 169     qq21             = _mm256_mul_ps(iq2,jq1);
 170     qq22             = _mm256_mul_ps(iq2,jq2);
 171
 172     /* Avoid stupid compiler warnings */
 173     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 174     j_coord_offsetA = 0;
 175     j_coord_offsetB = 0;
 176     j_coord_offsetC = 0;
 177     j_coord_offsetD = 0;
 178     j_coord_offsetE = 0;
 179     j_coord_offsetF = 0;
 180     j_coord_offsetG = 0;
 181     j_coord_offsetH = 0;
 182
 183     outeriter        = 0;
 184     inneriter        = 0;
 185
 186     for(iidx=0;iidx<4*DIM;iidx++)
 187     {
 188         scratch[iidx] = 0.0;
 189     }
 190
 191     /* Start outer loop over neighborlists */
 192     for(iidx=0; iidx<nri; iidx++)
 193     {
 194         /* Load shift vector for this list */
 195         i_shift_offset   = DIM*shiftidx[iidx];
 196
 197         /* Load limits for loop over neighbors */
 198         j_index_start    = jindex[iidx];
 199         j_index_end      = jindex[iidx+1];
 200
 201         /* Get outer coordinate index */
 202         inr              = iinr[iidx];
 203         i_coord_offset   = DIM*inr;
 204
 205         /* Load i particle coords and add shift vector */
 206         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 207                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 208
 209         fix0             = _mm256_setzero_ps();
 210         fiy0             = _mm256_setzero_ps();
 211         fiz0             = _mm256_setzero_ps();
 212         fix1             = _mm256_setzero_ps();
 213         fiy1             = _mm256_setzero_ps();
 214         fiz1             = _mm256_setzero_ps();
 215         fix2             = _mm256_setzero_ps();
 216         fiy2             = _mm256_setzero_ps();
 217         fiz2             = _mm256_setzero_ps();
 218
 219         /* Reset potential sums */
 220         velecsum         = _mm256_setzero_ps();
 221         vvdwsum          = _mm256_setzero_ps();
 222
 223         /* Start inner kernel loop */
 224         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 225         {
 226
 227             /* Get j neighbor index, and coordinate index */
 228             jnrA             = jjnr[jidx];
 229             jnrB             = jjnr[jidx+1];
 230             jnrC             = jjnr[jidx+2];
 231             jnrD             = jjnr[jidx+3];
 232             jnrE             = jjnr[jidx+4];
 233             jnrF             = jjnr[jidx+5];
 234             jnrG             = jjnr[jidx+6];
 235             jnrH             = jjnr[jidx+7];
 236             j_coord_offsetA  = DIM*jnrA;
 237             j_coord_offsetB  = DIM*jnrB;
 238             j_coord_offsetC  = DIM*jnrC;
 239             j_coord_offsetD  = DIM*jnrD;
 240             j_coord_offsetE  = DIM*jnrE;
 241             j_coord_offsetF  = DIM*jnrF;
 242             j_coord_offsetG  = DIM*jnrG;
 243             j_coord_offsetH  = DIM*jnrH;
 244
 245             /* load j atom coordinates */
 246             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 247                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 248                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 249                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 250                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 251
 252             /* Calculate displacement vector */
 253             dx00             = _mm256_sub_ps(ix0,jx0);
 254             dy00             = _mm256_sub_ps(iy0,jy0);
 255             dz00             = _mm256_sub_ps(iz0,jz0);
 256             dx01             = _mm256_sub_ps(ix0,jx1);
 257             dy01             = _mm256_sub_ps(iy0,jy1);
 258             dz01             = _mm256_sub_ps(iz0,jz1);
 259             dx02             = _mm256_sub_ps(ix0,jx2);
 260             dy02             = _mm256_sub_ps(iy0,jy2);
 261             dz02             = _mm256_sub_ps(iz0,jz2);
 262             dx10             = _mm256_sub_ps(ix1,jx0);
 263             dy10             = _mm256_sub_ps(iy1,jy0);
 264             dz10             = _mm256_sub_ps(iz1,jz0);
 265             dx11             = _mm256_sub_ps(ix1,jx1);
 266             dy11             = _mm256_sub_ps(iy1,jy1);
 267             dz11             = _mm256_sub_ps(iz1,jz1);
 268             dx12             = _mm256_sub_ps(ix1,jx2);
 269             dy12             = _mm256_sub_ps(iy1,jy2);
 270             dz12             = _mm256_sub_ps(iz1,jz2);
 271             dx20             = _mm256_sub_ps(ix2,jx0);
 272             dy20             = _mm256_sub_ps(iy2,jy0);
 273             dz20             = _mm256_sub_ps(iz2,jz0);
 274             dx21             = _mm256_sub_ps(ix2,jx1);
 275             dy21             = _mm256_sub_ps(iy2,jy1);
 276             dz21             = _mm256_sub_ps(iz2,jz1);
 277             dx22             = _mm256_sub_ps(ix2,jx2);
 278             dy22             = _mm256_sub_ps(iy2,jy2);
 279             dz22             = _mm256_sub_ps(iz2,jz2);
 280
 281             /* Calculate squared distance and things based on it */
 282             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 283             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 284             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 285             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 286             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 287             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 288             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 289             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 290             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 291
 292             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 293             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 294             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 295             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 296             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 297             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 298             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 299             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 300             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 301
 302             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 303             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 304             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 305             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 306             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 307             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 308             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 309             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 310             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 311
 312             fjx0             = _mm256_setzero_ps();
 313             fjy0             = _mm256_setzero_ps();
 314             fjz0             = _mm256_setzero_ps();
 315             fjx1             = _mm256_setzero_ps();
 316             fjy1             = _mm256_setzero_ps();
 317             fjz1             = _mm256_setzero_ps();
 318             fjx2             = _mm256_setzero_ps();
 319             fjy2             = _mm256_setzero_ps();
 320             fjz2             = _mm256_setzero_ps();
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             r00              = _mm256_mul_ps(rsq00,rinv00);
 327
 328             /* Calculate table index by multiplying r with table scale and truncate to integer */
 329             rt               = _mm256_mul_ps(r00,vftabscale);
 330             vfitab           = _mm256_cvttps_epi32(rt);
 331             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 332             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 333             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 334             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 335             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 336             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 337
 338             /* REACTION-FIELD ELECTROSTATICS */
 339             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 340             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 341
 342             /* CUBIC SPLINE TABLE DISPERSION */
 343             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 344                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 345             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 346                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 347             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 348                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 349             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 350                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 351             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 352             Heps             = _mm256_mul_ps(vfeps,H);
 353             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 354             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 355             vvdw6            = _mm256_mul_ps(c6_00,VV);
 356             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 357             fvdw6            = _mm256_mul_ps(c6_00,FF);
 358
 359             /* CUBIC SPLINE TABLE REPULSION */
 360             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 361             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 362             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 363                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 364             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 365                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 366             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 367                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 368             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 369                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 370             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 371             Heps             = _mm256_mul_ps(vfeps,H);
 372             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 373             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 374             vvdw12           = _mm256_mul_ps(c12_00,VV);
 375             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 376             fvdw12           = _mm256_mul_ps(c12_00,FF);
 377             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 378             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 379
 380             /* Update potential sum for this i atom from the interaction with this j atom. */
 381             velecsum         = _mm256_add_ps(velecsum,velec);
 382             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 383
 384             fscal            = _mm256_add_ps(felec,fvdw);
 385
 386             /* Calculate temporary vectorial force */
 387             tx               = _mm256_mul_ps(fscal,dx00);
 388             ty               = _mm256_mul_ps(fscal,dy00);
 389             tz               = _mm256_mul_ps(fscal,dz00);
 390
 391             /* Update vectorial force */
 392             fix0             = _mm256_add_ps(fix0,tx);
 393             fiy0             = _mm256_add_ps(fiy0,ty);
 394             fiz0             = _mm256_add_ps(fiz0,tz);
 395
 396             fjx0             = _mm256_add_ps(fjx0,tx);
 397             fjy0             = _mm256_add_ps(fjy0,ty);
 398             fjz0             = _mm256_add_ps(fjz0,tz);
 399
 400             /**************************
 401              * CALCULATE INTERACTIONS *
 402              **************************/
 403
 404             /* REACTION-FIELD ELECTROSTATICS */
 405             velec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
 406             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
 407
 408             /* Update potential sum for this i atom from the interaction with this j atom. */
 409             velecsum         = _mm256_add_ps(velecsum,velec);
 410
 411             fscal            = felec;
 412
 413             /* Calculate temporary vectorial force */
 414             tx               = _mm256_mul_ps(fscal,dx01);
 415             ty               = _mm256_mul_ps(fscal,dy01);
 416             tz               = _mm256_mul_ps(fscal,dz01);
 417
 418             /* Update vectorial force */
 419             fix0             = _mm256_add_ps(fix0,tx);
 420             fiy0             = _mm256_add_ps(fiy0,ty);
 421             fiz0             = _mm256_add_ps(fiz0,tz);
 422
 423             fjx1             = _mm256_add_ps(fjx1,tx);
 424             fjy1             = _mm256_add_ps(fjy1,ty);
 425             fjz1             = _mm256_add_ps(fjz1,tz);
 426
 427             /**************************
 428              * CALCULATE INTERACTIONS *
 429              **************************/
 430
 431             /* REACTION-FIELD ELECTROSTATICS */
 432             velec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
 433             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
 434
 435             /* Update potential sum for this i atom from the interaction with this j atom. */
 436             velecsum         = _mm256_add_ps(velecsum,velec);
 437
 438             fscal            = felec;
 439
 440             /* Calculate temporary vectorial force */
 441             tx               = _mm256_mul_ps(fscal,dx02);
 442             ty               = _mm256_mul_ps(fscal,dy02);
 443             tz               = _mm256_mul_ps(fscal,dz02);
 444
 445             /* Update vectorial force */
 446             fix0             = _mm256_add_ps(fix0,tx);
 447             fiy0             = _mm256_add_ps(fiy0,ty);
 448             fiz0             = _mm256_add_ps(fiz0,tz);
 449
 450             fjx2             = _mm256_add_ps(fjx2,tx);
 451             fjy2             = _mm256_add_ps(fjy2,ty);
 452             fjz2             = _mm256_add_ps(fjz2,tz);
 453
 454             /**************************
 455              * CALCULATE INTERACTIONS *
 456              **************************/
 457
 458             /* REACTION-FIELD ELECTROSTATICS */
 459             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 460             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 461
 462             /* Update potential sum for this i atom from the interaction with this j atom. */
 463             velecsum         = _mm256_add_ps(velecsum,velec);
 464
 465             fscal            = felec;
 466
 467             /* Calculate temporary vectorial force */
 468             tx               = _mm256_mul_ps(fscal,dx10);
 469             ty               = _mm256_mul_ps(fscal,dy10);
 470             tz               = _mm256_mul_ps(fscal,dz10);
 471
 472             /* Update vectorial force */
 473             fix1             = _mm256_add_ps(fix1,tx);
 474             fiy1             = _mm256_add_ps(fiy1,ty);
 475             fiz1             = _mm256_add_ps(fiz1,tz);
 476
 477             fjx0             = _mm256_add_ps(fjx0,tx);
 478             fjy0             = _mm256_add_ps(fjy0,ty);
 479             fjz0             = _mm256_add_ps(fjz0,tz);
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             /* REACTION-FIELD ELECTROSTATICS */
 486             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 487             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 488
 489             /* Update potential sum for this i atom from the interaction with this j atom. */
 490             velecsum         = _mm256_add_ps(velecsum,velec);
 491
 492             fscal            = felec;
 493
 494             /* Calculate temporary vectorial force */
 495             tx               = _mm256_mul_ps(fscal,dx11);
 496             ty               = _mm256_mul_ps(fscal,dy11);
 497             tz               = _mm256_mul_ps(fscal,dz11);
 498
 499             /* Update vectorial force */
 500             fix1             = _mm256_add_ps(fix1,tx);
 501             fiy1             = _mm256_add_ps(fiy1,ty);
 502             fiz1             = _mm256_add_ps(fiz1,tz);
 503
 504             fjx1             = _mm256_add_ps(fjx1,tx);
 505             fjy1             = _mm256_add_ps(fjy1,ty);
 506             fjz1             = _mm256_add_ps(fjz1,tz);
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             /* REACTION-FIELD ELECTROSTATICS */
 513             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 514             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velecsum         = _mm256_add_ps(velecsum,velec);
 518
 519             fscal            = felec;
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm256_mul_ps(fscal,dx12);
 523             ty               = _mm256_mul_ps(fscal,dy12);
 524             tz               = _mm256_mul_ps(fscal,dz12);
 525
 526             /* Update vectorial force */
 527             fix1             = _mm256_add_ps(fix1,tx);
 528             fiy1             = _mm256_add_ps(fiy1,ty);
 529             fiz1             = _mm256_add_ps(fiz1,tz);
 530
 531             fjx2             = _mm256_add_ps(fjx2,tx);
 532             fjy2             = _mm256_add_ps(fjy2,ty);
 533             fjz2             = _mm256_add_ps(fjz2,tz);
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             /* REACTION-FIELD ELECTROSTATICS */
 540             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 541             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             velecsum         = _mm256_add_ps(velecsum,velec);
 545
 546             fscal            = felec;
 547
 548             /* Calculate temporary vectorial force */
 549             tx               = _mm256_mul_ps(fscal,dx20);
 550             ty               = _mm256_mul_ps(fscal,dy20);
 551             tz               = _mm256_mul_ps(fscal,dz20);
 552
 553             /* Update vectorial force */
 554             fix2             = _mm256_add_ps(fix2,tx);
 555             fiy2             = _mm256_add_ps(fiy2,ty);
 556             fiz2             = _mm256_add_ps(fiz2,tz);
 557
 558             fjx0             = _mm256_add_ps(fjx0,tx);
 559             fjy0             = _mm256_add_ps(fjy0,ty);
 560             fjz0             = _mm256_add_ps(fjz0,tz);
 561
 562             /**************************
 563              * CALCULATE INTERACTIONS *
 564              **************************/
 565
 566             /* REACTION-FIELD ELECTROSTATICS */
 567             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
 568             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
 569
 570             /* Update potential sum for this i atom from the interaction with this j atom. */
 571             velecsum         = _mm256_add_ps(velecsum,velec);
 572
 573             fscal            = felec;
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm256_mul_ps(fscal,dx21);
 577             ty               = _mm256_mul_ps(fscal,dy21);
 578             tz               = _mm256_mul_ps(fscal,dz21);
 579
 580             /* Update vectorial force */
 581             fix2             = _mm256_add_ps(fix2,tx);
 582             fiy2             = _mm256_add_ps(fiy2,ty);
 583             fiz2             = _mm256_add_ps(fiz2,tz);
 584
 585             fjx1             = _mm256_add_ps(fjx1,tx);
 586             fjy1             = _mm256_add_ps(fjy1,ty);
 587             fjz1             = _mm256_add_ps(fjz1,tz);
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             /* REACTION-FIELD ELECTROSTATICS */
 594             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
 595             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
 596
 597             /* Update potential sum for this i atom from the interaction with this j atom. */
 598             velecsum         = _mm256_add_ps(velecsum,velec);
 599
 600             fscal            = felec;
 601
 602             /* Calculate temporary vectorial force */
 603             tx               = _mm256_mul_ps(fscal,dx22);
 604             ty               = _mm256_mul_ps(fscal,dy22);
 605             tz               = _mm256_mul_ps(fscal,dz22);
 606
 607             /* Update vectorial force */
 608             fix2             = _mm256_add_ps(fix2,tx);
 609             fiy2             = _mm256_add_ps(fiy2,ty);
 610             fiz2             = _mm256_add_ps(fiz2,tz);
 611
 612             fjx2             = _mm256_add_ps(fjx2,tx);
 613             fjy2             = _mm256_add_ps(fjy2,ty);
 614             fjz2             = _mm256_add_ps(fjz2,tz);
 615
 616             fjptrA             = f+j_coord_offsetA;
 617             fjptrB             = f+j_coord_offsetB;
 618             fjptrC             = f+j_coord_offsetC;
 619             fjptrD             = f+j_coord_offsetD;
 620             fjptrE             = f+j_coord_offsetE;
 621             fjptrF             = f+j_coord_offsetF;
 622             fjptrG             = f+j_coord_offsetG;
 623             fjptrH             = f+j_coord_offsetH;
 624
 625             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 626                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 627
 628             /* Inner loop uses 323 flops */
 629         }
 630
 631         if(jidx<j_index_end)
 632         {
 633
 634             /* Get j neighbor index, and coordinate index */
 635             jnrlistA         = jjnr[jidx];
 636             jnrlistB         = jjnr[jidx+1];
 637             jnrlistC         = jjnr[jidx+2];
 638             jnrlistD         = jjnr[jidx+3];
 639             jnrlistE         = jjnr[jidx+4];
 640             jnrlistF         = jjnr[jidx+5];
 641             jnrlistG         = jjnr[jidx+6];
 642             jnrlistH         = jjnr[jidx+7];
 643             /* Sign of each element will be negative for non-real atoms.
 644              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 645              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 646              */
 647             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 648                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 649
 650             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 651             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 652             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 653             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 654             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 655             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 656             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 657             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 658             j_coord_offsetA  = DIM*jnrA;
 659             j_coord_offsetB  = DIM*jnrB;
 660             j_coord_offsetC  = DIM*jnrC;
 661             j_coord_offsetD  = DIM*jnrD;
 662             j_coord_offsetE  = DIM*jnrE;
 663             j_coord_offsetF  = DIM*jnrF;
 664             j_coord_offsetG  = DIM*jnrG;
 665             j_coord_offsetH  = DIM*jnrH;
 666
 667             /* load j atom coordinates */
 668             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 669                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 670                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 671                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 672                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 673
 674             /* Calculate displacement vector */
 675             dx00             = _mm256_sub_ps(ix0,jx0);
 676             dy00             = _mm256_sub_ps(iy0,jy0);
 677             dz00             = _mm256_sub_ps(iz0,jz0);
 678             dx01             = _mm256_sub_ps(ix0,jx1);
 679             dy01             = _mm256_sub_ps(iy0,jy1);
 680             dz01             = _mm256_sub_ps(iz0,jz1);
 681             dx02             = _mm256_sub_ps(ix0,jx2);
 682             dy02             = _mm256_sub_ps(iy0,jy2);
 683             dz02             = _mm256_sub_ps(iz0,jz2);
 684             dx10             = _mm256_sub_ps(ix1,jx0);
 685             dy10             = _mm256_sub_ps(iy1,jy0);
 686             dz10             = _mm256_sub_ps(iz1,jz0);
 687             dx11             = _mm256_sub_ps(ix1,jx1);
 688             dy11             = _mm256_sub_ps(iy1,jy1);
 689             dz11             = _mm256_sub_ps(iz1,jz1);
 690             dx12             = _mm256_sub_ps(ix1,jx2);
 691             dy12             = _mm256_sub_ps(iy1,jy2);
 692             dz12             = _mm256_sub_ps(iz1,jz2);
 693             dx20             = _mm256_sub_ps(ix2,jx0);
 694             dy20             = _mm256_sub_ps(iy2,jy0);
 695             dz20             = _mm256_sub_ps(iz2,jz0);
 696             dx21             = _mm256_sub_ps(ix2,jx1);
 697             dy21             = _mm256_sub_ps(iy2,jy1);
 698             dz21             = _mm256_sub_ps(iz2,jz1);
 699             dx22             = _mm256_sub_ps(ix2,jx2);
 700             dy22             = _mm256_sub_ps(iy2,jy2);
 701             dz22             = _mm256_sub_ps(iz2,jz2);
 702
 703             /* Calculate squared distance and things based on it */
 704             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 705             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 706             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 707             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 708             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 709             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 710             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 711             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 712             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 713
 714             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 715             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 716             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 717             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 718             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 719             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 720             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 721             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 722             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 723
 724             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 725             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 726             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 727             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 728             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 729             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 730             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 731             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 732             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 733
 734             fjx0             = _mm256_setzero_ps();
 735             fjy0             = _mm256_setzero_ps();
 736             fjz0             = _mm256_setzero_ps();
 737             fjx1             = _mm256_setzero_ps();
 738             fjy1             = _mm256_setzero_ps();
 739             fjz1             = _mm256_setzero_ps();
 740             fjx2             = _mm256_setzero_ps();
 741             fjy2             = _mm256_setzero_ps();
 742             fjz2             = _mm256_setzero_ps();
 743
 744             /**************************
 745              * CALCULATE INTERACTIONS *
 746              **************************/
 747
 748             r00              = _mm256_mul_ps(rsq00,rinv00);
 749             r00              = _mm256_andnot_ps(dummy_mask,r00);
 750
 751             /* Calculate table index by multiplying r with table scale and truncate to integer */
 752             rt               = _mm256_mul_ps(r00,vftabscale);
 753             vfitab           = _mm256_cvttps_epi32(rt);
 754             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 755             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 756             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 757             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 758             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 759             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 760
 761             /* REACTION-FIELD ELECTROSTATICS */
 762             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 763             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 764
 765             /* CUBIC SPLINE TABLE DISPERSION */
 766             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 767                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 768             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 769                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 770             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 771                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 772             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 773                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 774             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 775             Heps             = _mm256_mul_ps(vfeps,H);
 776             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 777             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 778             vvdw6            = _mm256_mul_ps(c6_00,VV);
 779             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 780             fvdw6            = _mm256_mul_ps(c6_00,FF);
 781
 782             /* CUBIC SPLINE TABLE REPULSION */
 783             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 784             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 785             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 786                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 787             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 788                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 789             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 790                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 791             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 792                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 793             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 794             Heps             = _mm256_mul_ps(vfeps,H);
 795             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 796             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 797             vvdw12           = _mm256_mul_ps(c12_00,VV);
 798             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 799             fvdw12           = _mm256_mul_ps(c12_00,FF);
 800             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 801             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 802
 803             /* Update potential sum for this i atom from the interaction with this j atom. */
 804             velec            = _mm256_andnot_ps(dummy_mask,velec);
 805             velecsum         = _mm256_add_ps(velecsum,velec);
 806             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 807             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 808
 809             fscal            = _mm256_add_ps(felec,fvdw);
 810
 811             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 812
 813             /* Calculate temporary vectorial force */
 814             tx               = _mm256_mul_ps(fscal,dx00);
 815             ty               = _mm256_mul_ps(fscal,dy00);
 816             tz               = _mm256_mul_ps(fscal,dz00);
 817
 818             /* Update vectorial force */
 819             fix0             = _mm256_add_ps(fix0,tx);
 820             fiy0             = _mm256_add_ps(fiy0,ty);
 821             fiz0             = _mm256_add_ps(fiz0,tz);
 822
 823             fjx0             = _mm256_add_ps(fjx0,tx);
 824             fjy0             = _mm256_add_ps(fjy0,ty);
 825             fjz0             = _mm256_add_ps(fjz0,tz);
 826
 827             /**************************
 828              * CALCULATE INTERACTIONS *
 829              **************************/
 830
 831             /* REACTION-FIELD ELECTROSTATICS */
 832             velec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
 833             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
 834
 835             /* Update potential sum for this i atom from the interaction with this j atom. */
 836             velec            = _mm256_andnot_ps(dummy_mask,velec);
 837             velecsum         = _mm256_add_ps(velecsum,velec);
 838
 839             fscal            = felec;
 840
 841             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 842
 843             /* Calculate temporary vectorial force */
 844             tx               = _mm256_mul_ps(fscal,dx01);
 845             ty               = _mm256_mul_ps(fscal,dy01);
 846             tz               = _mm256_mul_ps(fscal,dz01);
 847
 848             /* Update vectorial force */
 849             fix0             = _mm256_add_ps(fix0,tx);
 850             fiy0             = _mm256_add_ps(fiy0,ty);
 851             fiz0             = _mm256_add_ps(fiz0,tz);
 852
 853             fjx1             = _mm256_add_ps(fjx1,tx);
 854             fjy1             = _mm256_add_ps(fjy1,ty);
 855             fjz1             = _mm256_add_ps(fjz1,tz);
 856
 857             /**************************
 858              * CALCULATE INTERACTIONS *
 859              **************************/
 860
 861             /* REACTION-FIELD ELECTROSTATICS */
 862             velec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
 863             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
 864
 865             /* Update potential sum for this i atom from the interaction with this j atom. */
 866             velec            = _mm256_andnot_ps(dummy_mask,velec);
 867             velecsum         = _mm256_add_ps(velecsum,velec);
 868
 869             fscal            = felec;
 870
 871             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 872
 873             /* Calculate temporary vectorial force */
 874             tx               = _mm256_mul_ps(fscal,dx02);
 875             ty               = _mm256_mul_ps(fscal,dy02);
 876             tz               = _mm256_mul_ps(fscal,dz02);
 877
 878             /* Update vectorial force */
 879             fix0             = _mm256_add_ps(fix0,tx);
 880             fiy0             = _mm256_add_ps(fiy0,ty);
 881             fiz0             = _mm256_add_ps(fiz0,tz);
 882
 883             fjx2             = _mm256_add_ps(fjx2,tx);
 884             fjy2             = _mm256_add_ps(fjy2,ty);
 885             fjz2             = _mm256_add_ps(fjz2,tz);
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             /* REACTION-FIELD ELECTROSTATICS */
 892             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 893             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 894
 895             /* Update potential sum for this i atom from the interaction with this j atom. */
 896             velec            = _mm256_andnot_ps(dummy_mask,velec);
 897             velecsum         = _mm256_add_ps(velecsum,velec);
 898
 899             fscal            = felec;
 900
 901             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 902
 903             /* Calculate temporary vectorial force */
 904             tx               = _mm256_mul_ps(fscal,dx10);
 905             ty               = _mm256_mul_ps(fscal,dy10);
 906             tz               = _mm256_mul_ps(fscal,dz10);
 907
 908             /* Update vectorial force */
 909             fix1             = _mm256_add_ps(fix1,tx);
 910             fiy1             = _mm256_add_ps(fiy1,ty);
 911             fiz1             = _mm256_add_ps(fiz1,tz);
 912
 913             fjx0             = _mm256_add_ps(fjx0,tx);
 914             fjy0             = _mm256_add_ps(fjy0,ty);
 915             fjz0             = _mm256_add_ps(fjz0,tz);
 916
 917             /**************************
 918              * CALCULATE INTERACTIONS *
 919              **************************/
 920
 921             /* REACTION-FIELD ELECTROSTATICS */
 922             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 923             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 924
 925             /* Update potential sum for this i atom from the interaction with this j atom. */
 926             velec            = _mm256_andnot_ps(dummy_mask,velec);
 927             velecsum         = _mm256_add_ps(velecsum,velec);
 928
 929             fscal            = felec;
 930
 931             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 932
 933             /* Calculate temporary vectorial force */
 934             tx               = _mm256_mul_ps(fscal,dx11);
 935             ty               = _mm256_mul_ps(fscal,dy11);
 936             tz               = _mm256_mul_ps(fscal,dz11);
 937
 938             /* Update vectorial force */
 939             fix1             = _mm256_add_ps(fix1,tx);
 940             fiy1             = _mm256_add_ps(fiy1,ty);
 941             fiz1             = _mm256_add_ps(fiz1,tz);
 942
 943             fjx1             = _mm256_add_ps(fjx1,tx);
 944             fjy1             = _mm256_add_ps(fjy1,ty);
 945             fjz1             = _mm256_add_ps(fjz1,tz);
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             /* REACTION-FIELD ELECTROSTATICS */
 952             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 953             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 954
 955             /* Update potential sum for this i atom from the interaction with this j atom. */
 956             velec            = _mm256_andnot_ps(dummy_mask,velec);
 957             velecsum         = _mm256_add_ps(velecsum,velec);
 958
 959             fscal            = felec;
 960
 961             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 962
 963             /* Calculate temporary vectorial force */
 964             tx               = _mm256_mul_ps(fscal,dx12);
 965             ty               = _mm256_mul_ps(fscal,dy12);
 966             tz               = _mm256_mul_ps(fscal,dz12);
 967
 968             /* Update vectorial force */
 969             fix1             = _mm256_add_ps(fix1,tx);
 970             fiy1             = _mm256_add_ps(fiy1,ty);
 971             fiz1             = _mm256_add_ps(fiz1,tz);
 972
 973             fjx2             = _mm256_add_ps(fjx2,tx);
 974             fjy2             = _mm256_add_ps(fjy2,ty);
 975             fjz2             = _mm256_add_ps(fjz2,tz);
 976
 977             /**************************
 978              * CALCULATE INTERACTIONS *
 979              **************************/
 980
 981             /* REACTION-FIELD ELECTROSTATICS */
 982             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 983             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 984
 985             /* Update potential sum for this i atom from the interaction with this j atom. */
 986             velec            = _mm256_andnot_ps(dummy_mask,velec);
 987             velecsum         = _mm256_add_ps(velecsum,velec);
 988
 989             fscal            = felec;
 990
 991             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 992
 993             /* Calculate temporary vectorial force */
 994             tx               = _mm256_mul_ps(fscal,dx20);
 995             ty               = _mm256_mul_ps(fscal,dy20);
 996             tz               = _mm256_mul_ps(fscal,dz20);
 997
 998             /* Update vectorial force */
 999             fix2             = _mm256_add_ps(fix2,tx);
1000             fiy2             = _mm256_add_ps(fiy2,ty);
1001             fiz2             = _mm256_add_ps(fiz2,tz);
1002
1003             fjx0             = _mm256_add_ps(fjx0,tx);
1004             fjy0             = _mm256_add_ps(fjy0,ty);
1005             fjz0             = _mm256_add_ps(fjz0,tz);
1006
1007             /**************************
1008              * CALCULATE INTERACTIONS *
1009              **************************/
1010
1011             /* REACTION-FIELD ELECTROSTATICS */
1012             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
1013             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1014
1015             /* Update potential sum for this i atom from the interaction with this j atom. */
1016             velec            = _mm256_andnot_ps(dummy_mask,velec);
1017             velecsum         = _mm256_add_ps(velecsum,velec);
1018
1019             fscal            = felec;
1020
1021             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1022
1023             /* Calculate temporary vectorial force */
1024             tx               = _mm256_mul_ps(fscal,dx21);
1025             ty               = _mm256_mul_ps(fscal,dy21);
1026             tz               = _mm256_mul_ps(fscal,dz21);
1027
1028             /* Update vectorial force */
1029             fix2             = _mm256_add_ps(fix2,tx);
1030             fiy2             = _mm256_add_ps(fiy2,ty);
1031             fiz2             = _mm256_add_ps(fiz2,tz);
1032
1033             fjx1             = _mm256_add_ps(fjx1,tx);
1034             fjy1             = _mm256_add_ps(fjy1,ty);
1035             fjz1             = _mm256_add_ps(fjz1,tz);
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             /* REACTION-FIELD ELECTROSTATICS */
1042             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
1043             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1044
1045             /* Update potential sum for this i atom from the interaction with this j atom. */
1046             velec            = _mm256_andnot_ps(dummy_mask,velec);
1047             velecsum         = _mm256_add_ps(velecsum,velec);
1048
1049             fscal            = felec;
1050
1051             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1052
1053             /* Calculate temporary vectorial force */
1054             tx               = _mm256_mul_ps(fscal,dx22);
1055             ty               = _mm256_mul_ps(fscal,dy22);
1056             tz               = _mm256_mul_ps(fscal,dz22);
1057
1058             /* Update vectorial force */
1059             fix2             = _mm256_add_ps(fix2,tx);
1060             fiy2             = _mm256_add_ps(fiy2,ty);
1061             fiz2             = _mm256_add_ps(fiz2,tz);
1062
1063             fjx2             = _mm256_add_ps(fjx2,tx);
1064             fjy2             = _mm256_add_ps(fjy2,ty);
1065             fjz2             = _mm256_add_ps(fjz2,tz);
1066
1067             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1068             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1069             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1070             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1071             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1072             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1073             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1074             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1075
1076             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1077                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1078
1079             /* Inner loop uses 324 flops */
1080         }
1081
1082         /* End of innermost loop */
1083
1084         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1085                                                  f+i_coord_offset,fshift+i_shift_offset);
1086
1087         ggid                        = gid[iidx];
1088         /* Update potential energies */
1089         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1090         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1091
1092         /* Increment number of inner iterations */
1093         inneriter                  += j_index_end - j_index_start;
1094
1095         /* Outer loop uses 20 flops */
1096     }
1097
1098     /* Increment number of outer iterations */
1099     outeriter        += nri;
1100
1101     /* Update outer/inner flops */
1102
1103     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*324);
1104 }
1105 /*
1106  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3W3_F_avx_256_single
1107  * Electrostatics interaction: ReactionField
1108  * VdW interaction:            CubicSplineTable
1109  * Geometry:                   Water3-Water3
1110  * Calculate force/pot:        Force
1111  */
1112 void
1113 nb_kernel_ElecRF_VdwCSTab_GeomW3W3_F_avx_256_single
1114                     (t_nblist                    * gmx_restrict       nlist,
1115                      rvec                        * gmx_restrict          xx,
1116                      rvec                        * gmx_restrict          ff,
1117                      t_forcerec                  * gmx_restrict          fr,
1118                      t_mdatoms                   * gmx_restrict     mdatoms,
1119                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1120                      t_nrnb                      * gmx_restrict        nrnb)
1121 {
1122     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1123      * just 0 for non-waters.
1124      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1125      * jnr indices corresponding to data put in the four positions in the SIMD register.
1126      */
1127     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1128     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1129     int              jnrA,jnrB,jnrC,jnrD;
1130     int              jnrE,jnrF,jnrG,jnrH;
1131     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1132     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1133     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1134     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1135     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1136     real             rcutoff_scalar;
1137     real             *shiftvec,*fshift,*x,*f;
1138     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1139     real             scratch[4*DIM];
1140     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1141     real *           vdwioffsetptr0;
1142     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1143     real *           vdwioffsetptr1;
1144     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1145     real *           vdwioffsetptr2;
1146     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1147     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1148     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1149     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1150     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1151     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1152     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1153     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1154     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1155     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1156     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1157     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1158     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1159     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1160     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1161     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1162     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1163     real             *charge;
1164     int              nvdwtype;
1165     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1166     int              *vdwtype;
1167     real             *vdwparam;
1168     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1169     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1170     __m256i          vfitab;
1171     __m128i          vfitab_lo,vfitab_hi;
1172     __m128i          ifour       = _mm_set1_epi32(4);
1173     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1174     real             *vftab;
1175     __m256           dummy_mask,cutoff_mask;
1176     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1177     __m256           one     = _mm256_set1_ps(1.0);
1178     __m256           two     = _mm256_set1_ps(2.0);
1179     x                = xx[0];
1180     f                = ff[0];
1181
1182     nri              = nlist->nri;
1183     iinr             = nlist->iinr;
1184     jindex           = nlist->jindex;
1185     jjnr             = nlist->jjnr;
1186     shiftidx         = nlist->shift;
1187     gid              = nlist->gid;
1188     shiftvec         = fr->shift_vec[0];
1189     fshift           = fr->fshift[0];
1190     facel            = _mm256_set1_ps(fr->epsfac);
1191     charge           = mdatoms->chargeA;
1192     krf              = _mm256_set1_ps(fr->ic->k_rf);
1193     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
1194     crf              = _mm256_set1_ps(fr->ic->c_rf);
1195     nvdwtype         = fr->ntype;
1196     vdwparam         = fr->nbfp;
1197     vdwtype          = mdatoms->typeA;
1198
1199     vftab            = kernel_data->table_vdw->data;
1200     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
1201
1202     /* Setup water-specific parameters */
1203     inr              = nlist->iinr[0];
1204     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1205     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1206     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1207     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1208
1209     jq0              = _mm256_set1_ps(charge[inr+0]);
1210     jq1              = _mm256_set1_ps(charge[inr+1]);
1211     jq2              = _mm256_set1_ps(charge[inr+2]);
1212     vdwjidx0A        = 2*vdwtype[inr+0];
1213     qq00             = _mm256_mul_ps(iq0,jq0);
1214     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1215     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1216     qq01             = _mm256_mul_ps(iq0,jq1);
1217     qq02             = _mm256_mul_ps(iq0,jq2);
1218     qq10             = _mm256_mul_ps(iq1,jq0);
1219     qq11             = _mm256_mul_ps(iq1,jq1);
1220     qq12             = _mm256_mul_ps(iq1,jq2);
1221     qq20             = _mm256_mul_ps(iq2,jq0);
1222     qq21             = _mm256_mul_ps(iq2,jq1);
1223     qq22             = _mm256_mul_ps(iq2,jq2);
1224
1225     /* Avoid stupid compiler warnings */
1226     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1227     j_coord_offsetA = 0;
1228     j_coord_offsetB = 0;
1229     j_coord_offsetC = 0;
1230     j_coord_offsetD = 0;
1231     j_coord_offsetE = 0;
1232     j_coord_offsetF = 0;
1233     j_coord_offsetG = 0;
1234     j_coord_offsetH = 0;
1235
1236     outeriter        = 0;
1237     inneriter        = 0;
1238
1239     for(iidx=0;iidx<4*DIM;iidx++)
1240     {
1241         scratch[iidx] = 0.0;
1242     }
1243
1244     /* Start outer loop over neighborlists */
1245     for(iidx=0; iidx<nri; iidx++)
1246     {
1247         /* Load shift vector for this list */
1248         i_shift_offset   = DIM*shiftidx[iidx];
1249
1250         /* Load limits for loop over neighbors */
1251         j_index_start    = jindex[iidx];
1252         j_index_end      = jindex[iidx+1];
1253
1254         /* Get outer coordinate index */
1255         inr              = iinr[iidx];
1256         i_coord_offset   = DIM*inr;
1257
1258         /* Load i particle coords and add shift vector */
1259         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1260                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1261
1262         fix0             = _mm256_setzero_ps();
1263         fiy0             = _mm256_setzero_ps();
1264         fiz0             = _mm256_setzero_ps();
1265         fix1             = _mm256_setzero_ps();
1266         fiy1             = _mm256_setzero_ps();
1267         fiz1             = _mm256_setzero_ps();
1268         fix2             = _mm256_setzero_ps();
1269         fiy2             = _mm256_setzero_ps();
1270         fiz2             = _mm256_setzero_ps();
1271
1272         /* Start inner kernel loop */
1273         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1274         {
1275
1276             /* Get j neighbor index, and coordinate index */
1277             jnrA             = jjnr[jidx];
1278             jnrB             = jjnr[jidx+1];
1279             jnrC             = jjnr[jidx+2];
1280             jnrD             = jjnr[jidx+3];
1281             jnrE             = jjnr[jidx+4];
1282             jnrF             = jjnr[jidx+5];
1283             jnrG             = jjnr[jidx+6];
1284             jnrH             = jjnr[jidx+7];
1285             j_coord_offsetA  = DIM*jnrA;
1286             j_coord_offsetB  = DIM*jnrB;
1287             j_coord_offsetC  = DIM*jnrC;
1288             j_coord_offsetD  = DIM*jnrD;
1289             j_coord_offsetE  = DIM*jnrE;
1290             j_coord_offsetF  = DIM*jnrF;
1291             j_coord_offsetG  = DIM*jnrG;
1292             j_coord_offsetH  = DIM*jnrH;
1293
1294             /* load j atom coordinates */
1295             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1296                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1297                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1298                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1299                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1300
1301             /* Calculate displacement vector */
1302             dx00             = _mm256_sub_ps(ix0,jx0);
1303             dy00             = _mm256_sub_ps(iy0,jy0);
1304             dz00             = _mm256_sub_ps(iz0,jz0);
1305             dx01             = _mm256_sub_ps(ix0,jx1);
1306             dy01             = _mm256_sub_ps(iy0,jy1);
1307             dz01             = _mm256_sub_ps(iz0,jz1);
1308             dx02             = _mm256_sub_ps(ix0,jx2);
1309             dy02             = _mm256_sub_ps(iy0,jy2);
1310             dz02             = _mm256_sub_ps(iz0,jz2);
1311             dx10             = _mm256_sub_ps(ix1,jx0);
1312             dy10             = _mm256_sub_ps(iy1,jy0);
1313             dz10             = _mm256_sub_ps(iz1,jz0);
1314             dx11             = _mm256_sub_ps(ix1,jx1);
1315             dy11             = _mm256_sub_ps(iy1,jy1);
1316             dz11             = _mm256_sub_ps(iz1,jz1);
1317             dx12             = _mm256_sub_ps(ix1,jx2);
1318             dy12             = _mm256_sub_ps(iy1,jy2);
1319             dz12             = _mm256_sub_ps(iz1,jz2);
1320             dx20             = _mm256_sub_ps(ix2,jx0);
1321             dy20             = _mm256_sub_ps(iy2,jy0);
1322             dz20             = _mm256_sub_ps(iz2,jz0);
1323             dx21             = _mm256_sub_ps(ix2,jx1);
1324             dy21             = _mm256_sub_ps(iy2,jy1);
1325             dz21             = _mm256_sub_ps(iz2,jz1);
1326             dx22             = _mm256_sub_ps(ix2,jx2);
1327             dy22             = _mm256_sub_ps(iy2,jy2);
1328             dz22             = _mm256_sub_ps(iz2,jz2);
1329
1330             /* Calculate squared distance and things based on it */
1331             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1332             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1333             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1334             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1335             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1336             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1337             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1338             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1339             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1340
1341             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1342             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1343             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1344             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1345             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1346             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1347             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1348             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1349             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1350
1351             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1352             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1353             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1354             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1355             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1356             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1357             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1358             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1359             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1360
1361             fjx0             = _mm256_setzero_ps();
1362             fjy0             = _mm256_setzero_ps();
1363             fjz0             = _mm256_setzero_ps();
1364             fjx1             = _mm256_setzero_ps();
1365             fjy1             = _mm256_setzero_ps();
1366             fjz1             = _mm256_setzero_ps();
1367             fjx2             = _mm256_setzero_ps();
1368             fjy2             = _mm256_setzero_ps();
1369             fjz2             = _mm256_setzero_ps();
1370
1371             /**************************
1372              * CALCULATE INTERACTIONS *
1373              **************************/
1374
1375             r00              = _mm256_mul_ps(rsq00,rinv00);
1376
1377             /* Calculate table index by multiplying r with table scale and truncate to integer */
1378             rt               = _mm256_mul_ps(r00,vftabscale);
1379             vfitab           = _mm256_cvttps_epi32(rt);
1380             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1381             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1382             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1383             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1384             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1385             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1386
1387             /* REACTION-FIELD ELECTROSTATICS */
1388             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1389
1390             /* CUBIC SPLINE TABLE DISPERSION */
1391             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1392                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1393             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1394                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1395             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1396                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1397             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1398                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1399             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1400             Heps             = _mm256_mul_ps(vfeps,H);
1401             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1402             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1403             fvdw6            = _mm256_mul_ps(c6_00,FF);
1404
1405             /* CUBIC SPLINE TABLE REPULSION */
1406             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1407             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1408             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1409                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1410             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1411                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1412             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1413                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1414             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1415                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1416             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1417             Heps             = _mm256_mul_ps(vfeps,H);
1418             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1419             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1420             fvdw12           = _mm256_mul_ps(c12_00,FF);
1421             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1422
1423             fscal            = _mm256_add_ps(felec,fvdw);
1424
1425             /* Calculate temporary vectorial force */
1426             tx               = _mm256_mul_ps(fscal,dx00);
1427             ty               = _mm256_mul_ps(fscal,dy00);
1428             tz               = _mm256_mul_ps(fscal,dz00);
1429
1430             /* Update vectorial force */
1431             fix0             = _mm256_add_ps(fix0,tx);
1432             fiy0             = _mm256_add_ps(fiy0,ty);
1433             fiz0             = _mm256_add_ps(fiz0,tz);
1434
1435             fjx0             = _mm256_add_ps(fjx0,tx);
1436             fjy0             = _mm256_add_ps(fjy0,ty);
1437             fjz0             = _mm256_add_ps(fjz0,tz);
1438
1439             /**************************
1440              * CALCULATE INTERACTIONS *
1441              **************************/
1442
1443             /* REACTION-FIELD ELECTROSTATICS */
1444             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1445
1446             fscal            = felec;
1447
1448             /* Calculate temporary vectorial force */
1449             tx               = _mm256_mul_ps(fscal,dx01);
1450             ty               = _mm256_mul_ps(fscal,dy01);
1451             tz               = _mm256_mul_ps(fscal,dz01);
1452
1453             /* Update vectorial force */
1454             fix0             = _mm256_add_ps(fix0,tx);
1455             fiy0             = _mm256_add_ps(fiy0,ty);
1456             fiz0             = _mm256_add_ps(fiz0,tz);
1457
1458             fjx1             = _mm256_add_ps(fjx1,tx);
1459             fjy1             = _mm256_add_ps(fjy1,ty);
1460             fjz1             = _mm256_add_ps(fjz1,tz);
1461
1462             /**************************
1463              * CALCULATE INTERACTIONS *
1464              **************************/
1465
1466             /* REACTION-FIELD ELECTROSTATICS */
1467             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1468
1469             fscal            = felec;
1470
1471             /* Calculate temporary vectorial force */
1472             tx               = _mm256_mul_ps(fscal,dx02);
1473             ty               = _mm256_mul_ps(fscal,dy02);
1474             tz               = _mm256_mul_ps(fscal,dz02);
1475
1476             /* Update vectorial force */
1477             fix0             = _mm256_add_ps(fix0,tx);
1478             fiy0             = _mm256_add_ps(fiy0,ty);
1479             fiz0             = _mm256_add_ps(fiz0,tz);
1480
1481             fjx2             = _mm256_add_ps(fjx2,tx);
1482             fjy2             = _mm256_add_ps(fjy2,ty);
1483             fjz2             = _mm256_add_ps(fjz2,tz);
1484
1485             /**************************
1486              * CALCULATE INTERACTIONS *
1487              **************************/
1488
1489             /* REACTION-FIELD ELECTROSTATICS */
1490             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1491
1492             fscal            = felec;
1493
1494             /* Calculate temporary vectorial force */
1495             tx               = _mm256_mul_ps(fscal,dx10);
1496             ty               = _mm256_mul_ps(fscal,dy10);
1497             tz               = _mm256_mul_ps(fscal,dz10);
1498
1499             /* Update vectorial force */
1500             fix1             = _mm256_add_ps(fix1,tx);
1501             fiy1             = _mm256_add_ps(fiy1,ty);
1502             fiz1             = _mm256_add_ps(fiz1,tz);
1503
1504             fjx0             = _mm256_add_ps(fjx0,tx);
1505             fjy0             = _mm256_add_ps(fjy0,ty);
1506             fjz0             = _mm256_add_ps(fjz0,tz);
1507
1508             /**************************
1509              * CALCULATE INTERACTIONS *
1510              **************************/
1511
1512             /* REACTION-FIELD ELECTROSTATICS */
1513             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1514
1515             fscal            = felec;
1516
1517             /* Calculate temporary vectorial force */
1518             tx               = _mm256_mul_ps(fscal,dx11);
1519             ty               = _mm256_mul_ps(fscal,dy11);
1520             tz               = _mm256_mul_ps(fscal,dz11);
1521
1522             /* Update vectorial force */
1523             fix1             = _mm256_add_ps(fix1,tx);
1524             fiy1             = _mm256_add_ps(fiy1,ty);
1525             fiz1             = _mm256_add_ps(fiz1,tz);
1526
1527             fjx1             = _mm256_add_ps(fjx1,tx);
1528             fjy1             = _mm256_add_ps(fjy1,ty);
1529             fjz1             = _mm256_add_ps(fjz1,tz);
1530
1531             /**************************
1532              * CALCULATE INTERACTIONS *
1533              **************************/
1534
1535             /* REACTION-FIELD ELECTROSTATICS */
1536             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1537
1538             fscal            = felec;
1539
1540             /* Calculate temporary vectorial force */
1541             tx               = _mm256_mul_ps(fscal,dx12);
1542             ty               = _mm256_mul_ps(fscal,dy12);
1543             tz               = _mm256_mul_ps(fscal,dz12);
1544
1545             /* Update vectorial force */
1546             fix1             = _mm256_add_ps(fix1,tx);
1547             fiy1             = _mm256_add_ps(fiy1,ty);
1548             fiz1             = _mm256_add_ps(fiz1,tz);
1549
1550             fjx2             = _mm256_add_ps(fjx2,tx);
1551             fjy2             = _mm256_add_ps(fjy2,ty);
1552             fjz2             = _mm256_add_ps(fjz2,tz);
1553
1554             /**************************
1555              * CALCULATE INTERACTIONS *
1556              **************************/
1557
1558             /* REACTION-FIELD ELECTROSTATICS */
1559             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1560
1561             fscal            = felec;
1562
1563             /* Calculate temporary vectorial force */
1564             tx               = _mm256_mul_ps(fscal,dx20);
1565             ty               = _mm256_mul_ps(fscal,dy20);
1566             tz               = _mm256_mul_ps(fscal,dz20);
1567
1568             /* Update vectorial force */
1569             fix2             = _mm256_add_ps(fix2,tx);
1570             fiy2             = _mm256_add_ps(fiy2,ty);
1571             fiz2             = _mm256_add_ps(fiz2,tz);
1572
1573             fjx0             = _mm256_add_ps(fjx0,tx);
1574             fjy0             = _mm256_add_ps(fjy0,ty);
1575             fjz0             = _mm256_add_ps(fjz0,tz);
1576
1577             /**************************
1578              * CALCULATE INTERACTIONS *
1579              **************************/
1580
1581             /* REACTION-FIELD ELECTROSTATICS */
1582             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1583
1584             fscal            = felec;
1585
1586             /* Calculate temporary vectorial force */
1587             tx               = _mm256_mul_ps(fscal,dx21);
1588             ty               = _mm256_mul_ps(fscal,dy21);
1589             tz               = _mm256_mul_ps(fscal,dz21);
1590
1591             /* Update vectorial force */
1592             fix2             = _mm256_add_ps(fix2,tx);
1593             fiy2             = _mm256_add_ps(fiy2,ty);
1594             fiz2             = _mm256_add_ps(fiz2,tz);
1595
1596             fjx1             = _mm256_add_ps(fjx1,tx);
1597             fjy1             = _mm256_add_ps(fjy1,ty);
1598             fjz1             = _mm256_add_ps(fjz1,tz);
1599
1600             /**************************
1601              * CALCULATE INTERACTIONS *
1602              **************************/
1603
1604             /* REACTION-FIELD ELECTROSTATICS */
1605             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1606
1607             fscal            = felec;
1608
1609             /* Calculate temporary vectorial force */
1610             tx               = _mm256_mul_ps(fscal,dx22);
1611             ty               = _mm256_mul_ps(fscal,dy22);
1612             tz               = _mm256_mul_ps(fscal,dz22);
1613
1614             /* Update vectorial force */
1615             fix2             = _mm256_add_ps(fix2,tx);
1616             fiy2             = _mm256_add_ps(fiy2,ty);
1617             fiz2             = _mm256_add_ps(fiz2,tz);
1618
1619             fjx2             = _mm256_add_ps(fjx2,tx);
1620             fjy2             = _mm256_add_ps(fjy2,ty);
1621             fjz2             = _mm256_add_ps(fjz2,tz);
1622
1623             fjptrA             = f+j_coord_offsetA;
1624             fjptrB             = f+j_coord_offsetB;
1625             fjptrC             = f+j_coord_offsetC;
1626             fjptrD             = f+j_coord_offsetD;
1627             fjptrE             = f+j_coord_offsetE;
1628             fjptrF             = f+j_coord_offsetF;
1629             fjptrG             = f+j_coord_offsetG;
1630             fjptrH             = f+j_coord_offsetH;
1631
1632             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1633                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1634
1635             /* Inner loop uses 270 flops */
1636         }
1637
1638         if(jidx<j_index_end)
1639         {
1640
1641             /* Get j neighbor index, and coordinate index */
1642             jnrlistA         = jjnr[jidx];
1643             jnrlistB         = jjnr[jidx+1];
1644             jnrlistC         = jjnr[jidx+2];
1645             jnrlistD         = jjnr[jidx+3];
1646             jnrlistE         = jjnr[jidx+4];
1647             jnrlistF         = jjnr[jidx+5];
1648             jnrlistG         = jjnr[jidx+6];
1649             jnrlistH         = jjnr[jidx+7];
1650             /* Sign of each element will be negative for non-real atoms.
1651              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1652              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1653              */
1654             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1655                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1656
1657             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1658             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1659             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1660             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1661             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1662             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1663             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1664             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1665             j_coord_offsetA  = DIM*jnrA;
1666             j_coord_offsetB  = DIM*jnrB;
1667             j_coord_offsetC  = DIM*jnrC;
1668             j_coord_offsetD  = DIM*jnrD;
1669             j_coord_offsetE  = DIM*jnrE;
1670             j_coord_offsetF  = DIM*jnrF;
1671             j_coord_offsetG  = DIM*jnrG;
1672             j_coord_offsetH  = DIM*jnrH;
1673
1674             /* load j atom coordinates */
1675             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1676                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1677                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1678                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1679                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1680
1681             /* Calculate displacement vector */
1682             dx00             = _mm256_sub_ps(ix0,jx0);
1683             dy00             = _mm256_sub_ps(iy0,jy0);
1684             dz00             = _mm256_sub_ps(iz0,jz0);
1685             dx01             = _mm256_sub_ps(ix0,jx1);
1686             dy01             = _mm256_sub_ps(iy0,jy1);
1687             dz01             = _mm256_sub_ps(iz0,jz1);
1688             dx02             = _mm256_sub_ps(ix0,jx2);
1689             dy02             = _mm256_sub_ps(iy0,jy2);
1690             dz02             = _mm256_sub_ps(iz0,jz2);
1691             dx10             = _mm256_sub_ps(ix1,jx0);
1692             dy10             = _mm256_sub_ps(iy1,jy0);
1693             dz10             = _mm256_sub_ps(iz1,jz0);
1694             dx11             = _mm256_sub_ps(ix1,jx1);
1695             dy11             = _mm256_sub_ps(iy1,jy1);
1696             dz11             = _mm256_sub_ps(iz1,jz1);
1697             dx12             = _mm256_sub_ps(ix1,jx2);
1698             dy12             = _mm256_sub_ps(iy1,jy2);
1699             dz12             = _mm256_sub_ps(iz1,jz2);
1700             dx20             = _mm256_sub_ps(ix2,jx0);
1701             dy20             = _mm256_sub_ps(iy2,jy0);
1702             dz20             = _mm256_sub_ps(iz2,jz0);
1703             dx21             = _mm256_sub_ps(ix2,jx1);
1704             dy21             = _mm256_sub_ps(iy2,jy1);
1705             dz21             = _mm256_sub_ps(iz2,jz1);
1706             dx22             = _mm256_sub_ps(ix2,jx2);
1707             dy22             = _mm256_sub_ps(iy2,jy2);
1708             dz22             = _mm256_sub_ps(iz2,jz2);
1709
1710             /* Calculate squared distance and things based on it */
1711             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1712             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1713             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1714             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1715             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1716             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1717             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1718             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1719             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1720
1721             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1722             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1723             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1724             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1725             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1726             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1727             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1728             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1729             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1730
1731             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1732             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1733             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1734             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1735             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1736             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1737             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1738             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1739             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1740
1741             fjx0             = _mm256_setzero_ps();
1742             fjy0             = _mm256_setzero_ps();
1743             fjz0             = _mm256_setzero_ps();
1744             fjx1             = _mm256_setzero_ps();
1745             fjy1             = _mm256_setzero_ps();
1746             fjz1             = _mm256_setzero_ps();
1747             fjx2             = _mm256_setzero_ps();
1748             fjy2             = _mm256_setzero_ps();
1749             fjz2             = _mm256_setzero_ps();
1750
1751             /**************************
1752              * CALCULATE INTERACTIONS *
1753              **************************/
1754
1755             r00              = _mm256_mul_ps(rsq00,rinv00);
1756             r00              = _mm256_andnot_ps(dummy_mask,r00);
1757
1758             /* Calculate table index by multiplying r with table scale and truncate to integer */
1759             rt               = _mm256_mul_ps(r00,vftabscale);
1760             vfitab           = _mm256_cvttps_epi32(rt);
1761             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1762             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1763             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1764             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1765             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1766             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1767
1768             /* REACTION-FIELD ELECTROSTATICS */
1769             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1770
1771             /* CUBIC SPLINE TABLE DISPERSION */
1772             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1773                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1774             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1775                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1776             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1777                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1778             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1779                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1780             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1781             Heps             = _mm256_mul_ps(vfeps,H);
1782             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1783             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1784             fvdw6            = _mm256_mul_ps(c6_00,FF);
1785
1786             /* CUBIC SPLINE TABLE REPULSION */
1787             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1788             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1789             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1790                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1791             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1792                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1793             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1794                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1795             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1796                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1797             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1798             Heps             = _mm256_mul_ps(vfeps,H);
1799             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1800             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1801             fvdw12           = _mm256_mul_ps(c12_00,FF);
1802             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1803
1804             fscal            = _mm256_add_ps(felec,fvdw);
1805
1806             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1807
1808             /* Calculate temporary vectorial force */
1809             tx               = _mm256_mul_ps(fscal,dx00);
1810             ty               = _mm256_mul_ps(fscal,dy00);
1811             tz               = _mm256_mul_ps(fscal,dz00);
1812
1813             /* Update vectorial force */
1814             fix0             = _mm256_add_ps(fix0,tx);
1815             fiy0             = _mm256_add_ps(fiy0,ty);
1816             fiz0             = _mm256_add_ps(fiz0,tz);
1817
1818             fjx0             = _mm256_add_ps(fjx0,tx);
1819             fjy0             = _mm256_add_ps(fjy0,ty);
1820             fjz0             = _mm256_add_ps(fjz0,tz);
1821
1822             /**************************
1823              * CALCULATE INTERACTIONS *
1824              **************************/
1825
1826             /* REACTION-FIELD ELECTROSTATICS */
1827             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1828
1829             fscal            = felec;
1830
1831             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1832
1833             /* Calculate temporary vectorial force */
1834             tx               = _mm256_mul_ps(fscal,dx01);
1835             ty               = _mm256_mul_ps(fscal,dy01);
1836             tz               = _mm256_mul_ps(fscal,dz01);
1837
1838             /* Update vectorial force */
1839             fix0             = _mm256_add_ps(fix0,tx);
1840             fiy0             = _mm256_add_ps(fiy0,ty);
1841             fiz0             = _mm256_add_ps(fiz0,tz);
1842
1843             fjx1             = _mm256_add_ps(fjx1,tx);
1844             fjy1             = _mm256_add_ps(fjy1,ty);
1845             fjz1             = _mm256_add_ps(fjz1,tz);
1846
1847             /**************************
1848              * CALCULATE INTERACTIONS *
1849              **************************/
1850
1851             /* REACTION-FIELD ELECTROSTATICS */
1852             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1853
1854             fscal            = felec;
1855
1856             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1857
1858             /* Calculate temporary vectorial force */
1859             tx               = _mm256_mul_ps(fscal,dx02);
1860             ty               = _mm256_mul_ps(fscal,dy02);
1861             tz               = _mm256_mul_ps(fscal,dz02);
1862
1863             /* Update vectorial force */
1864             fix0             = _mm256_add_ps(fix0,tx);
1865             fiy0             = _mm256_add_ps(fiy0,ty);
1866             fiz0             = _mm256_add_ps(fiz0,tz);
1867
1868             fjx2             = _mm256_add_ps(fjx2,tx);
1869             fjy2             = _mm256_add_ps(fjy2,ty);
1870             fjz2             = _mm256_add_ps(fjz2,tz);
1871
1872             /**************************
1873              * CALCULATE INTERACTIONS *
1874              **************************/
1875
1876             /* REACTION-FIELD ELECTROSTATICS */
1877             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1878
1879             fscal            = felec;
1880
1881             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1882
1883             /* Calculate temporary vectorial force */
1884             tx               = _mm256_mul_ps(fscal,dx10);
1885             ty               = _mm256_mul_ps(fscal,dy10);
1886             tz               = _mm256_mul_ps(fscal,dz10);
1887
1888             /* Update vectorial force */
1889             fix1             = _mm256_add_ps(fix1,tx);
1890             fiy1             = _mm256_add_ps(fiy1,ty);
1891             fiz1             = _mm256_add_ps(fiz1,tz);
1892
1893             fjx0             = _mm256_add_ps(fjx0,tx);
1894             fjy0             = _mm256_add_ps(fjy0,ty);
1895             fjz0             = _mm256_add_ps(fjz0,tz);
1896
1897             /**************************
1898              * CALCULATE INTERACTIONS *
1899              **************************/
1900
1901             /* REACTION-FIELD ELECTROSTATICS */
1902             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1903
1904             fscal            = felec;
1905
1906             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1907
1908             /* Calculate temporary vectorial force */
1909             tx               = _mm256_mul_ps(fscal,dx11);
1910             ty               = _mm256_mul_ps(fscal,dy11);
1911             tz               = _mm256_mul_ps(fscal,dz11);
1912
1913             /* Update vectorial force */
1914             fix1             = _mm256_add_ps(fix1,tx);
1915             fiy1             = _mm256_add_ps(fiy1,ty);
1916             fiz1             = _mm256_add_ps(fiz1,tz);
1917
1918             fjx1             = _mm256_add_ps(fjx1,tx);
1919             fjy1             = _mm256_add_ps(fjy1,ty);
1920             fjz1             = _mm256_add_ps(fjz1,tz);
1921
1922             /**************************
1923              * CALCULATE INTERACTIONS *
1924              **************************/
1925
1926             /* REACTION-FIELD ELECTROSTATICS */
1927             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1928
1929             fscal            = felec;
1930
1931             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1932
1933             /* Calculate temporary vectorial force */
1934             tx               = _mm256_mul_ps(fscal,dx12);
1935             ty               = _mm256_mul_ps(fscal,dy12);
1936             tz               = _mm256_mul_ps(fscal,dz12);
1937
1938             /* Update vectorial force */
1939             fix1             = _mm256_add_ps(fix1,tx);
1940             fiy1             = _mm256_add_ps(fiy1,ty);
1941             fiz1             = _mm256_add_ps(fiz1,tz);
1942
1943             fjx2             = _mm256_add_ps(fjx2,tx);
1944             fjy2             = _mm256_add_ps(fjy2,ty);
1945             fjz2             = _mm256_add_ps(fjz2,tz);
1946
1947             /**************************
1948              * CALCULATE INTERACTIONS *
1949              **************************/
1950
1951             /* REACTION-FIELD ELECTROSTATICS */
1952             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1953
1954             fscal            = felec;
1955
1956             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1957
1958             /* Calculate temporary vectorial force */
1959             tx               = _mm256_mul_ps(fscal,dx20);
1960             ty               = _mm256_mul_ps(fscal,dy20);
1961             tz               = _mm256_mul_ps(fscal,dz20);
1962
1963             /* Update vectorial force */
1964             fix2             = _mm256_add_ps(fix2,tx);
1965             fiy2             = _mm256_add_ps(fiy2,ty);
1966             fiz2             = _mm256_add_ps(fiz2,tz);
1967
1968             fjx0             = _mm256_add_ps(fjx0,tx);
1969             fjy0             = _mm256_add_ps(fjy0,ty);
1970             fjz0             = _mm256_add_ps(fjz0,tz);
1971
1972             /**************************
1973              * CALCULATE INTERACTIONS *
1974              **************************/
1975
1976             /* REACTION-FIELD ELECTROSTATICS */
1977             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1978
1979             fscal            = felec;
1980
1981             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1982
1983             /* Calculate temporary vectorial force */
1984             tx               = _mm256_mul_ps(fscal,dx21);
1985             ty               = _mm256_mul_ps(fscal,dy21);
1986             tz               = _mm256_mul_ps(fscal,dz21);
1987
1988             /* Update vectorial force */
1989             fix2             = _mm256_add_ps(fix2,tx);
1990             fiy2             = _mm256_add_ps(fiy2,ty);
1991             fiz2             = _mm256_add_ps(fiz2,tz);
1992
1993             fjx1             = _mm256_add_ps(fjx1,tx);
1994             fjy1             = _mm256_add_ps(fjy1,ty);
1995             fjz1             = _mm256_add_ps(fjz1,tz);
1996
1997             /**************************
1998              * CALCULATE INTERACTIONS *
1999              **************************/
2000
2001             /* REACTION-FIELD ELECTROSTATICS */
2002             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
2003
2004             fscal            = felec;
2005
2006             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2007
2008             /* Calculate temporary vectorial force */
2009             tx               = _mm256_mul_ps(fscal,dx22);
2010             ty               = _mm256_mul_ps(fscal,dy22);
2011             tz               = _mm256_mul_ps(fscal,dz22);
2012
2013             /* Update vectorial force */
2014             fix2             = _mm256_add_ps(fix2,tx);
2015             fiy2             = _mm256_add_ps(fiy2,ty);
2016             fiz2             = _mm256_add_ps(fiz2,tz);
2017
2018             fjx2             = _mm256_add_ps(fjx2,tx);
2019             fjy2             = _mm256_add_ps(fjy2,ty);
2020             fjz2             = _mm256_add_ps(fjz2,tz);
2021
2022             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2023             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2024             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2025             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2026             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2027             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2028             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2029             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2030
2031             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2032                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2033
2034             /* Inner loop uses 271 flops */
2035         }
2036
2037         /* End of innermost loop */
2038
2039         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2040                                                  f+i_coord_offset,fshift+i_shift_offset);
2041
2042         /* Increment number of inner iterations */
2043         inneriter                  += j_index_end - j_index_start;
2044
2045         /* Outer loop uses 18 flops */
2046     }
2047
2048     /* Increment number of outer iterations */
2049     outeriter        += nri;
2050
2051     /* Update outer/inner flops */
2052
2053     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*271);
2054 }