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